The Syntax of Specifiers and Heads: Collected Essays of Hilda J. Koopman (Routledge Leading Linguists 3)

THE SYNTAX OF SPECIFIERS AND HEADS Since the 1980s there has been remarkable progress in syntactic theory and particul...

Author: Hilda J Koopman

54 downloads 630 Views 2MB Size Report

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

DOWNLOAD PDF

THE SYNTAX OF SPECIFIERS AND HEADS

Since the 1980s there has been remarkable progress in syntactic theory and particularly in the understanding of the structure of the theory. This book brings together the most important of Hilda J.Koopman’s essays in the field, written during this time of rapid theoretical development. The Syntax of Specifiers and Heads covers such topics as: • • • • •

interpretation and distribution of pronouns ECP effects specifiers and phrase structure the role and functioning of head movement the architecture of grammar

Each chapter draws syntactic arguments from phenomena in a broad range of languages and brings these to bear on the structure of syntactic theory and the understanding of crosslinguistic variation. Among the languages studied are the African languages, Welsh and Irish, Norwegian, French, English and Dutch. Professor Koopman has added to this collection an overview of advances in syntactic theory since the early 1980s and offers her analysis of future directions in the field. The Syntax of Specifiers and Heads will become a valuable resource for all those interested in syntactic structures and the development of linguistic theory. Hilda J.Koopman received her PhD from the University of Tilburg in 1984 and held a research position of the University of Quebec for several years. She is currently Professor of Linguistics at the University of California, Los Angeles.

ROUTLEDGE LEADING LINGUISTS Series editor: Carlos Otero 1 ON SYNTAX AND SEMANTICS James Higginbotham 2 PARTITIONS AND ATOMS OF CLAUSE STRUCTURE Subjects, agreement, case and clitics Dominique Sportiche 3 THE SYNTAX OF SPECIFIERS AND HEADS Collected essays of Hilda J.Koopman 4 CONFIGURATIONS OF SENTENTIAL COMPLEMENTATION Perspectives from Romance Johan Rooryck

THE SYNTAX OF SPECIFIERS AND HEADS Collected Essays of

Hilda J.Koopman

London and New York

First published 2000 by Routledge 11 New Fetter Lane, London EC4P 4EE Simultaneously published in the USA and Canada by Routledge 29 West 35th Street, New York, NY 10001 Routledge is an imprint of the Taylor & Francis group. This edition published in the Taylor & Francis e-Library, 2005. “To purchase your own copy of this or any of Taylor & Francis or Routledge’s collection of thousands of eBooks please go to www.eBookstore.tandf.co.uk.” © 2000 Hilda J. Koopman All rights reserved. No part of this book may be reprinted or reproduced or utilised in any form or by any electronic, mechanical, or other means, now known or hereafter invented, including photocopying and recording, or in any information storage or retrieval system, without permission in writing from the publishers. British Library Cataloguing in Publication Data A catalogue record for this book is available from the British Library. Library of Congress Cataloguing in Publication Data A catalogue record for this book has been requested. ISBN 0-203-02786-8 Master e-book ISBN

ISBN 0-203-17366-X (Adobe eReader Format) ISBN 0-415-16183-5 (Print Edition)

CONTENTS

Acknowledgments Introduction PART I Pronouns and WCO 1 Variables and the Bijection Principle with Dominique Sportiche 2 Pronouns, logical variables, and logophoricity in Abe with Dominique Sportiche 3 The internal and external distribution of pronominal DPs

vi 1

16 37 74

PART II ECP 4 ECP effects in main clauses 5 Control from COMP and comparative syntax 6 A note on long extraction in Vata and the ECP with Dominique Sportiche

117 122 147

PART III Specifiers and phrase structure 7 On the absence of case chains in Bambara 8 Prepositions, postpositions, circumpositions, and particles: the structure of Dutch PPs

164 198

PART IV Heads 9 Licensing heads 10

On Verbs that fail to undergo V-second

PART V Specifiers and Heads 11 The Spec head configuration 12 Unifying predicate cleft constructions Index

257 295

323 357 375

ACKNOWLEDGMENTS

“Variables and the bijection principle” appeared in The Linguistic Review, 2:3, 39–160; “Pronouns, logical variables and logophoricity in Abe” in Linguistic Inquiry, 20:4, 555–589; “ECP effects in main clauses”, in Linguistic Inquiry, 14: 2, 346–350; “Control from COMP and comparative syntax”, in The Linguistic Review 2:4, 365–391; “A note on long extraction in Vata and the ECP”, in Natural Language and Linguistic Theory, vol. 4, no. 3, 357–374; “On the absence of case chains in Bambara” in Natural Language and Linguistics Theory, 10: 555– 594; “Licensing heads”, in D.Lightfoot and N.Hornstein (eds) Verb-Movement, 261–296, Cambridge University Press; “On Verbs that fail to undergo V-second”, in Linguistic Inquiry, 26: 1, 137–163. “The internal and external distribution of pronominal DPs”, will appear in K. Johnson and I. Roberts (eds) Beyond Principles and Parameters. Essays in Memory of Osvaldo Jaeggli, Kluwer Academic Publishers; “Unifying predicate cleft constructions” will appear in Kevin Moore (ed.) Proceedings of the twenty-third annual meeting of the Berkeley Linguistics Society, February 14–17: special session on syntax and semantics in Africa. We are grateful for relevant permissions to reprint from Cambridge University Press, Kluwer Academic Publishers, MIT Press, and Mouton de Gruyter.

INTRODUCTION

The articles collected in this volume span my work from the 1980s till the middle of the 1990s. This introduction contains a general perspective on the theoretical developments over this period, and provides some background on the issues discussed in these chapters, so as to place them in the context of their times. 1.1 Development of the theory I was fortunate to be an apprentice syntactician at the end of the 1970s and the beginning of the 1980s. This was a truly exciting time in syntax within the general framework developed by Noam Chomsky. Chomsky’s Pisa lectures, published in 1981 as Lectures on Government and Binding, marked the beginning of an extremely productive era. The Government Binding (GB) framework provided powerful analytical tools that allowed substantial broadening of empirical coverage of language internally and crosslinguistically, leading to a much better understanding of the properties of syntactic representations. It lent itself to the discovery of many new patterns in ever expanding areas of inquiry and in languages that had been little studied previously, or that had not been studied at all, leading to explosive growth of the field. The theory itself was sufficiently flexible to extend to new phenomena and to establish correlations and new connections. Kayne’s work, collected in his 1984 book, played a particularly important role in this respect. By the end of the 1980s, it had become clear that syntactic representations were large structures, much larger than one would naively expect based on the “physical” properties of a particular sentence (i.e. based on the presence of overt material in the sentence). With larger structures also came simplification of the basic structural design: syntactic structures are constructed out of basic binary branching structures, with both lexical heads and functional heads projecting. During this period, it became clear as well that there had to be much more movement (both XP and head movement) than previously thought. Movement came to play an ever increasingly important role in capturing the different types of dependencies. It became more general, but also more constrained, operating both in the overt syntax and the covert syntax (LF). This allowed the incorporation of scope phenomena (LF-syntax), and played an important role in viewing language variation as resulting from whether languages use overt or covert movement to establish the necessary dependencies. With the rise of movement the Spec Head

2 THE SYNTAX OF SPECIFIERS AND HEADS

configuration emerged as an increasingly important structural configuration, often at the expense of the canonical government relation, i.e. the head complement relation. The former emerges as the syntactic licensing configuration for XP movement (phrasal movement), the latter as the syntactic licensing condition for head movement (compare Koopman 1994). This general picture crystallized over the years because of research in many different areas: the studies of the distribution of heads and the development of the theory of head movement,1 the development of Case theory,2 the development of VP internal subjects,3 of VP external objects,4 of VP shells,5 the studies on the lexicon, resulting in an improved understanding of the mapping of thematic properties onto syntactic structure,6 the studies on the internal structure of DPs,7 the studies on scrambling,8 leading to a more refined picture of the different distributional properties of various types of phrases, the studies on the syntax semantics interface9 (LF, QR, wh-raising), the incorporation of adverbs,10 of negation,11 and so on. In the beginning of the 1990s, syntax was ready for a new cycle of theoretical progress. With larger structures, too many analytical options had become available. The theory seemed too flexible and contained too many redundancies to allow making sense out of the bewildering complexity of patterns. The different modules of the Government Binding Theory had also been sufficiently explored to see what they could capture and what they could not. Not only had it become possible, in fact it had become necessary to ask general questions about the structure of the theory and theoretical parsimony. There were several proposals in the immediate pre-Minimalist period which did exactly this. Because of a certain dissatisfaction with the Barriers Theory of Chomsky (1993), Sportiche (1990) proposed that a simpler and more elegant system could be constructed out of the same building blocks. In 1993, Sportiche proposed a ‘reductionist’ research program reducing all licensing conditions to Spec head licensing with highly constrained movement responsible for deriving the distributional properties (published as Sportiche 1995b). In Koopman (1991), published in 1994, I proposed to reduce a number of syntactic properties, including binary branching structures, to the mechanics of head movement. These developments, driven by considerations of parsimony, were close in spirit to the Minimalist Program (MP) of Chomsky (1995), though different in execution. The Minimalist Program takes our understanding of syntactic properties at the end of the 1980s as basically correct, and develops a program that yields these results parsimoniously. The MP has led to a much improved and cleaner theory. However, compared to the GB framework, the Minimalist Program led to relatively few new insights in our understanding of phenomena in the first half of the 1990s. This is probably because it did not generate new analytical tools, and thus failed to generate novel ways of looking at well-known paradigms or to expand and solve old problems, an essential ingredient for progress to be made at this point. Our understanding of empirical phenomena has grown steadily since the early 1990s. As a result the general picture has changed, and the object that the theory must account for has taken on a different shape. This is due on the one hand to many careful analyses, independently of Kayne (1994) (Nkemnji (1992, 1995), Sportiche (1995b), Koopman (1993b)). On the other hand, it results from analyses inspired by Kayne (1994). From

INTRODUCTION 3

these works, it is clear that languages make extensive use of overt pied-piping of all kinds of big constituents, dubbed by Nkemnji (1995) as “heavy pied-piping”. These movements are often “remnant” movements,12 i.e. movements of a constituent containing overt material and one or several traces. This indicates that the picture of the early 1990s, which involves articulated structures and quite a lot of movement, was simply incomplete. Structures are even larger,13 and there is more overt movement (and maybe only14 overt movement). Heavy pied-piping and remnant movement are the norm rather than the exception. Heavy pied-piping, which universal grammar makes available, plays no role in many standard analyses for well-studied languages like English. The realization that it exists dates from the 1990s, well after these analyses had been established. This implies that all old results should be reconsidered in the light of this possibility: we might simply have the wrong analysis. In my present research, I look at some old and unsolved problems15 from this perspective, and initial results are exciting and encouraging. We seem to be heading toward a different understanding of the object under investigation: syntactic theory seems yet at the beginning of another cycle of theoretical progress, with greatly improved empirical coverage. My current theoretical assumptions are largely minimalist in character, and converge in many respects with Kayne’s recent work.16 The following differences with respect to the “standard” Minimalist Program are important. While Chomsky’s MP tries to minimize the number of functional categories, I assume that structures are maximally large, constructed of a unique ever repeating design, with each feature projecting. I fully exploit large structures, and the space they provide. In particular, I assume that all functional categories are strong, forcing massive overt movement. Movement itself is not subject to Economy, but proceeds blindly and mechanically. Paradigms should fall out from the mechanics of movement, and the interaction of the numerous movements that must take place in a derivation. 1.2 Contents The chapters included in this volume are grouped together according to themes. Within each theme, chapters are organized in chronological order. A large part of my research is inspired by analytical problems of the languages I have worked on. 1.3 Pronouns The chapters in this section represent different aspects of my research on pronouns. Most of this research is directly inspired by the languages I have worked on: what pronominal features are realized, what form does the pronoun take (is it homophonous with a D or not, is it silent or not); should the overt realization of the pronominal features be analyzed as an agreement marker, a clitic, or a weak pronoun; what is the relation of pronouns and agreement; what does the pronoun cooccur with, i.e. how does it interact with its


environment. This includes the distribution and behavior of resumptive pronouns (A-bar bound pronouns), WCO (weak crossover) phenomena and pronominal binding. Chapter 1, “Variables and the Bijection Principle”, is coauthored with Dominique Sportiche, and was originally presented at the GLOW conference in Göttingen in April 1982. It introduces the Bijection Principle to account for WCO (weak crossover) phenomena. The Bijection Principle prohibits a single operator from locally binding two variables. The idea for the Bijection Principle became possible because of the properties of resumptive pronouns in some of the languages that I had worked on. Haitian has resumptive pronouns in relative clauses in the genitive and object of P position (Koopman 1982). Vata has resumptive pronouns in subject position in wh-questions (Koopman 1984).17 Resumptive pronouns in these languages distribute like overt traces: they are homophonous with weak pronouns, they obey strong islands, and occur in particular syntactic configurations only. In wh-questions in Vata, the resumptive pronoun licenses parasitic gaps, and triggers WCO effects. It therefore shows all the characteristics of being a locally A’ bound variable, except that it is overt, not covert. We proposed accordingly that overtness or covertness of variables is irrelevant to the notion of variable itself. This made it possible to explain the WCO phenomenon (*whoi does hisi mother like [e]i) by the Bijection Principle which prohibits an operator from locally A’ binding two variables. The Bijection Principle became an important tool to probe the structure of the theory through the study of parasitic gaps (Chomksy 1982). Chapter 2, “Pronouns, logical variables and logophoricity in Abe”, is coauthored with Dominique Sportiche as well. This article grew out of fieldwork on Abe, a Kwa language spoken in the Ivory Coast. Abe has an extremely interesting and complex pronominal system, with two series of third person pronouns interacting with each other in complicated ways. One series of pronouns functions as logophoric pronouns in precisely defined embedded contexts. We motivate an analysis for the pronominal system and logophoricity. As far as I know, the analysis of logophoric pronouns holds up very generally and extends to many African languages with logophoric pronouns. However, broader questions about pronominal systems remain on the general research agenda, and insights will no doubt come from the analysis of complex pronominal systems. Many of these systems have simply not yet been explored. Chapter 3, “The internal and external distribution of pronominal DPs”, specifically addresses distributional properties of pronouns. What is the relation between pronouns, clitics and agreement? I argue that an analysis of the form of pronominal elements is extremely important, because of the intimate connection between the DP internal syntax (i.e. the derived constituent structure) and the DP external syntax. I develop a detailed analysis of the pronominal system and agreement in Welsh, which yields insight into the general problem. This chapter anticipates some of the ideas of Chapter 11, “The Spec head configuration”, included in this volume. This chapter quite generally illustrates the line of argumentation that I adopt in much of my work. The form of overt material, no matter how small, matters. This is what native speakers work with, and this is what restricts the space of the possible hypotheses that the language learner can entertain. Superficial differences between languages result

INTRODUCTION 5

from the overt form of particular items to a large extent, maybe in conjunction with some extremely restricted choices at an early point in the derivation. 1.4 On the ECP The chapters in this section reflect my work on the ECP from the early to mid 1980s. They are concerned with subject, object and adjunct asymmetries under wh-movement. Most of the patterns described in these chapters remain basically unexplained. Chapter 4, “Control from COMP and comparative syntax”, examines patterns of extraction of simple wh-phrases from subject position, in main clauses (who left), and from embedded tensed complements selected by bridge verbs (Who do you think left?) in both overt and in covert syntax (i.e. wh-in situ) (*what did who see?). It examines the patterns in various languages: Vata, a Kru language spoken in the Ivory Coast, on which I did original fieldwork (Koopman 1984), English, French, Dutch, Italian, Mahou, a Mande language spoken in the Ivory Coast, and Chinese. This chapter makes several points. First, the ECP applies to both overt and covert movement. Second, though wh-phrases may remain in situ in French or in Vata, they may not remain in the syntactic subject position, and third, subject extraction to a local C node is only possible if the language uses a process that Aoun, Hornstein and Sportiche (1981) called “COMP indexing”. This rule allows the C node to head govern the trace in subject position depending on the internal structure of the COMP node (note that this rule was formulated before the incorporation of CP into X-bar theory in Chomsky 1986a): COMP indexing (Aoun, Hornstein and Sportiche 1981) [COMP[Xi”.] → [COMPi X”…] iff COMP dominates only i-indexed elements. The rule of COMP indexing, formulated as a percolation rule, with did and that in English blocking the rule but not zero C, is essentially the precursor of today’s Spec head agreement rule (Rizzi 1990). In Chapter 5, “ECP Effects in main clauses”, I show that subject extraction from main clauses and embedded tensed complement clauses pattern in the same way, and should thus receive the same explanation. In English, do-support is blocked with subject extraction in matrix clauses, in the same configuration that must be silent in embedded clauses. In Dutch, all dialects allow that-t violations,18 a fact which I correlated with V to C movement (V-second). I return to subject wh-extraction from a different angle in current work.19 The point of departure for this work can be found in the “Spec head configuration” included in this volume. Chapter 6, “A note on long extraction in Vata and the ECP” (coauthored with Dominique Sportiche) examines subject/object/adjunct asymmetries under whextraction. It provides a strong argument for the separation of the constraint which yields


locality of movement (“antecedent government”, or “Minimal Chain Link”) from the licensing condition on empty categories (i.e. head government does not obviate the need to have a local antecedent). In Vata, objects, but not subjects, adjuncts or focused Vs, may escape from wh-islands. We analyze patterns of crosslinguistic variation, and propose that these are captured by the “Condition on Long extraction” (long extraction iff extraction takes place from a theta-position, (Koopman and Sportiche 1985)). The question why certain languages show strong subject object asymmetries under whmovement (French, English, Vata), but others do not (Japanese/ Chinese) initially motivated our proposal for VP internal subjects made at GLOW 1985 in Brussels. There, we argued that the difference would follow from the Condition on Long extraction if the subject occupied different types of positions in the overt syntax, with French/English subjects occurring in a non-thematic position outside of the theta-position in the VP, and Japanese/Chinese subjects in their thematic position.20 Once we had the idea of the VP internal subject hypothesis, other motivations, already in place in the theory, suggested themselves immediately. Several version of the VP internal subject hypothesis circulated, reflecting the history of our thinking. However, as is often the case, our original motivation (Condition on long extraction) plays no role in the final version of Koopman and Sportiche (1991) (not included in the present volume, but included in Sportiche (1998)). Indeed we argued there that subjects always move and develop Case theory in such a way that subjects may move to different positions, depending on how Case is assigned. 1.5 Specifiers Much of my research revolves around figuring out structure. The idea that there is a movement dependency between the surface position of DPs in any regular active sentence and a VP internal position is illustrative of my research. I argued in Koopman (1984) that the surface position of finite verbs is not to be trusted because verbs enter into a dependency with another head position (V to I movement). In general, the surface distribution of specific elements is just that, and should not be trusted as revealing a deep property. If lexical subjects are forced to move out of the VP because of the disassociation of theta-properties and Case properties, this could hold for objects,21 objects of Ps,22 or pronouns,23 or clauses,24 as well, though not necessarily for the same reasons. My current view is that everything needs to move to get licensed.25 Case theory is a particular instantiation of a much more general licensing theory.26 Quite generally, I try to reduce analytical problems to structural properties in conjunction with simple movements, rather than looking for explanations in the mechanics. The chapters in this section explore issues in phrase structure related to Spec and head positions. Chapter 7, “On the absence of Case chains in Bambara”, was originally circulated in 1987, and appeared in 1992. This chapter looks at a number of distinctive properties of Bambara (a Mande language spoken in Mali). In Bambara, verbs that do not assign Case move to I, but verbs that do, remain in situ and a dummy inflectional element appears in I. Intransitive verbs can appear in causative constructions, but transitive verbs

INTRODUCTION 7

cannot, unless they are first passivized. These patterns abstractly resemble properties of languages with ergative/absolutive Case marking patterns, which can be viewed as resulting from problems with the licensing of accusative Case by the V.27 I proposed in this chapter that these patterns are due to the failure of verbal traces to assign Case. Case chains are quite generally blocked in Bambara (Case chains are also unavailable in the nominal system: Bambara forces overt movement in more configurations than English does). These properties (and some others), I argued, would fall out if Case chains were impossible in Bambara, but possible in English.28 In this chapter, I also proposed that overt agreement is always and only a reflection of a relation between a Spec position and a head,29 a position that I still maintain in its strongest form. This proposal, taken together with Pollock (1989) and Kayne (1984) led to the introduction of AgrO into the clausal structure (Chomsky 1991). Chapter 8, “Prepositions, postpositions, circumpositions and particles: the structure of Dutch PPs”, (Koopman 1998a) grew out of two earlier papers: “Verb particle construction and the syntax of PPs”, (Koopman 1991) and “The Structure of Dutch PPs” (Koopman 1993b), neither of which is included in the present volume. In the former, I tried to show that PPs (= Particle Phrases) in English obey the same structural design as other categories: in particular PPs contained Spec positions that could be exceptionally Case marked by V thus yielding the familiar alternation in particle constructions. In the context of its time, this article inscribed itself in the rise in importance of the Spec positions as hosts for moved material. A further question arose: if particle phrases are PPs, how are particle phrases and prepositional phrases related, and how should we capture the different distributions? I argued that the architecture of full prepositional phrases was similar to that of a clause, with the PP projection dominated by functional projections. Particle Phrases are similar to small clauses/bare VPs. “The structure of PPs in Dutch” (Koopman 1993b) develops this idea in detail, motivating it with overt empirical evidence. It was originally written as an illustration of how the theoretical assumptions of Sportiche (1990), restrict and (successfully) inform possible analytical options. I thus pleaded for theoretical rigor—it seemed at the time that almost any proposal could be made to accommodate a particular problem arising in the data— and argued in favor of the theory proposed by Sportiche (1990) as having the right properties. In this version, I proposed a head initial structure for all PPs, and relied on heavy pied-piping (i.e. movement of large structure) to derive postpositional and circumpositional structures. The chapter included in this volume is a much revised version of the (1993b) paper. It contains a detailed discussion of the behavior of directional PPs which allows us to sharpen considerably the issues with respect to directionality. 1.6 Heads The distribution of heads represents a central area of my research. My book The Syntax of Verbs: From Verb Movement Rules to Universal Grammar (Koopman 1984) appeared right at the beginning of the head movement era. I argued on the basis of Vata and Gbadi, two Kru languages spoken in the Ivory Coast, that there are several types of verbal dependencies


paralleling A-movement and A’-movement dependencies for DPs: A-type verbal dependencies between tensed C, I and V, and A’ type verbal dependency between C and I/V in predicate cleft (focused verb) constructions.30 I showed that the theoretical apparatus of the Government Binding theory, which was completely DP-centric, extended naturally to capture the properties of verbal dependencies. I argued in particular that the locality of head movement, which Travis (1984) independently attributed to the Head Movement Constraint, followed from the ECP. Included in this volume are two closely related chapters on head movement. Chapter 9, “Licensing heads”, presented at the verb movement conference in Maryland, October 1991, addresses the question why head movement exists. I explore the idea that head movement represents the fundamental licensing mechanism for heads. In particular, I develop the idea that lexical items are similar to molecules, with lexical properties represented as open positions (receptors) that must be satisfied. Satisfaction is achieved when a head moves onto the receptor. I develop the way in which head movement onto the receptor proceeds, which yields a theory of excorporation. This chapter explores a wide array of consequences of this idea. Chapter 10, “On Verbs that fail to undergo V-second” argues how this theory of head movement accounts for complex patterns of V-movement in Dutch. 1.7 Recent work The final section in this volume contains two recent chapters which represent the direction of my current work: Chapter 11 “The Spec head configuration” (Koopman 1996) and Chapter 12, “Unifying predicate cleft constructions” (Koopman 1998b). In the Spec head configuration chapter, I examine the distribution of overt material over Spec and head positions, and argue that a slight modification of Kayne’s (1994) LCA (Linear Correspondence Axiom) prohibits both a Specifier and a head of a particular projection from containing overt lexical material at spell-out: if this configuration were to arise the items would be in too symmetrical a relation to allow linearization. This in essence derives the doubly filled COMP filter, and makes it inviolable. I also argue that each projection must be activated by lexical material at some stage in the derivation (the “PPA” or “Principle of Projection Activation”). This principle forces massive movement. This chapter explores the numerous consequences of these proposals, ranging from the analysis of doubly filled C phenomena, that-t violations, neg-drop, PRO-drop, to the structure of main clauses and imperatives. In Chapter 12, “Unifying predicate cleft constructions”, I return to the predicate cleft constructions analyzed in my book as involving A-bar head movement (Koopman 1984). I propose a reanalysis in terms of remnant XP movement for these constructions, which fit the properties much better. This analysis is only available given larger structures and more movement. In this paper, I examine a particular problem that arises with respect to what has been called the Universal Base hypothesis: all languages have identical underlying orders. This hypothesis is extremely attractive (it prohibits structural variation, and thus restricts the possible space for variation), but it meets with severe problems For example,

INTRODUCTION 9

it has been argued that some languages have a high focus position, whereas others have a low focus projection. In this chapter, I show, through the patterns of so-called verbal focus construction (also known as the predicate cleft construction) in two different African languages (Vata and Nweh), that superficially low focus not only can be reduced to the same position as superficially high focus, but that it is actually desirable to do so: as a byproduct of the derivation specific cooccurrence restrictions of wh-phrases and focused verbs fall out. This chapter also describes the research program, which is to a large extent compatible with the Minimalist Program, but also differs from it in certain respects. It converges with Kayne’s recent work (Kayne 1997, 1998). I assume large uniform structures (and thus no multiple Spec positions), with each feature projecting (this is what Sportiche (1998) calls the “atomization of syntactic structure”). I exploit these large structures by moving massively, with pied-piping and remnant movements the norm. Movement “rolls” up the structure, forming “spirals”, with some element extracting out of a constituent, and the remnant moving around it to some higher position. Movement is not subject to Economy conditions, but proceeds mechanically, blindly and to the closest available attractor. In that sense movement is “wired in” (see also Chomsky (1995)). In my present research, I am pursuing the idea that all movement is overt. If this idea is correct, there is no Economy Principle Procrastinate (Chomsky 1995),31 nor is there any need to decide if a particular functional category is strong or weak. All functional categories can be assumed to be strong, meaning that they must be checked against appropriate lexical material. Initial exploration32 seems highly promising and exciting. 1.8 Some biographical notes The author was born and raised in Nijmegen in the Netherlands, the fifth child of six. She started as a student in general linguistics in 1977 in Amsterdam. She wanted to specialize in language deficits, aphasia in particular, but quickly realized the need to understand how language in normal people worked. She worked on syntax with Henk Van Riemsdijk, who later directed her PhD. She worked on the syntax of pidgins and creoles with Pieter Muysken, and with Catherine Snow and Pieter Muysken on issues related to languages in contact. In 1978, she went to Montreal to work on her MA thesis on Haitian for a couple of months. She stayed seven years. Her immediate goals were two-fold. First, she wanted to follow up some ideas in her first syntax paper, where she argued (without much evidence) that NPs had Cs, and that there was a particular particle la in Haitian which showed up whenever extraction took place. If she were right, the distribution of this particle would provide very strong evidence for successive cyclic movement, and solve the problem of how movement proceeds.33 Her second goal was to explore the syntactic properties of African languages to determine what influence, if any, they had on the formation of creole languages.34 During her stay in Montreal, she worked on Haitian creole with Claire Lefebvre, and finished her MA thesis on a different topic (wh-questions and relative clauses in Haitian). Together with Jonathan Kaye, who taught her phonology, she started doing fieldwork on Vata with Doua Blé Siméon, a native speaker of Vata and


at that time a student in the linguistics department of the Université du Quebec a Montreal. She discovered she loved fieldwork, and has been involved in fieldwork ever since. The initial work on Vata grew into a big project on African languages. During her time in Montreal, she visited MIT many times, and felt extremely fortunate to be included in the incredibly lively and active atmosphere in Building 20. She met Dominique Sportiche during this time. They have two children, Noémie and Sophie. She moved with her family to Los Angeles in 1986, where she is a Professor in the Department of Linguistics at UCLA. Notes 1 Koopman (1984), Baker (1988), Travis (1984), Pollock (1989), among others. 2 Chomsky (1980), Chomsky (1986b), among others. 3 Koopman and Sportiche (1985, 1991), Sportiche (1988), Kuroda (1988), Fukui and Speas (1986), Zagona (1982), among others. 4 Koopman and Sportiche (1985), Koopman (1986 (published in 1992)), Mahajan (1990), among others. 5 Larson (1988), Pustejovsky (1995), among others. 6 Hoekstra (1984), Belletti and Rizzi (1988), Larson (1988), Levin (1993), Pesetsky (1995), Hale and Keyser (1993), Sportiche (1995b), Kural (1996). Compare also the publications of the Lexicon project at MIT. 7 Abney (1987), Longobardi (1994), Ritter (1991), Szabolcsi (1994), Valois (1991), among others. 8 Mahajan (1990), Webelhuth (1992), among others. 9 Huang (1982), May (1985), Heim (1982), Beghelli and Stowell (1997), Szabolcsi (1997), among others. 10 Cinque (1997). 11 For example, Zanuttini (1991, 1997), Laka (1990). 12 den Besten and Webelhuth (1989, 1990). 13 See also Rizzi (1995), and Cinque (1998). 14 Koopman (1996, 1998b), Kayne (1997, 1998), Koopman and Szabolci (1998). 15 On subject-object asymmetries and heavy pied-piping (talks presented at the pied-piping conference in Jena (May 1979) and at Irvine (November 1997)). 16 Kayne (1997, 1998). 17 We now know that this is an extremely common feature of many Niger Congo languages. 18 Contra Dutch A and Dutch B (Maling and Zaenen 1978). That-t violations are possible in specific configurations, depending on ill-understood properties of the clause, including the number of words! (For additional discussion of this largely unsolved problem see Reuland (1983) and Bennis (1986)). 19 “On Pied-piping and procrastinate”, paper presented at the conference on pied-piping in Jena, (May 1997), “Subject extraction and pied-piping”, paper presented at Irvine (November 1997). 20 We no longer believe that this is correct for Chinese/Japanese type languages. It is clear that in Chinese for example subjects clearly occur very high in the functional domain (Chiu 1992).

INTRODUCTION 11

21 This is precisely what we argued for Dutch objects in the GLOW paper in Brussels in 1985 where we introduced VP internal subjects. While this idea is now standardly accepted, it did not exactly meet with approval at that time. 22 Koopman (1993b, 1998a) included in this volume. 23 Koopman (1993a) included in this volume. 24 “Topics in imperatives”, talk presented at Ibbs a/d Donau, and Koopman and Szabolcsi (1998). 25 Compare also Kayne (1996) UCLA lectures. 26 Sportiche (1995a, 1995b). 27 For this insight, see in particular Bok-Bennema and Groos (1984) and Bok-Bennema (1991). 28 It is not clear how to capture these data or contrasts given my present theoretical assumptions. 29 This idea is close to Kayne’s (1989) proposal that participle agreement in Romance languages is a reflection of the relation between a subject and the participle. 30 See Chapter 12, “Unifying predicate cleft constructions”, for an updated discussion. 31 As Carlos Otero points out, it is highly relevant that Economy is not mentioned once in Chomsky (1998). 32 Compare Koopman and Szabolcsi (1998) Verbal complexes. My work in progress includes work on wh-in-situ and partial wh-movement constructions (Class lectures, spring 1997), on subject object asymmetries and heavy pied-piping (talks presented at the pied-piping conference in Jena (May 1979) and at Irvine (November 1997) and imperatives (talk presented at the workshop on Transformations in Ibbs a/d Donau (June 1997)). 33 She later came across several African languages which showed this pattern much more transparently. 34 See Koopman (1987), not included in this volume.

References Abney, S. (1987) “The noun phrase in its sentential aspect”, unpublished PhD dissertation, MIT. Aoun, J., N.Hornstein and D.Sportiche (1981) “Aspects of wide scope quantification”, The Journal of Linguistic Research 1:69–95. Baker, M. (1988) Incorporation: A Theory of Grammatical Function Changing, University of Chicago Press, Chicago. Barbiers, S. (1995) “The syntax of interpretation”, University of Leiden, Holland Institute of Linguistics. Beghelli, F. and T.Stowell (1997) “Distributivity and negation: the syntax of each and every”, in A.Szabolcsi (ed.) Ways of Scope Taking, 71–109, Kluwer, Dordrecht. Belletti, A. and L.Rizzi (1988) “Psych verbs and theta-theory”, Natural Language and Linguistic Theory 6:291–352. Bennis, H. (1986) “Gaps and dummies”, unpublished PhD thesis, University of Tilburg. Besten, H.den and G.Webelhuth (1989) “Adjunction and remnant topicalization in the Germanic SOV languages”, paper presented at the 1987 GLOW conference in Venice. Besten, H. den and G.Webelhuth (1990) “Stranding”, in G.Grewendorf and W.Sternefeld (eds) Scrambling and Barriers, 77–92, John Benjamins, Amsterdam. Bok-Bennema, R. and A.Groos (1984) “Ergativiteit”, GLOT 7: 1–49.


Bok-Bennema, R. (1991) “Case and agreement in Inuit”, unpublished PhD thesis, Tilburg University. Chiu, B. (1992) “The syntax of functional words in Chinese”, unpublished PhD dissertation, UCLA. Chomsky, N. (1980) “On binding”, Linguistic Inquiry (11)1: 1–46. Chomsky, N. (1981) Lectures on Government and Binding, Foris Publications, Dordrecht. Chomsky, N. (1982) Some Concepts and Consequences of the Theory of Government and Binding, MIT Press, Cambridge, Mass. Chomsky, N. (1986a) Barriers, Cambridge, Mass: MIT Press. Chomsky, N. (1986b) Knowledge of Language: Its Nature, Origin, and Use, Praeger, New York. Chomsky, N. (1991) “Some notes on economy of derivation and representation”, in R. Freidin (ed.) Principles and Parameters in Comparative Syntax, MIT Press, Cambridge, Mass. Chomsky, N. (1993) “A minimalist program for linguistic theory”, in K.Hale and S.J. Keyser (eds) The View from Building 20: Essays in Linguistics in Honor of Sylvain Bromberger, MIT Press, 1–52, Cambridge, Mass. Chomsky, N. (1995) The Minimalist Program, MIT Press, Cambridge, Mass. Chomsky, N. (1998) “Minimalist inquiries: the framework”, unpublished manuscript, MIT. Cinque, G. (1998) Adverbs and Functional Projections: A Cross-linguistic Perspective, Oxford University Press, New York. Fukui N. and Speas, M. (1986) “Specifiers and projection”, MIT Working Papers 8. Hale, K. and S.J.Keyser (1993) “On argument structure and the lexical expression of syntactic Relations in K.Hale and S.J.Keyser (eds) The view from Building 20: Essays in Linguistics in Honor of Sylvain Bromberger, 53–109, MIT Press, Cambridge, Mass. Heim, I. (1982) “The semantics and syntax of definite and indefinite noun Phrases”, PhD dissertation, University of Amherst, Amherst (published (1988), Garland, New York). Hoekstra, T. (1984) Transitivity: Grammatical relations in Government-Binding Theory, Foris, Dordrecht. Huang, C.T. J. (1982) “Logical Relations in Chinese and the Theory of Grammar”, unpublished PhD dissertation, MIT. Kayne, R.S. (1984) Connectedness and Binary Branching, Foris Publications, Dordrecht. Kayne, R.S. (1989) “Facets of past participle agreement”, in P.Benincà (ed) Dialect Variation and the Theory of Grammar, Foris Publications, Dordrecht. Kayne, R.S. (1994) The Antisymmetry of Syntax, MIT Press, Cambridge, Mass. Kayne, R.S. (1997) “Overt vs. covert movement”, unpublished manuscript, New York University. Kayne, R.S. (1998) “Prepositional complementizers”, to appear in Probus. Koopman, H. (1982) “Les constructions relatives”, in C.Lefebvre, H.Magloire-Holly, and N.Piou (eds) Syntaxe de l’haitien, 167–203, Karoma publishers, Ann Arbor. Koopman, H. (1983a) “ECP Effects in Main Clauses”, Linguistic Inquiry 14(2): 346–350 (Chapter 4 in this volume). Koopman, H. (1983b) “Control from COMP and Comparative Syntax”, The Linguistic Review 2(4): 365–391 (Chapter 5 in this volume). Koopman, H. (1984) The Syntax of Verbs: From Verb Movement Rules in the Kru Languages to Universal Grammar, Foris Publications, Dordrecht. Koopman, H. (1987) “The genesis of Haitian: implications of a comparison of some features of the syntax of Haitian, French and West-African languages”, in P.Muysken et al. (ed.) Universals versus Substratum Influence in Creole Genesis, Benjamins, Amsterdam, 231–259.

INTRODUCTION 13

Koopman, H. (1991) “The verb particle construction and the syntax of PPs”, unpublished manuscript, UCLA. Koopman, H. (1992) “On the absence of case chains in Bambara”, in Natural Language and Linguistics Theory (10):555–594 (circulated in 1986, Chapter 7 in this volume). Koopman, H. (1993a) “The internal and external distribution of pronominal DPs”, to appear in Kyle Johnson and Ian Roberts (eds.) Beyond Principles and Parameters (Chapter 3 in this volume). Koopman, H. (1993b) “The structure of Dutch PPs”, unpublished manuscript, UCLA. Koopman, H. (1994) “Licensing heads”, in D.Lightfoot and N.Hornstein Verb-Movement, Oxford and New York: Oxford University Press, Cambridge and New York: Cambridge University Press, 261–296 (Chapter 9 in this volume). Koopman, H. (1995) “On Verbs that fail to undergo V-second”, Linguistic Inquiry, 26 (1): 137–163 (Chapter 10 in this volume). Koopman, H. (1996) “The Spec head configuration”, in F.Lee and E.Garret, Syntax at Sunset, 37–65 (Chapter 11 in this volume). Koopman, H. (1998a) “Prepositions, postpositions, circumpositions and particles: the structure of Dutch PPs” (Chapter 8 in this volume). Koopman, H. (1998b) “Unifying predicate cleft constructions”, in K. Moore (ed.) Proceedings of the twenty-third annual meeting of the Berkeley Linguistics Society, February 14–17, 1997: Special session on syntax and semantics in Africa (Chapter 12 in this volume). Koopman, H. and D.Sportiche, (1983) “Variables and the bijection principle”, The Linguistic Review 2 (3): 139–160 (Chapter 1 in this volume). Koopman, H. and D.Sportiche (1985) “Theta theory and extraction”, abstract of a talk given at the 1985 GLOW colloquium in Brussels, GLOW Newsletter, February 1985, Dordrecht: Foris Publications (included in Sportiche 1998). Koopman, H. and D.Sportiche (1986) “A note on long extraction in Vata and the ECP”, in Natural Language and Linguistic Theory, 4 (3): 357–374 (Chapter 6 in this volume). Koopman, H. and D.Sportiche (1989) “Pronouns, logical variables, and logophoricity in Abe”, Linguistic Inquiry 20:555–589. Koopman, H. and D.Sportiche (1991) “The position of subjects”, Lingua 85:211–258 (published in Sportiche 1998). Koopman, H. and A.Szabolcsi (1998) “Verbal complexes”, unpublished manuscript, UCLA. Kural, M. (1996) “Verb incorporation and elementary predicates”, unpublished PhD dissertation, UCLA. Kuroda, S.-Y. (1988) “Whether we agree or not: a comparative syntax of Japanese and English”, Lingvistica Investigationes 12: 1–47. Laka, I. (1990) “Negation in syntax: on the nature of functional categories and projections”, unpublished PhD dissertation, MIT. Larson, R. (1988) “On the double object construction”, Linguistic Inquiry 19: 335–392. Levin, B. (1993) English Verb Classes and Alternations: A Preliminary Investigation, University of Chicago Press, Chicago. Levin, B. and M.Rappaport Hovav (1995) “Unaccusativity: at the syntax-lexical interface”, Linguistic Monograph Series 26, MIT Press, Cambridge, Mass. Longobardi, G. (1994) “Reference and proper names: a theory of N-movement in syntax and logical form”, Linguistic Inquiry 25: 609–708. Mahajan, A. (1990) “The A/A-bar Distinction and Movement Theory”, unpublished PhD dissertation, MIT.


Maling, J. and A.Zaenen (1978) “The Non-universality of a surface filter”, Linguistic Inquiry 14: 421– 446. May, R. (1985) Logical Form: Its Structure and Derivation, MIT Press, Cambridge, Mass. Nkemnji, M. (1992) “Issues in the syntax of negation in Nweh”, unpublished MA thesis, UCLA. Nkemnji, M. (1995) “Heavy pied-piping in Nweh”, UCLA PhD dissertation series. Pesetsky, D. (1995) Zero Syntax: Experiencers and Cascades, MIT Press, Cambridge, Mass. Pollock, J-Y. (1989) “Verb movement, universal grammar and the structure of IP”, Linguistic Inquiry 20: 365–424. Pustejovsky, J. (1995) The Generative Lexicon, MIT Press, Cambridge, Mass. Reuland, E. (1983) “Movement versus merger: relations between inflection and verb”, NELS XIII. Ritter, E. (1991) “Two functional categories in noun phrases: evidence from modern Hebrew” unpublished manuscript, UQAM. Rizzi, L. (1982) Issues in Italian Syntax, Foris Publications, Dordrecht. Rizzi, L. (1990) Relativized Minimality, MIT Press, Cambridge, Mass. Rizzi, L. (1995) “On the fine structure of the left periphery”, unpublished manuscript, Université de Genève. Sportiche, D. (1988) “A theory of floating quantifiers and its corollaries for constituent structure”, Linguistic Inquiry 19: 425–449. Sportiche, D. (1990) “Movement, agreement, and case” (published in Sportiche 1998). Sportiche, D. (1995a) “Clitic constructions”, in L.Zaring and J.Rooryck (eds) Phrase Structure and the Lexicon, Kluwer Academic Publishers (published in Sportiche 1998), Dordrecht. Sportiche, D. (1995b) “Sketch of a reductionist approach to syntactic variations and dependencies”, in H.Campos and P.Kempchinsky (eds) Evolution and Revolution in Linguistic Theory: Essays in Honor of Carlos Otero, Georgetown University Press (published in Sportiche 1998), Washington, D.C. Sportiche, D. (1998) Partitions and Atoms of Clause Structure, Routledge, London. Szabolcsi, A. (1994) “The Noun Phrase”, in F.Kiefer and K.E.Kiss (eds) The Syntactic Structure of Hungarian, Syntax and Semantics 27, Academic Press, San Diego, Calif. Szabolcsi, A. (1997) Ways of Scope Taking, Kluwer, Dordrecht. Travis, L.deMena (1984) “Parameters and effects of word order variation”, unpublished PhD dissertation, MIT. Valois, D. (1991) “The internal syntax of DP”, unpublished Doctoral Dissertation, UCLA. Webelhuth, G. (1992) Principles and Parameters of Syntactic Saturation, Oxford University Press, New York. Zagona, K. (1982) “Government and proper government of verbal projections”, doctoral dissertation, University of Washington at Seattle. Zanuttini, R. (1991) “Syntactic properties of sentential negation: a comparative study of Romance languages”, unpublished doctoral dissertation, University of Pennsylvania. Zanuttini, R. (1997) Negation and Clausal Structure, Oxford Studies in Comparative Syntax, Oxford University Press, Oxford and New York. Zwart, C.J.-W. (1993) “Dutch syntax: a minimalist approach”, unpublished PhD dissertation, University of Groningen. Zwart, C.J.-W. (1997) Morphosyntax of Verb Movement. A Minimalist Approach to the Syntax of Dutch, Kluwer, Dordrecht.

Part I PRONOUNS AND WCO

1 VARIABLES AND THE BIJECTION PRINCIPLE* with Dominique Sportiche

1 Introduction The term variable has been introduced into recent linguistic theoretical frameworks by analogy with standard logic usage. In the following forms: (1)

a. b.

[Everyone]i [ ]x, xεS,

[αXi likes strong tea] [αP(x)]

The constituent α is an open sentence: its truth value cannot be evaluated since the reference of x or the domain over which x may range is not determined within [α]. In standard logic, x in (1)b is called a variable and is said to be bound by the universal quantifier. Analogically, the term xi in (1)a is called a variable, bound by the quantifier phrase everyone. This is not sufficient to define what a variable is, however. By extension from cases like (1)a, the implicit characterization of variables up to (and not including) Chomsky (1981) has been (2):1 (2) a is a variable iff a is the trace of movement from an A-position to an Ā;-position. By an A-position, we mean a position bearing a grammatical relation (subject of …, object of…). An Ā;-position is a position which is not an A-position (COMP, adjunct…). Beyond terminology, the introduction of definition (2) in linguistic theory embodies an implicit empirical claim: it characterizes a class of elements which, one would expect, should exhibit some distinctive property(-ies). This has indeed been argued to be the case (e.g. in Chomsky (1976)) with the so-called weak crossover phenomena (henceforth WCO) illustrated in (3) below:2 (3)

a. b.

Who does his mother love His mother loves everyone

VARIABLES AND THE BIJECTION PRINCIPLE 17

c. d.

His mother loves JOHN Sa mère aime qui

(focal stress on John) (non echo wh-in situ)

In these sentences, the pronoun his or sa cannot be bound by who (or by everyone, JOHN, qui respectively). At LF, the respective representations of the sentences in (3) are: (4)

a. b. c. d.

whoi [hisi mother loves xi] everyonei [hisi mother loves xi] Johni [hisi mother loves Xi] quii [sai mère aime xi]

(4)a is identical to its S-structure (3)a. (4)b is derived from (3)b by QR (the rule of Quantifier Raising compare May (1978)), i.e. by adjunction of the QP to S3, (4)c from (3)c by the rule of Focus Interpretation (cf. Chomsky (1976) and infra section (5)), and (4)d from (3)d by wh-raising, i.e. wh-movement in LF (compare Aoun Hornstein and Sportiche (1981), Huang (1980)). By definition (2), which we assume holds at LF, the xi in (4) are all variables: they are traces of movement from an A-position to an Ā;-position. The consideration of such examples as (3) led Chomsky (1976) to express the WCO property as follows: (5)

The Leftness Condition (LC) A variable cannot be an antecedent for a pronoun to its left

It is quite clear how the Leftness Condition rules out the interpretations (4) for the sentences in (3). In Chomsky (1981), a modification of definition (2) is proposed, in accordance with the general idea that the nature of empty categories should be determined functionally, by looking at their closest antecedent, if any. Thus, variables are defined as in (6), a definition that we also assume to be operative at LF (at least): (6)

α is a variable iff:

a is an empty category and α is in an A-position and α is locally Ā;-bound4

As can be seen, the account given for the paradigm in (3) and (4) holds exactly as before, since (6) characterizes the same elements (2) does.5 However, we will argue that, given (6), a different account of this paradigm can be offered which carries over to other cases for which the LC and other proposals have nothing to say. In section 2, we will show that (6) is too narrow and must be extended so as to include certain occurrences of pronouns as well. In section 3, we will examine the consequences of this extension of the concept of variables. A new explanation of the WCO facts will be constructed on the basis of what we call the Bijection Principle. In section 4, we will


compare the Bijection Principle with other proposals and show why it seems superior. We will also discuss the problem of whether the Bijection Principle is a condition on rule application and/or on representations at some level, and if so, at which level. In section 5, we will briefly examine the case of focused phrases and the various alternative treatments of the WCO phenomenon in these constructions. In the remainder of this chapters, we basically adopt the theoretical framework laid out in Chomsky (1981, 1982) which presents in particular the following picture:

where the ECP as formulated as:6 (8)

ECP: An empty category must be governed by a proper governor.7

Note finally that we will restrict our attention to NPs and NP positions throughout, whether A or A, thus excluding PPs, and so on from our discussion. 2 Resumptive pronouns The existence of languages using a resumptive pronoun strategy poses a superficial challenge to definition (6), which requires variables to be empty categories. What happens when a locally Ā;-bound empty category is filled by a resumptive pronoun, especially if this type of pronoun exhibits the WCO characteristics of variables as is the case for example in Haitian (compare Koopman (forthcoming))? It would seem that definition (6) should be modified accordingly. This argument only holds however, if it can be shown that these resumptive pronouns are not empty categories at LF: otherwise, it


could simply be assumed that the phonetic realization of these resumptive pronouns takes place in PF. We will now show that Vata, a West African language of the Kru family spoken in the Ivory Coast,8 displays a configuration of data forcing the conclusion of the above argument. 2.1 In Vata, wh-movement from subject position of a tensed clause requires the insertion of a resumptive pronoun in the vacated position: (9)

* mlì l àl who *(he) left wh “who left” gū;gū; yī; what you think “what did you think happened”

a.

b.

nā; that

*(ì) *(it)

lì fell

l wh

This resumptive pronoun differs from regular third person subject pronouns in that it bears a low tone ( , ì,…) instead of a mid-high tone ( , ,...).9 It is shown in Koopman (1980) that the obligatory character of this insertion is a direct consequence of the constraint on LF representations imposed by the ECP: the structure of Vata (superficially resembling Italian in this instance) is such that the subject position of a tensed clause is never properly governed by material in COMP position (contrary to, say, English or French). Thus, in cases like (9) above, insertion of a resumptive pronoun in subject position is obligatory, and must take place no later than S-structure in order to avoid an ECP violation. On the other hand, such pronouns exhibit what we assume to be a characteristic property of variables at LF, the WCO property, which is shown by (10): (10)

a.

*

b.

*

n gù gù nā; àl i ih i whoi hisi mother think that hei “who did his mother think left” yr n àl i i whoi you tell hisi mother “who did you tell his mother left”

mlì left

l wh

nā; that

hei

i

mlì left

l wh

Neither in (10)a nor in (10)b can the low tone pronoun be interpreted as coreferential with the underlined pronoun. (Note that there is no general prohibition forbidding coindexing of a low tone pronoun with a mid-high tone pronoun, compare 2.2). So, this paradigm indicates that: The low tone pronoun behaves as a variable at LF.


The low tone pronoun position cannot be empty at LF (it must be filled by a phonologically realized pronoun). In order to preserve the coextensiveness of the set of elements obeying the WCO property and the set of variables, we take the obvious step of allowing α in (6) to be a pronoun as well. Let us assume (11) for the time being: α is a variable iff

(11)

a is empty or a pronoun and α is locally Ā;-bound and α is in an A-position 2.2

This configuration of data also shows that it is not plausible to define an NP as a variable only by looking at its content, as would be the case for example if (12) were adopted:10 α is a variable at LF iff α is empty and α is in an A-position and α is case marked

(12)

The argument above shows that (12) should be supplemented by a statement of the form (13) in order to accommodate the variable-like character of the low tone pronoun: (13)

Certain pronouns with a feature [+F] are variables

The choice of the feature [+F] must be such that it properly distinguishes the low tone pronouns which behave as variables from the mid-high pronouns which do not. (14)

a.

mlì left

à1 who “who left”

l wh

mlì left

b. “he left”

The nature of the feature [+F] is not clear however. It could neither be Case (since pronouns are all Case marked) nor the feature realized as a low tone instead of a mid-high tone, which is the feature [+wh]. The reason is that occurrences of low tone pronouns are not restricted to sites of wh-phrases (and in fact, this is what justifies our calling them pronouns): a low tone pronoun may also occur in a position which is, informally speaking, neither too close, nor too far from the site of a wh-element provided that it is coindexed with a wh-trace, or a low tone pronoun [+wh].11 (15)

a.

àl

i

i

gū;gū;

nā;

j

/*

i/

i

ní

yà

l


whoi think i “who thinks he is sick”

that

j

/*

i

/

NEG

i

healthy

wh

Furthermore, these last occurrences of the low tone pronoun do not obey the WCO property as shown by the well-formedness of (16): (16)

yr n àl i i i whoi hei tell hisi mother “who told his mother that he left”

nā; that

mlì left

i

hei

l wh

In (16), the underlined low tone pronoun may be coreferent with , which is to its left and does not c-command it. Thus, because there is no independently motivated difference in the internal constituency of these occurrences of the low tone pronoun, (13) itself would have to be supplemented by some property distinguishing them. It seems that the only one is that those behaving as variables are locally Ā;-bound while those that do not are not, i.e. only the first occurrence of in (16) but not the second. So (13) should be replaced by(13’): (13’)

A pronoun is a variable iff it is locally Ā;-bound and is marked [+F]

It appears thus that reference to local Ā;-binding is difficult to avoid. But if such is the case for pronouns, it extends naturally to empty categories as well, making reference to internal structure redundant (although it may still be true that Case-marked empty categories are variables). In the case of pronouns in Vata, reference to the feature [+F] is clearly not redundant, but we shall see directly that dropping reference to it will allow a simple and elegant treatment of the WCO property, while allowing us to keep the maximally simple definition for variables that we have adopted in (11). 3 The Bijection Principle 3.1 Let us now consider more closely a case of WCO in Vata, exemplified by the following form: (17)

*

àl i whoi

w they

yrá tell

i

hisi

n mother

nā; that

i

hei

mlì left

l wh

In (17), the low pronoun is locally Ā;-bound by the wh-phrase in COMP position and is thus interpreted as a variable at LF by virtue of (11). It is clear how the LC rules out the interpretation of (17) indicated, since the pronoun is to the left of the variable and coindexed with it. Definition (11) however, also implies that the pronoun is interpreted


as a variable since it too is locally Ā;-bound by the wh-phrase in COMP. We are now in a situation where an Ā;-binder binds two positions which are both interpreted as variables. This immediately suggests an alternative approach to WCO violations. Before spelling it out, let us first examine some necessary properties of variables. 3.2 We have seen that prior to the introduction of such definitions as (6) and (11), variables were understood as A-positions created by movement to an Ā;-position. It was in fact also implicitly assumed that the NP moved to an Ā;-position was the local Ā;-binder of the trace in the A-position in question.12 Under this conception of variables, the following generalization is true: (18)

A variable is locally bound by one and only one Ā;-position.

Indeed, if α is a variable, it is a trace resulting from movement to an Ā;-position and is therefore Ā;-bound, and in particular is bound. Since there exists a binder for α, there exists a local binder, call it β. If β is not in an Ā;-position, the resulting configuration is ruled out by condition C of the Binding theory (Chomsky 1981) which requires variables to be locally A-free (i.e. non Ā;-bound). Thus β is in an Ā;-position.13 Hence the existence of the local Ā;-binder. The uniqueness of the local Ā;-binder follows from the following considerations. Suppose Ā;1 and Ā;2 are two local binders for α. Then A1 and Ā;2 must c-command each other. But quite generally, this situation does not arise: if an Ā;-position A1 c-commands another position Ā;2, the converse is not true. Thus any element has at most one local Ā;binder.14 Hence (18). Under definition (2) with the extra assumption spelled out right above (18), the “converse” of (18) is also true, namely (19): (19)

An Ā;-position locally binds one and only one variable.

Suppose that NP1 in an Ā;-position locally binds two variables A1 and A2. By assumption, A1 and A2 must be the traces of NP1. But no phrase can locally bind two of its traces because of constraints on movement rules imposed by the Binding conditions and Θtheory (compare Chomsky (1981:203) for an argument). Therefore this situation is not possible. Furthermore, NPs appear in an Ā;-position only if they have been moved there from an A-position. Note now that under definitions (6) or (11) the generalization (18) remains essentially true,15 but not (19), since (19) allows the introduction of variables in a representation by simple coindexing of a category with an Ā;-position. (19) will be falsified just in case two NPs, NP1 and NP2, are coindexed with each other, and locally bound to a same Ā;position.


3.3 Let us now return to structures like (17): (17) illustrates the fact that precisely when (19) is violated, under definition (11) for variables, ungrammatical structures are generated. So suppose that we take (19) as an axiom: let us assume that (19) is a wellformedness principle holding at LF.Putting (18) and (19) together, we can state this principle as in (20):16 (20)

The Bijection Principle There is a bijective correspondence between variables and positions.

Or, to put it differently, every variable is locally bound by one and only one Ā;-position and every Ā;-position locally binds one and only one A-position.17 It is clear how the Bijection Principle (BP) rules out the ungrammatical Vata form (17), and also how it handles WCO violations in general. As an illustration, consider (3)a repeated below: (3)

a.

*

whoi

does

hisi

mother

love

[e]i

At LF, both his and ei are locally Ā;-bound by who, which is ruled out by the Bijection Principle. Of course, this account works provided that we extend definition (11) to English and other languages.18 But this is clearly the better move: one wishes to restrict particular language definitions with respect to such fundamental notions as variables as much as possible. Whatever the correct definition of variable, it is reasonable to suppose that it is universal.19 3.3.1 Let us consider (11) again. Although the modification from (6) to (11) is under-standable, we may realize under closer examination that it really makes no sense. For, if a pronoun is locally Ā;-bound, it is no longer a pronoun since by definition (compare Chomsky (1981: 330)) pronominals are either free or locally A-bound to an NP with an independent Θrole. In fact, (11) means that variables need not be empty categories, they may also assume the shape of pronouns. This suggests that we replace (11) by the simpler, more general and more adequate (21): (21)

α is a variable if a

is in an A-position is locally Ā;-bound

Clearly, some independently needed principle will exclude entities with intrinsic reference, such as proper names and so on, from being locally Ā;-bound with the effect that variables may be only pronominals and/or empty categories (lexical anaphors are excluded by the Binding Theory, which requires them to be locally Ā;-bound).


Summing up, we assume that the BP (20) and the definition of variables (21) hold universally. Furthermore, keeping the simplest possible assumptions, we will suppose that (21) holds at any syntactic level. 3.3.2 It is important to note that the BP only makes sense if one adopts (21) as the definition of variables: these two hypotheses reinforce each other. It is worth pointing out however that (21) is independently motivated by the fact that it permits deriving the Strong Crossover effects (SCO), i.e. part of principle C of the Binding Theory, from Principles A and B. The SCO configuration is illustrated below:

Note that, by (21), it is the pronoun he which is interpreted as a variable, and no longer the trace ei of the wh-phrase who. How then is ei interpreted? In each case, it will be locally A-bound to he, ignoring traces of successive cyclic movement in the intermediate COMPs which appear to play no particular role. He has an independant Θ-role, so ei is an empty pronominal, i.e. a PRO. But principles A and B of the Binding Theory (compare Chomsky (1981: Chapter 3)) require PRO to be ungoverned and ei in (22b, 22c and 22d) is governed: hence they are ruled out by these principles. The SCO violations are thus explained by (21) and principles A and B of the Binding Theory. 3.3.3 We have provided above motivations for (21) independent of the Bijection Principle and WCO. We now present evidence for the Bijection Principle independent of WCO. The relevant type of ungrammatical structures are those involving parasitic gaps. Extensive discussion of these structures can be found in Chomsky (1982), Engdahl (1980) and Taraldsen (1981). We will limit ourselves here to one particular type of parasitic gap structures obeying the schema (23)a illustrated in (23)b:20


Clearly, structures like (23)b can be generated: base generation of an empty category in one of the two gap positions and movement to an Ā;-position from the other. Furthermore, as the reader may verify, nothing rules out these structures, since they are equivalent to the following two grammatical structures: (24)

a. b.

Whoi Whoi

did did

you give you give

a picture of ei to John a picture of John to ei

Whatever mechanisms permit these structures will carry over to (23)b and the like.21 Note in particular that the proposal presented in Chomsky (1981: 203) to rule out (23) b in terms of Θ-theory, in fact rules out a particular derivation of these structures (namely by movement from one e-position into the other and from there to the Ā;-position), but not (23)b itself. Clearly, structures like (23) are deviant, and it is obvious that they violate the Bijection Principle, since an Ā;-position locally binds two A-positions. In conclusion, constructions like (23) justify, quite independently of WCO, the need for some principle like the BP. In other words, given the definition of variables (21) and the independent justification for the BP due to the parasitic gap constructions, we get a free explanation for the traditional cases of WCO. Summing up, the BP treats uniformly: (25)

a. b. c.

Whoi did you give a picture of himi to ei Whoi did you give a picture of ei to himi Whoi did you give a picture of ei to ei 4 The Bijection Principle and Alternatives

There are a number of proposals that have been made in recent years to deal with the WCO violations, for example, the Leftness Condition (LC) of Chomsky (1976), the Revised Leftness Condition (RLC) (Higginbotham (1980a)), the Accessibility Condition (AC) (Higginbotham (1980b)), the C-command condition (CCC) (Reinhart (1976), (1979), (1981)) advocated for in Haïk (1982). Some of these are wellformedness conditions, e.g. the CCC, while others are conditions on rule application, e.g. the RLC,


the AC. We will limit our comparison here to the AC, the CCC and the BP, and refer the reader to Higginbotham (1980b) for criticisms of the LC and the RLC. Each approach specifies two parameters: first the particular kind of structural configurations leading to WCO violations, second the mode by which these ungrammatical structures are ruled out. 4.1 Let us begin with the CCC. The CCC states in essential that: (26) A non-definite NP (trace, QP,…) can be coindexed with a pronoun if the pronoun is c-commanded22 by the non-definite NP at S-structure. The BP, taking into account the effects of QR, cannot be an S-structure condition. Furthermore, contrary to the CCC, the BP does not include any necessary condition for pronominal binding by a quantifier. We in fact assume that the independent principle (27) does this work: (27) The Scope Condition A pronoun may be coindexed with a variable bound by a (quasi-)quantifier (i.e. wh-phrases, quantifiers subject to QR),23 only if it is in the scope24 of the (quasi-) quantifier at LF. In fact, our approach is very similar to the CCC. Consider the following configuration at LF (linear order irrelevant), where P is a pronoun, there are no phrases indexed other than those indicated, and e is the trace of whmovement, QR or wh-raising. (28)

Qi…Pi…ei

Suppose first Qi c-commands Pi. There are three possibilities, discussed in (29)a-(29)c. (29) a. Pi c-commands ei. Under the BP approach, this will be excluded as a SCO violation—cf. 3.3.2 —since ei is locally bound by Pi. Under the CCC approach, we must further distinguish between two cases: if ei does not c-command Pi, the structure is excluded by the CCC since Qi will not c-command Pi at S-structure. If ei does command Pi, the structure respects the CCC. However, if ei and Pi c-command each other, they govern each other. They therefore belong to the same Governing Category. This configuration is then ruled out by the Binding Theory, since the pronoun Pi is not free in its Governing Category.


b. Neither Pi nor ei c-command the other. The BP is violated since, at LF, Qi locally binds both ei and Pi. The CCC is also violated since at S-structure, Qi occupies the position ei which does not c-command Pi. c. ei c-commands Pi and not conversely. This is allowed by both the BP and the CCC as the reader may verify. This leaves us to discuss a last case: (29) d. Qi does not c-command Pi. This is excluded by the CCC under the CCC approach, and by (27) under the BP approach. The BP and the CCC are therefore very close in the empirical predictions they make, but they make them in very different ways. Consider examples of (29)a, b, c and d above: (30)

a. b. c. d.

* * *

Whoi did hei see [ei] Whoi does hisi mother like [e]i Whoi [e]i saw hisi mother Hisi mother wonders whoi [ei] left

Under the CCC theory, the ungrammatical examples in (30) are excluded by the CCC. Under the BP approach, (30)a is a Binding Theory violation, (30)b a BP violation, and (30) d, a violation of (27). We believe the CCC makes the wrong claim, for (30)b and (30)d appear to us to be violations of a very different nature: the violation of the Scope Condition (30)d appears to us to be much stronger than WCO violation (30)b. Similarly, violations like (30)a appear much stronger to us than violations like (30)b, a judgment that is reflected under the BP approach by the fact that the first one is a Binding Theory violation, while the second one is a BP violation. No such distinction is made by the CCC, unless of course we adjoin to it (redundantly in this case) a version of principle C of the Binding Theory, which, as we have shown in 3.3.2 follows from the BP approach and principles A and B of the Binding Theory. There are two other differences between the two approaches. We see that the CCC states a necessary and sufficient condition for a pronoun to be understood as coreferent with a non-definite NP. Clearly the BP agrees that it is a sufficient condition, but it predicts that it is not a necessary condition. Although we have been unable to find sentences illustrating this point, it can be abstractly illustrated. Consider the following structure:


In (31)a, neither the pronoun, nor the QP c-command the other and the pronoun is in an A-position. (31)b is derived by QR from (31)a. The CCC rules such structures out since the pronoun is not c-commanded by the QPi at S-structure, i.e. in (31)a. For the BP the structure is well formed: looking at (31)b—LF— QPi locally Ā;-binds Pi and the Ā;-position locally Ā;-binds ei. Finally, and most importantly, the CCC has nothing to say about parasitic gap constructions. As we argued, a simple principle excluding them would automatically extend to the WCO cases, as does the BP. 4.2 Let us now turn to the AC. Higginbotham (1980b) actually proposes two accessibility conditions: The Strong Accessibility Condition (SAC) and a weaker variant, the Accessibility Condition (AC), to which we return in the next section. In fact, Higginbotham (1980b) notes that the following modification of the CCC: (Note that if A = C below, (32) is the CCC.) (32) If A is a quantifier and B a pronoun, A can bind B only if A is co-indexed with some C that c-commands B at S-structure. “…quite apart from the assumption of QR, appears to be equivalent to assuming the SAC with QR.” Consequently, the main criticism that we have offered of the CCC in favor of the BP holds against the SAC as well, i.e. the fact that the treatment of parasitic gap structures like (23), when properly formulated, carries over to the WCO cases. There is an important difference however between the SAC (or the Revised LC) and the BP. Higginbotham (1980a, b) argues that the SAC (or the RLC) is best viewed as a condition on rule application, namely a rule reindexing a pronoun with the index of some other NP, rather than a wellformedness condition applying at a particular level. So far, we have assumed without particular argument that the BP is a wellformedness principle holding at the level of LF. Although there is no intrinsic bar to viewing the BP as constraining rule application, such a theory would appear less desirable on several counts. First, if it is a constraint on the applicability of some rules, we must assume the existence of the relevant rules, in particular, a rule of reindexing or identification of indices. But this is a needless complication of the optimal indexing algorithm, namely index freely at S-structure, which


is all we need if the BP is an LF filter. Moreover, conditions on rule applicability (such as, perhaps, Subjacency) constrain the analyzability of terms in a string with respect to a given structural description, whereas the BP viewed as constraining rule application would rule out the output representation of some rules at any stage of a derivation and would thus in effect be a new type of condition which is made, in principle, available to the language learner, thus leading to a less restrictive theory of grammar. Consequently, we take the minimal assumption to be that the BP is an LF filter. We will now show that the arguments presented in the literature for taking whatever condition involved in WCO as constraining derivations, i.e. rules rather than representations, do not hold with respect to the BP, an LF filter. Such arguments are based on sentences like the following (compare Higginbotham (1980, 1980b)): (33)

a. b.

*

which pictures of which mani did hisi mother see Someone i will play every piece you want himi to

Instead of reviewing the arguments, we will simply show that such sentences pose no problem for the BP. Consider first (33)a. If which man has narrower scope than which pictures (which is possible and comes out more clearly with men instead of man) it will not c-command his, so binding is impossible by principle (27). Consider next the inversely linked case, i.e. the reading of (33)a corresponding to LF (34): (34)

[which man]i [ [swhich pictures of ei]j [s hisi mother saw ej]]

Such structures violate the BP: which man locally Ā;-binds ei which is in an A-position, namely, argument of pictures, and also locally Ā;-binds his. So structures like (33)a, or similar structures where which man is further embedded (e.g. which daughter of which uncle of which man) pose no problem for the BP taken as a condition on LF representations. Consider next (33)b. In (33)b some can only bind him if some has wider scope than every. How do we account for this fact? Suppose every has wider scope than some. Then, the only possible LF representation of (33)a corresponding to this reading is one in which QR moves the object as in (35). (35)

[every piece you want himj to]i [ [ssomeone]j [s ej will play ei]]

Here, some cannot bind him, since, by assumption, him is not in the scope of some. Suppose every has narrower scope than some. If both quantified expressions are adjoined to S as in (36): (36)

[sSomeone]i [sEvery piece you want himi to]j [ei will play ej]]

the resulting structure will be ruled out by the BP (as noted in Haïk (1982)). Note however the implicit assumption for which no independent argument, neither empirical


nor conceptual has been offered, namely that QR must move all quantifiers and adjoin them to S. If either of these assumptions is dropped, no problems for the BP arise. Suppose for example, that QR may adjoin quantifiers to VP. Then we may derive: (37)

Someonei [s ei [VP [NP every piece you want himi to]j ]]

VP[played ej]

In (37) binding of him by someone is permitted without BP violation, since it is not local binding. Pending arguments to the effect that, say, QR adjunction to VP is forbidden and so on, we see that the BP can be thought of as an LF filter. 4.3 Some asymmetries Consider the following examples:

These examples have the structure indicated in (38)d, where, moreover, ei is inside an NP which c-commands Pi. As is clear, the BP excludes such structures. For some languages, e.g. Mandarin Chinese (Higginbotham 1980a), Vata, our respective dialects of French and Dutch, some dialects of English (Lasnik 1976)), there is nothing more to say. For some other, for example some dialects of English (Reinhart (1976), Higginbotham (1980a)), sentences like (38), although marginal, are not felt as unacceptable as parallel structures such as (3)a, where the linear order of ei and Pi is reversed. Higginbotham (1980b) captures this difference by weakening the SAC to the Accessibility Condition (AC). Essentially, the effect of this weakening is to permit coindexing of a pronoun B with a quantified A only if A is coindexed with some C c-commanding B or being in an NP ccommanding B. Although some stipulation or other may be added to the BP account, for example, one extending the c-command domain of an empty category to that of an NP containing it,25 the logic of our account would seem to imply that these sentences should be considered ungrammatical in the unmarked case, even if somewhat acceptable to some speakers. First, because they are ruled out by the BP, but more importantly because they display an


empty category/pronoun asymmetry with respect to Ā;-binding, which, given our arguments on how to define variables, would be unique to these sentences. 5 Focus In this last section, we consider focus constructions. As (3)c shows, focus constructions exhibit WCO properties, which lead Chomsky (1976) to postulate a LF movement rule of Focus Interpretation (FI), converting (3)c into (4)c. Clearly, forms like (4)c are excluded by the BP. The problem is, however, that focused elements should, on this account, behave as empty categories at LF. However, they fail to do so on two counts: First, with respect to the ECP (39)a is translated into (39)b by FI for example: (39)

a. b.

Mary claims that SARAH should stay SARAHi [Mary claims that ei should stay

(39) should be ill formed, contrary to fact, since ei is not properly governed. Note that, if we assume, as is quite natural, that the scope of Focus, i.e. the minimal constituant not containing it at LF (or perhaps LF’) expresses the presupposition of the sentence, FI should be able to freely adjoin its target phrase to any S, i.e. the embedded S or the matrix S in (39). (So the two possible derived structures would be semantically similarly to It is Sarah that Mary claims should stay and Mary claims it is Sarah who would stay.) If this assumption about presupposition is dropped, one could claim that the Focus is always adjoined to the first S containing it, thus avoiding the ECP problem, as suggested in Sportiche (1980). However, even this assumption would not overcome the other problem facing FI, to which we turn now. It has been argued that a specific (or referential) NP cannot contain a variable whose local binder is external to this NP at LF (compare The Specificity Condition of Fiengo and Higginbotham (1971), The Name Constraint of Guéron (1981) and so on). Focused NPs however are not subject to this constraint as (40) shows: (40)

a.

She appreciates Vermeer’s last painting of DELFT.

Applying FI to the focus NP yields violations of the mentioned constraints, unless one assumes that FI adjoins the focus to NP so that the specification of the trace is NP internal. Sticking to the most natural assumption, namely that the scope of Focus at LF determines the presupposition of the sentence, we can make two alternative assumptions to overcome the problem. The first one is to assume that, although the Focus is somehow related to some Ā;-position, this is not done through movement. This is for example what is suggested for the scope of quantifiers in van Riemsdijk and Williams (1981), who propose that the scope of a phrase is marked by adjoining the index of this phrase to the relevant constituent. Such a move is consistent with the BP, even if this mode of expressing scope


relations is true across the board, i,e. for quantifiers, wh-in situ, Focus NP, as is suggested in van Riemsdijk and Williams (1981). The second possibility is to assume, following a suggestion of Brody (1981) that there is a rule of FI freely adjoining a Focus phrase to any S, but that this rule maps LF onto LF’, rather than S-structure onto LF. At LF, where the ECP and the Specificity Condition are relevant, no empty category is present. This would of course force us to assume that the BP is an LF’ filter (rather than or) as well as an LF filter. If this were the correct move, we might extend the BP to a well-formedness condition constraining all levels of representations (D-structure, S-structure, LF and LF’). We will leave the problem as it stands, simply noting that the BP is consistent with either of these possibilities. 6 Summary Let us conclude by briefly summing up our proposals. We have argued that WCO phenomena as well as some ill-formed parasitic gap constructions can be handled by the Bijection Principle, which requires a one-to-one correspondence between Ā;-binders and variables. We have argued that variables must be defined as locally Ā;-bound categories (with no extra requirement), a definition that allows the reduction of SCO phenomena to principles A and B of the Binding Theory. We have furthermore argued that the BP can be viewed as a filter on LF representations (and perhaps D-structure, S-structure and LF’ as well). Notes * We would like to thank Y.Aoun, M.Brody, N.Chomsky, R.Freidin, I.Haïk, N. Hornstein and J.R.Vergnaud for their various contributions to our understanding of the matters discussed in this article. The content of this article has been presented at the Göttingen GLOW Colloquium (April 1981), which was made possible in part by the Deutsche Forschungsgemeinschaft. This research is supported in part by the Social Sciences and Humanities Research Council of Canada, grant number 410–81–0503. 1 Note that we restrict ourselves throughout to the syntactic sense of variable. We exclude from consideration cases of pronouns bound to variables and therefore interpreted semantically as variables, as he in Everyonei said hei left. 2 The name “Weak Crossover” was chosen in opposition to “Strong Crossover” cases exemplified in (a) below:

a.

*

Whoi did hei see ei

the unacceptability of the WCO being “weaker” than that of the SCO.


3 We need not assume QR is obligatory: the reason is that if the QP does not move, his will not be able to be interpreted as bound by the QP since it will not be in its scope at LF (compare section 4, principle (27)). 4

x x x

c-commands y iff the first maximal projection which dominates x also dominates y (compare Aoun and Sportiche (1981)). (locally) binds y iff x c-commands y and x and y are coindexed (and if z has the same properties as x, z c-commands x) (locally) Ā;-binds y iff x (locally) binds y and x is an Ā;-position (resp. for Ā;-binding).

5 Chomsky (1981) also stipulates that (6) might be extended so as to include cases of resumptive pronouns, although no argument is presented. We return to this point in section 2. 6 Compare Kayne (1981), Rizzi (1980) and Koopman (1980, 1982) for arguments that the ECP applies at LF. 7 x governs y iff the first maximal projection dominating y is xmax. Proper governors are X° in the system (compare Aoun and Sportiche (1981) and Aoun, Hornstein and Sportiche (1981) for details). 8 For our purposes, it suffices to assume that Vata is identical to English in the relevant respects except as indicated in the text. For details on the structure of Vata, compare Koopman (1980) and Koopman and Sportiche (1981). 9 Vata has four lexical tones T high, T midhigh, T mid and T low. On the selection of pronouns in relation to nouns, compare Kaye (1981). 10 Compare Borer (1981) for a similar argument based on Hebrew data. 11 For a detailed discussion, compare Koopman (1980) and Koopman and Sportiche (1981). 12 Note that we exclude altogether from our discussion intermediate traces of successive cyclic movement and constructions with clitics which, we may assume, only involve superscripting in the sense of Chomsky (1981), Chapter 4 and no coindexing. Compare Sportiche (forthcoming) for detailed discussion. 13 In fact, by assumption, β can be nothing else than the phrase which has moved (ignoring successive cyclic movement). 14 Note that this was true due to the definition of c-command that used to be adopted, namely a c-commands β iff the first branching node dominating a dominates β, and a does not contain β. However, given the definition of c-command we adopt (Note 4) (as well as Reinhart’s, compare Note 22), the situation may arise in which two A-positions c-command each other. A detailed discussion would take us too far afield. Let us just indicate the relevant type of cases. The only type that would not be ruled out independently is of the following nature (with two ultimately A-bound positions, neither of which c-commands the other):

a. S-structure. I bought a picture of everyonei for everyone b. LF: Everyonei [Everyonej [I bought a picture of ei for ej] ] Nothing in the present theory so far prevents the interpretation corresponding where i corresponds to j (i=j), namely


x, I bought a picture of x for x.

As can be checked, the BP (20) will rule this out since ei and ei will be both locally Ā;-bound by everyonei (and in fact may be construed to do so in two different ways) compare Sportiche (forthcoming) for more details. 15 Note again that this holds under the c-command definition given in Note 14. Under the one we adopt, this is stricto sensu false. The only generalization remaining true after the changes of definition is that a variable is locally bound by at least one Ā;-position, which is true by definition. The others are false as the examples in Note 14 show. What this means is that the BP (20) must be stipulated in toto, which, in fact, provides further independent evidence for it. 16 An idea similar in spirit if not in extension was expressed in an unpublished version of Aoun, Hornstein and Sportiche (op. cit.). A modification of (2) quite different from (and in fact incompatible with) (6) or (11) was proposed (the idea was to exclude successive cyclic LF movement by assuming that each trace created in LF was a variable and by applying the Binding Principle C to the resulting configuration). 17 Recall that we exclude clitics and intermediate traces of successive cyclic movement into COMP from consideration. Note that the BP as stated is too strong, given our definition of Ā;-position, if positions such as TOPIC and the like are considered A-positions, viz. Johni, hisi mother saw himi on TV and more generally dislocations, some resumptive pronoun constructions. Perhaps the defining characteristic of these A-positions is that the NP they contain inherits a Θ-role, contrary to preposed quantifiers, wh-phrases in COMP…. For concreteness, let us assume that NPs in these positions bear no index at LF, and therefore do not participate in any of what we say. They might get coindexed with some other position through a Predication rule mapping LF to LF’, as Chomsky (1982) suggests. 18 We implicitly assume here that the specifier position of NPs in English, French, Dutch and Chinese (compare Infra) is an A-position. This seems plausible given that such positions appear to be able to bear independent Θ-roles, i.e. Θ-roles not inherited from another position by coindexing. This assumption becomes necessary if one adopts the version of the Θ-criterion given in Chomsky (1981) in conjunction with his Projection Principle. 19 As it stands, the BP suffers from some of the defects of the LC. The following examples seem to indicate that it might be too strong:

a. Any man that his friends help will succeed. b. PRO sleeping late pleases everyone. These should be on a par with WCO violations but they seem more acceptable. See Chomsky (1982) for a proposal concerning the relative clause case (a). In the (b) case, we adopt the Aoun and Sportiche (op. cit.) suggestion that coindexing of the PRO with the QP at LF is not necessary to get the relevant interpretation but rather that it is constructed derivatively from the predication of the whole sentential subject with the QP as in: c. Bad marks bother every serious student with the reading: every serious student is bothered by his getting bad marks.


Finally, let us mention that across-the-board contexts yield superficial violations of the BP as in: d. I wonder whoi you saw ei and you talked to ei At worst, we would have to stipulate that coordinate structures are special. 20 It turns out that these are the only structures containing parasitic gaps which are not independently excluded. 21 Some questions might arise with respect to P-stranding, because of the double stranding.

The following examples escape this problem: a. Whoi did you persuade friends of ei to see ei b. Whoi did you persuade friends of ei to see John c. Whoi did you persuade friends of John to see ei 22 The notion of c-command used by Reinhart is different from the one adopted here (compare Note 4). For her, x c-commands y iff the first branching node (or the node immediately dominating it if they are of the same category) which dominates x dominates y. Compare Aoun and Sportiche (1981) for a criticism of this notion. 23 We wish to exclude from consideration quantifier-like exessions like any which are not subject to QR (compare Aoun, Hornstein and Sportiche (1981) and Hornstein (1981)). 24 We define scope of a quantifier (at LF) as the maximal constituent not containing this quantifier. 25 Reinhart (1976) describes these two dialects of English. She points out that there is a general tendancy for speakers who reject (resp. accept) sentences in (38)a, (38)c to accept (resp. reject) sentences like hisi mother likes Johni. She argues that this correlation follows from the assumption that in one dialect, the c-command domain of the specifier is extended to that of the NP containing it.

References Aoun, Y., N.Hornstein and D. Sportiche (1981) “Some aspects of wide scope quantification”, Journal of Linguistic Research 1(3). Aoun, Y. and D.Sportiche (1981) “A formal theory of government”, to appear in The Linguistic Review. Belletti, A., L.Brandi and L.Rizzi, (eds) (1981) Theory of Markedness in Generative Grammar, proceedings of the 4th GLOW Colloquium at Pisa, Scuola Normale Superiore, 1979. Borer, H. (1981) “On the definition of variables”, Journal of Linguistic Research 1. Brody, M. (1981) “Binding theory and the generalized ECP”, unpublished manuscript presented in part at NELS XII at MIT and 1982 GLOW in Paris. Chomsky, N. (1972) “Deep structure, surface structure and semantic interpretation”, Studies on Semantics in Generative Grammar, Mouton, The Hague. Chomsky, N. (1976) “Conditions on rules of grammar”, Linguistic Analysis 2.


Chomsky, N. (1981) Lectures on Government and Binding, Foris Publications, Dordrecht. Chomsky, N. (1982) Some Concepts and Consequences of the Theory of Government and Binding, MIT Press, Cambridge, Mass. Engdahl, E. (1980) “Parasitic gaps” to appear in Linguistics and Philosophy. Fiengo R. and J.Higginbotham (1981) “Opacity in NP”, Linguistic Analysis 7(4). Guéron, J. (1981) “Logical operators, complete constituent and extraction transformations”, Levels of Syntactic Representation, R.May and J.Koster (eds) Foris Publications, Dordrecht. Haïk, I. (1982) “Indirect binding”, unpublished article, MIT, read at 1982 GLOW Colloquium in Paris. Higginbotham, J. (1980a) “Pronouns and bound variables”, Linguistic Inquiry 11(4). Higginbotham J. (1980b) “Anaphora and GB: some preliminary remarks”, NELS, X. Hornstein, N. (1981) “Two ways of interpreting quantifiers”, unpublished article, Columbia University. Huang, J. (1980) “Move wh in a language without Wh-movement”, The Linguistic Review 1:369– 416. Kaye, J. (1981) “On the selection of pronouns in Vata and other eastern Kru languages”, Revue Québécoise de Linguistique I(1). Kayne, R. (1981) “Two notes on the NIC”, in A. Belletti, L.Brandi and L.Rizzi, (eds). Koopman, H. (1981) “Theoretical implications of the distribution of quoi”, NELS XII. Koopman, H. (1980) “Subject/object asymmetries in Vata”, unpublished article, UQAM, presented at the 5th GLOW Colloquium at Nijmegen. Koopman H. (1982) “Les relatives en Haitien”, Syntaxe du Haitien, C.Lefèvre et al. (eds) Karma, Ann Arbor. Koopman, H. and D.Sportiche (1981) “Binding theory and pronouns in Vata”, unpublished article, MIT and UQAM. Lasnik, H. (1976) “Remarks on coreference”, Linguistic Analysis 2(1). May, R. (1977) “The grammar of quantification”, unpublished PhD dissertation, MIT. Reinhart, T. (1976) “The syntactic domain of anaphora”, unpublished PhD dissertation, MIT. Reinhart, T. (1979) “Syntactic domains for semantic rules”, in F.Guenthner and S.Schmidt (eds) Formal Semantics and Pragmatics for Natural Languages. Reinhart, T. (1980) “Coreference and anaphora: a restatement of the anaphora Questions”, unpublished article, Tel Aviv University. Reinhart, T. (1981) “Definite NP anaphora and C-command domains”, in Linguistic Inquiry, 12(4). Riemsdijk, H. van and E.Williams (1981) “NP structure”, The Linguistic Review 1: 171–217. Rizzi, L. (1980) “Negation, Wh-movement and the null subject parameter”, Issues in Italian Syntax, Foris Publications, Dordrecht. Sportiche, D. (1980) “A short note on focus”, unpublished article, MIT. Sportiche, D. (1983) PhD dissertation, MIT. Taraldsen, T. (1981) “The theoretical implication of a class of marked exceptions”, in A. Belletti, L.Brandi and L.Rizzi (eds).

2 PRONOUNS, LOGICAL VARIABLES, AND LOGOPHORICITY IN ABE* with Dominique Sportiche

Introduction 1.1 Preliminaries In this chapter we describe and analyze the properties of the pronominal system of Abe, a Kwa language spoken in the Ivory Coast, which we view as part of the study of pronominal entities (that is, of possible pronominal types) and of pronominal systems (that is, of the cooccurrence restrictions on pronominal types in a particular grammar). Abe has two series of third person pronouns. One type of pronoun (O-pronoun) has basically the same properties as pronouns in languages like English. The other type of pronoun (n-pronoun) very roughly corresponds to what has been called the referential use of pronouns in English (see Evans (1980)). It is also used as what is called a logophoric pronoun —that is a particular pronoun that occurs in special embedded contexts (the logophoric contexts) to indicate reference to “the person whose speech, thought or perceptions are reported” (Clements 1975). This dual use of n-pronouns makes this pronominal system particularly interesting because it permits placing the logophoric properties in a wider perspective. This chapter contains four major sections. In section 1 we present some back-ground assumptions. In section 2 we present the data on referential interactions between pronouns and nominals. In section 3 we offer an analysis of the pronominal system excluding logophoric effects. We motivate our proposal that the n-pronouns should in fact be treated as LF (or logical) variables, that is, as being Ā;-bound by an operator at LF. This proposal provides a key to the complexity and apparent contradictions of Abe’s pronominal system. It is against this background that we address the problem of logophoric pronouns in section 4. The rather surprising properties of logophoric pronouns in Abe follow from the conjunction of two independent factors. First, they are instances of pronouns that are treated as logical variables. Second, certain types of clausal complements in Abe (as in many West African languages) are introduced by verbal complementizers. We will show how these apparently unrelated syntactic factors conspire to yield the typical logophoric effects. In Abe, then, there is no logo-phoric pronoun as

38 PRONOUNS, LOGICAL VARIABLES, AND LOGOPHORICITY IN ABE

such. If our analysis can be successfully extended to other languages, there is no “logophoric pronoun”; there are pronouns treated as logical variables yielding “logophoric” effects in certain syntactic contexts. 1.2 Background Following Chomsky (1981) and much other work, we suppose that pronouns realize (phonologically or not—that is, overtly or covertly) the possible bundles of features drawn from the set of so-called -features, including number, gender, person, Case, and so on (or whatever is the right set of features generating the space of possible NPs, apart from lexical properties; see; for example, the class systems of Bantu languages). Furthermore, we assume that the relevant interpretive properties are described as follows: Pronouns can be used independently; this requires no special statement. An independent pronoun makes reference (with all the usual provisos about reference) to an entity present in the shared perceptual environment or rendered salient in some other way. Speaking about John, we may say He left, interpreting it exactly like John left, that is, treating he as a kind of proper name. A pronoun can also depend for its reference upon another NP. This NP is called the antecedent of the pronoun. In a certain domain D (governing category or GC) a pronoun cannot depend for its reference upon a c-commanding NP: this property is accounted for by Principle B of the binding theory. We take referential dependency to be ambiguous between coreference (or more generally referential overlap) and binding; that is, pronouns so used are either referential or bound variables (see, for example, Lasnik (1976), Reinhart (1983), Sells (1986), Evans (1980)). The referential case is exemplified by a sentence like John likes his mother, where the speaker intends his to refer and to have the same reference as John. The bound variable case is exemplified when a pronoun has a quantified expression as antecedent and is used in such a way as to be analogous to the bound variable of the logician: Every man/No man saw his mother, x/No x, x a man, x saw x’s mother. Naturally, in such a case, neither the pronoun nor its antecedent is referential. We will say that the pronoun is bound by its antecedent or that it is a case of pronominal binding. We will assume that the bound pronoun interpretation can arise iff the pronoun is ccommanded1 by its antecedent (Condition on Pronominal Binding).2 We represent all cases of referential dependence as coindexing: we assume with Lasnik (1981) and Higginbotham (1983) that both the coreference and the bound variable cases are syntactically represented (but see Reinhart (1983) for a different view). We assume furthermore that referential dependence of X upon Y can be interpreted as coreference only if Y is a referential NP and Y is in an A-position. We therefore have the following possibilities: (i) Y is in an A-position and Y is referential: coindexing is either coreference or binding.

THE SYNTAX OF SPECIFIERS AND HEADS 39

(ii) Y is not referential, or Y is not in an A-position: coindexing can only be interpreted as binding. If X is in an A-position, these are standard assumptions. However, we need to specify what happens if X is in an Ā;-position. The case we are interested in is the case where X is an operator in an Ā;-position. We assume that coindexing of such an operator with some Y can in principle be interpreted exactly as above. It can be bound, which means that its range is identified with the value of some variable. This situation is illustrated by a sentence like No onei is too stubborn [Opi for [John to talk to ei]]. It can also be coreferential with some Y, provided that the range of such an operator can be set in the absence of any antecedent (just like the reference of a pronoun can be set in the absence of any antecedent); that is, if, intuitively speaking, it is “referential.” Such a case might be illustrated in English with a focalized pronoun: Johni thinks that himi, Mary likes. Note again that we simply extend current assumptions about arguments to operators. Unlike pronouns, anaphors usually require some antecedent in order to have their reference fixed. This is why an anaphor must be bound by its antecedent, but a pronoun may either be bound by or corefer with its antecedent. Finally, we assume that if X is referentially dependent upon Y, X cannot c-command Y. This assumption is made explicit in various ways by different authors (see, for example, Higginbotham (1983) or the work on Principle C of the binding theory). Consequently, if X and Y are coindexed and neither c-commands the other, referential dependency may go either way. If X c-commands Y, Y is referentially dependent upon X. 1.3 The pronominal system of Abe We will be exclusively concerned with third person singular pronouns in Abe; first and second person pronouns behave like English or French first and second person pronouns. There are two series of morphologically distinct third person singular pronouns. We will refer to these third person pronouns respectively as pronouns of the O-series (or Opronouns) and pronouns of the n-series (or n-pronouns). The distinction is not marked in the plural. (1)

O-series [± animate] a. b.

Nominative/Inalienable Elsewhere

O

n-series [+ human] n n

The nominative pronoun of the O-series and the genitive pronoun expressing inalienable possession are null pronouns3 (context (la)); all other forms are represented as O (context (1b)). Pronouns of the n-series are invariably represented as n. The tone on both series of pronouns varies and depends on mood, tense, and aspect features (nominative n), on construction type (genitive, in conjunction with tonal properties of the following noun), and on Case (accusative pronouns). Tone will not be indicated here. Pronouns of the O-


series can stand for or replace both animate and inanimate nouns; those of the n-series refer exclusively to humans. As is to be expected, pronouns of the O-series and the n-series can be used as independent pronouns: to questions like those in (2), the answers in (3) are appropriate.

2 Referential interactions We now turn to the referential interaction between a pronoun, call it the target, and another NP, call it the antecedent. In order to describe the properties of the pronominal system of Abe, we must take into account the nature of the antecedent and the nature of the target. The antecedent can be a referential NP, a non-referential NP, an O-pronoun, or an n-pronoun. The target can be either an O-pronoun or an n-pronoun. The initial description, summarized in section 2.3 and analyzed in what follows, will be restricted to configurations in which the antecedent c-commands the target. Other configurations will be discussed in section 3. Until section 3.3 we will further limit our discussion to coindexing, not taking the coreference/bound variable difference into account except when directly relevant. Nonreferential antecedents will be discussed in section 3.3. We will use the following indexing convention: lack of coindexing between two NPs denotes disjoint reference (that is, coindexing is excluded). In other words, when two NPs cannot overlap in reference, they are contraindexed. In order to clarify the exposition, we will sometimes explicitly indicate that coindexing is impossible. 2.1 The target is an O-pronoun 2.1.1 The antecedent is a lexical NP or an O-pronoun and the target is an Opronoun Let us start with the properties of a target O-pronoun, with respect to a lexical NP or Opronoun antecedent. Ignoring the specific properties of Abe—in other words, putting aside the n-pronoun and the logophoric contexts–pronouns of the O-series basically behave in the same way as pronouns in English or French. First, pronouns of the O-series obey Principle B of the binding theory and must be contraindexed with c-commanding NPs in their GC:


(4)

yapii/ i Yapi/he(O)

wu Oj,*i saw him (O)

As expected, they may be coindexed with an antecedent outside their GC: (5)

a. b.

yapii/ i Yapi/he(O) yapi/ i Yapi/he(O)

wu saw wu saw

[NPOi,j his (O) orovi snake

wo dog [PP Oij him

n] Det SE] (O) near

When the target occurs in an embedded complement, its behaviour depends on the type of complement. In an indicative embedded complement introduced by the complementizer ye the O-pronoun may be coindexed with an antecedent outside its GC: (6)

nkui/ i Nku/she (O) Nkui/ i Nku/she(O)

a. b.

bO wu believed bO wu believed

ye ye ye ye

Oi,j she f you

mU (O) mU knew

api knew Api Oi,j her(O)

In two other embedded environments, however, the O-pronoun behaves differently: if it occurs in subject position of a subjunctive complement, and if it is contained in a complement that induces logophoric effects (kO-complements). Here we will only describe the referential possibilities in subjunctive complements, reserving the discussion of logophoric effects in kO-complements for section 4. We call the clausal complements of certain Abe verbs subjunctive complements. These verbs translate in languages with subjunctive morphology (say, French) as verbs typically taking control complements or subjunctive complements. In Abe these subjunctive complements contain a special mood marker, which is realized tonally on Infl and morphologically in the form of certain nominative pronouns. The following examples illustrate the referential possibilities of the O-pronoun in subjunctive complements: (7)

a.

b.

ye api b- wu yapii kolo Yapi wants ye Api Subj see him(O) “Yapi wants Api to see him.” ye O-bi, *j yapii kolo Yapi wants ye he(O) Subj “Yapi wants to see Api.”

Oi,j

wu api see Api

An O-pronoun in object position of a subjunctive complement behaves like the O-pronouns in (6). In subject position, however, an O-pronoun must be coindexed with the subject of the main verb. Disjoint reference is indicated by the n-pronoun (see section 2.2.1).


2.1.2 The antecedent is an n-pronoun and the target is an O-pronoun So far the O-pronoun behaves like a regular pronoun with respect to the binding theory, except when it occurs in subject position of a subjunctive complement. Quite a different situation obtains, however, if the antecedent is not a lexical NP or an O-pronoun but an npronoun. The following examples establish that the O-pronoun must be taken as disjoint from a c-commanding n-pronoun: (8)

a. b.

wu saw wu saw

ni he(n) ni he(n)

Oj, *i his(O) orovi snake

wo dog Oj,*i him(O)

n Det sE near

Moreover, disjoint reference is not restricted to a local domain, but holds throughout: (9)

a.

ni he(n) ni he(n)

b.

bO wu believed bO wu believed

ye ye ye ye

j,*i

api Api

mU he(O)knew mU knew

api Api Oj, *i him(O)

Finally, subject orientation (or anti-orientation) is not involved here. Disjoint reference also holds with respect to a c-commanding n-antecedent that is not in subject position: (10)

m I

hE told

Oj,*i his(O)

ni him(n)

wo dog

n Det

jO about

2.2 The target is an n-pronoun 2.2.1 The antecedent is a referential NP or an O-pronoun and the target is an npronoun We now illustrate the behavior of an n-target with respect to a referential NP or Opronoun antecedent. If the potential antecedent is too “close” to the n-pronoun in a way to be defined below, contraindexing is required ((11d) illustrates the subject of a subjunctive complement): (11)

a. b.

yapi/ i Yapi/he(O) yapii/ i Yapi/he(O)

wu saw wu saw

nj,*i him(n) [NP nj,*i wo n] his(n) dog Det


c. d.

yapii/ i Yapi/he(O) yapii/ i Yapi/he(O)

wu saw kolo wants

orovi [pp nj,*i sE] snake him(n) near ye nj,*i wu ye he(n) see

api Api

If the antecedent is not too close—say, if it occurs in the matrix clause and the pronoun is in an embedded indicative complement, or in object position of a subjunctive complement —disjoint reference is preferred, but coindexing is possible: (12)

nkui/ i Nku/she(O) nkui/ i Nku/she(O)

a. b.

(13)

yapii/ i Yapi/he(O)

bO wu believed bO wu believed kolo wants

ye n(i),j ye she(n) ye f ye you

ye ye

f you

mU knew mU knew

bo Subj

wu see

api Api n(i),j her(n) n(i),j him(n)

A clear and consistent contrast holds between cases like (11a-c), in which co-indexing is impossible and cases like (12) and (13), in which coindexing is possible. The n-pronoun behaves differently from the O-pronoun: whereas the O-pronoun can be coindexed with an antecedent NP in the same clause but outside the NP or PP containing it (in contexts like (5)), the n-pronoun must be disjoint in reference from it in these environments (contexts (1 1b, c)). 2.2.2 The antecedent and the target are n-pronouns Surprisingly, if the antecedent itself is an n-pronoun, the referential possibilities of the target n-pronoun change. ((14d) illustrates the subject of a subjunctive complement and (15) embedded complements.) (14)

a. b. c. d.

(15)

a.

wu ni he(n) saw n; wu he(n) saw wu ni he(n) saw kolo ni he(n) want “He wants to see Api.” ni he(n)

bO wu believed

n*i, *j him(n) ni, *j his(n) orovi snake ye ye

ye ye

wo dog [ni, *j him(n)

n Det sE] near wu api see Api

mU knew

ni,(j) him(n)

ni, *j he(n)

f you


b.

ni he(n)

bO wu believed

ye ye

ni,(j) he(n)

mU knew

api Api

An n-target must be coindexed with a close—but not too close—n-antecedent (see the contrast between (14a) and (14b-d)). A preference exists to construe two n-pronouns as referring to the same person, but in contexts like (15) different references are possible. An n-target is also necessarily coindexed with an n-antecedent that is not in subject position: (16)

m I

hE told

ni him(n)

ni, *j his(n)

wo dog

n Det

jO about

2.3 Summary The data discussed so far are summarized in (17) and (18). Boundaries intervening between the antecedent and the first constituent containing the pronoun are indicated. Variable context elsewhere can contain any material. (17)

(18)

The c-commanding antecedent is a lexical NP or an O-pronoun: a. NP/ i…[VP V Oj,*i/nj, *i . . .] b. NP/ i…[NP[NP Oi,j/n*i,j] N] c. NP/ i…[PP[NPOi,j/n*i,j] P] d. NP/ i…[ye Oi, *j/ni,j, V (Subj)…] e. NP/ i…[ye Oi,j/n(i),j V(Indic)…] f. NP/ i…[ye…V(Indic/Subj) Oi,j/n(i),j] The c-commanding antecedent is an n-pronoun: a. ni…[VP V O *i,j/n*i, *j] b. ni…[NP[NP Oj/ni] N] c. ni…[PP[NP Oj/ni,] P] d. ni…[ye Oj/ni V(Subj)…] e. ni…[ye Oj/ni,(j) V(Indic)…] f. ni…[ye…V(Indic/Subj)Oj/ni,(j)]

Let us now examine what generalizations emerge from (17) and (18) and which of these are accounted for. Certain patterns in (17) and (18) follow straightforwardly from Principle B of the binding theory: disjoint reference in (17a) and (18a), and the referential options of the Opronoun in (17b) and (17c), assuming that NP and PP count as GCs, that is, as the domain in which the (O-) pronoun must be free. We will return to this in the next section. It is also to be expected that the O- and n-pronouns in an embedded clause are free to coindex or not with an NP outside of their clause (for instance, cases (17e,f)). An


explanation must be given here for the reading preferences of the n-pronoun (preferred coindexing with another occurrence of an n-pronoun, preferred disjoint reference from another occurrence of a lexical NP or O-pronoun). Some problems arise. The first concerns the n-pronoun in NPs and PPs: contrary to expectations, the n-pronoun must be coindexed with the antecedent, if the antecedent is an n-pronoun (see (18b,c)). Moreover, if the antecedent is not an n-pronoun, the n-pronoun must be disjoint in reference from it (see (17b,c)). Descriptively, then, there is a certain domain D such that in D n is O-free or coindexed with an occurrence of n.4 This explains (18a): in domain D n must be coreferential with an occurrence of n. However, in (18a) it must also be disjoint from its antecedent by virtue of Principle B of the binding theory. n therefore cannot occur in object position if the subject is also an n-pronoun. (18f) raises a second problem. The O-pronoun must be n-free regardless of its relative degree of embedding. A third problem arises with respect to the referential possibilities of a pronoun occurring in the subject position of a subjunctive clause. In the context (17d) an Opronoun must be coindexed and an n-pronoun contraindexed with a lexical NP or Opronoun antecedent. In the same context (that is, (18d)) the n-pronoun must be (and the O-pronoun cannot be) taken as coindexed with an n-pronoun antecedent. We will return to this problem in section 3.4. In sum, then, the following generalizations emerge: (19) a. b. c. d. e. f.

O/n is free in its GC. n is O/lexical NP-free in some domain D. Subjunctive facts Two occurrences of n in domain D are necessarily coindexed. O is n-free throughout. n is preferably coindexed with some other occurrence of n not in domain D, and preferably contraindexed with some NP or O-pronoun antecedent not in domain D.

(19a) needs no further discussion. We now turn to the problem of (19b,d,e,f). (19c) will be discussed in section 3.4. 3 The syntax of pronouns We now develop our analysis, where we will start with the generalizations directly involving the n-pronoun in (19b,d,e). The first question we need to answer is how domain D should be characterized. We can greatly facilitate the discussion by noticing that the data presented so far are consistent with the assumption that the domain D of some item is the first clause containing that item. Let us adopt this preliminary and informal characterization of domain D.


We will establish in section 3.1 that the characterization of domain D is not dependent on binding-theoretic notions. Rather, the domain D of some item should be characterized as the first clause containing this item and a Comp. In section 3.2 we will present the analysis of the pronominal system. In section 3.2.1 we will motivate an analysis according to which n is an LF variable that needs to be operatorbound at LF. We will show how this proposal, together with some independently motivated assumptions, accounts both for the n-freeness of O (19e) and for the characterization of domain D as a clause with a Comp. In section 3.2.2 we will demonstrate how the behavior of a target n-pronoun with respect to an O-pronoun antecedent follows. Moreover, we will show that some surprising predictions about configurations in which there is no (S-Structure) c-command between the antecedent and the target are borne out and thus provide strong support for the analysis. In section 3.2.3 we will discuss the analysis of an n-pronoun with respect to an n-antecedent in domain D. In section 3.3 we will turn to the pronominal binding properties of O and n, which will straightforwardly follow from our analysis. Finally, in section 3.4, we will show that the subjunctive facts are consistent with our analysis. 3.1 The nature of domain D 3.1.1 GC Given the data so far, it is not implausible to suppose that domain D is somehow related to the binding theory. A simple assumption would take the domain of an item to be this item’s governing category. This would contradict the account of the O-pronoun given in (19a), where we claimed that taking NP and PP as GCs could account for the referential options of an O-pronoun occurring in NP or PP. Here, for the n-pronoun, we would need to say that NP and PP are not GCs; rather, the first clause must be. Independent evidence, based on the distribution of the reflexive anaphor Ose/nse, suggests that our initial position is more plausible. The reflexive anaphor is, at least historically, composed of one of the third person pronouns O or n and a noun meaning “body”. The surface form of the anaphor depends on the features of the antecedent (O-se with a name or O-pronoun antecedent, n-se with an n-pronoun antecedent): (20)

a. b.

yapi Yapi n he(n)

mU knows mU knows

Ose/*nse him(O)self nse/*Ose him(n)self

Reflexive anaphors are excluded from positions inside NPs and PPs in which pronouns are permitted, showing that NP and PP must count as the GC for the reflexive anaphor:


(21)

a. b.

*yapi Yapi *n he(n)

mU knows mU know

[NP [NP

Ose him(O)self nse him(n)self

erenyi] house erenyi] house

Given an anaphor and a pronoun in the same position, the GC of the anaphor is always larger than or equal to the domain of the pronoun (see Chomsky (1986)). In other words, the domain in which an anaphor must be bound is larger than or equal to the domain in which a pronoun must be free. This is simply illustrated by pairs of English examples such as They like their shoes/They like each other’s shoes, They think that pictures of them are on sale/ They think pictures of each other are on sale. If NPs and PPs count as GCs for anaphors, we can conclude that they should also count as GCs for pronouns. Thus, it might be possible that the binding theory is involved, but not in such a simple way. Let us consider an alternative binding-theoretic account. 3.1.2 (Accessible) SUBJECT If domain D is clausal, it is tempting to try to relate this to characteristics of clauses that the kind of NPs and PPs that we have seen so far do not have. One such characteristic is the presence of an (accessible) SUBJECT. We could then state: (22) In the domain of an (accessible) SUBJECT, n must be (lexical NP and O-pronoun) free and two occurrences of n are necessarily coindexed. This would work for clauses, but consider the predictions it makes in (23a,b) with respect to pronouns that are embedded in NPs or gerunds that contain a subject (see below for evidence bearing on the existence of gerunds in Abe). Or, if accessibility is involved, consider the predictions in (23c) with respect to pronouns that would have no accessible SUBJECT in their clause: (23)

a. b. c.

[NP NP [N[NP n…] N] [GERUND NP [VP n]] [CP…[ye [[NP n N]…]]]

(22) predicts that the n-pronoun in (23a,b) need not be necessarily O-free or coindexed with n in its clause; moreover, in (23c) it would need to be O-free or coindexed with n in the superordinate clause. Structures like (23a) are prohibited in Abe: only one argument may precede the noun, and none may follow. (The meaning equivalent to (23a) is rendered in Abe by a relative clause.) Examples of the structures in (23b) and (23c) do not support (22). The following examples show that the notion (accessible) SUBJECT is too strong, since it creates a domain D when one is not present:5


(24)

a.

b.

/apii fa [OERUND mj she/Api forbade me “She/Api forbade my hitting her.” [ fa ni GERUNDmj she(n) forbade me “She forbade my hitting her.”

[VP n*i,k her(n) [VPni,*k her(n)

e]] hit e]] hit

We analyze the preceding structures as gerunds, for the following reasons: (a) The head bears nominal tonology. (b) Just like direct objects, objects in this construction can be extracted by Wh Movement, leaving a trace. Extraction of the object of N must leave a resumptive pronoun. This suggests the presence of a verbal projection. (c) Extraction of adjuncts is excluded, as it is from within English NPs or gerunds (but not infinitivals). (d) The verbal head cannot be extracted under predicate clefting (Wh Movement of the verb; see Koopman (1984)), whereas it can be in clauses. Moreover, the following examples show that the notion accessible SUBJECT is too weak. It wrongly predicts that the embedded clause should not count as a domain D, when it does: (25)

a.

b.

bO wu ye n(i),j ceewu yapii Yapi believe ye his(n) friend “Yapi believed that his friend was coming.” bO wu ye ni,(j) ceewu ni he(n) believe ye his(n) friend “He believed that his friend was coming.”

n Det

a come

n Det

a come

Given these data, (22) does not seem adequate. 3.1.3 The binding theory The notions used by the binding theory to define binding domains (governing category or accessible SUBJECT) do not seem relevant for the definition of domain D. However, one could still try to tailor the definition of domain to this particular case, so that even though D is not defined as a usual binding domain, the behavior of n is a binding theory effect. In fact, there are strong reasons to believe that the binding theory is not involved in the computation of domain D. Any analysis that attempts to account for (19b) by means of disjoint reference runs into the problem with (19d): coindexing of n with an n-antecedent is obligatory in exactly the same domain. Consider for concreteness an attempt to derive the asymmetry between the O-and npronoun in (17b) from the way in which the respective binding domains of O-pronouns and n-pronouns are computed: (17)

b.

NP/ i…[NP[NP Oi,j/n*i,j] N]


The GC would be NP* for the O-pronoun, but the CP for the n-pronoun. A larger disjoint reference domain for n-pronouns predicts that the n-pronoun has to be disjoint from any antecedent in this domain. This is simply false: n must be coindexed with an npronoun antecedent in the same domain D in which it must be disjoint from a lexical NP or O-pronoun antecedent: b. ni…[NP[NPni,j] N]

(18)

In order to solve this problem, we would have to define the binding domain of n not only in terms of the properties of n but also by making crucial reference to the internal properties of the potential antecedent. That the binding theory is not the right module to look at for an account of these data is further supported by the fact that (S-Structure) c-command within domain D does not play a necessary role in accounting for the coindexing possibilities (as we will show in section 3.2.2). 3.1.4 Domain D is CP At the beginning of this section we noted informally that taking domain D to be a clause seemed to fit all the data that had so far been presented. Let us therefore assume that (26)

a. b.

Domain D of x is the first CP containing x, and consequently Within its CP, an n-pronoun must be obligatorily disjoint from a c-commanding lexical NP or O-pronoun, and coindexed with a c-commanding n-pronoun.

What is the significance of this fact? What characterizes CPs as opposed to other domains like NPs? Clauses are typically domains over which operators (for example, wh-phrases) take scope. Furthermore, the fact that CPs but not gerunds (see (24)) count as domain D suggests that the presence of Comp is crucial, since gerunds do not seem to have (available) Comps (see Stowell (1981)). 3.2 n as an LF variable 3.2.1 The behavior of O Having established that domain D is a CP, we now develop our analysis, beginning with the generalization (19e) that O must be n-free.


There are two surprising aspects to this generalization. First, the prohibition is nonlocal. Second, even when it is interpreted as coreference, coindexing is impossible despite the fact that both pronouns are third person. It turns out that there is a simple way to tie these two facts together. Recall that coreference is possible only if the antecedent is a referential element in an A-position. Coindexing with a nonreferential element must be a case of binding. We will assume that the n-pronoun is not a referential element at the appropriate level of representation. In particular, we will assume that the n pronoun is a variable: (27)

The n-pronoun is a (LF) variable.

Since it is only possible for an element to be coreferential with referential elements, O cannot be coreferential with n, regardless of the distance between them. Of course, this does not yet derive (19e). Judgments on which this generalization is based are judgments prohibiting identity of referential value. Since identity of referential value can arise either by coreference or by pronominal binding, we must still rule out binding. Whatever the reason is, there is independent and direct evidence that binding is ruled out. As (21) shows, there is an anaphor corresponding to each pronoun. But the n-anaphor cannot be bound by an O-pronoun and the O-anaphor cannot be bound by the n-pronoun. (19e) follows. Why should binding be ruled out? O-pronouns and n-pronouns must be distinguished by some feature (say, [± n), and binding requires nondistinctness of certain features (for instance, gender). The feature [± n] could be among them. The n-anaphor cannot be bound by a lexical NP, either (or by a quantified NP; see section 3.3). This indicates that lexical NPs, QPs, and O-pronouns are [− n], whereas n-pronouns are [+ n]. From now on, we will suppose that [+ n] elements can never be bound by [− n] elements and vice versa.6 So far, then, (19e) follows if the n-pronoun is in fact a variable. If n is a variable, we expect the existence of some Ā;-binder or operator to bind it. The preceding discussion shows that this operator must be assumed to be [+ n]. Let us call it the n-operator. Where is this n-operator? If we assume that it must, like a wh-operator, appear in Comp at LF, we will be able to derive why domain D is clausal. Let us then further specify the assumption in (27) as follows: (28)

At LF the n-pronoun is bound by an n-operator in some Comp.

(19e) still follows from assumption (28): O cannot be bound by the n-pronoun or by the noperator because of feature conflict. O cannot be coreferential with n because n is not referential but a variable. Since O is c-commanded by the n-operator, the only option left is that O be referentially dependent upon this n-operator. But binding is excluded (by feature conflict) and coreference with the operator is impossible since it is in an Ā;position. In other words, each time an n-operator c-commands an O-pronoun, the Opronoun and the operator must be contraindexed.


3.2.2 The behavior of n Let us now see how (19b) (n is O-pronoun/lexical NP-free in CP) follows. Consider the following LF representation: [CP OPi […Oj…ni…]]

(29)

Suppose that i=j. The O-pronoun will be coindexed with a c-commanding n-operator. This is ruled out, as we just demonstrated. Of course, this also holds in case O is replaced by a lexical NP. (19e) thus straightforwardly follows from the analysis given so far. Assumption (28) furthermore derives (26a), namely, that the domain in which n must be O-free is the first CP. By (28), some Comp c-commanding n contains an operator binding n at LF. If this operator also binds O, the resulting configuration is ill-formed. Consider the following configuration: (30)

[CP Comp1…[CP Comp2…[CP Comp3…O…n…]]]

The n-operator could be in Comp,1 Comp,2 or Comp3. But no matter where it is, it will c-command O if Comp3 is the first Comp c-commanding O. In other words, the configuration will be ill-formed if the first Comp c-commanding n is also the first Comp ccommanding O, that is, if they are clausemates. This derives (26b) in part: n must be disjoint from O in CP, because O must be n-free. If this analysis is correct, we expect that c-command between the pronouns is irrelevant in certain configurations—for example, in the configuration in (31), where neither n nor O c-commands the other, but where the Comp c-commanding n ccommands O:

We expect that the O-pronoun and the n-pronoun must be contraindexed, since otherwise the O-pronoun would be bound by the n-operator. This prediction is borne out:7 (32)

a.

b.

[ni tEEwu foto n] his(n) enemy picture Det “The picture of his enemy bothered him.” tEEwu foto n] [Oi his(O) enemy picture Det

1E bother

Oj,*i him(O)

tE Part

1E bother

nj,*i him(n)

tE Part

In such examples the O-pronoun must be disjoint from the n-pronoun. The correct generalization then is the following:


(33) An O-pronoun cannot be coindexed with an n-pronoun if the Comp of the first clause containing n c-commands O. However, the analysis we are developing predicts that an O-pronoun may very well be coindexed with an n-pronoun if the O-pronoun is not c-commanded by the Comp of the nclause. Indeed, consider such a configuration: [CP[CP OPi…ni]…Oi,j]

(34)

n is used like a third person pronoun. This means that the operator that binds it ranges over a set containing only one element—say, the singleton w. We know that the operator can have its range (the value of w) fixed in the absence of any antecedent, since n can be used independently. Suppose all three elements are coindexed. Since no c-command obtains between the operator and the O-pronoun, this cannot be a case of binding. However, it could be a case of coreference. If w is assigned the same referent as the Opronoun, coreference will arise. Coreference is permitted, since the O-pronoun may be referential and occurs in an A-position.8 The following examples show that this is indeed the case: (35)

a.

b.

a su, hE na ni,j i he(n) arrive he(O) told the “After he arrived, he told the truth.” [kolo n f kolo ni,j] love Rel you love him(n) “The fact that you love him bothers him.”

hOrE truth 1E bothers

Oi him(O)

tE Part

Reversing the order of the pronouns in this context should yield different results, as will be clear from (36): [CP OPi [CP…Oj]…ni]

(36)

The n-operator will c-command O, which will therefore be in the scope of the n-operator. Coreference should therefore be impossible and contraindexing obligatory. This prediction is borne out, as the following examples illustrate: (37)

a su, nj hE na hOrE i he(O) arrive he(n) told the truth “When he arrived, he told the truth.” b. [kolo n f yo kolo Oj] love Rel you Neg love him(O) “The fact that you don’t love him bothers him.”

a.

These data then strongly support our analysis.

1E bothers

ni him(n)

tE Part


Concluding this part, we have established that (i) The descriptive generalization in (19b) should be replaced by (33): an O-pronoun cannot be coindexed with an n-pronoun if the Comp of the first clause containing n ccommands O. (ii) (33) and (19e) follow from the requirement that n must be Ā;-bound from Comp at LF. 3.2.3 n in domain D So far we have discussed the behavior of an n-pronoun with respect to an O-antecedent and vice versa. We now turn to (19d,f): two occurrences of n in the same domain D (that is, clausemates) must be coindexed; n is preferably coindexed with some other occurrence of n not in domain D, and preferably contraindexed with some NP or Opronoun antecedent not in domain D. Our analysis rests on the assumption that n is a variable at LF and that it is Ā;-bound at LF. However, we have not said what we mean by variable. A syntactic or formal variable is a category locally A- or operator-bound: a wh-trace, for example. A logical variable need not be a formal variable, however. An English example like Who thinks he is sick? can be interpreted as For which x, x thinks that x is sick. The pronoun he is treated as a (logical) variable even though it is not locally Ā;-bound. If we interpret (27) and related statements as requiring that n be treated as a logical variable, we conclude that it is not necessary that there be as many n-operators as there are n-pronouns. If n-operators are needed, it suffices to introduce the minimal number of such operators so that all occurrences of n-pronouns that need to be bound end up being Ā;-bound. Consider the case of two occurrences of n in a simple clause: (38)

[CP Comp[…n… …n…]]

In order for them to be Ā;-bound, we need to postulate the presence of at least one noperator in Comp to bind them. Moreover, if there is only one such operator, the two occurrences of n will both have to be bound by it and therefore will have to be coindexed. Thus, if no more than one n-operator is allowed per Comp, property (19d) follows. We will therefore assume: (39)

At most one n-operator is allowed per Comp.9

If this analysis is correct, we expect (19d) to hold, even if there is no c-command relation between the pronouns. In a configuration in which one Comp node is available, where the clause contains two n-pronouns n1 and n2 and where n1 does not c-command n2, n1 and n2 will have to be coindexed: (40)

[CP OP [[n1]…n2]]


The following example illustrates that this is indeed the case:10 (41)

ni his(n)

ceewu friend

n Det

kolo likes

ni ,*j him(n)

Obligatory coindexing in domain D follows from (39) and the requirement that n be a variable at LF. This analysis also explains the impossibility of (18a): n cannot occur in object position if the subject is also an n-pronoun: [CP OP [n [V n]]]

(42)

Since only one operator is allowed, both instances of n will have to be bound by the same n-operator. Therefore, the n in direct object position will also be bound by the subject n, and Principle B is violated. Actually, property (19d) has been established only on the basis of simple clauses. In such structures, only one Comp node is available. According to (39), only one n-operator may occur per Comp. Suppose now that we embed a simple clause like (38), so that the structure contains two Comp nodes: [CP ComP1…[CP Comp2 […n… …n…]]]

(43)

Since we have two Comp nodes, we may have two distinct operators, one in each Comp, and each binding one occurrence of n. Nothing forces coindexing in this case. Our account therefore makes the curious prediction that two occurrences of n must be coindexed if they occur in the same domain D and if they are both c-commanded by only one Comp node. This prediction appears to be correct. The sentences in (44) illustrate the possibilities left open by this prediction: (44)

a. b.

api Api api Api

bO wu believes bO wu believes

ye ye ye ye

n he(n) n his(n)

kolo likes (ceewu) (friend)

n he(n) kolo likes

n his(n)

erenyi house

In each of these sentences the two occurrences of n may be contraindexed. In (44b) this is possible but not preferred. In (44a) it is the only option; this is especially significant because its nonembedded counterpart is ungrammatical. Let us now turn to the case where two occurrences of n are found in different clauses: (45)

[CP ComP1 […n…[CP Comp2 […n…]]]]

Recall that in such configurations the two occurrences of n are preferably coindexed. In order to minimally satisfy the requirement that each n be Ā;-bound, it suffices to postulate the presence of an n-operator in Comp1. If so, the two ns will both be interpreted as variables bound by the n-operator in Comp1 and will therefore be


coindexed. But there is another option, namely, that we postulate one n-operator per Comp. In that case the two pronouns will not have to be coindexed. Each n will be bound by its own operator and will have its referential value determined by the range of its operator, and the two operators do not have to be coindexed. The preference for the coindexing option will follow if we suppose that the requirement for Ā;-binding is preferably satisfied by giving as wide a scope as possible to the necessary n-operators. It is clear that this is minimally satisfied by postulating the presence of one n-operator in the matrix Comp. If it is preferred that there be only one operator, it is preferred that the two occurrences of n be coindexed. Given that each Comp contains at most one such operator, any other postulated noperator will fail to have widest possible scope. The same assumptions entail that in (46) (46)

[CP ComP1 […O/lexical NP…[CP Comp2 […n… …]]]]

it is preferred to postulate only one n-operator, situated in the matrix Comp (although it is possible in addition to postulate one operator situated in Comp2). From this, we expect it to be preferred to take the n-pronoun as being disjoint from O or the lexical NP in a higher clause (see the discussion of (19e)), as is indeed the case. 3.3 Pronominal binding The analysis presented so far disregards whether coindexing is interpreted as coreference or binding. We now turn to this aspect of the problem. Recall that a pronoun may be either coreferential with some other NP (if this NP is referential and in an A-position) or bound by another NP (as for instance when this NP is a quantified expression). In examples like John saw his mother, the pronoun can be either a referentially used pronoun (John saw John’s mother) or a bound variable (John=x, x saw x’s mother). Following Reinhart (1983) (see also Lasnik (1976), Sag (1976), Williams (1976)), we will assume that, to a certain extent, VP deletion provides a test for this difference. (47)

John saw his mother and Bill did too.

We understand (47) as meaning that “John saw his mother and Bill saw his mother too”. The existence of some level of mental representation at which the elliptical conjunct is reconstructed by filling in information taken from the first one is postulated. Assuming that the first conjunct means that “John saw his own mother”, the second one is ambiguous between “Bill saw John’s mother” (the nonsloppy identity reading) or “Bill saw his own mother” (the sloppy identity reading). If his is indeed used either referentially or as a bound variable, we have an explanation for the fact that we get exactly these two interpretations. In the first case John saw his mother in fact stands for “John saw John’s mother”. Reconstruction of the missing part in the second conjunct will yield “Bill saw


John’s mother”. In the second case the first conjunct stands for “John=x (x saw x’s mother)”. Reconstruction will yield “Bill=x (x saw x’s mother)”. However, elaborating on Sells (1987), we also assume that it is not true that the nonsloppy identity reading arises iff the pronoun is coreferential with its antecedent. Sells provides cases in which nonsloppy readings arise with nonreferential pronouns in what he terms the cospecificational reading: (48) With each new Hollywood hit, the lead actress thinks she is the new Monroe and the director does too. Here, nonsloppy identity is possible (the director thinks that the lead actress is the new Monroe). The NP antecedent of the pronoun is in the scope of the quantifier and thus not referential. The pronoun is not referential either. This shows that having a nonreferential antecedent does not guarantee sloppy identity. In sum, if sloppy identity is available, we can only conclude that the pronoun is a bound variable. Reciprocally, if the pronoun has a referential antecedent, nonsloppy identity is available.11 3.3.1 Binding of O-pronouns We expect that O-pronouns will basically behave like English pronouns. The examples in (49) show that O-pronouns can be interpreted as pronouns bound to a ([− n]) nonreferential antecedent (quantifier or wh-phrase): (49)

a.

b.

bO γ0 wu apOUNi γo nobody neg take Neg see For no x, x believed that x knew Api bO γo wu apOUNi γo nobody Neg take Neg see For no x, x believed that you knew x

Ye ye

he(O)

mU knew

api Api

ye ye

f you

mU knew

Oi him(O)

i

Besides being bound to a quantificational NP, the O-pronoun can also be bound to a whquantifier and can occur as a resumptive pronoun in subject position,12 indirect object position, and genitive contexts: (50)

f mU cai who you know “Whose house do you know?”

Oi his

erenyi house

e wh

The data on sloppy identity corroborate these conclusions. The examples in (51) show that the O-pronoun can give rise to sloppy identity interpretation: that is, it can be bound by a referential NP: (51) yapi/

bO wu

ye

mU

api,

yavo

ese


Yapi/he (O) take see ye he(O) knew Api “Yapi believed he knew Api, and Yavo too.” (i) believed Yapi knew Api (nonslop py) (ii) believed he (Yavo) knew Api (sloppy)

Yavo too

In these examples the antecedent of the bound pronoun c-commands it. In general, the structural conditions for pronominal binding in Abe seem to be identical to those found in English. Thus, in (52) the pronoun cannot be interpreted as bound, due to lack of ccommand: (52)

[kolo n f γo kolo apOUNi] love Rel you Neg love nobody “The fact that you love nobody bothers him.”

1E bothers

Oj, *i him(O)

tE Part

To sum up: The O-pronoun can be bound to a QP or a referential NP and occurs as a bound pronoun (resumptive pronoun) in wh-questions, relatives, and topicalizations in certain positions (subject, indirect object, genitive). The last possibility to consider is whether O can be bound by n. This question has already been addressed in the previous section: O can never be bound by a [+ n] antecedent. When an O-pronoun and an npronoun are coindexed, sloppy identity is not possible (ignoring the data discussed in section 4). 3.3.2 Binding of n-pronouns This raises the question of whether the n-pronoun can occur as a bound pronoun at all. n cannot be bound by a quantifier like nobody:13 (53)

a.

b.

apOUNi γo bO γo wu nobody Neg take Neg see “Nobody believed that he knew Api.” γo bO γo wu apOUNi nobody Neg take Neg see “Nobody believed that Api knew him.”

ye ye

n*i,j he(n)

mU know

api Api

ye ye

api Api

mU know

n*i,j him(n)

As we would expect, the sloppy identity interpretation is not available in elliptical contexts: (54)

bO wu ye n mU api yapi/ Yapi/he(O) take see ye he(n) know Api “Yapi believes that he(n) (Yapi) knows Api, and Yavo too.”

yavo Yavo

ese also


(i) knows that he(n), Yapi knows Api (nonsloppy) (ii) * sloppy identity interpretation This follows straightforwardly: n cannot be bound by a [− n] element, and binding is a prerequisite for the sloppy identity interpretation. Binding of an n-element should be possible, however, if the antecedent is an n-element. And in fact, an n-pronoun can be a bound pronoun if the antecedent is also an n-pronoun. Several constructions illustrate this possibility. In wh-constructions the n-pronoun occurs as a resumptive pronoun in certain extraction sites (subject, genitive, indirect object) if the antecedent is an n-pronoun. This is the case if an n-pronoun is topicalized (note that these examples are syntactic counterparts to the proposed LF representation of sentences containing n-pronouns): (55)

a.

b.

n O m bO wu he(n) Top I take see “It is him who I believe knows Api.” n O f mU him(n) Top you know “It is his house you know.”

ye ye n his(n)

n he-R

mU know

api Api

erenyi house

Moreover, in elliptical constructions the sloppy identity interpretation is available if the antecedent is an n-pronoun: (56)

a.

b.

n kolo n ceewu n, yavo ese he(n) like he(n) friend Det Yavo too “He likes his friend and Yavo does too.” (i) Yavo likes Yavo’s friend (sloppy) (ii) Yavo likes n’s friend (nonsloppy) n bO wu ye n mU api, yavo he(n) take see ye he(n) know Api Yavo “He(n) believes that he(n) knows Api, Yavo too.” (i) believes that Yavo knows Api (sloppy) (ii) believes that he(n) knows Api (nonsloppy)

ese too

In fact, both the sloppy and the nonsloppy interpretation are available in (56b).14 In sum, the n-pronoun cannot be bound by a non-n antecedent, because the n-element cannot be bound by a [− n] NP. The n-pronoun can be bound by an n-operator as in (55). These properties follow without any additional assumptions. A final remark on the occurrence of several n-pronouns in the same domain D: recall that only one n-operator is allowed per CP so as to account for obligatory coindexing of all the n-pronouns in CP. This correctly predicts that in elliptical constructions the sloppy interpretation should be possible regardless of c-command relations, since both instances of the pronoun are operator-bound (note that Yapi in the elliptical conjunct is an object):


(57)

n ceewu his(n) friend (sloppy and nonsloppy)

kolo loves

n, him(n)

yapi Yapi

ese too

Because of the presence of the operator, the first clause is treated as “The x, x human, such that x’s friend loves x”. Reconstruction of the second conjunct may yield the sloppy identity reading, which is the relevant reading for our purposes. The nonsloppy reading is also available, possibly a case of cospecification in Sells’s sense. However, the non-sloppy reading can also arise simply if the second conjunct is understood as conjoined to the object of the first clause, rather than to the whole clause. Parallel facts hold in English. In a sentence like His friends like him, and John ‘s friends too, only the nonsloppy reading is available because the first pronoun fails to c-command the second. This contrasts with a case like Him I would not ask his boss to listen to t, but John, I would, where both readings are available. The topicalized pronoun binds the genitive pronoun and its own trace (at S-Structure), exactly as we claim the n-operator binds the two occurrences of the n-pronoun at LF in (57). Again, this provides strong evidence for the presence of such an operator. 3.4 Subjunctive complements We now briefly consider the subjunctive facts (19c). Although we do not have a full analysis of these facts, they seem to be consistent with and support some of our assumptions. In the configuration (58) (58)

…NP*…[CP[IPO/n…V(Subj)…]]

an O-pronoun in subject position of a subjunctive clause must be coindexed with some [− n] NP* (lexical NP or O-pronoun) in the immediately superordinate clause. An n-pronoun must be disjoint in reference from the same [− n] NP* in this context. The n-pronoun must be (and the O-pronoun cannot be) taken as coindexed to this NP* when it is an npronoun. Let us examine where our analysis takes us. The behavior of the O- and n-pronouns here recalls that of pronouns in subject position in subjunctive complements in Romance languages: an overt pronoun must be disjoint in reference from some c-commanding NP (Jeani veut qu’ilj, *i vienne “Jean wants that he come”), and coreference can only be expressed by using a control structure (Jean veut venir “Jean wants to come”). Abe does not have complement control structures with infinitivals as French does (but compare the logophoric contexts discussed in section 4). Rather, obligatory control is expressed by using subjunctive subordinate clauses. The analogy with control is reinforced by the fact that obligatory coindexing or contraindexing of the subject of the embedded clause holds with respect to the NP corresponding to the controller in a French obligatory control structure: NP* in (58) corresponds to the controller. If this is a structure of obligatory


control, it is probably more accurate to talk about the subject of the embedded clause as being bound by the controller rather than as being (obligatorily) coreferential with it. However, it is not really possible to totally assimilate these structures to obligatory control structures: coindexing between NP* and the embedded subject is not obligatory. For example, the embedded subject can be a name (see (7)). A more accurate generalization would be (59): (59) If the subject of the subjunctive clause can be bound by NP*, it must be bound by NP*. With this assumption, some of the facts follow. If the embedded subject is O, it will be bound by NP* if NP* is [− n]. Otherwise, if NP* is n, binding is impossible. Furthermore, coreference is also excluded, since n ccommands O. So O cannot be coindexed with NP*=n. If the embedded subject is an n-pronoun, binding of this pronoun by NP* is impossible unless NP* is itself an n-pronoun. But suppose NP* is [− n]. Why is coreference between n and NP* excluded? Some alternatives come to mind. The first would be to interpret (59) more broadly, as in Reinhart (1983): since the subject is an n-pronoun and the intended reading involves coreference with the matrix subject, use the O-pronoun instead since it expresses the same meaning but can also be bound. According to a second alternative, the behavior of n would follow if the embedded Comp position did not permit the appearance of an noperator. To see this, suppose that the subjunctive Comp is unavailable (Kempschinsky (1986) proposes that the subjunctive Comp is not available for wh-phrases). The noperator binding the embedded n subject will then be one Comp up or higher. This is equivalent to extending the domain D of n (at least) to the higher clause. And in D, n is obligatorily O-free and n-bound (if there is an n-binder). The two approaches make different predictions in a structure like (60): (60)

…NP*…[CP[IP[NP O/nN]…V(Subj) …]

Neither alternative says anything about the O-pronoun. The first deals only with subjects, the second only with the n-pronoun. We therefore expect that the O-pronoun can be coindexed or contraindexed with a [− n] NP*, and this seems to be the case. The first alternative says nothing about the n-pronoun in this case, either, so its options should be the same. The second approach rightly predicts that a genitive n in (60) should behave with respect to NP* exactly as a subject n would, since they have the same domain: n must be contraindexed with a [− n] NP* and coindexed with NP*=n in such configurations. 4 Logophoricity We now turn to the analysis of logophoric pronouns.


A number of West African languages, like Abe, have what are called logophoric pronouns or constructions (for instance, Yoruba (Pulleyblank 1986), Mundang (Hagège 1974), Ewe (Clements 1975), Gokana (Comrie and Hyman 1981): also see the references cited in these works). The puzzling problem of logophoric pronouns can be informally described as follows. There is a special class of verbs that typically, but not exclusively, includes verbs of saying like say and report and verbs of perception like see. Let us call these verbs logophoric verbs. In contexts embedded under a logophoric verb and only in these contexts, a special pronominal form, called the logophoric pronoun, must be used to indicate reference to “the person whose speech, thoughts or perceptions are reported” (Clements 1975). Usually, but not necessarily, the subject of the verb of saying also refers to this person. Following Sells (1987), we will call this person the logocentric referent. The regular third person pronoun, often a weak or clitic pronoun, which otherwise behaves like its English counterpart, must be construed as not referring to the logocentric referent. This is illustrated in the following Yoruba example (from Pulleyblank 1986): (61)

a.

oi he oi he

b.

ri saw ri saw

pe that pe that

oj he ouni he(L)

ni had ni had

owo money owo money

The logophoric verb here is ri “see”, the logophoric pronoun is oun, and the logocentric referent is denoted by the main subject. For ease of reference, the NP denoting the logocentric referent (here it is the subject of the main verb) will be designated NP*. This raises the following questions: (a) What characterizes the set of logophoric verbs? (b) What determines the logocentric referent? and (c) What accounts for the behavior of pronouns in logophoric contexts? 4.1 Logophoric pronouns in Abe and kO-complementation In embedded indicative sentences introduced by ye (ye-complementation), pronouns were shown to have the reference possibilities in (62): (62)

a. b.

NP/Oi…[ye…Oi,j/n(i),j] V(Indic)…] ni…[ye…Oj/ni,(j) V(Indic)…]

ye-complements are not the only type of embedded complements in Abe, however. Another frequently occurring type of embedded complement is introduced by kO: (63)

yapi hE kO Yapi says kO “Yapi says that you are nice.”

f you

ye are

sE nice


We will refer to this type of complementation as kO-complementation. The pronouns in kOcomplements behave quite differently from pronouns in ye-complements, and as the following examples show, they immediately recall the behavior of logophoric pronouns in other West African languages: (64)

a. b. c.

ye sE yapii hE kO Oj/ni,(j) Yapi said kO he is handsome wu Oi/ni,(j) yapii hE kO f Yapi said kO you saw him bO wu ye Oj/ni,(j) yapii hE kO f Yapi said kO you take see ye he “Yapi said you believe that he is handsome.”

ye is

sE handsome

Here the O-pronoun, which as we have seen is the regular pronoun, must be disjoint from the main subject. The n-pronoun is used to express coreference with it. In this sense, the referential n-pronoun is used as a logophoric pronoun. The logocentric referent is the referent of the main subject, noted NP*. Seen from the perspective of languages with logophoric pronouns, this is hardly surprising. In Abe, however, these properties are unexpected. A comparison of (62) and (64) shows that a reversal of the normal properties of O- and n-pronouns is observed in kOcomplements. Whereas the O-pronoun is free in reference in (62), it must be disjoint from NP* in (64). Furthermore, although an embedded n in (62) is preferably taken as free, it is preferably taken as coindexed with NP* in (64). This reversal is only observed with respect to a [− n] antecedent. When the antecedent is an n-pronoun, the data are exactly parallel to the usual pattern when the antecedent is an n-pronoun: (65)

a.

ni he(n) b. ni he(n) c. ni he(n)

hE said hE said hE said

kO kO kO kO kO kO

Oj/ni,(j) he f you f you

ye is wu saw bO wu believe

sE handsome Oj/ni,(j) him ye Oj/ni,(j) ye he

ye sE is handsome

(65) suggests some preliminary observations. The first runs counter to what we have seen of Abe’s pronominal system: O- and n-pronouns in kO-complements behave uniformly, disregarding the NP type of their antecedent. Second, the pattern of behavior is exactly the one we would expect in normal contexts when the antecedent has the feature [+ n]. The analysis developed below exploits this generalization: we will motivate an analysis for logophoric effects in which a [+ n] element is always present. Before we do so, however, let us first determine the exact conditions that trigger logophoricity. A first question concerns the class of logophoric verbs. What makes a verb qualify as a logophoric verb in Abe? So far as we have been able to determine, such verbs in Abe all are verbs of saying. However, not all verbs of saying select a kO-complement. A verb like


ka “tell”, for instance, selects a ye-complement. As the following examples show, pronouns in this complement behave just like pronouns in regular ye-complements, and do not show logophoric effects. This indicates that being a logophoric verb is not (exclusively) a semantic property: (66)

a. b.

yapii Yapi yapii Yapi

ka tell hE said

api Api kO kO

ye Oi,j/n(i),j ye he Oj/ni, (j) ye he is

ye sE is handsome sE handsome

Logophoric effects thus seem to depend on the arbitrary syntactic property of taking a kOcomplement.15 (Clements (1975) makes a similar observation about Ewe.) Second, logophoric effects are not always observed in kO-complements. They occur only in a subset of kO-complements, with respect to a particular argument of the logophoric verb that We will call the designated argument.16 This can be concluded from the following examples, in which the pronouns behave in a normal fashion: (67)

a. b.

kO Oi, j/ni,j m hE apii I said to Api kO she ce kO Oi,j/ni,j ye yapii Yapi heard kO he is

ye sE is handsome sE handsome

In (67a) no logophoric effect is observed with respect to the object of the logophoric verb (here the source in Sells’s sense would be the subject). In (67b) no logophoric effect is observed at all. There is no way in Abe to express the implicit source as an argument of the verb “hear”. If the implicit source is expressed, it is by inserting someone said: k hE “they said”.17 Then, logophoric effects obtain with respect to the subject of the verb “say”, but not with respect to the subject of “hear”: (68)

yapii Yapi

ce heard

apik Api

hE say

kO kO

Oi,j,*k/ni,j,k a s/he

a had

arrived

This phenomenon thus qualifies in all respects as a case of logophoricity. 4.2 Analysis Let us summarize the logophoric effects, noting that they hold everywhere, regardless of how far under a logophoric verb the pronouns are embedded: (69) a. The O-pronoun must be contraindexed with a particular argument (NP*, if expressed) of the logophoric verb. b. n-pronouns may be coindexed with NP* regardless of what NP* is.


We will propose an analysis that leaves our treatment of the pronominal system intact. The different behavior in kO complements will be a consequence of (a) some lexical property related to the particular class of verbs that select a kO-complement and (b) a property of the kO-complement itself. As mentioned earlier, the pronouns’ behavior in logophoric complements would be entirely regular if the antecedent were not a lexical NP but rather an n-pronoun. The following examples—(70a,b) versus (70c)—illustrate the parallelism: (70)

NP/ni … [kO…Oj/ni,(j)…] ] ni…[ye…Oj/ni,(j)…] NP/Oi…[ye…Oi,j/n(i),j…]

a. b. c.

(70a) shows that the NP type of the pronoun’s antecedent does not matter, as long as it is the designated argument. As a first approximation, it appears that the target pronouns behave as though they had an n-pronoun antecedent. Let us therefore assume the following property of verbs selecting for a kO-complement: (71) A verb that selects a kO-complement assigns the feature [+ n] to its designated argument. It can easily be seen how (71) accounts for the data discussed so far: (72)

a.

b.

c.

yapii [+n] Yapi yapii [+n] Yapi yapi [+n] Yapi

hE

kO

Oj/ni

ye

sE

said hE

kO kO

he f

is wu

handsome Oj/ni

said hE

kO kO

you f

saw bO wu

him ye

Oj/ni

ye

sE

said

kO

you

believe

ye

he

is

handsome

If the designated argument is different from the antecedent or not syntactically represented, the pronouns will behave in a regular fashion. We thus obtain the contrast between (72) and (67), repeated here as (73): (73)

a. b.

m I yapii Yapi

hE said

apii to ce heard

kO Api kO kO

Oi,j/ni,j kO Oi,j/ni,j he

ye she ye is

sE is handsome sE handsome

Finally, nothing needs to be said if NP* is an n-pronoun. Although (71) constitutes the core idea of the treatment of logophoric pronouns in Abe, its actual form needs revision. (71) looks suspicious: we have assumed that the feature [+


n] is incompatible with any NP type other than n-pronouns. (71) has as a consequence that all NPs could be assigned a hidden [+ n] feature. In any case, (71) is not descriptively adequate. By virtue of (71), the feature [+ n] is assigned to some argument (NP*) of the verb that selects for kO. This implies that any pronoun c-commanded by NP* should treat NP* as though it were an n-pronoun. In particular, one would expect that pronouns not contained in the kO-complement, but c-commanded by NP*, show the typical logophoric effects. For example, one would expect that an n-genitive pronoun in the following structure would be coreferential with NP*, whereas an O-pronoun would be disjoint from NP*: (74)

a. b.

Prediction: Fact:

NP*i… hE yapii Yapi said

[ni *j, O*i,j…] n*i,j/Oi,j/Oi,j to his

ceewu friend

[kO… [kO… that

As (74b) shows, this prediction is not borne out; the pronouns behave in their usual way with respect to a [− n] antecedent (NP*). Consider (69a) again. Which module of syntax makes reference to a particular argument of a predicate? Reference to a particular argument recalls the kind of condition that we find in control constructions. In control constructions it is often the case that the controlled NP must have as antecedent a particular argument of the control verb (see Jackendoff (1987)). The logophoric effect is in a sense an anticontrol property: the pronoun O must be disjoint from a particular argument of the logophoric verb. We can keep the basic insight of the above analysis and relate it to control theory. The basic insight of the proposed analysis was the covert presence of a [+ n] NP. Suppose now that this covert NP is not NP* but rather an NP that is controlled by NP*. The relation between O and NP* will then in fact be mediated through this covert NP. We will have the following structure: (75)

[NP*…

…kO…O….] media

We know that O cannot be coindexed with [e]. If [e] is controlled by NP*, two properties are explained: first, that concepts relevant to control theory are involved; second, that O cannot be coindexed with NP*. This proposal also naturally covers (69b). If O in (75) is replaced by n, it will follow that n is preferably coindexed with [e], hence with NP*, but also optionally not (provided that [e] and n are not in the same domain D). The next question is, Where is this covert NP? This [+ n] NP, we will now propose, is a silent NP subject of kO. As in many languages, the sentence introducer kO is homophonous with the verb “say” (which is a defective verb). Koopman (1984) argues that a similar sentence-introducing element na “say” in Vata is not a realization of the complementizer node (Comp) but rather a verb heading its own clause and taking a sentential complement in turn. Suppose that this is also true for kO-complements. If kO is the verbal element of a clause, this clause would presumably contain a silent subject NP, given the Extended


Projection Principle. This silent subject would receive as θ-role the θ-role the verb “say” assigns its subject. The semantic import of the verb kO seems totally redundant. We would like to suggest that kO is to the superordinate verb what a cognate object like a dance in dance a dance is to its verb. Since the verbs taking kO are all verbs of saying, it makes kO a hitherto unrecognized type of verbal cognate object. The structure of a clause with a verb that selects for a kO-complement would really be as follows: NP I V [e kO [CP[Comp[IP…]]]]

(76)

If this structure is correct, we can now propose that the NP carrying the n-feature is in fact the subject of the verb kO (or, equivalently, the n-operator introduced at LF to bind this empty [+ n] category, if there is one; see below). To this effect, we modify (69) to (77), with (78) as result: (77)

kO selects an external argument with the feature [+ n].

This silent category being [+ n], it requires the introduction of an n-operator to bind it. Furthermore, the simplest assumption seems to be that this kO-clause is a CP: NP* V [CP[Comp

(78)

[e kO [CP…]]]] [+n]

Some independent evidence for this full clausal structure (and therefore for the presence of a subject) derives from topicalization constructions, and from the wh-type of Vmovement construction (predicate cleft, see Koopman (1984)): a topicalized NP or a topicalized verb can be moved to clause-initial position. If a kO clause is involved, this topicalized element can occur sentence-initially, immediately following kO, or immediately preceding kO: (79)

a.

b. c.

(80)

a.

b. c.

kolo o yapi hE api love-Top Yapi said Api “Yapi told Api that he loves her.” yapi hE api kolo O Yapi told Api love-Top yapi hE api kO kolo O Yapi told Api kO love-Top he love her sakaa yapi hE api rice-Top Yapi told Api “Rice, Yapi told Api that he ate.” yapi hE api sakaa Yapi told Api rice-Top yapi hE api kO saka

kO kO

n he

kolo loves

O her

kO kO

n he n

kolo loves kolo

O her O

kO kO

n he

di ate

kO kO

n he a

di ate ne di


Yapi

told

Api

kO

rice-Top

he

ate

Crucially, examples like (79b) and (80b) are only possible with kO-complements: they are not possible with ye-complements (though both the equivalents of (79a,c) and the equivalents of (80a,c) are): (81)

*

yapi Yapi

ka told

api Api

kolo O love-Top

ye ye

O he

kolo loved

O her

This follows if we treat ye as a regular complementizer and kO as a verb. If kO is a verb in a full clausal structure and is selected by a higher verb, a problem arises. It seems that selectional (and other semanticolexical) restrictions should be checked under government at LF (as suggested for example by the distribution of wh-words in English and Chinese): in order to fulfill this requirement, we are led to assume that the verb kO moves from a position in which it is ungoverned by a verb that selects for kO to a position in which it can be governed by such a verb; that is, it will have to move to the Comp position of its CP in LF. In a real sense, then, it is both a verb and a complementizer.18 4.3 Pronominal binding in logophoric contexts We now turn to a problem related to pronominal binding in logophoric contexts. Consider first nonlogophoric contexts. We have seen that an n-pronoun cannot be bound by anything but another n-pronoun. When coindexing of an n-pronoun with a [− n] NP occurs, this NP cannot be quantificational (a wh-phrase or quantified expression). When this NP is referential, the tests for pronominal binding fail: sloppy identity is never an option. Consider now logophoric contexts. In such contexts an occurrence of n embedded under a logophoric verb is preferably understood as coindexed with NP*. So far, we have treated this as a case of coreference. The problem is that n can also be bound by NP*. This shows up in two ways: (82) a. apOUNi nobody b. yapi hE Yapi said

γe Neg kO kO

hE said ni he

kO kO ye is

Oj/ni ye he is sE, handsome

sE handsome api ese Api also (sloppy/ nonsloppy)

Thus, the reversal of behavior of the n-pronoun observed in logophoric contexts extends to pronominal binding as well. Let us examine the predictions of our proposals case by case. The basic structure to examine obeys the schema (83), (83)

NP*…V…[CP1 OP1

[e kO [CP2 Comp2 […NP**…


[+n] where V is the logophoric verb, NP* its designated argument, kO the verbal complementizer, e its silent subject, which is marked [+ n], Op1 the n-operator that must be introduced to bind this silent [+ n] subject, and NP** a pronoun. By assumption, NP* and e (and consequently Op1) are coindexed: this is the control property of the construction. If NP** is an O-pronoun, it cannot be coindexed either with e or, by transitivity, with NP*, since it would become bound by Op1 .An O-pronoun can neither be coreferential with nor bound by NP*. Suppose next that NP** is an n-pronoun. We now have two options: either introduce an n-operator in the second Comp to bind NP** or not. If we do not, we must coindex NP** with Op1 so as to turn it into a variable: by transitivity, it becomes coindexed with e and NP*. In this case NP** is bound by Op1, which is itself bound by NP* because of the control property of the construction: NP** is bound by NP* and we get sloppy identity in elliptic constructions. The second option consists in introducing a [+ n] operator Op2 in the second Comp to bind NP**. This operator can either be coindexed with NP*/e/Op1 ornot. If it is, the coindexation between NP* and OP2 can be either a case of coreference (the range of OP2 is identical to the reference of e, or of NP*, which of course must be referential) or a case of binding. In the first case we get the nonsloppy identity reading in elliptical constructions. In the second case we get the sloppy identity reading. We thus cover all the cases. Of course, one problem remains: Why can the [+ n] operator Op1 be bound to NP* by control even if NP* is [− n]? Why is the prohibition against binding with mismatching features lifted in logophoric contexts? The only answer we can offer at this point is an analogy with observations made by Montalbetti (1984) concerning Spanish (and also Brazilian Portuguese, Japanese, and Catalan): (84) There is a class C of positions such that a. In some position P of C i. Overt pronouns may freely alternate with covert pronouns. ii. An overt pronoun cannot have a formal variable as antecedent; in other words, a trace generated by Quantifier Raising or Wh Movement of a quantified expression cannot bind an overt pronoun (see (85a)). b. In some position P not in C i. Covert pronouns are not allowed. ii. Overt pronouns may have formal variables as antecedents (see (85b)). Roughly speaking, then, in positions where there is a choice between an overt and a covert pronoun, the overt pronoun cannot have a formal variable as antecedent:19


a.

b.

¿Quién ti cree que él*i/ei who believes that he (Montalbetti (1984:98)) quiere que María ¿Quién ti who wants that Maria “Who wants Maria to talk about him?” Binding OK (Montalbetti (1984:100))

es is

inteligente? intelligent

hable talk

de about

éli/*e? him

Although the Abe and Spanish situations differ, they are formally similar. In Abe there is no real overt/covert alternation. However, there is an O/n alternation. In all but one of the contexts we have described so far, either one of these two pronouns may occur. In contexts where O-pronouns can be bound by some antecedent A, n-pronouns cannot be bound by A. Reciprocally, in contexts where n-pronouns can be bound by some antecedent A, O-pronouns cannot be bound by A. The one exception is the subject position of kO. There, we claim, the silent subject must be marked [+ n], therefore not tolerating any alternation. By analogy with Montalbetti’s observations, we may propose to relativize our earlier claim that two elements must have the same feature specifications if a binding relation holds between them in the following way: feature value identity is required if there is a choice. If there is no choice, this restriction is overridden. Of course, we would not want to extend this to any kind of feature mismatch. In this case it does not seem unreasonable. After all, both n- and O-pronouns are third person pronouns that can be coreferential with each other. 5 Summary and conclusion Let us summarize the main features of our analysis of the referential interactions between pronouns and nominals in Abe. Binding theory accounts only for disjoint reference between pronouns in their GCs; it accounts for none of the complex referential interactions between pronominals and nominals. Central to understanding these is our hypothesis that pronouns of the n-series are logical variables. This implies that in sentences containing n-pronouns an appropriate operator must be introduced to turn the n-pronoun into a logical variable. We thus derive that the domain in which two (or more) n-pronouns are obligatorily coindexed is a domain with a Comp: n-pronouns are obligatorily coindexed when they are dominated by one and at most one Comp node. If more Comp nodes are present, more n-operators can be introduced and, as we have shown, coindexing is no longer obligatory. Furthermore, an O-pronoun that occurs in the scope of a Comp node immediately dominating an n-pronoun must be contraindexed with it, because the O-pronoun cannot be bound by an n-element (coindexing with an Ā;-position or with an n-pronoun (a variable) must be interpreted as binding). We further showed that our analysis extends to the facts concerning pronominal binding and that it is consistent with the particular behavior of pronouns in subject position of subjunctive complements.


It is crucial to our analysis that an n-operator occurs in Comp, so as to convert the npronoun into a logical variable, if the n-pronoun is not already converted into a logical variable. So far we have not specified how such an operator is introduced. One possibility is that it can be freely introduced in any Comp. Prohibition against vacuous quantification will require the presence of an appropriate n-pronoun in its scope. Alternatively, we could assume that the operator-variable relation arises through movement of some npronoun to Comp. Once an n-pronoun has moved to Comp, other n-pronouns can be treated as logical variables. There are no clear indications in favor of a movement analysis (no ECP effects or constraint violation effects). In any case the precise way in which the noperator is introduced does not really affect our analysis, and we will leave the matter open at this point. We started by assimilating the overt n-pronoun to a referentially used pronoun. It seems to be an exact third person counterpart to first and second person pronouns. In a given sentence it identifies particular participants (preferably one) in a discourse situation. This is why, just like first and second person, there is a strong preference (if not a requirement) to take it to have only human referents. Because of its syntactic properties, we paradoxically concluded that it was in fact a logical variable. It appears natural to conclude that its referential qualities in fact reflect those of the operator introduced to bind it. By uniformity, it would also appear natural to conclude that first and second person pronouns should also be treated as logical variables. A particularly interesting feature of Abe’s pronominal system is that the n-pronouns are also used as logophoric pronouns. We described the logophoric effect in detail and motivated a purely syntactic analysis for it. The logophoric effect arise from the particular properties of the n-pronoun, in conjunction with the properties of a particular type of complementation: the complementizer kO is in fact a verb with a [+ n] silent subject. Logophoricity was originally introduced as a descriptive category for the properties of certain pronouns in West African languages, which were deemed to be discourse properties (as the name logo-phoricity suggests). It has since been invoked as an important feature in understanding the properties of Japanese and Icelandic reflexives (see Sells (1987) and references therein), and it has again been argued to require the introduction of discourse considerations in the treatment of phenomena usually treated more syntactically, such as the constraints on referential dependencies between NPs. There is no a priori way to determine in terms of which concepts a given phenomenon should be treated. Logophoricity has often been treated in terms of discourse properties or representations. The Abe system suggests that it should be considered in the realm of syntax, at least as far as referential interactions are concerned. Certain other aspects of the problem do not seem to be syntactic. For example, we have left the meaning of the feature [+ n] open: it seems plausible that discourse considerations play an important role in determining it.


Notes * The Abe data have been collected in Montreal and in the Ivory Coast. Thanks to Philomène Nguessan, Christine Tellier, Jean François Prunet, Tanya Reinhart, Tim Stowell, Richard Kayne, Norbert Hornstein, Joseph Aoun, and two anonymous reviewers. We will transcribe e as ε, as O, and ω as U. 1 The actual condition could be more complex (compare donkey-sentences) in a way that does not affect our discussion. See, for example, Reinhart (1983) and Haïk (1984). 2 Although in English this referential/bound variable distinction is not overtly indicated, in some languages it is. Such is the case of Fula. In Fula two sets of [+ human] third person pronouns occur: the dum-series and the mO-series (Arnott 1970). (We abstract away from a third type of pronoun: oon “the other”.) Their relevant properties can be summarized as follows (for more details, see Koopman and Sportiche (in preparation)): (a) All obey Principle B of the binding theory: they must be free in their GC (that is, neither bound nor overlapping in reference with a c-commanding NP in their GC). (b) Pronouns of the dumseries are bound pronouns (bound to a c-commanding referential NP, a QP, or a wh-phrase). They cannot be bound by an mO-pronoun. (c) Pronouns of the mO-series are used as referential pronouns. Therefore, they can only be coreferential with other NPs, and in particular, they must be dum-pronoun free, since these pronouns are not referential. 3 In the latter case this situation is comparable to that found in French (for example, Il a levé le bras “He raised his arm”): it seems plausible to assume that the pronoun is syntactically represented, since it is semantically present. The question remains where this pronoun appears structurally: presumably as specifier of NP (see Sportiche (1988)). 4 As we will show, this is not entirely accurate. We will modify this generalization and the consequence for the explanation of (18a) below. 5 Alternatively, the structure is mi [gerund PROj [VP n*i,ke]]. This does not affect the argument. 6 See section 4 for a restatement of this assumption. 7 This is one more reason why the disjoint reference effects discussed in section 3.1.3 cannot be due to binding theory effects. 8 Note that we cannot say that O is referentially dependent upon the n-operator, since this operator is not in an A-position. 9 This restriction recalls the prohibition found in English, for example, that one clause cannot contain two topics. Note that other operators, like wh-operators, may cooccur with noperators in Comp. 10 Note that (40) poses an apparent problem, since it is a weak crossover configuration. Both instances of the n-pronoun will be bound by the operator in Comp and will thus violate the Bijection Principle of Koopman and Sportiche (1982). Sportiche (1983) argues that a distinction should be made among Ā;-binders and that the Bijection Principle should be reformulated accordingly. Certain Ā;-binders (true operators) induce weak crossover effects. Others do not. It suffices to assume that the n-operator is not a true operator in the required sense. Lasnik and Stowell (1987) make a similar proposal and suggest that weak crossover effects are found only with quantificational operators. Since the [+ n] operator is clearly not quantificational, the Abe facts are consistent with their proposal. 11 Sells argues that what we have called coreference is in fact cospecification, so that sloppy identity is available iff the pronoun is a bound variable, nonsloppy identity iff the pronoun is cospecified with its antecedent. See also Fauconnier (1984) for relevant discussion.


12 The subject O-pronoun is a null pronoun. It may seem a priori hard to decide whether the subject position contains a wh-trace or a zero resumptive pronoun. We assume it is a zero resumptive pronoun, because of the data described in section 3.3.2. 13 Nor by a wh-phrase:

(i)

caai f mU who you know “Whose house do you know?”

n*i his

erenyi house

The analysis developed so far already accounts for the impossibility of (i): the npronoun needs to be bound by a [+ n] operator, which the wh-phrase is not. 14 The availability of the sloppy reading, coupled with the requirement on the logical treatment of the n-pronoun, suggests that the formula reconstructed in the second conjunct does not contain the pronoun—that is, that the formula reconstructed is [x believes that x knows Api], with x bound by Yavo. 15 Although it might be the case that belonging to a certain semantic class is a necessary condition for taking a kO-complement. 16 Sells (1987) informally suggests that the role Source (of information) is the relevant notion determining what the designated argument is in African languages: Abe is consistent with this. 17 Indefinite pronouns are rendered as third person plural pronouns. 18 This might relate to the tendency for what appears to be homophony between the verb “say” and complementizers. 19 In Spanish the generalization is more complicated: when there is a choice, the overt pronoun cannot have a formal variable as local antecedent (see Montalbetti (1984)).

References Arnott, D.W. (1970) The Nominal and Verbal Systems of Fula, Oxford University Press, London. Chomsky, N. (1981) Lectures on Government and Binding, Foris, Dordrecht. Chomsky, N. (1986) Knowledge of Language: Its Nature, Origin, and Use, Praeger, New York. Clements, G.N. (1975) “The logophoric pronoun in Ewe: its role in discourse”, Journal of West African Languages 10:141–177. Comrie, B. and L.Hyman (1981) “Logophoric Reference in Gokana”, Journal of African Languages and Linguistics 3:19–37. Evans, G. (1980) “Pronouns”, Linguistic Inquiry 11:337–362. Fauconnier, G. (1984) Espaces mentaux, Editions de Minuit, Paris. Hagège, C. (1974) “Les pronoms logophoriques”, Bulletin de la Société de Linguistique de Paris 1:287– 310. Haïk, I. (1984) “The syntax of operators”, unpublished doctoral dissertation, MIT, Cambridge, Mass. Higginbotham, J. (1983) “Logical form, binding, and nominals”, Linguistic Inquiry 14: 325–420. Jackendoff, R. (1987) “The status of thematic relations in linguistic theory”, Linguistic Inquiry 18: 369–411. Kempschinsky, P. (1986) “Romance subjunctive clauses and Logical form”, unpublished doctoral dissertation, UCLA, Los Angeles, California.


Koopman, H. (1984) The Syntax of Verbs, Foris, Dordrecht. Koopman, H. and D.Sportiche (1982) “Variables and the bijection principle”, The Linguistic Review 2: 139–160 (Chapter 1 in this volume). Koopman, H. and D.Sportiche (forthcoming) The Pronominal System of Fula. Lasnik, H. (1976) “Remarks on coreference”, Linguistic Analysis 2:1–23. Lasnik, H. (1981) “On two recent treatments of disjoint reference”, Journal of Linguistic Research 1 (4): 48–58. Lasnik, H. and T.Stowell (1987) “Weakest crossover”, unpublished manuscript, UCLA, Los Angeles, California, and University of Connecticut, Storrs. Montalbetti, M. (1984) “After binding”, unpublished doctoral dissertation, MIT, Cambridge, Mass. Pulleybank, D. (1986) “Clitics in Yoruba”, in H.Borer (ed.) Syntax and Semantics, vol. 19, The Syntax of Pronominal Clitics, Academic Press, New York. Reinhart, T. (1983) Anaphora and Semantic Interpretation, Croom Helm, London. Sag, I. (1976) “Deletion and logical form”, unpublished doctoral dissertation, MIT, Cambridge, Mass. Sells, P. (1986) “Coreference and bound anaphora: a restatement of the facts”, in Proceedings of the Sixteenth Annual Meeting, NELS, GLSA, University of Massachusetts, Amherst. Sells, P. (1987) “Aspects of logophoricity”, Linguistic Inquiry 18:445–479. Sportiche, D. (1983) “Structural invariance and symmetry in syntax”, Doctoral dissertation, MIT, Cambridge, Mass. Sportiche, D. (1988) “Conditions on silent categories”, unpublished manuscript, UCLA, Los Angeles, California. Stowell, T. (1981) “Origins of phrase structure”, unpublished doctoral dissertation, MIT, Cambridge, Mass. Williams, E. (1976) “Discourse and logical form”, Linguistic Inquiry 8:101–139.

3 THE INTERNAL AND EXTERNAL DISTRIBUTION OF PRONOMINAL DPs

The form and distribution of pronouns varies considerably crosslinguistically. In this chapter, I will propose that there is a direct relation between the form (i.e. DP internal realization), and syntactic distribution (i.e. DP external realization).* Pronominal elements can be homophonous with determiners (French le/la/les, Dutch ‘t,…), or not (English him/her/them, French lui, Dutch ’m…), or they can be composed, as demonstrative pronouns in many African languages, or the Welsh reduplicated pronouns discussed below. Pronominal elements can be overt or covert (e.g. pro). Overt pronominal elements can distribute as DPs (i.e. their distribution is identical to that of (specific) DPs), as clitics (their distribution is that of a head which occurs in a particular syntactic position), or as agreement morphology (i.e. as a head that shows up if an XP is or has moved through its Spec position). The analysis of the syntax of pronominal elements meets with considerable analytical difficulties. Whether a particular pronominal element has the distribution of a DP or not, presupposes an understanding of the distribution of the relevant type of DP. Whether distributional differences can be accounted for by a combination of the distribution of DPs, and heads (i.e. whether a pronoun is a head that has undergone head movement from the DP position) further presupposes an understanding of the distribution of heads. Further questions arise beyond these analytical problems. If the pronominal element is a head, what position is it cliticized to? What distinguishes it from agreement? Why do pronouns, but not DPs, trigger obligatory agreement in some languages? Why do clitics and agreement occur in the particular positions they do? How should pro-drop be analyzed? How are clitic doubling constructions to be analyzed? How should the often different distribution of first and second person pronouns versus third person pronouns be accounted for? Most studies of pronominal elements base their analyses on their distribution in syntactic structures, and pay little or no attention to the actual form of the pronoun.1 This may be because there is not much overt evidence on which to base an analysis, or because there does not seem to be much at stake, or simply because the internal structure of DPs has not been developed enough to allow this question to be asked. I argue in this chapter that a proper understanding of the form and structure of pronominal elements is in fact extremely important: there is an intimate connection between the DP internal syntax and the DP external syntax.

INTERNAL AND EXTERNAL DISTRIBUTION 75

I will start out with an analysis of the internal structure of pronominal DPs, and show that pronominal DPs consist of (at least) two basic elements, NumP and D,2 which can be in an overt agreement relation, with NumP in Spec, DP. Languages vary with respect to overtness or covertness of NumP and of D. This defines a typology of possible pronouns, and has immediate consequences for the external distribution of pronouns. In particular, the structure of pronominal DPs, taken together with movement theory, predicts the existence of NumP movement with D-stranding. NumP movement, as I will show, is the key that leads to understanding many aspects of the distribution of pronominal elements. Since NumP movement plays such an important part in accounting for the syntax of pronominal elements, I motivate its existence carefully. I establish NumP movement and D stranding on the basis of the distribution of pronouns, agreement, and lexical DPs in Welsh, which has a sufficiently rich pronominal syntax to make the empirical argument for NumP movement and D-stranding. The NumP movement analysis straightforwardly accounts for the distribution of pronouns in Mainland Scandinavian (section 5); and can be extended to English in section 6. The analysis proposed in this chapter allows us to reduce the important superficial differences of languages like Welsh and English to identical structures, with differences in what gets lexicalized. 1 The form of the pronoun Pronouns are specific DPs, and should therefore be represented as specific DPs. Following Postal (1969), it is often assumed that pronouns are intransitive Ds (DP → D) (Abney (1987), Cardinalletti (1993), among others). This proposal explains the homophony of pronouns and Ds in some languages, but it fails to capture the fact that pronouns in some sense stand for nouns, encode person and number features, nor does it make explicit what kind of relation there is between D and N. This proposal furthermore cannot accommodate pronominal forms that are not homophonous with Ds, nor compound pronouns, and merely treats the form of the pronoun and the relation to the structure as accidental. There is no reason why pronouns should not fit into the regular DP structure. This is, in fact, the null hypothesis. Recent work on the internal structure of DPs (Abney (1987), Carstens (1991), Ritter (1991), Szabolcsi (1987), Stowell (1991), Valois (1991), among others) has shown that DPs are highly structured projections, similar to clausal projections, with at least the following structure (NumP stands for Number Phrase; AgrP, I assume, is related to the presence of a DP internal subject (i.e. a genitive DP) and is only present when a genitive DP needs to be licensed. I therefore omit AgrP from DPs not containing a genitive DP): (1)

[DP[Spec [D [AgrpAgr [NumP Spec [Num [Np]]]]

The linear order of constituents in the DP is derived by various movements and parallels the way in which the linear order in a clause is derived. Head movement accounts for displaced heads, and DP movement for the position of the structurally Case marked DP


(genitive) in the overt syntax. As I show below, there are additional DP movements, of constituents smaller than DP. Since pronouns are DPs, (1) defines the space for crosslinguistic variation. Pronouns are specific: they stand for a head noun, and they encode person, number and gender features. We therefore minimally expect the structure in (2) (see p. 80).3 It is reasonable to assume that pronouns are lexically marked for number, person and gender features. These features must therefore be checked. In principle this can be achieved in one of two ways: either by head movement to a functional category, or by movement of some projection containing the pronoun to a Spec position where the features of the pronoun can be checked under Spec head agreement, at LF at the latest. If we assume that specificity must also be checked at LF, and is checked by a particular D, we expect the LF representations and the associated derivations presented in (3). Certain movements are incompatible with others: this is indicated with a star in the table below.4 The symbol →indicates those derivations that are relevant for the particular pronouns under scrutiny in this chapter:


I should point out that, for the purposes of this chapter, head movement of the pronoun for checking purposes (3b) plays a limited role. Most of the work will be achieved by Spec head agreement (see (4) and (5). For reasons of simplicity, I abstract away from the internal structure of NumP in the remainder of this chapter, and simplify the structure to (6). The overt syntax can reflect the various stages between (2) and (4) and (5), depending on whether movement is overt or not, how much movement is overt and what elements are lexical. Given the background so far, I now turn to various pronominal systems. In English, I assume, and motivate below, NP moves to Spec NumP, and NumP to Spec DP in the overt syntax.5,6 NumP agrees with an empty D. The obligatory absence of D parallels the obligatory absence of C when Spec, CP is filled (compare Koopman (1991), Sportiche (1995)) (see 7).


In this structure, NumP, an element smaller than DP, moves to Spec, DP. Justification for NumP movement comes from other languages. Nkemnji (1993) for example, provides quite convincing arguments in favor of overt NumP movement, based on the derivation of DP


internal word orders and agreement patterns within the DP for Nweh, a Grassfield language spoken in Cameroon.7 I argue in Koopman (1994) that NumP movement to Spec, DP actually underlies a necessary LF representation which permits further movement of the head N to check selectional restrictions with the theta-role assigner. In short, languages differ whether NumP movement takes place in the overt syntax or at LF. In addition, in English, a pronominal NumP must move in the overt syntax, but a lexical NumP does not. As shown in Valois (1991), lexical Ns remain low in the hierarchy, possibly in situ:8 (8)

a. b.

pronominal NumPs moves to Spec, DP in the overt syntax; lexical NumPs move at LF.

This asymmetry is ultimately responsible for the different distributional properties of pronouns and lexical DPs. Pronominal NumP are governed by an external governor. This either allows further NumP movement (4, 5, and 6), or head movement of the pronoun. Lexical NumP are not in a position that allows for further syntactic movement. The lack of (lexical) N incorporation in English syntax is therefore a direct consequence of the overt syntax: a lexical N cannot undergo N incorporation, because it is too low in the structure.9 French pronominal DPs mirror English DPs (abstracting away from the obligatory cliticization of D): a covert (i.e small NumP pro) agrees with an overt D. It is typically the case in French that the head is overt, and the element in the Spec silent: wh-operators in French for example are silent and the head C is overt whenever possible (i.e. when “recoverable”).

Pronominal DPs thus far differ with respect to the internal structure of NumP, whether NumP moves to Spec, DP in the syntax or not, and whether NumP in Spec, DP is overt and D is overt or covert. Restricting attention to languages with overt NumP movement to Spec DP, other combinations of overt/covert are expected to occur. For example, a pronominal DP with both NumP and D silent is a likely candidate for the representation of pro:


Small pro is further constrained by the need to be identified: in (10) covert NumP pro is not identified, because of the absence of overt features; identification will depend on agreement with some external head. We further might expect a pronominal DP with both NumP and D overt. Welsh reduplicated pronouns fit into this slot, as I will show in the next section.10 2 The pronominal system of Welsh In the pronominal system of Welsh all possibilities in terms of overt/covert NumP and overt/covert D seem to be attested, as I will now show. Welsh pronominal paradigms are quite complex.11 There is a basic distinction between pronouns that can only be used in the absence of agreement (independent pronouns) and pronouns that cooccur with agreement (dependent or auxiliary pronouns). I start the analysis of the Welsh pronominal system with the paradigm of reduplicated pronouns, which belong to the series of so-called independent pronouns.12,13 (11)

Reduplicated pronouns (Williams 1980: 46) 1. myfi 2. tydi 3. (efe, efo, fe/fo (masc.)) hyhi (fem.)

nyni chwychwi hwynthwy

Third person masculine pronouns are between parentheses. They are identical to the auxiliary or affixed pronouns discussed below (except for efe which only occurs as a reduplicated pronoun).14 The second part of the reduplicated pronoun is italicized to draw attention to the (significant) homophony of this part and the so-called auxiliary pronouns presented in (12). Auxiliary or affixed pronouns cooccur with agreement. (12)

Auxiliary or affixed pronouns (Williams 1980: 48) 1. i/fi ni 2. di,ti chwi 3. (masc.) ef/efo hwy, hwynt fo/fe (fem.) hi


As before, auxiliary forms that appear as the second part of a reduplicated pronoun are in italics. Reduplicated pronouns are clearly composed of some pronominal element followed by an auxiliary or affixed pronoun, encoding person, number and gender as well. The similarity between auxiliary pronouns and the second part of reduplicated pronouns in fact seems to extend beyond their segmental forms. Phonology also seems to treat them in similar ways. The italicized forms in the reduplicated paradigm are accented, which is unusual since the accent usually falls on the penultimate syllable. Auxiliary pronouns can be accented as well.15 Suppose that the first part of the reduplicated pronoun is a NumP, and that the auxiliary pronoun is (some kind of) a D agreeing with the NumP. This would allow auxiliary pronouns and reduplicated pronouns to fit directly into the proposed structure for pronominal DPs, and describe their difference. They only differ with respect to overtness or covertness of NumP.16

The first part of the reduplicated pronoun should thus be analyzed as NumP. At this point, a comparison between the reduplicated independent (11), the (“short”) independent (14) and auxiliary pronouns (12) becomes important (as before, the forms in italics are auxiliary pronouns, i.e. the forms we have analyzed as agreeing Ds): (14)

Simple independent personal pronouns: 1. mi, fi

ni


2. 3.

ti, di ef (masc.) hi (fem.)

chwi hwy, hwynt

The non-italicized elements my, mi and ty, ti, and hwynt can unambiguously be analyzed as NumPs, since they also appear as the first part of the reduplicated pronoun, i.e. as the part that we analyzed as NumP. Welsh independent pronouns thus either have an Englishlike structure, with D covert and NumP overt, or a French-like structure, with an agreeing overt D, and silent NumP:

Besides these forms, there is another series of independent pronouns, called conjunctive17 pronouns: (16)

Singular 1. minnau 2. tithau 3. yntau (masc.) hithau (fem.)

Plural ninnau chwithau hwythau, hwyntau

These pronouns are important in as far as they appear to be composed of an overt NumP, as can be concluded from first person singular and second person singular forms (mi/ti, not fi, di) and some other head (thau). Finally, since Welsh also allows pro in agreement configurations, there are pronominal DPs with both NumP and D silent:


In sum, Welsh pronouns are composed of NumP and D, and all combinations of covert/ overt NumP and covert/overt D are attested: (18)

Welsh pronominal paradigms: NumP D+ = overt, − = silent + − : some independent pronouns (e.g. mi) + + : reduplicated pronouns, conjunctive pronouns (with special D) + + : auxiliary pronouns − − : pro

This analysis of the internal structure of Welsh pronouns becomes extremely important in the following sections. I will propose an account of overt agreement in Welsh, based on the idea that NumP in Spec, DP can undergo further movement to Spec, Agr, and thus trigger agreement, stranding D, causing apparent clitic doubling when D is overt. 3 External syntax Given the structure of pronominal DPs, there is a certain number of possibilities with respect to their external (i.e. syntactic) distribution. The entire pronominal DP could undergo further movement to a DP position (19a). The head of NumP in Spec, DP or D itself can undergo head movement when governed by an appropriate host (19b). A final possibility, given that NumP occurs in Spec, DP, and Spec positions are typically extraction sites, would be for NumP to move from Spec, DP to a further (designated) position, stranding D. The head of this position must be able to license NumP, a constituent that is smaller than a DP. I will assume that Agr, or D (as Sportiche’s (1995a) clitic projection) is such a position (19c):


In the remainder of this chapter, I present strong arguments in support of NumP movement and D-stranding (19c), based on the fact that it will provide a quite simple analysis of Welsh agreement. 4 Welsh agreement As in other Celtic languages,18 there are agreement and non-agreement configurations in Welsh. Agreement is realized in three different ways depending on the syntactic environment: as an inflectional suffix (subject verb agreement, and prepositional


agreement), as a prefix in nominal environments, and as an infix (representing the object between C and V in CVSO structures in highly formal Welsh).19 Agreement is asymmetric: it is obligatory with pronominals, whether overt or covert, but excluded with lexical DPs. Pro is licensed only with overt agreement. The following examples illustrate the different environments and their particular properties. A summary is presented in (20) on p. 90: (21) a.

Agorodd y dynion y drws opened-3S the men the door “The men opened the door.” b. y dynion a ddarllenodd/*ddralenasant y the men ptread-3S /*read-3Pl the “The men who read the book.” c. Agoront (hwy) y drws opened-3P (they) the door “They opened the door.”

(Sadler 1988:51) llyfr book (Sadler 1988:52)

(Sadler: 51)

As these examples show, Welsh finite clauses exhibit VSO order in finite clauses. Finite verbs come in either analytic or synthetic forms. The analytic form encodes information about the tense and mood of the clause, but is invariantly inflected as third person singular. The synthetic form encodes tense and mood and is fully inflected for person or number. The analytic form must be used with a lexical DP subject, as in (21a), or with a wh-trace, as in (21b). The synthetic form, fully inflected for person and number, must occur with pronouns, whether overt or covert. Thus, verbs fully agree with pronominal subjects, but not with lexical DPs. Since third person agreement on Vs does not encode gender difference, it cannot be seen if lexical DPs trigger gender agreement as is the case in Arabic. Welsh has a class of inflected Ps. A representative agreement pattern for P is given in (22) and (23) for the P ar “on”: (22)

a.

ar on

(23)

Singular 1S 2S 3SM 3SF

y the

plant children

arnaf (i) arnat (ti) arno (fe/fo) arni (hi)

b.

*

Plural arnon (ni) arnoch (chwi) arnynt

arno on-3S

y the

plant children

(Hendrick 1988:38)

An inflected P does not agree with a lexical DP (22a). Ps differ from finite V which must carry third person singular agreement with a lexical DP subject. As (22b) shows, this is impossible with Ps. Inflected Ps further distinguish themselves from inflected verbs in overtly encoding gender agreement for third person singular pronouns (arno fe “on3SM him” and arni hi


“on3SF her”). Gender agreement shows up in nominal environments as well: ei (3SM) causes lenition.20 Agreement shows up as a clitic-like prefix, in the case of genitive agreement in DPs (24), and object agreement with nonfinite verbs (socalled verbal nouns) (25): (24)

a.

c. e.

(25)

a.

car Aled car Aled “Aled’s car” ei gar (hi) 3SM car Aled ei car (hi) 3SF car (her)

Gwnaeth Emrys werthu did-3S Emrys sell “Emrys sold the book.” *Gwneath Emrys werthu did-3S Emrys sell Gwneath Emrys ei did-3S Emrys 3SM “Emrys sold it.”

b. c.

b.

*

ei gar Aled 3S car Aled

d.

*

car hi car her

y the

llyfr book

ef it werthu sell

(Sadler 1988:73–74)

(ef) it

The agreement pattern is identical to the one discussed so far: agreement is impossible with lexical DPs, but obligatory in pro-drop structures or structures with an overt pronoun. As with Ps, no third person agreement marker appears on N when its subject is a lexical DP. In highly literary Welsh, an infixed agreement form may optionally appear in front of the V when there is a so-called overt presentational suffix (C): (26)

a.

b.

Fe’th welodd pt-2S saw-3S “He/she saw you.” Fe’th welodd pt-2S saw-3S

di you

(Sadler 1988 (20): 75)

(Sadler (1988: (21): 75)

The presence of agreement licenses pro. If agreement is absent, pro is excluded: (27)

Fe pt

welodd saw-3S

*(di) you

Pronouns cooccurring with agreement are called “affixed” or “auxiliary” pronouns. These pronouns we have analyzed as agreeing Ds: (28)

Auxiliary pronouns cooccur with agreement.


Pronouns that occur in configurations without agreement, most notably object pronouns in sentences with VSO order, or complements of Ps that lack an inflectional paradigm, are called “independent” pronouns. These pronouns cannot cooccur with overt agreement: (29)

Independent pronouns cannot occur with agreement.

In verb initial VSO structures, object agreement is impossible in all varieties of Welsh: the independent form of the pronoun must be used, and pro is disallowed: (30)

a.

b. c.

Glywoch chi hear-PAST-2Pl you Sion/him “You heard Sion/him.” *Glywoch chi “You heard him.” Glywoch hear-PAST-2PL “You heard me.”

Sion/ef

chi you

mi me

As (30b) shows, pro is licensed only if there is overt agreement. (20) summarizes the distribution of lexical DPs, agreement, and auxiliary pronouns: (20) (a) overt pronouns and pro obligatorily trigger agreement in agreement environments; (b) auxiliary pronouns cooccur with agreement; (c) independent pronouns cannot occur with agreement; (d) lexical DPs do not trigger agreement; (e) pro is licensed iff there is overt agreement; (f) finite verb forms must carry agreement, either full agreement with a pronoun or third person agreement; other lexical categories have nonagreeing forms. (g) Finite verb forms do not agree in gender; P and N do. In this chapter, I present an account for the properties in (20a-e). I will not address the interesting problem of the source of third person agreement on tensed verbs in (20f),21 nor will I attempt to say anything about the problem of the distribution of a, wh-traces and the resumptive pronoun strategy (see Sadler (1988), Rouveret (1990) among others). 4 The analysis of Welsh agreement Agreement in Welsh is asymmetric: only pronouns trigger agreement, but lexical DPs do not. This suggests that overt agreement reflects the overt syntax, and therefore yields insight into the distribution of different elements in the overt syntax:


(31)

Agreement (in Welsh) reflects the configuration in the overt syntax.

If agreement is triggered in Spec, Agr in the overt syntax, Spec, Agr contains an overt agreement element in the overt syntax with pro and pronouns, but not with lexical DPs. Thus: (32) a. Agr is triggered by an overt pronominal element in Spec, Agr; b. there is no agreement triggering element in Spec, Agr in the overt syntax with lexical DPs. This simple account of the agreement asymmetries will receive strong support if it can be shown that lexical DPs and pronominal elements have different distributions, i.e. if lexical DPs do not occur in Spec, Agr, but pronominal elements do. What pronominal category could actually be occupying Spec, Agr? There are two possibilities: either the entire pronominal DP moves to Spec, Agr, as in Koopman (1989) or part of the DP (NumP) occurs in Spec, Agr, with D stranded in a DP position. I will now motivate the latter. 4.1 Are pronominal DPs in Spec, Agr? Analytical difficulties Suppose first that pronominal DPs, but not lexical DPs, move to Spec, Agr positions in the overt syntax. If this is correct, overt pronouns and lexical DPs should occupy different positions in the overt syntax. Is there independent empirical support for a different distribution of pronominal and lexical DPs? In this section, I outline the analytical difficulties that one encounters in trying to argue for pronominal DP movement. These analytical problems are actually quite representative of the kind of problems one runs into doing contemporary syntactic analysis. It is difficult to determine where a head is in the overt syntax, and where lexical DPs occur. There are many configurations in which pronominal DPs could very well occur in higher positions than lexical DPs, but which do not allow us to establish that this must be the case. I will briefly discuss these configurations here, and discuss some data in Welsh and other Celtic languages, which suggest that pronominal DPs are higher. Through this discussion, I also illustrate why the argument is difficult to make. Consider the following configuration, in which both a pronominal DP and a lexical DP raise, but in which the pronoun raises to a Spec position that is higher than the lexical DP. Suppose furthermore that the head does not end up in the position between the pronoun and the lexical DP, but raises to a position which is still higher than the pronoun:22 (33)

X [pronoun [e [lexical DP […

Many configurations in Welsh fit the schema in (33). Consider subjects for example. Lexical subject DPs must raise in Welsh, as argued in Koopman and Sportiche (1991:234– 235) since they precede sentence medial negation. But pronouns could still raise higher to


an Agr triggering position. This is only possible if V moved higher than Agr, possibly to C, as is often assumed for Celtic languages.23 Basically, then, some modifier must be found which occurs between the pronoun and the lexical DP. But even so, the different positioning of the pronoun could be due to head movement of the pronoun, and the argument cannot be made tightly. Awberry (1990) presents evidence based on the distribution of subjects with sentence medial negation in Welsh dialects which suggests that pronouns must always raise. In these dialects, pronouns must precede sentence medial negation, definite DPs may precede or follow negation,24 and indefinite DPs follow negation. However, this paradigm only shows that pronouns must raise. It does not show that pronouns must raise to a different position than DP subjects which precede medial negation. Other Celtic languages have phenomena that suggest that pronominals occupy different positions than lexical DPs. The problem here is that the distribution might be due to head movement of the pronoun. There is a paradigm in Breton for example which suggests that pronominal DPs move to Agr triggering positions which lexical DPs cannot reach. While Breton disallows overt pronouns with agreement,25 the Welsh pattern surfaces with emphatic pronouns, which obligatorily trigger agreement. The following distribution of DPs, agreement, and emphatic pronouns holds in negative sentences in Breton, which involve a negative C (ne) and a sentence medial negation (ket): (34)

a.

b.

c.

d.

Ne gousk ket ar PCL sleep not the “The boys do not sleep.” Ne gouskont ket PCL sleep3P not “They do not sleep.” Ne gouskont-int ket PCL sleep3P-they not “They do not sleep.” *Ne gousk-int ket PCL sleep-they not

baotred boys

(Stump 1984: (35a))

(Stump 1984: (34a))

(Stump 1984: (35b))

Lexical DP subjects follow medial negation ket (34a), and are thus quite low. Pronouns trigger agreement (34b). Importantly, emphatic pronouns precede negation (34c) while obligatorily triggering agreement (34d). This seems quite solid evidence that pronouns raise to a higher position than lexical DPs. The emphatic pronoun could have cliticized onto V, however, and could have moved with V.26 Two other potential arguments could be made for Irish, based on the distribution of object pronouns in tensed clauses and the position of overt subject pronouns. Irish pronominal objects tend to occur in clause final position. Chung and McCloskey (1987) analyse this positioning as resulting from a rule of pronoun postposing, which postposes only pronouns to the nonargument maximal projection that most immediately dominates them. There are two surprising facts about this rule: it applies only to pronominal DPs, and it shifts a pronoun to the right. If pronominal DPs must occur in Spec, Agr positions, but


Celtic lexical DPs cannot, we can assume that the pronoun postposing is actually an instance of the same process: the pronoun moves to a Spec position which is only available in tensed sentences, but to a final Spec position. Although these data are highly suggestive, I do not know why this Spec position would be final, whereas other Spec positions are initial, nor how to motivate its final position through obligatory movement of the complement of AgrO for example. Chung and McCloskey (1987) further note that certain parenthetical or interjective elements can intervene between the finite verb and the lexical subject, but not between the finite verb and a pronominal subject: (35)

a.

Tá arndóigh, saighdiúirú be(pres) of-course soldiers “There are, of course, soldiers on the road.” (Chung and McCloskey 1987: (133a)) b. *Tá arndóigh, siad ar be(pres) of-course they on “They are, of course, on the road.” (Chung and McCloskey 1987: (134a))

ar on

on the

bhelach road

on the

bhelach road

They argue that the pronoun is a phonological clitic. However, if pronouns must move to some Spec position higher than lexical DPs, the pronouns in (35) could actually be in a position higher than parentheticals, which in turn are higher than lexical subject DPs. The problem again is that the pronoun could have cliticized from the surface position of the lexical DP onto the inflected verb, and could have pied-piped with the V to the higher head position. This analysis has some appeal for Irish because the finite verb in these configurations does not agree with the pronoun. These examples illustrate some of the problems in determining where pronouns occur: either it cannot be ruled out that the head has moved to some head position higher than Agr, or a different distribution of pronouns could be the result of head movement from the position in which regular DPs occur. The question thus arises if it can be shown where pronominal DPs occur, and whether our hypothesis is actually testable. 4.2 Pronouns and agreement in DPs: pronominal DPs are not in Spec, Agr It can be shown through the structure of Welsh DPs that pronominal DPs have the same distribution as lexical DPs, and that lexical DPs do not occur in Spec, Agr (thus confirming (32b)). This argument requires some preliminary discussion and analysis of the internal structure of the DP in Welsh. In the DP, Agr surfaces as a prefix which can be separated from the head N by certain adjectives and numerals: (36)

ei his

tri three

hen old

lyfr book

(Sadler 1988: (48) 105)


“the three old books” While prenominal adjectives can be argued to form lexical compounds with N (Williams 1980:38), as McCloskey and Hale (1984:524) propose for the prenominal adjectives in Irish, a lexical compound analysis cannot be justified for numerals. Agr is thus an independent head, occurring higher than Num. The headhood of Agr is further confirmed by the fact that Agr can cliticize onto a preceding N. The data in (36) might suggest that the head N is lower than Num. The situation is slightly more complex, however, because the head N precedes adjectives, the subject DP and complements: (37)

a.

hanes bywiog Wyn am story lively Wyn about “Wyn’s lively story about the immigrants” (Awberry 1977:192 (200)) b. llyfr newyydd Dafydd book new David “David’s new book”

yr the

ymfudwyr immigrants

These examples show that subjects of DPs occur quite low in the structure, presumably in situ in Spec, NP. This explains why they precede complements of N. Since they are preceded by N, N must have moved to a position higher than Spec NP, at least to Num. If adjectives are adjoined to NumP, N must have moved to a higher head position, possibly Agr. N would therefore rightadjoin to Agr. This is in accordance with the fact that quite generally adjunction to a head in Welsh seems to be right adjunction, as the syntax of Welsh numerals shows. Numerals do not appear to behave as modifiers, but rather as heads in Welsh. Plural number is realized by an inflectional suffix on Ns. Welsh numerals, however, must be followed by a singular noun (Williams 1980:41). Compare three book in (36) not three books. Thus plural nouns and cardinal numerals are in complementary distribution. This suggests that plural and numerals occupy the same head position (Num). I assume that N with plural-base generated morphology raises to Num, where plural is checked by Num. When Num is occupied by a numeral, the numeral occurs in Num, and there is therefore no way to check plural morphology. The noun therefore cannot carry plural morphology. Since overt numerals (i.e. Num) precede the head N, N rightadjoins to Num.27 Agr occurs to the left of Num, and is therefore higher than Num. I assume that Num, containing Num and N, rightadjoins to Agr. Agr possibly excorporates and moves on to D. Since the lexical subject DP occurs in Spec, NP, and does not occur in Spec, Agr, there is no agreement with lexical DPs.28 Since pronominals obligatorily trigger agreement, there must be a pronominal element in Spec, Agr in the overt syntax. We can now test our hypotheses as to what pronominal element raises, DP or NumP. If the entire pronominal DP raises, overt pronouns should precede adjectives; if it does not, pronominal DPs should occupy the same position as lexical DPs.


(39)

b.

a.

ei hanes (*ef) bywiog ef Agr3 story (*his) lively his “his lively story about the immigrants”

ei llyfr Agr3 book “his new book”

(*ef) (*his)

am about

newyydd new

yr the

ymfudwyr immigrants

ef his

Overt pronominal elements can never precede adjectives, as the examples in (39) show. They therefore occur in exactly the same position as lexical DPs.29 Pronominal DPs are therefore not in Spec, Agr, and we are back to the question of why pronominals trigger obligatory agreement, but not lexical DPs. 4.3 NumP movement and D-stranding Suppose that overt agreement is triggered in Spec, Agr, not by the entire pronominal DP, but by a NumP, with the overt pronoun (the auxiliary pronoun) being a stranded D. If this is correct, (39b) would be assigned the following representation:


NumP moves DP internally to Spec, DP, triggering overt agreement on D (fi). Then, NumP moves on to Spec, Agr, triggering agreement on Agr. (It is very likely, but irrelevant for the present discussion, that fy further moves on to D, thus accounting for the obligatory absence of the determiner when Agr is present.) Note that the form tells us that Agr is overt with NumP silent, rather than NumP overt and Agr silent; f and d characterize first and second person agreement on D, m and t characterize first and second person NumP. Covert NumP thus occurs in Spec, Agr, with Agr overt. This actually always seems to be the case: only covert NumP can occur in Spec, Agr; there do not appear to be forms that are composed of overt NumP and silent Agr or with both NumP and Agr overt.30 I conclude tentatively: (41)

Only covert NumP, pro, can appear in Spec, Agr in the overt syntax.

The situation in (41) therefore slightly differs from that within the DP, where in some cases NumP can be overt. In fact, (41) holds for all Celtic languages: overt Agr cooccurs with Pro, i.e. silent NumP. The agreement asymmetry directly falls out from the different internal structure of pronominal and lexical DPs: only with pronominal DPs can NumP move to Spec, Agr and


trigger agreement. Lexical NumPs cannot escape from the DP simply because N undergoes head movement to Agr: there can therefore be no NumP in Spec Agr with lexical DPs, and lexical DPs therefore do not trigger agreement. The proposed analysis in terms of NumP movement and D-stranding makes specific predictions about the distribution of forms that can or cannot cooccur with Agr. Pronominal paradigms in Welsh can be analyzed in terms of whether NumP is overt or not, and whether D is overt or not ((18), repeated here for convenience): (18) Welsh NumP + + − −

pronominal paradigms: D+= overt,−= silent − : some independent pronouns (e.g. mi) + : reduplicated pronouns, conjunctive pronouns (special D head) + : auxiliary pronouns − : pro

This analysis is exclusively based on the internal syntax of pronominal DPs, i.e. on a comparison of the forms and the different pronominal paradigms. This analysis, however, makes direct predictions about the distribution of NumPs and Ds. Pronominal forms consisting of an overt NumP should never be able to cooccur with Agr, since there simply would be no source for overt Agr. Furthermore, since NumP movement underlies overt Agr, only those elements analyzed as stranded Ds should be able to cooccur with Agr. These predictions are borne out. Pronouns, with overt NumP, like mi (1P) for example, can only occur in environments without Agr: (42)

a. b.

gyda mi with me *arnaf mi on 1PS me

(42b) is ruled out for the simple reason that there is no NumP in Spec, AgrP. (42a) is fine since there is no overt Agr. Reduplicated and conjunctive pronouns contain an overt NumP and should never cooccur with agreement. This again is correct: reduplicated and conjunctive pronouns are defined as independent pronouns, i.e. pronouns that only occur in configurations without agreement. Overt pronominal elements cooccurring with Agr must necessarily be stranded Ds since NumP has moved away. This is correct; only the so-called auxiliary pronouns, i.e. precisely those pronouns analyzed as agreeing Ds, can cooccur with overt Agr. Some independent pronouns are composed of a null NumP and an overt D. These pronouns can be used as independent pronouns, as in (43a), where the pronoun is used as an answer to a question (the representation of (43 a) is given in (43b)):


(43)

a. b.

fi “I/me” [DP [NumPpro[fi] [NumPe]]

Finally, consider the distribution of pro, (i.e. silent NumP), which can be summarized as follows: (44)

DP-internal pro-drop a. [DP[NumP pro[Agr Agr] [NumPe]] b. *[DP[NumPpro[Agre] [NumP e]] DP-external pro-drop c. [[NumPproi] [AgrAgr […[[DP[e]i[DD] d. [[NumPproi] [AgrAgr[…[[DP[e]i[D[e]] e. *[[NumPproi] [Agr[e] […[ [Dp[e]i[DD] f. *[[NumPproi][Agr[e][…[[DP[e]i[D[e]]

With DP internal pro-drop, D must be overt (44a); similarly, with DP external pro-drop Agr must be overt (as (44e) shows it is not sufficient for D to be overt). This paradigm shows that silent NumP must be identified by an overtly agreeing head, or rather that the head of a silent NumP head must be identified by an overtly agreeing head: (45)

The head of a silent NumP chain must be identified by overt agreement.

The analysis is thus extremely simple. Overt agreement is triggered by NumP raising (and when overt agreement is chosen NumP must raise). Lexical DPs do not trigger agreement, because there is no possible NumP that triggers agreement. To conclude this section, let me illustrate the derivation. A sentence like (46a) has the corresponding derivation in (46b): The pronominal DP moves to the position definite DPs must move to in the overt syntax. Within the DP, NumP moves to Spec, DP triggering agreement on D, and further escapes to Spec, AgrP, stranding D. Tensed clauses must have agreement. This forces raising of NumP in the overt syntax. The V itself raises through the relevant intermediate functional categories to T and Agr.


This analysis is compatible with any analysis in which V, or any other lexical category for that matter, never raises higher than Agr. This is consistent with McCloskey’s (1992) arguments that C in Irish lowers to V (i.e. V is not in C). As far as I have been able to determine, similar arguments hold for Welsh. 4.4 Conclusion The analysis of the internal structure of Welsh pronouns and the external distribution of pronouns, agreement, pro and lexical DPs provides strong evidence both for the proposal that pronominal DPs consists of a NumP in Spec, DP and a D which are in an agreement relation, and for the existence of NumP movement to a higher Spec, Agr position where external agreement is triggered. It is precisely the richness of pronominal paradigms, and the occurrence of asymmetric overt agreement, which allowed me to construct an argument for overt agreement triggered through NumP movement. This analysis furthermore presents strong evidence in favour of analyzing agreement as resulting from a overt Spec head relation.


Let me consider more concretely how the generalizations in (20), repeated here for convenience, are accounted for. (20) (a) overt pronouns and pro obligatorily trigger agreement in agreement environments; (b) auxiliary pronouns cooccur with agreement; (c) independent pronouns cannot occur with agreement; (d) lexical DPs do not trigger agreement; (e) pro is licensed iff there is overt agreement. Obligatory (silent) NumP movement explains (20a). With overt agreement, NumP must move to Spec, Agr in the overt syntax. Auxiliary pronouns cooccur with agreement (20b), (i.e. there is clitic doubling with overt pronouns), because they are agreeing Ds, stranded by NumP movement to Spec, Agr. The statement in (20c) has to be adjusted somewhat: those independent pronouns that contain an overt NumP cannot cooccur with agreement. This explains why reduplicated pronouns, conjunctive pronouns, and simple independent pronouns with overt NumP do not cooccur with agreement: NumP is overt within the pronominal DP in all these pronouns, and (sentential) Agr can only cooccur with silent NumP (41). Lexical DPs do not trigger agreement, (20d), simply because they do not occur in Spec, Agr in the overt syntax. Their NumP is contained within the lexical DP (the head N moves to Num, Num and N move to Agr), and therefore there is no possible source for agreement. Agreement asymmetries therefore follow entirely from the structure of the DP in the overt syntax: if NumP is in Spec, DP, further movement is possible. If NumP is not in Spec, DP, as in the case of lexical DPs, further movement is of course impossible. Pro (empty NumP) is licensed iff there is overt agreement (20e). The head of an empty NumP chain must be licensed by overt agreement. Pro thus occurs in Spec, DP licensed by an agreeing D, (this is the case with certain independent pronouns that consist of an agreeing D in contexts without overt agreement), or when licensed by overt agreement. Note furthermore that we have further assumed that the lexical head occurs in Agr in the overt syntax: heads move to Agr, but no higher. Head movement will extend the domain, allowing NumP to reach Spec, Agr. Since heads do not move any higher, however, NumP cannot escape outside of these domains: there is therefore no clitic climbing or cases of multiple agreement beyond those already discussed. The proposed analysis strongly supports agreement as a reflex of an overt Spec head relation. There is a competing theory for agreement, however, which consists of treating agreement as an incorporated pronoun (Anderson (1981), McCloskey and Hale (1984), Taraldsen (1992), Rouveret (1991)). This theory is motivated primarily on the basis of Celtic languages, and aims to capture the complementary distribution of agreement (no full agreement with lexical DPs, full agreement with pronominal DPs); I have offered a different account for this asymmetry which is based on a fundamental asymmetry between the overt distribution of pronouns and lexical DPs in phrases (both clauses and DPs). This asymmetry is in fact widely attested in other languages, and the present theory is


therefore simpler in so far as it makes use only of Spec head agreement and the distribution of pronouns. I cannot actually show that the incorporation theory of agreement is incorrect: this theory would be consistent with the analysis in this chapter, provided that NumP movement to a position from which its head can be incorporated into some higher head position is assumed (i.e. agreement in DPs is much too high for direct incorporation). Proponents of the incorporation theory could capitalize on the fact that NumP triggering agreement must obligatorily be empty: if the head of NumP incorporated this would follow. However, if this is correct, there would be several ways of triggering agreement available in Universal Grammar, surely not an attractive hypothesis from the point of view of the language learner.31 4.5 NumP movement in Irish and other Celtic languages The NumP movement analysis quite straightforwardly extends to Irish and other Celtic languages: NumP movement must occur in configurations with overt agreement, but cannot occur in the absence of agreement.32 Thus, full pronouns must be used in nonagreement contexts. As McCloskey and Hale (1984) discuss, Irish verbal paradigms are defective and do not show full agreement for all forms. When an analytic verb form exists, or is chosen, full pronouns must be used, i.e. there can be no NumP movement, because silent NumP is not identified by an agreeing head. Although Irish differs from Welsh and lacks pronominal forms consisting of a NumP and an agreeing D, there are other forms which consist of a pronoun, i.e. NumP, preceding a particular head (reflexive or emphatic pronouns, demonstratives,33 and contrastively stressed pronouns (McCloskey and Hale 1984:493– 494). Only the contrastive suffix overtly agrees with NumP in its Spec. These heads can be stranded by NumP movement as well, in configurations with overt agreement (compare McCloskey and Hale (1984)). Note that in nonagreement contexts, the NumPs precede these heads, as expected: since NumP is in the Spec of the particular head, it has precisely reached the position from where it could undergo further movement. As in Welsh, the stranded head occupies the same position as lexical DPs,34 as the internal structure of DPs allows us to conclude: (47) a.

a shaol suarach féin MRS life wretched REFLEX “his own wretched life” (McCloskey and Hale 1984: (63b) 514) b. ar soal stoirneach-inne P1 life stormy CONTR P1 “our stormy life” (McCloskey and Hale 1984: (65c) 515)

Quite generally, lexical DPs cannot trigger agreement, for the same reasons as in Welsh: lexical NumPs are contained within Spec, DP in the syntax. They therefore cannot move to Spec, Agr. There is no possible source for overt AGR.


5 Mainland Scandinavian NumP movement So far, Celtic languages illustrate silent NumP movement to Spec, Agr, with overt Agr. The NumP movement analysis also straightforwardly accounts for the distribution of pronouns in Mainland Scandinavian languages, as I will now show. As is well-known (Holmberg (1986) among others), pronouns in the Mainland Scandinavian languages move to positions where lexical DPs cannot occur. Pronoun shift depends on Vmovement, in the sense that the pronoun never moves to a position outside the domain of a verbal chain. This is generally taken to indicate that V-movement extends the domain in which movement is possible. Because pronouns fail to license parasitic gaps, pronoun shift is generally assumed to be movement to an A-position. I only analyze Norwegian pronouns in this section, although I believe that the results extend to other Mainland Scandinavian languages, as well. Let us start with an analysis of Norwegian pronouns, based on Hestvik (1990). Hestvik argues that English pronouns and Norwegian pronouns, which are neither homophonous with Ds nor transparently composed elements, have different representations: English pronouns are undecomposable NPs (DPs in more modern terminology), and cannot be modified (*he with long hair, *she from upstairs). Norwegian pronouns are Ns and behave as Ns in the sense that they can be modified. The equivalent of he with long hair is fine in Norwegian. Translating Hestvik into our structures for pronouns, pronouns are either N or NumP, depending on whether the pronoun can be modified by a PP or not. Treating English pronouns as undecomposable NumPs and Norwegian pronouns as decomposable into N and Num, is rather unsatisfactory, however. After all, the NumP projection expresses the idea that there is both a Num head and a N head. English pronouns should therefore be treated as being composed of N and Num (and person) as well. The question thus arises how to account for the possibility/ impossibility for a pronoun to be modified. Let us maintain the account that English pronouns start out as N as well, and explore a different way to account for the possibility of pronominal modification. Assume that it follows from the particular overt syntax of the DP. Within the proposed pronominal structure, there are two likely candidates: PP modification is impossible if N movement occurs, or, PP modification is impossible when NumP movement occurs. The former will not explain the impossibility for modification. First, in comparable configurations V movement seems perfectly possible. And second, in languages with head movement of N, like French for example, N can move and still be modified by a PP modifier. (48)

l‘ami italien de Jean aux the friend Italian of John with “John’s Italian friend with long hair”

cheveux hair

longs long

The head N has raised out of the NP to a position higher than the position in which the possessor occurs, and the adjective. Yet PP modifiers can occur. If the presence of a PP modifier does not depend on N movement, could it depend on overt movement of NumP? For this option to be right, it must be shown that PP


modifiers may occur when NumP does not move in the overt syntax, as is the case with regular English DPs, and that PP modifiers are excluded if NumP moves to Spec, DP, as with English pronouns. The possibility of NP modification by a PP is not restricted to pronouns crosslinguistically, but actually also shows up with lexical Ns. While Ns in most IndoEuropean languages can be modified by PPs, Ns in many other languages cannot (Abe (Kwa) (Koopman, fieldnotes), Bambara (Mande) (Crook 1990) Nweh (Nkemnji 1993) and Turkish (Kural, personal communication)). In these languages, PP modifiers occur with verbal projections, but not with nominal projections35 (the equivalent of PP modification is expressed by a relative clause). Why then are PP modifiers allowed in certain languages, but not in others? Languages that do not tolerate PP modifiers of NP appear to be D final. Often D agrees with the head N. Since many of these languages are otherwise head initial, this suggests that a projection of N has moved to Spec, DP, triggering agreement on D, as Nkemnji (1995) argues for Nweh (Grassfield Bantu). If this is correct, the structure is roughly identical to the one proposed for English pronouns, but with lexical NumP movement:

Given these observations, we can now assume the descriptive generalizations in (50): (50)

a. b.

PP modifiers are not allowed if NumP moves to Spec, DP. PP modifiers are allowed if N moves to Num or a higher head, or if N is in situ.

Although it is unclear at this point what explains (50a), I conclude that PP modifiers cannot occur with English pronouns, not because they are unanalyzable NumPs, but because NumP obligatorily moves to Spec, DP. The impossibility of PP modification thus provides an additional argument for NumP movement in English pronominal DPs. Since Norwegian pronouns can be modified, NumP has not moved to Spec, DP when the pronoun is modified. This raises the question if NumP ever moves in Norwegian. I assume it does, because of the fact that NumP can undergo movement to a higher Spec position, yielding pronoun shift: this indicates DP internal NumP movement. Norwegian pronouns are thus analyzed as consisting of a silent D and an overt NumP, with optional DP internal movement of NumP to Spec, DP. However, when the pronoun is modified, it must be the case that overt NumP movement to Spec, DP has not taken place. Since


NumP is overt, overt NumP should be able to move further to Spec, Agr positions, as in Welsh, but now with Agr silent: this, I assume is precisely what happens with pronoun shift. NumP movement should furthermore only be possible when NumP moves to Spec, DP in the overt syntax. In particular, if NumP does not or cannot move to Spec, DP, the pronominal DPs should have exactly the same distribution as lexical DPs. These predictions are borne out. Although pronoun shift is strongly preferred, it is not obligatory (it is near obligatory). This immediately follows from the optionality of NumP to Spec, DP movement in the overt syntax. If NumP does not move to Spec, DP, a pronominal DP will have the same distribution as a lexical DP, because NumP will be unable to move to Spec, Agr in the overt syntax. This yields the effect of optionality of pronoun shift, which thus in fact reduces to the distribution of constituents within the DP. If NumP does move to Spec, DP, it will be able to undergo further movement and move to Spec, Agr, thus yielding object shift.36 Pronoun shift should be impossible in those cases where NumP movement to Spec, DP is impossible, i.e. with pronouns that are modified by PPs. Holmberg (1986) notes (but does not explain) the fact that pronouns that are modified by PPs cannot undergo object shift. The following paradigm shows this for Norwegian.37,38 (51)

a. b.

c. d.

* Jeg I * Jeg I

Jeg liker ham ikke I like him not Jeg liker ikke han med I like not him with “*I do not like him with long hair.”39 liker like liker like

han him han him

med with ikke not

langt long med with

langt long

haring hair langt long

haring hair

ikke not haring hair

(51a) involves DP internal NumP movement to Spec, DP followed by NumP movement to Spec, Agr. (51b) shows that a pronoun with a PP modifier has the distribution of a lexical DP: it cannot undergo pronoun shift (51c), nor, quite surprisingly, can the pronoun shift stranding the modified PP (51d). This paradigm is immediately explained: when a NumP is modified it is prohibited from moving to Spec, DP (50); (51c) is therefore excluded in the same way as shifting of a lexical DP, i.e. NumP cannot reach Spec, AgrO; (51d) is excluded in the same way: NumP is unable to reach Spec, DP in the first place, and movement to Spec, Agr is therefore impossible.40 A final property of object shift is that it fails to license parasitic gaps. Rather than attributing this to the A or A-bar status of the landing site, as is common practice, this could be due instead to the fact that the moved category is too small, i.e. NumP, not DP, to license a parasitic gap.


6 English pronouns Although the distribution of lexical DPs and pronouns in English overlaps to a large extent, their distribution differs in a number of constructions. As I will now show, the distribution of English pronouns also reduces the movement of NumP to some designated position (which I assume is Spec, Agr, as in Welsh and in Norwegian). Consider first, however, what to expect given the particular structure of the English pronominal DP. English pronouns are composed of overt NumP and silent D (7). If there is further NumP to Spec, Agr movement, as in Welsh and Norwegian, overt NumP should occur in Spec, Agr. Moreover, given the fact that NumP is overt, (and not silent as in Welsh), we expect Agr to be silent, given the fact that English seems to mark either Spec or head position, but not both (i.e. the doubly filled COMP filter): 52

a. b.

[NumPpronoun [Agre]…(English, Norwegian…) [NumP [e] [Agr Agr]…(Welsh, Irish…)

If this is correct, English is not a pro-drop language, simply because NumP is overt not silent. This does not explain why NumP cannot also be covert in configurations that do show agreement. Suppose however that pro (i.e. covert NumP) is only licensed if the head of NumP is in a Spec head relation with overt Agr (or rather with the head of an overt Agr chain), as in Welsh. This configuration is arguably never satisfied in English (except for the Spec, CP, and C of relative clauses, where NumP can actually be empty): English has no person agreement other than some restricted subject verb agreement. Agreeing main verbs are always lower in the structure than AgrS; AgrS is therefore always empty at Sstructure, and NumP must therefore be overt. If auxiliaries carrying Agr never occur in AgrS either, but either remain lower, or raise higher (in the case of SAI), overt Agr is actually never in a position to license the head of an empty NumP chain. NumP therefore must always be overt. It follows from (52b) and the discussion above that there will be no analytical help in English from the presence or absence of agreement; nor can we use stranded Ds, because D is silent. Arguments for NumP movement can be constructed, though, on the basis of the syntax of certain QPs containing [all, both] (henceforth allPs), which can be stranded. Consider the following paradigm: 53

a. b. c. d.

all of these people/all of them came all these people/*all they came I saw them all/*these people all I spoke to them all/*to these people all

All can be followed by an of-DP (53a) (which has no bearing on our discussion), or by a bare DP (53b), which is the configuration that interests us. When all is followed by an ofDP, pronouns and lexical DPs distribute in exactly the same way as they do in PPs. When all is followed by a bare DP, as in (53b), pronouns cannot follow all but must precede


(53c).The distribution of pronouns and lexical DPs is thus asymmetric: pronouns must precede all, but lexical DPs cannot (53c, 53d). Based on this paradigm Maling (1976) proposes that pronouns and certain quantifiers obligatorily undergo a rule of Quantifier Pronoun flip (Q-flip). How should Q-flip be analyzed? What position does the pronoun occupy? There are two possibilities: the pronoun could be either within the projection containing all, call it allP, as (53d) seems to suggests, or it could be outside the allP. Suppose first that pronouns are within the allP, and occupy Spec, allP. The problem arises why this position is restricted to pronouns: Spec, allP does not really seem to be restricted to pronouns, in the sense that DPs can move through this projection in the case of Q-float, Sportiche (1988). In Celtic languages, lexical DPs can never move through Spec, Agr positions, as can be concluded from the fact that full agreement is impossible in case of wh-extraction. Suppose however that, notwithstanding the appearances, the pronoun is not within the allP, but rather in Spec, Agr. More precisely, suppose that the situation is exactly as in Welsh, with NumP moving to Spec, Agr. That this is the correct analysis is what I will argue in the remainder of this section. Let us first consider support for analyzing the pronoun as being outside the allP. (54)

a. b.

(55)

Q: A:

56

Q: A:

All these candies I really like. *Them all I really like.

Which ones do you want, all the yellow candies, or all the green ones? a. all the candies b. all of them c. *them all d. I want them all. Which book do you want? This one or that one? a. both the books b. both of them c. *them both d. I want them both.

While the order all DP can be used as an independent DP constituent, the order pronoun Q cannot.41 This constitutes strong evidence that pronouns are not within the allP, but actually have raised outside. (57) thus follows: (57)

Pronouns must raise outside the allP.

The problem now arises how to reconcile (57) with (53d). I will assume that the pronoun in (53d) has raised to Spec, AgrP, with the lexical DP in a lower position, and P raising to some projection higher than the pronoun. (This would be possible, if one assumes PP shells, as in Koopman (1991).) If pronouns are not within the allP, where are they? In which position do pronouns occur? Insight into this problem comes from the environments that allow for pronouns


linked to allPs: subjects, objects or objects of P. Given the analysis of Welsh, we can therefore assume that the pronouns move to Spec, AgrS, Spec, AgrO and Spec, AgrP in the overt syntax . But why do only pronouns raise to Spec, Agr positions in the overt syntax? The same answer can be given to English as we provided for Welsh and Norwegian. The asymmetry reduces to an asymmetry in the DP structure of pronouns and lexical DPs. Only pronouns move because NumP is in Spec, DP in the overt syntax, and can therefore undergo further movement in configurations where Spec, Agr is structurally available. Lexical NumP are in situ and cannot therefore undergo any movement.42 We thus assume that, as in Welsh, NumP must move to Spec, Agr, whenever possible. (See (58) on p. 108.) This explains the asymmetry: pronouns are NumPs which appear in Spec, DP in the overt syntax. They are therefore in a position from which further syntactic movement not only is possible, but actually required in configurations in which Agr is projected.43


particle construction pronouns must occur in preparticle position, whereas lexical NPs can either precede or follow:

(59)

a. b. c. d.

I put him up. *I put up him. I put my family up. I put up my family.

If NumPs must occur in Spec, Agr, and lexical DPs cannot, there must be both a Spec, Agr position and a lexical DP position preceding the particle (as concluded in Johnson (1991)). NumPs must move to Spec, Agr, and lexical DPs may optionally move to a Spec position preceding the particle. How does NumP movement proceed? The following examples yield insight in the derivation: (60)

a. b. c.

I called them all up. *I called these people all up. *I called them up all.

The allP obligatorily precedes the particle when NumP extraction has occurred. This position is not explained by head movement of all left adjoining to P: if P adjoined to the left of the allP, the allP should be transparent for DP movement, and lexical DP movement to the pre-particle position should be possible. (60b) is therefore wrongly predicted to be possible. I therefore conclude that NumP extraction is preceded by DP movement to preparticle position: The syntax of allPs thus shows that pronouns (i.e. NumPs) must occur in Spec, Agr positions which cannot host lexical DPs (i.e. with these people all). If this is correct, it can be concluded that pronouns always occur in Spec, Agr positions. Support for this assumption comes from the verb particle construction: in the verb Finally, since English V in the verb particle construction obligatorily precedes pronouns it follows that V moves from a V position to some higher head position, which is higher than the Spec, Agr position to which the pronoun moves.


In sum, English pronouns occur in a position from which lexical DPs are excluded. Since distributional asymmetries show up exactly in the configurations where pronouns in Welsh trigger agreement, we argued that English NumPs overtly move to Spec, Agr positions when these positions are syntactically projected. Note that we have not entertained an analysis which accounts for the position of pronouns by head movement of the pronoun (NumP movement to Spec, allP followed by adjunction to the right of a higher host (V, P, and AgrS?).44 English pronouns do not look like clitics. A strong argument in favour of the NumP analysis is that it captures the similarities between English and Welsh: English exactly looks like Welsh, except that it has overt NumP, and silent Ds and Agrs. English and Welsh syntactic configurations are thus entirely identical, and the difference ultimately reduces to the internal structure of the pronominal DP: in English NumP is always overt and D and Agr are silent, NumP in Welsh can be empty and D and Agr are overt. 7 Conclusion In this article, I provided arguments that pronominal DPs consist of NumP in Spec, DP and D, and showed that this hypothesis has direct consequences for the syntactic distribution of pronouns. In particular, I argued that asymmetries in the distribution of pronouns and lexical DPs follow from NumP movement to Spec, Agr positions,


whenever possible. This analysis finds strong support in the internal analysis of Welsh pronominal paradigms and agreement, and also yields insight into the distribution of pronominal elements crosslinguistically. Thus, the distribution of pronouns in Norwegian and English is basically identical to that of Welsh and other Celtic languages, and reduces to the optionality of NumP movement DP internally (optional in Norwegian (pronouns can be modified) but obligatory in English/Welsh), and to whether the language is one in which NumP is overt (English/Norwegian) and D/Agr silent, or is one with NumP silent (Welsh and other Celtic languages), and D/Agr overt: (62)

A.

B.

NumP movement to Spec, DP or not: : obligatory: English/Welsh [NumP[D]… …] optional: Norwegian NumP overt/covert, Agr/D overt/covert: (i) [NumPpronoun [Agr/De]…(English, Norwegian…) (ii) [NumPpro [AgrAgr/D]…(Welsh45…)

The coexistence of both systems, as in the Welsh pronominal DP, in which all possible combinations of overt/covert NumP and overt/covert D are attested, provides the empirical basis on which NumP movement to Spec, Agr could be established. If there are languages in which NumPs never raise to Spec, DP, we would expect pronouns to have exactly the same distribution in the overt syntax as lexical DPs, and not to interact in any way with Spec, Agr, simply because they are too low in the structure. Our analysis provides new understanding of the syntax of pronouns and agreement, and shows that a common structure underlies both. The next question is how this proposal applies to clitic constructions. Romance clitic constructions quite straightforwardly fit into the structure presented so far. Sportiche (1995a) argues for an analysis of clitic constructions as involving a base generated head (the clitic) and an element in the Spec, Clitic which has undergone movement from a DP position: if the moved element is actually NumP, Romance languages fit into the schema above, with NumP silent and Agr overt (62b.ii), although quite interestingly, D in the pronominal DP is silent (or alternatively has undergone movement to the Agr position). If this is correct, there is a common structure underlying the distribution of pronouns, clitic and agreement, involving NumP movement and Agr. The analysis raises many new questions which go beyond the scope of this chapter. I will briefly address some of these. First, the analysis yields new insight in the distribution of pro. As a prerequisite for a pro-drop language, the language must be a silent NumP language. Furthermore, the language must be an overt Agr language.46 And finally, we speculated, the head of Agr must be in a Spec, head relation with the Spec containing pro: i.e. pro-drop is dependent on the position of the head carrying Agr.47 Agr licenses NumP in the syntax. This raises questions about its nature, its function, and its position (see Sportiche (1995a) for discussion of these issues). Agr, as used in this chapter, encodes person and number features, and is identical to what is sometimes called pronominal agreement. The content of Agr differs from participle agreement or adjectival agreement in Indo-European languages, which do not encode person agreement. Note that


I have also been using agreement as a reflex of a Spec head relation, as for instance the DP internal agreement between NumP and D. This suggests that the two views in the literature on Agr are both correct: there are certain instances of Agr that involve an overt syntactic head (i.e. pronominal agreement), and certain cases in which agreement is a byproduct of a particular Spec, head relation. In the languages under discussion, Spec, Agr can only host NumP in the overt syntax, but not lexical DPs. This raises two questions. Is this always the case? And if so, why? I will assume without further evidence that the agreement asymmetries reflect LF; Agr only licenses NumP, never DPs:48 this would follow if the LF function of Agr is the licensing of pronominal features. If this is correct, lexical DPs with no pronominal features to license occur in a different position at LF, probably the Case positions, and contrary to common assumptions, Agr and Case are two independent heads, having distinct licensing purposes (Sportiche (1995a). This would provide a motivation both for the obligatory raising of subject lexical DPs in Welsh and for additional NumP movement: raising of the lexical DPs to the Spec, Nom position, and NumP to Spec, Agr in the overt syntax, reflects the actual licensing positions at LF. A similar explanation extends to English objects: they can optionally move to Spec, AccP, but NumP must obligatorily move to Spec, Agr. Why does Spec, Agr only host NumP, not DP? We will leave this question open for future research. Where precisely is Agr located and what crosslinguistic differences are observed? Given the position of Agr in Welsh, we can conclude that AgrS occurs just below C, AgrN just below D, AgrP just below the “complementizer” position of P. The position of (pronominal) AgrO is less clear. It is lower than AgrS. Sometimes it seems to be higher than T, as for example in languages in which tense is realized independently of AgrS and AgrO, and in which AgrS AgrO are fused as portmanteau morphemes; sometimes it is lower than T (Basque, Bantu), and separated from AgrS by T. The problem of the position of AgrO thus recalls similar problems concerning the position of NegP.49 Notes * This chapter is dedicated to the memory of Osvaldo Jaeggli. I have been intrigued by the distribution of pronouns, and the relation between pronouns, clitics, and agreement for a long time now (Koopman 1984:73–76), Koopman and Sportiche (1982) and Koopman and Sportiche (1989). This chapter grew out of a presentation given at MIT in 1987 (“Restrictions on Spec positions”), where I proposed that pronouns but not lexical DPs occur in particular Spec positions in the overt syntax, in conjunction with ideas about the internal DP structure of pronominal DP (Koopman 1991). Special thanks to participants of several seminars at UCLA (Spring 1989, Spring 1991, and Winter 1993) where part of this material was developed, to audiences at UCLA and University of Geneva where part of the material was presented, and to Randall Hendrick, James McCloskey, Richard Sproat, for e-mail conversations on Welsh, to Ian Roberts for getting me help with judgments on some crucial Welsh examples, to Arild Hestvik for discussions on Norwegian, and to Kyle Johnson, Tim Stowell and Dominique Sportiche for comments on a previous version of this chapter (June 1993). The present chapter is identical to the December 1993 version.


1 This seems to have changed in the beginning of the 1990s. See several abstracts in the GLOW Newsletter 1993, Rouveret (1990), Cardinaletti (1993), Carstens (1993), among others. 2 Or some D-like head. The label is not important; the position within the hierarchy of the DP is. 3 For reasons of simplicity, I omit person and gender features, though I believe these should project a person and gender projection respectively. These projections probably enter in an account of the difference in distribution between first or second person pronouns and third person pronouns, but do not affect the main points made in this chapter. 4 For example, N to Num movement bleeds NP to Spec, DP movement: the NP would contain an unbound trace in the head position. In other words, NumP movement to Spec DP is forced in this case. 5 This generalization does not extend to emphatic pronouns. 6 Possibly through an intermediate position if this is the position where person features are licensed (PersP or AgrP). Note that it might be the case that this node is only present when non-third person features must be licensed. This might give some insight into the as yet unsolved problem why in so many languages the distribution of first and second person pronouns differs from that of third person pronouns. 7 See also Moritz (1993) for Ncufie, another Grassfield Bantu language. 8 I will not attempt to explain this asymmetry. 9 This may also explain why only NP can be a complement in morphology, but not NumP: only when a morpheme takes a NP complement can N incorporate. The presence of any higher projection blocks incorporation. This approach quite generally extends to derive incorporation asymmetries from the structural properties of the overt syntax (see Koopman (1994) for some discussion), and derives the relativized minimality effects for incorporation of Baker and Hale (1990). 10 I believe that reduplicated pronouns, which frequently occur in African languages, fit into this structure. It is interesting to note that superficially speaking African languages allow both Spec and head position to be overt at the same time. This is generally not the case in Indo-European languages. 11 The analysis of Welsh pronominal paradigms is based primarily on Williams (1980). 12 Williams (1980:46) writes: “the reduplicated forms are somewhat more emphatic than the simple”. I interpret this statement as evidence that there is no real difference between simple and reduplicated independent pronouns. 13 I assume, without argument, that vowel reduction accounts for the appearance of y and not i in the paradigm of the reduplicated pronouns. 14 Although efe might be composed of e and fe, I have not been able to come up with anything interesting for third person (non-feminine) pronouns. Arguments will be based primarily on the distribution of first and second person pronouns. 15 See Morris-Jones (1931:83): “The affixed pronoun has often an accent of its own. In poetry it is usually accented”. 16 A number of irregularities are glossed over: third person pronouns (non-feminine) do not appear to allow for an overt NumP in Spec DP. Some other forms in the auxiliary pronouns do not occur as reduplicated pronouns (i, ti, hwynt). The use of both i and ti appears to be phonologically conditioned (Williams 1980:50, 74 (i)). The occurrence of hwynt as an auxiliary pronoun is potentially problematic, but not attested in any of the other grammars I have consulted. I will ignore this form in the remainder of this chapter.


17 Conjunctive pronouns have a variety of uses: they are used in apposition to a noun (David “he too”), in case of a conjunction of a lexical DP and a pronoun, and have a variety of other uses (“even I”, “while she”, etc.). 18 For the distribution of agreement, pro, and pronouns in Celtic languages, see, among others, Anderson (1982), McCloskey and Hale (1984), Stump (1984, 1989), Hendrick (1989), Sadler (1988), Rouveret (1990). 19 This context differs from the others: agreement is only possible in highly formal varieties of Welsh. In addition, it is optional. There are further structural conditions on its appearance (Sadler 1988:76). 20 I have nothing to say about this asymmetry with respect to gender agreement. 21 I believe that this asymmetry is ultimately due to the presence of Tense. Tense causes there to be a pronoun in Spec, TP, which will raise to Spec, Agr, triggering third person agreement. There is indirect evidence for a null agreement in DPs containing a genitive DP. In this construction, the definite article must be silent, even though the genitive DP is quite low in the structure (in Spec, NP see 4.2 in this chapter). The absence of the definite article can be explained if there is a zero Agr in this construction that must move and substitute for D. 22 Similar difficulties arise of course when the head stays lower than lexical DPs and pronouns, as is the case with English subjects. 23 This analysis is not available if C lowers to I in Welsh, as McCloskey (1996) convincingly shows for Irish. The analysis arrived at below is consistent with V being in Agr, as in Koopman and Sportiche (1991). 24 I do not know if the position of the DP correlates with a specific reading for the DP or not. 25 See Anderson (1982), Stump (1984) and Hendrick (1988) among others. 26 If emphatic pronouns obligatorily moved to Spec, AgrP, (i) should be impossible, with a synthetic verb form:

(i)

*ne PCL

gouskont sleep-3P

ket not

int they

Stump (1989:414) points out that this surface string is acceptable, but only with int as a right-dislocated element. Borsley and Stephens (1989:414) argue that such examples are grammatical in Stephens’s dialect: (ii)

ne PCT

oantket were3P

int not

deut they come

See Stump (1989:437, footnote 7) for further discussion. If (ii) is fine in Stephens’s dialect, it suggests that cliticization in Stephens’s dialect is optional. The grammaticality of (ii) poses no problem for the analysis presented below. It simply implies there are two processes: NumP movement, and D cliticization, with the latter optional in Stephens’s dialect. 27 This is further confirmed by the position of the head N in composite numbers:

(i)

tair merch ac ddeg three(fem.) girl and ten


“thirteen girls” The head N adjoins to the right of the head of the compound numeral.

28 Agr does appear to contain an empty element which moves into D. This might explain the obligatory absence of the definite article in structures containing a lexical subject. See also Note 21. 29 Thanks to Ian Roberts for helping me check these examples. 30 Historically, though, there seem to have been some cases with NumP overt. In particular, some of the preverbal particles derive from pronouns, i.e. NumP in our account (mi, fe…). 31 The alternative, reducing all agreement to cases of incorporation, as proposed in Taraldsen (1992) does not seem exactly promising. 32 I will not attempt to account for the agreement patterns under coordination as described in McCloskey and Hale (1984). 33 I am aware of complications with the demonstrative pronoun (e seo ‘this’, and not *seo teach ‘this house’ but an teach seo ‘the house this’. An analysis of this construction will take me too far afield. 34 This is also true for Breton (compare Stump (1989: 438)). 35 This generalization actually extends to complements: many languages cannot both license a possessor and a theme in their DPs, and can only license an internal argument if the genitive position is available. 36 Here I assume that V movement somehow “activates” AgrO. 37 Thanks to Arild Hestvik for the examples. 38 Similar examples can be constructed for Dutch pronouns with PP modifiers. These pronouns have the distribution of lexical DPs. The pronoun cannot be a clitic form (d’r or ’m):

(i)

ik ken*d’r/ haar van boven I know her from upstairs

niet goed not well

This shows that when there is a clitic form, NumP to Spec, DP must take place, and when there is not, NumP to Spec, DP is optional. Note that this conclusion holds independently of the problem how these clitics should be analyzed: if they involve NumP movement, they must be preceded by NumP movement to Spec, DP; if they involve head movement, NumP movement to Spec, DP must be assumed as well.


39 The star stands for the interpretation “I do not like the person with long hair”, not for a small clause construal “I don’t like him when he has long hair.” 40 This account raises the question how object shift in Icelandic, where lexical DPs and pronominal DPs both shift, should be accounted for. I believe that this phenomenon is much closer to scrambling in Dutch, which applies to both pronouns and specific DPs, whereas the pronoun/lexical DP asymmetry is closer to clitic placement, i.e. involves NumP movement to a pronominal head projection (Dutch actually has both processes). This implies that there is a distinction between AgrO (which could be accusative Case) and pronominal AgrO, which could be much closer to Sportiche’s (1995) clitic projections. 41 The same is true for topicalization: (?)all these people I really like, *them all I really like. 42 Overt movement can be easily forced in Chomsky’s 1992 theory where only morphological features can cause overt movement in the syntax (strong morphological features must be eliminated through checking in the overt syntax). English pronouns differ morphologically from English lexical Ns in carrying overt morphological Case distinctions. Pronominal movement could therefore be motivated as DP movement to the position where these features are checked. Asymmetries between pronominal DPs and lexical DPs would arise from the principle of Procrastinate (Chomsky 1992: don’t move unless you are forced to). There are quite serious problems with Procrastinate, however. Consider for instance Chomsky’s assumption that English accusative DPs do not (or do not optionally) move in the overt syntax. The distribution of DPs in the verb particle construction quite strongly argues in favor of optional DP movement of lexical DPs (Johnson (1991), Koopman (1991) among others). But if DPs can optionally move, the impossibility of structures like (i) cannot be explained:

(i)

a.

*with these people all

An account which forces movement of the entire pronominal DP thus runs into the problem that optional DP movement should be able to move the DP to the same positions as well. 43 While pronouns move to Spec AgrS, Spec, AgrO and Spec, AgrPP, Spec, AgrN is absent from this list, in contrast with Welsh. In fact, interesting problems arise with this position in DPs in English.

(i)

a. b. c. d. e.

I spoke to the mother(sg) of all these children. I spoke to the mothers of all these children. ?*I spoke to all these children’s mother. (??) I spoke to all these children’s mothers. **I spoke to them/their all’s mother(s).

While ((i)e) is crushingly bad, the source sentences ((i)c) and ((i)a) are pretty unacceptable as well. It is unclear whether the interpretation in ((i)b) is available for ((i)d): this sentence feels like a garden path, and judgments are accordingly. However, regardless of what explains((i)c) and ((i)d), if, in addition, NumP must move to some higher Agr position, the ungrammaticality of ((i)e) is explained: the DP containing the pronoun is in Spec, AgrN: NumP cannot move to a higher Spec


position, say Spec, AgrO, or Spec, AgrP because it is not governed from the outside. 44 45 46 47

It is often argued that English pronouns are clitics. D can be overt or covert in Welsh. Or maybe a silent Agr language and a silent NumP language, as Chinese or Japanese. If French finite verbs do not occur in Agr, but lower, and the V in other Romance languages moves to Agr, it would follow French is not (subject) pro-drop. 48 There are typically Agr asymmetries with overt object agreement. Overt object agreement is often restricted to specific DPs. This might follow if agreement is always and only triggered by NumP. If the [NumP D] structure is a prerequisite for triggering of object agreement, you must belong to a particular class of DPs for NumP extraction to be possible. If the DP is lower than AgrO, there will be no agreement. If it is in a base generated A’position, there will be agreement with a NumP in Spec, Agr. 49 Finally, the question arises what light this analysis sheds on clitic doubling constructions. In Welsh, pronouns are obligatorily doubled by Agr: this clitic doubling was described as NumP movement to Spec, Agr with D stranding. This suggests a similar analysis for clitic doubling constructions quite generally. The problem in clitic doubling constructions is finding a source for the clitic. In clitic doubling constructions, the doubled DP must typically be preceded by a dummy P (Spanish a, Hebrew sel,…). Classical analyses of clitic doubling constructions (Jaeggli (1981), Borer (198) for example) assume that the clitic absorbs Case, and that the function of the dummy P is to Case mark the DP. The analysis presented in this chapter provides another way of looking at the appearance of P, however. In the absence of P, there is no source for NumP movement: the lexical NumP is within the DP. It could be proposed that the dummy P creates an additional position from which NumP extraction is possible. The dummy P creates an additional agreement projection, AgrP, in which an expletive NumP occurs, agreeing with NumP in the DP. It is the expletive NumP that moves to a higher Spec, Agr, yielding clitic doubling. In other words, these dummy Ps are parasitic on P, and project part of the PP structure. This allows the presence of an expletive NumP. This NumP can further move to Spec, Agr, only if there is no C like type position dominating Agr, i.e. these dummy Ps project transparent PP small clauses (in that way they resemble Ds).

References Abney, S. (1987) “The English noun phrase in its sentential aspect”, unpublished PhD thesis, MIT. Anderson, S. (1982) “Where is morphology”, Linguistic Inquiry 12:571–612. Awberry, G. (1976) The Syntax of Welsh: A Tranformational view of the Passive, Cambridge University Press, London. Awberry, G. (1990) “Dialect syntax: a neglected source for Welsh”, in R.Hendrick, M. Baker and K.Hale “Relativized Minimality and Pronoun Incorporation”, Linguistic Inquiry 21:289–297. Baker, M. and K.Hale (1990) “Relativized minimality and pronoun incorporation”, Linguistic Inquiry 21:289–297. Borer, H. (1984) Parametric Syntax, Foris Publications, Dordrecht. Borsley R. and J.Stephens (1989) “Agreement and the position of subjects in Breton”, Natural Language and Linguistic Theory 3:407–427. Cardinalletti, A. (1993) “On the internal structure of pronominal DPs”, unpublished manuscript, University of Venice.


Carstens, V. (1991) “The morphology and syntax of determiner phrases in KiSwahili”, unpublished PhD dissertation, UCLA. Carstens, V. (1993) “Deriving agreement”, unpublished manuscript, Cornell University. Chomsky, N. (1991) “Some notes on economy of derivation and representation”, in R.Freidin (ed.) Principles and Parameters in Comparative Grammar, MIT Press, Cambridge, Mass. Chomsky, N. (1992) “Towards a minimalist program for linguistic theory”, MIT Occasional Papers in Linguistics 1. Chung, S. and J.McCloskey (1987) “Government, barriers and small clauses in Modern Irish”, Linguistic Inquiry 18: 173–237. Corver, N. (1990) “The syntax of left branch extractions”, unpublished PhD dissertation, University of Tilburg. Crook, H. (1990) “The internal structure of DPs in Bambara”, unpublished MA thesis, UCLA. Grimshaw, J. (1991) “Extended projections”, unpublished manuscript, Brandeis University. Hendrick, R. (1988) Anaphora in Celtic and Universal Grammar, Kluwer, Dordrecht. Hendrick, R. (1990) The Syntax of Modern Celtic Languages, Academic Press, San Diego, 27–79. Hestvik, A. (1990) “LF movement of pronouns and the computation of binding domains”, unpublished PhD dissertation, Brandeis University. Holmberg, A. (1986) “Word order and syntactic features in the Scandinavian languages and English”, unpublished PhD dissertation, Stockholm. Jackendoff, R. (1990) Semantic Structures, MIT Press, Cambridge, Mass. Jaeggli, O. (1981) Topics in Romance Syntax, Foris Publications, Dordrecht. Johnson, K. (1991) “Object positions”, in Natural Language and Linguistic Theory, 9, 577–637. Kayne, R. (1985) “Principles of particle constructions”, in Guéron et al. (eds) NLLT 9: 577–636. Grammatical Representation, Foris Publications, Dordrecht. Koopman, H. (1984) The Syntax of Verbs, Foris Publications, Dordrecht. Koopman, H. (1988) “The structure of the VP in Dutch”, paper presented in Geneva (see Note *). Koopman, H. (1989) “Pronouns in Spec”, guest lecture at MIT. Koopman, H. (1991) “The verb particle construction and the syntax of PPs”, unpublished manuscript, UCLA. Koopman, H. (1993) The Structure of Dutch PPs, unpublished manuscript, UCLA. Koopman, H. (1994) “Licensing Heads”, in N.Hornstein and D.Lightfoot, Verb Movement, Cambridge University Press, Cambridge (Chapter 9 in this volume). Koopman, H. and D.Sportiche (1982) “Variables and the bijection principle”, in The Linguistic Review 2: 139–160 (Chapter 1 in this volume). Koopman, H. and D.Sportiche (1989) “Pronouns, logical variables and Logophoricity in Abe”, Linguistic Inquiry, 555–589 (Chapter 2 in this volume). Koopman, H. and D.Sportiche (1991) “The position of subjects”, in Lingua 85: 211–258. McCloskey J. and K.Hale (1984) “On the syntax of person-number inflection in Modern Irish”, Natural Language and Linguistic Theory 6: 143–189. McCloskey, J. (1996) “On the scope of verb movement in Irish”, Natural Language and Linguistic Theory 14: 47–104. Malung, J. (1976) “Notes on quantifier postposing”, Language Inquiry 7: 708–718. Morris-Jones, J. (1931) A Welsh Grammar: Phonology and Accidence, Oxford University Press, Oxford. Moritz, L. (1993) “DPs in Ncufie”, in Koopman and Kural (eds) The structure of Ncufie, UCLA working papers. Nkemnji, M. (1993) “The structure of noun phrases in Nweh”, unpublished manuscript, UCLA. Nkemnji, M. (1995) “Heavy pied-piping in Nweh”, unpublished PhD thesis, UCLA.


Postal, P. (1969) “On the so-called ‘pronouns’ in English”, in D.Reibel and S.Shane (eds) Modern Studies in English, Prentice Hall, Englewood Cliffs N.J: 201–224. Ritter, E. (1991) “Two functional categories in Noun Phrases: evidence from Modern Hebrew”, unpublished manuscript, UQAM. Rouveret, A. (1990) “X-bar theory, minimality and barrierhood in Welsh”, in R.Hendrick et al (eds) Linguistic Enquiry 21: 289–297. Rouveret, A. (1991) “Functional categories and agreement”, The Linguistic Review 8: 353–387. Sadler, L. (1988) Welsh Syntax. A Government-Binding Approach, Croom Helm, London. Sportiche, D. (1988) “A theory of floating quantifiers and its corollaries for constituent structure”, Linguistic Inquiry 19: 425–449. Sportiche, D. (1990) “Movement, agreement and case”, unpublished manuscript, UCLA. Sportiche, D. (1995a) “Sketch of a reductionist approach to syntactic variations and dependencies”, in H.Campos and P.Kempschinsky (eds) Evolution and Revolution in in Romance Syntax, Georgetown University Press, Washington D.C. Sportiche, D. (1995b) “Clitic constructions”, in L.Zaring and J.Rooryck (eds) Phrase Structure and the Lexicon, Kluwer, Dordrecht. Stowell, T. (1991) “Small Clause Restructuring”, in R.Freidin (ed.) Principles and Parameters in Comparative Grammar, MIT Press, Cambridge, Mass. Stump, G. (1984) “Agreement versus incorporation in Breton”, Natural Language and Linguistic Theory 2: 289–348. Stump, G. (1989) “Further remarks on Breton agreement”, Natural Language and Linguistic Theory 3: 429–472. Szabolcsi, A. (1987) “Functional categories in the noun phrase”, in I.Kenesei (ed.) Approaches to Hungarian vol. 2. Jate Szeged, Budapest. Taraldsen, K.T. (1992) Pronouns and Agreement, GLOW newsletter. Valois, D. (1991) ‘The internal syntax of DP”, unpublished PhD dissertation, UCLA. Watkins, T.A. (1972) “The Welsh personal pronoun”, Word 28: 146–165. Williams, S.J. (1980) A Welsh Grammar, University of Wales Press, Cardiff.

Part II ECP

4 ECP EFFECTS IN MAIN CLAUSES*

Recent literature dealing with that t or Empty Category Principle (ECP) effects has been concerned mainly with their occurrence in embedded sentences. In this squib, I will argue that ECP effects can also be observed in English main clauses. More specifically, I will propose that the nonapplicability of Subject-Aux Inversion (SAI), an otherwise obligatory rule, in cases of subject extraction (for example, *who did t come) can be explained in terms of the ECP. (For discussion of the ECP, see Chomsky (1981).) I will then show how such an account, relying crucially on the assumption that SAI moves the Aux into Comp (den Besten 1978), sheds some light on a language like Dutch, in that it establishes a nontrivial correlation between systematic that t violations in embedded clauses and the obligatory application of Verb Second in main clauses. Finally, I will briefly discuss the implications of the analysis for the acquisition problem. 1 In English, an ECP effect is illustrated by the that t phenomenon in (1): (1)

a. b.

whoi do you think [s’[Comp ti (*that)] [s ti left]] whoi do you think [s’[Comp ti (that)] [s Johnsaw ti]]

Under an ECP account of this phenomenon,1 that must be absent in (la) so as to allow the empty category in subject position to be properly governed by the trace in Comp (by means of coindexing, since the subject position is not governed by a lexical category). In (1b), however, that may be either absent or present, since the empty category is properly governed by the verb (government by a lexical category). In main clauses, wh-extraction of the subject or the object yields examples such as (2): (2)

a. b.

whoi ti left whoi did John see ti

In (2a), the preposed wh-phrase properly governs the trace in subject position (by virtue of their being coindexed; compare (1a)); in (2b), the trace is properly governed by V (compare (1b)).


(2b) also illustrates the effect of the root rule of Subject-Aux Inversion (SAI). Any analysis of English must contain mechanisms to ensure the application of SAI in whquestions and the appearance of do in Infl (= Aux) if Infl is not adjacent to V (Do Support) (or, alternatively, the disappearance of do if adjacent to V (Emonds 1976)). The examples in (2) reveal an asymmetry with respect to SAI: whereas it has applied in (2b), it has not applied and cannot apply in (2a), as the ungrammaticality of (3), with a nonemphatic reading of do, illustrates. (3)

*whoi did ti leave

Before I show that the ECP can explain this asymmetry in the same way as the phenomena in (1), some remarks on (3) are in order. First, it is not possible to exclude (3) by arguing that wh-phrases in subject position do not move into Comp, since it was shown in Koopman (1981) that wh-phrases in subject position not only can but actually must move into Comp prior to S-structure. Thus, since who in (2a) has moved into Comp, SAI should apply. Second, once SAI has applied, Do Support should in turn apply to yield (3), since the whtrace which acts as a phonologically realized noun phrase (Chomsky (1981), Jaeggli (1980)) intervenes between the preposed Aux and the main verb. The ungrammaticality of (3) as opposed to the grammaticality of (2b) thus leads to the conclusion that SAI is blocked if the subject is questioned, whereas it is obligatory elsewhere. 2 I will now show that the asymmetry discussed above follows from the theoretical framework as it stands, under the assumption proposed by den Besten (1978) that SAI moves the Aux into the complementizer position.2 To see this, suppose that SAI applied in cases of subject extraction, yielding the structure (4): (4)

[S′[Comp whoi, did] [ti leave]]

By virtue of the ECP, the trace in subject position must be properly governed. But in (4) the Comp contains both a wh-phrase and do; moreover, the structure is exactly parallel to (la), which is ruled out by the ECP. In addition, the ungrammaticality of (4) cannot be explained by the doubly filled Comp filter (Chomsky and Lasnik (1977)), given the grammaticality of (5): (5)

[s′[Comp whoi did] [John see ti]]

Extending the account given for (la) to the structurally identical (4), the impossibility of S AI can be immediately explained by the fact that, if S AI were to apply, the resulting structure would violate the ECP, since the trace in subject position would fail to be properly governed. I thus propose that the inapplicability of SAI with subjects is explained by the ECP and that the well-known ECP effect of embedded clauses is observed in exactly the same way

ECP EFFECTS IN MAIN CLAUSES 119

in matrix clauses: treating the inverted Aux and the lexical complementizer as occupying the same syntactic position allows a uniform explanation of the necessary deletion of that in structures like (la) and the impossibility of SAI with subjects.3 3 The analysis presented above for English main clauses opens new ways of looking at the problem of Dutch, a language in which that t violations occur.4 Why do English and Dutch differ with respect to that t phenomena? I propose that the difference can be reduced to a difference in the functioning of the similar rules of SAI and Verb Second: whereas Verb Second resembles SAI in moving the finite verb into the complementizer position (den Besten 1978), it differs in being obligatory in all Dutch main clauses, regardless of the original position of the preposed constituent. Its functioning is illustrated in (6): (6)

a.

b.

[s′[Comp

wiei heeftj] who has “Who saw him/John?” wiei [s′[comp who have “Who have you seen?”

[ti hem/Jan him/John hebj] [jij ti you

gezien tj]] seen

gezien tj]] seen

Contrary to what happens in the English equivalent (3), the movement of the finite verb into Comp in (6a) does not result in an ECP violation. I conclude accordingly that, unlike English, Dutch permits proper government from Comp in a configuration like (6a), yielding a that t violation in main clauses.5 If proper government is possible in structures like (6a), in which the Comp contains both a wh-phrase and the finite verb, we would expect it to be equally possible in embedded contexts. This prediction is borne out, as (7) illustrates: (7)

denk je [s′[Comp ti dat] wiei who think you that “Who do you think saw John?”

[ti hem/Jan him/John

gezien seen

heeft]] has

4 That t violations are systematically possible in Dutch in both matrix and embedded clauses, whereas they are systematically impossible in English in both matrix and embedded clauses. The difference can be reduced to the different functioning of the rule of Verb Second, which forces Dutch to allow a more liberal environment for proper government from Comp ((6a) and (7)) than English does. The fact that languages differ with regard to the exact conditions under which proper government from Comp takes place raises the interesting question of how the language learner is able to deduce these conditions. The account given here, making no distinction between ECP effects in matrix and embedded sentences, gives insight into this problem. Consider English, for example.


Depending on further assumptions about how language “learning” proceeds, we could assume either that English represents essentially the unmarked case (i.e. there is nothing to learn) or, admitting indirect negative evidence, that it would be sufficient for the language learner to be presented with the asymmetric behavior of SAI (Who came? vs. Who did John see?) to deduce that proper government from Comp requires Aux to be absent from Comp and, by extension, requires the complementizer to be absent as well, consistently yielding that t effects. In Dutch, however, the symmetric behavior of Verb Second forces the language learner to assume proper government in cases like (6a) even if the Comp node contains both a wh-phrase and the finite verb, or a trace and the complementizer, yielding that t violations in both matrix and embedded clauses alike. The appeal of this analysis lies in its ability to make the complex grammatical behavior with respect to that t phenomena readily deducible upon presentation of simple clauses to the language learner. Notes * Research for this squib is supported in part by the Social Sciences and Humanities Research Council of Canada, grant 410–81–0503. Many thanks to Dominique Sportiche for discussion of the issues presented here. 1 The ECP (Chomsky 1981) requires that an empty category be properly governed. Since the notion of proper government has received many definitions in the literature, I will present the initial one here:

(i)

ECP [β e] must be properly governed. α properly governs β iff α governs β a. α=X° or b. α is an NP coindexed with β

Proper government is a stricter requirement than government, for which I assume the definition presented in Sportiche and Aoun (1981). 2 For a different approach to this problem, see Safir (1981). 3 Note that it cannot be assumed that SAI has in fact applied in cases of subject extraction and that the lexicalization of do is blocked, given sentences in which a modal or auxiliary appears in Infl (who has come, who must come). By virtue of the argument given above, these sentences must have an S-structure representation like who t must come. We must therefore conclude that SAI is obligatory up to the violation of a principle. 4 The situation in Dutch is complicated and needs elaboration. Considering the most conservative dialect with respect to subject extraction (called Dutch B by Maling and Zaenen (1978) and Dutch by Bennis (1980)), Koopman (1982) argues that, in the case of intransitive verbs or transitive verbs with indefinite objects, the extraction takes place from a properly governed position in the VP.

ECP EFFECTS IN MAIN CLAUSES 121

Root and embedded sentences act alike with respect to (i)-(iii). In the case of transitive verbs with definite objects, however, the subject can be extracted in both main and embedded clauses (cf. (6) and (7)). Although (i)-(iii) raise many intriguing questions, it seems clear that extraction from subject position in (6) and (7) is acceptable in all Dutch dialects. 5 I assume that the Comp node properly governs the trace in subject position, and that it does so iff it is coindexed with the subject position. Comp can be indexed by means of a percolation rule applying at S-structure (Aoun, Hornstein and Sportiche (1981) Bennis (1980)). I furthermore assume (Koopman (1982)) that languages differ, first, in allowing Comp indexing at all (consider Vata vs. English and Dutch) and, second, in the conditions under which Comp indexing may occur.

References Aoun, J., N.Hornstein, and D.Sportiche (1981) “Some aspects of wide scope quantification”, Journal of Linguistic Research 1. Bennis, H. (1980) “Coindexing and complementizer trace phenomena”, paper presented at the 1980 GLOW conference, Nijmegen. Besten, H. den (1978) “On the interaction of root transformations and lexical deletive rules”, paper presented at the 1978 GLOW conference, Amsterdam. Chomsky, N. (1981) Lectures on Government and Binding, Foris, Dordrecht. Chomsky, N. and H.Lasnik (1977) “Filters and control”, Linguistic Inquiry 8: 425–504. Emonds, J. (1976) A Transformational Approach to English Syntax, Academic Press, New York. Jaeggli, O. (1980) “Remarks on to contraction”, Linguistic Inquiry 11: 239–245. Koopman, H. (1981) “Theoretical implications of the distribution of quoi”, NELS XII. Koopman, H. (1982) “Control from COMP and comparative syntax”, unpublished manuscript, Université du Quebec a Montreal (Chapter 5 in this volume). Safir, K. (1981) “Inflection, government, and inversion”, The Linguistic Review 1 (4): 417–467. Sportiche, D. and J.Aoun (1981) “On the formal theory of government”, The Linguistic Review 2 (3): 211–236.

5 CONTROL FROM COMP AND COMPARATIVE SYNTAX*

1 Introduction The conception of Universal Grammar (U9) as a system of principles and parameters from which a specific core grammar can be derived by fixing the parameters of the system, has lead to a renewed interest in comparative syntax. Indeed, the powerful analytic tools resulting from recent theoretical developments, not only contributes to discover deep similarities between superficially very different languages, but also permits the discovery of systematic patterns of variations and allows one to reduce them to different choices in the value of some parameter of the system. Here, we want to look at one particular principle of UG (the ECP) and the crosslinguistic variation surrounding its scope of application. More specifically, we will be concerned with subject/object asymmetries which obtain with respect to both syntactic and LF wh-movement, and with the nature of proper government of wh-traces in subject position. Chomsky (1981) proposes to account for subject/object asymmetries, exhibited in English for example by the so-called that-t phenomena, (*who do you think t that t came) by means of the Empty Category Principle (ECP), a principle governing the distribution of empty categories at LF. The ECP can be stated as: [αe] must be properly governed The notion of proper government has received many definitions in the literature. For heuristic reasons, the initial one will be presented here:

β properly governs α iff β governs α1 a) β=X° b) β is an NP coindexed with α. Essentially, this principle requires that empty categories (more precisely trace of NP-and wh-movement) be locally controlled.


In this view, the observed subject/object asymmetries are a consequence of government: whereas the object of a verb is governed by a lexical category (namely the verb itself), the subject is not. The question arises how the condition on proper government is fullfilled for traces in subject position. It is generally assumed that in cases of subject extraction in matrix clauses (e.g. who t came) the subject trace is properly governed by the wh-phrase in the adjacent COMP (by virtue of being coindexed). In cases of long extraction, the complementizer that in who do you think t (*that) t came must be absent so as to allow the trace in COMP to properly govern the trace in subject position. We will refer to proper government of wh-traces in subject position, our main concern here, as Control from COMP. Within the framework outlined above, restricting ourselves to wh-questions, we will first present data on subject/object asymmetries, discussing Vata, a West African language (Kru family) in section I, and French in section 2. We will argue that the ECP explains the configuration of data in both Vata and French and that our analysis supports the hypothesis that the ECP applies at the LF level of representation as first argued for in Kayne (1981) and subsequently in Rizzi (1982) and Jaeggli (1980). Our analysis shows that there is some crosslinguistic variation as far as Control from COMP is concerned: in Vata, contrary to French, a subject trace can never be properly governed from COMP, not even in matrix clauses. Many questions arise as to what Control from COMP is and what determines the possibility or the impossibility in a given language for Control from COMP. The analysis of the French data bears on these questions, in that they establish that movement to COMP in LF does not lead to proper government configurations, supporting a similar conclusion reached in Aoun, Hornstein and Sportiche (1981) (henceforth AHS). We will argue that, since the wh phrase in COMP cannot be the element which properly governs the trace in subject position, it must be the COMP node itself which properly governs it, by virtue of the existence of a COMP indexing rule. In section 3, we address the problem of how the crosslinguistic variation can be accounted for and propose the Control from COMP parameter. This parameter is discussed with respect to certain well-studied languages (English, Dutch and Italian) in 3.1 and 3.2. In 3.3 it is discussed from the point of view of the language learner. In 3.4, we address some apparent problems for our analysis which arise in languages like Chinese (Huang 1982) which lack a rule of syntactic wh-movement applying in the formation of wh-questions. In 4, finally, we summarize the major findings of this chapter. In Vata, a West African language spoken in the Ivory Coast, there exist two strategies for the formation of wh-questions, either by the application of (syntactic) wh-movement, moving the wh-phrase into COMP, or by leaving the wh-phrase in its A-position (henceforth wh-in situ). We will discuss both strategies in turn and show that they display subject/object asymmetries.2 1.1 Wh-movement The sentences in (1) contain examples which are representative of extraction out of subject and non-subject position:3

124 CONTROL FROM COMP AND COMPARATIVE SYNTAX

(1)

a.

b.

àl *( ) who *(heR) “Who is eating rice” kòfí yī; what Kofi “What is Kofi eating”

lē; eat

s ká rice

l wh

l eat

(*mí) (*it)

l wh

When the subject is questioned, a resumptive pronoun must occur in subject position.4,5 When non-subjects are questioned, resumptive pronouns are excluded. This generalization holds in all constructions which involve wh-movement, i.e. whquestions, focus constructions, and relative clauses. Data on extraction of subjects and non-subjects in focus constructions and relative clauses are presented in (2) and (3) respectively: (2)

a.

b.

(3)

a.

b.

lē; eat

m ( ) k man PRON *(heR) “It is the man who is eating rice.” s ká má k rice PRON man “It is rice the man is eating.” k mā;m ( ) man REL-PRON *(heR) “the man who is eating rice, …” mā;m ` k s ká rice REL-PRON man “The rice the man is eating, …”

s ká rice

l eat lē; eat

le eat

(*ma) (*it)

REL

s ká rice (*má) (*it)

REL

These data show that wh-movement of subjects requires the presence of a resumptive pronoun, notwithstanding the occurrence of a wh-phrase (la), a focused pronoun (2a) or a relative pronoun (3a) in COMP. Resumptive pronouns are excluded from any other position. The same asymmetry between subjects and non-subjects is observed with apparent unbounded movement: (4)

a.

b.

gū;gū; nā; àl who you think that “Who do you think arrived” àl gū;gū; nā; who you think that “Who do you think they saw”

*( ) *(heR) w they

yì arrive y ` see

l wh (*m ) (*him)

yé PART

l wh


Before turning to the analysis of these data, it is necessary to make some remarks on the resumptive pronoun. The resumptive pronoun has the segmental form of a nominative third person pronoun but differs from it in bearing a low tone instead of the mid high tone ( ) third person pronouns usually carry. (For a description and an analysis of this low tone pronoun, see Koopman and Sportiche (1983).) The resumptive pronoun agrees with the wh-phrase, as can be seen in the following examples (compare Kaye (1982) on the selection of pronouns in Vata). (5)

a.

b.

c.

d.

gū;gū; nā; yī ; what you think that “What do you think is happening” gb gū;gū; yī ; what thing you think “What do you think is the matter” k a gū;gū; yī ; what men you think “What kind of men do you think died” k ba zàl zaloko there thereR “It is Zaloko that is nice”

ì itR

lē; fall

l wh

nā; that

itR

lē; fall

nā; that

wà theyR

l wh

die

l wh

b kp `l beautiful-be

Finally, the resumptive pronoun behaves syntactically like a wh-trace (compare Koopman and Sportiche (1983)): constructions containing a resumptive pronoun behave in the same way as the cluster of properties which is characteristic of wh-movement (Chomsky 1977), as do constructions containing a wh-gap. In other words, although gaps are not allowed to occur in subject position, constraints (in particular Subjacency) may not be violated either. To sum up, wh-constructions in Vata display a subject/object asymmetry (which is in fact a subject/non-subject asymmetry): although wh-movement of non-subjects must result in a structure containing a gap, a gap is excluded in cases of extraction of the subject [NP, S] of a tensed clause, in which case a resumptive pronoun must occur.6 Interestingly, when put into proper perspective, this array of data bears striking resemblance to those of, say, English, which are dealt with in terms of the ECP. It seems therefore natural to explain the impossibility of a gap in subject position in terms of the ECP. To this effect, let us assume that in Vata the trace in subject position is not properly governed, even if it is coindexed with a wh-phrase in an adjacent COMP (contrary to English who t came). (Note incidentally that the absence of Exceptional Case Marking verbs or small clauses makes it impossible to find any examples of traces in subject position.) Thus, something must take place in Vata in order to save these structures, which is the insertion of a resumptive pronoun. Given the fact that Vata’s resumptive pronouns behave in all respects like wh-traces, it is attractive to treat the resumptive pronoun as the “spelling out” or the “lexicalization” of the trace, which is, if one adopts Chomsky’s hypothesis (1981) about the internal structure of empty categories (there is in fact one empty category NP which contains pronominal features) nothing else than the lexicalization of


the pronominal features of the wh-trace. Under this analysis, the agreement phenomena described in (5) are directly accounted for. The analysis raises a question however concerning Vata’s second strategy for forming whquestions. Why does Vata have to make use of the spelling out of wh-trace, which seems to be a marked move, instead of exploiting an already existing possibility in the language, which consists of not applying wh-movement to subjects? We turn to this question in the next section. 1.2 Wh-in situ As mentioned above, there exists a second strategy for the formation of wh-questions in Vata, which is to leave the wh-phrase in A-positions (wh-in situ).7 The distribution of whphrases in situ turns out to be very interesting and reveals a second subject/object asymmetry: although wh-phrases may appear in object position (6), they are excluded from subject position (7):8 lē; you eat “What do you eat”

(6)

(7)

a.

yí came gū;gū; think

àl who * you

b.

yī; what

l wh nā; that

àl who

l wh

y` came

l wh

This subject/object asymmetry only occurs with wh-quantifiers, as the contrast in grammaticality between (7) and (8) shows, in which the non-wh form àl “the other one” or a quantified expression occurs in subject position: (8)

a. b.

àl the k man

other kwlá each

one y` arrived

y` arrived

An adequate theory should account for the subject/non-subject asymmetry and for the difference in behavior between wh-quantifiers and other quantifiers. Let us follow Aoun, Hornstein and Sportiche (1981) (AHS) and Huang (1982) in assuming that structures like (6) and (7) are subject to an LF movement rule which moves the wh-phrase into COMP at LF (AHS’s rule of wh-raising). Application of wh-raising to the S-structures (6) and (7) yields LF representation (9) and (10): (9)

[S’[COMP wh-phrasei] [S[VP…[e]i]]]


(10)

[S’[COMP wh-phrasei][S[NPe] i…]]

Now, configuration (10) is similar to the one derived by syntactic wh-movement, which is ruled out by the ECP, as we argued above. If we adopt the hypothesis of Kayne (1979) (published in 1981) that the ECP applies at the LF level of representation, the explanation of the ungrammaticality of (7) is straightforward and identical to the one put forth to account for the impossibility of a wh-trace in subject position: the wh-phrase in situ has to move into COMP at LF by wh-raising, leaving behind a trace which falls under the ECP. Now, we already concluded that Control from COMP is excluded in Vata in configurations like (10). The ungrammaticality of (7) can thus be directly accounted for if we assume that the LF representation (10) violates the ECP, because the trace is not properly governed. (Note that in configurations like (9) no problems arise, since the trace is governed by the lexical category V.)9 Both subject/object asymmetries can thus be brought back to the fact that the configuration [COMP[wh-phrasei]] [ei] does not constitute a proper government configuration in Vata. Derivations containing a trace in subject position resulting from the application of wh-movement or wh-raising, are ruled out at LF by the ECP. It therefore follows that there does not exist any choice in Vata for questioning subjects, other than the application of syntactic wh-movement, after which the escape mechanism consisting in the lexicalization of features of the trace can and, in fact, must, take place. As noted above, non-wh-quantifiers contrast with wh-quantifiers in that they do occur in subject position. This contrast can be accounted for if we assume, following AHS, that both types of quantifiers are subject to different rules in LF: wh-quantifiers to wh-raising (moving the quantifier into COMP) and other quantifiers to Quantifier Raising (QR, May (1977)) adjoining the quantifier to S).10 1.3 In the preceding sections, the following account was given for both subject/non-subject asymmetries in Vata: wh-phrases move into COMP either by syntactic wh-movement or by wh-raising at LF, leaving behind an empty category which is subject to the ECP. On the assumption that in Vata a wh-phrase in COMP can never properly govern a trace in subject position, the explanation is simple: after syntactic wh-movement of the subject, the language must resort to an escape mechanism in order to escape the effects of the ECP (hence the lexicalization of the trace). Wh-phrases are excluded from subject position at Sstructure since the output of wh-raising will lead to an ECP violation. This account provides arguments for several hypotheses: – It provides an additional argument for the LF status of the ECP. – It provides evidence for the hypothesis that wh-phrases are moved in LF by wh-raising, leaving behind an empty category. – It provides evidence in favor of two separate processes in LF: wh-raising and QR (as AHS argue).


Interesting questions arise with respect to comparative syntax: what constitutes the difference between Vata and English with respect to Control from COMP? What type of language represents the unmarked case with respect to Control from COMP? What is Control from COMP? Where does it hold? Before addressing these questions, we will first show that the analysis concerning the subject/object asymmetries involving the French wh-word quoi sheds light upon the nature of Control from COMP, and is directly relevant for determining the Control from COMP parameter. 2 Control from COMP: the distribution of quoi The distribution of the French question word quoi has been extensively discussed in the literature, among others by Obenauer (1976, 1977), Hirschbühler (1978) and Goldsmith (1981).11 Here, we will show that quoi idiosyncratic distribution can be reduced to the existence of one language specific filter. The existence of this filter will allow us to study directly some properties of Logical Form (LF), and gain insight in the nature of Control from COMP. More specifically, we will argue that: a. It provides a simple and strong argument in favor of the hypothesis that the ECP applies at LF. b. It shows that movement to COMP in LF does not create proper government configurations for subjects. 2.1 In French, wh-questions are formed either by syntactic wh-movement, moving the whword into COMP, or by leaving the wh-word in situ. We discuss the distribution of quoi for standard French in both constructions. 2.2 Quoi in COMP Examples (11), (12) and (13) illustrate the distribution of quoi, as opposed to that of a “regular” wh-word such as qui, in direct and indirect questions. 11

a. b.

(12)

a. b.

Qui/*quoi as-tu vu Who/*what did you see A qui/à quoi penses-tu About who/about what are you thinking Je me demande qui/*quoi tu as vu I wonder who/*what you saw Je me demande à qui/à quoi tu penses I wonder about who/about what you are thinking


(13)

a.

Qui/quoi voir Who/what to see Je me demande qui/quoi voir I wonder who/what to see

b.

These examples illustrate the impossibility for quoi to appear in the COMP of a tensed sentence (1 la, 12a), unless it is preceded by a preposition (1 1b, 12b). Quoi may appear freely in the COMP node of an infinitival sentence (13a, 13b). The distribution of quoi can thus be summarized as follows: Quoi cannot appear exhaustively dominated by a [+Tense] COMP node.12 In order to account for this distribution, an odd fact, we adopt the following filter: (14)

*[COMPquoi], where [+T]

COMP exhaustively dominates quoi [+T]

This filter is in essence a reformulation (integrating Chomsky and Lasnik (1977)) of the rule PAS-DE-QUOI of Obenauer (1976), and can be considered the core of the analysis of the quoi/que and ce que alternations (see Koopman 1982). Since Chomsky and Lasnik (1977), it is generally assumed that filters apply in the phonological component (PR). What is important for the discussion here is to show that filters (14) cannot apply at the LF level of representation. The following argument shows this is in fact the case. Consider the (non-echo) question (15) in which a wh-word occurs in situ: (15)

Tu as fait quoi. You did what

As discussed in the section on Vata, structures like (15) are subject to the LF rule of whraising, which derives representation at LF like (16); (16)

[COMP

[S′

quoi]

tu

as

fait

[e]i]

Suppose now that (14) applies at LF. Then (16) would be marked ungrammatical, since quoi occurs in a Tensed COMP which exhaustively dominates it. But (15) is grammatical. We therefore conclude that (14) cannot apply to the output of wh-raising, hence not at LF. We may thus assume it applies at PR (or at S-structure). 2.3 Quoi in situ Examples (17), (18), and (19) illustrate the distribution of quoi and qui in situ. (17)

a.

Tu as décidé quoi


b.

(18)

a.

Tu comptes sur quoi You count on what Tu comptes sur qui You count on who

b.

(19)

You decided what Tu as vu qui You saw who

a. b.

*Quoi est tombé What fell Qui est tombé Who fell

The distribution of quoi is asymmetric: whereas quoi can appear in situ in object position (17a), or as the object of a preposition (18a), it is excluded from subject position (19a). Indeed, the ungrammaticality of (19a) allows us to establish that quoi may not appear in subject position. To see this, suppose first that in the surface structure (19a) quoi occurs in COMP. Then the sentence would be ruled out by the filter (14). Suppose next that quoi in (19a) does not occur in COMP, but rather in subject position. The ungrammaticality of (19a) shows however that this analysis is not available; quoi is therefore excluded from subject position. Before showing that a natural explanation can be given for the observed subject/object asymmetry in terms of the ECP, note that we cannot assume that the paradigm of quoi is defective in having no nominative form13 (besides being undesirable from a theoretical point of view). In the first place, it is possible to find quoi in subject position, in which case the sentence must receive an echo interpretation and intonation (showing incidently that the restriction that quoi be excluded from subject position only holds for general questions).14 QUOI a été décidé WHAT has been decided

(20)

A second argument can be constructed based on constructions which contain multiple whquestions and in which stylistic inversion has applied. (21)

a.? b.?

Je me demande où ei a été arrêté quii I wonder where has been arrested who Je me demande où ei est tombé quoii I wonder where has fallen what

In these sentences, a wh-word in subject position has been moved by stylistic inversion to postverbal position. Although not perfect, (21a) and (21b) exhibit no contrast in


judgment. Furthermore, the judgments in (21b) do not compare with the impossibility of sentences like (19a). We conclude that the question mark status of (21) is due to the postverbal wh-subject, rather than to quoi lacking nominative Case. The examples thus show it is impossible to stipulate that the paradigm of quoi is defective. 2.4 Returning now to the question why quoi cannot appear in situ in subject position in nonecho questions, suppose quoi is generated in subject position in D-structure. Depending on the application of wh-movement which is optional in French, the S-structures (22) are derived. (22)

[S′ [COMP quoii] [S′[COMP]

a. b.

[ [ei] est arrivé]] [quoi est arrivé]]

We know that (22a) will be filtered out by (14), since quoi cannot be exhaustively dominated by a tensed COMP. Nothing so far, however, rules out S-structure (22b). We will see that (22b) is in fact excluded by the ECP in LF, in exactly the same way as whphrases in situ are excluded from subject position in Vata. In LF, quoi in (22b) must move in COMP by wh-raising (which is obligatory), yielding (23): (23)

LF:

[S′ [COMP quoi [ [ei] est arrivé]]

If we assume now that the trace in (23) is not properly governed, the impossibility of quoi in subject position is immediately accounted for. We may thus conclude that in French, just as in Vata, movement to COMP in LF does not lead to a configuration in which proper government for subject traces holds. Independent motivation for the assumption that movement to COMP in LF never creates proper government configurations derives from the analysis of wh-in situ and Superiority put forth in AHS (1981). Let us briefly sum up their argumentation. They consider the following pair: (24)

a. b.

I know who saw what * I know what who saw

The difference in grammaticality between these two forms cannot be stated at LF because the application of wh-raising to them yields the representations (25 a) and (25b). (25)

a. b.

I know [COMP whoi whatj] [e]i saw [e]j I know [COMP what whoj] [e]i saw [e]j

These forms are identical, besides the order of the wh-phrases in COMP. The difference, they claim, must therefore lie at S-structure. It is at S-structure that there exist such rules as deletion of that in English, or the change of que→qui in French. These rules affect the


COMP node and have the effect of allowing proper government of the trace in subject position. They are furthermore specific to S-structure, since in English for example, the presence or the absence of the complementizer that does not have any effect on ECP violations created by LF rules like wh-raising, viz. (26a) (vs (26b) in which no ECP violation occurs): (26)

a.* b.

Who expects (that) who leaves Who expects (that) John likes who

This brings them to propose that: 1 It is the COMP node itself rather than the wh-phrase contained in it which properly governs the subject position. (A similar idea has been put forth in Bennis (1980)). 2 It does so iff it is coindexed with the subject position. The way it gets an index is by means of the following (optional) percolation rule applying at S-structure.15 (27)

[COMP Xi• …] → [COMPi Xi" "]

iff COMP dominates only i-indexed elements

Thus, the index of a phrase contained in COMP can optionally percolate up if the COMP dominates only i-indexed elements. The idea is clear: configurations of proper government involving COMP must be present at S-structure, or, in other words, movement to COMP in LF (i.e. wh-raising) never does create a proper government configuration. Extending this analysis to quoi yields the desired results: no well-formed sentence can correspond to the S-structure in (22b). In sum then, the impossibility of quoi appearing in subject position is excluded in the same way as wh-in situ subject position in Vata: quoi in subject position at S-structure, must move to COMP in LF by wh-raising. The ECP applies to the output of wh-raising. Since movement to COMP in LF does not create proper government configurations for the subject trace,16 which can only be created by the COMP indexing rule (27) at Sstructure, cases of movement to COMP from the subject position in LF will be ruled out by the ECP. Consider next a string like qui est arrivé, to which, by the same reasoning as above, the following S-structures can correspond: (28)

(29)

[S′ [COMPi (wh-movement and (27)) [S′ [COMP

quii

][

quii

] [ [e]i

est

est

arrivé]]

arrivé]]

Now note that (29b) is not a possible S-structure, since it will lead to an ECP violation in exactly the same way (22b) does. S-structure (29a) however leads to a well-formed


sentence because (27) can apply at S-structure. This is an interesting result: French is forced to apply syntactic wh-movement to wh-phrases in subject position, although whmovement is optional elsewhere, and we may conclude this is generally the case: if a language has both wh-movement and wh-in situ, it will not have any other choice for extracting the subject than to apply wh-movement to it. This account thus argues against prohibiting string vacuous rule application as has been proposed for example in George (1980). 2.5 The distributional properties of quoi bear on several theoretical issues, concerning the LF status of the ECP, the nature of proper government and the functioning of LF rules. The success of the theoretical framework we adopt is measured by the explanatory power of the (relevant parts of) the theoretical framework: besides filter (14), which excludes quoi if it is exhaustively dominated by a Tensed COMP, nothing further needs to be stipulated.17 The exclusion of quoi from subject position is explained in the same way as wh-phrases in subject position in Vata: the subject trace resulting from wh-raising in LF is not properly governed and hence violates the ECP. This account provides a simple and strong argument in favor of the hypothesis that the ECP applies at LF, as argued in Kayne (1981) and Rizzi (1982). This is a welcome result given the subtlety of grammatical judgments on which their arguments are based. The cases of Vata and French seem to be unique to us in the sense that their structure is such that there exists direct evidence for obligatory movement of wh-subjects in matrix clauses, notwithstanding the option of wh-in situ in both languages. 3 The control from COMP parameter 3.0 In two unrelated languages, Vata and French, which both have the rules of syntactic whmovement and wh-raising, subject/object asymmetries with respect to these rules are explained in a simple way by the ECP. The two languages differ however with respect to the possibility of Control from COMP for subject traces. In Vata, Control from COMP is never possible, neither at S-structure nor at LF. In French, Control from COMP is possible, but can only be created at S-structure by means of the COMP indexing rule (27). This implies that a wh-trace in subject position must be properly governed at S-structure. Thus, even in a language which allows Control from COMP, it is a restricted phenomenon involving a COMP indexing rule. Considering now Vata in the light of the French data, it appears that Vata is not that different from French as it may seem at first sight. Indeed, the only difference (in this respect) is the presence of a COMP indexing rule like (27) in French. If one assumes that Vata lacks a COMP indexing rule, the difference is immediately accounted for. Let us


assume accordingly that the Control from COMP parameter is the presence or the absence of a COMP indexing rule in a given language. How does this account extend to other languages? We will turn to this question now, discussing first languages which seem to lack Control from COMP (Italian 3.1), and second languages which have COMP indexing (English and Dutch 3.2), showing that systematic differences between these languages follow from the way COMP may be indexed. 3.1 Italian Before discussing Italian, let us mention some languages which seem to be potential candidates for lacking a COMP indexing rule, among others, several languages of the Kru family (Bete of Gbadi, Koyo (Kokora 1976), Dida de Lakota, and so on) Yoruba, Kikuyo,18 and Duala (Epée 1976). Clearly, only careful analyses will teach us if they do lack Control from COMP. It has been argued by Rizzi (1982) that wh-extraction out of the subject position [NP, S] is not possible in Italian, despite apparent cases such as (30): (30)

Quante pietre How many stones are fallen “How many stones fell down”

sono cadute

In fact, he argues, the wh-phrase in COMP in (30) has not been extracted from subject position, but rather from postverbal position. The evidence he provides for this claim is based on the interaction of wh-movement and the syntax of the clitic pronoun ne (Belletti and Rizzi 1981). In Italian, the clitic pronoun ne pronominalizes an N’ which is preceded by an indefinite quantifier. Crucially, this N‘must occur in the VP and be governed by the verb: (31)

a. b.

Mario *(ne) ha prese alcune Mario *(of-them) has taken some Alcune (*ne) sono cadute Some (*of them) are fallen

With one class of verbs (Burzio’s (1981) “ergative” verbs) postverbal subject NPs behave like objects: (32)

a. b.

Sono cadute are fallen *(ne) sono *(of them) are fallen some

alcune some cadute

pietre stones alcune


Noting that the obligatoriness of ne is preserved under wh-movement, pronominalization of the N’ and the occurrence of ne can now be used as a detector for the position whmovement takes place from. Rizzi then considers the following data: (33)

quante *(ne) how many *(of them) are fallen

sono

cadute

The obligatoriness of ne indicates that movement could not have taken place from the subject position (in which case ne would be excluded), but must have taken place from postverbal position. Rizzi concludes accordingly that extraction from subject position is not possible in Italian, even in matrix clauses.19 In terms of our terminology, Italian lacks a COMP indexing rule, and resembles Vata in this respect. 3.2 English and Dutch The account presented for French in section 2 carries over to English directly. Control from COMP exists in English (who t came). We can safely assume that COMP indexing is subject to the same conditions as in French: the COMP can only acquire an index at Sstructure, if it dominates a wh-phrase or a wh-trace (explaining the necessary absence of that in cases of long extraction), or if it dominates an operator and that in relative clauses (assuming that that appearing in relative clauses bears an index (Pesetsky 1982)). The situation in Dutch is somewhat different. As has been noted originally by Perlmutter (1971), violations of the that t filter occur in Dutch. Chomsky and Lasnik (1977) suggest that the that t violations in Dutch may be subject to dialectal variation. These dialects have subsequently been called Dutch A and Dutch B by Maling and Zaenen (1978) (or Belgian (sic) (Dutch A) and Dutch (Dutch B) by Bennis (1980)). Typical examples of subject extraction in Dutch A and Dutch B are presented in (34): (34)

a. b.

Wie denk je dat (er) gekomen is Who thinks you that there come is Wie denk je dat *(er) gekomen is Who think you that *(there) come is “Who do you think came”

Dutch A Dutch B

Thus, a trace is allowed in subject position in Dutch A, whereas it is excluded in Dutch B. Maling and Zaenen try to relate the possibilities of that t violations in Dutch A with the optionality of the dummy subject er in the same dialect. (The apparent violations in Dutch A would then be on a par with the apparent that t violations in PRO drop languages such as Italian.) Maling and Zaenen predict that both dialects of Dutch differ systematically with respect to the dummy project er. Bennis (1980) argues that this prediction is not borne out. He proposes instead that the difference between both dialects can be accounted for if one assumes that COMP indexing differs for both dialects. He proposes that in Dutch A, the COMP can get the index of the wh-phrase even if it dominates both the wh-phrase (or


trace) and a lexical complementizer. In Dutch B however, COMP indexing works as in English or in French, and can only take place if COMP dominates the wh-phrase or a trace. Here, we will show that, contrary to Bennis’s claim, that t violations occur in all dialects of Dutch. Moreover, we show that they occur in both matrix and subordinate clauses, and that the behavior of er does not directly bear on the problem of proper government for subject traces. First, we assume that the verb second rule in Dutch moves the finite verb into the complementizer position, as has been proposed by den Besten (1978). Consider now the following sentences which are drawn from the Dutch B dialect which seems to be the more restrictive one as far as extraction of subjects is concerned. (35)

a. b.

(36)

a.

b.

c.

d.

wiei is] *(er) [COMP “Who came” wiei denk je [ti dat] *(er) who think you that

gekomen

there

gekomen is come

[COMP wiei heeft]? *(er) iets gezien who has there something seen “Who has seen something” iets gezien heeft [wiei denk] je [ti dat]? *(er) who think you that there something seen has “Who do you think saw something” [wie heeft] (?*er) hem/Jan gezien who has him/John seen “Who has seen John/him” gezien [wie denk] je [ti dat] (?*er) hem/Jan who think you that him/John seen “Who do you think saw him/John”

is

heeft has

These examples show that er seems to be obligatory with intransitive verbs (35), and with transitive verbs with indefinite objects (36a, 36b), whereas marginally possible with transitive verbs with definite objects (36c, 36d). They show furthermore that there is no contrast for possibilities of extraction between matrix clauses and subordinate clauses. Assuming the appearance of er is dealt with by some independent mechanism, the examples in (36c, and 36d) show that even in Dutch B wh-traces are possible in subject position. We conclude accordingly that Control from COMP exists in Dutch, and that there exists a rule of COMP indexing (37) which allows the COMP node to acquire an index, even if it contains both the wh-phrase and the finite verb, or a wh-trace and the complementizer: (37)

[COMP wh-phrasei…] → [COMPi wh-phrasei…]


This analysis of Dutch thus implies that languages may differ as to the actual form of the rule for COMP indexing.20 According to our account, there is no difference in Dutch between extraction of subjects in matrix clauses or embedded clauses, the reason being that both contain doubly filled COMPs.21 Note that we predict that languages with verb second rules like Dutch will either not allow wh-extraction of subjects at all (lacking COMP indexing), or allow it showing that t violations in both matrix and, providing there exist bridge verbs, embedded clauses. The data in (35) and (36) show furthermore that it is necessary to dissociate Control from COMP and the appearance of er, which has to be accounted for in a different way. We will not elaborate further here on this hazy area of Dutch syntax.22 Consider again verb second phenomena. According to den Besten (1978), English subject/Aux inversion is to be analyzed as a verb second rule which moves the auxiliary into COMP. Still, Dutch and English differ systematically with respect to that t violations.23 Now, consider the following examples: (38)

a. b. c.

[COMPi * [COMP [COMP

whoi ] [ej whoi did ]ei come whoi did] John

came see ei

Curiously, do cannot move into COMP in (38b), whereas the finite verb must move into COMP in Dutch. The account developed so far provides an explanation for this fact, relying crucially on the different functioning of Control from COMP. Suppose that do did move into COMP in (38b). COMP indexing would be blocked and the sentence would violate the ECP in exactly the same way *who do you think t that t came does. Thus, instead of posing a problem for our analysis, the examples in (38) may in fact support it, by providing an account for a different functioning of verb second in English and Dutch. 3.3 In order to account for the crosslinguistic differences which obtain with respect to Control from COMP, we have proposed the Control from COMP parameter. Some languages, we have argued, lack a COMP indexing rule, while others have one. Moreover, there exists crosslinguistic variation as to the conditions under which COMP can acquire an index, (compare English and French versus Dutch). The questions about Control from COMP may be asked in a slightly different way, taking the point of view of the language learner. Given UG and the Control from COMP parameter, the following questions arise: What is the unmarked value or the Control from COMP parameter? What data are sufficient for the language learner to determine the value of the parameter? How does the language learner determine the actual form of the COMP indexing rule? Although we would like to relate the possibility for Control from COMP with independent properties,24 it is sufficient for our purposes here to show how the relevant information can be deduced. Note that we have shown that in a language with syntactic movement, one is forced to apply wh-movement to subjects.


Consider now Vata. The case of Vata is straightforward: given the impossibility of inversion and the obligatoriness of subjects in tensed sentences, the language learner, hearing the resumptive pronoun in subject position, and noting that similar facts hold in focus constructions and relative clauses (which have wh-pronouns in COMP), concludes there is no COMP indexing. Vata does not give any particular insights into the question what the unmarked option in UG is. Indeed, it could represent both the unmarked or the marked case. If it would represent the marked case, it could be argued that there exists positive evidence for concluding the lack of Control from COMP, i.e. the presence of the resumptive pronoun. We do not believe, without much argument however, that Vata represents the marked case. It seems to us that if we consider a hypothetical language in which there are always alternative ways for the extraction of subjects, this language will be of the Vata type, for there is no need for assuming Control from COMP. That is, we assume that the unmarked value for the Control from COMP parameter is the absence of a COMP indexing rule. In this view, Italian lacks Control from COMP, because due to free inversion and null subjects, there does not seem to be any need for assuming it. The language learner of French concludes on the basis of sentences like qui t a telephone (recall qui has to move in syntax) that Control from COMP exists. The actual form of the COMP indexing rule can probably be deduced from surface forms like qu’est-ce que tu penses t qui t a été donné a Pierre, i.e. the COMP node can acquire an index if it dominates more than one element, one of which lexical, with the same index. A similar story holds for English, where sentences like who t came and the door that t opened (that bears an index, following Pesetsky (1982)) lead to the same COMP indexing rule as in French. Moreover, the asymmetric behavior of do (who t came and who did John see t compare (38)) may be sufficient for the language learner to deduce the actual form of the COMP indexing rule (need for domination of elements bearing the same index). The situation in Dutch is slightly different. On the basis of the data discussed in 3.1, there is Control from COMP in Dutch. Now given the verb second rule which moves a finite verb into COMP, the actual rule for COMP indexing differs from the French or English rule: indeed, the COMP may be indexed even if it contains elements with different indices. It thus follows that Dutch systematically differs from English or French in allowing that t violations. Note that, while some important questions are simply begged, we showed that the information concerning the Control from COMP parameter can be deduced in a relatively simple way. 3.4 The grammar of the languages which have been discussed up to this point, all contain the syntactic rule of wh-movement. In most of these languages, wh-phrases occur in situ, either in restricted environments (multiple wh-questions), in general questions or in both. As we have seen, wh-phrases are excluded from subject position at S-structure, since the trace left by the LF rule of wh-raising fails to be properly governed. What happens now in languages which do not derive wh-questions by means of syntactic wh-movement, but always have wh-phrases in situ?


Chinese (Huang 1982) and Mahou25 (a Northern Mande language, spoken in the Ivory Coast) have this property. Our analysis so far leads us to expect that wh-phrases cannot occur in subject position, since the trace left by wh-raising would not be properly governed in LF. This prediction is not borne out: in both languages, there are no ECP effects with respect to wh-phrases in situ (Huang 1982). Given the ECP as a principle of UG, how can these data be analyzed in a way which is consistent with our findings above? It is generally assumed (Chomsky 1981) that the LF rules are subject to very little variation among languages, since direct evidence of their nature is extremely limited. It is possible, and entirely desirable, that whatever variation there may be is in fact limited in the sense that it is in a way a reflex of some aspects of the grammar for which overt evidence is presented to the language learner. Now, the fact that there are no ECP effects with wh-phrases in situ in subject position is only a problem if we assume wh-raising to be subject to wh-raising in all languages, or, put differently, if the only available way for interpreting wh-phrases in situ is by application of wh-raising. Suppose in fact that what is fixed is the actual LF vocabulary, possibly comprising QR (adjunction to S), wh-raising (movement into COMP) and focus interpretation (which, according to Koopman and Sportiche (1983) is not a movement rule, but adopting essentially a proposal of van Riemsdijk and Williams (1980) rather an interpretation rule which consists of the insertion of an operator binding a variable by means of co-indexing). Suppose furthermore that the language learner chooses between the available mechanisms for the interpretation of specific constructions, and that this choice is determined by some overt properties of the language. Concretely, this would mean that in French or in Vata, wh-raising moves a wh-phrase into COMP at LF, because there is overt evidence for doing so (syntactic wh-movement). In Chinese or Mahou however, no syntactic wh-movement applies to wh-phrases. There is thus no a priori reason to suppose wh-raising actually moves the wh-phrase into COMP (or adjoins it to S), while leaving an empty category. Rather, the absence of ECP violations suggest that the LF representations do not contain a trace, and resemble English focus constructions in this respect. We propose tentatively that wh-questions in Mahou or in Chinese are not interpreted by a rule of wh-raising, but rather by a rule which inserts a whoperator which binds the wh-phrase by means of it being coindexed. The LF representations corresponding to the Chinese or French S-structures in (39) and (40) would be roughly (41) and (42): (39)

Zhangsan xiangxin shei mai-le shu Zhangsan believe who bought book “Who does Zhangsan believe bought books”

(Huang 1982: (4))


(40)

(41) (42)

Jean John LF: LF:

[xi

croit believes

que that

[Zhangsan

xiangxin

[COMP quii]

Jean

croit

tu you

as vu saw shei

que

tu

qui who

mai-le as

shu] vu

[e]i

The problem we set out to discuss in this section can be analyzed in a way which is consistent with our analysis, and may lead to new insights into available processes at LF and how they are put to use.

Conclusion In this chapter, we have discussed one particular principle of UG, the ECP, and the crosslinguistic variation surrounding its scope of application. The discussion was entirely restricted to subject/object asymmetries which obtain with respect to both syntactic and LF wh-movement, and to the nature of proper government of wh-traces in subject position. The configuration of data can be summarized as shown in the table on p. 145. A first dichotomy concerns languages with syntactic wh-movement and languages without. Languages with syntactic wh-movement all show ECP violations whereas languages without do not. We proposed tentatively that this is a consequence of how available mechanisms at LF are put to use. Whereas languages with syntactic whmovement apply (in fact, have to apply) wh-raising to wh-phrases in situ, languages without wh-movement interpret their wh-questions by a process similar to Focus interpretation. Turning our attention now to languages with syntactic wh-movement, and leaving idiosyncratic differences aside, we proposed that the explanation of this, at first sight chaotic picture, involves the ECP, for whose LF character we provided strong evidence, and the Control from COMP parameter. First, all the ungrammatical sentences are excluded by the ECP, covering both short and long wh-extraction, and wh-raising in simple and multiple wh-questions. We argued, wh-raising in LF will never lead to proper government of the trace in subject position, but syntactic wh-movement will in some languages. In order to account for this variation, we proposed the Control from COMP parameter, which consists in the presence or the absence of a COMP indexing rule at Sstructure. This COMP indexing rule has the effect of permitting the COMP node itself to properly govern the subject position, by virtue of its being coindexed.



We assumed that the unmarked case for this parameter in UG is the absence of a COMP indexing rule, explaining both Vata and Italian as the unmarked case. We furthermore assumed that languages only make use of the marked option of a COMP indexing rule if no other option is available for the questioning of subjects. Interestingly, it turns out that languages which have Control from COMP differ as to the conditions under which it takes place. These conditions interact with language specific phenomena, such as verb second, yielding systematic patterns of variation (which show there is no difference between short and long wh-movement) which, therefore, place no burden on the language learner. Notes * Research for this paper is supported in part by the Conseil de Recherches en Sciences humaines du Canada (410–81–0503). I would like to thank Noam Chomsky, Dominique Sportiche and Henk van Riemsdijk for discussion on the matters presented here. 1 We adopt the following definition of government (Aoun and Sportiche (1981): α governs β iff φ a maximal projection then φ dominates α iff φ dominates β. 2 This section is a compact version of Subject Object Asymmetries in Vata (Koopman 1980). 3 Vata is a tone language which has four tones represented as T, (high), T (mid high) T (mid) and T (low). For the discussion in this chapter, the following information is relevant: all tensed sentences contain an overt subject, and no subject AUX inversion, nor free inversion occurs. The reader is referred to Kaye (1981, 1982) for details on the phonological system, and to Koopman (1980, 1982, 1983b), Koopman and Sportiche (1983) and Sportiche (1983a) for some aspects of its syntactic structure. 4 This is actually a simplification of the data, since a second surface structure corresponds to (la):

In (i), a low tone follows the wh-phrase. This low tone cannot be analyzed as a floating tone occurring in wh-constructions, but appears actually to be a residue of the optional deletion of the segmental form of the resumptive pronoun which is phonologically conditioned. We thus conclude (la) and (i) are basically similar. 5 It can be shown that the resumptive pronoun occurs in subject position and is not a clitic (Koopman 1983b). 6 The asymmetry is double: on the one hand the distribution of traces is asymmetric. On the other hand, the distribution of resumptive pronouns is so too. That is, resumptive pronouns are excluded from object position (46). A functional explanation may be given for this fact in exploiting a suggestion made by Chomsky (1981). There may be a general principle (the Avoid Pronoun Principle) which has the effect of excluding lexical pronouns whenever their non-lexical variant may be used. This would imply that resumptive pronouns are impossible in object position, since a gap can be used. 7 Although this is a possibility, the option of overt wh-movement is clearly preferred. 8 The tones in (7) and (8) are actually surface tones. We need to refer to surface tones here in order to show that the wh-phrase occurs in subject position for the following reasons. In


Vata, there is a phonological process by virtue of which tones are attracted by a low tone in the appropriate context. This rule can be formulated as follows: Now, as noted in Note 3, there may be just the tonal remnant of the resumptive pronoun in subject position. So, we actually have to assume the above rule of tone attraction took place in order to be able to show that the wh-phrase is excluded from subject position. (Compare (8) where attraction has occurred.) Note finally that the attraction rule is insensitive to the presence of wh-trace as the following example shows:

(i)

yī; what

w they

“Wha t

have

ní NEG -A not

[e]i

n´` do

they

done ”

9 As pointed out by an anonymous TLR reviewer, the fact that wh-phrases may occur as the object of a postposition implies either that P is a proper governor, or that there is no P-stranding in LF. 10 Assuming adjunction to S to yield a proper government configuration for the trace left by QR. 11 Part of this section has been presented at NELS XII, under the title “Theoretical Implications of the Distribution of quoi” (Koopman 1982). 12 Quoi is excluded if it is exhaustively dominated by a Tensed COMP, rightly predicting the occurrence of quoi with pied-piped material in COMP such as (i), taken from Hirschbühler (1978):

(i)

[COMP [NP quoi tu n’aimes pas]]

Pierre a-t-il acheté

13 Goldsmith (1981:562) for example argues that “quoi is unpredictably defective in its paradigm in having no nominative form”. He is however referring to a morphosyntactic use of the term nominative, arguing that the nominative clitic form which exists in the paradigm for pronouns lacks quoi. The arguments given in the text can therefore not be taken as refutations of Goldsmith’s hypothesis, as he has pointed out to me. 14 These examples have been suggested to me by J.R. Vergnaud. Dominique Sportiche has suggested the following examples to me, in which nominative quoi lands in a [Tense] COMP, escaping (14):

(i) (ii)

?? ??

Quii dire qui a est arrive Quoii dire qui ei est arrive

These examples have a marginal status, due to the (near) impossibility of a wh-word originating in a tensed clause to land in a COMP which dominates an infinitival sentence (crossing from a plus to a minus Tensed COMP). These examples show again that there is no contrast between (i) and (ii).


15 The condition that COMP dominates only i-indexed elements is necessary in order to account for cases of long extraction like:

qui tu penses ti quii ti a été battu

(i)

We adopt AHS’s proposal that qui is a complementizer base-generated with an index, contrary to que.

16 Note that this rules out in principle that that deletion or the que→qui rule would apply in LF. 17 Quebec French differs only minimally from standard French in this respect. There seem to exist two dialects with respect to quoi in Quebec French. In one dialect (1), described in Lefebvre (1982b), filter (14) has been generalized to all tensed COMPs (i.e. those containing the complementizer and those without a lexical complementizer). In the other dialect (2), filter (14) functions as in French, and quoi is allowed to appear in Tensed COMPs providing a lexical complementizer is present:

Dialect 1 * quoi que tu fais * quoi tu fais * Je me demande quoi (que) tu fais

Dialect 2. quoi que tu fais * quoi tu fais Je me demande quoi *(que) tu fais

18 I base myself on data from Yoruba provided by Pulleyblanc, and data from Kikuyo presented by Clements at the 1979 Conference on African Linguistics (Illinois). The interpretation of the data is strictly my own. 19 Our discussion of Dutch will show that this argument is not sufficient. In fact, for the argument to go through, it must be shown that extraction of subjects is equally impossible in non-ergative contexts. On this matter, see Sportiche (1983b). 20 This conclusion on Dutch is quite interesting, since it rules out in principle an account in which the COMP indexing rule (27) is taken as a universal rule, and which would derive the apparent absence of a COMP indexing rule in Vata to be the effect of the presence of an abstract complementizer blocking rule (27) from applying. 21 This account predicts the occurrence of indirect questions with doubly filled COMP nodes even if extraction took place from the subject position:

According to my judgment examples like (i) are perfectly acceptable and are given as such in the literature (Koster (1978) example (218a), Reuland (1981) example (39)). According to Bennis (1980) however, (i) is ungrammatical. 22 In the dialect under discussion, there seems to be a correlation between the distribution of indefinite NPs and the extraction site of wh-phrases. In wh-questions in which er is obligatory, the subject has rather been extracted from a properly governed position in the VP:

(i)

wiei

is

er

t

gekomen


who

is

there

come

Interestingly in the corresponding active sentences, indefinite NPs also seem to be excluded from subject position: (ii)

a. b.

?*

iemand is gekomen er is iemand gekomen

23 On these matters see also Koopman (1983a). 24 In Koopman (1983b), it is proposed that the possibility of Control from COMP is related to independently motivated differences concerning the COMP node. In Vata, the wh-position does not really form a unity with the COMP node which probably occurs sentence final. In other words, the sentence initial position in Vata really behaves as a specifier, whereas COMP in English displays certain headlike characteristics. Taking COMP to be the head of S’, and assuming only heads are governors, it follows Control from COMP is possible in English, but excluded in Vata since it does not qualify as a governor (being a non-head). This assumption is in no way incompatible with the account presented in the text. It would imply that Vata and Italian are similar, for different reasons. In Vata, Control from COMP would be excluded in principle, while in Italian Control from COMP is not excluded in principle, but absent since there is no need for assuming it. 25 Thanks to Bamba Moussa for the information on his language, Mahou.

References Aoun, J.N.Hornstein and D.Sportiche (1981) “Some aspects of wide scope quantification”, Journal of Linguistic Research 1(3). Aoun, J. and D.Sportiche (1981) “A formal theory of government”, The Linguistic Review 2(4). Belletti A. and L.Rizzi (1981) “The syntax of ‘ne’: some theoretical implications”, The Linguistic Review 1(2). Bennis, H. (1980) “Coindexing and complementizer-trace phenomena”, paper presented at the 1980 GLOW conference, Nijmegen. Besten, H. den (1978) “On the interaction of root transformations and lexical deletion rules”, paper read at the 1978 GLOW conference. Burzio, L. (1981) “Intransitive verbs and Italian auxiliaries”, unpublished PhD dissertation, MIT. Chomsky, N. (1977) “On Wh-movement” in P.Culicover, T.Wasow and K.Akmajian (eds) Formal Syntax, Academic Press, N.Y. Chomsky, N. (1981) Lectures on Government and Binding, Foris Publications, Dordrecht. Chomsky, N. and H.Lasnik (1977) “Filters and Control”, Linguistic Inquriy 8(3). Epée, R. (1976) “Generative syntactic studies in Duala”, unpublished PhD dissertation, Cornell. George, L. (1980) “Analogical generalizations of natural language syntax”, unpublished PhD dissertation, MIT. Goldsmith, J. (1981) “Complementizers and root sentences”, Linguistic Inquiry 12(4). Hirschbühler, P. (1978) “The syntax and semantics of Wh-constructions”, unpublished PhD dissertation, Amherst. Huang, J. (1982) “Move wh in a language without wh-movement”, The Linguistic Review 1(4). Jaeggli, O. (1980) “On some phonological null elements in syntax”, unpublished PhD dissertation, MIT.


Kaye, J. (1981) “La selection des formes pronominales en Vata”, Revue québécoise de linguistique 11 (1). Kaye, J. (1982) “Harmony processes in Vata”, Rapport de Recherches sur les langues Kru 1. Kayne, R. (1981) “Two notes on the NIC”, in Beletti, A., L.Brandi and L.Rizzi (eds) Theory of Markedness in Generative Grammar, Scuola Normale Superiore Pisa. Kokora, D.P. (1976) “Studies in the grammar of Koyo”, unpublished PhD. dissertation, Indiana University. Koopman, H. (1980) “Subject object asymmetries in Vata”, Publications de l’ILA, Abidjan. Koopman, H. (1981) “Subject object asymmetries, predicate cleft and the ECP”, unpublished manuscript, UQUAM. Koopman, H. (1982) “Theoretical implications of the distribution of quoi”, in NELS XII. Koopman, H. (1982) “Quelques problèmes concernant ‘que/quoi’ et ‘ce que’” in C. Lefebvre (ed). Koopman, H. (1983) “ECP effects in main clauses”, Linguistic Inquiry 14(2). Koopman, H. (1983) The Syntax of Verbs: From Verb-Movement Rules in Kru Languages to Universal Grammar, Dordrecht, Foris Publications. Koopman, H. and D.Sportiche (1980) “Pronouns and the binding theory”, unpublished manuscript, UQAM. Koopman, H. and D.Sportiche (1983) “Variables and the bijection principle”, The Linguistic Review 2 (3) (Chapter 1 in this volume). Koster, J. (1978) Locality Principles in Syntax, Foris Publications, Dordrecht. Lefebvre, C. (1982a) “Le repertoire des mots WH en français vernaculaire et leur insertion dans la grammaire du français”, in C.Lefebvre (ed.). Lefebvre, C. (1982b) (ed.) La syntaxe comparée du français standard et du français populaire: approches formelles et fonctionelles, Office de la Langue Française, Montreal. Maling, J. and A.Zaenen (1978) “The nonuniversality of a surface filter”, Linguistic Inquiry 9(3). May, R. (1977) “The grammar of quantification”, unpublished PhD dissertation, MIT. Obenauer, H. (1976) Etudes de syntaxe interrogative du français: quoi, combien et le complémenteur, Niemeyer, Tübingen. Obenauer, H. (1977) “Syntaxe et interpretation: ‘que’ interrogatif”, Le français moderne 45. Pesetsky, D. (1982) “Complementizer-trace phenomena and the nominative island condition”, The Linguistic Review 1(3). Perlmutter, D. (1971) Deep and Surface Structure Constraints in Syntax, Holt, Rinehart and Winston, New York. Riemsdijk, H.van and E.Williams (1981) “NP-structure”, The Linguistic Review 1(3). Reuland, E. (1981) “Empty subjects, case and agreement and the grammar of Dutch”, in F.Heny (ed.) Binding and Filtering, MIT press, Cambridge, Mass. Rizzi, L. (1982) Issues in Italian Syntax, Foris Publications, Dordrecht. Sportiche, D. (1983a) “Bete reciprocals and clitic binding”, in Kaye, Koopman, Sportiche and Dugas (eds) Current Approaches to African Linguistics, vol. 2, Foris Publications, Dordrecht. Sportiche, D. (1983b) Structural Invariance and Symmetry in Syntax, unpublished PhD dissertation, MIT.

6 A NOTE ON LONG EXTRACTION IN VATA AND THE ECP* with Dominique Sportiche

Arguments of a predicate and adjuncts behave differently with respect to Wh-extraction. Most current assumptions attribute this asymmetry to a single requirement on Wh-traces that adjunct traces fail to observe. Huang (1982) and Lasnik and Saito (1984) (henceforth HLS) suppose this requirement follows from the ECP.1 Aoun (1984, 1986) assumes it to be the principle A of the Generalized Binding Theory (henceforth GEBI). This chapter presents new data from Vata, a West African language of the Kru family. Our main purpose here is to show that such analyses as HLS’s and Aoun’s are not sufficient to account for the full range of extraction possibilities and the distribution of Wh-traces. Instead, we argue that two distinct principles must be postulated: first, a principle like the ECP, requiring empty categories to be governed in an appropriate sense and, in addition, a second principle regulating the distance between a target of Wh-extraction and its immediate antecedent. Many possible alternatives come to mind to solve the problems that the Vata data reveal. Here we will limit ourselves to showing that the Condition on Long Extraction, argued for in Koopman and Sportiche (1985, 1986) will handle these problems straightforwardly and better than some existing alternatives. The chapter is organized as follows. Section 1 briefly describes the HLS account. In section 2, we present the Vata data. We discuss its theoretical implications and show why they pose problems for the HLS account. In section 3, our analysis is sketched. In section 4, we show why GEBI is not sufficient by itself and we briefly discuss the extension of GEBI proposed in Wahl (1985).2 1 The HLS ECP analysis Most of the central facts concerning argument/adjunct asymmetries have been presented in Huang (1982). He observes that, in Chinese, the possible scopes of a Wh-phrase in situ inside a Wh-island depends on whether it is an argument or an adjunct. If it is an argument, it can have wide scope as in (1). But if it is an adjunct, it can only have narrow scope as in (2a), even though Wh-adjuncts can have wide scope in other clauses such as (2b). (1)

ni you

xiang-zhidao wonder

[shei mai-le who buy

ASP

sheme] what


(i) (ii)

Who is the person x such that you wonder what x bought? What is the thing x such that you wonder who bought x? (Huang: 525)

(2) a. ni xiang-zhidao [shei weisheme mai-le shu] (i) Who is the person x such that you wonder whether x bought the book? (ii)* What is the reason x such that you wonder who bought the book for x? (Huang: 525) b. ni renwei [ta weisheme meiyou lai] you think he why not come “Why do you think that he did not come?” (Huang: 534) Huang proposes that this general complement/non-complement asymmetry under LF movement follows from the ECP, in a fashion exactly parallel to the subject/ object asymmetry observed for syntactic movement in languages like English. Recall how this asymmetry is handled. Syntactic movement of an object leaves a lexically governed trace, hence satisfying the ECP. Syntactic movement of an (English) subject leaves a nonlexically governed trace. The ECP requires antecedent government of this trace. This requirement is satisfied when the adjacent COMP position inherits—through the COMPindexing rule of Aoun, Hornstein and Sportiche (1981) (henceforth AHS)—the index of the moved subject (or its trace) that it contains. This COMP coindexed with the subject trace becomes its antecedent governor. Huang generalizes this account to LF movement of all phrases in Chinese (and other languages), and to syntactic movement of adjuncts in all languages. Huang argues that after the application of Wh-movement in LF, traces of arguments are lexically governed (either by the verb, or—a property specific to Chinese—by INFL).3 Wh-arguments can thus be interpreted as having wide scope, as in (1). Traces of adjuncts, however, being adjoined to VP, are lexically governed neither by V nor by INFL. The ECP can therefore only be satisfied through antecedent government. Huang then proposes that (LF or syntactic) traces of adjuncts satisfy the ECP exactly as traces of (English) subjects do: a COMP that is coindexed with an adjunct trace acts as antecedent governor for it. This implies that a trace of an adjunct always needs an antecedent governor in the clause in which it occurs. The interpretation (2a), with why having wide scope, is then correctly excluded: why cannot move into the immediately adjacent COMP since that COMP will be occupied by who, and its trace will therefore not be antecedent governed. The interpretation in (2b) is possible, provided that LF Wh-movement is successive cyclic, and that COMP-indexing occurs at LF. (In this, Huang’s account differs from AHS.) Generalizing the usual account for subject-object asymmetries to complement/ noncomplement asymmetries creates a problem, as Huang notes. One would expect that, in a language like English, subjects and adjuncts behave in the same fashion, since neither subjects nor adjuncts are lexically governed, and this predicts that there is a two-way

LONG EXTRACTION IN VATA AND THE ECP 149

distinction (subject-object versus complement) with respect to the ECP. This is incorrect, since there is in fact a three-way distinction: COMP-indexing for antecedent government of subject traces is sensitive to the presence or absence of that, but COMP-indexing for antecedent government of the adjunct trace is not:4 (3)

a. b.

* Who do you think [t that [t left]] Why do you think [t that [John left t]]

Lasnik and Saito (1984) further develop and refine the analysis of Huang. They propose in addition that a trace in COMP itself must be antecedent governed, and develop a system that accounts for the three-way distinction noted above. Their system proposes essentially that the ECP applies at S-structure only to arguments, and to all empty categories (arguments, adjuncts, and traces in COMP) at LF (and in particular after LF deletion of that in (3b)). 2 The Vata Data Let us start by reviewing the basic Wh-movement patterns in Vata, as well as their analysis (see Koopman (1984) for more details). We will limit ourselves to cases of syntactic Wh-movement,5 and discuss in turn short Wh-movement (sections 2.1 and 2.2) and long Wh-movement (section 2.3). 2.1 Short Wh-movement of subjects, objects, subcategorized PPs and verbs The pattern of Wh-movement of a phrase to the initial position of its clause or, through “COMP-to-COMP” movements without skipping any intermediate COMP, to the initial position of a higher clause, is presented in (4) below.6 We will henceforth refer to this type of movement as SHORT WH-MOVEMENT, contrasting this with LONG WHMOVEMENT, such as movement out of a Wh-island. (4) a.

b.

Subject: àlÓ Ò/*[e] who he-R “Who did it?”

Subject: àlÓ n gū;gū; who you thought “Who did you think did it?” Direct Object:

nā; NA

nÙ mÍ l did

Ò/*[e] he-R

it

nÙ did

WH

mI it

l WH


c.

d.

yĪ; Kòfí nÙ what did “What did Kofi do?” Subcategorized PP: nánÍ` slé mlÍ y what that house in “In which house did Kofi put it?”

kòfí

[e]

l WH

zU ` put

mI [e] l it WH

If the target of extraction is in the subject position, a resumptive pronoun must appear; if it is the direct object (or the indirect object or a subcategorized PP), a trace occurs in the extraction site. Besides Wh-movement of the usual categories, there is another instance of Whmovement in Vata: Wh-movement of main verbs, a process that underlies the predicate cleft construction. In this construction, the main verb occurs in initial position, and is understood as focused. The sentence contains a copy of the verb, which we refer to as a RESUMPTIVE verb. This process exhibits the usual diagnostic properties of Whmovement, being apparently unbounded, subject to the usual constraints on Whmovement, and so on. It is discussed at length in Koopman (1984, chapter 6).7 (5) a.

b.

Wh-movement of V: nŪ; kòfí do “Kofi will DO it.”

gū;gū; nā; *nŪ; do I think NA “I think that Kofi will DO it.”

ká F-AUX

kòfí

mÍ

ká F-AUX

n it

do

mI it

*nŪ; do

Given that Subjacency is the diagnostic property for movement, we treat (4a, b) and (5) as cases of syntactic Wh-movement: the resumptive pronoun and the resumptive verb must be subjacent to their respective antecedents. The resumptive pronoun, in addition, behaves in all other respects like a Wh-trace: it participates in weak crossover effects (Koopman and Sportiche 1982), and does not prevent licensing of parasitic gaps (Sportiche 1983). So far, the distribution of empty categories (as opposed to resumptive pronouns or verbs) in Wh-constructions shows that traces (i.e., empty categories) are not allowed in precisely those cases in which the extraction site is not lexically governed. This follows from the ECP (see Koopman 1983, 1984), provided preposed Wh-phrases or verbs are not antecedent governors: a trace in subject position or verbal position cannot obey the ECP. To avoid an ECP violation, an overt category, i.e., a resumptive pronoun or verb, occurs. For concreteness assume that these resumptive forms arise through a process of trace spell-out.8 That preposed Wh-phrases are not antecedent governors need not be


stipulated; it follows from the fact that in Vata these elements do not occur in (or adjacent to) COMP. They occur clause initially but the COMP-position in Vata is clause final. Percolation of the index to the governing head—the mechanism of COMP indexing— cannot therefore occur. 2.2 Short movement of adjuncts Let us next turn to Wh-extraction of adjuncts. The following discussion is restricted to extraction of reason, cause and manner adjuncts.9 The syntactic behavior of manner and reason/cause adjuncts is illustrated in the examples (6)-(8): (6)

dtdÒdldÒ sū;Ō l y sC how you cut-M-cut-M tree-DET WH “How did you cut the tree?” y s k sū;Ō d dÒdlÌdÒ how you FUT-A tree-DET cut-M-cut-M “How will you cut the tree?”

a.

b.

l WH

If a manner adjunct is extracted, a special form of the verb must appear: a suffix (#CV or #1E)10 is attached to the base form of the verb and the newly formed verb is reduplicated. (7)

a.

b.

c.

d dÒ sū;Ō y s how you cut-M tree-DET “How come/why did you cut the tree?” d IÈ sū;Ō y s how you cut-M tree “How come/why did you cut the tree?”

n ká sÓ like-this I said I FUT-A “It is like this that I said I cut a tree.”

sū;Ō tree-DET

l WH l WH

* / dÌdÒ/d lÈ *cut / cut-M

If a reason or cause adjunct is extracted, as in the examples of Wh-question or topicalization in (7), the verb must carry a particular suffix (#CV or #1E). This is actually the same suffix as in (6). The verb, however, is not reduplicated. The Wh-phrase together with the particular type of morphology in (6) and (7), yields the interpretation of manner and reason or cause respectively. We will henceforth refer to the morphology in (6) and (7) as the adjunct morphology, and to these adjuncts as adverbial adjuncts or non-PP adjuncts. The presence or absence of the adjunct morphology depends on the internal structure of the preposed Wh-phrase: it is obligatory if the Wh-phrase yEsO - a non-PP adjunct - is


extracted, but excluded if a PP adjunct is extracted, as in the example in (8) where gbU is a postposition: (8)

k yĪ; gb what reason you FUT-A “Why are you going to cut the tree?”

sū;Ō tree-DET

/*dÌdÒ cut/*cut-M

The adjunct morphology, furthermore, occurs only if the adjunct has been extracted by Wh-movement. It may not occur in sentences like (9) and (10) with sO, the non-Whcounterpart of yEsO, which differ minimally from (6) and (7): (9)

(10)

d /*d dÒ(dÌdÒ) I cut/*cut-M “I cut the tree quickly.” n ká sÓ I said I FUT-A like-this “I said I will cut the tree like this.”

sū;Ō tree-DET

sū;Ō tree-DET

fáfá quickly

/* d dÒ(dIdÒ) cut/* cut-M

So far, then, adjunct extraction in Vata displays the following characteristics: (11)

The verb carries adjunct morphology if a non-PP adjunct is extracted.

Any analysis has to account for the two parts of (11): (i) why must adjunct morphology appear if a non-PP adjunct is extracted, and (ii), why cannot it appear if an adjunct is not extracted or if a PP adjunct is. Let us start with the first problem: why must adjunct morphology appear on the verb in Vata if a non-PP adjunct is extracted. The relevant difference between the examples in (6) and (7) and the examples in (9) and (10) seems to be the appearance of an adjunct trace. An adjunct trace must be properly governed: this requirement is usually fulfilled by antecedent government from COMP. But recall that in Vata there is no antecedent government from the initial position. Questions therefore arise: how can adjuncts be extracted at all (we will return to this problem below), and how can an adjunct trace satisfy the ECP? Since antecedent government does not obtain, the only other possibility is that the trace of an adjunct is in some other way properly governed. It is natural to assume that adjunct morphology precisely serves this purpose: it provides a way to fulfill the proper government requirement imposed by the ECP. There are several ways to obtain this result. Suppose for example that adjunct morphology allows the extension of the government domain of the verb to include the adjunct marked by the verbal morphology, i.e., [V+M] (compare 12a).


If this is correct, the adjunct will be lexically governed by the verb. Or, suppose alternatively that the projection of [V+M] is (12b), with V the head of the theta projection Vmax and M the head of the projection of [V+M]:

In this structure, the adjunct trace will be governed by M. This can be conceived of either as a case of lexical government—the morphology will subcategorize for it, and subcategorization is often considered the core case of lexical government— or as antecedent government. We will assume it is a case of lexical government. Whether it is lexical government or antecedent government, however, is irrelevant to our arguments, as we will show in section 2.3. What counts is that the adjunct trace is properly governed. The ECP thus accounts for the obligatory appearance of adjunct morphology: in Vata preposed Wh-phrases do not antecedent-govern their traces. The trace of an adjunct must therefore be licensed in a different way. Since a bare verb will not properly govern the adjunct trace, adjunct morphology must appear on the verb so as to make proper government possible. This analysis establishes a close parallel between the Vata adjunct morphology phenomenon and COMP-trace effects. When movement of an English subject takes place, that must delete, because of the ECP. When movement of a French subject takes place, the que→qui rule (Kayne 1972) must apply for the same reason. Let us next turn to the second problem of (11): why is adjunct morphology excluded in the following examples? (13)

a.

b.

c.

d /*d dÒ(d dÒ) sū;Ō fáfá I cut tree-DET quickly “I cut the tree quickly.” n ká sÓ sū;Ō I say I FUT-A like-this tree-DET “I say that I will cut the tree like this.” ká sū;Ō yĪ; gb What reason I FUT-A tree-DET “Why are you going to cut the tree?”

/*d dÒ(d dÒ)** cut /*d dÒ cut/*cut-M

The grammaticality of (13a, b) and the contrast with (6) and (7) indicates that the licensing conditions for the appearance of phonologically realized adjuncts and adjunct traces are different. It is sufficient for a lexical adjunct to be a sister of a predicate as in


(14a);11 this is insufficient for the licensing of an adjunct trace, as in (14b), basically, as we have shown above, because of the ECP:

The exclusion of adjunct morphology in (13a, b) is part of a pattern found much more generally across languages. It is an often made observation that languages seem to adopt “minimalist” strategies as unmarked strategies when possible: licensing processes are invoked only when necessary. The French que→qui rule mentioned earlier is a case in point. It applies only when necessary. Consider, in this light, the resumptive pronoun strategy in Vata. Resumptive pronouns may only appear in subject position (15a), and are excluded from object position (15b): (15)

a. b.

Wh-phrase [*[e]/ res. pron… Wh-phrase [NP…[VP… [e]/*res. pron. V].

Wh-traces in object position obey all the demands the relevant principles impose on them: they are locally A-bar-bound, they are subjacent to their antecedent, and they satisfy the ECP by being lexically governed. No other licensing condition is necessary. Because of the minimalist strategy, insertion of a resumptive pronoun is excluded. The impossibility of adjunct morphology in (13a, b) represents exactly the same case: these adjuncts may occur by virtue of their sisterhood to VP, the only licensing condition they need to fulfill. Applying adjunct morphology would lead to non-minimality, and is therefore ruled out. The absence of antecedent government from initial position yields an account of the obligatoriness of adjunct morphology in (6) and (7). But how can PP adjuncts be extracted in examples of the type illustrated in (13c), and why is adjunct morphology excluded in these cases? The contrast between (6 and (7) on the one hand and (13c) on the other shows that the internal structure of the adjunct must be taken into account. While the adjuncts in (6) and (7) cannot be related to any major syntactic category, the adjuncts of the type illustrated in (13c) are clearly PPs. If PP adjuncts may be extracted, it must be the case that they satisfy the ECP at the relevant level of representation, i.e., LF. To overcome a similar problem, Huang (1982) proposes an analysis that we adopt here. He assumes (i) the need for a LF process of reconstruction, and (ii) that Ps are proper governors—at least at LF.12 This implies that at the LF level of representation the S-structure (13c) is represented as follows: (16)

LF:

yI…[VP [PP t

gbU] [VP…V]


what

for

In (16), the trace will satisfy the ECP by virtue of being properly governed by P; PP adjuncts, therefore, can undergo short Wh-movement. The fact that adjunct morphology cannot appear with PP-adjunct extraction can be accounted for by the minimality requirement discussed above: it suffices for a PP-adjunct trace to be a sister of VP, to be locally A-bar bound, and to satisfy the ECP. All these conditions are fulfilled, and appearance of adjunct morphology would violate minimality. Our analysis makes an interesting prediction. Suppose that there exists a language which shares with Vata the absence of antecedent government from initial position, but which differs from Vata in not having any morphological process that can be used to circumvent the ECP. One would expect that all adjuncts under extraction are PPs. This case is illustrated exactly by Koyo, a language very closely related to Vata, with a much less rich system of verbal morphology. In particular, adjunct morphology is lacking. Just like Vata, Koyo possesses bare lexical adjuncts like sO “in this way”. However, contrary to what happens in Vata, it is impossible to topicalize it bare, as in (7c), exactly as we would expect. The only way it can be topicalized is in a PP of the form sO mI “in this way”. 2.3 Long Wh-movement HLS derive the difference in locality requirements on complements and adjuncts from the ECP. Complements are lexically governed, and can undergo long Wh-movement, provided that Subjacency is obeyed in the case of syntactic movement. Traces of adjuncts, however, (as well as of subjects in languages like English and French) can only satisfy the ECP through government by an antecedent or its trace in COMP. Moreover, a trace in COMP that ensures antecedent government must be antecedent governed in turn. As shown above, traces of (non-PP) adjuncts in Vata are properly governed by the verb carrying adjunct morphology. Furthermore, subject and verb extraction escape the ECP, since an overt category—either a resumptive pronoun or a resumptive verb—appears in the extraction site. Given this state of affairs, the HLS account makes the prediction that there should be no asymmetry between the long extraction possibilities of complements, non-PP adjuncts, subjects and verbs. This prediction can be very simply tested. As shown in Koopman (1984), Vata resembles French or Italian (Rizzi (1984); Sportiche (1981) in allowing the Wh-island constraint to be violated, thus indicating that 13 and NP are Bounding nodes for Subjacency: (17) a.

n [ zĒ [ k àlÓ who you NEG-A thing you FUT-A yì l know WH “Who don’t you know what you will give to?”

REL

[e]

ny ] give


b. táblÒ k nÍ [ zĒ table on THERE I NEG-A what I yì z put know “It is on the table that I don’t know what I will put.”

ká FUT-A REL

[e]

In (17a), a direct object and in (17b), a subcategorized PP, have been extracted. Their traces must—and do—satisfy lexical government, since antecedent government is blocked, the other Wh-phrase occupying the intermediate COMP position. Consider now the case of a Wh-moved non-PP adjunct. We have argued that its trace is properly governed by the adjunct morphology. It should therefore be extractable out of a Wh-island. The following examples illustrate that this prediction is not borne out: (18)

a.

yl nyn [zĒ d dÒdÌdÒ * y s ` how you wonder thing you cut-MORPH “How do you wonder what you cut t?”

yl nyn [nyÓ` Ò b. * y sÓ how you wonder who he-R d dÒd dÒ l cut-MORPH WH “How do you wonder who cut it t?” n [zĒ Ò c. * y sÓ why you NEG-A thing he yì l know WH “Why don’t you know what he cut t?”

ká FUT-A

ká FUT-A

REL

[e]] l WH

Ó [e] mÉ] REL it

- Ó [e] REL

d dÒ] cut-M

Even though the adjunct trace is properly governed, long movement is impossible. This pattern of data establishes that it cannot be the ECP which accounts for the strict locality of adjunct movement.14 The same conclusion can be reached when we consider long extraction of subjects and verbs, which also yield ungrammaticality: n [zĒ mĒm ` (19) a. * àlÓ who you NEG-A reason it-it t mÉ] yì l it know WH “Who don’t you know why t cut it?”

gb for

Ò d ` he-R cut REL

Thus, neither subjects nor verbs nor adjuncts may be extracted out of a Wh-island, despite the fact that their traces, being either non-null or lexically governed, obey the ECP.


PP-adjuncts, finally, cannot be long-extracted either: (20)

a. * yĪ; gb ni [zĒ what cause you NEG-A what you yì l know WH “Why don’t you know what to do t?”

ká FUT-A

REL

t t nū;] do

This constitutes an additional problem for Lasnik and Saito’s analysis: after reconstruction, the P properly governs its complement. Since the ECP only applies at LF to adjuncts, the structure should be possible.15 Adjuncts, subjects and verbs thus obey strict locality, they may only undergo short extraction; but objects and subcategorized PPs are long extractable. The ECP, as formulated in HLS, can therefore not account for the strict locality imposed on nonlexically governed targets of extraction. 3 Analysis The data discussed so far are summarized in the table below; recall that by short movement of a phrase, we mean movement to clause initial position of the first clause containing that phrase.

In the case of short movement, we have seen that the ECP accounts for the distribution of empty categories and forces the appearance of a resumptive pronoun (a.ii), the obligatoriness of adjunct morphology (c.ii), and the appearance of a resumptive verb (d.ii). But, as we have shown, it cannot account for the configuration of data found with long Wh-extraction. Another principle regulating the distance between the antecedent and the target of extraction must be involved. Various possibilities come to mind to cover the cases not handled by the ECP, e.g. require both lexical government and antecedent


government and separate the notion of antecedent government from the ECP so that it applies to overt traces in the desired way (as we in fact propose in Koopman and Sportiche 1985, 1986). Our purpose here is not to discuss the form such a principle might take. We would merely like to point out that, if it implies the following condition, it will cover the data successfully: Condition on Long Extraction:

(22)

x is a possible long extraction site if x is a theta-position.

We motivate this condition extensively elsewhere (Koopman and Sportiche 1985, 1986). Here we will show how it very simply accounts for the data discussed. Direct objects and subcategorized PPs are long extractable, as predicted by (22) since they appear by definition in theta-positions. Adjuncts, whether adverbial or PP, do not appear in thetapositions. They are modifiers rather than arguments: for them, only short extraction is possible. Finally, verbs are predicates rather than arguments. They do not appear in thetapositions either and are thus only short extractable. Subjects seem to pose a problem. It is usually assumed that a subject does appear in a theta-position when it is the external argument of some verb. We argue in detail in Koopman and Sportiche (1985, 1986) that this assumption is incorrect, at least as far as languages like English, French, and Vata are concerned. Rather we show that the subject position (of clauses) in these languages is never a theta-position,16 because their basic sentential structure is as in (23) below (where SC is a small clause with VP head), leaving order aside:

For convenience, we can describe t. as the base generated position of subjects, assigned the external theta-role of the VP, and NPi as the position in which lexical subjects or subject Wh-traces must appear at S-structure. Since this position is always a non-theta position, only short extraction is permitted. In our view, it is the independent strong plausibility of this sentential structure that makes the Condition on Long Extraction so appealing.


4 Remarks on GEBI We now briefly consider the Generalized Binding theory proposed by Aoun (1984, 1986). We have seen that two principles are required: one along the lines of the ECP, and a second along the lines of the Condition on Long Extraction. GEBI is an attempt to generalize principle A of the Binding theory so as to entirely eliminate the need for the ECP. Basically, this generalization requires elements of a certain set H to be bound in a certain structural domain D. H is defined to contain English subjects (but not Chinese subjects) and non referential elements (including adverbial adjuncts, and traces in COMP). Given x a member of H, x must be bound in D, where D is the first S containing x. Our previous argumentation shows that the ECP cannot be entirely reduced to GEBI.17 This can easily be seen. Movement of adverbial adjuncts is short and triggers the insertion of the adjunct morphology. GEBI cannot account for both properties. As formulated, it predicts short movement for adverbial adjuncts, but says nothing about the adjunct morphology. The same applies to subjects (or verbs). GEBI predicts short movement of subjects but says nothing about the obligatory spell-out of their traces. Note that taking domain D to be S in Vata rather than S would not work: it could explain why subject and verb traces must be spelled out or why the adjunct morphology must appear. (It would count as antecedent for the adverbial adjunct trace.) But it would fail to explain why movement of subjects, verbs, adverbial adjuncts and PP adjuncts must be short. In conclusion, GEBI must also be supplemented by some other principle. This conclusion has been independently reached by Wahl (1985), which proposes a system very close to ours in its division of labor.18 Specifically, it is argued that GEBI applies to LF representations, but that the ECP (limited to head government) applies on the PF side of the grammar. As it stands, the GEBI/ECP approach faces a problem that our present account does not face. The problem has to do with PP adjuncts. Since the ECP applies in PF, it must be assumed that PP adjuncts are head governed. In order to derive the fact that only short movement of PP adjuncts is possible, it must also be assumed that they fall under the GEBI requirement that they be bound in D at LF. Since they are not subjects, they must, in order to fall under GEBI, be considered nonreferential. Wahl otherwise assumes that PP adjuncts in, e.g., English, French, and Chinese are referential. The necessary assumption for Vata appears ad hoc: Wahl’s framework would therefore seem to require some modification.19 Note that adopting the idea that there is reconstruction of the P, as we have assumed, would not work in Wahl’s system. At LF, the trace governed by the reconstructed P would be an NP, and as such referential. This trace would therefore not be subject to the locality requirement imposed by GEBI: the impossibility of long movement of PP adjuncts would not be accounted for.


Notes * Research for this paper was supported in part by a grant from the Social Sciences and Humanities Council of Canada #410–84–503. We wish to thank three anonymous reviewers for their comments on an earlier version. 1 Recall that the ECP requires empty categories to be governed either by a lexical head or by a position containing an antecedent, this last case being called antecedent government. 2 There are of course other proposals not discussed here, in particular Stowell (1985) and Kayne (1983), but see Koopman and Sportiche (1986). 3 This assumption accounts for the absence of subject/object asymmetries in Chinese. 4 We assume that (3) is grammatical. For a different view, see Wahl (1985). 5 Wh-phrases in situ seem to have a very limited distribution (compare Koopman (1983) for a partial distribution); it is clear that they are excluded from subject and adjunct position (which is exactly what we expect—see below). It is also clear that they may appear in object position or inside a subcategorized PP in main clauses. They seem to be excluded from all other positions, but contrastive judgements are difficult to elicit. 6 For extensive discussion of the syntax of Vata, a language of the Kru family spoken in the Ivory Coast, see Koopman (1984). Here we use the following conventions:

– – – – – – – – –

Tones are indicated as:′ High Tone; 'Mid High;-Mid, and ` Low; I, U, E, and O are [-ATR] vowels with the [+ATR] counterparts i, u, e and o; DET: determiner; WH: particle indicating the clause is a Wh-question; REL: particle indicating that the clause is a relative clause; FUT-A, NEG-A: future or negative auxiliaries; IT, THERE: preposed pronouns in topicalizations; NA: marker of subordination; R: resumptive pronoun.

7 This construction shows that the Head movement constraint of Travis (1984, section 3.4.1), which states that a head can only move into a head position that properly governs it, is too strong: Wh-movement of the verb violates this condition: in examples like (5) the verb moves into a position from which it would not properly govern the extraction site (see below). It is argued in Koopman (1984) that the properties of verb movement covered by the Head movement constraint basically follow from the ECP. (See Koopman (1984:177ff.) for discussion.) 8 Note that there is no principled reason whatsoever why traces should be empty. A better name than resumptive pronoun or verb would be overt trace, as opposed to covert trace, since the term resumptive is usually applied to cases not involving movement. 9 Locatives and temporals behave in the same way as in Chinese, English, and so on (Huang 1982). For the behavior of other adjunct types, see Koopman and Sportiche (1985). 10 The actual form of the suffix is either lE/lO or CE/CO, with C a copy of the preceding consonant. The former morphology is productive; the latter only appears with certain frequently used verbs. 11 Note that languages vary in the licensing conditions governing the appearance of adjuncts. In Poular (also called Fula), for example, certain adjuncts may appear only if the verb carries the appropriate morphology (Arnott 1970).


12 Theta role assignment to a complement is the core case of proper government. It thus seems minimal to assume that Ps are proper governors. One reviewer points out that the assumption that P is a proper governor raises the problem of why there is no P-stranding in a language like French. Huang (1982) argues that the possibility of having P-stranding is a PF property, not necessarily related to the ECP. (See section 4 for additional remarks.) It appears plausible to try to relate the possibility of P-stranding to the existence of verb particle constructions, that is, with an overt difference independent of proper government as suggested by the Germanic and Kru languages. 13 Although indirect questions superficially resemble relative clauses, their behavior with respect to Subjacency is that of an S, not of a relative clause. We will therefore treat indirect questions as (syntactic) S (compare Koopman (1984) for discussion). 14 In section 2.2, we stated that it was irrelevant for our purposes whether adjunct morphology satisfies the ECP through lexical government or through antecedent government: in both cases long Wh-movement should be possible, either because the trace would be lexically governed, or because it would be antecedent governed by lexical material, i.e., by something that is not a trace, and that therefore need not be antecedent governed in turn. 15 This kind of example does not constitute a problem for Huang, since he assumes that the ECP applies at S-structure and at LF to traces; (20), therefore, would violate the ECP at Sstructure. (See Huang (1982) for discussion.) 16 We claim that this is precisely the difference between these languages and languages like Chinese and Japanese which allow long extraction of subjects. In Japanese, for example, we propose that subjects of clauses may appear in theta-positions: essentially, their basic clausal structure is reduced to I in (23) below, with subjects appearing in the position of t. 17 The same conclusion applies to the Connectedness approach of Kayne (1983), but for other reasons: in Vata, both subjects and objects are canonically governed from the right, yet behave differently with respect to extraction. 18 This is also true of Stowell’s (1985) proposal. 19 We believe that this by itself shows that the notion of referentiality is not the relevant notion, but see Koopman and Sportiche (1986) for additional discussion.

References Aoun, J. (1984) The Formal Nature of Anaphoric Relations, MIT Press, Cambridge, Mass. – –(1986) Generalized Binding, Foris Publications, Dordrecht. Aoun, J., N.Hornstein and D.Sportiche (1981) “Some aspects of wide scope Quantification”, Journal of Linguistic Research 1(3), 69–95. Arnott, D.W. (1970) The Nominal and Verbal Systems of Fula, Oxford University Press, London. Huang, C.-T. J. (1982) Logical Relations in Chinese and the Theory of Grammar, MIT PhD dissertation, to be published in revised form by D. Reidel, Dordrecht. Kayne, R. (1972) “French relative que”, Recherches Linguistiques 2, Université de Paris 8. – –(1983) “Connectedness”, Linguistic Inquiry 14(2) 223–249. Koopman, H. (1983) “Control from COMP and comparative syntax”, The Linguistic Review 2(3) 365– 391 (Chapter 5 in this volume). – –(1984) The Syntax of Verbs: From Verb Movement Rules in the Kru Languages to Universal Grammar, Foris Publications, Dordrecht. Koopman, H. and D.Sportiche (1982) “Variables and the bijection principle”, The Linguistic Review 2 (2) 139–160 (Chapter 1 in this volume).


– –(1985) “Theta-theory and extraction”, GLOW Newsletter 14, paper presented at the 1985 GLOW colloquium in Brussels. – –(1986) “Covert categories and theta-theory”, unpublished manuscript, UCLA. Lasnik, H. and M.Saito (1984) “On the nature of Proper Government”, Linguistic Inquiry 15(2) 235– 289. Rizzi, L. (1984) Issues in Italian Syntax, Foris Publications, Dordrecht. – –(1981) “Bounding nodes in French”, The Linguistic Review 1(2) 219–246. Sportiche, D. (1983) Structural Invariance and Symmetry in Syntax, MIT dissertation, to be published by Foris Publications, Dordrecht. Stowell, T. (1985) “Null antecedents and proper government”, communication presented at the 16th NELS Conference in Montreal. Travis, L. de Mena (1984) “Parameters and the effects of word order”, unpublished PhD dissertation, MIT. Wahl, A. (1985) Two Types of Locality, unpublished manuscript, University of Maryland and USC.

Part III SPECIFIERS AND PHRASE STRUCTURE

7 ON THE ABSENCE OF CASE CHAINS IN BAMBARA *

1 Introduction In this chapter, I want to argue that CASE CHAINS are generally absent in Bambara, a Mande language spoken in Mali. Case chains arise when an NP that needs to satisfy the Case filter occurs in a position where it is not governed by the element that assigns Case to it. VERBAL CHAINS arise as a consequence of V movement. For example, when a verb moves to INFL, a verbal Case chain is formed whereby the Case-assigning properties of the head of the chain are transmitted through the chain to the trace of the verb. Case chains also occur in the nominal system when an NP satisfies the Case filter through Chain formation with an expletive pronoun. (Compare There are people in the room.) In the spirit of Koopman (1984), I will argue that in a particular language, Case chains are either absent or present for all types of chains, i.e. both nominal chains and verbal chains. Case chains will be argued to be absent in a language like Bambara, but present in languages like French or English.1 In this chapter I will first focus on the description and analysis of certain aspects of the verbal and nominal syntax in Bambara. Discussion of parametric variation will be confined to section 4. Before turning to the actual discussion and analysis of the data, I will make explicit the particular theoretical assumptions I adopt, and present some necessary background information on the properties of surface word order in Bambara, as well as an analysis. The theoretical assumptions underlying the analysis presented here are the standard theoretical assumptions of the so-called GOVERNMENT-BINDING THEORY, unless indicated otherwise. The THETA CRITERION, the PROJECTION PRINCIPLE and the CASE FILTER will be presupposed. I further assume the EXTENDED X-BAR THEORY and the definition of GOVERNMENT of Chomsky (1986a), according to which both a head and its specifier position ([SPEC, XP]) are governed by an external governor. I will also assume the now-standard VP-INTERNAL-SUBJECT hypothesis (Koopman and Sportiche (1985, 1991); Kuroda (1986); Kitagawa (1985); Larson (1988); Fukui (1986); Speas (1990); among others), in which thematic subjects originate as sisters to their maximal projection, i.e. as sisters to XP. In languages with an obligatory [SPEC, IP] position, like English, the subject must raise to [SPEC, IP] (a non-theta position).

THE ABSENCE OF CASE CHAINS IN BAMBARA 165

This movement is forced by Case theory, [SPEC, IP] being the position to which nominative Case is assigned by INFL. The treatment of external arguments as sisters to maximal projections makes it possible for the [SPEC, XP] position, which is in principle available, given X-bar theory, to contain other material. I will argue below that this is the case in Bambara: the direct object occurs in the [SPEC, VP] position. Bambara is a language with extremely rigid word order. The linear word order in tensed sentences is presented in (1): (1)

(AdvP/S′) NP1 INFL (NP) V (PP) (ADV) S′

Tensed sentences must contain an overt NP (indicated as NP1 in (1)) preceding an INFL (for a complete list of elements occurring in INFL and their properties, see Tables I and II below). This shows that [SPEC, IP] must contain lexical material. (Affirmative imperative clauses form the only exception: these lack both NP1 and INFL.) Bambara is neither a null-subject nor a null-object language. There is no person or gender marking, nor overt subject-verb agreement. There is no overt Case marking, and the pronominal system shows no reflexes of the abstract Case relations. NPs always occur to the left of their Case assigners (i.e. I, V, P and N). Direct objects, or more precisely NPs that depend on the verb for structural (accusative) Case, must precede the verb. Furthermore, only one NP can precede the verb, and all other complements must follow. I adopt a variant of my earlier analysis (Koopman 1984) of similar facts in Mahou, namely that all internal arguments follow the verb at D-structure, that [SPEC, VP] precedes the verb, and that NPs which must be licensed by structural Case move to [SPEC, VP]. Accusative Case assignment now parallels nominative Case assignment, and structural Case is now uniformly assigned to SPEC positions. Movement to [SPEC, VP] is an instance of NP movement: movement is forced by Case theory, the head of the chain is Case marked, and the tail of the chain is theta marked.2 Given these particular theoretical assumptions about subjects, and the properties discussed above, the surface word order in (1) is assigned the following S-structure, as in (2) opposite. Both subject and direct object must undergo A-movement in order to satisfy the Case filter (to [SPEC, IP] and [SPEC, VP], respectively).3 2 The verbal system 2.1 The realization of perfective aspect Bambara has a rich system of morphologically distinct elements occurring in tensed INFL; all but one (the perfective aspect -ra) are realized as independent INFL (which I also will call AUXILIARIES). The phonological shape of INFL varies not only with the tense/ mood/aspect features of a clause, but also with the categorial features of the complement of INFL. Table I presents the INFLs that


Table I INFL followed by VP complements

are followed by a VP complement. The syntax of INFL will be completed in Table II in Section 3, which contains all INFL taking XP complements other than VP.

Table I shows the forms of the affirmative INFL and their negative counterparts, and indicates in purely descriptive terms the particular tense/aspect they express. All elements in INFL can be preceded by the past tense marker tun.4 Table I shows that INFL contains an independent auxiliary, except with perfective aspect in affirmative clauses. Perfective aspect is realized either as an independent INFL (ye), or as a suffix (-ra, with the variants -la, -na, depending on the phonological properties of the verb it is suffixed to). The choice of ye or -ra depends on the syntactic properties of the main verb; the determining factor is whether the verb is transitive (an accusative Case assigner) or not: -ra must be used with all verbs, except with transitive verbs that license an NP that they govern. With these verbs ye must be used.


(3)

(4)

(5)

Intransitive verbs: -ra/*ye a. A s/he “S/he cried.” b. *A s/he Unaccusative verbs: -ra/*ye a. A s/he b. *A s/he

kasicry

ra. PERF

ye PERF

kasi. cry

taago ye PERF

ra PERF taa. go

Passive verbs:5 -ra/*ye a. Ji minna sisan water drink PERF now “The water has been drunk now (by the child).” b. *Ji ye min sisan water PERF drink now

(den fè). child by (den fè). child by

The examples in (3), (4) and (5) show that the perfective suffix appears on all types of intransitive verbs; the perfective with ye may not be used. The feature that characterizes the verbs in (3), (4) and (5) is that they do not assign accusative Case. Let us next look at the distribution of the perfective with transitive verbs, where I use TRANSITIVE to refer to a verb which assigns both an external theta-role and an internal one. The pattern which obtains with the core case of transitive verbs (i.e. those verbs which c-select an argument of the category NP) is presented in (6); the pattern of those that c-select for PPs is presented in (7): (6)

Verbs c-selecting NP: *-ra/ye a. Den ye ji child PERF water “The child drank water.” b. *Den minna child drink PERF c. Den ye cè child PERF man “The child saw the man.”

min. drink ji. water ye. see

/*Den ji minna. child water drink PERF

With transitive verbs that take a direct object NP (6), the independent INFL ye must occur. Transitive verbs that take a PP, however, must take the suffix -ra.6 The choice of the perfective aspect is thus clearly related to the licensing of the internal NP argument. As shown above, the difference between the two types of transitive verbs


in (6) and (7) can be expressed in terms of their Case properties. Transitive verbs that cselect for a direct object NP must assign (structural) Case to this NP; those that c-select for a PP do not assign Case to this PP, since PPs do not need Case.7 How can we account for this generalization? Let us start with (8a), and suppose that the perfective aspect -ra is a realization of INFL. It will have to merge morphologically with the head of the VP, since it appears as a suffix on the verb. We know that there are in principle two ways in which the suffix and the verb can merge: by moving INFL to V (as in English see Emonds (1978)), or V to INFL, as in French (compare Emonds (op. cit.)) and Vata (Koopman 1984). Although distributional evidence, such as the distribution of adverbials, is not available in Bambara (adverbials cannot occur between INFL and V), I will assume that the V moves into INFL in Bambara: V movement seems to be the most economical option (compare Chomsky (1991)). Moreover, this assumption will allow us to account for the -ra/ye alternation.

Summing up, then, the distribution of the perfective aspect markers -ra and ye can be described as in (8):


(8)

a. b.

ye occurs if V assigns structural accusative Case. -ra occurs if V does not assign structural accusative.

The suffix -ra cannot occur on accusative-Case-assigning verbs: a “dummy” INFL ye appears, the direct object appears in [SPEC, VP], and the verb follows the direct object, i.e., the verb remains in situ. V movement appears to be blocked in this configuration. Examples (9a) and (9b) show the configurations in which V movement can apply and those in which it cannot. (For ease of exposition, movement of the subject from VPinternal position is omitted from the trees):

The structure in (9b) differs from that in (9a) in two ways: first, there is an overt NP in [SPEC, VP] in (9b), but not in (9a), and second, the verb has to assign accusative Case in (9b), but not in (9a). Furthermore, the ungrammatical (9b) contains a verbal trace, as opposed to the grammatical (6a), which contains a lexical verb in situ in the VP. What accounts for the pattern above? Clearly nothing in the present framework rules out to V-to-INFL movement in these configurations, since government is respected. Therefore, either the movement of the object somehow blocks verb movement, or, alternatively, movement of the verb leads to a configuration in which the object is not licensed. Consider the first alternative. The object NP has undergone NP movement to [SPEC, VP], leaving a trace subject to the EMPTY CATEGORY PRINCIPLE (ECP). The trace of the moved NP is antecedent-governed from [SPEC, VP]. If, however, traces not only need to be antecedent-governed, but must also be head-governed (Jaeggli (1982), Stowell (1986), Koopman and Sportiche (1986), and Aoun et al (1987)), and if a verbal trace is not


a head governor, as proposed in Aoun et al. (1987), verb movement could be blocked because the trace of the object would not be head governed by the trace of the verb. There are good reasons to reject this proposal. First, the Bambara situation is not parallel to the Spanish one as described by Torrego (1984), who proposed that a verbal trace is not a proper governor. In Bambara, we are dealing with “short” NP movement, and antecedent government is satisfied. Short (Wh-)movement of the object in conjunction with verb movement does not create an ECP problem for Torrego: an ECP violation only arises in the case of long Wh-movement of the object. Presumably, then, the verbal trace also counts as a head governor for the trace of the object in Spanish, and Torrego’s examples need to be explained differently.8 Second, the proposal that NP movement blocks verb movement or that the trace of V is not a head governor makes the wrong predictions with respect to NP movement in passive constructions: in these cases there is also an NP trace that needs to be properly governed. If a verbal trace is not a head governor, the verb should not be allowed to move in passive sentences either. As (5b) shows, this is incorrect: -ra, not ye must be used in passives.9 We conclude that verb movement is not blocked because the object moves, or because the verbal trace fails to head-govern the original object trace. The explanation must therefore lie elsewhere, and Case theory is an obvious candidate. Direct object NPs need Case by virtue of the Case filter. [SPEC, VP] is the position to which structural accusative Case is assigned. Suppose now that the verbal trace in (9b) cannot assign Case. This would explain immediately the impossibility of verb movement. If the verb moves, the direct object NP would not be assigned Case, and a Case filter violation would arise. If the moved verb cannot assign Case from INFL—a reasonable assumption given the proposal that Case is assigned strictly leftward—the direct object can only be assigned accusative Case if the verb remains in situ, and another INFL, ye, which I consider to be a suppletive form for -ra, appears. Ye-insertion is like do-insertion; it is a language-particular device allowing the language to resolve a particular problem. It will therefore only be used when the -ra option is unavailable. V movement with non Case-assigning verbs, as in (9a), does not create any problem of course, since the verbal trace does not need to Case-license an NP through Case assignment.10 The following property of the syntax of Bambara has now been established: (10)

Verbal trace cannot assign Case.

This implies that the verb in INFL cannot transmit its Case-assigning property to the tail of the chain. In other words, a Case chain cannot be formed between INFL and the verbal trace. I will next discuss the consequences of (10) for the syntax of causative constructions. I will then go on to show that (10) is in fact a subcase of a more general condition: Case chains are absent in Bambara, both in the verbal system and in the nominal system.


2.2 Causatives The absence of Case transmission in the verbal system also accounts for the syntax of the causative construction in Bambara, as I will now show. The analysis below is similar to the one advocated by Baker (1988), in that it involves verb-raising and the proposal that verbal trace does not assign Case (as in (10)) (first proposed in Rouveret and Vergnaud’s (1980) treatment of French causatives). Causative verbs in Bambara are formed by a causative prefix la-, bearing a high tone, followed by the main verb, which retains its tonal properties. It is interesting that the causative verb syntactically behaves as a single lexical item, yet, tonologically, it acts as two independent words. The following distribution characterizes the causative construction: (11) a.

b.

c. d.

e.

f.

g.

h.

Intransitive V A bè den lakasi. s/he INFL child make cry “S/he made the child cry.” Unaccusative V: A bè den lataa s/he INFL child make leave “She made the child go to the market.” Transitive verb c-selecting a PP complement: N ye den lason I PERF child make agree Bala ye den labò Bala PERF child make visit “Bala made the child visit you.” Transitive verbs c-selecting an NP complement: A ye daga laci. s/he PERF pot make break “S/he had the pot broken.”

A ye den labugo s/he PERF child make hit “S/he made the child be hit.” A ye den labugo s/he PERF child make hit “S/he made the child be hit (to the woman).” *A ye den lamin. s/he PERF child make drink “S/he made the child drink.”

(*cè man (*muso woman

(sugu la). market to

a ma. it to i you

ye. at

fè). by (ma)). to

As these examples show, causative verbs can be formed with verbs that do not need to assign structural accusative Case. A problem arises again with transitive verbs that c-select


for NPs (11e-h). In causative constructions, the Case-marked object of the complex causative verb must be the theme of the thematic verb (11d), and cannot correspond to its agent (11f, g, h).11 There are no double-object constructions in Bambara, and the external argument cannot be expressed (neither in a by-phrase (11h), nor as a goal argument (11g) comparable to the French faireà causative construction). The causative complex itself can be passivized, as the following example shows: (12)

Den lakasira sisan child make cry PERF now “The child has been made to cry now (by the woman).”

(musofè). woman by

In sum, then, the Case-marked object of the causative complex corresponds to: (13) 1. 2. 3. 4.

the external argument of an intransitive V (11a); the internal argument of an unaccusative V (11b); the external argument of a transitive V c-selecting a PP (11c); the internal argument of a transitive V; the external argument must remain implicit, and cannot be realized (11f, g, h). Furthermore, 5. the causative complex can be passivized (12).

The analysis for causative constructions is based on (14): 14 1. The causative la- assigns an external theta-role, and takes a VP complement (a syntactic property); 2. the causative prefix la- selects for V as a morphological property (i.e., it triggers verb raising). Condition (14.2) forces the main verb to raise and adjoin to the causative prefix. The main verb adjoins to the right of the causative prefix, a property which might very well be related to the fact that verbs c-select for elements to their right. Since, syntactically, causative la- assigns an external theta role and takes a VP complement, it will assign Case to an NP it governs. Case licenses the appearance of the NP in direct object position (13. 1-3). Moreover, since the causative assigns both an external theta role and Case, it can be passivized (13.5). These assumptions yield the following derivation for (11a) (for simplicity, (1 la) starts with the VP projection of make. The higher IP projection is omitted):


Let us next consider how the properties of transitive verbs in (13.4) follow. Consider the following structure, in which the external argument of the lower verb must be suppressed (indicated as NP*, [θ]), see (16). Suppose a transitive verb projects both its arguments; V* must move to adjoin to the causative verb, because of (14.2). Since the verbal trace does not assign Case, every possible outcome would violate the Case filter. There are two NPs that need Case and there is just one Case assigner present: the causative complex. There is neither inherent nor default Case marking in Bambara. Thus, the only possible outcome here is one in which one of the arguments is suppressed. Suppression of the internal argument will yield a violation of the Projection Principle and the Theta Criterion. The only possibility, therefore, is to use an independently attested mechanism in the language: suppression of the external argument, so that the internal argument can move to be assigned Case by the causative complex. This process then is basically the same as the passive. It differs from the passive, though, with respect to the possibility of expressing the suppressed argument in a fe-phrase (i.e., a by-phrase). This is possible in a ‘sentential’ passive, but not possible in the complement of a causative


verb. If the fe-phrase is licensed by INFL (or gets its theta role from INFL),12 the difference can be reduced to the absence of INFL in the causative complement (thus providing no possible licensing of the fe-phrase), and the presence of INFL in the regular clausal passive (providing a licenser for the fe-phrase). The ungrammaticality of (11h) remains to be accounted for. It is unlikely that its ungrammaticality is a problem that is specific to the causative construction. It seems rather to be related to the general impossibility in Bambara of using transitive verbs intransitively: (17) a. A bè ji min. s/he INFL water drink “S/he is drinking water.” b. *A bè min. s/he INFL drink “S/he drinks.” c. A bè minni kè. s/he INFL drinkNOM do “S/he is drinking.”


As these examples show, a transitive verb can only be used intransitively if it is embedded in a nominalization. Varying INFL does not appear to influence the pattern in (17), nor is it lexically restricted. 2.3 Summary I have argued that the impossibility of Case transmission is the key to understanding the distribution of perfective aspect and the syntax of causative constructions. Given (10), and configurations in which verb movement is forced, problems arise regarding transitive verbs that must Case-license their direct objects. The two different ways in which Bambara solves this problem are: either (1) by blocking verb movement, and having a dummy INFL appear, or (2) by obligatorily suppressing the external argument of the transitive verb embedded under the causative. This concludes the discussion of the verb system. I will now turn to the nominal system. 3 The nominal system Is the lack of Case transmission specific to the verbal system, or is it more general? In this section it will be shown also to characterize chains in the NP system. If a language allows for the formation of Case chains, two overt NPs can be licensed by one structural Case (as, for example, in English There is man in the garden). If a language lacks Case chains, each NP must be licensed by its own Case assigner. I will show how the lack of Case transmission between NP positions will yield an account of certain properties of the Bambara nominal system. In section 3.1, I will discuss the Case properties of the A-system (the term A-SYSTEM here in the sense of Koopman (1984), where it extends to all obligatorily projected positions). Three different relations in the A-system will be discussed. First, in section 3.1.3, regular cases of NP movement will be examined, where movement is forced by virtue of Case theory. In section 3.1.3, I will turn to the syntax of the existential inflectional particle bè and the presentational particle don “it is”, and demonstrate that NP movement obligatorily applies in these Bambara constructions. This will account for the absence of expletive pronouns that are related to NPs. Then in section 3.1.4, I would like to use the idea that Case chains cannot be formed as a tool to investigate some syntactic properties of predicate nominals (e.g., This is a table. John is the writer of the book.) and nominal small clauses (e.g., I find John a fool). In section 3.2, finally, I will explore some fundamental properties of Bambara’s A´-system, and suggest that the lack of syntactic Wh-movement can be at least partially explained if we assume the absence of Case chains.


Table II INFL selecting for XP other than VP

3.1 The A-system 3.1.1 NP movement: from a [-Case] to a [+Case] position So far, we have discussed two cases of NP movement: movement to [SPEC, IP] (this movement takes place in all the configurations in Table I), and movement of an NP that needs accusative Case to [SPEC, VP]. Both represent regular cases of NP movement forced by Case theory. As I will now show, NP movement to [SPEC, IP] also occurs in all the projections of the elements in Table II, which completes the syntax of Table I. Indicated are the (rough) semantics of the elements INFL (all these would involve the copula be in English), and their lexical properties. In Bambara, INFL can be directly followed by any XP complement (VP (see Table I), AP, NP, PP, S′). The features of INFL thus vary according to the categorial features of its complement, a property expressed by the c-selection frames in Table II. These INFLs are all restricted to tensed complements. Furthermore, although conceivably these INFL assign some kind of internal theta to their complements, each INFL licenses particular kinds of complements, and they do not assign an external theta role: they therefore do not assign (structural) accusative Case. In the following sections, we will first discuss ka and locative and progressive bè. Don and existential bè are discussed in 3.1.3, and ye…ye in section 3.1.4. 3.1.2 ka and locative and progressive bè Ka can only refer to the present or past state (never to a state that will be achieved in the future), and must be followed by an AP. This latter property can be expressed by


assigning to ka the property of c-selecting an AP. Since the external argument of the A is generated either as external argument of the AP, or as internal argument (recognizing the existence of ergative adjectives (Cinque 1990)), and since A does not assign Case, it will have to move to [SPEC, IP] to be Case marked. This implies that a sentence like (18) is assigned the following S-structure: (18)

[I ka kais

Balai Bala “Bala

[APti INFL healthy.”

[APkènè]]]. healthy

Locative bè and progressive bè must be followed by PPs, and do not assign an external theta-role; the surface subject of bè must therefore have raised out of the PP predicate. The sentence in (19) will therefore be assigned the following structures: (19)

[I bè Balaj Bala is “Bala is in the house.” [Ibè Balaj Bala is “Bala is running.”

a.

b.

[PP tj [PPSO house [lap tj run

kònò]]]. in [tj boli-la]]]. la

Again, the NPs in [SPEC, IP] have raised there from their respective theta-positions in the predicates so kònò “in the house” and boli-la “running”. They move into [SPEC, IP] to satisfy the Case filter. 3.1.3 Existentials and don Sentence illustrating existential bè and presentational don are given in (20). (20)

a.

b.

Tabali table “It is a table.” Here peace “There is peace.”

don. don INFL bè. bè INFL

The NP corresponding to the predicate NP in English must occur in [SPEC, IP] and expletive pronouns can never be used in these constructions. I will propose that examples like (20a) and (20b) should be analyzed as (21a) and (21b): the NP in pre-INFL position has moved there from post-INFL position, basically because it is the only way for it to pass the Case filter. (21)

a.

Tabalii

[Idon[NPti]].


b.

table “It is a table.” Herei peace “There is peace.”

don-INFL [Ibè[NP ti]]. bè-INFL

Before motivating the analysis in (21), I will review some properties of the existential and presentational constructions in English. Our discussion will be limited to two aspects of the sentences in (22): (22)

a. b. c. d.

There is peace. There is a man in the garden. It is John. It is a table.

First, how does the NP occurring in post-INFL position get Case, and second, what if any, is the relation between the expletive pronoun and this NP? As for the first question, we will assume that the expletive pronoun there licenses the NP to its right by forming a Case chain with this NP at S-structure.14 Let us assume that this analysis equally extends to the relation between the expletive pronoun it and the NP in (22c-d). As for the second question, it is well known that there exists a locality relation between the expletive pronoun there and the related NP. This relation basically mirrors the locality observed under NP movement. Chomsky (1986b) proposes an account for these locality restrictions which forces NP movement to apply at LF to a sentence containing an expletive. At LF, then, the NP forming a chain with an expletive moves to the position containing the expletive, eliminating it. NP movement at LF is forced by the PRINCIPLE OF FULL INTERPRETATION (PFI), which states that all elements at PF and LF must be INTERPRETABLE OR LICENSED. If LF is the level that contains contentive and meaningful elements, expletives, being neither contentive and meaningful elements, cannot occur at LF. Chomsky proposes that expletives are replaced by LF by the (meaningful) NPs with which they are related at S-structure.15 The locality relation between the expletive and the related NP thus reduces to regular principles constraining antecedent-race relations (i.e. Subjacency, ECP, Binding Theory). With these assumptions, part of the syntax of expletive-NP relations in English is explained as follows: (23) a. In EXPLi. . . NPi, EXPLi can transmit its Case properties to NPi (i.e. expletive and NP can form a Case chain at S-structure). b. The relation between the expletive pronoun and the NP is local, as a consequence of movement of the NP to the position of the expletive pronoun at LF. Let us now return to Bambara. Consider the following sentences:


(24)

a.

b.

(25)

a.

b.

c.

Tabali table “It is a table.” Bala Bala “It is Bala.”

don. don-INFL don. don-INFL

Here bè. peace bè-INFL “There is peace.” Tòòrò tè. harm NEG-INFL “There is no harm./Everything is fine.” Ni na tè,… if sauce NEG-INFL “If there is no sauce,…”

Both don and bè function as genuine INFLs heading IPs: they can be modified, they can be preceded by the past tense marker, and so on. Surprisingly, a full NP occurs in [SPEC, IP]. But [SPEC, IP] is not a position to which a theta role can be assigned, nor can it be cselected. Therefore, the NP occurring in pre-INFL position must have moved there. But where did it move from? Suppose that don c-selects for an NP (as suggested in Table II). This captures its distribution: don can only occur with NPs. The underlying structure for (24) must then be [IP [Idon [NPBala]]], and the NP in post INFL position must move to [SPEC, IP]. The surface order in (25) is therefore derived via NP movement. Exactly the same arguments carry over to existential bè. The NP in [SPEC, IP] must have moved there from its c-selected position, the post-INFL position. If NP movement must apply in sentences like (24) and (25), the next question is why this is so. One could simply propose that expletive pronouns are absent in Bambara. Since pro is not licensed (Bambara is not a pro-drop language), NPs would have to move to the position in which an expletive would occur. This, however, seems to miss the point. Bambara does have overt expletive pronouns related to CPs.16 It is precisely expletive pronouns related to NPs that are absent. We already argued that Case chains cannot be formed in the verbal system (10a). If Case chains can never be formed in Bambara, an explanation can be provided. The postverbal NP needs Case, precisely because it is an NP. Since the D-structure position is not a position in which it can be Case-marked by INFL (recall that Case is assigned leftwards), and Case chains cannot be formed, the NP must move at S-structure into the position that is Case marked by INFL: [SPEC, IP]. In English, expletive pronouns can exist, precisely because the Case filter can be satisfied at S-structure by the formation of a Case chain, and NP movement occurs at LF. We assume thus that there is a parameter (26) and that this parameter is set negatively in Bambara, but positively in English (but see section 4 for more discussion):


(26)

Case transmission parameter Given an A-chain Case properties to

can or cannot transmit its

From this follows the absence in Bambara of expletive pronouns that transmit Case to an NP, and the obligatoriness of NP movement in contexts where, in other languages, Casetransferring expletive pronouns may appear. Alternatively, expletive pronouns cannot exist if the particular INFL they would co-occur with fails to assign inherent Case.17 Thus, Bambara has no choice about how to satisfy the Case filter: NP movement is forced at Sstructure. As a consequence, Bambara overt NPs (in A-positions) will always occur as the head of a chain. In English, NP movement can wait until LF.18 3.1.4 Other Case chains In the preceding section, constructions were discussed in which a lexical NP (as opposed to a expletive NP) needs to be licensed by Case. We arrived at the conclusion that there is a one-to-one correspondence between Case assigners and Case assignees in Bambara. Since verb movement does not allow the formation of a Case chain, only one structural Case position is available in sentences where verb movement has applied. If two structural Case positions are needed, both I and V need to be lexical, and V must be in situ. In the nominal system, Case chains are excluded as well, again with the consequence that each NP needs to appear in a Case-marked position, governed by an overt Case assigner. In this section, I would like to explore the generalization that each NP needs to be licensed by a Case assigner, and see just how general it is. I will show that it holds in all kinds of constructions that are not usually assumed to obey the Case filter, or to involve Case chains. The goal of this section, then, is twofold. First, to complete the description of the distributional properties of NPs in the A-system, and second, to use Bambara as a test case for establishing which positions need to be licensed by Case. Let us first turn to predicate nominal constructions and nominal small clauses. So far, examples have been presented illustrating the use of the INFL don. Don is used if only one NP is realized. It is surprising that the situation changes in predicate nominal constructions where two full NS have to be licensed: (27)

a.

b.

Nin ye this ye-INFL “This is a table.” Nin ye this ye-INFL “This is Bala.”

tabali ye. table to Bala ye. Bala to


In these sentences, the INFL ye must be used, and the second NP must be followed by the postposition ye (glossed as to). The NPs in (27) satisfy the Case filter: they are assigned different Cases by INFL and by the postposition ye respectively. They are also consistent with the one-Case-assigner-per-NP hypothesis. Consider next the ungrammatical examples in (28): (28)

a. b. c.

*Nin this *Nin this *Nin this

ye ye-INFL don don-INFL don don-INFL

tabali. table tabali. table tabali ye. table to

Examples (28a) and (28b) contain two NPs. If both NPs need to satisfy the Case filter, and if Case chains cannot be formed in Bambara, they are straightforwardly excluded, since there is only one Case assigner available. The ungrammaticality of (28c) clearly cannot be Case related. Here, a second Case marker (the postposition ye) licenses the second NP. It can be explained, however, by our assumption that don c-selects for an NP, and not for a PP. That is, what excludes (28c) is the fact that the ye-complement is not licensed. (This complement can only be licensed by a ye-INFL.) We also predict that don can only license one NP, i.e., the NP complement selected by don cannot have an external argument. Don c-selects for an NP, which needs to be Case marked. As we saw before, this implies that it must obligatorily move to [SPEC, IP]. Suppose that the c-selected NP did have an external argument itself (i.e. don [NP [NP]]). If both NPs need Case, a Case filter violation arises, because only one Case assigner is available (INFL). Hence the complement of don must necessarily be unaccusative. Let us next consider the distribution of nominal small clauses like John a fool in I consider [John a fool]. If both NPs need to be assigned Case, we expect such forms to be absent in Bambara. This prediction seems to be borne out. I have not been able to find any nominal small clauses, but for one apparent exception to which I turn immediately below. Potential candidates for nominal small clauses are all realized as PPs: (29)

U ye a they PERF him “They made him chief.”

kè make

kuntigi chief

ye. to

Of course, examples like (29) are perfectly consistent with our analysis: the subject of the small clause is assigned Case by the main verb, and the predicate NP is assigned Case by the postposition. So far, then, the generalization that each Case assignee is governed by a Case assigner seems to be well-founded. And insofar as it follows from the lack of Case transmission in Bambara, it seems to support our analysis. I now turn to configurations in which the generalization does not seem to hold. I will propose that the NPs in these configurations are actually in an A-bar position, and as such do not need Case.


Consider first the following examples: (30)

a.

b.

N tògò my name “My name is Alima.” U bè they INFL “They call me Alima.”

ye/ INFL

kò say

Alima. Alima

n me

wele call

Alima. Alima

Here, two NPs occur, but there is only one Case assigner.19 This is only a problem, though, if the NP following INFL is in an A-position, i.e. if there is a genuine small clause complement here. Wh-questions indicate that the second NP is not in an A-position, but rather in an adjunct position (A-bar position). In Wh-questions, the form of the second NP corresponds to an adjunct (di ‘how’) and not to an argument (mun ‘what’). The latter must be used when questioning a NP predicate: (31)

a.

b.

c.

i tògò ye your name is “What is your name?” U bè i they INFL you “What do they call you?” Nin ye mun/ this INFL what “What is this?”

di */ how

mun? what

wele call

di how *di how

/*mun? what

ye? to

Suppose that NPs like di are in A-bar positions, not in A-positions (recall that predicates occur in A-type positions, i.e., positions from which a theta role is assigned). Suppose furthermore that NPs in A-bar positions do not need Case. Sentence (31a) then would not be a counterexample to the claim that Case chains are absent. Additional evidence that NPs in A-bar positions do not need Case comes from the distribution of bare NP adverbs like kunun “yesterday”. These adverbs occur as bare NPs; they are not followed by a P.20 3.1.5 Summary The hypotheses that Case chains are absent in the NP system in Bambara accounts for the obligatoriness of NP movement in cases in which expletive pronouns appear in Case-chainforming languages, and for the absence of expletive pronouns related to NPs. It furthermore accounts for the necessity of two Case assigners in the predicate nominal construction, and for the non-occurrence of nominal small clauses. It is not usually


assumed that the predicate NP needs Case.21 However, in languages with overt morphological case, the predicate nominal must be Case-marked. It appears that there are two patterns of predicate nominal Case marking: either the predicate nominal agrees with the external argument in Case, as in Latin, or the predicate nominal has some default Case (for instance, instrumental in Russian). In the latter pattern, the appearance of instrumental case might represent some language-specific default case, i.e., it is a way of getting Case to the predicate nominal, much as, in Bambara, ye can be used to license the predicate nominal. As for the Case agreement pattern, I will adopt a proposal by Hoekstra and Mulder (1990) and assume that the predicate acquires Case through SPEC-head agreement with the trace of the moved NP, i.e., Case is transmitted through the chain with the trace:

If Case chains are absent in Bambara, i.e., if Case cannot be transmitted through a chain to license the predicate NP, the predicate NP cannot occur unless it is assigned Case in another way, i.e., by an independent Case assigner. Further examination of the distribution of NP suggests that NPs in A-positions (NPs in SPEC or complement position, and predicative NPs) need Case: NPs base generated in Abar positions, however, need not be licensed by Case. 3.2 A•-A relations The absence of (A-type) Case chains in Bambara accounts for an important part of the syntax of simple clauses in Bambara. Is there any reason to assume that (26) also extends to A′-chains? In the previous section, I have discussed NPs that are base generated in A′position: these do not appear to need Case. In this section, I discuss the problems surrounding A′-movement in Bambara. One salient fact about the syntax of Bambara is that syntactic Wh-movement is absent (with the exception of reason adjuncts, discussed below). Wh-words in Bambara must remain in situ. This is illustrated for Wh-questions in (33a): (33)

a.

b.

c.

I ye jon ye? you PERF who see “Who did you see?” I ye cè min you PERF man RL “The man you saw is called Bala.” N ye cè de I PERF man FOC

ye, see, ye. see

o that one’s

tògò name

Bala. Bala


“I saw the man” In relative clauses, the head of a relative clause occurs clause-internally and is marked by the relative clausemarker min (33b). The entire relative clause is excluded from regular NP positions and occurs in a dislocated position (relative clauses are the only elements which may occur in this position). In focus constructions, the focused NP must occur in situ (where it is marked by de), as in (33c). Bambara is quite exceptional in that it not only has in situ constructions: it only allows in situ constructions. Bambara furthermore does not allow for heavy-NP shift or any kind of scrambling of NPs. In an earlier version of this chapter, I suggested that, again, the absence of Case chains might constitute the key to understanding this property. This explanation was, however, problematic and quite clearly cannot be the only variable involved here. Nevertheless, I have decided to include the following since the configuration of data is interesting. Let us suppose that Wh-words move to [SPEC, CP] in the syntax. This implies then that A′-movement would result in an A′-chain consisting of the Wh-phrase and its trace. If this Wh-phrase needs Case, it will have to satisfy the Case filter by forming a Case chain with the trace. Since Bambara does not allow for Case chains, the Wh-phrase would fail to satisfy the Case filter. It must therefore remain in situ, and the impossibility of Whmovement of NPs is explained.22,23 Failure of an NP in A-bar position to pass the Case filter only yields a partial explanation for the impossibility of Wh-movement, however. More needs to be said, as becomes clear upon considering PPs, which do not need Case. Under the scenario above, only NPs are expected to remain in situ. PPs should be able to remain in situ or to appear in [SPEC, CP]. This prediction is not borne out: PPs must remain in situ. Furthermore, any proposal blocking PPs from moving to [SPEC, CP] (either by arguing that the PP is not a possible pied-piper, or that the [SPEC, CP] is a restricted position which does not tolerate any overt material) runs into the following problems: although PPs may not move, adjunct PPs may move, or remain in situ, and “reason” adjunct PPs must move (possibly for ECP reasons): (34) a.

I nana mun na? you come-PERF what to/ in which way/ *for what reason “What did you come for?/How did you come?” (i.e. by foot or by train) b. Mun na i nana? what to you come -PERF “Why (For what reason)/How did you come?” c. Mun naj i ye a fo a fagara tj? what for you PERF it say he kill PERF “For what reason did you say that he was killed?”


Since these PPs can move successive-cyclically, as in (34c), it is likely that they actually are in [SPEC, CP]. In sum, then, the assumption that Case chains cannot be formed between A′- and A-chains only yields a partial explanation for the lack of A′-movement. 4 Conclusions: further issues This chapter is organized around the idea that Case chains are generally absent in Bambara, in both the verbal system and the nominal system. The absence of verbal Case chains yields an account for the distribution of perfective aspect and the syntax of causative constructions. In these contexts verb movement is forced, and problems arise with transitive verbs that need to license their direct objects by means of Case. Bambara solves these problems in two language-particular ways: in the case of V to INFL movement, by blocking verb movement, and inserting a dummy INFL; and with causative verbs, by obligatorily passivizing the embedded transitive verb. The absence of nominal Case chains leads to NP movement in contexts where expletive pronouns could appear in Case-chain-forming languages; it further accounts for the appearance of two Case assigners in predicate nominal constructions, and explains a gap in the pattern of small clauses: while PP small clauses do occur, NP small clauses are nonexistent. Descriptively speaking then, failure of Case transmission in the A-system leads to the appearance of an overt Case assigner with each NP requiring Case.24 Finally, I speculated that the failure to form Case chains might also extend to A′-chains, and suggested how this fact could yield a partial account of the absence of syntactic Whmovement in Bambara, and how it also might be the principal reason for the extreme rigidity of Bambara word order. I now turn to a number of questions that the analysis of Bambara raises. I will start out with the verbal system. First, I have assumed that, although verbal Case chains do not exist in Bambara, Case can be transmitted along a verbal Chain in other languages— French, for instance. The question arises as to whether this assumption is correct, i.e., do verbal Case chains ever exist.25 I will demonstrate that verbal Case chains must be assumed for a language like French. A second question arises: are there any other languages that do not allow verbal Case chains? I will argue that languages with ergative Case marking are likely candidates for languages that disallow verbal Case chains. A third question arises, with respect to the strong claim that is made in this chapter: if Case chains are allowed in the verbal system, they should also be allowed in the nominal system, and vice versa: if Case chains are not allowed, they should not be allowed either in the verbal or in the nominal system. We will discuss these predictions for some languages that, respectively, disallow or allow Case chains. But first, I establish that verbal Case chains must be assumed in a language like French or Vata (Koopman 1984). I will do so by showing that the head of the verbal chain is not in an S-structure government relation with its Case-dependent NP. It must therefore be assumed that the Case properties are transmitted through the verbal chain. Consider the following examples:


(35) a.

b.

French toujours Jean choisiti John chooses always “John always chooses white rice.” Vata26 O li [saka t] s/he ate-PERF rice “S/he ate rice.”

[ti

le the

riz rice

blanc]. white

Both French and Vata have V movement to INFL (for French, see Emonds (1978), among others; for Vata see Koopman (1984)). In the examples in (35), the verb is therefore in I at S-structure. How does the direct object receive Case? Suppose French disallows Case chains, and that consequently the verb in INFL directly licenses the direct object, which might have moved to some SPEC position in order to be Case-marked. This analysis will not work, if one adopts the proposal that INFL is split into different functional heads as in Pollock (1989) and much other recent work. The structurally Case-marked object does not raise to a position governed by the highest INFL projection (it follows the adverbs and negation). Accusative Case cannot therefore be assigned from INFL under government and it must therefore be assigned by the verbal trace. That is, a verbal trace is able to license a structurally Case-marked NP, depending on the properties of the verb in INFL. A similar conclusion can be reached for Vata: in Vata, Case is uniformly assigned to the left, but the direct object must occur to the right of INFL. With respect to these Case properties, Vata exactly parallels Bambara. A Case-dependent object can therefore never be Case-marked by the verb in INFL, and must be assumed to be Case-marked by the verbal trace, i.e., the trace of the verb inherits its Case properties from the verb. (The Case assigned by the verbal trace behaves like a structural Case and cannot be assumed to be inherent.) Since in these languages the verbal trace assigns structural Case, Case can be transmitted along verbal chains. However, the assumption that verbal trace in French can inherit structural Case properties leads to a conflict for the analysis of French causatives: properties of the faire-à causative construction, it is often argued, follow from V (or V-bar) preposing of the embedded verb, in conjunction with the inability of verbal trace to assign Case (Rouveret and Vergnaud (1980), and Baker (1988)). Let us assume that this analysis is correct. In French, then, Case chains must be allowed as a result of V to I movement, but must be disallowed under V/V-bar preposing in causative constructions. Following Rizzi and Roberts (1989), this could be explained if these movement rules represent different types of rules, yielding different types of chains. V to I movement could represent an A-type movement (since it is substitution), while verb movement in causative constructions is A′movement, since it is an adjunction operation. It could then be proposed that Case chains should be further parametrized according to whether the case chain can be formed along A-chains or A′-chains.


Although this alternative might be worth pursuing, it is difficult to see how it could be extended to account for the difference between French and Vata V to I movement and Bambara V to I movement. Although one might propose that Bambara V to I movement represents A-bar movement, and Vata/French V to I movement represents A movement, there seem to be no morphological or syntactic criteria justifying this classification. For the moment then, we conclude that French/Vata V to I movement must allow for Case transmission, but Bambara V to I movement does not, i.e., there is a way in which French and Vata objects are licensed at S-structure which is absent in Bambara. Further discussion will be restricted to case chains formed by V to I movement. Are there other languages like Bambara, with no Case chains formed under V to I movement? Salleh (1987) argues that this is the case in Malay, where a verb can adjoin27 to I, except when the verb must license an “object”. Whether a verb can be attached to I can be determined in questions where I precedes the subject. An accusative-Case-assigning verb may not adjoin to I, but must remain in situ, because, as Salleh argues, verbal trace cannot assign Case. In our terms, then, Malay lacks verbal Case chains.28 Although the analysis of I-preposing is not without problems, and Malay does not seem to have any of the other properties of Bambara (as will be discussed below), I will assume for the sake of the argument that Salleh’s analysis is correct. The Bambara problem is an accusative Case assignment problem: accusative Case can only be assigned if the verb does not move: it may not be assigned if the verb has moved. There is of course another class of languages with accusative Case assignment problems: languages with ergative Case-marking systems.29 Let me briefly review their properties, and indicate how the absence of case chains in these languages seems to be the source of the problem. In a typical language with ergative Case marking, the subject of an intransitive or unaccusative verb is assigned what is traditionally called ABSOLUTIVE Case. There are good arguments that absolutive Case corresponds to nominative Case in nominative/ accusative languages. In clauses with transitive verbs, however, the subject is assigned some “extra” case, the ergative Case, and the object is assigned absolutive (nominative). Depending on the language, the ergative case is genitive or dative. Often ergative case is related to the appearance of a particular morpheme on I. Sentences with transitive verbs and ergative Case marking behave as active sentences in nominative/accusative languages, and not as passive sentences (with the exception of Dyirbal, compare below). In particular, the NP with ergative Case behaves like the external argument and not as a byphrase: it can be a controller and can correspond to PRO in non-finite sentences. Depending on the language, it can be the antecedent for a subject-oriented anaphor, and it can be missing under conjunction reduction. In languages with syntactic incorporation, it cannot incorporate, in contrast to internal arguments. Let us assume that V to I movement is obligatory in these languages, and that languages with ergative/absolutive Case marking are like Bambara in that verbal trace does not assign Case. As a consequence, only one structural Case is available per sentence. This of course leads to problems with transitive verbs, and languages appear to solve these problems in different ways.


In Dyirbal, as convincingly argued in Bok-Bennema and Groos (1984), transitive structures are obligatorily passivized or antipassivized, i.e., only one structural Case is available. In West Greenlandic, as a consequence of verb movement and the absence of Case chains, there is only one structural case available from I (absolutive = nominative). In sentences with transitive verbs, then, the object is assigned nominative in [SPEC, IP], and the subject is assigned genitive in a language-specific way, i.e., by the addition of an extra IP projection which licenses genitive Case assignment to its SPEC. In this case, INFL carries both subject agreement with the genitive NP and object agreement with the nominative NP. There are other ways of licensing syntactic transitive structures in Greenlandic: incorporation of the NP that needs structural accusative, or the appearance of antipassive morphology on the verbal complex. In these cases, the external argument is always assigned nominative Case. If incorporated objects do not need to be licensed by syntactic Case, as proposed in Baker (1988), incorporation represents a language-specific way to solve the problem caused by V to I movement. If the antipassive morpheme can assign inherent Case (the so-called INSTRUMENTAL or MODALIS Case) to an NP that is usually assigned accusative, the antipassive morpheme represents another languagespecific way of licensing objects.30 We now have some potential languages without verbal Case chains (Malay, West Greenlandic), and some potential languages with verbal Case chains (e.g. French, English, Vata). We can now discuss whether the strong predictions made by parameter (26) are borne out. On the basis of our discussion of Bambara, the following predictions are made (given the discussion on French causatives above, we will henceforth disregard causatives): (36)

(i)

(ii)

Languages without Case chains a. V in situ or transitivity problems under V movement to I b. no existential pronouns (i.e. no Case transmitting pronouns) c. nominal small clauses excluded d. Wh in situ e. general lack of A′/A relations a. no transitivity problems resulting from V movement to I b. existential pronouns may exist c. nominal small clauses are possible d. syntactic Wh-movement e. availability of A′-positions

As discussed above, Salleh argues that Malay has movement to I. Cursory inspection of Malay shows that it does not display any of the properties in (36 (i)), expect for (36 (ia)). Malay has existential constructions with an obligatorily empty [SPEC, IP]; these constructions could involve Case transmission. However, it is also conceivable that the post-I NP directly receives Case from I, which might explain the necessary absence of [SPEC, IP]. Malay appears to allow for nominal small clauses, and has syntactic Whmovement. If indeed Malay does not allow for verbal Case chains, but allows for nominal


Case chains, the proposed parameter makes predictions which are too strong. As noted earlier, however, V movement to I in Malay involves adjunction, and results in quite large verbal complexes. It could therefore be the case that this type of movement is similar to the one observed in French causatives, which also leads to the apparent unavailability of accusative Case. I will leave this problem for future research. West Greenlandic represents a more interesting test case. Above, it was assumed that West Greenlandic has obligatory V movement to I, and no verbal Case chains. Several language-particular ways are available to solve the problem of the syntactic absence of accusative Case-assigning verbs (an additional INFL projection licensing genitive Case, incorporation of the object or getting Case from the antipassive morpheme). Does Greenlandic allow for nominal case chains? Greenlandic has an existential construction involving the affixal verb -qar, “have”: (37)

Nutaa mik imiarsuarnut tallittarvi-qarnew MOD ship DAT harbour QAR “There is a new harbour for ships.” (Fortescue 1984:83)

puq. IND

3SQ

However, this construction has two properties suggesting that the NP “harbour” which is related to the (silent) expletive is not licensed through Case chain formation. First the NP must obligatorily incorporate to -qar. Incorporation is a way to circumvent S-structure Case requirements. The NP “harbour” is therefore not in a Case chain with the expletive pro, but licensed through incorporation. (Bambara lacks incorporation and would require NP movement to [SPEC, IP] in the equivalent to (37) “A new harbour is for ships”.) A second property of this construction is that the agreement on -qar never varies with the predicate NP: (38)

Apirisuqar -puq ask part QAR IND 3SQ tu -nik. PART MOD PL “There are some angry questioners.” (Fortescue 1984:84)

kamatbe angry

The Case of the stranded modifier of the incorporated noun (-nik) shows that NP is plural. INFL, however, carries third person singular agreement. This might be taken as further evidence that the pro triggering agreement and the NP that incorporates never form a Case chain at S-structure.31 Thus, the existential construction in West Greenlandic does not involve a Case chain at S-structure. Rather, the NP is licensed through incorporation. Interestingly, West Greenlandic does not allow an NP to remain in situ in unaccusative constructions either (as in English There arrived a man), as can be concluded from the fact that these NPs must have absolutive Case, fully agree with INFL, and can never be incorporated. Turning now to (36 (id)), West Greenlandic has nominal small clauses. But again, the predicate NP must be incorporated. Since incorporation is a way to avoid case assignment, the occurrence of small clauses is actually unproblematic.32


West Greenlandic appears to be a Wh- in situ language and thus seems to be consistent with (36 (ie)). Finally, (36 (if)) appears to be too strong: West Greenlandic (as do many other ergative Case marking languages) has quite free word order, indicating that a fair amount of scrambling is allowed. Clearly, then, something else needs to be said about the availability or non-availability of A and A′ scrambling positions. This certainly cannot be an effect of the absence of Case chains in itself. Given the discussion above, and modifying (36 (i)) accordingly, Greenlandic seems to correspond quite closely to the parameter (26): (39)

(i)

Languages without Case chains a. V in situ or transitivity problems under V to I b. no Case transmitting pronouns c. nominal small clauses excluded (under Case transmission) d. Wh in situ e. general lack of A′/A relations

Greenlandic yes yes yes yes no

West Greenlandic, then, seems to be consistent with the parameter, except for the availablity of A′/A relations (scrambling). It has been argued in the literature (Mahajan (1990) among others) that there are two types of scrambling: A scrambling (i.e. movement to an A- position) and A‘scrambling (scrambling to an A′-position). Note that the unavailability of A scrambling in Bambara (i.e., movement of the object to an A-position that is higher than the subject at Sstructure) follows from the fact that the highest A-position must be filled in Bambara (a Case A-chain cannot be formed). Thus, the only way an NP can be licensed is by moving to the structural Case position in the syntax. If Greenlandic is like Bambara and does not allow for Case chains, we predict A-scrambling characteristics to be absent. This also implies that in languages with A scrambling, NP chains at LF must probably be allowed, i.e., an NP with oblique Case can appear in a position which is not the position in which structural Case is assigned; it will then have to move to this position to eliminate the expletive pro in SPEC position. It appears, then, that there is no general ban on expletive pronouns (remember that Bambara does have expletive pronouns related to CPs). What is claimed is that the expletive pronoun is not allowed to form a Case chain with the NP, which therefore must be licensed in some other way (incorporation or some other language-specific Case assigning mechanism). The main differences between Bambara and Greenlandic, then, are the availability of an “extra” structural Case for the subject, the availability of a language-specific inherent Case (modalis), and the incorporation option. Finally, consider Vata as an example of an non-Indo-European language with Case chains. As discussed earlier, Vata has V-to-I-movement, and accusative Case can be assigned without any problem. Moreover, Vata appears to have expletive pronouns related to NP, and allows for nominal small clauses in copula constructions. Vata makes


extensive use of Wh-movement, and also allows VP-internal scrambling. These properties are represented in (40): (40)

(ii)

Languages with Case chains a. no transitivity problems under V to I movement b. existential (Case-transmitting) pronouns c. nominal small clauses are possible d. Wh-movement e. availability of A-bar positions

Vata yes yes yes yes yes

In conclusion, then, with the exception of Malay, parameter (26) seems to have some crosslinguistic support, insofar as there is a cluster of properties that seem to pattern together. As is often the case, though, it also appears to be too early to determine whether the parameter holds in its strongest form, i.e., for both the verbal and nominal systems. I will leave further questions for future work, and conclude this chapter by pointing out that the parameter seems to have the right properties: its value can be set on the basis of very simple and extremely frequent constructions in main clauses. The present study suggests a number of properties that are diagnostic for the setting of the parameter: problems with transitivity in the verbal system, the existence or non-existence of pleonastics/NP chains (where the pleonastic and the NP agree and the NP is not incorporated), the surface form of predicate nominals and nominal small clauses, and the syntax of Case in Wh-movement constructions. Until we gain more insight into these problems, this parameter has, if nothing else, allowed me to describe, analyze and present a substantial part of the grammar of Bambara in a simple fashion: the syntax of I, including the interesting properties of copula-like constructions, the problems of the interaction of V movement and transitivity, a problem reminiscent of ergative Case-marking systems, the surface form of existential and presentational constructions, the distribution of nominal small clauses, the syntax of causative constructions, and some intriguing properties of the A-bar system. Notes * This chapter was first circulated in 1987 as an article. The present version contains some minor changes and an all new conclusion. I would like to thank the audiences at UCLA where this material was first presented and developed, as well as Mark Baker, Noam Chomsky, Harold Crook, Richard Kayne, Dominique Sportiche and various anonymous reviewers for their comments and suggestions. The Bambara data discussed here are drawn from Bird, Hutchison and Kante (1976), Bird and Kante (1976), Bird (1966), Courtenay (undated), and my own fieldwork. Standard Bambara orthography will be adopted, except for tones, which will not be marked. 1 Case chains have been argued to exist by Safir (1985), Chomsky (1986b), Hoekstra and Mulder (1990) among others. Belletti (1988), Borer (1984), Chomsky (1991), and Pollock


2

3

4

5

6

(1981) argue that there is no Case chain between the existential pronoun there and the NP that is related to there in existential constructions. This proposal seems well integrated into the theory now (compare Larson (1988), Sportiche (1990), and Chomsky (1991), who suggests that the object in Bambara could actually be in [SPEC, AGR-0], the position that triggers object agreement). In an earlier version of this chapter, I suggested that the idea that direct objects can appear in [SPEC, VP] has direct consequences for the analysis of languages in which the verb agrees with the direct object. If agreement is always the reflection of a SPEC-Head relation, the direct object must appear in [SPEC, VP] when the verb agrees with it. Kayne (1985) essentially argues for the same point: he argues that what looks like object agreement in past participle constructions in Romance is, in fact, better analyzed as agreement with the structural subject of the participle. The rigid word order of Bambara seems to be related to the limited use Bambara can make of A′-positions. A′-positions can only contain adjuncts and internally headed relative clauses. Syntactic Wh-movement is absent (except for one particular case: why adjuncts, compare section 3.2). I will refer to INFL as one syntactic constituent, although, following Pollock (1989) and others, it is presumably made up of several heads (Tense, Negation, Aspect, and so on). I will refer to INFL as either INFL or I. Although there is no overt passive morphology in Bambara, these verbs behave exactly like passive verbs: the external argument can optionally be reassigned through a by-phrase, the [SPEC, IP] is a non-thematic position, and accusative Case is not assigned. Note that I assume that the verbs in (7) assign an external theta role, as well as an internal theta role, and not two internal theta roles. The property of assigning an external theta role allows an account of the fact that these verbs are potential Case assigners. Precisely this class of verbs can optionally license “cognate objects”: if a cognate object is present, the verb has to assign it accusative Case, and ye is selected:

7 I exclude from the discussion transitive verbs that c-select for CP complements. Tensed (but not infinitival) CP complements must be related to an overtly realized Case position, and the verb acts like a transitive Case-assigning verb with respect to the selection of yel -ra:

8 Note also that the Spanish facts are not paralleled by the facts of V movement in Vata (see Koopman (1984) for discussion). 9 Moreover, the trace of INFL in English also needs to be considered an appropriate head governor, as one can conclude from the interaction of raising to subject from the VP adjoined position to [SPEC, IP] and SUBJECT AUX INVERSION (SAI). In Did John tdid tJohn see Bill?, John has moved from tJohn [SPEC, IP]. TJohn can only be properly governed by tdid’ the trace of INFL which has moved into C. Yet there is no problem with this sentence. We must thus assume that the trace of an INFL is an appropriate head governor.


10 But why is affix hopping, supposing it is a universally available option, not a possible option in Bambara? We might resort here to my (1984) proposal that the head of a Case-assigning chain must be lexical. Affix hopping would result in a chain with a covert head, and nominative Case cannot be assigned. This will not explain the possibility of affix hopping in English, if indeed affix movement is the correct analysis. 11 Bird, Hutchison and Kante (1976) cite one counterexample: ladun ‘make someone eat’. Here the external argument of the embedded verb corresponds to the object of the causative. 12 I thank an anonymous reviewer for this suggestion. The reviewer also suggests that my analysis may yield an account for the fact that no overt morphology needs to appear with the passive: if the verb raises to INFL it will not be able to assign accusative Case. The only possible outcome would then be not to project the external theta role of the verb. 13 The termlaP is used to indicate the maximal projection of the element la which independently occurs as postposition (P). 14 Another possible answer to the Case problem does not involve Case transfer or a Case chain, but rather (inherent) Case assignment by the copula to the postverbal NP (compare Pollock (1986); Belletti (1988)). If this is correct, the Bambara data could be explained as follows: the INFLs do not assign Case. This in turn could either be related to the general absence of a lexical category “copula”, or to the fact that cop-ulas must appear in INFL (for example, this property would follow if they have undergone some movement to INFL, together with the general absence of Case chains). 15 Chomsky (1991) proposes a slightly different analysis in which the expletive pronoun is not actually replaced at LF. It is treated as an LF affix instead, to which the NP adjoins. 16 CPs are not Case marked at S-structure. They do not need to be in a Case chain. We can assume that they move at LF to replace the expletive. 17 Woolford (1989) argues that expl INFL NP cannot exist in Bambara, because INFL governs to the left. INFL not only governs to the left: its complement, the XP which it s-selects, occurs to its right in D-structure, so it also governs to the right. It could be argued that INFL Casegoverns only to the left, and that no inherent Case is available. 18 This does not explain why *peace is or *John is are excluded at S-structure in English, as pointed out to me by R.Kayne. It seems to be the case that be requires an overt sister in the VP at S-structure. This situation actually parallels that surrounding Wh-movement: there are languages in which Wh-movement obligatorily applies at S-structure (e.g. Italian), languages in which Wh-movement can apply at S-structure or at LF (French), and languages in which Wh-movement only applies at LF (Bambara). Similarly, there are languages in which NP movement obligatorily applies at S-struc-ture (Bambara), languages in which NP movement can apply at S-structure or at LF (English), and languages in which NP movement applies only at LF (Dutch), unless other factors force the subject to move at S-structure, (compare the definiteness of the object for example). 19 Note also that examples like (30) are the only type of examples that I am aware of that can be used without an overt INFL: N tògò Alima, “My name is Alima.” These should probably be treated as NPs, and not as projections of IP, with the second NP modifying the first. Support for this assumption comes from the fact that the corresponding question must contain INFL I tògò ye/ko/*ø di?, “Your name is how?” 20 Internally headed relative clauses occur in left-dislocated position, without overt Case marking. If they are NPs, they represent another instance of NP in an A-position not followed by an overt Case marker. 21 But see Emonds (1986) and Tremblay (1991) for arguments that they need Case.


22 Note that the insertion of a specific Case assigner for the Wh-phrase, which often takes the form of a copula construction, might be another way of solving this problem, i.e., it represents a way to license the initial constituent. 23 This hypothesis seems to contradict the conclusion reached in the preceding section: NPs in A′ positions do not need Case. It is possible to distinguish between these cases, however, since Wh-phrases are arguments, and the cases under discussion in the previous section are adjuncts to start with. 24 Although this generalization holds in tensed and infinitival complements, present and past participle clauses are problematic: both the subject and the direct object may occur before the participle.

How then are these two NPs assigned Case? I do not understand the internal structure of these complements and their distribution well enough to present a complete analysis of them. It seems likely, however, that the participial morphology is in I, that V movement has applied, and that the direct object in this construction is actually an incorporated N. This would explain some restrictions on the direct object in this construction, which basically can only be a noun (including proper names): it cannot appear with any modifiers, nor can it be a Wh-phrase or be marked with the relative clause marker. Full direct object NPs can only be licensed in the following construction:

The examples in (ii) are more expected: both NPs are assigned Case by their own Case assigner. If my analysis for (i) is correct, then (i) would be characterized by the availability of noun incorporation in this construction. Of course, it is unavailable in regular tensed or infinitival clauses. This could possibly be related to the nominal character of the type of I that appears in these clauses, which also allows for compounding. 25 Chomsky (1989), for example, argues that Case chains never exist. He proposes that the correct formulation of the parameter here is not whether Case chains can be formed, but whether empty elements (traces and expletives) can participate in the Case system or not; in Bambara-type languages empty elements cannot participate in the Case system, in Englishtype languages they can. There are two problems with this proposal: first, there are expletive pronouns in Bambara that are linked to CP complements. Second, the language


26 27

28

29

30

31

32

allows for bare NP adverbs, which according to Larson (1985) are assigned Case by an empty P. Perfective aspect is formed by a tonal affix (compare Koopman (1984)). Salleh proposes in fact that a verb can either substitute into I if I is empty, or adjoin to I if I contains other material. The resulting I can be quite big. As far as I was able to determine, there does not seem to be any reason to assume two different processes. In order to account for the difference between Vata and Malay, Salleh exploits the difference in the positioning of the head in the VP in both languages: in languages with a head-initial VP, verbal trace would not be a Case assigner, whereas in V-final languages, verbal trace is a Case assigner. This generalization does not seem to hold, however. There are many headinitial languages in which verbal trace must be assumed to function as a Case assigner in (35): e.g. French, the verb-second languages (like the Scandinavian languages, which have V to INFL movement and V to C movement in root clauses, and INFL to V movement in embedded environments), and Yiddish (which has V to INFL movement in all clause types, and V to C movement in root clauses; see den Besten et al.(1985)). A far as I know, Bok-Bennema and Groos (1984) were the first to propose that the absence of accusative Case was the key to understanding ergative Case marking patterns. The following discussion relies heavily on their (1984) discussion as well as on the discussion in BokBennema (1991) and references cited therein. Bok-Bennema and Groos (1984) propose that in ergative languages, verbs are not Case assigners. Our proposal essentially derives this property from the obligatoriness of verb movement, and the absence of Case chains. Note that this construction causes problems for expletive replacement at LF. If the NP whose head has been incorporated must move to replace the expletive in [SPEC, IP], an ECP problem arises, since incorporation is not possible from this position. Possibly the verb moves to C and incorporates the N from [SPEC, IP] after expletive replacement. I have nothing to say here about the particular fixed position (immediately post-INFL) in which stranded modifiers must appear in this construction (see Sadock (1985) and BokBennema and Groos (1988) for discussion).

References Aoun, Joseph, Norbert Hornstein, David Lightfoot and Amy Weinberg (1987) “Two types of locality”, Linguistic Inquiry 18: 537–577. Baker, Mark (1988) Incorporation, University of Chicago Press, Chicago. Belletti, Adriana (1988) “The case of unaccusatives”, Linguistic Inquiry 19: 1–34. den Besten, Hans and Corretje Moed van Walraven (1985) “The syntax of verbs in Yiddish”, in Haider and Prinzhorn (eds) Verb Second Phenomena in Germanic Languages, Foris, Dordrecht. Bird, Charles (1966) “Aspects of Bambara syntax”, unpublished PhD dissertation, UCLA. Bird, Charles and Mamadou Kante (1976) Intennediate Bambara, Indiana University Linguistics Club, Bloomington. Bird, Charles, John Hutchison and Mamadou Kante (1976) Beginning Bambara, Indiana University Linguistics Club, Bloomington. Bok-Bennema, Reineke (1991) “Case and Agreement in Inuit”, unpublished PhD thesis, Tilberg. Bok-Bennema, Reineke and Anneke Groos (1984) “Ergativiteit”, GLOT 7: 1–49. Bok-Bennema, Reineke and Anneke Groos (1988) “Adjacency and incorporation”, in Everaert et al. (eds) Morphology and Modularity, Foris Publications, Dordrecht, 33–56.


Borer, Hagit (1984) “I-subjects”, Linguistic Inquiry 17: 375–416. Chomsky, Noam (1986a) Barriers, MIT Press, Cambridge, Massachusetts. Chomsky, Noam (1986b) Knowledge of Language: Its Nature, Origin, and Use, Preager Publishers, New York. Chomsky, Noam (1991) “Some notes on the economy of derivation and presentation”, in Robert Freidin (ed.) Principles and Parameters in Comparative Grammar, MIT Press, 417–459. Cinque, Guillermo (1990) “Ergative adjectives and the lexicalist hypothesis”, NLLT 8: 1–39. Courtenay, Karen (undated) “Bambara Class material”, UCLA. Emonds, Joseph (1978) “The verbal complex V’-V in French”, Linguistic Inquiry 9: 151–175. Emonds, Joseph (1986) A Unified Theory of Syntactic Categories, Foris, Dordrecht. Fortescue, M. (1984) West Greenlandic, Croom Helm, London. Fukui, Naoki (1986) “A theory of category projection and its application”, unpublished PhD dissertation, MIT. Hoekstra, Teun and Rene Mulder (1990) “Unergatives as copular verbs: locational and existential predication”, The Linguistic Review 7: 1–79.) Huang, C.-T.James (1982) “Logical relations and the theory of grammar”, unpublished PhD dissertation, MIT. Huang, C.-T. James (1984) “On the distribution and reference of empty pronouns”, Linguistic Inquiry 15: 531–75. Jaeggli, Osvaldo (1982) Topics in Romance Syntax, Foris, Dordrecht. Kayne, Richard (1985) “L’accord du participe passé en français et en italien”, Modèles Linguistiques 7: 73–80. Kitagawa, Yoshihisa (1986) “Subject in Japanese and English” unpublished PhD dissertation, distributed by the Graduate Linguistic Student Association, Amherst. 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 (1985) “Theta theory and extraction”, GLOW newsletter. Koopman, Hilda, and Dominique Sportiche (1986) “A note on long Extraction in Vata and the ECP”, NLLT 4: 357–374 (Chapter 4 in this volume). Koopman, Hilda, and Dominique Sportiche (1991) “The position of subjects”, Lingua 85: 211–258. Kuroda, S.Y. (1986) “Whether you agree or not; rough ideas on the comparison of English and Japanese”, Linguisticae Investigationes XII: 1–47. Larson, Richard (1985) “Bare-NP adverbs”, Linguistic Inquiry 16: 595–621. Larson, Richard (1988) “On the double object construction”, Linguistic Inquiry, 19: 335–391. Mahajan, Anoop (1990) “On the A/A-bar distinction”, unpublished PhD thesis, MIT. Pollock, Jean-Yves (1981) “On case and impersonal constructions”, in Robert May and Jan Koster (eds) Levels of Syntactic Representation, Foris, Dordrecht, 219–252. Pollock, Jean-Yves (1989) “Verb movement, UG and the structure of IP”, Linguistic Inquiry 9: 101– 136. Rizzi, Luigi and Ian Roberts (1989) “Complex inversion in French”, Probus 1: 1–30. Rouveret, Alain and Jean-Roger Vergnaud (1980) “Specifying reference to the subject”, Linguistic Inquiry 9: 197–260. Sadock, Jerry (1985) “Autolexical syntax: a theory of noun incorporation and similar phenomena”, in NLLT 3: 300–319. Safir, Kent (1985) Syntactic Chains, Cambridge University Press, Cambridge.


Salleh, Ramdi M. (1987) “Fronted constituents in Malay; base structure and move A in a configurational Non-Indoeuropean Language”, unpublished PhD dissertation, University of Washington. Speas, Margaret (1990) Phrase Structure in Natural Language, Studies in Natural Language and Linguistic Theory, Kluwer, Dordrecht. Sportiche, Dominique (1988) “Conditions on silent categories”, unpublished manuscript, UCLA. Sportiche, Dominique (1990) “Movement, agreement and case”, unpublished manuscript, UCLA. Stowell, Timothy (1983) “Subjects across categories”, The Linguistic Review 2: 285–312. Stowell, Timothy (1986) “Null antecedents and proper government”, NELS 16: 476–492. Torrego, Ester (1984) “On inversion in Spanish and its effects”, Linguistic Inquiry 15: 103–127. Tremblay, Mireille (1991) “Dative and possession: binary branching from the lexicon to the syntax”, unpublished PhD dissertation, McGill University. Williams, Edwin (1984) “There-insertion”, Linguistic Inquiry 15: 131–153. Woolford, Ellen (1989) “The position of unaccusative subjects in SVO languages: the sole NP constraint”, unpublished manuscript, University of Pennsylvania (revised version to appear in NLLT 11).

8 PREPOSITIONS, POSTPOSITIONS, CIRCUMPOSITIONS, AND PARTICLES The structure of Dutch PPs

0.1 Introduction The substantial broadening of the empirical basis of syntactic theory in the 1980s has resulted in a better understanding of the general architecture of syntactic structures and syntactic theory.* Syntactic structures are large structures, assembled out of small simple building blocks with a unique structure design. With large structures and simple design, the hypothesis that structural variation between languages is minimal or non-existent can be reasonably entertained. Linguistic variation can be seen as the result of different movement options being exercised (which constituents move, how “big” are these constituents, Kayne (1995), Koopman (1994), Koopman (1996), Sportiche (1995), Chomsky, (1995)) operating on fixed and crosslinguistically invariant structural skeletons (Sportiche (1995), Koopman (1996), Cinque (1997)). Seriously testing this hypothesis, however, presupposes a good understanding of the invariant skeletal structure, which is all too lacking. In this chapter, I explore the architecture of PPs, a quite modest syntactic category, with the ultimate goal being a better understanding of the structure of Ps universally. I will do so not by analyzing patterns of crosslinguistic variation and drawing conclusions based on these patterns, but rather by attempting to provide a uniform analysis of the syntax of Ps in Dutch. As is well-known from the extensive literature on this subject, starting with the seminal work of van Riemsdijk (1978), the syntax of Ps in Dutch is extremely rich. It thus lends itself well to this enterprise. There are many different types of Ps (prepositions, postpositions, particles and circumpositions (complex Ps)). The distribution of modifiers and pronouns within the PP is intricate, providing a good starting point for determining the internal structure of PPs. Different overt movement processes apply to Ps and PPs: movement out of PPs resulting in P-stranding, head movement of Ps (incorporation), pied-piping of PPs, scrambling of PPs, and PP over V. Taken together these should allow us to form a solid picture of the structure of PPs, and how everything hangs together. The development of a unified analysis for Dutch Ps should further our understanding of the necessary properties of the underlying invariant structure, which by hypothesis, underlies the syntax of all human languages.


0.2 Expectations and theoretical assumptions The structure of PPs will be established by using as analytical tools those aspects of the theory that are relatively well understood, in particular the basic form of syntactic structures and movement theory. I depart from much current syntactic practice, which imposes additional methodological restrictions on possible analyses. In accordance with my recent research, (Koopman (1996), Koopman and Szabolcsi (1998)), I avoid explanations using Economy, and rely on purely mechanical solutions instead. Since syntactic structures are binary branching (Kayne 1985), Ps minimally project a PP projection, with a possible Spec and a complement position. Usually PP is taken to be the maximal projection of P as well for the purpose of external syntax. However, just as work on the internal structure of clauses1 and DPs2 has established that the lexical projections of V and N, VP and NP, are dominated by a number of functional categories, PPs might be expected to be dominated by functional categories as well. This is directly confirmed by the existence of inflected Ps3 in many languages, showing that the extended projection of PP can contain at least an Agr projection. Different word orders are derived by movement of different constituents from a common skeleton. Given the essential role of movement, it is important to spell out the theory of movement adopted in this chapter. This chapter assumes strict locality of movement and domain extension: (1) a. XP movement proceeds through the local Spec.4 b. Head movement is strictly local.5 c. Head movement extends the domain of movement turning the Spec position of the landing site into a locally accessible Spec.6 Head movement makes direct movement to some higher Spec position possible. 1. Dutch PPs 1.1 The problem Dutch superficially has prepositions, postpositions, circumpositions (which are made up of a preposition and a postposition or a postpositional element), and particles: (2)

b.

a.

op on

de the

de berg op the mountain on “onto/up the mountain”

tafel table

preposition

postposition

200 THE STRUCTURE OF DUTCH PPS

Table 1

c.

d.

op iemand af komen on someone from come “come towards someone” Ik heb jou opgebeld I have you upcalled “I called you up.”

preposition and postposition

particle

Since these all look alike, the null hypothesis is that prepositions, postpositions and particles belong to one and the same syntactic category P (Jackendoff (1973), van Riemsdijk (1978), and Emonds (1976, 1985)). All Ps, including semantically empty Ps,7 therefore minimally project PP. This is what all Ps have in common. However, each of the PP projections in (2) behaves differently with respect to the internal syntax, i.e. the distribution of PP internal material, and the external syntax, i.e. with respect to piedpiping, i.e. movement of a PP containing a wh-phrase, PP-over-V,8 i.e. the possibility for a PP to occur to the right of the verbal complex, P-stranding and P-incorporation. The following table summarizes the complex distribution in anticipation of the sections below. As this table shows, it is necessary to distinguish between nondirectional and directional PPs. Pied-piping under wh-movement, or scrambling, is possible for (non-idiomatic) prepositional phrases, but basically excluded for postpositional and particle phrases. Piedpipable PrepPs can in principle also occur in the PP-over-V position, except for directional (prepositional) PPs. P stranding reveals asymmetries: Dutch Preps can be stranded in the right structural configuration, but only a class of morphologically distinguished elements, the so called [+R] -pronouns, may escape from the projection of a preposition. DPs and PPs can escape the projection of postpositions and particles in the


right configuration. P incorporation to V reveals an asymmetry as well. Given the right structural environment, Dutch postpositions and particles can incorporate to V but prepositions cannot. The analytical problems that arise are complex, as the table above shows. How should one account for the head initial or head final character of the respective projections (i.e. this is a problem of the internal structure), how should one account for the distributional properties of the different parts of the PP (the complement of P, the P head, and so on), and for their external syntactic distribution (as being able to undergo syntactic movement or not)? I will develop a unified account for the different types of Ps, which will all be argued to project head initial structures, in accordance with Kayne (1994). In other words, the orientation of the basic building block is fixed. I will establish that Prepositional PPs contain a functional category Place as well as two other functional projections. Postpositional phrases combine a functional projection Path with some projection of Place. This structure, motivated on purely syntactic grounds, mirrors Jackendoff s (1990) conceptual argument structures for PPs quite closely. Different word orders fall out from independently motivated movements operating within the extended projection of P. The external syntactic distribution follows from the amount of functional structure present within the PP. PPs differ in the same way as sentential complements do, resulting in CP, IP or VP complements. Prepositional PPs will be shown to be parallel to full CPs, directional Ps to IPs and PartPs to bare VPs. 2 Non-directional prepositional phrases The analysis starts with non-directional prepositional phrases: of all PP projections, their properties are probably best understood.9 The internal structure of prepositional phrases is established in the sections below, based on locative Ps. Each structural level is motivated, and the entire structure is summarized in (45) (section 2.4). 3.1.3.1 discusses how this structure accounts for the external distribution of prepositional PPs. 2.1 R-pronouns While DP objects follow prepositions, inanimate pronominal objects precede them. Inanimate pronouns belong to a particular morphological paradigm, which earned them the name r-pronouns.10 The general locative pronouns also belong to this paradigm. (3)

a. b. c. d.

op on *de the op on Hij

de the tafel table Jan John heeft

tafel table op on op on him er

op on *het it hem

*het it op on *Jan *John gewoond.

op *er on there er op there on op *hem op up *him up (locative pronoun)


he has there “He has lived there.”

lived

The following analytical questions arise: (4)

a. b. c.

Where is er? Why is this position restricted to r-pronouns, and what explains the homophony of the inanimate and locative pronoun? 2.1.1 R-pronouns are in Spec

R-pronouns show the typical behavior of elements occupying some Spec position. They are to the left of P (5), yet still within the PP, as pied-piping of PP in (6) shows. (5)

(6)

Ik heb dat boek I have that book “I have put that book on there.” de tafel, waarop ik the table, whereon I “the table, on which I put the book”

daarop there

het the

boek book

gelegd. op put

heb have

gelegd put

They can undergo further movement, either to the position where other clitics occur (7a) or to Spec, CP (7b), stranding P: (7)

a.

b.

Ik heb er dat boek op gelegd. I have there that book on put “I have put that book on it.” Waar heb jij dat boek op gelegd. Where have you that book on put “What did you put that book on?”

R-pronouns should thus be analyzed as occurring in some Spec position, as argued in Van Riemsdijk (1978), from where they may escape further. This Spec position is restricted to R-pronouns, and unable to host non-R DPs, as shown in (8). (8)

*deze this

tafel table

op on

(versus daarop)

The ability to escape from PP correlates with the ability to reach a designated Spec position. Non-R DPs cannot strand P, in contrast to r-pronouns:


(9)

*Welke tafel heb je dat Which table have you that “Which table did you put the book on?”

boekje book

op on

gelegd? put

Van Riemsdijk (1978) explains this as a locality effect. A lexical DP cannot strand P, because it cannot reach the escape hatch of the PP. This situation is comparable to the distribution of English main verb: main Vs may not invert (e.g. occur in C), because of an an intermediate landing site which is “hostile” to main Vs. We return to this issue in section 2.3.3 where we propose a different account. So far, the data are compatible with a dual analysis of r-pronouns, either as occupying a Spec position, or, as the spelling suggests, as being incorporated to some head. However, given the necessity of the Spec analysis, and in the absence of arguments for incorporation other than the spelling, I assume that the incorporation analysis is simply unavailable to the native speaker. 2.1.2 Which projection hosts r-pronouns? There are three potential hosts for r-pronouns. First, they could be in the Spec of the projection containing the P. I will reject this option, since it can be shown that r-pronouns occur higher than this. Second, they could be in Spec, AgrP, i.e. the Case position where lexical DPs within the PP are licensed. This option will be rejected, because this position must be reserved for regular pronouns (compare 2.1.2.2). This leaves a third option: rpronouns move to Spec of a designated projection, which will be labeled Place. 2.1.2.1 R-PRONOUNS ARE HIGHER THAN SPEC, PP There is empirical evidence that r-pronouns occupy a position higher than Spec, PP. The location of the P can be further specified (van Riemsdijk 1978): (10)

omdat ik ze boven in de because I them up in the “because I have put them up in the drawer”

la drawer

gelegd put

heb have

In this configuration the r-pronoun must precede the place specification (boven) (van Riemsdijk 1978): (11)

omdat ik ze er boven because I them there up “because I have put them up in there”

(*er) in in

heb have

gelegd put


Since the r-pronoun cannot follow boven, it cannot be in the projection containing the P in. If r-pronouns are attracted to some higher Spec position, this distribution would fall out from the geometry of the tree. I take this as an argument for locating er in a projection on top of the projection containing the lexical P.11 2.1.2.2 R-PRONOUNS ARE NOT IN SPEC, A GRP R-pronouns correspond to DP objects of P, and Spec, AgrP is thus a reasonable candidate, pointing to a hierarchical structure AgrP>P. There is evidence based on Q float within PPs that (non R) pronominal pronouns occupy this position (Koopman 1993). Non rpronouns follow the P, (P>Agr), but r-pronouns precede (r-pronoun>P>Agr) and are therefore not in Spec, AgrP. I summarize the argument for Dutch. At issue is the question whether rpronouns are in Spec, Agr or in some other Spec position, to be labeled Place below. Floated quantifiers provide important clues of the internal organization of phrases, as the influential work of Sportiche (1988) established. A floated Q can be associated with the object of a P (it is important to read all examples below without stress on the pronoun and stress on the Q allemaal; lexical DPs can also strand Q, but are left out of consideration because of very interesting but ill-understood behavior): (12)

a.

Hij he Hij He

b.

heeft has heeft has

met with tegen against

jullie you

állemaal gepraat all talked hun állemaal gestemd them all voted

The floated Q is within the PP, as the entire string may appear in the first position of a root clause, traditionally taken as tolerating only a single constituent: (13)

a.

met with tegen against

b.

jullie you

állemaal heeft all has hun állemaal heb them all have

hij he ik I

gepraat talked gestemd voted

There are several possibilities as to the internal constituent structure of the PP constituent P pron Q. The pronoun could be in Spec, QP/DP (14a). It could form a constituent with the P (14b), or it could be outside of the QP, but lower than P (14c): (14)

a. b. c.

[…P [QP/DPproni [Q[e]i…] […[P pron]…[Q…] [PP…[P [XP pron X [QP/DP…Q…]]]…]

I will not consider (14b) seriously,12 but concentrate on (14a) versus (14c). (14a) predicts that the string pron+accented Q has the same distribution as the QP. This prediction is not


borne out as one can conclude from root clauses. Root clauses tolerate a single constituent before the finite verb, and allow a QP, but not a pronoun (+ accented) Q:13 (15)

a. * zij they c. * ons us

állebei/zij állemaal zijn gekomen both/they all are come állebei/ons allemaal belt zij iedere both us all calls she

dag everyday

op up

The ungrammaticality of (15b) shows that the pronoun is not in Spec, QP, but rather outside the QP. If pronouns must obligatorily raise to Agr, as argued in Koopman (1993), the stars in (15) follow from the fact that there is no Agr position available in root initial position. In other words, whenever a weak pronoun precedes a stressed floated Q, the pronoun is in Spec Agr outside of the QP. Yet, a preposition precedes this sequence. This eliminates the structure in (14a), and fits (14c), with XP=AgrP. Since pronouns follow the lexical P, Agr must be lower than the projection where P surfaces. I will assume P has raised to some head position higher than AgrP (which I will simply call PP for convenience), hence the structure in (16). (Positions containing overt lexical items are boldfaced. I leave lexical DPs out of consideration: they could be in Spec, Agr or lower. Nothing hinges on this decision.) Given this structure, then, r-pronouns are not in Spec, Agr, because they precede the overt P. 2.1.2.3 R-PRONOUNS AND PLACEP I have shown that r-pronouns are higher than the projection containing the lexical P, and eliminated PP and Agr as potential landing sites. I will assume that r-pronouns agree with a locative head, call it Place, following Jackendoff. R-pronouns are morphologically distinct and can be assumed to have a strong Place feature (an r-feature) which forces overt movement to Spec, Place:14


Non-R DPs do not encode Place morphologically, and this surely plays a role in why they may not move to Spec, Place. The motivation for movement depends on a property of the item that moves: r-pronouns have a feature that satisfies the Place head, but DPs do not.


We might therefore conclude, as I did in earlier versions of this paper, that DPs may procrastinate, but r-pronouns may not. It is unclear however, how this is consistent with Chomsky’s (1995) (highly desirable) proposal that movement is only driven by properties of functional heads: given overt movement to Place, Place must have a strong r-feature. (18)

Place has a strong feature.

Hence Place must attract some lexical material. But what happens then when there is a DP complement? A proposal that would make strength of Place dependent on the moved element (Place has a strong feature when there is an r-pronoun, and a weak feature elsewhere) should of course be rejected. We can simply assume that the feature is checked by pied-piping the entire PP to Spec, Place.15 Since the Place head itself is silent, the effect of this movement does not yield a different word order. This yields the following configurations for the Place head:

In other words, either the PP or the r-pronoun may satisfy the Place head. PP must move when it contains a regular DP, because the regular DP does not have what it takes to satisfy the Place head.16


The differences in derived constituent structure between (19a) and (19b) are important, because they give insight into what causes a basic P-stranding asymmetry in Dutch. Pstranding is possible with r-pronouns, but not with lexical DPs. In (19a) the r-pronoun and the P are “split” in the sense that they occur in two different projections, Place and PP. The r-pronoun is in Spec, PlaceP, a canonical extraction configuration, and can indeed extract further, yielding P-stranding. In (19b), P and DP are not split, but are contained together in Spec, Place. Extraction of PP (i.e. the string dominating P and DP) might be allowed, but extraction of DP out of Spec, PlaceP can be straightforwardly excluded as a left branch violation.

Thus, regular DPs may not strand P because the necessary separation from P cannot be created before the DP gets carried along and frozen on a left branch.17 Pied-piping is forced.


Finally, this representation provides room to express the surely non-accidental homophony of r-pronouns and locative pronouns. R-pronouns are in Spec, Place, where they are licensed. With P overt, prepositional er arises, with P covert, the locative pronoun:18,19 In sum, then: (22) a. b. c d. e. f.

R-pronouns are morphologically specified for Place. DPs do not encode a morphological Place feature. Place is strong. R-pronouns are attracted to Spec, PlaceP. PP is attracted to Spec, Place. Locative pronouns and r-pronouns are homophonous because they occur in the same structural configuration. 2.2 Deg(place) and C(place)

Apart from PlaceP, the extended projection of a PP can contain at least one, and probably two additional levels of projection dominating PlaceP. This can be established on the basis of the distribution of r-pronouns and certain bare adverbial modifiers of (locative or temporal) P (pal “right”, vlak “just”). R-pronouns either precede or follow such bare adverbs modifiers (van Riemsdijk 1978), with no meaning difference: (23)

a.

vlak

bij

het

huis

b.

(er)

vlak

(er)

bij


close (24)

a.

pal right

near

the

house

achter behind

het the

huis house

there b.

(er) there

close pal right

there (er) there

near achter behind

If modifiers always occupy the same structural position, there must be two Spec positions within the PP capable of hosting er: one preceding and one following er (as van Riemsdijk (1978) concludes). Since the modifier can precede er, it is higher than PlaceP. Er can also precede the modifier, hence there must be an additional Spec preceding the modifier. Dominique Sportiche (personal communication) suggests that the bare overt modifier actually is in a head position, heading a projection comparable to Degree phrase, specifying the “degree” of the Place specification.20 A zero modifier gives rise to an arbitrary PRO interpretation (behind the house (somewhere behind the house)) and vlak achter het huis would mean that “the degree” or “the distance” to “(the points) behind the house” is “very small”. den Dikken (1992:106) also suggests that the bare adverb is in a higher head position, and presents strong empirical support for the treatment of these bare adverbs as heads: bare modifiers block P-incorporation (the examples in (25) are adapted from den Dikken (1992:106)).21 (25) a.

dat Jan de bal pal/vlak that Jan the ball right/right “that Jan shot the ball right over”

b.

dat Jan de bal (?*pal/*vlak) that Jan the ball right/right “that Jan shot the ball right over”

over over

heeft has

(no P-incorporation) heeft geschoten has shot (P- incorporation) over geschoten over shot

Bare adverb modifiers can occur with an intransitive P (25a), but block P-incorporation, as (25b) shows. (25b) follows if the bare modifier occupies a head position dominating Place, call it Deg(place): P-incorporation is blocked because the P is too low within the PP and V is not the closest c-commanding head. These data thus argue in favor of treating bare adverbs as heads (Deg(place)). Is er preceding Deg(place) in Spec, Deg(place), or in yet a higher Spec position? The distribution of er with XP modifiers, which I take as occurring in Spec, Deg(place), reveals the presence of yet another projection dominating the modifier. Instead of a bare adverb, as in (25), an XP modifier can modify Place: (26)

dat Jan de bal twee meter over geschoten that Jan the ball two meters over shot “that Jan shot the ball two meters over the fence”

het

hek

heeft

the

fence

has


Head movement is not blocked in this case, as expected: (27)

dat Jan de bal twee meter that Jan the ball two meters “that Jan shot the ball two meters over”

heeft has

over over

geschoten shot (den Dikken 1992)

If both the XP modifier and the r-pronoun occupy Spec, Deg(place), measure phrases and r-pronouns should not be able to cooccur, contrary to fact: (28)

[daar twee meter achter] begint het there two meters behind starts the “Two meters behind it, no-man’s-land starts.”

niemandsland no-man’s-land (van Riemsdijk 1978)

(28) therefore reveals the presence of an additional projection dominating Deg(place). I will call the head of this projection C(place), to express the parallelism with CPs and DPs, and refer to its maximal projection as CP(place). I assume that the CP(place) level turns a PP into an “independently” licensed constituent, which enables it to undergo PP over V, scrambling or pied-piping under wh-movement (see 2.3.2). A final question concerns the position that P occupies in the overt syntax. Since P always follows er and the modifiers, and precedes pronouns, P at most can be as high as Place or Spec, Place if PP contains a regular DP. A full-blown structure for Prepositional Phrases is presented in 2.4. 2.3 External syntax of P and PrepPs Different aspects of the external syntactic distribution of Ps and their constituents, are discussed in sections 2.3.1 (P-incorporation), 2.3.2 (P-stranding) and 2.3.3 (Pied-Piping) respectively. 2.3.1 P-incorporation As is well known, Dutch has overt P incorporation: the position occupied by P within the verbal complex is restricted to bare heads only (P, N and A). When P is within the verbal complex following the finite verb, P incorporation has taken place (the incorporated P is boldfaced):22 (29)

dat ik Jan Marie (op) heb that I John Mary up have “that I wanted to let Mary call up John”

willen want

laten (op) let

bellen call


When P incorporates, the incorporator, say V, “governs” the position from which P incorporates, i.e. V must be the closest c-commanding head of the position containing P. P may not incorporate if the structural conditions for incorporation are not met. This can come about if there simply is no c-command between V and PP, or, more interestingly, if V c-commands the PP, but P is too low within the projection, i.e. V is not the closest ccommanding head of the position containing the overt P. This situation can arise if there is an additional structure between V and the position where P is spelled out, i.e. if the structure of PP is more complex.23 The structure motivated so far immediately explains why lexical prepositions fail to incorporate, even when V c-commands the extended projection of the PP and P stranding is possible: (30) a. dat zij er vroeger vaak mee heeft that she has there earlier often “that she often played with it a long time ago” b. dat zij er dit vaasje op heeft that she there this vase up has “that she wanted to put this vase up there”

(*mee) (with) has

gespeeld (with) played

willen (*op) want

zetten put

The head P is lower than Deg(place), and not in C(place) or in Deg(place). V is not the closest c-commanding head of P, and intervening projections are unable to host P. P therefore cannot incorporate. In other words, P incorporation can only occur if the following structural configuration holds: (31)

(i) (ii)

P raises to C(place). CP(place), DegP(place), or PlaceP are absent.24

Since P does not raise higher than Place in Dutch, (31 (i)) is never available in Dutch. Asymmetries with respect to P-incorporation must therefore fall out from the PP internal structure (31 (ii)). 2.3.2 Pied-piping Prepositional phrases can undergo pied-piping under wh-movement, scrambling, and appear to the right of the verbal complex (PP-over-V): (32)

a.

Met welke ouders heb with which parents have “Which parents did you talk to?”

jij you

b. Zij heeft met Jan maar heel she has with John just a short “She spoke only for a short while with John.”

gesproken spoken

eventjes while

(wh-movement)

(scrambling) gesproken spoken


c.

omdat ik gesproken because I spoken “because I spoke with John”

heb have

met with

Jan John

(PP-over-V)

Syntactic mobility has traditionally been taken as evidence for the constituency of a moved string. Failure to undergo wh-movement or pied-piping does not show that a projection is not a syntactic constituent, however. The extended projection of a PP consists of several syntactic constituents, which are all maximal projections (XPs). Yet, none of the projections smaller than CP(place), like PlaceP or PP, can undergo any of the processes illustrated above. This is shown in the following examples (since C(PP) is empty, it cannot be tested if DegP(place) can be extracted): (33)

a.

het niemandsland begint twee meters the noonesland starts two meters “No-man’s-land starts two meters behind it.”

daar there

achter behind

In sum, a constituent may very well be a maximal projection, but fail to undergo external syntactic movement (i.e. topicalization, wh-movement, or scrambling). This raises the question what property enables a projection to be able to “count” as wh-phrase, focus or topic. In DP and APs, these properties are located at the left edge, in the C/D domain, suggesting that this is where these properties are represented. Thus, the property enabling a constituent to undergo movement to the wh-landing site, or to FP, or TopP, is located at the CP (type) level. CP itself is a more articulated structure (Rizzi 1995). However, for the purposes of this chapter (34) is sufficient: (34) The property of being a wh-phrase, a topic, or a focus is represented at the C level of a particular phrase. Thus, if a PP has undergone wh-movement, scrambling, it must have the appropriate C level. Projections without the appropriate CP levels fail to undergo these external movements. This immediately accounts for the well-known restriction that idiomatic PPs can neither be wh-moved, scrambled or topicalized. Idiomatic PPs simply do not have what it takes, i.e. they are not “full” PPs, and lack the C level. Idiomatic PPs may not occur in the PP-over-V position either. This suggests that the PP-over-V position is a position that can only host CP(place), i.e. PPs with a CP level. This will become relevant in the discussion of directional PPs below. Given (34), external syntactic movement is a diagnostic criterion for the presence of CP (place): (35) PP has CP(place) level if it can move to Spec, CP, scramble, or occur in the PPover-V position.


In other words, the internal structure of PPs is comparable to that of clauses and DPs. Differences between types of PPs in mobility will follow from the amount of internal structure that is present. 2.3.3 P-stranding R-pronouns can strand P, because they are separated from the projection that contains the lexical P. DPs cannot strand P, because they are contained within the PP in Spec, Place. They have pied-piped before they could get a chance to separate from P (20). Thus, the asymmetry with respect to what elements can strand P falls out from the internal structure of P. It is of course well-known that the constraint on internal structure is not a sufficient condition. The extended projection of the PP (CP(place)) must be transparent, i.e. to use Barriers terminology, it must be “L-marked”, as well. In the remainder of this section I summarize the paradigm of P-stranding in Dutch, i.e. which configurations allow for P-stranding, and lay out how transparency could be achieved in terms of head movement. Stranded Ps must precede the verbal complex, but do not need to be adjacent to the verbal complex: (36)

Hij is er toen (mee) naar he is there then (with) to “He then went to the doctor with it.”

de the

dokor doctor

(mee) (with)

gegaan gone

Stranded Ps thus cannot be in the PP-over-V position: (37)

b.

a.

*Zij she

Zij heeft vroeger vaak gespeeld she has earlier often played “Earlier she often played with Lego.” heeft has

er there

vroeger earlier

vaak often

met with

legos Lego

gespeeld played

(PP over V)

mee with

Stranded Ps cannot be “too high” in the clausal skeleton, where “too high” refers to any position higher than NegP or focus particles like maar.


PPs that count as too high in this sense includes adjunct Ps (temporal, cause, and reason Ps),25 scrambled PPs (as in (42a)) as well as any other PP that has undergone A’ movement. (39)

a.

b.

(40)

a.

b.

(41)

a.

b.

(42)

b.

a.

Ik zal hoogstwaarschijnlijk daarna weggaan I will probably thereafter away go “I will probably leave after that.” *Ik zal daar hoogstwaarschijnlijk na weggaan I will there probably after away go de reden waarom de reason wherefore “the reason for which he left,… *de reden waar the reason where

hij he hij he

de manier waarop hij the manner whereup he “the manner in which he left,…” *de manier waar hij the manner where he Zij heeft vroeger (met legos) she has earlier (with Lego) (with Lego) “She once played with Lego often.”

Zij heeft er vroeger (*mee) she has there earlier “She once played often with it.”

vertrokken left om for

vertrokken left

vertrokken left op up

is,… is

is is

vertrokken left

vaak often

vaak often

(met legos) (with Lego)

mee with

In sum, stranded Ps must end up between NegP and the verbal complex.

is is

is,… is gespeeld played

gespeeld played


What allows P-stranding, i.e. extraction of an r-pronoun out of PP(CP)? There are basically two types of proposals in the literature. Head movement of the lexical P voids the barrierhood of PPs (see Zwart (1993) for a recent proposal); or PPs are transparent for extraction when they are “L-marked” (van Riemsdijk (1978), Sportiche (1988)). Lmarking itself has been related to incorporation by several linguists (Uriegereka (1988), Koopman (1994)). Head movement thus seems to be somehow involved in P-stranding. It can easily be shown that incorporation of the lexical P is not involved in P-stranding: stranded and incorporated Ps have different distributions. Stranded Ps cannot occur in the verbal complex (incorporated Ps can), but precede the entire verbal complex: (43)

de man waar Jan Piet gisteren the man where John Piet yesterday (*tegen) zien (*tegen) praten seen talk “the man who John saw Peter talk with yesterday”

tegen against

heeft has

heb have

gelegd put

In addition, the stranded P may be preceded by DegP: (44)

omdat ik het er zojuist because I it there just “because I just put it right on top of it”

vlak right

boven high

op up

As established in section 2.2, P is no higher than Place within the extended projection, and therefore cannot have been incorporated.26 Since the structure of PPs themselves is more articulated however, we can still maintain head movement’s involvement in P-stranding. PPs are topped off by a nonlexical C level: incorporation of C would allow escape of the r-pronoun from the PP projection. In fact, it would not only allow it, but it would also force it if incorporation of the C level “deactivates” the level that makes pied-piping of PP impossible. We leave deriving the condition that head movement of the C node must meet the structural condition of being lower than Neg/AgrO, but higher than the verbal complex, for future research.


2.4 Summary

Sections 2.1 and 2.2 motivated the following structure for locative PPs. Positions that may contain overt lexical material are boldfaced. PP shells or AgrP will play no role in the remainder of this chapter. This structure is taken to hold for all prepositional PPs that are not directional, and that allow for r-pronouns: How this structure accounts for the properties of non-directional prepositional phrases, presented in Table 2, is indicated overleaf.


Table 2

3 Directional PPs The syntax of directional PPs is quite complex and poorly understood, with many facts unexplained.27 Dutch has prepositional directional PPs and postpositional directional PPs. The latter consists of both circumpositional PPs, and simple postpositional PPs. Postpositional PPs must receive a directional interpretation, while prepositional PPs can receive a directional interpretation:29 (46)

(unambiguously directional) Zij is meteen in het water gesprongen she is immediately in the water jumped “She jumped into the water immediately.” (unambiguously directional) b. Zij is meteen het water in gesprongen she is immediately the water in jumped “She jumped into the water immediately.”

a.

The alternation between prepositional and postpositional PPs is not free, but restricted to specific syntactic environments. Prepositional and postpositional directionals only alternate when they occur as complement of a (motion) verb and the selected auxiliary is be.30 In other contexts, prepositional PPs are unambiguously locative. This is illustrated within DPs in (47):


(47)

a.

b.

de weg in the road in “the road in the forest” the weg the road “the road into the forest”

het the het bos in the forest

bos forest

(locative only)

(directional only) in

When the auxiliary have is selected, the PP can only be interpreted as locative and, concomitantly, a postpositional phrase is disallowed: (48) a.

Zij heeft in het water (op en neer) she has in the water up and down “She jumped up and down in the water.” b. *Zij heeft het water in gesprongen she has the water in jumped “She jumped in the water.”

(locative reading only) gesprongen jumped

This suggests the following generalization: (49) Prepositional directional PPs are only allowed when selected by motion verbs. Since postpositional phrases are one particular type of directional PP, their syntax can only be understood against a general understanding of the distribution and properties of directional PPs as a whole. The sections below determine the properties of each type of directional PP, using as analytical tools the distribution of the DPs, r-pronouns, modifiers, the incorporability of P, and the mobility under wh-movement. Prepositional directional phrases are examined in 3.1.1, circumpositional PPs in 3.1.2, and postpositional PPs in 3.1.3. 3.1 The structure of directional PPs Directional PPs are often complex (into, onto, toward,…), pointing to a complex underlying syntactic structure. Jackendoff (1990: 45) suggests the following conceptual structure for a sentence like John ran into the room: (50)

[EventGO [Thing JOHN], [Path TO([Place IN ([[Thing ROOM])])])])

As argued in 2.1.2.3, Place is syntactically represented as a functional head. Suppose that Jackendoff s Path is represented in a similar way, with Path selecting some PlaceP complement. This hypothesis is attractive because it entails that syntactic structure closely mirrors the conceptual structure. If Path is head initial and selects some projection of


Place, (either CP(place) or some smaller complement, say PlaceP, or PP) we are led to expect the syntactic structures in (51), (for convenience only head positions are indicated).

As I will show, the basic properties of directional PPs can be derived from these three structures, in a quite simple fashion. (51a) underlies prepositional directional phrases and


circumpositional PPs (section 3.1.1 and 3.1.1), (51b) simplex postpositional phrases (section 3.1.3) and (51c) directional particles (section 4.2).


3.1.1 Prepositional directional PPs The structure in (51a) is a good candidate for directional prepositional phrases, with P remaining in Place within the CP(place).

If P is not higher than Place, lexical items in higher head or Spec positions should cooccur with, and precede P. This is correct: (53)

a.

b.

Ben jij er langs gelopen? are you there along walked “Did you walk along it?” jij bent vlak langs de afgrond you are right along the precipice “You walked right along the precipice.”

gelopen walked

The projection dominating P therefore contains at least a PlaceP (er occupies Spec, PlaceP) and a Deg(place), where bare adverbial modifiers like vlak occur. The presence of a CP(place) level can be determined on the basis of the external syntax: CP(place) projections can be scrambled or wh-moved (2.3.2). Some CP level must be present, because the preposition and its complement can undergo wh-movement: (54)

Langs welke afgrond along which precipice “Along which precipice did you walk?”

ben are

jij you

gelopen? walked


There are two potential candidates for CP levels in (52): either Path is dominated by some C-like projection and the entire PathP has moved, or CP(place) has extracted out of the PathP. The former option can be ruled out: other PathPs selected by verbs of motion may never be wh-moved (3.1.3.1). This follows if PathP selected by motion verbs lack the CP (Path) level necessary for mobility. If PathP cannot have moved in (54), it must be CP(place) that escaped from PathP. Movement out of the PathP is allowed because of the accessible Spec, Path position, which is a licensing position for CP(place), as discussed below (3.1.2.2). Prepositional directionals therefore contain a CP(place) projection: [ [pathe] [CP(place)……P]

(55)

(55) contains a silent Path head, with a CP(place) complement. This raises two further questions: what is the distribution of the silent Path head and is there any need to determine its location with respect to CP(place) in the overt syntax i.e. is the path head postpositional, or prepositional? 3.1.1.1 THE DISTRIBUTION OF THE SILENT PATH HEAD Directional prepositional phrases only cooccur with motion verbs (49). This suggests that the Path node is in a chain relation with the motion verb. Following Koopman (1994) we interpret this as evidence that silent Path is licensed through overt incorporation into a motion verb. (56)

Silent Path is a trace whose antecedent is incorporated in a verb of motion.

Let us consider how (56) excludes directional prepositional phrases from DPs. Consider the representation in (58): (58)

*

[N

[PathP

[Path

e][CP(place)

]


This violates (56) since Path is not a trace whose antecedent is incorporated into a verb of motion. Thus, silent Path in (58) cannot head a chain. It cannot count as a trace of a Path that has incorporated into N either, as below, because this requires N to be an appropriate host (licensing head in the terminology of Koopman (1994)): (59)

*[[[Path e]i N][PathP [e]i [CP(place)]

N, in contrast to V, is quite generally not a licensing head, as argued in Koopman (1994). If N cannot license the Path feature, there is simply no way to satisfy (56). Postpositional directional phrases are possible in complement position of N (compare (47b)). This means that the Path node escapes (56), and must be independently licensed in this environment. The way it escapes (56) is by being overt, not silent. As shown in 3.1.3, P raises via Place to Path in this configuration (in addition, the complement moves to the left of Path): (60)

[ [PathP[Path P]i [PP [P e]i ]]

Thus, either P raises to Path, or else silent Path raises to V. This could suggest that the Path head must be attached to an appropriate lexical host, where P and V are appropriate hosts, but N and A are not.31 In other words, Path would act like a “bound” morpheme which attaches to either V or P, i.e. which selects for a [−N] category: (61)

Path is a bound morpheme selecting for a [−N] category.

(61), however, is to be rejected as an explanation for the following reason. It predicts that P either incorporates to Path, or that Path incorporates to V, and that they cannot cooccur. However, incorporation of Path into the motion V is still possible even if Path is lexical, i.e. if (61) is satisfied. P to Path is therefore independent of the relation between Path and a selecting verb. We conclude: (62)

Silent Path cannot be licensed within the projection of the PathP.

Path containing lexical P escape the effects of (62). 3.1.1.2 THE ORDER OF PATH AND CP(PLACE) Since Path is silent, the relative ordering of CP(place) and Path cannot be determined on the basis of these directional PPs alone. However, all other cases of directional phrases involve movement of a phrasal projection to Spec, Path (3.1.2 and 3.1.3) which result in


(63)

[PathP CP (place)i

[path e] [CP(place) ei] 3.1.2 Circumpositional PPs

Circumpositional PPs fit into the proposed structure for prepositional directionals, but contain more lexical items, hence slightly more structure. Some circumpositional PPs involve a postposition which is homophonous with a preposition. These therefore contain an additional PP where the P originates. Others contain a specific lexical postpositional element which I will assume is a direct lexicalization of the Path node.32 All have a regular preposition at the left edge: (64)

a.

“door, op, aan”

b.

“heen, vandaan”

onder de brug door under the bridge through over de stoel heen over the chair heen

tegen het huis op against the house up (van) onder het bed vandaan from under the bed from

The postposition associated with the Path reading or Path element acts as head of the entire projection. It can be incorporated into the verbal complex (compare 3.1.2.1 for examples and discussion). Path thus combines with a PlaceP complement that precedes it. The PlaceP complement contains at least a DegP(place) complement, as shown by the possible presence of a degree modifier: (65)

Het vliegtuig is vlak onder de The airplane is right under the “The airplane flew right under the bridge.”

brug bridge

door through

gevlogen flown

This is consistent with a full CP(place) in a Spec position to the left of Path, say Spec, Path (or alternatively in any other projection higher than Path):33 (66)

[[CP(placeP)/DegP(place) onder de brug

[Path Pi [PP [Pe]i [CP(place)] door

Thus emerges a second property of the Path projection: Spec, Path attracts lexical material. Spec, Path is not insensitive to the category that ends up there: it must be some projection of Place. This can be demonstrated by the following ungrammatical string: (67)

* [door [CP(place) onder de brug] [Pathe] [PP]

Underlying this string is a derivation in which door, instead of moving to Path, has piedpiped to Spec, Path in an effort to make Spec, Path content. The empty Path is licensed by


incorporating into V. Nothing so far excludes this derivation. The movement of the PP containing door is strictly local, and empty Path can be licensed by incorporating into V. What seems wrong here, intuitively, is that the moved constituent is of the Path category, and not of the Place category. Although Path contains the Place projection, this projection is far too embedded in the pied-piped constituent. In all good cases, Spec, Path contains a projection of the PlaceP; we therefore conclude: (68)

Spec of Path attracts a projection of Place.

This is the basic price to pay for a head initial Path. 3.1.2.1 PATH CONTAINS P The postposition in circumpositional PP is in Path, as shown by its incorporability into V.34 (69) a. dat

zij

gisteren

onder

de

brug

is

door

she

yesterday

under

the

bridge is

throu gh

“that she walked under the bridge” b. dat zij snel achter

het

konijn zijn aan

that

that they quickly behind the rabbit “that they chased the rabbit” c. dat de plant tegen het huis is that the plant against the house is

gelop en walke d gelop en walk

be

at

op up

gegroeid grow n

“that the plant grew up the side of the house” d. dat zij de fiets weer tegen de muur heeft aan gezet that she the bike again against the wall has at put “that she put the bike against the wall again” e. dat de kinderen stilletjes onder het balkon zijn langs gelopen that the children quietl under the balcony are along y walked “that the children quietly walked along under the balcony” (70) a. dat

zij

de

jas

over

de

stoel

heeft

heen


gelegd that they the coat over the chair prt (particle) has put “that they laid the coat over the chair” b. dat dit book (van) onder het bed is (? *vandaan) gekomen 35

that this book from under come “that this book came from under the bed”

the

bed is from

The postposition is therefore in the head position of this constituent, and there are no intervening projections between Path and V. These examples also show that the incorporability of Path into V is independent of the needs of the Path node. Path can be incorporated into V, even if Path is independently licensed by the postposition, as shown by its ability to occur in DPs: (71)

a.

b.

dat fietspad onder de that bikepath under the “that bikepath under the bridge” de reis door Europa the trip through Europe

brug bridge

door through

heen prt (particle)

3.1.2.2 THE COMPLEMENT OF PATH IS CP(PLACE) OR DEGP(PLACE) Path takes a PlaceP complement, which can be at least as big as DegP(place) in the case of circumpositional PPs: (72)

Het vliegtuig is vlak onder de the airplane is right under the “The airplane flew right under the bridge.”

brug bridge

door through

gevlogen flown

The moved constituent could be either Deg(place) or CP(place). The external syntax can differentiate between these two options. If it is a CP(place), further movement should be possible, if not, further movement should be blocked. The following examples show whmovement is possible for me and many other Dutch speakers: (73)

a.

Onder welke brug is het vliegtuig Under which bridge is the airplane “Under which bridge did the airplane fly?”

door through

gevlogen? flown


b.

Achter welk konijn zijn Behind which rabbit are “After which rabbit did they quickly run?”

zij they

snel quickly

aan on

gerend? run

For these speakers, it must be CP(place) that moves to Spec, PathP. Not all speakers accept such sentences however. Similar examples are given as ungrammatical in Koster (1987: 177). This suggests that these speakers analyze the constituent in Spec, Path as smaller than CP(place), i.e. as Deg(placeP). This is not at all implausible, since there are other cases of Path selecting a complement smaller than CP(place) in the language as well as we will see in (3.1.3). In sum, the overt syntax of directional PPs is driven by properties of the Path projection. Path attracts a projection of Place to its Spec; a silent Path head must attach to a [-N] category, causing it to either incorporate to V, or to attract P to it. This forces either P movement to Path, or Path incorporation to V. 3.1.2.3 EXTERNAL SYNTAX OF CIRCUMPOSITIONAL PPS Pied-piping of the entire directional PP under wh-movement36 is impossible, indicating that CP(Path) is absent: (74)

a.

*Onder welke brug door is Under which bridge through is “Under which bridge did the airplane fly?”

het the

vliegtuig airplane

gevlogen? flown

This goes well together with the fact that the head of the circumpositional PP can be incorporated, showing that it is in Path, and that no other head positions intervene between it and the V. Thus, PathPs lack the C level which would enable them to undergo pied-piping (compare 3.1.3.1). Scrambling of the entire PathP is impossible as well, pointing to the same conclusion: (75) a.

*Ik heb [e] gezet I have put b. *Ik heb gezet I have put

toen tegen

de

muur

aan maar

then

the

wall

to

FocP your bike

against

mijn

fiets

toen mijn

fiets

tegen

de

muur

aan

maar [e]

then

bike

against

the

wall

to

FocP

my


3.1.3 Simple postpositional phrases The structures so far determine the possible analyses for simple postpositional phrases to a large extent. Postpositions can optionally incorporate: (76)

a.

b.

c.

omdat zij de boom is in because she the tree is in “because she climbed into the tree” omdat zij het bos (door) because she the forest through “because she walked through the forest” omdat jij de kamer (uit) because you the room out “because you walked out of the room”

geklommen climbed is is

(door) gelopen walked

bent are

(uit) gelopen walked

If the place P occurs in Path in (76), the projection of PlaceP must have allowed P to escape. We know from prepositional phrases that P can move no higher than Place within CP(place). It follows that the complement of Path can be no bigger than PlaceP, respecting locality of head movement (78a), and cannot contain any of the higher projections, see (78b) opposite. The structure in (78a) finds additional empirical support. R-pronouns can be licensed (77b) revealing the presence of PlaceP. (77)

b.

a.

omdat zij de boom because she the tree “because she climbed into the tree”

omdat zij er because she there “because she climbed into it”

(in) in

is (in) is

in in

is is

geklommen climbed

geklommen37, 38 climbed

In sum, circumpositional PPs lack a CP(Path) level. Path being in final position. This suggests that Spec, Path always contains overt material, and that prepositional directionals are in fact hidden postpositional structures, with CP (place) in Spec, Path:


Postpositional phrases cannot contain overt realizations of Deg(place).39 This follows simply from the locality of head movement, which forces projections higher than Place to be absent: (79)

b.

a.

Overt Place can precede a directional prepositional phrase omdat zij boven in de boom is because she up in the tree is “because she climbed up into the tree”

Postpositional phrases cannot contain overt Place omdat zij (*boven) de boom geklommen because she up the tree climbed

geklommen climbed

(*boven)

in

is

up

in

is


“because she climbed into the tree” (80) a. Overt Deg (place) can precede a directional prepositional phrase omdat zij (vlak) langs de afgrond is because she right along the precipice is “because she walked right along the precipice” b. Postpositional phrases cannot contain overt Place omdat zij de afgrond (*?vlak) langs “because she the precipice right along c. omdat zij (*vlak) de afgrond langs is because she right the precipice along is

gelopen walked

is gelopen is walked” gelopen walked

Phrasal degree modifiers however can precede the postpositional object, and the P can incorporate: (81)

dat that

Jan John

drie three

meter meters

de the

boom tree

(in) in

is (in) is

geklommen climbed

However, the phrasal modifier modifies Path, not Place, i.e. it is in Spec, Deg(Path) which is higher than Path. Since it is phrasal, it does not block further head movement of the P that has reached Path. Postpositional directionals can appear within DPs, showing again that the Path node is licensed within the PathP: (82)

de weg the road “the road into the forest”

[het the

bos forest

in] in

P must have raised to Path to provide a lexical host for Path. This is of course consistent with the fact that simple postpositions can incorporate. Postpositional phrases thus represent the skeleton in (83). We next consider Spec, Path, which so far was argued to attract a PlaceP constituent (67). Postpositional order arises when some phrasal constituent containing the DP shows up in Spec, PathP. The category in Spec, PathP could in principle be a DP, (84a), a structure close to the traditional postpositional phrase, or a “remnant” PlaceP, or PP (84b):


In Koopman (1993), I argued in favor of (84a), mainly on theoretical grounds. In particular, I argued that (84b), with remnant PlaceP or PP in Spec, Path, was excluded by the ECP which subsumes the Proper Binding Condition. Since these projections contained a head trace, the ECP kept them in the c-command domain of Path in the overt syntax.


This conclusion, however, seems no longer tenable. There are good cases of head movement with exactly this configuration. Nkemnji (1995), for example, presents excellent evidence that such configurations are fine.40 How then can we distinguish between these two possibilities?

It turns out to be quite difficult to determine which structure is correct. (84b) is attractive since it allows maintaining in its most general and simple form that Spec, Path demands a PlaceP constituent. This would make the structure in (84a) simply unavailable. Moreover, given that remnant movement is allowed, it would simply be difficult to block (84b). What would we need to say if (84a) were correct? First, it would require a complication of the statement of what can satisfy Spec, Path. Not only PlaceP, but a DP “contained” in PlaceP would do as well. Furthermore, (84a) requires a way to block the derivation in (84b). On general grounds, then, (84b) seems the simplest and hence the preferred analysis. One might explore blocking (84b) and forcing (84a) by finding a reason that (84b), though the simpler analysis, would not lead to convergence. This could be tied to the licensing of DP in (84b). If DP fails to satisfy the Case filter in this configuration, perhaps (84a) could be forced. Thus, the DP in (84a) would satisfy the Case filter in PathP, not in PlaceP, and Case is unavailable within the PlaceP in (84b). This is a priori an attractive move, given the existence of languages in which directional Ps license their own Case. In German, for example, directional Ps license the accusative Case. This cooccurrence restriction should be structurally captured by moving the DP to a Case position in the PathP. We look at this property below. The analysis just outlined predicts that Case is satisfied external to the PlaceP but internal to the PathP. There is evidence that suggests that this is incorrect for Dutch. Case


in PathPs is (almost) always determined within the PlaceP. If it is not, it is determined outside PathP. Pronominal DPs within postpositional phrases can show up either as r-pronouns or, in restricted cases, as accusative pronouns. The distribution is difficult to establish: …there are unfortunately, many stylistic, dialectal, and other factors that influence the judgments on the choice of r-pronouns or non-r-pronouns in such examples. For some reason, relative clauses show the ambivalent behavior most clearly. (van Riemsdijk (1978:98–99)) It seems clear however that r-pronouns have the widest distribution, and that accusative pronouns are quite restricted. R-pronouns are licensed PlaceP internally, which we take as evidence that case is assigned PlaceP internally, accusative pronouns PlaceP externally. Given this background, consider the fact that accusative pronouns are possible in clauses, but excluded from parallel DPs: (85)

a.

hij is de boom/ hem/ he is the tree/ him/ de klim de boom in the climb the tree in “the climb into the tree” *de klim hem in the climb him into (?)de klim er in the climb therein

b.

c. d.

er in there

geklommen inclimbed

(85c) shows quite straightforwardly that accusative case does not depend on properties of the Path projection, but on properties outside the PathP. If this is correct, the DP (the tree) in (85b) should not get its Case within PathP either. Rather its Case should be determined within PlaceP, as the occurrence of r-pronoun (85d) shows:41 (86)

a. b.

Accusative case is not assigned within PathP. Case on DPs is determined within PlaceP.

Case does not provide any support for (84a). We will therefore assume (84b) must be available as an analysis for postpositional phrases. There are of course further important questions relating to accusative Case. What is clear, minimally, is that there must be a way in which DPs can escape from the PathP so that they can have (Path external) accusative in particular circumstances in clauses. We leave open the problem of how these derivations proceed, but would like to suggest that there is a stage in the derivation where Spec, PathP contains a remnant PP. Directly related to the previous point is the fact that further DP movement is possible in postpositional structures, resulting in P-stranding. (As (87b) shows, DP extraction is independent of P incorporation):


(87)

a.

welk bos is hij ingelopen? which forest is he into walked “Into which forest did he walk?” b. omdat hij zo ‘n donker bos niet (in) lopen because he such a dark forest not in walk “because he doesn’t dare walk into such a dark forest”

durft (in)

te

dares in

to

We leave the problem of how exactly the DP is able to escape from the postpositional PP and separate from P unsolved: we do assume that it involves the step in (84b). 3.1.3.1 EXTERNAL SYNTAX OF POSTPOSITIONAL PPS PathPs could be further dominated by a CP(Path) level, or not, i.e. they could parallel fully articulated clauses, or reduced clauses. Postpositional PPs with a CP(Path) level should show the diagnostic properties associated with this level, i.e. they should be able to scramble or pied-pipe under wh-movement, or occur in the PP-over-V position. Postpositional PPs selected by verbs fail to pied-pipe under wh-movement, scramble or undergo PP over V, and thus behave as lacking a CP(Path) level (just like circumpositional PPs). (88) a.

*Welk bos in ben jij Which forest in are you “Into which forest did you walk?”

b.

*Ik ben de kamer uit I am the room out “I did not walk out of the room.”

(pied-piping under wh-movement) gelopen? walked (scrambling)

niet not

gelopen walked

(PP-over-V) c.

*Zij zijn gelopen het they are walked the “They walked through the forest.”

bos forest

door through

Is this property restricted to PathPs selected by verbs, or does it hold for PathPs in general? Postpositional PPs can occur independent of motion verbs (van Riemsdijk (1978, 1990)). Yet, they cannot be wh-moved when this can be tested: (89)

De the

kamer room

uit out

met with

jou! you


“Out of the room with you!” (90)

(91)

de weg the road “the road into the city” a.

b.

de the

omdat hij meegereden is, because he withdriven us, “because he drove with us, up the mountain” *Welke berg op is hij which mountain on is hij “Onto which mountain did he ride with you?”

stad city

de the

in into

berg mountain

op up

meegereden? with rode

I conclude that postpositional PPs always lack a C level, and that this is a general property of PathPs (compare 3.1.2.2).42 In this respect, the Path projection resembles verbal projections, say like VP, which cannot be wh-moved or scrambled either. I return to further similarities between Path and V in section 5.3.1. 3.2 Summary: directional PPs The properties of directional PPs of Table 2 have now been discussed. The account can be summarized in the following table: In the literature, verb particle constructions are either base generated as part of a complex V (and therefore do not project a P-type syntactic projection (Koster (1975), Johnson (1991), among others), or they project some syntactic projection. The projection containing the particle is argued either to be the projection of an intransitive P (a P with no complement, as in Emonds (1976), (1985)), or starting with Kayne (1985), of some type of small clause in which the argument of the verb particle combination originates, either in subject position of the particle (Kayne 1985), or in the complement position of the particle (Taraldsen (1983), Guéron (1986), den Dikken (1992) and Koopman (1991), among others). For the purposes of this chapter, any small clause structure will do


the job: what matters is the categorial nature of the particle (P), and the fact that the complement originates within the PP. The question I would like to address is how the PP projection of the particle relates to the general structure of PPs established in this chapter.


4 Particles Particles homophonous with prepositions have a variety of uses: idiomatic, directional and aspectual. This section shows how idiomatic and directional particles fit into the structures established so far, but will not go into any of the other issues particles raise. 4.1 Idiomatic particles Idiomatic particles form a thematic complex with V, and lack autonomous thetaproperties (see also Kayne (1985)). Particles are like unaccusative verbs, and do not assign accusative Case. Given the absence of independent lexical properties, a reasonable hypothesis is that idiomatic particles project a PP without any functional layers:

This structure accounts for the syntactic distribution of particles. Particles can incorporate into V, because V is the closest c-commanding head of V: (94)

omdat ik Jan niet because I John not “because I have not called John up”

heb have

opgebeld upphoned

The complement of P can escape the PP via Spec, PP as usual. When the complement is a DP, as in (94), DP movement is obligatory: the unaccusativity of the particle (compare Koopman (1991)) implies lack of case properties. Thus, there is no AgrP within the PP itself. Depending on which Case is available externally, the DP will move in search of an accessible Case licensing position. (93) does not contain a PlaceP level. This is supported by the failure of PlaceP material to surface in verb particle constructions: (95)

omdat ik het/*er because I it/*there “because I looked it up”

heb have

opgezocht uplooked


Particles cannot be accompanied by bare modifiers, establishing the absence of Deg (place): (96)

omdat ik het (*vlak) because I it right “because I looked it right up”

op up

heb have

gezocht looked

Idiomatic particle verbs therefore consist of a V selecting a bare PP complement. Since the CP(place) level is absent, the PP cannot be preposed,43 scrambled, or occur in the PPover-V position. (97)

a. b. c.

*op heb ik het niet gezocht up have I it not looked *omdat ik het op niet heb because I it up not have *omdat hij het heeft gezocht because he it has looked

gezocht looked op up

4.2 Directional particles Directional particles express Path, and therefore contain a projection of Path. Thus far, Path was shown to take a CP(place) complement, or a PlaceP complement. Directional particles fill the gap in the paradigm, with Path taking a PP complement (see (98)). P raises to Path in directional PPs. From there, it can further incorporate into V. DP moves to Spec PathP and continues on its journey in search of an appropriate licenser. Since the complement of Path is a bare PP, r-pronouns cannot be licensed (see (99)).


(99)

Ik heb niets/* I have nothing/ “I picked up nothing.”

nergens nothing+er

opgepakt. up picked

Since the particle cannot be modified by a Deg(place), Deg(place) must be absent: (100)

Hij heeft het He has it “He picked it right up.”

(*vlak/*pal) right

op up

gepakt. picked

CP(place) must be absent, and directionals always lack a CP(Path) node: directional particle PPs can basically not be preposed, scrambled or occur in the PP-over-V position. Directional particles resemble idiomatic particle constructions except that they are embedded under Path. The projection of directional particles differs from other directional PPs in that the complement of Path is a bare PP, not a PlaceP nor a CP(place). 5 General issues This chapter focused on the architecture of PPs in Dutch, the development of a unified account for the different types of surface PPs, and the distribution of their constituent parts. What looks like a relatively simple syntactic category turns out to be quite complex, as usual. In this section, I briefly summarize the major results, and address some general issues. 5.1 Structures As was shown, functional categories are not only expected within the extended projection of a P: their existence can in fact be quite firmly established on the basis of the overt syntax. Two basic semantic types of PPs must be distinguished: directional PPs and nondirectional PPs. To these semantic categories correspond functional categories, for which Jackendoff s (1990) labels Place and Path seem entirely appropriate. The syntactic structure, motivated by distributional evidence, closely mirrors the conceptual argument structure (Jackendoff 1990). This strongly supports the idea that the syntax builds the structure necessary for the semantic interpretation. Other functional categories involve Agr, degree modification, and C-like categories. The following structures were motivated: Within this structure, the preposition is never spelled out any higher than Place. From this, it follows that Ps can never incorporate: P is simply not high enough within its extended projection to enter into the necessary structural relation to V (or whatever PP external category it would incorporate to).


The discussion focused on locative PPs, and never addressed the problem how semantically empty Ps fit into the picture. Empty Ps function in a variety of roles: as case markers and as Cs. In general, they have no particular semantic relation with the complement they license. Are these Ps Cs, as Kayne (1994) proposes, or Ps, as Emonds (1985) argues. What precisely is at stake here? Whether something is a C or a P depends on the functional structure associated with the categories dominating the head, not necessarily on the complement structure, since both C and P can take surface clausal complements. For concreteness consider a grammatical P comparable to of, i.e. Dutch van. Van looks like a P, and shares with P the property that it projects at least a PlaceP projection, in which r-pronouns can be licensed: 5.2 PlacePs


5.2.1 Semantically empty Ps (102)

de verwoesting van de stad the destruction of the city

de verwoesting ervan the destruction of it

Van is also dominated by a CP(place) since it can be wh-moved, scrambled or occur in the PP-over-V position. This shows unambiguously that van has properties in common with P. Although this might appear incompatible with Kayne’s (1994) proposal that Ps like of or van are Cs, it is in fact not. Kayne proposes that elements like van are Cs in that they select for a clausal complement: the following DP is not a direct complement of van, but occurs in some Spec position in the clausal complement. Nothing prevents analyzing van as a P (hence showing external syntax of CP(place)) which somehow combines with an IP, out of which a DP has raised (i.e. there is no direct complement relation between van and the DP).44 English C for can be treated in much the same way as Dutch van. Since it licenses accusative Case, there must be at least a P shell and an Agr shell present:

Prepositional complementizers raise the problem of exceptional Case marking. English for licenses Case on the subject of an infinitival, but Romance de or Dutch om do not. This could be taken as evidence that they do not provide any structural position for Case, i.e. these Ps would not project an AgrP projection. This proposal is unattractive since it still raises another question: how does a language learner determine whether P projects AgrP or not. Alternatively, these prepositional Cs project the same structure as for, including AgrP. The reason why Romance languages and Dutch do not allow for overt subjects in these infinitivals is not due to a structural difference, but follows from the different status of infinitivals in the languages in question. Infinitivals in Dutch and Romance languages have nominal morphology, but not in English. As a consequence, the entire infinitival complement is forced to raise to Spec, AgrP in Dutch and Romance languages whereas DPs raise in English.


This proposal reduces parametric variation (the structures are identical) and shifts the locus of variation to the size of the moved constituent. 5.3 PathPs Directional PPs have the skeletal structure in (106), with Path combining with some projection of the PlaceP:


PathP is never dominated by a CP type level, at least not by a CP level that makes whmovement or scrambling possible. This hypothesis is useful in that it accounts for why PathPs never undergo pied-piping under wh-movement or scrambling. The different constituents of the PlaceP that can be selected by Path are illustrated in (107) overleaf. (107) raises two questions: why exactly these categories and what determines selection. The latter involves general issues about complementation, and this chapter presents no new insights into these. The former question should be answerable, however. Spec, PathP must contain a PlaceP projection. A projection which does not carry this property recognizably on its sleeve, will simply not be selectable, because it will have nothing to offer to Spec, Path. Thus, AgrP is not selectable because PP is embedded under it. The lexical PP is, by virtue of its lexical place properties and PlaceP: it contains either the lexical P or the PP. It is less clear how CP(place) satisfies it, since in the derivations it is not structurally close to either PlaceP or PP.45


5.3.1 Path: P and V Although Path looks like a P, and not like a V, it has both P-like behavior and V-like behavior. Dutch has verb particle constructions, with P optionally incorporating to V. The overt P in PathP can optionally incorporate into V as well.46 This suggests that the Path head is part of a verb particle construction. If this is correct, PathPs are never dominated by a (wh-type of) CP projection because these projections are excluded with particles as well. V-like behavior includes the fact that non-CP complements of V precede V, and so does the complement of Path. Path and V do not have parallel Case properties: Path in Dutch is never responsible for accusative Case (compare 3.1.3).


Taken together, these observations might suggest that Path projections may involve both a verbal projection and a particle construction. In other words, the Path projection would be a verbal small clause headed by a light verb taking a particle.

This structure allows us to sharpen the issues. Which properties are due to which projections? Are the verbal characteristics due to the presence of the light verb? Are the P characteristics due to the projection of the particle? Consider the V projection. VP small clauses are generally excluded from DPs, yet PathPs are fine within DPs (de weg de stad in “the road into the city”). This indicates the absence of the VP small clause projection, leaving a “bare” PathP present. DPs then become a good test case to tease properties apart, as I have in fact been doing all along. Properties of the V projection should disappear within DPs, but properties of the Path projection should be present. PathPs within DPs are always postpositional. The leftward movement of PlaceP to Spec, Path is therefore not a property of V, but, as assumed all along, a property of Path. The shared property with V is accidental. Pronominal DPs within postpositional phrases can show up either as r-pronouns or, in restricted cases, as accusative pronouns. When accusative pronouns are possible in clauses, they are excluded from parallel DPs: (109)

a. b.

hij is de he is the de klim klim hem the climb climb him

boom/hem/er in geklommen tree/him/there inclimbed de boom in /de klim in into the tree /the climb into

er

in *de

there

n /the

This shows that accusative Case does not depend on properties of the Path projection, but on properties of the light verb (or other characteristics of the clause). Since this projection is missing within DPs, accusative Case is simply unavailable.


The similarity with verb particle constructions, which yields optional incorporation of Particles and PathPs, could in fact be due to the presence of the light V in verbal Path constructions. The presence of the light V would of course also be extremely important in light of the fact that languages with serial verbs typically use lexical verbs in directional constructions: if a structure like (106) underlies clausal directionals universally, questions about the overt forms of directionals crosslinguistically become discussable and answerable in precise ways. 5.3.2 Further questions In this chapter, I set out to explore the syntax of Ps, with the ultimate goal of getting a better understanding of the architecture of Ps universally. I did not do so by hopping around from one language to another, but tried instead to provide a uniform analysis of the syntax of Ps in Dutch. I arrived at a reasonable, coherent picture of the properties of the different types of PPs in Dutch, accounting for a large amount of data in a unified fashion. Although progress has been made, the last word has not been said about many of the issues raised here. In particular, it seems that we are at the beginning of understanding the extremely interesting issues surrounding the syntax of Path. Some problems that remain in this chapter are in fact general old theoretical problems, not problems related to my analysis in particular: the theory of complementation, what accounts for the optionality of incorporation into the verbal complex, and so on. My analysis does not yield any new insight either into the well-known problem that DPs can ccommand out of their PPs (see among others, Pesetsky (1995)). Arguments for the architecture of PPs should not only come from careful language internal analyses, but also from success or ease in handling crosslinguistic variation. Indeed, if structural variation between languages is minimal or non-existent, the structures motivated for Dutch should extend directly to PPs in other languages. Unfortunately, serious investigation of this issue goes beyond the scope of the present chapter. Notes * This paper grew out of an earlier proposal for the structure of English PPs (Koopman 1991). Koopman (1993) was originally written as an illustration that restrictive theoretical assumptions completely determine particular analyses and do so quite successfully. The present chapter supersedes and differs from earlier versions (“The structure of Dutch PPs”, 1993 and 1996). I would like to thank Hans Bennis, Marcel den Dikken, Teun Hoekstra, Ed Keenan, Henk Van Riemsdijk, Jeannette Schaeffer and Dominique Sportiche for their comments, as well as the participants of my seminars on particles (UCLA, Winter 1990) and on head movement (UCLA, Winter 1992) where the analysis presented in this chapter was developed. The usual disclaimers apply. Financial support from the Academic Senate of UCLA is gratefully acknowledged. 1 Pollock (1989), Chomsky (1991), Sportiche (1995a), Cinque (1997) and many others.


2 Abney (1987), Carstens (1991), Longobardi (1994), Ritter (1991), Szabolcsi (1987, 1994), Valois (1991), among others. 3 Interestingly, though, there always appear to be two classes of Ps: inflected Ps and uninflected Ps. 4 See in particular, Sportiche (1990, 1995b, 1998). 5 In particular, I adopt the theory of Head movement outlined in Koopman (1994, 1995a). 6 See Chomsky (1995). Precursors to domain extension are the Head Constraint of van Riemsdijk (1978), and the Government Transparency Corollary of Baker (1988). 7 Semantically empty Ps are basically used to create X-bar structure (or shell structures), in the same way as semantically empty Vs can be used to create subordinated structures (compare the cases of indirect complementation discussed in Koopman (1984), and Koopman and Sportiche (1989)). 8 This chapter presupposes a head initial VP for Dutch (following Zwart (1993), Koster (1993), and many others). PP-over-V therefore cannot be a rightward movement rule. I continue to use the term PP over V as a descriptive term to talk about PPs that can occur to the right of the verbal complex. For an interesting account of restrictions on PP over V, see Barbiers (1995). His proposal is incompatible with the proposal in this chapter. 9 The analysis of Dutch PrepPs presented below updates van Riemsdijk’s (1978) analysis and insights to a large extent. 10 Besides the general locative pronoun er, this paradigm contains the demonstrative (daar op “thereon”), [+ wh] (waar op “whereon”), negative (nergens op “nowhere on”), and the universal quantifier (overal op “everywhere on”) (van Riemsdijk 1978). 11 I assume that (11) is impossible because boven takes a “small clause” PP, i.e. a projection of P that is smaller than the projection where r-pronouns are licensed. 12 I thank an anonymous reviewer for the important observation that the Q can also be floated outside of the PP.

This might support the constituent structure in (14b) if the source of the floated Q can only be within the DP. These facts remain unaccounted for in this chapter, and merit further study. The text only considers floated Qs that are clearly PP internally. 13 These examples are acceptable with focal stress on the pronoun, and no stress on the Q. I leave these cases out of consideration. There is a slight contrast between subject and nonsubjects. A remnant VP preposing analysis might be available for non-subjects (with the preposed VP containing only the object and the floated Q) rendering the judgment less clear in the latter case. 14 It is well known that not all Ps allow for r-pronouns. van Riemsdijk (1978) argued that Ps select for the features of their Specs. In my terms, Ps that disallow +R lack a lexical feature +Place and therefore fail to license the PlaceP. The distribution of the +Place feature is interesting. Ps that express notions not transparently related to location in time or space, and allow for +R, can also all be used as locative prepositions. 15 For similar proposals, see Koopman (1996). 16 What remains to be explained is why an r-pronoun must move if it is contained within a PP, i.e. what explains the ill-formedness of * op er. My inclination is to not follow the Economy


line of explanation, but to pursue an account by which the r-pronoun “robs” the P of the structural property that satisfies Place. 17 What is therefore crucial for P-stranding is the separation of DP and P at an early point in the derivation. For English, this can be achieved in the way the chapter describes it for rpronouns: in English DP extracts to Spec, Agr, and the remnant PP goes to Spec, Place. DP, now separated from PP, can extract further. 18 I assume that silent P is licensed in Place, yielding the following structure of the locative: [er [P e]i [PP [e]]i

That the overt P is in PP, not in Place, is shown by PP to Place movement discussed above. 19 It follows that there must be a Place projection licensing the existential pronoun in existential sentences. 20 This recalls Corver’s (1990) proposals for a DegP in APs. 21 den Dikken attributes the examples in (25) to Bennis (1991). The judgments reported in Bennis differ though, as he judges the (b) examples as grammatical. My judgments presented here in (25) coincide with den Dikken’s. 22 Traditional descriptions recognize two positions for incorporated heads which are underlined in (i):

As traditional analyses acknowledge, there is dialectal variation not covered by the schema above. Bennis (1991) assumes that P can be anywhere in the verbal complex, as long as it is preverbal. In my dialect, there are three positions for incorporable elements: pre-finite verb, immediately following the finite (auxiliary) V, and preceding the verb it is theta-dependent on, but not in the starred-position below ((ii) representing the verbal complex of (29).

The difference between my dialect and the one described in traditional terms can be reduced to the distribution of the finite auxiliary. My dialect seems to allow the auxiliary to optionally raise to a higher head position than the traditional dialect described in (i): (iii)

a. b.

F X° aux V V. aux X° [e] V V


For more discussion on this issue, see Koopman (1995b). 23 Incorporation asymmetries can be derived in this purely structural way. I do not follow Baker and Hale’s (1990) proposal for parametrization of functional and lexical heads with respect to relativized minimality, nor do I assume that there are two different types of incorporation as argued in Uriegereka (1988). 24 Or alternatively, the smallest projection containing the lexical P pied-pipes to a Spec position where it is locally c-commanded by the incorporator. 25 Marcel den Dikken informs me that this generalization might be too strong since he accepts examples like the following:

I do not accept such examples, and I do not know how much variability on the judgments there is either. 26 The alternative analysis of treating the degree and Place modifier as a complex P undergoing incorporation should be rejected, since complex Ps, or complex heads, fail to incorporate. 27 The basic behavior of postpositions with respect to incorporation, extraction, and so on, is discussed in van Riemsdijk (1978). Koster (1987) contrasts extraction possibilities from postpositional PPs and prepositional PPs. The external syntactic properties of directional phrases in relation to have/be selection are discussed in Hoekstra (1984) and Hoekstra and Mulder (1990). van Riemsdijk (1991) was the first to propose that postpositional order derived from (rightward) moving the preposition to some (functional) P projection. My analysis maintains the idea that prepositional PPs and postpositional PPs are related through movement, (leftward though), and quite generally strives to present a uniform structural account of the whole class of directional PPs. 28 We do not discuss properties of P taking PP complements, as in (i).

These Ps introduce a temporal or locative argument which behaves as an adjunct with respect to islandhood (r-pronouns cannot escape). PP complements are contained within this argument, and cannot escape either. 29 The meaning difference between the prepositional and postpositional PPs in (46a) and (46b) is not clear. According to my intuition the object of a postposition receives a literal interpretation obligatorily (and the object of a preposition optionally). This accounts for the following contrast:

In (i.a.) the path described by the motion V involves the room. In (ii.b) the path cannot involve my eyes (you were never literally in my eyes), and the sentence is therefore ill-formed.


30 This is generally taken to show that any V which combines with a directional is unaccusative (Hoekstra (1984), Hoekstra and Mulder (1990)). 31 There do not seem to be underived adjectives in Dutch that take directional PrepPs. There are deverbal adjectives in Dutch that can take directional PrepPs (compare Broekhuis (1998)). Interestingly, these PPs are only prepositional, and cannot be circumpositional nor postpositional. This suggests that the Path feature is obligatorily incorporated into the verbal part of the adjective. 32 Some contain an optional postpositional element (toe, heen, vandaan).

(i)

Hij He

loopt walks

(ii)

de

weg

the

road

naar towar ds naar

Nijmegen. (toe) Nijmegen.

(leadin g)

to

Nijme gen (toe) Nijme gen

Optionality holds for clausal and DP contexts alike, showing Path is licensed Path internally. This suggests that the absence of the postposition is due to a PF deletion process. As Joost Zwart (1995) discusses, the postpositional element becomes obligatory when r-pronouns are extracted: (iii) (iv)

Hij He de the

loopt walks weg road

er there er there

naar. *(toe) towards naar *(toe) towards

If the absence of the postpositional element is handled at PF, then, blocking deletion should be handled at PF as well. I believe that deletion in (iv) is blocked for prosodic reasons. The preposition in circumpositional structures is always followed by a stress-bearing element (naar Gróningen (toe); er naar tóe). Optional postpositional elements can only be absent when non-stressed, i.e. in (i) and (ii), not when stressed in (iii) and (iv). 33 For simplicity, I have put the entire PP in Spec, Path. However, the PP could land in a still higher projection, (in accordance with the generalized doubly filled C filter (Koopman (1996)). This is not important in the present chapter. 34 The acceptability of P incorporation in this context seems to vary somewhat across speakers and within speakers. Judgments may vary depending on individual lexical items. All native speakers that I have consulted, accept at least several, if not all, cases of P incorporation. There is an extremely sharp contrast, however, between incorporation of the postposition and incorporation of the preposition in this structure. Any attempt at incorporating the latter yields total unintelligibility. Relative ease of incorporation therefore shows for all speakers that the postpositional element is the head.


35 The compound postposition vandaan does not incorporate, a property that holds for compounds in many languages. I take the failure of incorporation of vandaan to show that it is not really a single complex head, but rather a sequence of two heads in different head positions (i.e. it has more syntactic structure), with van being in the syntactically higher position. 36 The first position in non-interrogative root clauses can contain non-interrogative circumpositional PPs, but not interrogative circumpositional PPs:

This contrast might be explained if (i) is embedded in a different constituent, say VP, out of which the participle has been extracted, i.e. it would be a case of remnant movement (see also Note 43.) 37 Particularly interesting is the fact that in acceptable sentences like (i), P incorporation is blocked:

These facts follow. Since P is preceded by boven, P cannot have raised to Path. P must therefore be within the PlaceP: thus, CP(Place) is in Spec, PathP in these examples, and incorporation is simply impossible because of locality. 38 The existence of two derivations for directional PPs renders the analysis of the examples in (78) tricky. If the place P is within the CP(place), the structure represents a prepositional directional, and would reveal nothing about the postpositional structure. This derivation can be ruled out because P can be incorporated (compare the bold-faced P in (78b)). This is a diagnostic for P to Path movement yielding postpositional structures. 39 As a reviewer points out, phrasal degree modifiers can precede the postpositional object:

If the phrasal modifier in (i) modifies Path, not Place, this would be entirely unproblematic. 40 See also Mueller (1997) for extensive discussion 41 I have no account for the fact that (85d) is slightly awkward, though infinitely better than (85c). 42 Directional PPs thus are some kind of “small clause”. This conclusion is similar to that of Hoekstra (1984), and Hoekstra and Mulder (1990), who argue on the basis of auxiliary selection that verbs taking directional PPs are unaccusative and that directional PPs are small clauses with the subject of the main V originating within it. My analysis is neutral with respect to their particular proposals. 43 The PartP can be contrastively focused, and occur in first position in root sentences, showing phrasal behavior. I have nothing to say about such cases: this is consistent with a


bare PP analysis, or a remnant movement analysis; the preposed constituent is a VP containing an incorporated P, and a trace of V. The latter analysis again raises the questions about the condition on proper binding (i.e. how the verbal trace in the preposed constituent is licensed).

44 Strong empirical support for a Kaynian analysis is presented in Hoekstra (1995). 45 Following Koopman (1996), empty projections must be licensed at a point in the derivation. This implies that some category containing overt lexical material is sitting in Spec, CP (place). Pied-piping PlaceP to Spec, CP(place) will make CP be recognizable as PlaceP, in the same way as having a wh-phrase in Spec, DP allows the DP to count as wh-phrase. 46 Precisely this fact motivated van Riemsdijk’s rule of P-shift (1978), which turns a postposition into a particle.

References Abney, S. (1987) “The English noun phrase in its sentential aspect”, unpublished MIT PhD thesis. Baker, Mark (1988) Incorporation. A Theory of Grammatical Function Changing, University of Chicago Press, Chicago. Baker, Mark and Kenneth Hale (1990) “Relativized minimality and pronoun incorporation,” Linguistic Inquiry 21: 289–297. Barbiers, Sjef (1995) The Syntax of Interpretation, Holland Institute of Generative Linguistics, Leiden. Bennis, Hans (1986) Gaps and Dummies, Foris Publications, Dordrecht. Bennis, Hans (1991) “Theoretische aspekten van Partikelvooropplaatsing”, Tabu 21: 89–95. Bennis, Hans (1992) “Long head movement: the position of particles in the verbal clus-ter in Dutch”, in R.van Hout, and R.Bok-Bennema (eds) Linguistics in the Netherlands 1992, John Benjamins, Amsterdam. Besten, Hans den and Gert Webelhuht (1990) “Stranding”, in G.Grewendorf and W. Sternefeld (eds) Scrambling and Barriers, 77–92, Academic Press, Amsterdam. Broekhuis, Hans (1998) “Adjectives and adjective phrases”, unpublished manuscript, University of Tilburg (to appear as Modern grammar of Dutch occasional working paper 2). Carstens, Vicki (1991) “The morphology and syntax of determiner phrases in KiSwahili”, unpublished PhD dissertation, UCLA. Chomsky, Noam (1991) “Some notes on economy of derivation and representation”, in R. Freidin (ed.) Principles and Parameters in Comparative Grammar, MIT Press, Cambridge, Mass. Chomsky, Noam (1995) The Minimalist Program, MIT Press, Cambridge, Mass. Cinque, G. (1997) Adverbs and Functional Projections: A Cross-linguistic Perspective, Oxford University Press, New York, forthcoming. Corver, N. (1990) “The syntax of left branch extractions”, unpublished PhD dissertation, Tilburg University. Dikken, Marcel den (1992) Particles, Holland Institute of Generative Linguistics, Leiden. Emonds, Joseph (1985) A Unified Theory of Syntactic Categories, Foris Publications, Dordrecht. Emonds, Joseph (1976) A Transformational Approach to English Syntax, Academic Press, New York. Grimshaw, Jane (1991) “Extended projections”, manuscript, Brandeis University.


Guéron, Jacqueline (1986) “Clause union and the verb-particle construction in English”, paper presented at NELS 17. Hoekstra, Teun (1984) Transitivity. Grammatical Relations in Government Binding Theory, Foris Publications, Dordrecht. Hoekstra, Teun (1988) “Small clause results”, in Lingua 74: 101–139. Hoekstra, Teun (1995) “The status van van”, Tabu 15: 160–164. Hoekstra, Teun and Rene Mulder (1990) “Unergatives as copular verbs: locational and existential predication”, The Linguistic Review 7: 1–79. Jackendoff, Ray (1973) “The basic rules for prepositional phrases”, in S.Anderson and Paul Kiparsky (eds) A Festschrift for Morris Halle, Holt, Rinehart and Winston, New York. Jackendoff, Ray (1990) Semantic Structures, MIT Press, Cambridge, Mass. Johnson, Kyle (1991) “Object positions”, Natural Language and Linguistic Theory 9: 577–636. Kayne, Richard (1985) “Principles of particle constructions”, in Guéron et al. (eds) Grammatical Representation, Foris Publications, Dordrecht. Kayne, Richard (1993) “Towards a modular theory of auxiliary selection”, Studia Linguistica 47: 3– 31. Kayne, Richard (1994) The Antisymmetry of Syntax, Linguistic Inquiry Monograph, MIT Press, Cambridge, Mass. Koopman, Hilda (1984) The Syntax of Verbs, Foris Publications, Dordrecht. Koopman, Hilda (1991) “The verb particle construction and the syntax of PPs”, unpublished manuscript, UCLA. Koopman, Hilda (1993) “The structure of Dutch PPs”, unpublished manuscript, UCLA (Chapter 8 in this volume). Koopman, Hilda (1994) “Licensing heads”, in Norbert Hornstein and David Lightfoot, Verb Movement, Cambridge University Press, Cambridge (Chapter 9 in this volume). Koopman, Hilda (1995a) “On verbs that fail to undergo V-second”, Linguistic Inquiry 26: 37–63 (Chapter 10 in this volume). Koopman, Hilda (1995b) “De plaats van geinkorporeerde hoofden in de werkwoords-kluster”, Tabu 25(4): 174–179, also published in English as Koopman (1996) “The position of incorporated heads within the verbal cluster”, in F.Lee and E.Garret Syntax at Sunset, 31–36, UCLA, Los Angeles. Koopman, Hilda (1996) “The Spec head configuration”, in E.Garrett and F.Lee (eds) Syntax at Sunset, UCLA Working Papers in Syntax and Semantics 1 (Chapter 11 in this volume). Koopman, Hilda and Dominique Sportiche (1989) “Pronouns, logical variables and logophoricity in Abe”, Linguistic Inquiry 20(4): 555–589) (Chapter 2 in this volume). Koopman, Hilda and Dominique Sportiche (1991) “The position of subjects”, Lingua 85(2/3). Koopman, H. and A. Szabolcsi (1998) Verbal Complexes, to be published by MIT Press. Koster, J. (1975) “Dutch as an SOV language”, Linguistic Analysis 1(2): 111–136. Koster, J. (1987) Domains and Dynasties, The Radical Autonomy of Syntax, Foris Publications, Dordrecht. Koster, J. (1993) “Predicate incorporation and the word order of Dutch”, unpublished manuscript, University of Groningen. Longobardi, G. (1994) “Reference and proper names: a theory of N-movement in syntax and logical form”, Linguistic Inquiry 25: 609–708. Mueller, G. (1996) Incomplete Category Fronting: a Derivational Approach to Remnant Movement in German, SfS, Tübingen. Nkemnji, Michael (1995) “Heavy pied-piping in Nweh”, unpublished PhD dissertation, UCLA. Pesetsky, D. (1995) Zero Syntax. MIT Press, Cambridge, Mass.


Pollock, Jean-Yves (1989) “Verb movement, universal grammar and the structure of IP”, Linguistic Inquiry 20: 365–424. Riemsdijk, Henk van (1978) A Case Study in Syntactic Markedness: the Binding Nature of Prepositional Phrases, Foris Publications, Dordrecht. Riemsdijk, Henk, van (1990) “Functional prepositions”, in H.Pinkster and I.Genee (eds) Unity in Diversity, Foris Publications, 229–241. Ritter, E. (1991) “Two functional categories in noun phrases: evidence from modern Hebrew”, unpublished manuscript, UQAM. Rizzi, L. (1995) “On the fine structure of the left periphery”, unpublished manuscript, Université de Genève. Sportiche, Dominique (1988) “A theory of floating quantifiers and its corollaries for constituent structures”, Linguistic Inquiry 19: 425–449. Sportiche, Dominique (1989) “Le mouvement syntaxique: contraintes et paramètres”, Languages 95: 35–80. Sportiche, Dominique (1990) “Movement, agreement, and case” (included in Sportiche 1998). Sportiche, Dominique (1995a) “Clitic Constructions”, in L.Zaring and J.Rooryck (eds) Phrase Structure and the Lexicon, Kluwer Academic Publishers, Dordrecht (included in Sportiche 1998). Sportiche, Dominique (1995b) “Sketch of a reductionist approach to syntactic variations and dependencies”, in H.Campos and P.Kempchinsky (eds) Evolution and Revolution in Linguistic Theory: Essays in honor of Carlos Otero, Georgetown University Press, Washington, D.C. (included in Sportiche 1998). Sportiche, Dominique (1998) Partitions and Atoms of Clause Structure, Routledge, London. Stowell, Tim (1991) “Small Clause Restructuring”, in R.Freidin (ed.) Principles and Parameters in Comparative Grammar, MIT Press, Cambridge. Szabolcsi, Anna (1987) “Functional categories in the noun phrase”, in I. Kenesei (ed.) Approaches to Hungarian, vol. 2, Jate Szeged, Budapest. Szabolcsi, A. (1994), “The noun phrase”, in F.Kiefer and K.E.Kiss, (eds) The Syntactic Structure of Hungarian, Syntax and Semantics 27, Academic Press, San Diego, Calif. Taraldsen, K.T. (1983) “Parametrix variation in phrase structure”, unpublished dissertation, University of Tromsø. Uriegereka, Juan (1988) “On government”, unpublished PhD dissertation, University of Connecticut. Valois, Daniel (1991) “The internal syntax of DP”, unpublished PhD dissertation, UCLA. Zwart, Jan Wouter (1993) “Dutch syntax: a minimalist approach”, unpublished PhD dissertation, University of Groningen. Zwart, Joost (1995) “Complex prepositions and P-stranding in Dutch”, unpublished manuscript, Utrecht University.

Part IV HEADS

9 LICENSING HEADS

1 Introduction The principles governing the structure of syntactic representations can be seen as a collection of licensing conditions that categories of various types have to meet. Arguments, for example, need to be licensed by the theta-role assignment, overt NPs by Case, small pro by being identified, negative polarity items by negative elements, anaphors by being appropriately bound to antecedents, expletive pronouns by being replaced by an NP at LF. Licensing conditions are stated in terms of a number of primitive features (nominal or verbal, phonetic or covert, and so on) and relations (government, ccommand, Spec-head agreement, coindexing, and such). In recent work, Sportiche (1992) has suggested that an entire set of licensing conditions could be reduced to one, namely Spec-head licensing. As illustration, consider two major cases of syntactic XP-movement: wh-movement and NP movement. They represent movement to Spec positions (Spec of CP and Spec of AGRsP, Spec of AGRoP). This type of movement represents a way to license particular XPs, either because the XP is a wh-phrase and has to be licensed in Spec of CP, or because the XP needs Case. Spec positions thus typically function as “licensing” positions, and (overt or covert) movement represents the means to fulfill the licensing requirement: a particular XP is licit because it “counts” as being in the Spec position of a particular (licensing) head. This general idea brings together a number of results from recent work that has established the privileged status of Spec positions as licensing positions. Some formulations of Case theory assume that (at least structural) Case is attributed to a property of Spec positions which are in a relation with a particular head (Chomsky (1991), Sportiche (1990)). Rizzi’s (1991) update of May’s (1985) wh-criterion states that wh-phrases are licensed under agreement with a wh-head. Moritz and Valois (1992) argue that negative quantifier phrases are licensed in Spec of NegP and derive the distribution of negative elements from movement theory (see also Zanuttini (1991) and references therein). Koopman (1991) argues that pure phi-categories (pronouns and anaphors) must be licensed in Spec positions at S-structure. This line of research is pushed furthest in Sportiche (1992), where it is proposed that Spec-head licensing is the most general


licensing configuration extending to all cases of movement, theta and case licensing, and scope relations. Head movement plays a more humble role in this general theory of licensing.1 Although it has sometimes been proposed that heads must move in order to license their Spec (V-movement was forced in this way in Koopman (1984) for instance), head movement is most often seen as a way to fulfill some kind of morphological licensing, driven by the status of some head as a bound affix or not, or by the need for some stem to get an affix. Head movement in this view is quite accidental depending on languageparticular morphological properties. This chapter argues for a much more prominent role of head movement. It represents the way through which a more fundamental licensing requirement is satisfied. Although I subscribe to the reductionist program outlined in Sportiche (1992), which seeks to reduce licensing to Spec-head licensing, I shall argue that Spec-head licensing does not represent the only licensing mechanism: it might be the only licensing relation for XPs, mediated by XP movement. Certain head positions will be argued to provide the licensing positions for other heads. Just as an XP must be licensed in a Spec position to which it is allowed to move given movement theory, thus accounting for the distribution of these XPs, so must a head X be licensed in another head position accessible to it by head movement, accounting in this way for the distribution of XP. This leads to the following picture: (1)

a. b.

Spec positions are licensing positions for XPs Head positions are licensing positions for X

where (1a) and (1b) constitute the (only) licensing mechanisms, and movement represents the only way to satisfy them. I will show that head movement is a driving force behind phrase structure. The focus of this chapter is on V-movement in particular and on head movement more generally. There are two types of V-movement: V-to-I-to-C and V-to-V movement, also referred to as complex verb formation, V-incorporation or V-raising (in the Dutch tradition). The questions that I will attempt to answer are why V-movement exists, at which level V-movement must occur, why languages vary with to respect to V-movement in the syntax and what the reason for head movement in general is. In order to address these questions, I first spell out my assumptions about head movement (section 2). In section 3, I turn to the distribution of VPs in English, and to the question of how their distribution should be accounted for. In 3.1, I argue that a V headed by T and C (a “clausal” VP) is licensed through head movement. I motivate a particular proposal according to which head movement is forced to satisfy the inherent lexical properties of V, and discuss in detail how this licensing is achieved. 3.2 discusses the distribution of bare VP complements in English, which will be shown to fall out from constraints on head movement (this generalizes a proposal by Stowell (1991a) for the distribution of complement small clause APs). This leads to the conclusion that the head of a predicate must be licensed through incorporation (see also Uriegereka (1988)). Section 4 addresses the question at which level the verbal head must incorporate in English (Structure or LF), and reaches the surprising result that while the head of a VP

LICENSING HEADS 259

complement of T and C incorporates at LF, the head of a bare VP complement must incorporate at S-structure. Discussion of verb raising in the Dutch causative construction brings support for this conclusion. I will suggest that obligatory S-structure incorporation follows from the basically defective nature of verbs selecting bare VPs. Section 5 considers the head that serves as the host of the incorporation. Not all heads will be shown to count as licensing heads for a particular X. This is parallel to the fact that the licensing nature of Spec positions depends on the nature of the head; a wh-XP is only licensed in the Spec of an XP whose head bears the feature [+wh], a negative XP only in Spec of NegP. Whether a head is a licensing head or not yields a new account for the wellknown differences between noun complementation and verb complementation. This account allows us to dispense with the Uniformity Condition of Chomsky (1986). Section 6 generalizes the proposal that V-movement represents the way in which lexical properties are satisfied to head movement in general: head movement is the means by which all lexical properties are satisfied. Lexical items resemble molecules, with a number of positions or receptors (“lexical properties”) which must be bound (“satisfied”). The receptors are bound by a head that attaches to them. The mechanism which allows a head to attach to a receptor is head movement. (This proposal represents a particular implementation of what Stowell (1981) called the coindexation of the elements of the theta-grid with the particular arguments.) This proposal has far-reaching consequences: it derives the fact that theta-roles are assigned under government from the way head movement operates and it accounts for the binary branching nature of phrase structure. 2 Head movement Head movement is strictly local (Koopman (1984), Travis (1984) and Baker (1988)). This follows from the fact that traces must be governed to satisfy the ECP. Government, following arguments in Sportiche (1988,1990) and Koopman and Sportiche (1991), is always defined in terms of i-command (A i-commands B if a sister of A contains B). (2a), (2b) and (2c) satisfy government and represent a possible head movement configuration, but (2d) and(2e) do not:

I adopt Sportiche’s (1990) definition of barrierhood, according to which Xmax is never a barrier for Xmax: (3)

Xmax is a barrier for B if any non-L-marked projection of X includes B.


Since V1 is the head of V1P, V1P is not a barrier for movement of V1. V2, however, is dominated by V1P; V2 is not the head of V1P; V1P therefore is a barrier for V2. Dutch verbraising constructions illustrate the configuration in (4), but, because of Dutch word order, the VPs represent the mirror image of (4). A causative verb like laten “let” selects a bare VP complement, and triggers verb raising: V2 adjoins to the causative laten (V1). Laten subsequently excorporates, as can be concluded from the morphology (laten is inflected for past tense in (5a)) and from the fact that it can only be laten that undergoes Vto-C movement (verb second).2 This implies that the head of an X category can be extracted from Xmax if Xmax is Lmarked. It follows from (3) that the head of a projection can be extracted, even if some other head has adjoined to it. A head can excorporate in the following configuration, where V1 represents a verb selecting a bare VP complement (V2P), and where V2 adjoins to V1:

LICENSING HEADS 261

A head can therefore be both a target of incorporation, and a target of excorporation (compare Roberts (1991) for a different view of excorporation). Excorporation will be extensively used in this chapter. (4) restricts the possibilities of head movement either to moving the entire X complex or to moving the head of the X complex. Successive cyclic (adjunction) head movement therefore cannot exist. This implies that clitic-climbing cannot be analyzed as successive cyclic head movement (as in Kayne (1989)). It must rather be analyzed as involving movement of XP and possibly subsequent head movement (compare Sportiche (1988, 1990 and 1992) for arguments to this effect). Long distance anaphora cannot be analyzed as successive cyclic head movement, as in Lebeaux (1983) and Cole, Hermon and Sung (1990) among others. The cases analyzed in Lema and Rivero (1990) and Roberts (1991) as involving long head movement, and the cases of excorporation of adjoined elements discussed in Roberts (1991) must be treated differently as well (compare Koopman (1991) for an analysis which reduces these cases to local head movement and excorporation). Head movement further requires that the landing site be an appropriate host for the incorporee, call it a licensing head (compare section 5 for more discussion). Two conditions on head movement must thus be satisfied. First, the target of incorporation, i.e. the incorporee, must be governed by the host (6a), because head movement requires government. And second, the host must be a licensing head for the incorporee (6b): (6)

a. b.

The incorporee must be governed by the host The host must be a licensing head 3 The distribution of VPs

What is the distribution of VP and how should it be accounted for? VPs occur as complement of a string of functional categories, the highest of which is C, and as complement of a restricted number of verbs. Specific accounts for the distribution of VPs have been proposed by several authors (Fabb (1983), Zagona (1988 a,b), and Guéron and Hoekstra (1988) among others). Zagona (1988b) and Fabb (1983) argue that the distribution of VPs follows from Case theory, in particular from the requirement that VPs (or Vs) need Case: since Case is assigned under adjacency, VPs will be adjacent to the Case assigning head,


either I or V. This proposal cannot be maintained, as Zagona (1988b) shows. VPs do not have the distribution of Case-marked NPs. VPs cannot occur in Case positions, overt Case-assigners comparable to of-insertion cannot be inserted and VPs cannot undergo rightward movement comparable to heavy-NP shift. Instead, Zagona argues, the distribution of VPs resembles that of theta-marked elements, and should therefore be accounted for in terms of theta-theory. She develops an account by which VPs need to be assigned temporal theta-roles, which they receive from T. Guéron and Hoekstra (1988) follow Zagona (1988b) and Evers (1982): they assume that Vs need to be T(ense)marked, and develop the notion of T-chain, an S-structure concept. In their account, certain heads (auxiliaries) must be assumed to be transparent for T-marking of V. The analysis developed below shares aspects of Zagona’s theta-marking by T, or Evers’ and Hoekstra and Guéron’s T-marking. It differs from these accounts by assuming that verbs can only satisfy T-marking through head movement, and that they do so, depending on language, at S-structure or at LF. In 3.1, I first develop an account of licensing of VP complements by T/C. This will give the necessary background for discussion of bare VP complements in 3.2. 3.1 V to I to C VPs occur as complements of a chain of functional categories (which will all be referred to as Is), the highest of which is C in independent (or root) clauses: (7)

[ C [ I1 [ I2 [ I3 […[V*P V*[VP…]]]]]]]

The structure in (7) assumes layered VPs (Larson (1988)), combined with Sportiche’s (1990) proposal for the particular way in which layered VPs are structured. Internal arguments are assigned in VP; the external argument is assigned in a VP-shell, V*P, with V moving to V*. V-to-V* movement is forced by the necessity to assign a theta-role to the external argument (VP shells will be shown to follow from incorporation in section 6).3 Languages differ with respect to the S-structure position V occupies. The task of the language learner basically is to figure out how high the V moves up in the I/C chain. In languages like English, it is standardly assumed that V does not move to an I category at all (Emonds (1976) and Pollock (1989)).4 In French, the V moves up higher than in English, but not as high as in Vata or Italian, where the V seems to move to the highest I category.5 In V-second languages, the V moves up even higher to C if C is ungoverned (i.e. in matrix clauses), although in non-root contexts English-type behavior is found (Norwegian and Swedish) as well as Romance-type behavior (Icelandic and Yiddish).6 Although languages vary at S-structure, it is standardly assumed that languages are invariant at LF (compare Huang (1982), Chomsky (1991), Stowell (1983), May (1977). I assume with Stowell (1983) and others that the verb moves to C at LF in all languages. Language variation with respect to the position of the verb is thus purely an S-structure phenomenon.

LICENSING HEADS 263

What forces V-to-C movement at LF? Stowell (1983), following den Besten (1983), proposes that T is located in C and that Vs must be associated with T, i.e. T-marked. Verbs must move to C because they need to be T-marked. While I adopt the basics of Stowell’s idea, I assume that T is located under IP and that the C projection is required to license TP: e.g. TPs need to be licensed further, while CPs, at least matrix CPs, do not. I will refer to this property of C as “the ability to close a predicate,” and return in section 6 to a possible account of this property of C: (8)

a. b.

T is located in I C closes the predicate

Why do Vs need to be T-marked? Suppose that this follows from inherent lexical properties of the V. Verbs are not only specified for a theta-grid, yielding a classification of Vs in terms of intransitive, unaccusative, transitive verbs, but also for a temporal and aspectual grid (compare Higginbotham (1985)), yielding basic verb classes: statives, activities, achievements, accomplishments.7 For reasons of simplicity, I collapse the temporal and aspectual grids and refer to them as the temporal grid.8 (9)

Vs have an argument-grid (theta-grid) and a “temporal” grid (T-grid)

The T-grid consists of a number of slots or “receptors,” depending on the verb class. Each of these receptors must be bound (i.e. licensed), because lexical properties must be satisfied. This forces the appropriate licensers to be projected in the syntax as I heads. Thus, just as the VP structure depends on the theta grid of a particular verb, the I structure depends on the T-grid of a verb. How exactly is licensing of the T-grid achieved? Each receptor needs to be bound by a designated head. Suppose that a receptor is licensed when the appropriate head attaches to the relevant receptor, i.e. binding reduces to head movement. This will force verb movement to a position from which the designated head can incorporate to the receptor. As illustration, consider a particular derivation of a verb with a T-grid containing an aspectual slot ASP and a tense slot T. The particular lexical representation of this V forces the presence of both an ASP head and a T head, which will enter into X-bar theory, and project an ASPP and TP respectively. Since ASP appears to be syntactically lower than T, the underlying syntactic structure is as in (10) (irrelevant details aside). The T-grid of V is represented by double bars attached to the core V:


Adjunction of V to ASP yields the structure in (11):

In this representation, ASP is not attached to the ASP-receptor of the V: thus, the ASPreceptor is not licensed. Let us assume that licensing is achieved by further movement of the ASP head onto the ASP-receptor:

LICENSING HEADS 265

This movement yields a violation of the head movement constraint given the standard assumption that ASP is the head of the complex in (11). I will assume, however, that there is possible ambiguity in this structure about what counts as the head ASP or V. More particularly, I will assume that ASP can incorporate onto the verbal receptor ASP, precisely because V can count as the head of the newly formed complex as in (12), where ASP is sister of V and governed by V. V can therefore attach to the ASP receptor and will be licensed as a trace.

Consider next the licensing of the T-receptor of V. V will excorporate from the structure in (12) (this is allowed because V is the head of the newly formed complex) and adjoins to T. T will subsequently move onto the T-receptor of the verb.9 This step will again force V to count as the head of the verbal complex: The verbal receptors are thus eventually filled with heads, and this is achieved through head movement. The verbal “molecule” heads the clause, with all its receptors filled, and all lower heads are licensed as traces. (8b) further forces V to move to C (see section 8 for a possible motivation of this movement).


Satisfying lexical properties requires at least two local steps of head movement: (14)

Receptor-binding requires two steps: a. Adjunction of the lexical item to the head b. Head movement of the licenser onto the receptor of the lexical item

(14) will play an important part in this chapter. If (14) is true, we expect to observe both types of incorporation at S-structure. I believe that this is correct; (14a) gives rise to the adjunction type of head movement involved in complex verb formation, (14b) to the substitution type of V-movement involved in V-to-C. Furthermore, the two steps in (14) can describe the crosslinguistic variation in the position of V at S-structure. As pointed out above, the analysis forced me to make the non-standard assumption that a functional category can only move onto the receptor of V if V is the head of that newly formed complex. However, this is only necessary at the point where (14b) must apply, which I assume to be at LF. Suppose that only (14a) applies at S-structure in a particular language. In this case there is no reason not to assume that ASP is the head of the derived complex. But if ASP is the head, the V cannot excorporate to any higher position, precisely because V is not the head. The result then is that V cannot move on. However, if both (14a) and (14b) occur at S-structure, V must be the head (as in (12)). But now, since V is the head of the projection, it can excorporate and move on to higher I projections. This opens up a new way to account for crosslinguistic variation with respect to the S-structure position of the head: only if both steps have occurred can the lexical head excorporate to a higher projection. If only (14b) takes place at S-structure, the verb will not be able to do so. As a final remark, it is easy to see why the I nodes must be arranged in a head-complement relation: each V must undergo head movement to some I, so that this I in turn can incorporate onto the receptor of V.

LICENSING HEADS 267

3.2 Bare VP complements We are now ready to address the question how bare VPs which are not dominated by C are licensed. The distribution of bare VPs is extremely restricted (Zagona (1988b)): (i) The VP complement contains no overt I if the governing verb is active. (15)

a. I made Mary visit her mother b. I let Marie go home (ii) Bare VPs cannot occur in adjunct positions: (16)

a.

*I finished my work rapidly [PRO go home] (I finished my work rapidly [(in order) to go home]) b. *I listened to the radio [PRO drive home] (I listened to the radio driving home)10 c. *This doctor examines his patients [PRO be nude] d. This doctor examines his patients [PRO nude] (iii) Bare VPs are excluded from subject position (Spec of IP): (17)

a. * [Mary visit her mother] bothers me b. *[PRO visit one’s mother] is nice (iv) Bare VP complements cannot undergo “rightward” movement: (18)

*Mary made John yesterday leave the country

Bare VP complements are thus restricted to the complement domain of V, i.e. they must be governed by V at S-structure:11 (19)

Bare VPs must be governed by V

(19) exactly represents the configuration allowing for head government. We can therefore assume that (19) holds precisely because head movement must occur, just as Baker (1988) proposed for languages with overt morphological causatives.12 (20)

The head of a bare VP must incorporate

This raises the question why head movement must occur. The head of a bare VP is not dominated by T/C. Given the arguments in section 3.1, the predicate is not closed by a C head. Let us assume therefore that incorporation represents a way to license the predicate, and put off the details to section 4.5. (21)

The head of a bare VP complement is licensed by incorporating to V


A further question arises: at which level must head movement occur? At LF, as Stowell (1991a) argues for AP complement clauses, or at S-structure? I return to the latter question in section 4. It is easy to see how (21) accounts for the distribution of bare VP complements. Bare VPs can occur in complement position because head movement can proceed from this position. Bare VPs are excluded from adjunct position. Because adjuncts are not L-marked, the head of a VP adjunct cannot be licensed by incorporating into a governing head.13 Bare VPs are excluded from subject position because of the lack of government between the licensing head V and the head that must incorporate. Bare VPs cannot undergo rightward movement, because of the absence of S-structure government.14 From the discussion above, it also follows that V is a licensing head for the head of a bare VP: (22)

V is a licensing head for the head of a bare VP

We return to (22) in section 5. 4 S-structure or LF incorporation Since the head of a bare VP must be licensed by head movement, its distribution is restricted to the complement position of V. Since V moves to T at LF in English, not at Sstructure, the conclusion that V-to-V incorporation should also occur at LF, not at Sstructure, seems straightforward. In this section, I argue that English V-to-V incorporation does indeed happen at LF. However, I will also present evidence that causative verbs must trigger incorporation at Sstructure, even in English: they trigger S-structure incorporation of some I that is contained in the bare VP complement. Incorporation at LF requires government between the licensing head and the incorporating head. Government, however, does not need to hold at S-structure. In particular, the XP can occur in an S-structure position which allows for reconstruction at LF.15 I assume, without going into the details here, that Spec positions allow for reconstruction16 but adjunction sites do not. Rightward adjunction rules do not seem to allow for reconstruction, and rightward VP movement as in (18) is therefore excluded. S-structure incorporation, however, requires government at S-structure, because the trace of the incorporated head must be antecedent-governed. Any XP with a trace in head position will therefore have a rigidly fixed S-structure position, and movement rules will not be able to apply to them. The following diagnostic criteria therefore distinguish between S-structure or LF incorporation: (23) If X incorporates to Y at LF, syntactic movement of XP to a Spec position is allowed (24) If X incorporates to Y in the syntax, XP must be governed by Y at S-structure.

LICENSING HEADS 269

English VPs can undergo VP preposing, stranding T at S-structure. (25)

[visit her doctor every week] she did

VP-preposing is possible precisely because English V-to-T movement occurs at LF.17 If Vto-T movement occurs at LF, the null hypothesis is that V-to-V incorporation also occurs at LF. Bare VP complements should therefore be able to undergo VP-preposing. Yet they cannot: (26)

*[Visit her doctor every week] I made Mary (to)

I will argue that (26) is impossible because of (24): S-structure incorporation has occurred. In order to make this point, though, discussion of VP-preposing is necessary, and alternative explanations for (26) must be shown to be inadequate or insufficient. 4.1 VP-preposing The presence of more inflectional projections in IP raises the question which projection undergoes the process traditionally known as VP-preposing: VP, ASPP, or yet another projection? VP-preposing affects a constituent lower than T, Neg, or ASP, since these elements must be stranded: (27)

[Writing a letter] she may not have been [– –]

This restricts the projection to either VP, V*P, or to some function projection immediately dominating VP, but lower than any of the projections in (27). The lower VP shell cannot be preposed, because it does not contain the V at S-structure: as assumed above, V always raises V*. This leaves two possibilities: VP-preposing is preposing of V*P, or preposing of some functional projection dominating V*P: call this projection I*P. Because the temporal grid of V needs to be licensed at LF, the V*P or I*P will reconstruct, and head movement of V to T will subsequently occur.18 Bare VP complements fail to undergo VP-preposing: (28)

a. b.

*[Mary visit her mother every week] I made *[Visit her mother every week] I made Mary

(28a) can be straightforwardly excluded as a Case filter violation and will not be discussed any further. To exclude (28b), it must be decided what the projection of bare VPs is. This is not an easy task and basically all possibilities ranging from bare VPs to CPs have been


Suppose that bare VPs were V*Ps. Under this assumption, (28b) could be accounted for if Mary is contained within this V*P: the NP could not have been stranded. This analysis runs into problems with AP-preposing: AP-preposing in the same environment is perfect: (29)

How angry did you make Mary?

If make takes a V*P in (28), make should also take an A*P in (29) since the number of shells is determined by the lexical property of the lexical item. If V must move to V* for external theta-role assignment reasons, A must move to A* for the same reason. If APpreposing affects A*P, NP should not be able to be stranded. The following variant of this idea would work. Suppose that VP-preposing is I*P-preposing, and that causative verbs take I*P complements. (30)

a. b. c. d. e.

I made [I*P Mary [I*] [V*P]e [visit her mother]] I*P-Preposing → *Mary visit her mother you made *Visit her mother I made Mary I made [I*P Mary [I*] [AP] e [angry e]] AP-preposing → How angry did you make [Mary [I* [APe]]]

The paradigm now follows. I*P-preposing is excluded, because the NP Mary must piedpipe, and fails to get Case (30b).19 The NP cannot be stranded because it is part of the projection that undergoes movement (40c). AP-preposing on the other hand is possible, because the subject of the AP has raised out to Spec of I*P; the AP can therefore be preposed, leaving its subject stranded. There are several problems with this analysis. First, it is based on the fact that the NP is contained in the I*P and has not raised to some higher Spec position. Indeed, if the NP could raise outside I*P, the result should be grammatical, just as raising out of the AP to Spec of IP is in (30e). Second, it forces a treatment of all causative complements as equal to or smaller than I*P, which is almost certainly incorrect. 4.2 Causatives trigger S-structure incorporation (24) offers a different explanation for the paradigm in (28). Suppose that causative verbs indeed take I*P complements (or even big complements), and that VP-preposing is I*Ppreposing. Suppose moreover that I*P cannot be preposed, because its head incorporates at S-structure. The underlying structure then is as in (31): (31)

[V*P(1) Spec V* [VP Spec [V, make V [I*P I* V*P(2)]]]]

LICENSING HEADS 271

The causative verb triggers S-structure incorporation of the head of the I*P, I*. I*P therefore cannot undergo I*P-preposing, because the trace in the head position would not be antecedent-governed. There are two ways to work out this idea, depending on whether V moves to I* or not. If V moves to I*, English would literally have V-incorporation at S-structure. If V moves to I* at LF, as I assume below, English has I*-incorporation at S-structure, but Vincorporation at LF. Suppose first that V moves to I* and that English has literal V-incorporation at Sstructure. Verbs like make take an I*P complement, and the embedded V moves to I*. From I*, V incorporates into make, and make in turn excorporates to V *, the position from which accusative Case is assigned to Spec of VP, and the external theta-role to Spec of V*P:

(32)

a.

[V*P Spec [V*make]k [VP [Spec Maryi [visitj ek]k [I*P [ [ [ [NP e] [*I [e]j [VP ej]]]]]]]

Suppose therefore that V does not move to I* in the syntax in English, and that it is I* which incorporates.21 Since V cannot raise to I*, it appears as if incorporation has not taken place, and V2 acts in all respects as if occurring in the usual V-position: I*P of course can still not be preposed, because the trace in the head position would fail to be antecedent-governed at S-structure. There are problems with this analysis. First, the causative verb must excorporate within the thematic projection of the causative. This raises the question why excorporation does not appear to be forced in the same way in other languages (compare Dutch, Italian, French), where the full verbal complex seems to proposed in the literature. Moreover, in order to account for differences between different types of bare VP complements, it has been argued that bare VPs can actually represent different projections, depending on the verb selecting for the bare VP complement. normal sentence, and it assigns exactly the same cases as a verb heading any VP. And third, this solution is rigid in the sense that it disallows treating some causative complements as bigger units like IPs or even TPs with empty heads. This is because these projections


move to V* yielding strings like make sleep John.20 Second, V-movement in this configuration often affects the Case-assigning properties of the embedded verb. In English, however, the verb behaves as if it occupies the same position as it does in a

LICENSING HEADS 273

contain heads which the English V cannot reach: V-to-V movement should therefore be blocked.

This is the solution that I will adopt. At LF, the V moves through the available head positions to the position where I* is at S-structure. (I return below to the licensing of the T-grid of the embedded verb in (32)b.) So far, then, V-to-T movement in English occurs at LF when T is governed by C, because preposing of I*P (VP-preposing) at S-structure is possible: I* is therefore not a trace at S-structure. But I have also argued that I*-to-V movement occurs at S-structure, when I* is governed by V, because preposing of I*P cannot occur at S-structure: I* therefore is a trace at S-structure. This yields the picture at (33) opposite on page.


(33) seems to cause serious problems for the analysis. Why should it hold? What forces Sstructure incorporation with causatives? Why should it be the licenser that determines if incorporation can occur at LF or at S-structure? Before suggesting an explanation for (33b) within the general framework of licensing developed here, let me first present additional evidence for (33b) based on Dutch causatives. 4.3 The causative in Dutch In Dutch verb raising constructions, verb raising is generally optional in the following context: (34)

In a string V1V2 where V2 governs V1,V1 raises optionally to V2

V-raising becomes obligatory when more than two verbs are involved. (35)

b.

a.

dat Jan gekomen is (no V-raising) that John come is “John came”

dat Jan is gekomen (V-raising) that John is come “John came”

LICENSING HEADS 275

(36)

a.

b.

c.

dat Marie haar moeder bezoeken wou (no V-raising) that Mary her mother visit wanted “that Mary wanted to visit her mother” dat Marie haar moeder wou bezoeken (V-raising) that Mary her mother wanted visit “that Mary wanted to visit her mother” dat Marie haar moeder heeft willen bezoeken (obligatory V-raising) that Mary her mother has want visit “that Mary wanted to visit her mother”

In the (a) examples above, the embedded verb precedes the auxiliary or the modal: Vraising has not taken place. In the (b) examples, the embedded verb follows the inflected verb as a result of V-raising. (36c) illustrates obligatory V-raising. Dutch also has a process of I*P-preposing. I*P-preposing should be possible when V has not raised higher than I*, i.e. when V-raising is optional (compare also Koster (1987)). This is correct, as the following examples show. (37)

a.

b.

[naar huis gekomen] is Jan niet to home come is John not “come home, John did not” [haar moeder bezoeken] wou Marie niet her mother visit wanted Mary not “visit her mother Mary didn’t want to”

Given the discussion above on incorporation at S-structure or LF, I*P should not be possible when V-raising must apply (either V-to-C, or V-to-V), since the I*P would contain the trace of the incorporated verb in the head position. This is borne out, as the examples in (38) show: (38)

(39)

a.

a.

b.

(40)

a.

*[iemand koekjes e] gaf ik niet (main V to C) someone cookies gave I not dat Jan niet naar huis heeft willen gaan that John not to house has want go “that John has not wanted to go home” *[naar huis gaan] heeft Jan niet willen to house go has John not want dat Marie haar moeder heeft willen bezoeken (obligatory V-raising: *bezoeken willen heeft) that Mary her mother has want visit

(obligatory V-raising: *gaan willen heeft)


b.

“that Mary wanted to visit her mother” *haar moeder bezoeken heeft Marie vast wel willen her mother visit has Mary certainly want

With this background, let us turn to the Dutch causative verb, laten which has both a permissive and causative meaning (Dutch does not truly have a causative verb corresponding to make). In the configuration in (34), laten acts differently from other Vraising verbs, in the sense that V-raising generally seems to be preferred: (41)

a.

b.

omdat ik de kinderen boeken laat lezen (V-raising) because I the children books let read “because I make/let the children read books” ? omdat ik de kinderen zulke boeken best lezen laat (no V-raising) because I the children such books read let “because I have no problem letting the children read such books”

Interestingly, however, (41b) only seems to allow the permissive reading, not the causative one. This suggests that causative laten obligatorily triggers V-incorporation, but permissive laten does not. How does the complement of laten fare under I*P-preposing? As the examples in (42) show, the complement of laten can be preposed. However, in this case only the permissive reading is available and the causative reading is excluded. (42)

a.

b. c.

de auto wassen laat ik Jan nooit (only permissive/*causative) the car wash let I John never *wash the car I never let John *zulke boeken door de kinderen lezen laat ik nooit22 such books by the children read let I never het huis shoonmaken laat Marie hem iedere week the house clean let/*makes Mary him every week “*clean house Mary lets him every week”

The unavailability of the causative reading in (41b) and (42) follows if causative laten obligatorily triggers S-structure incorporation of I*. Since the V moves to I*, V-raising is obligatory with causative laten. We thus arrive at a somewhat surprising result that Dutch causative laten obligatorily triggers S-structure incorporation, but permissive laten does not: (43)

a. b.

Causative laten must incorporate I* at S-structure Permissive laten may incorporate I* at S-structure

(43a) accounts for the obligatoriness of V-raising at S-structure (41a), and the unavailability of the causative reading under I*P-preposing illustrated in (42). The

LICENSING HEADS 277

examples also lead to the conclusion that permissive laten does not need to incorporate I* at S-structure. Other verbs selecting for bare VP complements, like perception verbs, act like permissive laten. Thus S-structure V-raising is optional in the configuration in (34), and I*P-preposing is possible: (44)

a.

b.

dat ik Marie de auto zag wassen/wassen zag that I Mary the car saw wash/wash saw “that I saw Mary wash the car” de auto wassen zag ik Marie nooit the car wash saw I Mary never “*Wash the car I never saw Mary”

A similar conclusion holds for French. While French has both a causative faire and (permissive) laisser, only causative faire triggers obligatory V’-preposing (V-fronting). Laisser (and perception verbs) do so optionally: (45)

a.

b.

J’ai fait/laissé manger la soupe à Jean I have made let eat the soup to John “I made John eat the soup” J’ai *fait/laissé Jean manger la soupe I have made/let John eat the soup

Suppose that the French verbal complex results from V-to-I* movement and I* adjunction to faire followed by excorporation of faire (Bok-Bennema (1991)). (45a) then shows that faire must incorporate I* at S-structure, and that laisser may incorporate I* at S-structure. In conclusion, the Dutch causative verb laten obligatorily triggers S-structure incorporation, and so does the French causative faire. This property of Dutch can surely not be a learned property given the high degree of opaqueness of the data. If it cannot be learned, it must be universal. We thus get strong support for the proposed analysis of English in (33b), at least as far as causative verbs are concerned; English causative verbs must trigger S-structure incorporation. However, the discussion on Dutch and French also shows that other verbs selecting bare VPs do not need to trigger S-structure incorporation. This implies that failure of preposing a bare VP complement cannot always be due to the fact that the selecting verb triggers S-structure incorporation. The question thus arises again what excludes I*P-preposing in the following cases: (46)

a. b.

*play handball I see her every day *wash the car I never let John

These examples could be excluded if the stranded NP is contained in the I*P, as suggested in section 4.1, and has not raised to the matrix object position. The Dutch equivalent of (46) is grammatical (as shown in (45)), because Dutch accusative NPs scramble to a position quite high in the matrix clause, whereas English NPs do not.


It is well known that French and the other Romance languages do not have I*Ppreposing. This can in fact be explained if the verb in Romance languages must always move to a projection higher than I*: I*P-preposing cannot exist, because there will always be a trace in head position of I*. 4.4 Licensing of bare VPs In section 4.1, the V of a bare VP complement was licensed by incorporating to the selecting V at LF (21). How is this licensing formally achieved? And why must causative verbs obligatorily trigger S-structure incorporation? Why does S-structure incorporation versus LF incorporation depend on the licenser? I will try to provide answers to these questions here. Consider the following underlying representation (where V2 has both an ASP and a T receptor):

At LF, the embedded V, V2, moves through the intermediate I projections, and adjoins to V1, yielding (48):

LICENSING HEADS 279

The T-grid of V2 is licensed by incorporating the relevant heads onto the lexical receptors. The predicate is not “closed” by C, however. I assume that adjunction of V2 to V1 plays the same role in this respect as V moving to C. The matrix V will excorporate and move on to satisfy its own T-grid receptors. But why must causative verbs trigger S-structure incorporation?23 Suppose that causative verbs have a “defective” temporal argument grid, and that S-structure incorporation represents the means to “repair” this defective grid. Suppose more concretely that causative make lacks the I* receptor and that the causative verb is unable to project a well-formed syntactic structure precisely for this reason: it needs an I* receptor to project a well-formed syntactic representation. Because it needs it for syntactic purposes, it needs to get one in the syntax, and the way this is achieved is by raising the I* from its complement in the syntax:


This kind of account could carry over to all verbs selecting for a bare VP. Suppose that these verbs have a defective temporal argument grid; the defect could affect different types of receptors. Suppose, for example, that a particular verb lacks a T receptor. This forces incorporation of T from a complement, basically leading to T-raising, a process Stowell (1991a) argues applies to the complement of perception verbs. Under this view, verbs take bare VP complements precisely because they have to incorporate some I node from their complement: i.e. superficially these complements are bare VPs because the Is are traces. It would follow that it is the deficiency of the T-grid of the verb that determines the projection of the complement clause. This analysis would leave the desired flexibility with respect to the categorical status of the different complements: so-called bare VP complements could represent different categorical projections, ultimately depending on the kind of governing causative verb. This type of account should yield an account for why in language after language the same class of Vs takes this type of complement. The proposal in this chapter that causative verbs trigger obligatory S-structure incorporation, but perception verbs, for example, do not, should also yield an account for why causatives typically show up as bound morphemes (triggering obligatory S-structure incorporation), but perception verbs do not. Finally, Sportiche (personal communication) suggests that auxiliary switch in Italian for instance could be treated in a similar fashion: the restructuring verb incorporates some aspectual feature of the lower verb in the syntax, and this is shown by the fact that auxiliary selection is sensitive to the aspectual properties of the lower verb. 4.5 Language variation Since both Dutch and English causatives trigger obligatory S-structure incorporation, Dutch and English are more similar than previously assumed. The difference between the languages reduces to an independent one: the S-structure position of V. In English, the V cannot reach the I*position governed by the licensing V, therefore there are no overt signs of incorporation at S-structure (other than, maybe, the empty I*s). In Dutch, V moves to I (and C); therefore syntactic incorporation can take place and complex verbs or verbal complexes can show up at S-structure. Since causatives universally trigger S-structure incorporation, and since the projection of the complement is driven by the lexical properties of the causative, causative constructions provide crucial information as to whether the language allows V-to-I* movement or not. This is a welcome result. Since languages vary with respect to the Sstructure position that the verb occupies, one of the language learner’s tasks is to find out what this position is. Arguments that allow one to establish this in a particular language often appear difficult to construct.24 This appears to be particularly true for head-final languages. However, if the analysis proposed in this chapter is correct, the form of causative constructions becomes extremely important: if causatives involve overt syntactic V-incorporation, the language must have syntactic V-to-I* movement. Only when V can move to I* can incorporation be overtly seen. If the causative does not overtly show V-

LICENSING HEADS 281

incorporation, it follows that the language does not allow for V-to-I* movement. This leads us to the following generalization: (50)

Overt V-to-V incorporation iff syntactic V-to-I* movement

Similarly, if perception verbs involve syntactic T-incorporation, and if the language has visible V-incorporation with perception verbs, it follows that the language has syntactic Vto-I*-to-T-movement. If the language has optional incorporation with perception verbs, it has optional syntactic V-to-I*-to-T movement. (50) holds more generally: syntactic incorporation (X-movement to a Y head) can only take place if the head can move up to a position governed by the host. Assuming that a lexical category X is governed by a functional one F, it follows that syntactic X movement to Y is only possible if syntactic X-to-F movement is allowed. That is, N can only move out of DP if N to the head position of D is possible in the syntax. P can only incorporate if P can move to the functional projection dominating P in the syntax (compare Koopman (1991)). 5 Licensing heads: N/V asymmetries We have presented ample motivation for an approach under which a head must be licensed through incorporation. Let us now turn our attention to what can count as a licensing head. The discussion of bare VP complements in 3.2 led to (22): (22)

V is a licensing head for the head of a bare VP

An examination of those contexts from which bare VP complements are excluded shows that there are restrictions on what element can count as a licensing head. Bare VP complements are excluded from the complement position of verb-particle combinations (Koopman 1991): (51)

*John made Mary out visit her mother

And bare VP complements are excluded from the complement position of N or P: (52)

a. b. c.

*the vision [of John leave] *my making [of Mary visit her mother every week] *without [Mary see Bill]

What excludes VP complements from the complement position of verb particles? Kayne (1985) shows that the object of a verb-particle construction is not a direct argument of the verb, but rather forms a projection with the particle. I assume, more specifically, that verb-particle constructions have the structure in (53), with the complement the verb-


particle combination originating in the complement position of P (see also den Dikken (1990), Koopman (1991), Taraldsen (1983)): V[PP[P [XP]]

(53)

As (51) shows, the XP cannot correspond to a bare VP. The V of the bare VP must undergo head movement to V, but will never be able reach the licensing head V, however, because it can move no further than the governing P. If it is assumed that P is not a licensing head for V, (51) is explained: (22)

a. b.

V is a licensing head for the head of a VP complement P is not a licensing head for the head of a VP complement

(22b) also accounts for the ungrammaticality of (52c). It immediately suggests a way to account for (52a) and (52b) as well: (22)

c.

N is not a licensing head for the head of a VP complement

(22) suggests a new account for the well known fact that N and V do not allow for the same range of complements. This asymmetry is generally attributed to a failure of the head N to license an NP with which it has no thematic relation. Kayne (1984), for example, proposes that Ns have different government properties from Vs: verbs can, but nouns cannot, govern across a small clause boundary. Chomsky (1986) assumes that Ns and Vs have different Case-assigning properties: Ns assign inherent Case, which is subject to the Uniformity Condition on Case assignment, restricting assignment of inherent Case to the dependent elements. Verbs assign structural Case which is not subject to this restriction. The discussion above suggests a different account, however: it is the head of the predicate that fails to be licensed, because N is not a licensing head. Consider the paradigm that needs to be accounted for: (54)

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

*Johni’s belief [ei to be crazy] *The belief of John to be crazy *Johni’s likelihood [ei to be a fool] *The consideration of John a fool *The consideration of John foolish *The call of John up *My gift of Mary a book *My claim John is sick *His taking of advantage of Bill

In (54a)-(54c) the predicates are not headed by C. They are therefore not independently licensed and the V must incorporate to a licensing head, V. The examples are ungrammatical since there is no V available in this structure, and N is not a licensing head.25

LICENSING HEADS 283

AP predicates need to be licensed by incorporating to V and, if we assume the same holds for NP predicates, (54d) and (54e) are excluded because N is not a licensing head. (54f) is excluded if a particle needs to be licensed under incorporation (as argued in Koopman (1991), particles must incorporate at LF). (54h) can be excluded either because it is a small clause predicate (as (54d)), or in the same way as (54f), if double object constructions are actually headed by an empty P (compare Koopman (1991)). Complementizer deletion is unavailable in Ns, suggesting complementizer deletion should be reduced to S-structure incorporation as well. N is not a licensing head for C. Idioms are excluded from complement position of N, suggesting again that the idiomatic part needs to be licensed under incorporation. Asymmetries in the complement type of Ns and Vs thus reduce to the fact that N is not a licensing head, but V is. This captures an old insight that Vs can trigger incorporation, but other categories cannot. Whether this difference can be further reduced will remain to be seen in future work. An important implication of my analysis is that it obviates the need for the Uniformity Condition, a welcome result.26 My proposal quite generally entails that complements which are not independently licensed by C, like those in (54), are licensed by incorporation to V either at S-structure or at LF. This leads to an incorporation account of raising complements, idiomatic expressions, restructuring constructions (compare wanna contraction), ECM verbs, particle constructions, double object constructions, and complementizer deletion. English allows for P-incorporation (Koopman 1991) in particle constructions, double object constructions, and probably wanna contraction, as well as C-incorporation (yielding empty complementizers) and possible ECM verbs. French has neither P-incorporation nor C-incorporation: French lacks particle constructions, double object constructions and empty Cs. This should explain certain differences between French and English first discussed in Kayne (1981) with respect to the existence of verb-particle constructions, double object constructions, and ECM verbs like believe. Note that since raising verbs exist in both French and English, both French and English have T-incorporation, possibly because the defective nature of the raising verb forces T-incorporation in the syntax. However, particle constructions, double object constructions, ECM type verbs and C deletion only occur in English, suggesting that these should all be reduced to cases of syntactic P- and C-incorporation;27 English, but not French, has syntactic P- and Cincorporation. So far, I have used the impossibility of occurring in the complement position of N as a diagnostic for elements that need to be licensed through head movement to a licensing head. Complements which may occur as complements of Ns (genitive NPs, PPs, tensed CPs and control CPs) need not be licensed through incorporation to N. Discussion in this section has focused on the fact that N is not a licensing head, whereas V is. We have also established that P is not a licensing head for predicates. P, however, is probably a licensing head for NPs, since NPs occur as complements of P. This, of course, raises questions about what precisely determines whether some element is a licensing head, and at which level. I will leave these questions for future research.


6 X0 movement to functional categories Although V-to-I-to-C movement is probably the best studied case of head movement to a functional category, similar arguments can be constructed for LF movement of the lexical category to the highest functional category, F, closing the projection of that category. Consider DPs for example. Recent work on the internal structure of NPs/DPs has shown that DPs are highly structured projections, resembling clausal projections. Universally, the structure of a DP seems to obey the following schema (abstracting from the relative order of elements; NumP stands for Number Phrase) (see Carsens (1991), Ritter (1991), Szabolcsi (1987), Valois (1991) among others): (55)

[DP [Spec [D [NumP Spec [NUM [NP]]]]

This structure is quite successful in capturing the internal syntax of DPs, such as internal agreement patterns, the distribution of pronouns, the position of different types of adjectives, and extraction possibilities, and also allows space to account for differences between languages which are argued to follow from the position to which the head N moves (see Giorgi and Longobardi (1991), Ritter (1991), and Valois (1991)). In English, it is assumed that the N does not raise (or does not raise very high), just as the V does not raise; in French, the N moves to Num; in Italian, to a position higher than Num. In the Hebrew construct state construction the N moves to D and the genitive is structurally Case-marked in Spec of NumP; in the Hebrew free genitive, the N moves to Num, and Spec of NP is marked with šel. However, although the assumption that D is the head of the NP is well motivated and useful in accounting for the internal structure of a NP and observable crosslinguistic differences, it also poses problems: standard tests for establishing what the head is, that is selectional restrictions, show that N, not D, functions as the head of the projection. It must be assumed accordingly that at S-structure the intermediate heads are transparent for government, as Grimshaw (1991) argues for in Extended Projections, although for the internal syntactic structure they are not. This problem can be solved if selectional restrictions do not need to be satisfied at S-structure, but must rather be checked at LF under government (as proposed by Koopman and Sportiche (1989:584)). N-movement into the government domain of the checking V will force N-movement into a position governed by V, the D-position of Num, with NumP moving into Spec of DP. If N moves into D, DPs are headed by N at LF, just as CPs are headed by V. Note also that Ns probably have inherent lexical properties, just like Vs: these properties must be satisfied through head movement (forcing movement to the determiner), in the same way as the T-grid of V is satisfied. Similar arguments can be made for the internal structure of APs, if these are also headed by functional categories (for proposals that this is the case for APs, see Abney (1987), Corver (1990), Stowell (1991b) and for PPs see Koopman (1991)). Languages differ as to where the head occurs at S-structure. They will be identical, though, at LF, in the sense that all XPs will be headed by lexical categories:

LICENSING HEADS 285

(56)

At LF, FPs are headed by lexical categories

S-structure differences with respect to the position of the head disappear at LF. Differences between languages should therefore be explained in terms of S-structure constraints, such as how high the X moves up in the projection. 7 Generalizing incorporation Verbs have been compared to molecules with a number of slots (the T-grid receptors) that need to be bound. This binding is accomplished by head movement. In the preceding section, I argued that lexical categories which are selected move up into the government domain of their projection so that selection can be satisfied under government. But note now that the head is exactly in the configuration that allows for head movement. This opens up the possibility that selectional restrictions are also satisfied by incorporation. This suggests that selection and theta-role assignment can also be reduced to head movement (Dominique Sportiche, personal communication).28 Let us explore this possibility further. Assume some standard version of the Projection Principle: (57)

Projection Principle: Lexical properties must be satisfied

Suppose furthermore that we extend the ideas about licensing of the temporal argument grid, namely that licensing of lexical properties can only be achieved through head movement: (58)

Lexical properties are satisfied through head movement

According to the Projection Principle (which I take to hold at LF), a lexical item is well formed if its lexical properties are satisfied, that is if its receptors are filled (or, maybe, if empty receptors are in an agreement relation with an XP in Spec position, as one might argue for wh-trace or pronouns). The syntactic structure would now follow to a large extent from the fact that lexical properties must be satisfied through head movement. Let me illustrate these points, starting with a derivation of a verb with a theta-grid consisting of two arguments, N1, the external argument, and N2, a NP internal argument. For ease of exposition, I ignore the T-grid.


Since the verb has two (N) arguments, these two arguments must be projected in the syntax, so as to provide possible binders for the receptors. Let us assume that they are projected according to a thematic hierarchy, with the external argument the highest. In order for the Projection Principle to be satisfied, N receptors must move onto the verbal molecule. This will force N to move into a DP projection position from which it can undergo head movement. There are two ways to achieve this, either by moving N into the D position, or by moving the complement of D, NumP, into Spec of DP, and N to Num. (60)

a. b.

N-to-D movement [DP Ni [NP ei]] NumP to Spec of DP and N to Num [DP [NumPj Ni[NP ei]]Dej]

Once in a governed position, N can adjoin to V, which implies V must i-command the DP: It thus follows that theta-roles must be assigned under government. In the structure in (61), however, the N receptor is not satisfied. This is achieved in a second step in which the head N moves onto the N- receptor. This step is parallel to the one for incorporation of T, discussed in section 3.1 (see in particular (14)):

There are two justifications for the two-steps process. First, both steps exist as the result of overt syntactic incorporation. Incorporation after step 1 (61) yields an adjoined structure from which the head can further excorporate. Incorporation after step 2 yields a syntactically inseparable word. And second, the two-step account plays a crucial role in accounting for the binary branching nature of phrase structure (7.1). How does the selected N move onto the external argument receptor? The V must excorporate and move to a higher V-position from where it governs the DP that corresponds to the external theta-role. As to the incorporation, the two steps described above are repeated: N adjoins to V, and N incorporates onto the open N-receptor:

LICENSING HEADS 287

Since the head N must be incorporated, V must be the head governing the DP, i.e. the verb must move to a verbal head position: this has the nice result of forcing the existence of VP shells. 7.1 Deriving binary branching The analysis developed in the preceding section makes theta-role assignment to the external argument no longer exceptional: theta-role assignment reduces to incorporation, and incorporation requires i-command. The verb is thus forced to move to a higher position from where it can incorporate the head of the external argument. A question remains, however: why must the DP carrying the external theta-role occur in a separate VP shell? Why could the structure not be flat? This seems to raise the more general question about phrase structure, namely that it seems to be binary branching (Kayne 1984). Incorporation is possible as long as the head i-commands the incorporee. In principle this allows structures like (64) where the external argument is assigned under V’, for example, or any other structure where XP and YP are both internal arguments:


(64)

[V XP NP]

Yet, such structures do not seem to exist. Head movement requires i-command. Icommand is respected in (64), and step 1 should therefore derive a licit structure (65):

In the next step, however, both X and Y need to move onto the lowest V, since this is the molecule to which the lexical receptors are attached. But in structure (65) only X is an immediate sister to V; Y is not. If only X undergoes head movement in this structure, since only X is the sister of V, a structure like (64) cannot exist: if both XP and YP need to be licensed through incorporation, both XP and YP need to be governed by a head position, i.e. both need to be projected in their own shell. Binary branching structures are thus actually forced for any XP whose head must be licensed through incorporation in order to satisfy lexical properties: there must be as many head positions as there are XPs whose head must be incorporated. 8 Conclusion In this chapter, head movement has been argued to play a central licensing role: head movement represents the way to license heads and is the driving force behind phrase structure. I argued that head movement accounts for the distribution of VPs. If one assumes that the head of a predicate must undergo head movement, the distribution of VP is explained. Moreover, the level at which head movement applies (S-structure or LF) can be determined by examining whether syntactic operations can affect the XP or not. If X undergoes syntactic head movement, the position of XP is fixed in the syntax; if not, XP can appear in an S-structure position that allows for reconstruction at LF. This led to a discussion of bare VP complements in English, where I argued (on the basis of whether VP-preposing was possible or not) that causative verbs in English and in fact universally must trigger S-structure incorporation. I furthermore showed the need to distinguish between licensing heads and non-licensing heads: N/V asymmetries follow from the fact that predicates must be licensed by moving to a licensing head (V), and from the fact that

LICENSING HEADS 289

N is not a licensing head for a particular type of head. Finally, I generalized the account and argued that all lexical properties are satisfied by incorporation. I have attempted to provide answers to a number of questions. Why does V-movement exist? V-movement exists because this is the way by which the inherent lexical properties of the verb are satisfied. At which level must V-movement occur? V-movement must hold at LF for those cases in which V-movement is driven by the satisfaction of the lexical properties of the governed verb. I argued that there is another type of head movement, involved in causative formation of the make type, that must universally take place at Sstructure. Causative verbs trigger S-structure incorporation, not because the lexical properties of the complement must be licensed (this licensing requirement is satisfied at LF), but because the causative verb itself is defective and needs to incorporate some type of I from its complement. That causatives appear to trigger S-structure incorporation of V is a by-product of V-to-I movement in that language. Languages will have morphological causatives or not depending on whether V must move to I in the syntax or not, yielding a diagnostic criterion for V-to-I movement. What accounts for language variation with respect to syntactic V-movement? The licensing of the lexical properties of V, i.e. of the T-grid of V, was argued to require two steps of head movement: (i) adjunction of the V to a particular I head and (ii) incorporation of the I onto the receptor of the lexical item. Language variation can now be described as follows: if the verb remains low in its clause, as in English, it implies that the head is stuck in adjoined position, and only (i) has applied. If V moves to a high position in the clause, it must be the case that both steps (i) and (ii) have applied in the syntax. Indications about how high the verb can move exactly come from the S-structure properties of different types of complex verb constructions (causatives, perception verbs, raising verbs and control verbs), in particular the surface shape of the complex verbs, and the syntactic properties of the constructions (restructuring, auxiliary switch, and so on). In languages in which you always move as high as you can, both steps (i) and (ii) apply in the syntax. The system described in this chapter also allows for mixed languages, like Swedish and Norwegian, that allow for both (i) and (ii), but only if V-to-C is forced, otherwise the V basically remains low (you do not apply (ii) in the syntax, unless you have to). What is the reason for head movement in general? Head movement in general exists because this is the means by which lexical properties are satisfied at LF. In the introduction I outlined a research program reducing primitive relations to movement theory, with two types of movement for licensing purposes: movement to Spec positions and head movement. Questions arise as to the compatibility of both approaches. A first problem arises in the following configuration. Suppose that for Spechead licensing the complement of X is licensed in Spec of XP but that the head of the complement must also be licensed under head movement. More concretely, suppose that T takes a complement IP, and this complement is licensed in Spec of TP. The problem that arises is that V must move through the head of IP position to T. Therefore, the head of IP is a trace which is not antecedent-governed by V in T (T does not c-command Spec of TP). Note, however, that both approaches are compatible, as long as V moves on to some higher F projection, from where it can antecedent-govern both the trace of the head in Spec position and the trace of the head in T. Basically then, the end projection can only


be one like V (as in (49)) or C, i.e. a position that licenses a particular XP, which is not its complement. Thus both approaches are compatible as long as there is a higher head projection to which V can move. This chapter raises many new questions which I hope to pursue in future work. Notes * An earlier version of this paper was presented at the V-movement conference in Maryland. I would like to thank Norbert Hornstein, David Lightfoot, Juan Uriegereka, Luc Moritz, Tim Stowell and Dominique Sportiche for their comments, questions, and suggestions. 1 This does not apply to Uriegereka (1988) where it is argued that head movement plays an extremely central role. Many of the conclusions that I reach are similar to his but differ in interesting ways. I shall not discuss these differences here. 2 Step (ii.a) is also attested in Dutch verb-raising constructions, when an additional verb is added. 3 It is possible that the structure of VP is universally invariant; what may vary is the position of functional categories (I/C/, etc.). If this were true, the head initial-/head-final parameter should be restricted to functional categories. 4 Alternatively, English V might move up a little, but surely not to the highest I category. 5 But compare Belletti (1994) for a proposal that the highest I category (AGR) can be recursive: only auxiliaries may move to this higher empty AGR. 6 It could be argued that in certain governed contexts the empty I nodes are in fact licensed by incorporation to C at S-structure. 7 Lexical properties appear to be hierarchically organized with arguments projected in the most deeply embedded shells, the aspectual grid on top of the theta-structure, and the temporal grid on top of the aspectual structure. The question arises why this is so. Dominique Sportiche has suggested to me that the hierarchical position of T with respect to Aspect would follow if Aspect is specified for a T-grid. 8 A theory of the temporal grid is presupposed. See Stowell (1982) for discussion. 9 V-to-T movement provides a second argument as to why V must count as the head at this point in the derivation. If ASP were the head in (12), the T head would not be able to attach to the T-receptor. 10 Fabb argues that bare VPs are licensed (Case-marked) by ing in adjuncts of the following kind: the man buying groceries. These adjuncts are clausal, however, as Zagona argues, based on the fact that they admit the aspectual have. 11 This does not cover examples like rather than visit her mother, as Peter Culicover has pointed out to me. 12 Uriegereka (1988) also proposes that small clauses are licensed through incorporation. He argues for incorporation by appealing to Baker’s (1988) idea that incorporation is an alternative to Case-marking to make items visible. 13 Note that APs can occur as adjuncts. This shows that the A can be dominated by a functional projection comparable to C which is able to license A independently. As argued in Stowell (1991b), adjectives incorporate at LF. 14 If verb projection raising in Flemish or Swiss German is indeed VP-raising, it must be assumed that the main verb governs the raised VP at S-structure. 15 Reconstruction of predicates is forced to the original position, whereas reconstruction of NPs can occur in intermediate Spec of CP positions (Barrs (1986), Huang (1991)):

LICENSING HEADS 291

(i)

a. b. c. d.

visit each otheri[ I believe [t [theyi did t]]] *visit each otheri [theyi believe [t [ I did t]]] Which pictures of each otheri [do you think [t [ theyi like t]]] Which pictures of each otheri [did theyi think [ t [I liked t]]]

If the head must be licensed under head movement to a licensing head, reconstruction is forced to the original position. NPs may reconstruct in intermediate positions, basically because the original position can be licensed as a pronoun at LF; there are no pronouns corresponding to V. 16 I do not distinguish between A and A’ positions for the purpose of reconstruction. The logic of this chapter requires idiomatic expressions to be licensed through head movement as well. Since an idiom chunk can appear in subject position, advantage appears to have been taken of John, reconstruction of the subject must be forced. 17 (24) yields an account for the impossibility of VP-preposing in, for example, Romance languages. If the verb has moved out of the VP, the projection corresponding to the English VP that could prepose would contain a trace in head position. 18 Suppose alternatively that reconstruction does not apply to (25a). Instead, the I categories will move to C at LF, yielding the representation:

(i) [visit her mother] did she t i.e. the VP is in a Spec-head relation with C, the element that we have assumed licenses the V, and the V can be licensed in this position. Reconstruction must still be assumed, however, given examples like: (ii) [visit her mother] I believe that Mary did 19 I*P-preposing after passivization is also impossible:

(i) *[Visit her mother every week] Mary was made (to) No attempt to analyze (i) will be made. The remnant should be headgoverned by to or by the verb. However, as shown by Zagona (1988a), the infinitival to only counts as an appropriate head-governor (in her terms a temporal marker) for an empty VP if it has raised to C of a theta-marked category. This accounts for the following paradigm: (ii)

a. b. c.

And leave John tried to *And lie repeatedly John believes Mary to *And lie repeatedly John seems to

20 This argument does not stand if in French the accusative NP does not raise to [Spec, make], while it does in English, as argued in Sportiche (1990). 21 Syntactic movement of I, although a higher I, is non-problematic and independently observed: compare I-to-C in both main and embedded clauses (SAI), and to-to-C, as argued for example in Zagona (1988b), Sportiche (1988) and Lightfoot (1991: ch. 4), and C-to-V movement (see below).


22 The ungrammaticality of (42) is related to the unavailability of the permissive reading in the source: 23 I leave open for now how true causative verbs taking tensed CPs should be incorporated into this picture. 24 Ideally, the information should come from other categories, like the internal syntax of DPs/ NPs. 25 This forces an analysis by which belief is not derived in the syntax by head movement. The following problems arise with an analysis by which belief is derived in the syntax from [N [believe [to be crazy]]] with head raising of believe into the N slot. Although the embedded verb cannot reach believe in the syntax, the raised V must be prevented from reconstructing and licensing the embedded predicate at LF; in some sense the T-slots of the V are not available as licensers. This may be related to the fact that a N can license a verbal stem in the morphology, but not in the syntax. 26 Uriegereka (1988) arrives at a similar conclusion from a quite different angle. 27 Maybe C-incorporation further reduces to P-incorporation. 28 Uriegereka (1988) proposes a similar idea. He proposes that only D incorporates, substituting for the argument, while the N remains in its projection. Although Dincorporation is attested in the syntax, I believe it is D-incorporation to the head of the case projection or to the head of the Clitic Voice projection (Sportiche 1992), and unrelated to argument substitution (as in Uriegereka (1988)), which is overtly attested as N-, V-, P- or Aincorporation.

References Abney, S. (1987) “The noun phrase in its sentential aspect”, unpublished PhD Dissertation, MIT. Baker, M. (1988) Incorporation: a Theory of Grammatical Function Changing, University of Chicago Press, Chicago. Barrs, A. (1986) “Chains and anaphoric dependence”, unpublished dissertation, MIT. Belletti, A. (1994) “Verb positions: evidence from Italian”, in D.Lightfoot and N. Hornstein (eds) Verb Movement, Cambridge, Cambridge University Press. Bennis, H. (1986) “Gaps and dummies”, unpublished PhD thesis, University of Tilburg. Besten, H. den (1983) “On the interaction of root transformations and lexical deletive rules”, in W.Abraham (ed.) On the Formal Syntax of Westgermania, John Benjamins, Amsterdam. Castens, V. (1991) “The morphology and syntax of determiner phrases in KiSwahili”, unpublished PhD dissertation, UCLA. Chomsky, N. (1986) Knowledge of Language: Its Nature, Origin, and Use, Praeger, New York. Chomsky, N. (1991) “Some Notes on Economy of Derivation and Representation”, in R. Freidin (ed.) Principles and Parameters in Comparative Syntax, Cambridge, Mass., MIT Press. Chomsky, N. (1995) The Minimalist Program, Cambridge, Mass, MIT Press. Cole, P., G.Hermon and L.M.Sung (1990) ‘Principles and parameters of long distance reflexives”, Linguistic Inquiry 20: 1–22. Corver, N. (1990) “The syntax of left branch extractions”, unpublished doctoral dissertation, University of Tilburg. Dikken, M. den (1990) “The structure of English complex particle constructions”, in R. BokBennema and Coopmans (eds) Linguistic in the Netherlands 7: 23–32. Emonds, J. (1976) A Transformational Approach to English Syntax, New York: Academic Press.

LICENSING HEADS 293

Evers, A. (1982) “Twee funtionele principes voor de regel ‘verschuif het werkwoord’ ”, GLOT 5: 11–30. Fabb, N. (1983) “Syntactic affixation”, unpublished doctoral dissertation, MIT. Giorgi, A. and G.Longobardi (1991) The Syntax of Noun Phrases: Configuration, Parameters and Empty Categories. Cambridge University Press, Cambridge. Grimshaw, J. (1991) “Extended projections”, unpublished manuscript, Brandeis University. Guéron, J. and T.Hoekstra (19888) “T-chains and the constituent structure of auxiliaries”, in A.Cardinaletti, G.Cinque and G.Giusti (eds) Constituent S-structure, Foris Publications, Dordrecht. Higginbotham, J. (1985) “On semantics”, Linguistic Inquiry 16: 547–593. Huang, C.-T. (1982) “Logical relations in Chinese and the theory of grammar”, unpublished doctoral dissertation, MIT. Huang, C.-T. (1991) “Reconstruction and structure of VP: some theoretical consequences”, Proceedings of NELS 21. Kayne, R.S. (1981) “On certain differences between French and English”, Linguistic Enquiry 12. Kayne, R.S. (1984) Connectedness and Binary Branching, Foris Publications, Dordrecht. Kayne, R.S. (1985) “Principles of particle construction”, in J.Guéron, H.G.Obenauer and J.Y.Pollock (eds) Grammatical Representation, Foris Publications, Dordrecht. Kayne, R. (1989) “Null subjects and clitic climbing”, in O.Jaeggli and K.Safir (eds) The Null Subject Parameter, Foris Publications, Dordrecht. Koopman, H. (1984) The Syntax of Verbs: from Verb movement rules in the Kru languages to Universal Grammar, Foris Publications, Dordrecht. Koopman, H. (1991) “The verb particle construction and the syntax of PPs”, unpublished manuscript, UCLA (Chapter 8 in this volume). Koopman, H. and D.Sportiche (1989) “Pronouns, logical variables and logophoricity in Abe”, Linguistic Inquiry 20, 555–589 (Chapter 2 in this volume). Koopman, Hilda and Dominique Sportiche (1991) “The position of subjects”, Lingua 85: 211–258, included in Sportiche (1998). Koster, J. (1987) Domains and Dynasties, Foris Publications, Dordrecht. Larson, R. (1988) “On the double object construction”, Linguistic Inquiry 19: 335–392. Lebeaux, D. (1983) “A distributional difference between reciprocals and reflexives”, Linguistic Inquiry 14: 723–73. Lema, J. and M.-L.Rivero (1990) “Long head movement: ECP vs. HMC”, in J. Carter et al. (eds) Proceedings of NELS 20: 333–437. Lightfoot, D. (1991) How to set Parameters: Arguments from Language Change, MIT Press, Cambridge, Mass. May, R. (1977) “The grammar of quantification”, unpublished doctoral dissertation, MIT. May, R. (1985) Logical Form: Its Structure and Derivation, MIT Press, Cambridge, Mass. Moritz, L. and D.Valois (1992) “Negation, LF movement and subjacency”, paper presented at NELS 22, Delaware (published as Moritz, L. and D. Valois (1994) “Negation, LF movement and subjacency”, Linguistic Inquiry 25). Pollock, J.-Y. (1989) “Verb movement, Universal Grammar and the structure of IP”, Linguistic Inquiry 20: 365–424. Ritter, E. (1991) “Two functional categories in noun phrases: evidence from modern Hebrew”, unpublished manuscript, UQAM. Rizzi, L. (1991) “Residual verb second and the Wh-criterion”, Technical Reports in Formal and Computational Linguistics 2, Faculty of Letters, Geneva (also published in Belletti, A. and L.


Rizzi (eds) Parameters and Functional Heads: Essays in Comparative Syntax: 91–116, Oxford University Press, New York). Roberts, I. (1991) “Excoporation and minimality”, Linguistic Inquiry 22: 209–218. Sportiche, D. (1988) “Conditions on silent categories”, unpublished manuscript, UCLA. Sportiche, D. (1990) “Movement, agreement, and case”, unpublished manuscript, UCLA (included in Sportiche (1998)). Sportiche, D. (1992) “Clitics, voice and Spec head licensing”, GLOW Newsletter 28: 46–47. Sportiche, D. (1995) “Clitic constructions”, in L.Zaring and J.Rooryck (eds) Phrase S-structure and the Lexicon, Kluwer Academic Publishers, Dordrecht (included in Sportiche (1998)). Sportiche, D. (1998) Atoms and Partitions of Clause Structure, Routledge, London. Stowell, T. (1991b) “The realignment of arguments in adjective phrases”, in S.Rothstein (ed.) Perspectives in Phrase Structure: Heads and Licensing, Academic Press, New York. Stowell, T. (1981) “Origins of phrase structure”, unpublished PhD dissertation, MIT. Stowell, T. (1982) “The tense of infinitives”, Linguistic Inquiry 13: 561–570. Stowell, R. (1983) “Subject across categories”, The Linguistic Review 2: 285–312. Stowell, T. (1991a) “Small clause restructuring”, in R.Freidin (ed.) Principles and Parameters in Comparative Grammar, MIT Press, Cambridge, Mass. Szabolcsi, A. (1987) “Functional categories in the noun phrase”, in I.Kensei (ed.) Approaches to Hungarian, vol. 2, Jate Szeged, Budapest. Taraldsen, K.T. (1983) “Parametric variation in phrase structure”, unpublished PhD dissertation, University of Tromsø. Travis, L. de Mena (1984) “Parameters and effects of word order variation”, unpublished PhD dissertation, MIT. Uriegereka, J. (1988) “On government”, unpublished doctoral dissertation, University of Connecticut. Valois, D. (1991) “The internal syntax of DP”, unpublished doctoral dissertation, UCLA. Zagona, K. (1988a) “Proper government of antecedentless VP in Spanish and English”, Natural Language and Linguistic Theory 6: 95–128. Zagona, K, (1988b) Verb-Phrase Syntax: A Parametric Study of English and Spanish, Kluwer Publishers, Dordrecht. Zanuttini, R. (1991) “Syntactic properties of sentential negation: a comparative study of Romance languages”, unpublished doctoral dissertation, University of Pennsylvania (published as Zanuttini, R. (1997) Negation and Clausal S-structure. A Comparative Study of Romance languages, Oxford Studies in Comparative Syntax).

10 ON VERBS THAT FAIL TO UNDERGO VSECOND*

In this chapter I develop an analysis for the syntactic distribution in Dutch of particle verbs, verbs with an inseparable prepositional prefix, verbs with the prefix her- “again”, and particle verbs prefixed with her-. Empirical evidence based on the distribution of particle verbs and the prefix her- will establish that the finite verb in nonroot environments in Dutch is in situ (or more correctly quite low in the structure). I will show that the proposals developed in Koopman (1994) for the way in which lexical properties are satisfied yield a direct and simple explanation for the paradigms under discussion and solve certain long-standing problems in Dutch syntax. Insofar as these proposals uniquely rely on the mechanism of head movement and are intended as a general theory for the way in which lexical properties need to be satisfied, the analysis developed here yields strong support for my 1994 proposals, which can be seen as a particular implementation of Chomsky’s (1993) checking domain. 1 Where is the finite verb in Dutch nonroot clauses? From the extensive work on V-movement in the different Germanic languages, it appears that these languages vary with respect to the position of the finite verb in nonroot environments. Within the Germanic VO languages, the situation can be roughly summarized as follows:1 (1) Root environments V-second Restricted V-second Nonroot environments V-in-situ V-to-I movement2

Swedish, Norwegian, Danish, Icelandic, Faroese, Yiddish English Swedish, Norwegian, Danish, English, Faroese Icelandic, Yiddish

With the exception of English, these languages are all V-second (V2) languages. Any finite verb must move through intermediate head positions to the highest functional head position in root clauses. Any finite verb is therefore in principle able to move to I. The position of the finite verb in nonroot environments varies, however. In some languages


the verb stays in situ; in others it must raise to I. Overt V-to-C movement in root environments therefore does not imply overt V-to-I movement in nonroot environments. In light of these data, it is difficult to decide exactly where the finite verb is in nonroot environments in the V2 Germanic OV languages like Dutch and German. If some I categories are final in Dutch and German, then I and V are adjacent. Empirical arguments that establish whether the finite verb is in some final I category or in some position lower than I have been extremely difficult, even impossible, to construct.3 This basically unsettled issue has been present in the literature on Dutch and German for a long time. In the older literature it takes the form of discussions over whether there is an I projection or not: since there is little or no distributional evidence for an I-final category, it is often concluded that such a category must be lacking and that the finite verb simply is in the VP. The standard view is probably that the language has V-to-final-I movement in nonroot environments. This view is based on the idea that since the language has overt Vto-C movement in root environments, it also has V-to-(final)-I movement in nonroot environments. This view is not empirically supported, however, as can be concluded from (1): overt V-to-C movement does not entail overt V-to-I movement in nonroot environments.4 In the absence of clear empirical arguments, the issue could be decided on the basis of other phenomena. Dutch specific DPs, for example, must scramble to some position quite high in the clause. If, as overt DP-movement in Icelandic suggests, such movement is contingent on overt V-movement, the Dutch verb must have moved in the overt syntax to some position in order to allow domain extension. This argument presupposes that the Icelandic phenomenon is fully understood. In the end, then, empirical arguments are needed, and the question arises whether such arguments can be constructed. 2 On verbs that fail to undergo V-second There are empirical data leading to the conclusion that the finite verb in Dutch can be in situ. Haider (1993:3–24) presents the argument in rudimentary form.5 He notes that there is a class of verbs in Dutch and German that fail to undergo V2 but that nevertheless can have finite forms in nonroot environments. These verbs consist of a particle verb (a verb with a separable P prefix) preceded by a prefix her- “again” (comparable to English re-). Her- can combine not only with certain classes of noncomplex verbs (2a) but also with particle verbs (2b).6 (2)

b.

a.

kauwen openen schrijven lezen

uitgeven indelen uitzenden

“publish” “divide” “transmit”

“chew” “open” “write” “read”

herkauwen heropenen herschrijven herlezen heruitgeven herindelen heruitzenden

“ruminate” “reopen” “rewrite” “reread” “republish” “redivide” “retransmit”

VERBS THAT FAIL TO UNDERGO V-SECOND 297

invoeren indijken

“introduce”, “import” “put in dikes”

herinvoeren herindijken

“reintroduce” “put in dikes again”

The verbs in (2a) and (2b) are syntactically and morphologically distinct. Thus, their syntactic distribution differs. Finite forms of the verbs in (2a) can undergo V2.7 (3)

a.

Koeien herkauwen hun eten. cows rechew their food “Cows ruminate.” b. Vandaag heropent de stad het museum. today reopens the city the museum “The city reopens the museum today.” c. De dichter herschreef deze passage eindeloos. the poet rewrote this passage endlessly “The poet endlessly rewrote this passage.” d. Gisteren las hij de eerste dertig bladzijden; vandaag herleest hij ze yesterday read he the first thirty pages; today reread he them already “Yesterday he read the first thirty pages; today he reread them already.”

But finite forms of the verbs in (2b) cannot, whether the entire verb preposes or whether part of the complex verb is stranded. (4) a. *Vandaag heruitgeven ze het. *Vandaag heruit. today reoutgive they it today reout b. *Deze klas herindeelden ze vandaag in this class redivid they today in ed “They redivided this class today in smaller groups.” c.

d.

e.

*Dit programma heruitzenden this program reoutsend “They rebroadcast this program regularly.” *Deze wet herinvoerden ze this law reimported they “They reinstated this law last year.” *Deze polder herindijken this polder redike “They redike this polder next week.”

ze they

geven

ze

het

give

they

it

kleinere smaller

groepen. groups

ze they

regelmatig. regularly

vorig last

jaar. year

volgende next

week. week


It is intuitively clear why these verbs cannot undergo V2: a particle must be stranded under V2, and her- must pied-pipe. This intuition will be formalized in sections 5, 6, and 7. It is not lack of finite verb forms that explains the restriction on the verbs in (4): the verbs in (2b) can be inflected as long as they are in nonroot environments, and this is in large part why this paradigm is interesting. (5)

a.

omdat deze uitgeverij zulke boeken vandaag de dag heruitgeeft because this editor such books today republishes “because this company does not presently republish such books”

niet not

Though the forms in (5) are fine, these verbs occur most naturally as bare infinitives or participles.8 (6)

d.

a.

omdat ze deze boeken niet because they these books not “because they will not republish these books”

omdat ze vorig jaar deze wet because they last year this law “because they have reinstated this law last year”

zulle will

hebben have

heruitgeven republish

heringevoerd reimported

The different syntactic distribution is not the only distinction between the verbs in (2a) and (2b). A morphological distinction shows up in their participles. The verbs in (2a) are mostly incompatible with the participial prefix ge-, as is typically the case for verbs with an unstressed prefix (like be- and ver-, for instance). Her-simply attaches to the verb with the participial suffix (indicated as part).9 (7) a. b. c. d.

ge- Vpart her- V part geopend heropend “opened, reopened” gekauwd herkauwd “chewed, rechewed” gelezen herlezen “read, reread” geschreven herschreven “written, (re)written”

her- ge- Vpart *hergeopend

ge- her- V part10 *geheropend

*hergekauwd

?geherkauwd

*hergelezen

*geherlezen

*hergeschreven

*geherschreven

With the verbs in (2b) there is no morphological interaction between the prefixes her- and ge-. The particle precedes ge-, which in turn obligatorily precedes the participial verb form. (8) a.

P- ge- V part uitgegeven

her- P- V part *heruitgeven

her- P- ge- V part heruitgegeven

ge- her- P- V part *geheruitgeven


b. c. d.

“published, republished” ingedeeld *herindeeld heringedeeld “divided, redivided” uitgezonden *uitzonden heruitgezonden “broadcast, rebroadcast” ingevoerd *herinvoerd heringevoerd “imported, reimported/ reinstated”

*geherindeeld *geheruitzonden *geherinvoerd

Thus, when her- attaches to a nonparticle verb, the resulting verb can undergo V2 and acts like a regular verb with a prefix. As with other compound verbs, the participial prefix gedoes not attach to her-V part. When her- attaches to a particle verb, the resulting verb fails to undergo V2, yet finite forms of these verbs can occur in environments where V2 is not triggered. With these verbs, there is no interaction between her- and ge-. 3 The finite verb in Dutch nonroot environments is in situ Particle verbs prefixed with her- fail to undergo V2, because the requirements that a particle must be stranded under V2 and that her- must pied-pipe somehow conflict. (9)

a.

Ze zonden they sent “They broadcast it.”

het it

uit. out

Although a precise analysis for this paradigm will be presented in sections 6, 7, and 8, we can draw a conclusion about the position of the finite verb in (9e) at this point. With particle verbs, the verbal part carrying finite morphology must undergo V2 (9a). Since it is the finite part of the particle verb that must move, it is natural to tie this movement to the licensing of the finite morphology. Suppose therefore that when the finite verb undergoes V2, the finite morphology is satisfied, or checked, in the overt syntax, in the appropriate head positions (say, T and Agr), which occur lower than C.11 The failure of P to pied-pipe under V2 (9b) and the lack of finite verb forms of her-P-Vs in root environments (9d) can now be explained if the resulting configuration does not allow the finite morphology to be checked in the overt syntax (see sections 6.1 and 8). If this is correct, it must be assumed that the finite morphology is obligatorily checked in the overt syntax when V moves to C. However, since these verbs can have finite forms in nonroot environments, it also follows that the finite morphology of V is not checked in the overt syntax in (9e). Therefore, the verb in (9e) has not undergone overt movement to T and Agr: the Dutch finite verb therefore is (or rather can be) lower than T and Agrs in nonroot environments. I refer to the position of the finite verb in nonroot environments as the in situ position for convenience. Note, however, that this is a sloppy use of the term in situ: the finite verb could have undergone some raising, as long as it raises to a position lower than T. Dutch thus basically patterns with Swedish and Norwegian.


Dutch specific DPs scramble to a position quite high in the clause (section 2). Although the verb in nonroot contexts is quite low, the DPs happily scramble high. In (5a), (5b), and (5e), for example, the objects have visibly scrambled. In fact, there does not seem to be any difference between scrambling of specific DPs in root and nonroot environments. This implies that the movement of specific DPs is not necessarily contingent on overt Vmovement. Since the verbal morphology of her-P-Vs cannot be checked in the overt syntax, it follows that any verbal morphology compatible with these verbs in nonroot environments (the infinitival suffix, the participle, and probably the te infinitive)12 is base-generated and later checked at LF. The paradigms discussed so far raise many questions. Why does attaching her-to a particle verb bleed V2? How are we to account for the syntactic distribution of particle verbs? If the finite verb can be base-generated, why is it not possible to check the morphology when the entire V moves? If the V is in situ in nonroot environments but raises at LF, why does movement at LF yield the correct results? What is the difference between the representation in the overt syntax and the representation at LF? In order to answer these questions, I present the proposals developed in Koopman (1994), which allow a quite elegant analysis of particle verbs (section 6) and the paradigm of her- (section 8). 4 Licensing heads In Koopman (1994) I pursue a research program based on the idea that all licensing principles are reduced to two fundamental licensing configurations spec-head licensing (Sportiche 1990, 1992) and head incorporation—and I explore the role of the latter. The spec-head configuration is achieved through (XP-) movement, incorporation through head movement. I argue in particular that head movement plays a fundamental role in licensing lexical properties. Lexical items resemble atoms: they come with a number of open positions (morphological, thematic, and tense-aspect slots)—call them receptors—that need to be satisfied by virtue of the Projection Principle. A receptor is satisfied when an element attaches to it, forming some kind of molecule. Elements attaching to receptors are heads. A head reaches its designated receptor by means of head movement. Head movement is severely constrained, as a consequence of the theory of locality. I will basically assume Sportiche’s (1988, 1989, 1990) theory of locality, which is summarized in (10). (10)

a.

b.

I-command α i-commands (immediately commands) β if the first constituent containing α contains β. Government α governs β if a i-commands β and no barrier intervenes between α and β.


c.

Barrier Xmax is a barrier for β (β • X0) if any non-L-marked projection of X0 includes β.

A constituent will have a chance to be governed by a head only if it is included in a constituent that is sister to a head; thus, head movement is possible only from sisters, whether XP or X0 sisters. Moreover, in the case of complex adjunction structures, this theory predicts that only the head can undergo further movement (excorporate);13 since an adjoined element is not the head of the complex word, the projection of the head will be a barrier for the adjoined head (see Koopman (1994) for discussion). For reasons that will become clear, I must assume a definition for antecedent government that is very close, but not identical, to (10b).14 (11)

Antecedent government α antecedent-governs β if the first category containing α contains β and no barrier intervenes between α and β.

How are lexical receptors satisfied? As an example, consider a V, V*, selecting for some VP complement. V* has a V receptor that must be filled at LF by the head of the selected complement. The embedded V must reach the receptor through head movement; the selected VP must therefore be in a complement relation with V*. Similarly, when V selects for an NP or a PP, the N or P must incorporate into the N or P receptor of V through head movement. This implies that the N or P must move within its extended projection to a position from where it can be incorporated. Finally, and crucially, I assume that attachment to the receptor involves at least two local steps of head movement, with (12a) necessarily preceding (12b). (The atomic lexical item is boldfaced; receptors attached to the atom are indicated with double lines. The representations below are incomplete; I will not be concerned with the internal structure of the “molecule” and will concentrate only on the immediate syntactic environment. For discussion, see Koopman (1994).)


The representations in (12) can be seen as representations of the two types of head movement assumed in the literature (see Rizzi and Roberts (1989), for example): adjunction (12a) and substitution (12b). I claim that the adjunction type of head movement always precedes the substitution type of head movement; this will play an important role in this article. Since lexical properties must be satisfied at LF, (12b) must apply at LF at the latest. However, given the hypothesis that covert processes can be anticipated in the overt syntax, we expect to find the following cases: (13) a. b. c.

Overt syntax (12a) (12a) and (12b)

Covert syntax (LF) (12a) and (12b) (12b)

Example No visible incorporation Dutch particle verbs her-V; inseparable [P-V]

(13a) yields no visible incorporation structures in the overt syntax. As I will show, (13b) illustrates Dutch verb-particle constructions, and (13c) illustrates the verbs prefixed with her- that were discussed earlier, as well as other verbal forms preceded either by a prefix (be-, ver-) or by an inseparable P (see section 7). I discuss (13b) and (13c) in detail below and show how their syntactic distribution is accounted for. 5 Particle verbs The derivation in (14) exemplifies (12a) for the P-incorporation that takes place with particle verbs.15 In the examples below I represent only the P receptor and the T receptor (used to indicate finiteness for convenience) of the atomic V, which is boldfaced; I will maintain a head-final analysis for the Dutch VP throughout.


Crucially, the particle, P, does not move onto the P receptor until LF.16 This structure accounts for the syntactic behavior of verb-particle constructions. As is well known, but as yet unexplained, the finite V must excorporate under V2, leaving the P stranded. (15)

a. b.

*Hij opbelt he upphones Hij belt he phones “He calls Jan up.”

Jan Jan

Jan. Jan op. up

Why is this so? What forces movement of V, and what prevents movement of V? To make things even more complicated, head movement of V in another process, V-raising, is possible. Either V or V may undergo V-raising.17 (16)

a.

b.

dat Jan Marie wil that Jan Marie want “that Jan wants to call Marie” dat Jan Marie op that Jan Marie up “that Jan wants to call Marie”

opbellen upphone wil want

bellen phone

(V-raising of V)

(V-raising of V)

I will address these questions in turn. Specifically, I will show that pied-piping of V is impossible, leaving excorporation of V as the only possibility, and I will provide an explanation for the asymmetry between V2 and V-raising structures.


5.1 Pied-piping of V What excludes pied-piping of the entire V, with subsequent checking of the finite morphology? Suppose that the entire verbal complex adjoins to some functional category —say, T—as in (17a). This yields the i-command configurations in (17b). In this structure, the lexical verb, V, i-commands P, since the first node dominating V contains P. Crucially, V does not i-command T; V therefore does not govern T, and T cannot incorporate into the appropriate receptor of V, even if V counted as the head of T/ V. If we assume that T must incorporate into the T receptor when the syntactic configuration is met (section 4),18 pied-piping of the particle verb is excluded in a quite simple fashion. Movement of the entire particle verb can therefore never lead to a configuration where a receptor must be satisfied in the overt syntax: the syntactically adjoined P blocks any further receptor binding in the syntax. This raises the question of how receptor binding can be satisfied at LF, when the finite verb remains in situ in the overt syntax. At LF the P receptor of V must be filled. If we assume that P incorporates into its receptor as soon as it can, P will incorporate into V in the configuration in (14). Once P has incorporated, V can excorporate and adjoin to T at LF. It is interesting that this LF derivation basically mirrors the overt syntax, as discussed in section 6.2.

5.2 Excorporation of V Given the structure in (14), V can excorporate. V is the head of the complex V, and therefore of the VP; as a result, it can move and adjoin to some higher head, where, say, T and Agr are checked. As we have seen, Dutch allows tense and agreement morphology to be


satisfied at LF in nonroot environments. However, when V moves to C, tense and agreement morphology must be assumed to be obligatorily checked in the overt syntax. Thus: (18) a. T, Agr must incorporate into their receptors in the overt syntax when the checking configuration is met.19 b. T, Agr incorporate into their receptors in the covert syntax elsewhere. If T and Agr receptors are properties of V, it follows that V must become a sister of T or Agr so as to allow receptor binding (only in this configuration does the element with the receptor i-command the element that must incorporate into the receptor). This forces head movement followed by excorporation of the finite verb for each step, as in (19).

V finally moves on to the highest functional projection in the root clause, stranding the elements that are adjoined to the head projections it moves through (like the clitics, which are adjoined to functional positions just below C). In the configuration in (19a) V icommands T, and T can attach to its receptor, satisfying the morphological properties of


V. In order for V to excorporate, however, V must also count as the head of the complex word at this point. I have proposed, precisely for this reason, that there is a certain ambiguity regarding what element counts as the head in cases of mutual i-command. If only (12a) applies in the overt syntax, V will not be the head of the complex word. However, in order for V to be able to excorporate, V must count as the head of the complex word. This assumption is not unreasonable: since T incorporates into V, TP becomes in effect a projection of the features of V, which now includes T. Note that this problem only arises in very restricted configurations: there must be mutual i-command, and the item with the receptor must move to some higher head position that contains an element that must incorporate into a verbal receptor.20 In other words, when V adjoins to a higher functional head, V counts as the head only when it is forced to, namely, when the functional head has to incorporate into the receptor. If incorporation into the receptor does not take place overtly, V is unable to excorporate to any higher position in the overt syntax (although the whole complex may undergo further head movement). If incorporation into the receptor does take place, V is able to move to higher functional projections, as in the case of V2.21 5.3 On the asymmetry between V-raising and V-second As mentioned above, there is an asymmetry between movement in V2 structures and movement in V-raising structures. In V2 structures, excorporation of V is the only possibility allowed by the theory, once it is established that the finite morphology must be checked in the overt syntax when V moves to C. In V-raising structures, either the complex particle verb V or V may undergo V-raising, respectively pied-piping or stranding P (compare (16)). This asymmetry can be accounted for if V-raising is another instance of head adjunction (12a), but not an instance of receptor binding (12b), in the overt syntax. Support for this analysis comes from the fact that the verbal complex behaves like a complex syntactic word on the one hand, but like a sequence of morphosyntactically independent words on the other. As is the case with particle verbs, only the finite verb excorporates under V2. Consider the derivations for the examples in (16), shown in (20), where the verb that triggers V-raising is indicated as V*. Both movement of V and movement of V are allowed in the overt syntax. V is the head of VP and can therefore undergo head movement. Because V is the head of the complex V, VP is not a barrier and V can excorporate. Next consider the covert syntax. At LF V needs to attach to the V receptor of V*. Thus, V must undergo head movement to the receptor of V*. This is possible in either (20b) or (20c). In both cases V* i-commands and governs V, and V can therefore incorporate into the V* receptor at LF. Movement of the entire verbal complex under V2 is excluded in exactly the same way as movement of the particle verb under V2 discussed in the previous section: the T receptor of the molecular V* must be filled by T, which requires the molecular V* to i-command T. This configuration can only be achieved by excorporation of the molecular V*. The asymmetry between what can move under V2 and what can move under V-raising thus follows from a


basic structural asymmetry: when the receptor licenser is structurally higher than the molecule, excorporation is the only possibility; when it is lower, excorporation or piedpiping of a complex syntactic word yields a configuration from which receptor binding can take place.

Before we consider incorporation structures where both (12a) and (12b) apply in the overt syntax, it might be useful to discuss briefly how one determines that (12a), but not (12b), applies in the overt syntax. The application of (12a) in the syntax leads to schizophrenic syntactic behavior. It relies on evidence that a head has overtly incorporated into another head. Evidence for overt incorporation can come from the fact that in certain constructions (like V-raising constructions) the complex head can undergo further head movement; from the syntactic immobility of the incorporated head; from the syntactic


immobility of the constituent whose head has undergone head movement (see Koopman (1994) for discussion); and possibly from the presence of “transparency” (i.e. restructuring) phenomena. Evidence for overt incorporation is often obscured by the fact that the complex syntactic word consists of morphosyntactically independent words, with part of the complex syntactic word typically excorporating for the “checking” of overt morphological properties; as shown in this section, though, this is exactly what the theory predicts.

6 Inseparable P-V Her-prefixation, be-prefixation,22 and verbs with an inseparable P prefix represent instances where both steps in (12) apply in the overt syntax of Dutch. Discussion in this section is restricted to the formation of verbs with an inseparable P prefix, as illustrated in (21) (see van Riemsdijk (1978) for discussion).


(21)

a. b. c.

ondermijnen onderzoeken overkomen

“undermine” “research” (lit.: under-search) “happen” (lit.: over-come)

As shown in (22), phonologically the verbs with an inseparable P prefix in (21) and particle verbs are distinct: P is accented in the latter, but V is accented in the former (this is generally the case for verbs with verbal prefixes). I have nothing to say about accent here, but simply use the latter pattern as a diagnostic criterion for receptor binding. (22)

a.

overkomen

“come over”, “visit”

accent

b.

overkomen

“happen”

accent

on P verbs on V verbs

with

particle

with

[PV]

The verbs in (21) also differ morphologically from particle verbs: the participial prefix gecannot attach either to the entire V23 or to V (recall that the particle attaches to a ge-V part form with particle verbs; compare (8) and (23c)). (23)

[P-V part] ondermijnd overkomen versus *overkomen

a. b. c.

[P-ge-v part] *ondergemijnd *overgekomen

ge-P-V-part ?geondermijnd *geoverkomen

“undermined” “happened”

overgekomen

*geoverkomen

“visited”

Finally, the syntactic distribution of the verbs in (21) is distinct from that of particle verbs. The verbs in (21) undergo V2 and P-stranding is disallowed. (24)

a.

b.

Mij overkomt dit niet vaak. me overcomes this not often “This does not happen often to me.” *Mij komt dit niet vaak over. me comes this not often over

Both the morphological and the syntactic evidence show that the P is truly part of the molecular V in the overt syntax. What syntactic structure do these verbs project? I assume that these verbs have basically the same syntactic structure as the verb-particle construction, but with the P head incorporating into the P receptor in the overt syntax. The problem with these verbs is not so much showing that P is part of the V, but showing that P is related to a syntactic projection in the overt syntax. Arguments in favor of a syntactic analysis derive from the complementation properties of these verbs. Verbs with P prefixes and particle verbs impose similar restrictions on the type of complement they take. These restrictions can be explained if a syntactic PP projection is assumed.24 I indicate the property that the P


receptor must be filled in the overt syntax by arbitrarily assigning a * to the receptor, as in the derivations in (25). Equivalently, we might assume that the complex verb is basegenerated, the syntactic structure being projected from the complex lexical item.

Since P incorporates into the receptor of V, it becomes part of the verbal molecule. Consequently, it pied-pipes under V2. Truly incorporated Ps and the participial prefix ge- are in complementary distribution in standard Dutch. Suppose that ge- is an expletive element25 licensing some morphological (or perhaps phonological) property of the molecular participial verb form, Vpart’26 which is essentially that the participial molecule fit the template in (26). (26)

[Vpart X0+VPART]


If some X0 incorporates into the receptor of the molecular verb, (26) will be satisfied, and ge- need not be present. However, if no X incorporates, X will be spelled out as the morphological expletive element ge-.27 7 The paradigm of herWe now have all the ingredients needed to account for the paradigm of her-. I will treat her(and re- as well) as a syntactic head that is higher than V, with V incorporating to her-. There are several reasons for this. The first reason is semantic: the semantics of her- are aspectual. If aspect is higher than V, her-should be higher than V as well. The second reason is based on my work in progress. Treating her- and re- as higher heads yields an account for the classes of Vs to which her- and re- cannot attach that is more adequate than the one presented in Keyser and Roeper (1992) and that allows us to capture crosslinguistic differences with respect to her- and re- quite nicely. The third reason is that the proposed analysis yields a direct explanation for the paradigm under discussion. Consider attachment of her- to a verb other than a particle verb. Since her- is basically in complementary distribution with the participial ge- prefix and pied-pipes under V2, it is reasonable to assume that her- actually attaches to a lexical receptor of V, call it REC for convenience. This implies that two local steps of head movement are involved: adjunction

of V to her- and incorporation of her- into the appropriate verbal receptor. (I arbitrarily assume that her- is initial in the derivations below; whether it is head-initial or head-final is immaterial to my argument.) The structure in (27c) allows further excorporation of V to T (recall that her- incorporation to V turns V into the head of the projection) and whatever other head positions V is forced to move through. Since her- becomes part of the verbal molecule, the morphological template for participles in standard Dutch is satisfied, and the morphological expletive ge- will not appear on the participle.


Next consider attachment of her- to a particle verb, with the initial structure presented in (28). In (30) her- does not incorporate into a verbal receptor in the overt syntax. In fact, we already know this from the verbal morphology: there is no interaction between her- and ge- in this case (e.g., heruitgegeven (lit. reoutgegiven “republished”). With simple verbs her-


Next, V adjoins to her-, as in (30).28 But now, her- cannot attach to REC, because V does not i-command her-. It is clear that V in (30) cannot undergo any further overt movement. Since P is adjoined to V, V does not i-command her-. Therefore, her- cannot incorporate into V. Since her- does not incorporate, it remains active as the head of the newly formed V. Therefore, V cannot get out of the projection to satisfy the tense morphology and move on to satisfy the V2 requirement: the verb is stuck. The only option therefore is to satisfy the base-generated morphology at LF, that is, to leave the V in situ—an option that is available only in nonroot environments.


and ge- are in complementary distribution, showing that her- incorporates into a verbal receptor. This implies that her- incorporates into a verbal receptor when it can, but is not required to do so. Stress patterns on her- seem to support this: incorporated her- is unstressed, unincorporated her- is stressed.

The distribution of particle verbs prefixed with her- is thus accounted for in the following way. Finite forms of particle verbs prefixed with her- cannot occur in root environments. V is forced to move in root environments to satisfy the V2 requirement; it cannot remain in situ. When V moves in root environments, tense and agreement morphology must be checked in the overt syntax. V, however, cannot escape the projection of her-. V could escape only if her- incorporated. But her- cannot incorporate into V because P is adjoined to V. Movement of the entire verbal complex will not lead to licensing of the finite morphology in the overt syntax either, for the same reason that movement of the entire particle verb or the entire verbal complex does not yield the appropriate configuration for the checking of tense features in the overt syntax. Leaving the verb in situ in nonroot environments should be possible, however. Particle verbs prefixed with her- have finite forms that are restricted to nonroot environments. This raises the question of how the finite morphology is checked at LF. Why can the finite morphology be checked at LF, but not in the overt syntax? At LF, P-incorporation into the P receptor as in (31) clears the way for her- incorporation. In (30) V i-commands her-, which can therefore incorporate into the REC receptor. V is now able to excorporate as in (32), and the finite morphology can be checked in the usual manner.


If this analysis is correct, it predicts that Vs that must incorporate P in the overt syntax into the P receptor should be able to be prefixed with her- and undergo V2. This prediction seems to be borne out (though the examples are slightly degraded). (33)

b.

a.

Zij

Deze mensen herondergaan iedere week dezelfde procedure. these people reundergo every week the same procedure. “These people reundergo the same procedure every week.” heronderzoeken

het

bloed

van

de

patient

iedere


paar dagen. they re-research the blood of the pair days “They reanalyze the blood of the patient every few days.”

patient

every

Thus, by treating her- as a higher head, the theory outlined here explains that Vmovement of verbs prefixed with her- is impossible for overt checking reasons. It also effortlessly accounts for the fact that these verbs have finite forms and therefore lead to well-formed LF representations. The incorporation of the particle at LF allows the incorporation of her- and subsequent excorporation of the molecular verb. 8 Conclusion In this chapter I have presented an analysis for the syntactic distribution in Dutch of particle verbs, verbs with an inseparable P prefix, verbs prefixed with her-, and particle verbs prefixed with her-, based on my proposals in Koopman (1994). In Koopman (1994) I pursue the idea that lexical properties are literally satisfied through head movement, and I try to maintain the theory of head movement in its simplest and strongest form: the way incorporation works follows from the general theory of movement as outlined in Sportiche (1988, 1990). A crucial ingredient of my proposals is the hypothesis that the satisfaction of lexical properties, which I call receptor binding, necessarily requires two local applications of head movement. If receptor binding requires two local applications of head movement, and if the overt syntax can anticipate the covert syntax, then the existence of two different types of complex heads in the overt syntax is predicted. Strong empirical support comes from the existence of both types of incorporation in Dutch. Overt head adjunction, as in Dutch verb-particle constructions, and Dutch V-raising represent the first step toward receptor binding taking place in the overt syntax: verbs prefixed with her-, verbs with an inseparable P prefix, are derived when both steps take place in the overt syntax. The proposals yield a straightforward and general account for the syntactic distribution: they explain what heads can excorporate, what heads must excorporate, and what type of movement complex heads can undergo. Thus, my proposals explain why the verbal part of a particle verb must excorporate, why in other cases of V-movement (such as V-raising) either the entire particle verb or the verbal part may move, and why her-prefixation to a particle verb has the syntactic effect of blocking V2. They also explain why LF movement of a finite verb prefixed with her- is possible. As far as I am aware, no formal account has previously been given for any of these facts; it is achieved by making minimal and extremely general assumptions about the structure of the theory. Particle verbs prefixed with her- do not have finite forms in root environments, yet they do in nonroot environments. This shows that these forms cannot undergo V2, which is obligatory in root environments. Under the assumption that the finite morphology must be checked in the overt syntax when V undergoes V2 (i.e., moves through the functional


projections into the highest functional head position of a root clause), the inability of a particle verb prefixed with her- to undergo V2 follows from the configuration that would result: overt movement of such a particle verb does not allow such checking. Since finite forms of these verbs can occur in nonroot environments, it also follows that the morphology of the finite verb in nonroot environments is not checked in the overt syntax (i.e., by overt V-movement) but rather at LF. Thus, the finite verb in Dutch can remain in situ (or rather lower than T and Agrs) in nonroot environments. Interestingly, as pointed out, there does not seem to be any difference between root and nonroot environments with respect to scrambling, which suggests that scrambling in Dutch is not directly dependent on movement of V. Although much work remains to be done, especially in the domain of morphology (i.e., the internal structure of the molecular verb), this chapter quite strongly supports the general ideas pursued in Koopman (1994), as well as their particular implementation. In fact, as pointed out earlier, these proposals represent the way in which checking of lexical properties is achieved. They can therefore be seen as a particular instantiation of Chomsky’s (1993) checking domain, which is unable to handle the data described here. In particular, Chomsky’s definition of checking domain allows a head to check lexical properties of an adjoined head, irrespective of the structure within this head. The data described here establish the need for structure within the checking domain, and do so minimally. Finally, I should note that the theory of head movement presented here is quite incompatible with Kayne’s (1994). In particular, it is crucial here that the head ccommand an adjoined sister and that a head can excorporate, leaving adjoined elements stranded. In Kayne’s account, the head does not c-command an element adjoined to it, nor is excorporation possible.29 Notes * I would like to thank two anonymous reviewers, and Jeanette Schaeffer and Dominique Sportiche for their comments. 1 See among others Vikner (1994) and references cited there. 2 V-to-I is used to indicate V-movement to some functional head position quite high in the clause. 3 This is true for V-final languages in general. The picture might change if Kayne (1994) is correct, and if there are no underlying final head positions. In that case, V-final languages are either languages in which V is in situ, and complements of V have moved to the left of V, or languages in which V is quite high, and a complement has moved to some specifier position to the left of the head position containing V. I will not entertain this possibility here for Dutch (but see Zwart (1993) for an analysis of Dutch as a head-initial language). It will not affect the basic arguments that I present. 4 In some works, it is proposed that V-to-I movement in embedded contexts is optional; in others, that the finite verb is in situ (Koopman 1984) or undergoes some raising but not to the head-initial functional categories (Koopman 1989, 1990, 1993). 5 Haider reaches the conclusions (a) that the V is in situ in nonroot environments and (b) that I is therefore absent. The latter conclusion does not follow.


6 As discussed in Keyser and Roeper (1992), re- cannot attach to particle verbs in English. In work in progress, I develop an account for the contrast between English and Dutch that exploits the respective absence or presence of P-incorporation in the overt syntax. 7 Inflection of verbs prefixed with her- is awkward in some cases. However, there are many examples that seem perfectly natural. 8 See Broihier et al. (1994) for similar effects in Germanic child language. Jeanette Schaeffer (personal communication) informs me that this distribution resembles the distribution of particle verbs in Dutch child language: particle verbs are disproportionally realized as infinitives, and finite forms in root environments are avoided (25 percent of particle verbs are inflected, as opposed to 50 percent of other verbs). 9 In the few instances in which ge- and her- cooccur, ge- precedes her-, and her- is stressed.

(i)

gewaardeerd *herwaardeerd *hergewaardeerd geherwaardeerd “reappreciated”

10 Verbs prefixed with her- allow ge-prefixation more freely in colloquial Dutch; in these cases ge- prefixes to the compound V. 11 The clausal architecture of Dutch is by no means settled. I assume the structure in (i); only boldfaced heads are important for the discussion in this article.

Nom, Dat, and Acc are the positions where specificity is licensed (Sportiche 1992). Of these head positions, only Wh, C, Nom, Dat, and Acc are visibly head-initial in the overt syntax; whether the other heads are initial or final in the overt syntax is unclear. 12 Although the judgments are not crystal clear, her-P-Vs can be preceded by the infinitival prefix te. Taking V-raising into account, the following pattern seems to arise:

This pattern is consistent with the paradigm described here and argues for base generation of the infinitival prefix (just like base generation of the participial prefix). (ii) is ruled out, because V would have to excorporate in case of V-raising, leaving heruit stranded; as shown in section 8, V cannot escape the projection of her-.


13 For discussion of excorporation, see Roberts (1991) and Koopman (1994). Although Roberts’s account allows excorporation of adjoined heads, the stronger theory presented here disallows such cases. 14 In this definition of antecedent government, it is necessary to refer to categories, so that a Y adjoining to a head antecedent governs its trace. For incorporation, sisterhood is needed: it is important that in a configuration like (i), X0 i-commands Y0 (and Y0 can therefore incorporate to X0) but not Z0.

I have not found a way to unify i-command and antecedent government. Alternatively, following a suggestion by Dominique Sportiche (personal communication), the i-command domain for receptor binding could be called the checking domain, and one could stipulate that checking requires sisterhood. This yields correct results for checking through head movement, but leaves the problem of checking by means of spec-head agreement (the specifier position is not a sister of the head). Percolation of the features of the head to the sister category of the specifier could be invoked to give sisterhood here, but this assumption does not appeal to me. 15 For arguments that particles project a syntactic projection, see Kayne (1985) and den Dikken (1992). That they project a bare PP is argued in Koopman (1991, 1993). Whether particles are base-generated as complex syntactic verbs, as in Johnson (1991), or derived in the syntax through head movement is of little importance to the present theory, as long as the particle is linked to a syntactic PP projection. If P is base-generated as part of a complex verb, this can be achieved by assuming that P still needs to be licensed. Since it is a head, it can only be licensed by being an antecedent of a head, thus forcing the presence of a P0 trace, and a PP projection in the overt syntax. 16 Within the Minimalist framework the question arises as to what drives P-incorporation in the overt syntax. Since there is no obvious morphological motivation for overt movement, syntactic incorporation of P cannot be a literal syntactic operation. Rather, the syntactic structure in (14) must be considered to arise from the lexical representation, in the way discussed in Note 15. 17 I assume that Dutch V-raising is needed irrespective of whether the VP is verb-initial or verbfinal. 18 This assumption will exclude adjunction of a verbal complex to T in general. 19 Apparently, then, there is an asymmetry between receptor binding of lexical categories (P does not incorporate into the P receptor until LF) and that of morphological categories (T must incorporate into the T receptor when the appropriate configuration is met). 20 This is actually quite close to Chomsky’s checking theory. Chomsky (1993) assumes that functional heads that check some property of V have V features. The functional head disappears, or becomes inactive, after it checks the relevant morphology. The system described here provides a precise mechanism by which the head “disappears”: since it incorporates, it is a trace.


21 It is worth pointing out that there would he no problems for T checking in a structure like (17) given Chomsky’s (1993) checking domain. 22 Hoekstra (1988) pioneered small clause analyses for be-Vs, according to which the prefix beoriginates in a small clause and incorporates into V. 23 Although, here again, ge- prefixes more freely in colloquial Dutch. 24 In particular, there are no particle verbs or verbs with an inseparable P prefix that trigger Vraising—that is, that take “bare VP” complements. This gap can be explained quite straightforwardly if there is a syntactic PP projection blocking incorporation of V into the V receptor of V (for more complete discussion, see Koopman (1991, 1994)). 25 The same analysis can possibly be extended to account for the appearance of ge- in other forms, like the nominal forms in (i), where it attaches to the verbal root.

(i)

(al dat) gedoe, geloop, gehuil,… (all that) doing walking crying “all that movement, walking, crying,…”

26 The distribution of ge- resembles the distribution of verb reduplication in nominalized structures in many African languages. For an example in Gbadi, see Koopman (1984:20). 27 In more colloquial Dutch, where ge- can cooccur more easily with P prefixes, geobligatorily precedes P. The difference between standard Dutch and more colloquial varieties of Dutch can be described as a change in the template, whereby ge- becomes a higher head that purely licenses participial morphology and thus does not interact with the molecular structure of V. 28 There is another logical possibility in this structure: V excorporates to her-, stranding the adjoined P; her- incorporates into V; and V further excorporates, yielding incorrect results.

(i) (ii)

*Hij herdeelt de klas in. he redivides the class in *Zij hergeven dit boek uit. they republish the book out

This derivation might not be distinguishable from one in which P has not incorporated. This problem recalls the fact that re- cannot attach to English particle verbs (Keyser and Roeper 1992) and suggests a common explanation for both Dutch (i)-(ii) and the English facts. 29 There are further differences. Kayne derives the fact that right adjunction to a head is impossible. I have assumed adjunction of V to the right of the higher her-/re- head, because V shows up to the right of the prefix. One can always get around this problem by basegenerating the complex verb and left-adjoining it to an empty Asp head. Finally, multiple adjunctions to a head are disallowed in Kayne’s proposals. I had to assume multiple adjunction in the case of her- prefixation to a particle verb ([her [v P [V]]]); this complex acts like a syntactic word and can undergo V-raising.

(i)

omdat hij het boek heruitgegeven heeft/heeft heruitgegeven because he the book reoutgiven has/has reoutgiven “because he has republished the book”


References Broihier, Kevin, Nina Hyams, Kyle Johnson, David Pesetsky, David Poeppel, Jeanette Schaeffer, and Kenneth Wexler (1994) “The acquisition of the Germanic verb particle construction”, paper presented at the 1993 Boston University Conference on Language Development. Chomsky, Noam (1993) “A minimalist program for linguistic theory”, in The View from Building 20: Essays in Linguistics in Honor of Sylvain Bromberger (ed) Kenneth Hale and Samuel Jay Keyser, 1– 52, MIT Press, Cambridge, Mass. Dikken, Marcel den (1992) Particles, Holland Institute of Generative Linguistics, Leiden. Haider, Hubert (1993) “The basic branching parameter”, unpublished manuscript, University of Stuttgart. Hoekstra, Teun (1988) “Small clause results”, Lingua 74:101–139. Johnson, Kyle (1991) “Object positions”, Natural Language and Linguistic Theory 9: 577–637. Kayne, Richard S (1985) “Principles of particle constructions”, in Grammatical representation, Jacqueline Guéron, Hans-Georg Obenauer, and Jean-Yves Pollock (eds) Dordrecht: Foris 101–140. Kayne, Richard S (1994) The Antisymmetry of Syntax, MIT Press, Cambridge, Mass. Keyser, Samuel Jay, and Thomas Roeper (1992) “Re: the abstract clitic hypothesis”, Linguistic Inquiry 23:89–125. Koopman, Hilda (1984) The Syntax of Verbs, Foris, Dordrecht. Koopman, Hilda (1989) “The structure of the Dutch VP”, paper presented at the Summer Syntax Workshop, University of California at Irvine. Koopman, Hilda (1990) “The internal structure of the Dutch VP”, paper presented at MIT, Cambridge, Mass. Koopman, Hilda (1991) “The verb particle construction and the syntax of PPs”, unpublished manuscript, UCLA, Los Angeles, Calif. Koopman, Hilda (1993) “The structure of Dutch PPs”, unpublished manuscript, UCLA, Los Angeles, Calif (Chapter 8 in this volume). Koopman, Hilda (1994) “Licensing heads”, in Verb Movement, David Lightfoot and Norbert Hornstein (eds) 261–296, Cambridge University Press, Cambridge (Chapter 9 in this volume). Riemsdijk, Henk van (1978) A Case Study in Syntactic Markedness: The Binding Nature of prepositional Phrases, Foris, Dordrecht. Rizzi, Luigi, and Ian Roberts (1989) “Complex inversion in French”, Probus 1: 1–30. Roberts, Ian (1991) “Excorporation and minimality”, Linguistic Inquiry 22: 209–218. Sportiche, Dominique (1988) “Conditions on silent categories”, unpublished manuscript, UCLA, Los Angeles, Calif. Sportiche, Dominique (1989) ‘Le mouvement syntaxique: contraintes et paramètres”, Langages 24: 35–80. Sportiche, Dominique (1990) “Movement, agreement, and case”, unpublished manuscript, UCLA, Los Angeles, Calif. Sportiche, Dominique (1992) “Clitic constructions”, unpublished manuscript, UCLA, Los Angeles, Calif. Vikner, Sten (1994) “Finite verb movement in Scandinavian embedded clauses”, in Verb movement, David Lightfoot and Norbert Hornstein (eds) 117–149, Cambridge University Press, Cambridge. Zwart, Jan-Wouter (1993) “Dutch syntax: a minimalist approach”, unpublished doctoral dissertation, University of Groningen.

Part V SPECIFIERS AND HEADS

11 THE SPEC HEAD CONFIGURATION

1 Introduction It became clear in the beginning of the 1990s, that syntax was at the beginning of a new cycle of theoretical progress.1 It had become possible, and necessary, given the analytical options available to the linguist and hence the language learner, to ask general questions about the structure of the theory and theoretical parsimony. In the immediately preMinimalist area, there were a number of ideas about theoretical parsimony, which got overshadowed by the somewhat different answers to the same basic question of theoretical parsimony pursued by Chomksy in the Minimalist Program (Chomsky 1993), and further developed in Bare phrase structure and Chapter 4 (Chomsky 1995). Dominique Sportiche (1992, 1995a, 1995b), for example, proposed that all syntactic relations reduced to two licensing relations: the Spec head licensing relation and the head head licensing relation, which were mediated by movement (overt or covert). In Licensing Heads (Koopman (1994), Chapter 9 in this volume), I pursued the idea that all complement relations reduce to the head movement configuration, and that the binary branching nature of phrase structure itself falls out from the properties of the way in which head movement proceeds. Kayne (1994), from his side, pursued the extremely interesting idea that the relation between hierarchical and linear order is rigidly fixed with linear order corresponding to asymmetric c-command relations (the LCA), and tackled the problem of how to derive the properties of X-bar theory itself. It became possible to ask these theoretical questions because of a better understanding of the empirical phenomena in many human languages. This better understanding in turn was made possible by developments in the mid 1980s: the VP internal subject hypothesis (Koopman and Sportiche (1985, 1991), Speas and Fukui (1986) and many others), the VP internal object hypothesis (objects move to a position outside the VP, Koopman and Sportiche (1985, 1991) and many others), the development of VP shells (Larson 1988), the split I hypothesis (Pollock 1989), in conjunction with V movement, the development of the DP (Abney (1987), Szabolcsi (1987)), in conjunction with head movement within the D (Ritter (1991), Valois (1991)), and so on, the importance of the Spec head relation (Kayne (1989), Koopman (1987), Kinyalolo (1990), Mahajan (1990), Rizzi (1994),


Sportiche (1992) among others).The Minimalist framework is based on the assumption that this understanding is basically correct and complete. However, our understanding of empirical phenomena has continued to grow and as a result has changed since the early 1990s. This is due on the one hand to many careful analyses, independently of Kayne (1994); on the other hand it results from new insights deriving from analyses inspired by Kayne (1994). From these works, it is clear that languages appear to make extensive use of pied-piping of all kinds of constituents. This indicates that the picture that we had in the beginning of the 1990s, which involves articulated structures and quite a lot of movement, was simply incomplete and even more general than previously thought. There is actually more overt movement than previously thought, with pied-piping being the norm rather than the exception: it looks as if everything needs to move (at least once). The question is of course why this should be so. In this chapter, I will be concerned with the type of configurations that one seems to find in languages, taking into account the existence of heavy pied-piping. I will focus on the distribution of overt elements over Spec and head positions, propose a revision of Kayne’s LCA which allows us to derive the doubly filled Comp filter, and develop a proposal which forces massive movement. The proposal that I will develop is much in the spirit of work by Sportiche (1995a) Koopman (1994) and Kayne (1994) and Kayne’s recent work (UCLA class lectures, January 1996). 2 The problem of pied-piping Languages appear to make extensive use of pied-piping of all kinds of constituents, other than DP and PP. This can be shown for languages that would typologically be classified as head initial languages. The arguments for pied-piping are therefore independent of the type of movement that must be assumed if Kayne (1994) is correct in assuming that all languages are underlyingly head initial. (Kayne shows that the standard view that languages have symmetrical structures (i.e. OV is symmetric with VO) does not yield the expected patterns in crosslinguistic variation). 2.1 Pied-piping of constituents within the DP The existence of pied-piping can perhaps be best illustrated within DPs. This kind of piedpiping is robustly found in languages with overt agreement patterns (compare Bantu languages), and in fact also in English and French (Kayne 1994) and Dutch (den Dikken (1998) and Hoekstra (1995) for some extremely strong empirical arguments) to name but a few. I will illustrate this property for Bantu languages. In Bantu languages, the “genitive” DP agrees in class and number features with the head N. A socalled “associative” marker a carries the agreement and precedes the genitive DP.2 This pattern of agreement can be reduced to the familiar Spec head agreement relation if it is assumed that a phrasal

THE SPEC HEAD CONFIGURATION 325

constituent containing the head N pied-pipes to the left of the associative marker, triggering the usual Spec head agreement. (Compare Nkemnji (1995) Moritz (1993)):

Preposing of a phrasal constituent containing the head N, triggers agreement on the associative head as a regular Spec head relation, and derives the correct word order patterns. The derivation in (1) is by no means an isolated case where we need to assume phrasal movement of some XP constituent to the left. Many other cases are found (Koopman 1993, 1994, Androutsoupoulou 1994). And of course, Kayne (1994) proposes similar derivations for a picture of John’s, and a host of other structures in English and Romance languages:

Pied-piping analyses are successfully applied to certain Dutch DPs (den Dikken 1998) and find strong empirical support (Hoekstra 1995). Pied-piping is not restricted to DP internal structure, but holds quite generally for all projections. I briefly discuss two more examples. 2.2 Pied-piping of IP Sportiche (1995a) argues that French intonational questions like the following (3)

Tu vas venir demain?

involve a derivation where IP pied-pipes to the Spec of the Q morpheme, whichseems to occur clause finally as sentence final rising intonation:

Although this analysis might seem less obviously correct,—after all the Q head is only intonationally present, and thus not an overt terminal element—Sportiche gives some extremely interesting empirical support for this analysis, which concerns the licensing of negative polarity items. While head initial yes/no questions in French license negative


polarity items (as do negative environments), intonational yes/no questions do not (as do statements): (5)

a. b.

A-t-il-vu Has-he-seen *I1 a He has

vu seen

qui que ce soit? anyone? qui que ce soit? anyone?

This quite surprising fact receives an immediate explanation if the structure is as in (4): clausal pied-piping removes the negative polarity item from the c-command domain of the Q head. Intonational questions therefore cannot contain a negative polarity item. 2.3 Pied-piping of the complement of Neg Nkemnji (1995) argues for phrasal pied-piping of the complement of NegP to Spec, NegP in Nweh, a Grassfield Bantu language spoken in Cameroon. Nweh looks like a regular head initial language. Clausal negation is expressed in two places: by a marker preceding the verb te and by a clause final element (b ). (6)

Njikem N

a Agr

ke? P-1

te Neg

pfεt eat

akend η plantains

b Neg

At first sight, it looks as if te equals French ne and b equals French pas. It turns out however that the initial element te has the same properties as pas, while the final element b has the properties of a head and is thus comparable to ne (b alternates with the verb in certain configurations, as in the example below). (7)

Njikem N

a Agr

ke? P-1

te Neg

akendo η banana

pfεt eat

Nkemnji convincingly argues for an analysis which treats b as the head of NegP, with the complement marked by te moving to Spec, NegP:

What is interesting here is that the pied-piped complement is extremely “heavy”, containing clausal complements, and adjuncts.3


This is expected, given the fact that it is some high constituent that moves. 2.4 Two configurations In conclusion, then, languages make use of overt pied-piping of constituents. Some of these constituents are surprisingly big and clause-like. Besides pied-piping (movement of some XP to some Spec position), a different configuration is well-attested: languages make use of head movement (movement of some head to another head position). If I were to pursue the discussion for each category (some of which I will be doing below), and present a complete “inventory” of what one finds and does not find in human languages, I believe that the following two configurations would emerge:4 (10)

a.

[ [Y [X]…[x] . .] ]

b.

[ [XP] [Y […[XP] ] ] ]

Head movement to some head position Movement of some XP to SPEC, YP

The overall picture then is the following: besides the familiar pattern in which languages raise a head to some high position, languages also make use of phrasal pied-piping by which some constituent ends up in a very high Spec position, possibly pied-piping entire clauses or IP-like constituents. This raises the question how widespread pied-piping really is. I suspect that rather than being marginal or isolated, pied-piping of predicates and clausal constituents occurs extremely frequently in human languages. This raises a number of interesting questions. (11)

a.

–If pied-piping is so heavily used, why did it take us (linguists)


b. c.

so long to discover it? –Why precisely do we get the two patterns in (10)? –What motivates pied-piping?

As for the last question, Chomsky’s feature driven theory of movement does not seem to yield much insight into the problem of the motivation of pied-piping. This is to be expected, given that the standard assumption about what kinds of movements one finds and does not find simply do not include the type of pied-piping of heavy constituents that natural languages instantiate. A different kind of answer is therefore called for. I will try to provide some insights into these questions by examining the Spec head relation, in particular by examining the distribution of overt and covert terminal elements over the Spec and head positions. Before doing so, I briefly spell out my background assumptions: • Syntactic structures are binary branching (Kayne 1994).

• There are no multiple Spec positions. • Adjuncts are accommodated within this structure (as in Sportiche (1994)). • There is no variation in hierarchical structure between languages. This implies that there can be no parameters distinguishing languages as having a high negation or a low negation, or as having a high focus or a low focus. I take the basic ordering to be given, and will not address the interesting research question how it arises. Linear order in languages arises from movement operating on the underlying skeleton. Here the assumptions that play a role in this chapter are: • head movement (left adjunction only) • XP movement (leftward only) • LCA (The Linear Correspondance Axiom, Kayne (1994), or some version thereof). The LCA is based on the idea that the relation between hierarchical structure and linear order is rigidly fixed, with asymmetric c-command mapping into linear precedence.


3 The spec head relation Given the existence of extensive pied-piping, and head movement, and large underlying structures, analyses are often underdetermined. As linguists we encounter the problem that too many options seem to be available. (Developments in the 1990s can all be seen as reducing these options.) We can state the problem from the point of view of the language learner: how does the language learner know where some phonologically overt element is pronounced? He/she knows that there is a hierarchical order of projections, that this structure is quite substantial, with each feature corresponding basically to a projection, and that extensive movement has taken place. The language learner gets much help from Kayne’s (1994) Linear Correspondence Axiom LCA and knows that linear order corresponds to asymmetric c-command relations. I will start with a suggestion that a slightly modified LCA further restricts possible analyses, and thus reduces the space of possible hypotheses entertained by the language learner. This proposal will lead me to a further examination of the type of configurations that are expected to arise. 3.1 The LCA and deriving the doubly filled Comp filter In his 1994 book, Kayne encounters the problem of how Spec positions are possible at all under the LCA. This problem can be illustrated in the following structure:

In this structure, X’/XP (the actual label of this node is not important for the present discussion) asymmetrically c-commands Y, and therefore x should precede y. However, YP also asymmetrically c-commands X, which means that y should precede x. This configuration cannot therefore arise. In order to allow for this structure, Kayne proposes to restrict c-command to categories. This removes X’/XP from the set because it is a segment. This leaves YP asymmetrically c-commanding X, and y therefore precedes x. For reasons having to do with the theory of head movement (Koopman 1994), I wanted to keep segments as c-commanders. I therefore explored a different solution to this problem. Suppose that we let X’/XP participate in c-command. Consider the configuration in (13) again. This configuration says that there can be no terminal elements in the Spec position of a projection, and in the head position. But it has often been proposed that this is correct, and there is even a filter which prohibits precisely


this configuration: the doubly filled Comp filter, generalized in Sportiche (1992, 1993) to the doubly filled Voice filter, and generalized to the Spec head configuration in Koopman (1991, 1993). The doubly filled Comp filter itself is as yet unaccounted for. The structure in (13) provides a simple way to derive the doubly filled Comp filter from the LC A, provided we find a way to allow for the configuration. One way to achieve this is to restrict the application of the LCA to overt terminal elements only (see also Chomsky (1995: Chapter 4)). I will call this version of the LCA the modified LCA: (14)

Segments participate in c-command.5 Modified LCA: the linear order of overt terminal elements corresponds to asymmetric c-command. 3.2 Spec head configurations

The modified LCA allows for the configuration in (13) iff one of the terminal elements is covert. Let us examine what the modified LCA has to say about the different Spec head configurations, distributing overt (boldfaced) and covert (italics) elements over Spec and head positions. Within covert categories, we further distinguish between traces (covert categories which are not the head of a chain) and covert heads of chains. This will play a role in section 3.6. It follows from the modified LCA that no projections can have terminal elements in Spec and in the head (15a). The doubly filled Comp filter is one particular example of this configuration. According to (15a), however, it should be entirely general and apply to all projections. This is the Double Voice filter of Sportiche (1992) for clitic constructions and Romance negative constructions, and the generalized doubly filled Comp filter, used in Koopman (1991) and (1993) for the structure of pronouns. Many linguists (including myself, Koopman (1983, 1993)) have taken the doubly filled Comp filter to be some type of economy principle which can be violated in certain cases. If the proposal above is correct, however, this simply cannot be true: the configuration can never occur, because there could be no linear ordering. This raises the question what to do about violations of the doubly filled Comp filter. I will argue for alternative analyses of these violations in section 3.3. Configurations that are allowed by the LCA are presented in (15b) and (15c). These in fact represent the expected cases: projections can have terminal elements in Spec. This yields so-called Spec marking languages (15b). Projections can have a terminal element in the head position. This yields so-called Head marking languages (15c). Languages can (and presumably must) exhibit any mixture of these two patterns, as long as neither Spec nor head projections end up containing overt material. It is important to point out that the question of whether (15a) is inviolable is in fact logically distinct from the expectation that languages conform to pattern (15b) or (15c). This expectation is still true if the generalized doubly filled Comp filter were an economy principle. (15d) represents the fourth and last possibility. Projections can have a covert Spec and covert head (15d), with further distinctions between types of covertness (is the covert element the head of a chain


or is it not). The modified LCA has nothing to say about this configuration, since it does not apply to covert elements. In 3.6, I pursue the idea that this configuration is actually excluded if both the Spec position and the Head position are the head of a chain.

This chapter then contains three interrelated, but logically distinct proposals. 1 The LCA applies only to terminal elements, and c-command is not restricted to categories. This proposal derives the doubly filled Comp filter. I briefly examine violations of the doubly filled Comp filter, and argue for a general solution for these cases. 2 Languages are expected to be have overt heads with silent Specs or silent Specs with overt heads. This expectation is still valid independent of the question of whether the doubly filled Comp filter derives from the LCA, or whether it is some economy principle. I turn to cases that fit the configurations well, and show how patterns which have been felt to be problematic so far, in fact represent the norm. I also discuss some cases where the actual surface word orders derive from leftward movement to Spec and the presence of a head. I argue that two phenomena fall under this heading: head drop and pro-drop. 3 Languages disallow projections to be headed by silent covert heads and Specs (15d). In particular, I will propose a principle that requires a projection to be associated with lexical material at some point in the derivation.6 The idea here is quite simple: projections must be activated to be semantically interpretable. Projection Activation


happens by associating overt lexical material to either Spec or head at some point in the derivation. This principle provides the motivation for pied-piping, and derives the two configurations in (10). 3.2.1 Why did it take us so long to discover the extent of pied-piping? I am now ready to provide an answer to the first question that we raised in (11a): (11) a. If pied-piping is so heavily used, why did it take us (linguists) so long to discover it? Pied-piping is XP movement to a Spec position; it is quite easy to see if some small DP or PP has moved to some Spec position as in Whose mother did you visit? (although it might not be immediately obvious that the head position of that projection is empty (compare 3.3.3)). The DP can occur in other positions within the clause. This type of pied-piping has of course been part of syntactic theory since the very beginning. Pied-piping of bigger constituents is simply much harder to detect, even if the bigger constituent occurs in different positions (as is for example the case with French yes/no questions). An overtly pied-piped element in Spec position forces the head of that projection to be silent, because it creates the configuration in (15b). The effects of pied-piping are therefore often hard to detect. 3.3 Problematic cases: doubly filled Comp phenomena The doubly filled Comp filter was formulated in Chomsky and Lasnik (1977) to account for the impossibility of both an overt wh-phrase and an overt C, and was instrumental in accounting for the form of relative clauses: (16)

*I wonder [who that/if] left *the man [who that John saw]

It is well known that the doubly filled Comp filter does not seem to hold universally. There are some languages which allow for more than one overt element in the C domain (Dutch, Quebec French, and so on). Even within English the doubly filled Comp filter does not seem to hold for all types of Cs. For example, no doubly filled C filter violation arises in root wh-questions with non-subject wh-phrases and inverted I (Who did you see?). The crosslinguistic variability has been taken by many to show that the doubly filled Comp filter can be violated. In that respect, it is more like an economy principle (“avoid lexical material”) that languages could choose to disobey. If the doubly filled Comp filter derives from the LCA, as proposed above, this simply cannot be true. The question then arises if we can find alternative analyses for doubly filled Comp filter violations, and how plausible these analyses are. The remainder of this section examines some violations of the doubly


filled Comp filter, and sketches analyses for these that are not only compatible with the LCA, but are in fact necessary independent of the LCA. I will discuss V-second phenomena in Dutch, doubly and triply filled Comps in Dutch, turn to apparent violations in English, and finally discuss verb movement in tensed clauses in French. 3.3.1 Verb second Consider the verb second phenomena, illustrated in the examples below: (17)

(i) (ii)

Jan komt morgen John comes tomorrow Morgen komt Jan Tomorrow comes John “Tomorrow, John comes”

(Dutch)

Two ingredients go into the standard analyses of verb second phenomena. First, some XP constituent pied-pipes to an initial Spec position in root clauses (say, Spec, TopicP). And second, the finite verb moves to the highest head position (say, Top°). As we can see in the following structure, this yields a doubly filled Comp effect:

My proposal makes this analysis unavailable, since the phrasal projection XP and the finite verb cannot be within the same projection. This problem might seem serious enough to simply abandon the proposal that the modified LCA derives the generalized doubly filled Comp effect. Yet, I do not think that this is necessary or even desirable. There is an alternative analysis which is consistent with the LCA: the initial XP is to the left of the preposed V, say in Spec, TopP. Since the Spec is overt, the head of this projection must be empty. The finite verb therefore occupies the head position of a lower projection, which by the LCA must have a silent Spec: (19)

XP Jan Morgen

[Y e

[

Spec e

[Vf komt komt

This proposal therefore forces the presence of more (rather than less) structure in the C domain: it forces a split C approach. This might seem ad hoc. However, quite independently from my proposal, there is mounting evidence that we need to assume a


split C approach anyway (Bhatt and Yoon (1992), Sportiche (1995a), Rizzi (1995), Nkemnji (1995), Hallman (1996)). Moreover, my proposal leaves the necessary space to account for the fact that the V-second phenomenon in Germanic languages is not a unified phenomenon, and that there is quite a bit of variation in the initial domain. So, the argument is that it is not only desirable, but also unavoidable to assume more structure. If there is more structure, then one would expect the situation in (19) to arise. In addition, we expect the modified LCA to further restrict possible analyses that can be entertained by the language learner: one therefore expects to find some systematic patterns in the seemingly chaotic patterns of violations of the doubly filled Comp filter that languages seem to exhibit. I turn to this problem now. 3.3.2 Doubly filled Comp in embedded clauses In embedded wh-questions in Dutch, overt wh-phrases and complementizers may (but do not need to) cooccur: (20)

a.

Ik I b. Ik I c. Ik I d. Ik I

vraag me af wonder vraag me af wonder vraag me af wonder vraag me af wonder

wie who wie who wie who wie who

of if of if er there dat that

dat er morgen komt that there tomorrow comes er morgen komt there tomorrow comes morgen komt tomorrow comes er morgen komt there tomorrow comes

These again cannot be analyzed as occupying the Spec position and the head position of the same C projection. Rather, an analysis is forced which has a separate WH projection, hosting the wh-phrase, a Q projection, hosting the head of (of typically occurs in +Q) and a declarative C projection hosting the head dat: (21)

[XP wie

[WH[ (cov) (cov)

[Q[ of

(cov)

[C[… dat…

(=(20a))

Since I am assuming that languages do not vary structurally, I will assume that the order of projections is universally as in (21). We must further account for the fact that in all Dutch dialects, these complementizers can be absent. If the language learner encounters an empty C, s/he might assume the C is empty because it occurs in the head position of a projection that hosts an overt Spec (i.e. the configuration (15a) obtains). The absence of further overt C heads can be accounted for by assuming that the other C heads are traces. This yields the following representations for (20c) and (20b) respectively:


If Dutch makes a distinction between a Wh projection and a Q projection, and if Dutch reflects the underlying hierarchical order of these projections, all languages, including English, should have these projections in this order. If there is more structure in the C domain, the question arises why English does not behave in the same way as Dutch. I turn to these questions in the next section. 3.3.3 Doubly filled Comp in English Let us next turn to English. As pointed out above, the doubly filled Comp seems to hold in standard English embedded Cs, but not in root wh-questions, where a non-subject has undergone wh-movement. The problem of matrix wh-questions immediately disappears given the structure that we have determined for Dutch. In matrix wh-questions, the whword occurs in Spec, WhP and the inverted I in Q: (25)

[XP who

[WH[ (cov) (cov)

[Q[ did

[C[ (cov)

There is no doubly filled Comp filter violation, because the wh-phrase is in Spec of the Wh-projection, and the auxiliary is in Q (or in C). What about embedded wh-questions in English? In embedded questions with overt wh-phrases, that if and whether are silent. I assume that English is like Dutch (21) with the Wh head silent, heading a chain in Q and C.

Note that the Q head, which must be lexical in English (it is strong and therefore attracts I) also counts as “lexical” in (26), since it is a trace of the raised Q (it is not the head of a chain, and therefore does not fall under (15d)).


There are dialects of English which allow different patterns. Belfast English, for example, seems to exhibit the following paradigm. This paradigm is based on Alison Henry (1995): (27)

a. b. c.

I wonder which dish that they picked *I wonder which dish whether they picked *I wonder which dish if they picked

This can be analyzed quite straightforwardly. Contrary to standard English, that can remain in C and fail to raise (27a) in Belfast English. Belfast English is like Dutch (20d) in this respect. Belfast English also differs from Dutch however: if cannot remain in Q, but must obligatorily merge with the +WH head. This accounts for the impossible cooccurrence of if and a wh-phrase in (27c). Whether is an element which must be associated with the Wh projection; whether will not be allowed to surface in this structure, either because it is a wh-phrase in Spec, WhP (and this position is therefore unavailable for whether), or more correctly, I think, because whether occurs in the head position of the Wh projection, (with a silent wh-phrase in Spec which is responsible for the movement characteristics associated to whether (Larson 1985) and the LCA disallows lexical material in both Spec and head positions. (For an argument that whether acts as a head, not as a whphrase, see the discussion on inversion in embedded questions below.) Therefore, no English dialects should allow wh-phrases and whether to cooccur. So far, then, the only difference between standard English and Belfast English is that that is allowed to stay in situ:

There is an additional difference between Belfast English and “standard” English or Dutch, and this concerns the possibility of inversion in embedded wh-questions. The following paradigm holds for Belfast English (Alison Henry 1995), who shows that these whquestions are true embeddings. (Inversion is optional, and non-inverted indirect whquestions are fine as well.) (29)

a. b. c. d.

I wonder which dish had they picked *I wonder whether had they picked the dish *I wonder if had they picked the dish She wonders had she picked the dish

Thus overt whether and if always block inversion. But when whether and if are covert, inversion is possible. Overt whether and if block inversion because they occupy the Q head at some stage of the derivation, and the Q morpheme is thus satisfied. It is interesting that whether patterns with if in this respect, and not with wh-phrases. This shows that whether is a portmanteau head which is both +Q and +Wh. Inversion is only possible, if the Q head


is truly empty: that is, if the Q morpheme is allowed to be absent from the initial array. When this happens, I to Q movement becomes obligatory.

This brief discussion is intended to make clear that the split C approach is in fact necessary if we simply want to address the problem of apparently doubly filled C phenomena and make sense of the crosslinguistic and dialectal variation. It is also clear that the modified LCA severely restricts possible hypotheses that the language learner can entertain. It is clear, then, what kinds of solutions are compatible with the revised LCA: we need to assume a more complex structure that provides enough heads and Spec positions. This conclusion is in agreement with Kayne (1994), Sportiche (1995a), Koopman (1994) and logically continues work on split I (Pollock 1989) and many others (Bhatt and Yoon (1992), Rizzi (1995), Nkemnji (1995), Hallman (1996)) but seems incompatible with the attempt to reduce the number of head positions, and make use of multiple Spec positions, as in Chomsky’s Chapter 4 (Chomsky 1995a). Chomsky’s proposal encounters the problem that many languages actually instantiate these head positions overtly: this seems to be excellent evidence that these head positions indeed exist. It is also clear that we should avoid postulating more structure, because this allows us to derive the doubly filled Comp filter (but note that if UG says that you cannot have both Spec and head overt, the language learner will be simply forced to do so, even if there would be no other supporting empirical evidence whatsoever). It is quite clear where to look for supporting empirical evidence. For example, the structure of split Cs should provide us with the means to account for the various and quite complex variations that one finds between languages with respect to the C domain. It is quite clear for example, that V-second is not a unified phenomenon, and that verb second languages vary from each other with respect to the initial field. Empirical support for this claim then derives from the insights that this structure provides in crosslinguistic differences with respect to the initial field. 3.3.4 The distribution of subjects and tensed Vs in French Apparent doubly filled Comp violations also occur in clauses in French, if we consider the distribution of French subject DPs and the finite verb. According to the standard analysis the subject DP occurs in the Spec position of the projection containing the finite verb, say, Spec, AgrS: (31)

[AgrSP [Marie] [AgrS adore] […[…camper “Marie loves camping”

As in the V second case, we must conclude that this is not the correct analysis: the subject DP is not in the Spec position containing the finite verb, but rather occurs in some higher Spec position:


(32)

[YP [Marie] [Y… [AgrSP [AgrS adore] [… [… camper “Marie loves camping”

This conclusion is reached in Sportiche (1992) (1995a) and Kayne (1994) for completely different reasons: clitic pronouns and negation intervene between the subject and the finite verb. These clitics are argued to occupy either the head or the Spec position of their own projections, which forces the subject DP even higher (Sportiche argues they are in NomP (nominative P) which is distinct from the position in which Agr is triggered (AgrSP)). If this is correct, the finite verb is never in the head position of the projection that contains the subject DP. In sum, it is clear what the general solution for this class of problems is: one must assume more structure, and support this structure with both language internal and crosslinguistic evidence. 3.4 Good cases: head marking and Spec marking 3.4.1 Where the head is overt and the Spec is silent The modified LCA allows heads of projections to be overt, but their Spec positions to be silent. Many configurations that are generally thought of as somewhat exceptional, are actually completely transparent in this view. I discuss several such cases in this section, starting the discussion with cases where the head is overt, and the Spec is silent. I will consider English and Dutch yes/no questions, Dutch imperatives, and VSO languages. English yes/no questions exhibit either I to C movement in root clauses or if/whether in embedded clauses. It is generally assumed that a silent operator occurs in Spec of the +Q projection. This yields the following representation. (See the preceding section for the particular conclusions about the position which whether occupies.) (33) V first (yes/no questions), with V to C movement (main clauses or if/whether) [Q Vf] [… (did John leave) a. [QP OP [if[ (if John left) b. [QP [QP [e][ (whether John left) [whP whether These configurations conform to the modified LCA: since the head of the projection is overt, the Spec position is silent. Dutch yes/no questions operate in much the same way. In root yes/no questions, the finite V moves to the +Q head. Since the V occurs in the +Q head, Spec, QP is silent. In embedded yes/no questions, the +Q head is realized as a lexical complementizer (of) that may optionally be followed by a declarative C dat: (34)

[OP

[Q[

[C[


(cov)

of

(cov)

dat…

A new problem arises with respect to root yes/no questions, which is in fact a very old problem. Given that Dutch is a V-second language, why are these structures V first and not V second? In section 3.3.1, we have seen that V second arises by obligatorily moving some XP to a projection higher than the one containing the finite verb. Note that this is also possible in (root) yes/no questions, as expected, given the analysis: (35)

Met With

Marie Marie

heb have

je you

ook also

gepraat? talked?

It is however not necessary. The question of why these structures may be V-first can be rephrased in terms of the availability of the higher projection or not. If the projection is unavailable, V-first arises. If it is available, V-second arises. What then determines the availability of the higher projection? Here there are two possibilities. Either the higher projection may simply be absent from the structure, or perhaps more interestingly, the projection must always be there, but the entire QP has pied-piped to the Spec of the projection. If the QP is in the highest Spec position, the head of this position must be silent by the modified LCA:

Imperatives also conform to the pattern that we expect. In Dutch, for example, imperatives involve V movement to a C-type position above the subject. The resulting string is V-initial. This pattern is compatible with an analysis that postulates an empty operator in the Spec position of the imperative head: (37)

Imperatives (V-second:) (Dutch) a. Geef (jij) dat maar terug Give you that only back! b. [OP [IMP Vf [jij…

There is of course much more to be said about imperatives, but the main point is clear. Imperatives are verb first because the head occurs in the imperative head position, and the Spec of this projection must therefore be silent. 3.4.2 VSO languages Consider next VSO languages, like Irish, and Welsh. In tensed clauses in these languages, the finite verb is preceded by a number of heads that express in linear order clause type complementizers: negation, the finite V, followed by tense and Agr morphemes, followed by the lexical subject. The order of the preverbal heads corresponds to the basic


hierarchical order. Since head movement can only be left adjunction, it follows that these elements occur in the head position of distinct projections. This yields the following structure, which assumes the standard analysis for the subject DP: it is outside the lower VP, in Spec, TP (or maybe in some lower functional projection, Koopman and Sportiche (1991) McCloskey (1994)): Projections containing overt head positions, cannot have their Spec positions filled with overt lexical material. This yields strictly head initial structures. The lexical subject can be in Spec, TP because T is covert: T has incorporated into Agr. This structure directly accounts for a salient property of Celtic languages, stated below:

(39)

Pro triggers full agreement. Lexical DPs do not trigger full agreement.

Lexical DPs fail to trigger agreement, simply because they cannot reach the position in which Agr is triggered. They cannot reach this position because AgrS is overt (and therefore Spec, AgrS must be silent). Pro on the other hand is covert; since it is covert it can reach Spec, AgrS. Hence full agreement is possible with pro.7,8 Comparing Irish with a verb second languages like Dutch raises a number of interesting questions. If declarative root clauses in Dutch contain a high topic like position that relates to discourse, and if languages do not vary with respect to the initial structure, what happens to this projection in Irish? Why does Irish not allow some XP in root clauses to go with the initial Spec, Topic position, just as Dutch does? Since this does not seem to be possible, it must mean that the initial position is unavailable in Irish. But if it is available in Dutch, it should be in principle available in Irish as well: we cannot accept an answer that simply says that this projection is unavailable in Irish. I would like to pursue the idea that no Topic XP can appear to the left of the verbal complex in declarative root clauses, because the entire clause has pied-piped and is occupying this position.


Under this view, the difference between Irish and Dutch reduces to a difference in the category that pied-pipes, with clausal pied-piping in Irish, and pied-piping of some smaller constituent in Dutch.9 3.4.3 Where the Spec is overt and the head is silent We have seen many examples of this configuration. As soon as a language learner has a reason to analyze an XP containing overt material as occupying a Spec position, s/he also automatically knows that the head position of that projection must be covert. This configuration is compatible with an analysis of topicalization in English which involves an empty topic head. 3.5 On interactions between heads and Spec marking Let us next turn to interactions between head and Spec marking. Consider a language with the following characteristics: (41) (i) There is a lexical item in the head position (say, negation, focus, Agr). (ii) There is independent leftward XP movement to the Spec of this projection under certain conditions. These two properties can conspire to yield a forbidden configuration, with both Spec and head containing overt material. Several things can happen. Either the head moves higher (yielding VO structures), or the Spec moves higher (yielding scrambling). Alternatively either the head or the Spec is forced to be silent in such configurations (head drop or Spec drop). I restrict the discussion to head and Spec drop, and discuss the disappearance of negative heads as a particular example of head drop, and pro-drop as a particular case of Spec drop.


3.5.1 Where heads disappear 3.5.1.1 NEGATION In certain Romance languages the negative head disappears but only when a negative XP precedes the finite verb and the clitics. This is illustrated for the Italian in the following example: (42)

a.

b. c.

nessuno e venuto /*nessuno non ha visto Mario Nobody is come “Nobody has come.” Non ho visto nessuno Not have seen nobody Niente ho detto Nothing, I have said

Based on the fact that only one negative XP may occur pre-verbally, which points to a unique landing site of this constituent, the standard analysis has the preverbal negative XP occupying Spec, NegP. If this is correct, we can now understand the disappearance of the negative head as forced by the revised LCA. (This analysis is identical to Sportiche’s (1992) proposal to treat the disappearance of the negative head as a doubly filled Comp effect (his doubly filled Voice filter).) Italian seems to allow the two configurations predicted by the LCA, with either the head or the Spec overt, but not both: (43)

a. b. c.

[ [NegP NegXP *[NegP NegXP

[non[ [e][ [non[

The LCA proposal makes very specific predictions; it only blocks (43c), if NegXP occurs in Spec, NegP. This means that no lexical material should be able to intervene between the negative XP and the position in which non should surface (i.e. the position preceding the accusative clitics). If the NegXP were able to reach a higher Spec position, the negative head should be able to reappear. There are indications, both internal to Italian and across Romance languages that indicate that this is basically correct, although the matter deserves more careful investigation. In Italian, the examples cited in the literature where a non-negative negative XP precedes the finite verb are always of the form in (42c), in which pro-drop has applied. It is highly suggestive that overt pronouns do not seem to be able to intervene between the negative XP and the finite verb: (44)

a.

Niente ho detto Nothing, I have said


b.

*Niente io ho detto Nothing, I have said

This seems to indicate that the analysis above is on the right track, although there are also indications that the point to a greater complexity.10 Variation across Romance languages seems to indicate that this prediction is basically correct however. Some languages (Catalan) allow preverbal negation to be absent or present with a preposed negative XP. This would be consistent with optional further movement of the Negative XP. Second, Susan Garrett (1995) presents a nice historical argument in favour of this view on the basis of the history of negation in Spanish. She shows that twelfth-sixteenth-century Spanish allowed the negative head to cooccur with an initial negative XP or to drop. She further shows that the obligatory disappearance of the negative head from the sixteenth century onward correlates with the disappearance of an initial Spec position higher than negation. This pattern would be entirely expected: (45)

Spanish (twelfth to [NegXP [NegXP[Neg […

a. b.

sixteenth century) Y[NegXP[Neg[

Depending on whether the NegXP occurs in Spec NegP or higher, the negative head is forced to be absent or present. When the higher YP in (45a) became unavailable, negative head drop became obligatory with preverbal negative XP: (46)

[NegXP [Neg [ 3.5.2 Where Spec disappears (dependent on configurations)

The mirror image also occurs, where it is not the head that drops, but rather the Spec. This Spec drop again is configurationally determined. I discuss in turn topic drop in Dutch and pro-drop in general. 3.5.2.1 TOPIC DROP (DUTCH) Dutch and German allow D-type pronominals to be dropped in root clauses in which Vsecond has applied: (47)

a.

[pro

b.

*ik I

[ weet [ik niet Know I not weet know

pro niet not


It is well known that the pronoun can only be dropped in a particular structural configuration. Given our discussion so far, we have a logical candidate: the pronoun is dropped in the Spec position of the projection containing the finite verb: (48)

[pro

[Vf[

It is often assumed that topic drop is optional, and that the same structure underlies (47) and (49): (49)

a.

dat that

weet know

ik I

niet not

Again, this analysis is unavailable, because of the revised LCA: the overt d-pronoun cannot be in the Spec position of the projection that contains the finite verb. Rather, the situation must be assumed to be as in (50): (50)

a. b.

[pro [ei

[datie

[Vf [Vf

There is in fact quite interesting evidence that there are two distinct projections: some Dutch constructions, like imperatives, allow for topic drop, but not for overt topicalization of the d-pronoun.11 3.5.2.2 PRO-DROP Dutch and German show extremely clear cases of configurationally determined pro-drop. It has been argued in the literature that Chinese object pro is licensed in a particular structural configuration as well (Cheng 1989). This raises the possibility that pro-drop in general falls under this generalization (compare Koopman (1993) for some discussion). Pro-drop is forced by the LCA if there is an overt element in the head position of that projection. (51)

[pro

[X

Can (51) be strenghtened to (52)? (52)

pro-drop iff there is an overt element in the head position

There certainly should be cases of pro-drop that fall under this general configuration. It is less clear that all cases of pro-drop can be brought back to the configuration in (51), although this is certainly an idea that I would like to explore more seriously in future work. Here are some problems. The LCA only has something to say about head or Specdrop in cases where both would be lexical: if the spec is silenced pro-drop occurs. The


LCA has nothing to say about whether pro is allowed in Spec position with a covert head. I return to this general type of configuration below. Furthermore, any account of pro-drop needs to be able to capture the relation between “rich agreement” and pro-drop: in many cases, there must be rich agreement, i.e. there must be an overt head to “identify” pro. In many languages, some tenses will license pro-drop, others will not. If the head is not “strong” enough, Spec must win, i.e. the pronoun must be overt. One way to make this consistent with (51) is to say that rich agreement licenses pro-drop, because the agreement head is occupying the head position of the projection in which pro is licensed. With “poor” agreement, the head simply does not end up in the right structural configuration, but stays lower. This again presupposes a more articulate structure, with basically split agreement heads (person, number and gender). Independent support for this comes from the fact that languages do not treat all pronouns as being equal. There are often differences in the distribution of first, second and third person pronouns, and singular and plural pronouns. Besides this type of solution, there could still be another different way to derive pro-drop: pro-drop could arise when overt pronouns and a bigger XP compete for the same Spec position:

If a bigger XP occupies the Spec position of the relevant agreement projection, the head of this projection will need to be silent (by the LCA). The pronoun needs to be silent as well, hence the appearance of pro-drop, since the position that allows it to be overt is not available. Such an account possibly extends to pro-drop in imperatives.

3.6 Covert categories: traces and heads of chains One configuration remains to be examined: the configuration where both Spec and head are covert. Here we must distinguish between several possibilities: the covert category is a trace (e.g. not the head of a chain), or covert HC (a covert element heading a chain). This yields four possibilities:


It is clear that the configuration in (54a) seems to arise quite often. In fact, it must be allowed: in many cases the thematic domain of V only seems to consist of traces, and everything has moved out higher. I will simply assume that (54b) and (54c) should be allowed, and will not discuss these configurations any further. Instead, I would like to concentrate on the configuration in (54d): can both the Spec and the head of a projection be covert heads of chains? Certainly standard practice allows this. For example, English declarative clauses are often considered to have a silent declarative C and a silent Spec as well. This silent projection thus would contain both a silent head of chain in its Spec, and a silent head of chain in the head position. I would like to explore the idea that this configuration is actually not allowed. (55) No projection can have a silent head of chain in Spec position and in the head position. Why would this configuration not be allowed? What is wrong with this configuration? In every day practice, silent heads of chains are standardly assumed and contribute to the semantic interpretation. But this raises a general question: why do there appear to be no languages in which all heads and Spec are silent, and in which everything simply remains in the thematic domain of the VP? Why does syntax make use of such elaborate movements? Why does heavy pied-piping exist? Why do we find the two configurations (spec are overt or heads are overt). Suppose that what is wrong with (55) is that these projections cannot be interpreted. In other words, suppose that for Full Interpretation to be satisfied the following must hold:12 (56)

A projection is interpretable iff it is activated by lexical material.


Each projection must be “activated”, i.e. lexical material must be associated with either the Spec or the head of a projection in the course of the derivation. It is quite easy to see that this proposal will force massive overt movement. This quite simple proposal has exciting and far-reaching consequences, and allows us to sketch some answers to the questions that I raised above. I will briefly discuss these below, but turn first to some analytical consequences of the proposal. 3.6.1 English main clauses An old and yet unsettled problem concerns the analysis of English main clauses. English main clauses look to be no bigger than IP. Yet they belong to a particular clause type, declarative clause (CP). In other languages, main clauses are clearly CPs. There have been two approaches to this problem in the literature. Some linguists argue that since English main clauses do not have the appearance of CPs, and look like IPs, they are what they look like, i.e. they are IPs. Under this view, languages may vary crosslinguistically with respect to the category of root declarative clauses. This analytical option is unavailable to me, because I assume clause types are universally represented in the same way: a declarative clause in language X is represented as a declarative clause in language Y. My view is more compatible with other linguists who argue that since declarative clauses are CPs in some languages, they are CPs in all languages. Therefore, these linguists conclude, English main clauses are CPs, with a silent C node, and presumably a silent declarative operator, and an IP complement.

This analysis is incompatible with (56), since the declarative head is not associated with any lexical material at any stage of the derivation, and therefore uninterpretable. If (56) is correct, the language learner is forced to analyze English main clauses as involving clausal pied-piping:


Since a clausal constant is in Spec, CP, C will be empty by the LCA. Embedded declarative complements trivially satisfy (56) when an overt C node is present.

When C is absent, as is possible in English, (56) is satisfied either because C is a trace of C which has incorporated into V (this means that CP was associated with lexical material at one stage of the derivation):13

Or alternatively, C can be absent, because IP has pied-piped to Spec, CP, as in (58). This latter proposal runs into the problem of explaining why C can only be absent in “head government” configurations (which reduce to incorporation configurations). Given (56), then, English has overt pied-piping of clausal constituents, just as Nweh has. (56) makes English into a much more mixed word order language than generally assumed, and has quite far reaching consequences for the analysis of many phenomena. In particular, it has consequences for clauses containing focus, negation, and Beghelli and Stowell’s (1997) DistP and ShareP, and so on (I am assuming that these projections are present iff they are needed for interpretation), which will all need to be activated by movement of some lexical material. Here I will concentrate on one particular consequence and propose a reanalysis of that-t effects in English. 3.6.2 Reanalyzing that-t effects The analysis of the that-t effect constitutes an old problem. Subject extraction in English root wh-questions does not trigger do-support, and extraction of the subject immediately adjacent to C forces that to be absent: (61)

a.

Who came yesterday? *Who did come yesterday?


b.

Who did you think came yesterday? *Who did you think that came yesterday?

The traditional analysis has the subject wh-phrase move to Spec, WhP, at least for (61b). With respect to the analysis in (61a), we find a familiar split. Some linguists argue that these wh-questions are simply IPs, others that they are CPs with the wh-phrase moved to Spec, CP. The impossibility of S AI or the obligatory absence of that is attributed to some property that the empty category in the extraction site must satisfy. Crosslinguistically, there are numerous problems arising with respect to subject extraction. One problem for instance is that some languages do not seem to allow for extraction out of the subject position at all. This is for example the case in Italian, where extraction must proceed from the postverbal subject position (Rizzi 1982). Suppose that the Italian situation is not exceptional, but in fact represents the norm: (62) A wh-phrase can never be extracted from the Spec position immediately ccommanded by C. (where by “subject” position, we understand the “highest” subject position, i.e. the Spec position immediately below the C node).14 We can now understand the English facts differently. Since the wh-phrase cannot be extracted from the subject position, the traditional analysis in which wh moves to Spec, WhP is unavailable:

This raises the question how the Wh projection is activated. There is no other option than to pied-pipe the entire clausal constituent to Spec, WhP (as I have argued before, the Wh projection combines with the Q projection).15

Thus English resorts to the analysis that Kayne (1995) argues for quite generally in certain wh-in situ languages. The Wh projection is activated by the lexical material in its Spec position. If this movement proceeds through Spec, QP, the Q projection is activated as well. The impossibility of do-support follows from the structure: the IP is simply too high, and I to C to Q movement would yield an improperly bound trace in the highest IP. The same analysis applies to subject wh-phrases in embedded complements. The whphrase cannot be extracted from the subject position, because that position universally cannot be extracted from. The IP therefore pied-pipes to Spec, CP, out of the c-command domain of C, from which the wh-phrase further extracts to Spec, WhP. Since IP is overt,


C must be silent. That must therefore be obligatorily absent: it is a doubly filled Comp filter effect:

The analysis in (65) can be easily extended to derive possible questions as in (66):16

It is interesting that the analysis of wh-questions in English is not uniform: within English, one finds the same variation as observed crosslinguistically. I take these patterns as very strong evidence in favor of (56). (56) obviously has consequences for the analysis of many other construction types. Imperatives will involve either head movement to the imperative head (as in Dutch), or pied-piping to the Spec of the imperative head. Negation will involve either an overt head, or overt pied-piping of a negative constituent to Spec NegP. Languages with no overt Ds will have to be piedpiping languages, where an XP is in Spec, DP, and so on. (56) forces the language learner to assume massive pied-piping in cases where the head is silent, and as such is an extremely powerful, yet restrictive, principle. I will explore these questions more fully in future research. 4 Some preliminary answers Let me now return to some of the more general questions. This chapter contains three interrelated, but logically distinct proposals. 1 The (modified) LCA applies only to terminal elements, and c-command is not restricted to categories. This proposal derives the doubly filled Comp filter. I have briefly examined apparent violations of the doubly filled Comp filter, and presented alternative analyses for these. I have tried to show that these analyses are independently needed, and are in fact necessary if we even want to start to talk about crosslinguistic variation in this domain.


2 Languages are expected to have overt heads with silent Specs or silent Specs with overt heads. This expectation is still valid regardless of whether the doubly filled Comp filter derives from the LCA, or whether it is some economy principle. I discussed some cases that fit the configurations well, and show how these configurations provide a simple analysis for patterns which have been felt to be problematic so far, but that in fact represent the norm. I also discuss some cases where the actual surface word orders derive from the interaction of leftward movement to Spec and the presence of a head. I argued that two phenomena fall under this heading: head drop and pro drop. 3 Languages disallow projections to be headed by silent covert heads and Specs. In particular, I proposed a principle that requires a projection to be associated with lexical material at some point in the derivation. This principle basically answers the question of why syntax makes use of so many movements, and why we seem to be heading toward the idea that everything must move, at least once. The idea here is quite simple: projections must be activated to be semantically interpretable. Projection Activation happens by associating overt lexical material to either Spec or head at some point in the derivation. Syntax thus uses few lexical items, but puts these to maximum use. Lexical items are used to form elaborate syntactic structures from which meaning can be computed. The formation of these elaborate syntactic structures can only be achieved in one way: by movement of some constituent containing a lexical item. By the LCA, movement is further restricted to movement of some XP to a Spec position of an empty head, or movement of some head position with a silent Spec position. This yields the two basic configurations that we appear to find within languages: some XP occupies the Spec position, and the head is silent, or a projection contains a head, and the Spec is silent. I have thus provided basic answers to the different questions raised in this chapter. I will repeat these here: (67) a. Why do we seem to find precisely the two configurations in (10), repeated here below, crosslinguistically? [[[YX] [… [X]]Head movement to some head position Y, or [[XP[Y[[XP]Movement of some XP to Spec, YP b. What motivates (heavy) pied-piping? c. If heavy pied-piping is so heavily used, why did it take us (linguists) so long to discover it? d. Why do there appear to be no languages in which all heads and Spec are silent, and in which everything simply remains in the thematic domain of the VP? The answer to these questions is simple: projections must be activated to be interpretable; this can be achieved either by moving a constituent to the Spec position of a projection, or


by moving a head to the head position of a projection: we thus find the two configurations above. Projection activation also partially explains (67b). (Heavy) pied-piping is a means to activate a projection if the head is silent. A question remains however. Why do languages pied-pipe different types of constituents? Take English wh-questions for example. According to my analysis, English wh-questions involve either pied-piping of a small wh-XP to Spec, WhP (who did you see), or pied-piping of an entire IP (who came [e]) to Spec, WhP. The latter is forced, only because the extraction of a smaller XP fails. A fuller answer to this question must wait until we have more insight into pied-piping. The modified LCA provides the answer to (67c). Pied-piping is often hard to detect, because the head must be silent in that case. So, for overt evidence for heavy pied-piping we must rely on certain configurations of data that do not seem to arise that frequently— movement of a big constituent to some Spec position to the left, and stranding of an overt head immediately below this projection to the right: (as usual, positions containing overt material are boldfaced). (68)

[[XP][Y

[

Z

[XP e]

Finally, my proposal also allows a beginning of an understanding of (67d): languages are not allowed to leave all lexical material in situ simply because this will yield a Full Interpretation violation. What needs to be explained though is why languages cannot simply raise their VPs through the different Spec positions, thus activating them on their way, and giving the appearance of a language that has everything in situ. Here I would bring in an additional consideration. Not every constituent can activate a particular head: there must be a semantic or maybe morphological relation beteen material contained within the pied-piped phrase and the head, which further restricts possible movement to Spec. Notes 1 Parts of this chapter have been presented at DIGS IV in Montreal (November 1995), “Antisymmetry and the doubly filled Comp filter”, at USC (1995) “Antisymmetry and the generalized doubly filled Comp filter” and at UCLA (1996). I thank the audiences for lively discussions, and Edward Garret and Felicia Lee for comments on a previous version of this chapter. 2 The parallelism with Romance de is actually even greater, in the sense that the associative marker also precedes postnominal adjectives. This recalls Kayne’s (1994) analysis of [quelqu’un [de [célèbre []. 3 Because-clauses form the only exception; see Nkemnji (1995) for an interesting discussion. 4 In Koopman (1994), I suggest that it is precisely these two configurations that are found, because they bring the head N in the domain of a c-commanding head, which would allow the head N to subsequently incorporate and satisfy the lexical properties of some predicate. 5 As Kayne (personal communication) points out, allowing X’/XP to be within the set of ccommanders raises a problem with respect to the ordering of the complement and the specifier, because these would be too symmetrical as well. I will not address this point here, but maybe complements are forced to move as well. From recent work it has become clear


that many terminal elements are in fact occupying Spec positions: the configuration [… [Spec H [complement] with positions containing overt material boldfaced seems to be omnipresent. Taken to apply derivationally, the LCA would force complements to move to Spec positions, and force further movement of the Spec or the head position. For the highest position in superficially head initial languages, it presumably forces both the head and the Spec of the complement to be silent:

[Spec

[H

[Spec H

[Spec [H

[Spec

6 This principle is reminiscent of Cheng’s (1991) clause-typing proposal. 7 As pointed out by Ian Roberts, this analysis does not account for another property of Celtic languages: why cannot wh-phrases move through this position and trigger agreement? This fact might be related to a much more general fact concerning extraction from Spec, IP. See 3.6.2 and Note 14 for more discussion. 8 This analysis carries over to the slightly more complex agreement facts in Arabic, if person, number and gender each head their own projections, as in Schlonsky (1989), with V moving to Number. [V+num[DP gen [. Thus, gender agreement can be triggered by the subject DP, but number agreement cannot, because the head of Num is overt. 9 If I were to pursue this issue in more detail in this chapter, I would show that Dutch also instantiates the Irish pattern in certain types of declaratives that do not have an overt topic. Quite generally, languages will exhibit much more mixed properties than traditional analyses assume. 10 Zanuttini (1991:61) cites examples like:

(i) A nessuno Gianni ha parlato It is unclear what to say about the absence of non in such cases. For other complex cases, see Zanuttini (1991:131). 11 This is illustrated in the following examples:

(i)

[pro

(ii)

*dat that

[geef[ give geef give

maar but maar but

terug [e back terug back

Interestingly, the situation in German is exactly the opposite: (iii)

das that *gib give

gib give mal but

mal but zurück back

zurück back

12 This proposal is similar in spirit to Cheng’s (1991) clause-typing proposal. Cheng proposes that a wh-question must be “typed” as such at S-structure. 13 Note that we can now almost understand why an incorporated C is never lexical. The head of the chain does not activate any projection, and only serves to activate the projection containing the trace of C. Head movement of a silent head, is thus another way to activate projections.


14 This also allows us to understand a well-known phenomenon in Celtic languages, which is that subject wh-phrases never trigger number agreement on V: only pro triggers number agreement. In order to trigger full agreement, wh-phrases must move through the highest subject position. Extraction from that position is impossible by (62); therefore wh-phrases cannot trigger number agreement. 15 SAI inversion would not help, since the inverted I ends up in the Q (see 3.3.3), not in Wh.

(i)

[

Wh [

[Q[did] [ who

It is not clear that the language learner will ever come up with such creative solutions as in (i) if these do not match the primary data. 16 As suggested by Anna Szabolcsi (1994:203).

References Abney, S. (1987) “The noun phrase in its sentential aspect”, unpublished PhD dissertation, MIT. Androutsoupoulou, T. (1994) “The distribution of the definite determiner and the syntax of Greek DPs”, CLS 30, University of Chicago Press. Beghelli, F. and T.Stowell (1997) “Distributivity and negation: the syntax of ‘each and ‘every’”, in A.Szabolcsi (ed.) Ways of Scope Taking, Kluwer, Dordrecht. Bhatt, R. and J. Yoon (1992) “On the Composition of COMP and Parameters of V2”, in Proceedings of the West Coast Conference on Formal Linguistics. Cheng, L. (1989) “Aspectual licensing of pro”, unpublished manuscript, MIT. Cheng, L. (1991) “On the typology of Wh-questions”, unpublished PhD dissertation, MIT. Chiu, B. (1992) “The syntax of functional words in Chinese”, unpublished PhD dissertation, UCLA. Chomsky, N. (1986a) Barriers, MIT Press, Cambridge, Mass. Chomsky, N. (1986b) Knowledge of Language: Its Nature, Origin, and Use, Praeger, New York. Chomsky, N. (1992) “A minimalist program for linguistic theory”, MIT Occasional Papers in Linguistics 1, Cambridge, Mass. (published (1993), in K. Hale and S.J. Keyser (eds) The View from Building 20: Essays in Linguistics in Honor of Sylvain Bromberger, MIT Press, Cambridge, Mass, 1–52. Chomsky, N. (1995a) Chapter 4 in N.Chomsky The Minimalist Program, MIT Press, Cambridge, Mass. Chomsky, N. (1996) “Bare phrase structure” in G.Webelhuth (ed.) Government and Binding Theory and the Minimalist Program, Blackwell, Oxford, and Cambridge, Mass. Chomsky, N. and H.Lasnik (1977), “Filters and Control”, Linguistic Enquiry LI 8(3). Dikken, M.den (1998) “Predicate inversion in DP”, in A.Alexiadou and C.Wilder (eds) Possessors, Predicates and Movement in the Determiner Phrase, John Benjamin, Amsterdam. Fukui, N. (1986) A Theory of Category Projection and its Applications, unpublished PhD dissertation, MIT. Garrett, S. (1995) “A history of negation in Spanish”, paper presented at DIGS IV, Université du Quebec a Montreal. Haegeman, L and R.Zanuttini (1991) “Negative heads and the NEG criterion”, The Linguistic Review 8: 233–251. Hale, K. and J.McCloskey (1984) “On the syntax of person-number inflection in modern Irish”, Natural Language and Linguistic Theory, 487–533.


Hallman, P. (1996) “German complex complementizers”, in F.Lee and E.Garret (eds) Syntax at sunset, UCLA Working Papers in Syntax and Semantics 1: 89–109. Henry, A. (1995) Belfast English and Standard English. Dialect Variation and Parameter Setting, Oxford University Press, New York and Oxford. Hoekstra, T. (1995) “The status van van”, Tabu 15: 160–164. Jaeggli, O. and K.Safir (1989) The Null Subject Parameter, Kluwer, Dordrecht. Kayne, R. (1989) “Facets of romance past participle agreement”, in P.Beninca (ed.) Dialect Variation and the Theory of Grammar, Foris Publications, Dordrecht. Kayne, R. (1994) The Antisymmetry of Syntax, MIT Press, Cambridge, Mass. Kinyalolo, K. (1990) “Syntactic dependencies and the spec/head agreement hypothesis in Kilega”, unpublished doctoral dissertation, UCLA. Koopman, H. (1983) “ECP effects in main clauses”, Linguistic Inquiry 14: 346–351 (Chapter 4 in this volume). Koopman, H. (1987) “On the absence of case chains in Bambara”, unpublished manuscript, UCLA (published in 1992 NLLT 10: 555–594 and as Chapter 7 in this volume). Koopman, H. (1984) The Syntax of Verbs: From Verb Movement Rules in the Kru Languages to Universal Grammar, Foris Publications, Dordrecht. Koopman, H. (1991) “The verb particle construction and the syntax of PPs”, unpublished manuscript, UCLA. Koopman, H. (1993) “The internal and external distribution of pronominal DPs”, unpublished manuscript, UCLA (Chapter 3 in this volume). Koopman, H. (1994) “Licensing heads”, in N.Hornstein and D.Lightfoot (eds) Verb Movement, Cambridge University Press, Cambridge (Chapter 9 in this volume). Koopman, H. and D.Sportiche (1985) “Theta theory and extraction”, abstract of a talk given at the 1985 GLOW colloquium in Brussels, GLOW Newsletter, February 1985, Foris Publications, Dordrecht. Koopman, H. and D.Sportiche (1991) “The position of subjects”, Lingua 85 (2/3). Larson, R. (1985) “On the syntax of disjunctive scope”, Natural Language and Linguistic Theory 3: 217–64. Larson, R. (1988) “On the double object construction”, Linguistic Inquiry 19: 335–392. Mahajan, A. (1990) “The A/A-bar distinction and movement theory”, PhD dissertation, MIT. McCloskey, J. (1994) “Subjects and subject positions in Irish”, in Robert Borsley and Ian Roberts (eds) The Syntax of Celtic Languages, Cambridge University Press, Cambridge. Moritz, L. (1993) “The internal structure of the Ncufie DP”, in Hilda Koopman and Murat Kural (eds) Aspects of Ncufie Grammar, UCLA Working Papers. Nkemnji, M. (1995) “Heavy pied-piping in Nweh”, UCLA dissertation series. Pearson, M. (1996) “A raising and small clause analysis of existentials in Malagasy”, unpublished MA thesis, UCLA. Pollock, J-Y. (1989) “Verb movement, universal grammar and the structure of IP”, Linguistic Inquiry 20: 365–424. Ritter, E. (1991) “Two functional categories in noun phrases: evidence from modern Hebrew”, unpublished manuscript, UQAM, Montreal. Rizzi, L. (1982) Issues in Italian Syntax, Foris Publications, Dordrecht. Rizzi, L. (1994) “Residual verb second and the wh-criterion”, in A.Belletti and A.Rizzi (eds) Parameters and Functional Heads: Essays in Comparative Syntax, Oxford University Press, New York. Rizzi, L. (1995) “On the fine structure of the left periphery”, unpublished manuscript, Université de Genève.


Schlonsky, U. (1989) “The hierarchical representation of subject-verb agreement”, unpublished manuscript, University of Haifa, Israel. Speas, M. and Fukui N. (1986) “Specifiers and projections”, MIT Working Papers, vol. 8. Sportiche, D. (1992) “Clitic constructions”, to appear in L.Zaring and J.Rooryck (eds), Phrase Structure and the Lexicon, Kluwer Academic Publishers, Dordrecht. Sportiche, D. (1994) “Adjuncts and adjunctions”, unpublished manuscript, UCLA. Sportiche, D. (1995a) “Sketch of a reductionist approach to syntactic variations and dependencies”, in H.Campos and P.Kempschinsky (eds) Evolution and Revolution in Romance Syntax, Georgetown University Press, Washington D.C. Sportiche, D. (1995b) “French predicate clitics and clause structure”, to appear in A. Cardinaletti and M.T.Guasti (eds) Small Clauses, Syntax and Semantics, vol. 28, Academic Press, New York. Sportiche, D. (1996) “Pronominal clitic dependencies”, in Henk van Riemsdijk (ed.) Language Typology: Clitics in the European Languages, Mouton de Gruyter, Berlin. Szabolcsi, A. (1987) “Functional categories in the noun phrase”, in I. Kenesei (ed.) Approaches to Hungarian, vol. 2, Jate Szeged, Budapest. Szabolcsi, A. (1994) “The noun phrase”, in F.Kiefer and K.E.Kiss, (eds) The Syntactic Structure of Hungarian, Syntax and Semantics, vol. 27, Academic Press, San Diego, Calif. Valois, D. (1991) “The internal syntax of DP”, unpublished doctoral dissertation, UCLA. Zagona, K. (1982) “Government and proper government of verbal projections”, unpublished doctoral dissertation, University of Washington at Seattle. Zanuttini, R. (1991) “Syntactic properties of sentential negation: a comparative study of Romance Languages”, unpublished doctoral dissertation, University of Pennsylvania.

12 UNIFYING PREDICATE CLEFT CONSTRUCTIONS

1 The problem Like many West African languages, Vata, a Niger Congo language of the Kru family, has a particular verbal focus construction, sometimes referred to as the predicate cleft construction. This construction involves contrastive focus on V or a predicate: a V is understood as contrasting with some verb implicit in the discourse:1 (1)

(2)

pā; ká m pá ā; (Vata) (Koopman 1984) throw you FUT it throw Q “Are you going to THROW it” (throw as opposed to roll) ká will

pā; throw I “I will THROW it”

m it

pá throw

The following properties characterize the contrastive verb focus construction in Vata. Morphology: the clause contains two copies of the verb. The verb in initial position carries special morphology associated with the construction (realized in Vata as a “construction” tone). The verb in the clause looks and acts like any regular V. Order. The contrastively focused verb occurs in clause initial position. Omission of the focused verb yields a regular sentence without focus. Dependency: The dependency between the focused verb and the copy obeys the same locality as manner and reason adjuncts (Koopman (1984), Koopman and Sportiche (1986)). A very similar construction involving contrastive focus on V, is found in Nweh, a Grassfield Bantu language spoken in Cameroon (Nkemnji 1995):2 (3)

a s/he

k ? P

ncù boil

ká crab

c boil

(Nweh)


“She BOILED the crab” (as opposed to frying it) (4)

à k ? ncù ká c s/he P boil crab boil “Did she BOIL the crab?” (as opposed to frying it)

l (Q)

(Nweh)

As in Vata, the clause contains two copies of the same V. The leftmost verb in Nweh has the form and distribution characteristic of Vs in clauses without verbal focusing. The rightmost verb carries particular verbal morphology (a tonal prefix and suffix and a segmental suffix).3 Apart from linear order, there is a further difference between Vata and Nweh which concerns cooccurrence restrictions of wh-phrases and focused verbs. In Vata, a focused V cannot cooccur with any wh-phrase, regardless of whether the wh-phrase is a subject, an object, or an adjunct (Koopman 1984): (5)

a. b.

*pā; throw *àl who

àl who pā; throw

he-R he-R

ká FUT ká FUT

m it m it

p throw pá throw

lá wh la wh

(Vata) (Vata)

In Nweh subject wh-phrases can cooccur with predicate cleft (6), but non-subject whphrases cannot (7): (6)

(7)

k ? ju b jù *àw who P1 n-buy fufu buy “Who BOUGHT the fufu?” (as opposed to who sold the fufu) k ? ju k *àt m Atem P1 n-buy what “What did Atem BUY?” (as opposed to sell)

jù buy

l Q

l Q

The data above raise the questions that I will try to answer in this chapter: (8)

How should one account for the difference in linear order?

(9) How should one account for the different cooccurrence restrictions of focused verb and wh-phrases? I will present an analysis of the predicate cleft construction, and argue for an optimally simple analysis of the crosslinguistic variation which derives both the differences in word order and the differences in cooccurrence restrictions from a common underlying structure.


2 Theoretical assumptions The theoretical assumptions below are Minimalist in spirit, but differ in the general shape of the theory. The overall picture is closest to the work of Sportiche (1995), Kayne (1994), Rizzi (1995), Cinque (1997). • Syntactic structures are Binary Branching structures, obeying X-bar theory. Whether the properties of X-bar theory can be derived (Kayne (1994), Chomsky (1995)) is of no concern to the present paper. • Each feature projects. This is what Sportiche (1996) calls the “atomization” of syntactic structures. The one-projection-per-feature theory is a logical continuation of work in the 1980s on the architecture of clauses, DPs, APs and PPs. • All languages are underlyingly identical (Universal Base Hypothesis). • There is no head initial head final parameter: all languages are Spec head complement underlyingly (Kayne 1994). For the purposes of this chapter it is sufficient that there are no underlyingly mixed languages. • There is no crosslinguistic difference in hierarchical structure (Sportiche (1995), Cinque (1997), Koopman (1996)). Language variation cannot be attributed to different hierarchical locations of projections associated with the same semantic interpretation. This rules out analyzing the difference between Vata and Nweh in terms of a different location of Focus (say high focus versus low focus). • There is no Procrastination: everything must move overtly. Crosslinguistic differences do not derive by overt or covert movement, but by movement of different sized constituants (see Koopman (1996) for more discussion as well as this chapter). Movement (copy and deletion) is of the usual kind: • head movement (left adjunction only, no base generated morphology) • XP movement (leftward only to designated Spec positions, no adjunctions). It is important to keep in mind that there is much more XP movement (pied-piping) than we are used to in standard analyses of, say, English, with big parts of sentences, and sentences themselves moving around (Sportiche (1995), Koopman (1995, 1996), Kayne (1994), Nkemnji (1995)). Much of this pied-piping is transparent in African languages both within DPs (final determiners, quantifiers, demonstratives, and so on) and within the clause (final negation, final question particles) (see in particular Nkemnji (1995)). • Movement obeys locality. Head movement obeys the Head Movement Constraint, and XP movement can only reach a local Spec. Head movement extends the domain of movement, and brings the next local Spec in the local domain (the Head Constraint of van Riemsdijk (1978), The Government Corollary of Baker (1988), Equidistance of Chomsky (1995)). Locality is ‘wired-in’, and not subject to Economy.


Movement takes place for licensing purposes, either for the familiar morphological reasons, or for semantic reasons (scope). In addition, I assume that movement can be forced because of a principle that I have called the PPA (Principle of Projection Activation (Koopman 1996)). (10) Principle of Projection Activation (PPA) (Koopman 1996) A Projection is interpretable iff it is associated with lexical material at some stage in the derivation. The PPA prevents representations with truly empty projections (where neither Spec nor head contains a lexical item or a trace) and forces movement. A translation of the PPA into the standard Minimalist terminology comes close to (11): (11)

Functional heads are strong.

• Overt material must be linearized. I assume that the distribution of overt lexical items over these huge universal structures is determined by some version of the LCA (Linear Correspondence Axiom (Kayne 1994)). In Koopman (1996), I modify the LCA and show that this modification yields the doubly filled C filter.4 (12) Modified LCA has as a consequence that no Spec and head position can simultaneously contain overt lexical material. 3 The analysis of predicate cleft The verbal focus construction in Vata and Nweh receives the same contrastive focus interpretation (which Larson and Lefebvre (1991) analyze as quantification of the event). This construction never yields an emphatic reading, (He DID want to…as opposed to He did NOT want to…), and cannot be used with individual level predicates. Since the focused verb occurs in a particular position in the clause, I will assume that it is “associated” (in a sense to be made precise below) with the Focus Projection (FocP). Since the same semantic interpretation arises, I will assume that it is associated with the same FocP in both Vata and Nweh. 3.1. Predicate cleft in Vata In Vata, the focused V appears at the left edge of the sentence, pointing to a head initial FocP (FocP>IP). The focused verb is “associated” with the FocP, which implies that the focused verb is either in the Spec position of the FocP, or in the head position. In Koopman (1984), I argued that the focused verb moved to COMP via head movement (at that point basically the only available analysis). I called this type of head movement the whtype of head movement (A’ head movement), because it behaved like phrasal A’


movement, and not like V to I movement which I called the A type head movement. With the subsequent development of the “middle” field, an XP analysis of the predicate cleft construction has become feasible and desirable: instead of head movement, predicate cleft involves XP movement of a “small” VP containing nothing but V to Spec, FocP. This analysis immediately accounts for the A’ properties of the construction: the predicate cleft construction patterns with XP movement, because it is XP movement. An XP analysis of predicate cleft makes the distinction between two types of head movement unnecessary, a welcome result. Finally, the XP analysis finds empirical support: some adverbs and aspectual markers may optionally accompany the focused V (Koopman 1984). There is no evidence that these should be analyzed as forming a complex head with the focused V. Since arguments and small clause predicates may never accompany the focused V, it must be the case that all arguments and complex predicates must obligatorily vacate the VP.5 As I argue elsewhere on independent grounds, arguments and predicates must always be licensed in specific landing site positions outside of the minimal VP. The following annotated tree illustrates the derivation for Vata (English words are used for convenience). The resulting sentence contains two overt copies of the same V, each carrying different morphology. This is an old and well-known problem of this construction that requires a new explanation. Under a head movement analysis, as in Koopman (1984), it was the spell-out of the V in the clause that required an explanation. Under a remnant movement analysis, it must be explained why the focused verb cannot be silent. Let us briefly consider the “spell-out” problem taking into account the hierarchical relations. The V within the VP does not c-command the V in I, and therefore does not form a V chain with it. The V within the clause is thus spelled out for the same reason as any V in the head of a chain position. What is unexplained is why the focused V cannot be silent. The V moves outside the VP to get tense morphology, just as arguments move out of the VP, prior to movement of the VP to Spec FocP. The V must be spelled out within the VP, (and within IP) but the copies of the arguments that are contained in the preposed VP cannot be spelled out and can only be spelled out in the IP. I will assume that spell-out of V is forced by recoverability (after all, if the focused verb were silent, nothing would signal verbal focusing), and that spelling out of arguments is prohibited in the absence of a local licenser. The morphology associated with the verbal focus plays a crucial role in that it makes the spell-out of the focused verb possible.


3.2 Predicate cleft in Nweh Vata and Nweh have the same verbal focus construction, with the same meaning, and therefore involve the same underlying hierarchical structure. The languages differ however as to the position in which the clefted predicate (= small VP) appears. In Vata it surfaces at the left edge of the sentence, pointing to a hierarchical order FocP>IP, in Nweh it surfaces somewhere toward the right edge. This is depicted in (14), which also includes information about the morphological structure of the clefted predicate: (14)

Vata: Nweh

[V+tone] DP T V-foc morphology DP T Vf DP… V1

…V1…(Q) V1 [focus tone+V+segment] (Q) [focus morphology–V–focus morphology]

The FocP in Nweh cannot be underlyingly head final, because of the assumption that there are no underlyingly head final languages (see section 2). The surface order in Nweh must therefore be derived by some leftward movement. At the surface it looks as if Nweh is using a low FocP, and Vata a high FocP. However, since by assumption there is no crosslinguistic difference in hierarchical order (this is really a “Minimalist” assumption) and since the construction in Nweh and Vata yield the same interpretations, FocP must be higher than IP in Nweh as well. See (15) on 372.


IP movement around the focused VP in Nweh raises the question of the landing site for IP. IP cannot land in Spec, FocP, because it hosts the clefted VP. Since it precedes the clefted V, it must be in the Spec of some higher position, YP. YP itself must be lower than Q, because the entire complement of Q precedes Q and Q scopes over FocP. Hence Q>Y>Foc. Although I will continue to label this projection YP, it is probably part of the focus projection, which should thus be viewed as a two layered projection. The YP possibly plays a role in pseudocleft constructions: what John boiled is a crab. The predicate cleft construction in Vata would be comparable to a cleft construction; the predicate cleft construction in Nweh to a pseudocleft construction.

The movement of IP to Spec, FocP obeys locality. If Foc moves to Y both Spec, FocP and Spec, YP are equidistant to IP. Empirical evidence for Foc to Y head movement consists of a low tone preceding the focus constituent. Thus, Spec, YP and Spec FocP are “active” in Nweh.


3.3 Crosslinguistic variation involving YP If there is a YP above FocP in Nweh, this projection must be present in Vata as well (there is no crosslinguistic variation in structure, see section 2). Since the PPA requires that all projections be activated by lexical material (i.e. all functional projections are strong) the question arises as to how the YP is licensed in Vata. There is no indication of any head preceding the focused verb in Vata. It must therefore be the case that the Spec of YP is filled. I assume that the entire FocP moves to Spec, YP in Vata, thus giving the appearance of the head initial character of the FocP.


In both Vata and Nweh, Spec, YP contains a constituent, but the size of this constituent varies: (18)

in Vata: in Nweh:

FocP is in Spec, YP IP is in Spec, YP

The difference between Vata and Nweh is not a structural difference, nor a difference involving head initial/head final character of a particular head, nor a difference involving covert versus overt movement. The difference lies in the size of the constituent that occupies YP, with the entire FocP in YP Vata, and the IP (the complement of Foc) in Nweh. This is depicted in the following structure:

It is easy to see that this analysis yields the different linear orders of Vata and Nweh from a common structure. I show in the next section that it does more: the incompatibility of wh-


phrases and predicate cleft in Vata, and the compatibility of subject wh and predicate cleft in Nweh fall out from the derived structures in (19). 4 Deriving cooccurrence restrictions from necessary structural properties Predicate cleft and wh-phrases are always incompatible in Vata. In Nweh, subject whphrases can cooccur with predicate cleft, but no other wh-phrases can. The incompatibility of focused verbs and wh-phrases in Vata was accounted for quite simply in earlier versions of the theory with a single landing site position for A’ moved elements (COMP). Complementary distribution followed from competition for the same landing site. This solution obviously cannot work for Nweh since the cooccurrence of wh-phrases and predicate cleft is configurationally determined. Intuitively speaking, subject whphrases are able to reach the wh-position, but object wh-phrases are not, and this is precisely what the structures give us, as I will show below. In the one projection per feature theory, these cooccurrence restrictions must be derived in a different way. For a similar problem arising in Italian, Rizzi (1995) proposes that the incompatibility of focus and wh-phrases follows from the fact that wh-phrases are inherently focused. This type of explanation predicts that wh-phrases and focus can never cooccur, and runs into trouble because Nweh subject wh-phrases and focused verbs can cooccur. It is unlikely that subject wh-phrases, and object wh-phrases receive a different focus interpretation, and we are dealing with contrastive focusing on V throughout. The Nweh data suggest a structural explanation which should have the effect that subject whphrases are able to reach the wh-position, but object wh-phrases are not. I will now argue that the structures in (19) exactly yield this effect. What must be explained is the following: (20)

a. b. c.

Wh-phrases cannot cooccur with predicate cleft in Vata. Subject wh-phrases can cooccur with predicate cleft in Nweh. Object wh-phrases cannot cooccur with predicate cleft in Nweh.

In Koopman (1996), I have argued that wh-questions consist of a Wh projection, where whphrases are licensed and a Q projection, with Wh>Q. This yields the structure Wh>Q>YP>FocP. Wh-question formation involves the appearance of a sentence final matrix question particle in both languages (la in Vata, 1ε in Nweh) indicating leftward movement of the complement of Q.


The structures in (21) are going to diverge, when movement to YP takes place (FocP moves to YP in Vata), IP moves around FocP to Spec, YP in Nweh, yielding the structures below: pied-piping. It follows that predicate cleft and wh-phrases are incompatible in Vata: the wh-phrase always remains trapped under the FocP and can never reach the WhP in this configuration.


The cooccurrence restrictions fall out from these structures, as I will show in more detail below. In a nutshell: • Wh-phrases in Vata can never cooccur with predicate cleft, because the wh-phrase will be unable to reach Spec, WhP (section 4.1). • Subject wh-phrases in Nweh can cooccur with FocP, because the movement of IP around FocP brings the wh-phrase in the local realm of the Spec, wh (section 4.2). • Non-subject wh-phrases in Nweh are also moved around focus, yet cannot cooccur with focused Vs. The question why they cannot cooccur with predicate cleft cannot be answered in the same way as in Vata, since the wh-phrase in Nweh is no longer trapped under focus (section 4.3). 4.1 The non-occurrence of predicate cleft and wh-phrases in Vata As shown in (22) the wh-phrase in the predicate cleft constructions in Vata will always be trapped under focus. In order for a licit wh-interpretation to arise, the wh-phrase must move to Spec, WhP. But in order to do so, the wh-phrase must cross an intervening A’position, yielding a locality violation. The wh-phrase cannot trigger pied-piping of the entire FocP complement, because it is not in the right structural configuration to trigger


4.2 The cooccurrence of subject wh-phrases and predicate cleft in Nweh Let us look at the next stage in the derivation in (22), when the complement of Q has raised to Spec, QP.

problems, because no projection contains lexical material in both Spec and head position simultaneously. Note that the option of subextracting the wh-phrase is blocked by the modified LC A (which derives the doubly filled C filter, see section 2). If the subject were


to extract, there would not be enough space for linearization: the QP projection would contain lexical material in both the head and the Spec position movement, and thus violate the doubly filled C filter. Subject wh-phrases and predicate cleft can thus cooccur in Nweh because of the movement of IP around the focused constituent, and the particular positon the wh-phrase occupies within the IP which allows for pied-piping of the bigger constituent.

4.3 The non-cooccurrence of wh-phrases and predicate cleft in Nweh Let us consider the structure of a cooccurring predicate cleft and non-subject wh-phrase in Nweh at the point in the derivation where YP (containing the wh-phrase) has inverted with Q (moved to Spec, QP): No grammatical sentence results from this structure. Two questions arise: why is piedpiping of the entire YP not possible (as it is for subjects), and why cannot the wh-phrase move to Spec, WhP on its own: one certainly must allow for movement of wh-objects! Non-subject wh-phrases are not in a pied-piping configuration (they are not in the Spec position of the entire constituent). The YP therefore cannot undergo pied-piping, and the surface string that would be derived by pied-piping is ungrammatical (*njikem boil what boil (compare example (7)). If the wh-phrase is too deeply embedded to trigger piedpiping, why cannot it extract by itself to Spec, WhP? This certainly seems to be a normal configuration for object extraction: it crucially differs from the Vata configuration, in that there is no intervening A’ position. Suppose then that the wh-phrase is able reach Spec, WhP in this configuration, and let us see if the resulting structure can be excluded on other grounds: The wh-phrase in the IP occurs in the Spec of the Spec position. This is a well-known piedpiping configuration (compare Whose brother’s picture did you take?), allowing it to pied-pipe the YP to Spec, WhP. (For arguments that English subject extraction involves pied-piping of the entire clause, see Koopman (1996).) This structure can be linearized without any


The structure will be ruled out by the doubly filled C filter, if nothing else happens: both Spec, QP and Q contain overt lexical material. This accounts for the illformedness of


(26): (26)

*akO njikEm kE? njuO juO IE (without further inversion) what Njikem P1 N-buy buy Q It is interesting that this structure improves substantially if it is embedded in an overt cleft construction (which seems to create the additional space for the realization of lexical material). (27)

?akO m azea njikEm what Foc Rel Atem What is it that Atem BOUGHT?

kE? P1

njuO n-buy

juO buy

IE Q


The structure in (25) is blocked as well if further inversion takes place, showing that such inversion (IP movement) must be impossible as well in this structure (presumably because of the intervening wh-phrase). (28)

*njikEm Njikem

kE? P1

njuO bought

juO buy

kO what

IE Q

(with further inversion)

5 Conclusion In this chapter, I have provided strong support for a unified analysis of the predicate cleft construction in Vata and Nweh. A common hierarchical structure can be assumed to underly Vata and Nweh. The difference in surface order reduces to a difference in the size of the constituent that occupies a particular Spec position: in Nweh IP moves around FocP to YP, whereas in Vata, the entire FocP occurs in YP. This analysis not only gets the different linear orders, but also, quite surprisingly, yields a simple explanation of the different cooccurrence restrictions of wh-phrases and focus Vs. This explanation uses noncontroversial assumptions about locality, a conservative assumption about pied-piping configurations (a wh-phrase can pied-pipe a constituent iff it is “associated” with the Spec of that constituent), and (restricted) appeal to the generalized doubly filled C filter. My analysis does not appeal to a head initial head final parameter (such an analysis in fact would not allow the same explanation of the cooccurrence restrictions), nor to a different hierarchical organization (the analysis shows that it is not necessary to assume a different hierarchical structure, it does not show that a different hierarchical structure cannot be assumed). All movements are overt and no appeal is therefore necessary to covert movement, nor to the strong-weak distinction of functional categories. Notes 1 This particular way of expressing contrastive focus on V is not widespread typologically. It is found in many West African languages of the Kwa family (for instance Yoruba, Fongbe, Ewe, Abe) and the Kru families (Vata, Gbadi,…), in Caribbean Creoles (Saramaccan, Sranan, Haitian, Jamaican…). The Vata data are based on my fieldwork, discussed and analyzed in Koopman (1984) and Koopman and Sportiche (1986). 2 The Nweh data are based on Nkemnji (1995) and data gathered during the UCLA field methods class on Nweh, UCLA Spring and Winter quarter, 1996. Thanks go to the participants of the class, Michael Nkemnji, Tonia Androutsopoulou, Edward Garrett, Matt Gordon, Catherine Crosswhite, Javier Gutierrez, Peter Hallman, Chai-Shune Hsu, and Matt Pearson, as well as to Manuel Español-Echevarría, Anna Szabolcsi, Ed Stabler, Andrew Simpson, and Dominique Sportiche. A computerized data base on Nweh is available on request. Working papers are in preparation. 3 Because of space limitations, I will not be able to go into the dependency between the two verbs in Nweh. It is difficult to show that the relation can be non-local, since many clause


types can contain focus. Straightforward island violations are observed for subject islands and purposive islands. 4 Maria Rita Manzini (personal communication) points out that the same conclusion can be reached in a particular version of the Minimalist Program: if each feature projects, then each projection will have exactly one feature to be checked. This can be achieved either by head movement or by XP movement. This might not be sufficient however. There are configurations with an overt head, where some constituent must still move to the Spec of that projection. Final question particles for example, are overt, but still trigger pied-piping of their complement. At the point of linearization, the projection no longer contains lexical material in both Spec and head position. One could say alternatively that the pied-piping is triggered to satisfy the features of yet another projection. 5 The verbal focus construction thus represents a case of remnant movement (compare den Besten and Webelhuth (1990)), i.e. a case in which extraction out of a constituent is followed by subsequent movement of that constituent to some higher position.

References Baker, M. (1988) Incorporation: A Theory of Grammatical Function Changing, University of Chicago Press, Chicago. Besten, H.den and G.Webelhuht (1990) “Stranding”, in G.Grewendorf and W. Sternefeld (eds) Scrambling and Barriers, John Benjamins, Amsterdam. Cinque, G. (1997) Adverbs and Functional Heads. A Cross-linguistic Perspective, to be published by Oxford University Press. Chomsky, N. (1995) The Minimalist Program, MIT Press, Cambridge, Mass. Kayne, R. (1994) The Antisymmetry of Syntax, MIT Press, Cambridge, Mass. Koopman, H. (1984) The Syntax of Verbs: from Verb Movement Rules in the Kru Languages to Universal Grammar, Foris Publications, Dordrecht. Koopman, H. (1995) “The syntax of Dutch PPs”, unpublished manuscripts, UCLA (Chapter 8 in this volume is a revised version). Koopman, H (1996) “The Spec head configuration”, in Edward Garrett and Felicia Lee (eds) Syntax at Sunset, UCLA Working papers in Syntax and semantics 1 (Chapter 11 in this volume). Koopman, H. and D.Sportiche (1986) “A note on long extraction in Vata”, NLLT 4: 357–374 (Chapter 6 in this volume). Larson, R. and C.Lefebvre (1991) “Predicate clefting in Haitian creole”, in Proceedings of NELS 21. Nkemnji, M. (1995) Heavy pied-piping in Nweh, UCLA PhD dissertation, UCLA dissertation series. Riemsdijk, H.van (1978) A Case Study in Syntactic Markedness: the Binding Nature of Prepositional Phrases, Lisse: The Peter de Ridder Press. Rizzi, L. (1995) “On the fine structure of the left periphery”, unpublished manuscript, Université de Genève. Sportiche, D. (1994) “Adjuncts and adjunctions”, unpublished manuscript, UCLA. Sportiche, D. (1995) “Sketch of a reductionist approach to syntactic variations and dependencies”, in H.Campos and P.Kempschinsky (eds) Evolution and Revolution in Romance Syntax, Georgetown University Press, Washington D.C. Sportiche, D. (1996) “The atomization and partitioning of syntactic structures”, unpublished manuscript, UCLA.

INDEX

Ā;-binding 21, 24, 31, 50 Ā;-position 16, 21, 123 Ā;-position 16, 21 Abe: logophoricity 60; pronouns 3, 36 Abney, S. 74, 284, 323 absolutive Case 187 AC see Accessibility Condition Accessibility Condition (AC) 25, 27 accessible subjects, pronouns in Abe 46 accusative Case: Bambara 164, 186; Dutch 233; French 185 affixed pronouns, Welsh 79, 86 agreement: NumP movement 106; pro-drop 344; pronominals in English 101; pronominals in Irish 97; pronominals in Welsh 79, 83, 90; Spec head configuration in VSO languages 339 allPs (English) 102 Anderson, S. 97 Androutsoupoulou, T. 325 antecedent government 5, 300; i-command 319n14; COMP indexing rule 147; wh-movement in Vata 152, 154, 155 antecedents 37, 39, 54 Aoun, J. 4, 120n5, 123, 125, 146, 147, 169 AP-preposing 269 arguments, wh-extraction in Chinese 146 aspect, T-grid 262

Aux, Subject-Aux Inversion 117 auxiliary pronouns, Welsh 79, 86, 96 Avoid Pronoun Principle 142n6 Awberry, G. 88, 90 Baker, M. 170, 186, 187, 258, 267 Bambara: Case assignment 6, 164; PP modification 99 Bantu languages: pied-piping 323; see also Nweh bare VP complements, licensing 265, 277, 280 barriers, theory of locality 300 be-prefixation (Dutch) 308 Beghelli, F. 348 Belfast English 335 Belletti, A. 133 Bennis, H. 120n4, 131, 134 Besten, H. den 117, 117, 118, 135, 136, 262 Bhatt, R. 333, 336 Bijection Principle 3, 17, 20, 71n10 binary branching 198, 287 Binding Theory, Strong Crossover phenomena 23, 26, 27, 31 Bok-Bennema, R. 187, 276 Borer, H. 113n49 BP see Bijection Principle Breton, pronominal DPs 88 Brody, M. 31 Burzio, L. 134 c-command: Linear Correspondence Axiom 328; pronouns in Abe 48, 50

375

376 INDEX

C-command condition (CCC) 25 Cardinalletti, A. 74 Carstens, V. 74 Case: clitic doubling 113n49; PPs in Dutch 233, 238, 240 Case chains 6, 164 Case filter: Case assignment in Bambara 164, 169, 172, 177; PPs in Dutch 233 Case theory 6; Case assignment 164, 169; distribution of VPs 261; Spec head licensing 257 Case transmission parameter 179 Catalan, negative head drop 342 causatives: Case assignment in Bambara 170; Case assignment in French 186; licensing 270, 288 CCC see C-command condition chains, silent heads 345 checking domain 294, 316 Cheng, L. 344 Chinese, wh-in situ 123, 138, 146 Chomsky, N. 1, 2, 3, 8, 16, 21, 24, 25, 30, 111n42, 117, 117, 119n1, 121, 124, 125, 134, 164, 177, 194, 257, 281, 294, 316, 323, 327, 331, 336 Chung, S. 89 Cinque, G. 177, 358 circumpositional PPs, Dutch 198, 217, 224 Clements, G.N. 36, 60, 62 clitic doubling 82, 96, 113n49 clitic-climbing 260 coindexing: Control from COMP 123; pronominal binding 37; pronouns in Abe 40, 41; sloppy/nonsloppy identity 54; that t phenomenon 117; wh-questions in English 131 Cole, P. 260 COMP indexing rule 4, 123, 137; antecedent government in Chinese 147; that t violations in Dutch 134; wh-movement in Vata 150;

wh- raising in French 131 Comrie, B. 60 Condition on Long Extraction 5, 146, 157 Condition on Pronominal Binding 36 conjunctive pronouns Welsh 81, 94, 96 Control from COMP 4, 123 coreference: pronouns 37, 48; sloppy/nonsloppy identity 54 covert pronouns 68 CP(Path), PPs in Dutch 234 CP(place), PPs in Dutch PPs 210, 222, 238 Crook, H. 99 D-stranding 74; agreement in Welsh 82, 92 Danish, V-movement 294 Deg(place), PPs in Dutch 209, 221, 229, 238, 239 dependent pronouns, Welsh 79 Dikken, M. den 209, 238, 281, 323, 325 do-support: Subject-Aux Inversion in English 117; that t effects in English 348 Double Voice filter 329, 331 doubly filled Comp filter 101, 329, 342, 348 DPs, pronominal 73 Dutch: V-second 294; causatives 273, 280; COMP indexing rule 137; doubly filled Comp filter 332; finite verb in nonroot clauses 294; inseparable P prefixes 308; participle verbs 310; Particle Phrases 6; pied-piping within DPs 323, 325; PP modification 110n38; PPs 197; pro-drop 343; Subject-Aux Inversion (SAI) 117; that t violations 117, 118, 119n4, 134, 137; topic drop 343; yes/no questions 338 Dyirbal, Case 187

INDEX 377

Economy Principle Procrastinate 8 ECP see Empty Category Principle Emonds, J. 168, 185, 199, 238, 240, 261 empty categories, variables 16, 17, 22 Empty Category Principle (ECP): Case assignment in Bambara 169; Control from COMP 121; Focus 30; long extraction in Vata 146; PPs in Dutch 233; that t phenomenon 117; variables 17, 19 Engdahl, E. 24 English: V-movement 261; Case assignment 164, 168, 177, 240; causatives 268, 276; Control from COMP 134, 136, 137; doubly filled Comp filter 331, 334; Empty Category Principle (ECP) 117, 127; main clauses 346; NumP movement 106; P-incorporation 280; Particle Phrases 6; pied-piping within DPs 323, 325; PP modification 98; pronouns 76, 101; sloppy/non-sloppy identity 54, 58; that t effects 117, 121, 136, 348; wh-questions 130; yes/no questions 337; see also Belfast English Epée, R. 133 ergative Case marking 134, 187 Evans, G. 36 Evers, A. 261 Ewe, logophoricity 60, 62 excorporation: inseparable P prefixes in Dutch 308; licensing 259, 271; particle verbs in Dutch 300, 303 existentials, Case assignment in Bambara 177 expletive pronouns, Case assignment 164, 174, 177, 182, 184, 189 Extended Projection Principle 4, 65 Fabb, N. 260

Faroese, V-movement 294 FI see Focus Interpretation FocP see Focus Projection focus: predicate cleft 356; subject extraction in Vata 123; WCO 30 Focus Interpretation (FT) 30 Focus Projection (FocP) 360 Fortescue, M. 189 French: V-movement 261; Case assignment 164, 168, 185; causatives 276; Control from COMP 123, 127, 137; doubly filled Comp filter 336; incorporation 282; pied-piping of IP 325; PP modification 99; pronominal DPs 79, 111; subjunctive complements 58; see also Quebec French Fukui, N. 164, 323 Fula, referential/bound variables 70n2 Full Interpretation 346, 351 Garrett, S. 342 GB see Government Binding Theory Gbadi 7 ge- (Dutch) 298, 308, 310 Generalized Binding Theory (GEB) 146, 158 genitive 74 George, L. 132 German: V-movement 294; pro-drop 344; topic drop 343 gerunds, pronouns in Abe 46 Giorgi, A. 283 Gokana, logophoricity 60 Goldsmith, J. 127, 142n13 government 258; history of theory 1; theory of locality 299; see also proper government Government Binding Theory (GB) vi, 1, 7 Grimshaw, J. 283

378 INDEX

Groos, A. 187 Guéron, J. 238, 260 Haider, H. 295 Haïk, I. 25, 29 Haitian, resumptive pronouns 3, 17 Hale, K. 90, 97 Hallman, P. 333, 336 head drop 331, 341 head incorporation 299 head marking 329, 337 head movement vi, 7, 323; DPs 75, 82; incorporation 299; licensing heads 258; predicate cleft in Vata 360; pronouns in Celtic languages 88; receptor binding 315; Spec head relation 326 head movement constraint 160n7 Hendrick, R. 85 Henry, A. 336 her- (Dutch) 295, 302, 310 Hermon, G. 260 Hestvik, A. 98 Higginbotham, J. 25, 27, 37, 262 Hirschbühler, P. 127 Hoekstra, T. 182, 260, 323 Holmberg, A. 98, 100 Hornstein, N. 4, 34, 120n5, 123, 125, 125, 147, 155 Huang, J. 123, 125, 138, 146, 262 Hyman, L. 60 i-command: antecedent government 319n14; licensing 258, 285; particle verbs in Dutch 303; theory of locality 299 Icelandic: V-movement 294; logophoricity 70; object shift 111n40 idiomatic PPs, Dutch 213, 238 imperatives, silent Spec in Dutch 338 incorporation: inseperable P prefixes in Dutch 308;

licensing 258, 259, 268, 277, 284; particle verbs in Dutch 302 independent pronouns, Welsh 79, 86, 94, 96 infinitivals, Dutch PPs 240 INFL: wh-movement in Chinese 147 instrumental Case 187 intonational questions, pied-piping in French 325 Irish: NumP movement 97; pronominals 89; Spec head configuration 339 Italian: Control from COMP 133, 137; negation 341; subject extraction 349 Jackendoff, R. 64, 199, 200, 204, 219, 240 Jaeggli, O. 113n49, 117, 123, 169 Japanese, logophoricity 70 Johnson, K. 104, 238 Kaye, J. 124 Kayne, R. vi, 2, 6, 8, 123, 126, 132, 152, 198, 200, 238, 240, 260, 280, 287, 316, 323, 325, 327, 336, 349, 358 Kempschinsky, P. 59 Keyser, S.J. 310 Kinyalolo, K. 323 Kitagawa, Y. 164 kO-complements (Abe) 40, 60 Kokora, D.P. 133 Koster, J. 228, 238, 274 Koyo, wh-movement 154 Kru languages, Control from COMP 133 Kuroda, S.Y. 164 L-marking 213, 259, 267 Larson, R. 164, 261, 323, 335, 360 Lasnik, H. 37, 54, 71n10, 117, 134, 146, 155, 157, 331 LC see Leftness Condition LCA see Linear Correspondence Axiom Lebeaux, D. 260 Lefebvre, C. 360 Leftness Condition (LC) 16, 19, 25

INDEX 379

Lema, J. 260 licensing heads 7, 260, 299 Linear Correspondence Axiom (LCA) 8, 327, 328, 332 locality theory 299 locative, Case assignment in Bambara 175 locative PPs, Dutch 200, 208, 215, 218 logophoric pronouns 3, 36, 60, 69 logophoric verbs 60 long extraction: Control from COMP 123; Vata 148, 155 Longobardi, G. 283 McCloskey, J. 89, 90, 95, 97, 339 Mahajan, A. 190, 323 Mahou: Case assignment 164; wh-in situ 138 Mainland Scandinavian languages, NumP movement 98 Malay, Case assignment 186, 188, 191 Maling, J. 102, 120n4, 134 May, R. 126, 257, 262 Minimal Chain Link 5 Minimalist Program (MP) 1, 2, 8, 323 modalis Case 187 Montalbetti, M. 67 Moritz, L. 257, 325 Morris-Jones, J. 109n15 Mulder, R. 182 Mundang, logophoricity 60 n-pronouns (Abe) 36, 38, 41, 56, 61 negation: head drop in Italian 341; head drop in Spanish 342; pied-piping in Nweh 325; pronominal DPs in Breton 88 Nkemnji, M. 2, 79, 99, 233, 325, 325, 333, 336, 356, 359 nominal small clauses, Case assignment 179, 182, 184, 189 nonsloppy identity 54, 58, 66 Norwegian: V-movement 294; NumP movement 98, 106 NP movement:

Case assignment in Bambara 164, 166, 174, 184; Spec head licensing 257 numerals, Welsh 90 NumP 74, 90, 106 Nweh: predicate cleft 356, 360, 361; pronouns 79, 99 O-pronouns (Abe) 36, 38, 55, 58 Obenauer, H. 127, 128 object shift: Icelandic 111n40; Norwegian 100 overt pronouns 68 P prefixes, inseparable, Dutch 308 P receptors: inseparable P prefixes in Dutch 308; particle verbs in Dutch 302, 313 P-incorporation: particle constructions in English 280, 282; particle verbs in Dutch 302; PPs in Dutch 199, 209, 217, 228, 234, 236 P-stranding: particle verbs in Dutch 302; PPs in Dutch 199, 201, 207, 213, 236 parasitic gaps: Bijection Principle 3, 24, 27, 31; object shift in Norwegian 101 Particle Phrases (PP), Dutch 6 particles: V-second languages 295; Dutch 198, 238, 244, 302, 310 PAS-DE-QUOI 128 Path, PPs in Dutch 200, 219, 238, 240, 243 PathP, PPs in Dutch 228, 234, 243 perfective aspect, Case assignment in Bambara 164, 174 Perlmutter, D. 134 Pesetsky, D. 134, 137 pied-piping 2, 6, 323, 331; English main clauses 346; particle verbs in Dutch 302, 306; PPs in Dutch 199, 207, 210, 218, 228, 234; predicate cleft construction 359, 368;

380 INDEX

Spec head configuration 350; that t effects in English 349 Place, PPs in Dutch 200, 202, 219, 228, 240 PlaceP, PPs in Dutch 211, 212, 219, 238 Pollock, J.-Y. 6, 185, 261, 336 Postal, P. 74 postpositional PPs, Dutch 198, 217, 223, 228 PP adjuncts, wh-movement in Vata 156, 158, 159 PP modification 98 PP over V, Dutch 199, 210, 211, 218 PPs: Case in Bambara 166, 180; Dutch 197 predicate cleft constructions 8, 356 prepositional PPs, Dutch 198, 200 Principle of Full Interpretation (PFI) 177 Principle of Projection Activation (PPA) 359, 364 pro: agreement in Welsh 82, 85, 93, 95; NumP movement 79; NumP movement in English 101 PRO, Strong Crossover phenomena 23 pro-drop: agreement in Welsh 85, 94; negation in Italian 342; NumP movement 101, 107; Spec head configuration 331, 343 Procrastinate 111 progressive, Case assignment in Bambara 175 Projection Activation 331, 350 Projection Principle 164, 172, 284 pronominal binding 37; Abe 47, 55; logophoric contexts 66; sloppy/nonsloppy identity 54 pronominal DPs, Mainland Scandinavian languages 98 pronoun shift, Norwegian 100 pronouns 3, 89; distribution 73; English 76, 98, 101; as intransitive Ds 74; logophoricity in Abe 3, 36, 66; Mainland Scandinavian languages 98; modification 98; overt 73;

overt/covert 68; referential dependency 37; resumptive 3, 17; as variables 16; wh-movement in Vata 123 proper government: ECP 119, 121; wh-movement in Vata 123; see also Control from COMP Pulleyblank, D. 60 Q-float 102, 203 QR see Quantifier Raising Quantifier Pronoun flip (Q-flip) 102; see also allPS Quantifier Raising (QR), WCO 16, 27, 126 Quebec French, quoi 143 quoi (French) 127, 143 r-pronouns (Dutch) 200, 213, 228, 233, 236, 238 receptors, licensing heads 258, 262, 284, 299, 315 reduplicated pronouns, Welsh 79, 94, 96 referent, logocentric 60 Reinhart, T. 25, 30, 37, 54, 59 remnant movement 2, 8 resumptive pronouns 3, 17; Abe 55, 57; Vata 123, 137, 149, 153, 155, 157 resumptive verbs, Vata 149, 155, 157 Revised Leftness Condition (RLC) 25, 28 Riemsdijk, H. van 146, 197, 199, 202, 208, 215, 233, 308 Ritter, E. 74, 283, 323 Rivero, M.-L. 260 Rizzi, L. 4, 123, 132, 133, 186, 212, 257, 302, 333, 336, 358, 365 RLC see Revised Leftness Condition Roberts, I. 186, 260, 302 Roeper, T. 310 Rouveret, A. 97, 170, 186 SAC see Strong Accessibility Condition Sadler, L. 85, 85, 90 Sag, I. 54 SAI see Subject-Aux Inversion

INDEX 381

Saito, M. 146 Salleh, R. 186, 188 SCO see Strong Crossover phenomena scope, wh-in situ 146 Scope Condition 25 scrambling: Case assignment 182, 188; particle verbs in Dutch 316; PPs in Dutch 199, 210, 234 Sells, P.41, 55, 60, 62, 70 short wh-movement 148, 157, 169 sloppy identity 54, 56, 66 small clauses see nominal small clauses Spanish: Case assignment 169; negative head drop 342; overt/covert pronouns 67; wh-movement 169 Speas, M. 164, 323 Spec drop 341, 342 Spec head agreement 75, 96, 182 Spec head licensing 257, 299, 323, 327 Spec marking 329, 337 Specificity Condition 31 Sportiche, D. 1, 3, 4, 8, 16, 30, 36, 76, 88, 102, 107, 123, 125, 146, 164, 169, 203, 215, 257, 258, 260, 261, 299, 315, 323, 325, 329, 331, 333, 336, 337, 339, 342, 358 Stowell, T. 48, 71n10, 74, 169, 258, 262, 267, 280, 348 Strong Accessibility Condition (SAC) 27 Strong Crossover phenomena (SCO) 23, 26 Stump, G. 88 subject extraction: Dutch 119; proper government 121; Subject-Aux Inversion 117; that t effects in English 348; wh-movement in Vata 123 Subject-Aux Inversion (SAI) 117, 136 subjunctive complements, Abe 40, 44, 58 Sung, L.M. 260 Swedish, V-movement 294 Szabolcsi, A.78, 198, 323 T receptors, particle verbs in Dutch 302

T-chain 261 T-grids 262, 269, 277 T-marking 261, 262 Taraldsen, K.T. 97, 238, 281 Tense, agreement in Welsh 109n21 that t effects: doubly filled Comp filter 348; Dutch 117, 118, 119n4, 134, 137; ECP 117; English 117, 121, 134, 136, 348; proper government 121 Theta Criterion 33n18, 164, 172, 177, 178 theta-grids 262, 285 theta-position, long wh-movement in Vata 157 theta-roles: Case assignment in Bambara 171; licensing heads 261, 286; WCO 22, 33n18 theta-theory: distribution of VPs 261, 262; WCO 21, 24 topic drop 343 Torrego, E. 169 transitive verbs: Case assignment in Bambara 166, 170; ergative Case 187 Travis, L. de Mena 161n7, 258 Turkish, PP modification 99 Universal Base hypothesis 8 Uriegereka, J. 215, 258 V-in situ 294, 299, 303, 312, 316 V-incorporation 258, 268 V-movement: V-second languages 294; Case assignment 164, 168, 185; Dutch 299; licensing of heads 258, 261, 288 V-raising: Case assignment in Bambara 170; causatives in Dutch 259, 273; Dutch 259; licensing heads 258, 259; particle verbs in Dutch 302, 306; receptor binding 315 V-second 261;

382 INDEX

doubly filled Comp filters 332, 337; finite verbs in nonroot clauses 294; particle verbs in Dutch 306, 315; Spec head configuration in Dutch 338; topic drop in Dutch 343; wh-extraction in Dutch 135 Valois, D. 74, 79, 257, 283, 323 variables: Bijection Principle 16; logophoric pronouns in Abe 36; pronouns in Abe 52, 69 Vata: Case assignment 168, 185, 190; Control from COMP 123, 137; logophoricity 65; long extraction 146; predicate cleft 356, 360; resumptive pronouns 3, 19; tones 142n3, 142n8; wh-questions 123, 132, 148 ver-prefixation (Dutch) 308 Verb Second, that t violations in Dutch 117, 118, 135 verb-particle constructions 103, 280 verbal focus construction see predicate cleft constructions verbs, logophoricity 60 Vergnaud, J. 170, 186 VP-internal subject hypothesis 164 VP-preposing 268 VPs: distribution 260; see also bare VP complements VSO languages 85, 339 Wahl, A. 146, 159 WCO see weak crossover phenomena weak crossover phenomena (WCO) 16, 19, 22, 71n10 Welsh: agreement 83; clitic doubling 113n49; pronominal system 74, 79, 107; Spec head configuration 339 West Greenlandic, Case assignment 187 wh-gaps, Vata 124 wh-in situ :

Bambara 183; Chinese 138, 146; French 128; Vata 125, 131; West Greenlandic 189 wh-islands, Vata 155 wh-movement: argument/adjunct asymmetry 147; Case assignment in Bambara 174, 182; Control from COMP in French 127; doubly filled Comp filters in English 334; ECP 4, 121, 139; Italian 133; PPs in Dutch 199, 210, 211, 228, 234; Spec head licensing 257; Vata 19, 123, 131, 148 wh-phrases, predicate cleft constructions 365 wh-questions: Case assignment in Bambara 181, 182; Chinese 138; doubly filled Comp filters 332; Dutch 333; English 130, 334, 348; French 79, 127; Italian 133; predicate cleft construction 358; proper government in Vata 123; resumptive pronouns in Vata 3; Subject-Aux Inversion (SAI) 117; that t effects in English 348 wh-raising: Chinese 138; French 128, 138; Vata 126, 138 wh-traces: argument/adjunct asymmetry 146; Control from COMP 123, 133; Dutch 134; ECP 121; long extraction in Vata 146; resumptive pronouns in Vata 124; Subject-Aux Inversion in English 117 whether (English) 335, 337 Williams, E. 31, 54, 138 Williams, S.J. 79, 90, 91 ye-complements (Abe) 60, 66

INDEX 383

yes/no questions: Dutch 338; English 337; French 325 Yiddish, V-movement 294 Yoon, J. 333, 336 Yoruba, logophoricity 60 Zaenen, A. 120n4, 134 Zagona, K. 260, 265 Zanuttini, R. 257 Zwart, Jan Wouter 215

The Syntax of Specifiers and Heads: Collected Essays of Hilda J. Koopman (Routledge Leading Linguists 3)

The Syntax of Specifiers and Heads: Collected Essays of Hilda J. Koopman (Routledge Leading Linguists 3)

Towards an Elegant Syntax (Routledge Leading Linguists)

Theoretical Comparative Syntax: Studies in Macroparameters (Routledge Leading Linguists)

Unification of Morphology and Syntax: Investigations into Italian, Romantch and Albanian Dialects (Routledge Leading Linguists)

The Routledge Handbook of Syntax

Generative Grammar: Theory and its History (Routledge Leading Linguists)

Minimalist Investigations in Linguistic Theory (Routledge Leading Linguists)

Collected Essays of George Orwell

Parameters and Functional Heads: Essays in Comparative Syntax (Oxford Studies in Comparative Syntax)

Collected Essays [essays-47]

Collected Essays

Collected Essays

Collected Essays

Collected Essays

Collected Essays

Collected Essays

Collected Essays

The Collected Essays of A. K. Ramanujan

The Collected Essays of Robert Creeley

The Heads of Cerberus

Configurations of Sentential Complementation: Perspectives from Romance Languages (Routledge Leading Linguists)

Collected Works 3 - The Mirror Of Relationship

Productivity Growth, Inflation, and Unemployment: The Collected Essays of Robert J. Gordon

Hilda and Pearl

Pointing the Way: Collected Essays

The Political and Social Theory of Max Weber: Collected Essays

The Syntax of French (Cambridge Syntax Guides)

Admitting the Holocaust Collected Essays

Hilda and Zelda

The Syntax of Specifiers and Heads: Collected Essays of Hilda J. Koopman (Routledge Leading Linguists 3)