Scrambling, Remnant Movement, and Restructuring in West Germanic (Oxford Studies in Comparative Syntax)

Scrambling, Remnant Movement, and Restructuring in West Germanic Roland Hinterhölzl OXFORD UNIVERSITY PRESS Scrambl...

Author: Roland Hinterholzl

30 downloads 871 Views 1MB Size Report

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

Report copyright / DMCA form

DOWNLOAD

Scrambling, Remnant Movement, and Restructuring in West Germanic

Roland Hinterhölzl

OXFORD UNIVERSITY PRESS

Scrambling, Remnant Movement, and Restructuring in West Germanic

OXFORD STUDIES IN COMPARATIVE SYNTAX Richard Kayne, General Editor The Higher Functional Field: Evidence from North Italian Dialects Cecilia Poletto The Syntax of Verb-Initial Languages Edited by Andrew Carnie and Eithne Guilfoyle Parameters and Universals Richard Kayne Portuguese Syntax: New Comparative Studies Edited by João Costa XP-Adjunction in Universal Grammar: Scrambling and Binding in Hindi-Urdu Ayesha Kidwai Inﬁnitive Constructions: A Syntactic Analysis of Romance Languages Guido Mensching Subject Inversion in Romance and the Theory of Universal Grammar Edited by Aafke Hulk and Jean-Yves Pollock Subjects, Expletives, and the EPP Edited by Peter Svenonius A Uniﬁed Theory of Verbal and Nominal Projections Yoshiki Ogawa Functional Structures in DP and IP: The Cartography of Syntactic Structures, Volume 1 Edited by Guglielmo Cinque

Syntactic Heads and Word Formation Marit Julien The Syntax of Italian Dialects Christina Tortora The Morphosyntax of Complement-Head Sequences: Clause Structure and Word Order Patterns in Kwa Enoch Oladé Aboh The Structure of CP and IP: The Cartography of Syntactic Structures, Volume 2 Edited by Luigi Rizzi The Syntax of Anaphora Ken Saﬁr Principles and Parameters in a VSO Language: A Case Study in Welsh Ian G. Roberts Structures and Beyond: The Cartography of Syntactic Structures, Volume 3 Edited by Adriana Belletti Movement and Silence Richard S. Kayne Restructuring and Functional Heads: The Cartography of Syntactic Structures, Volume 4 Guglielmo Cinque Scrambling, Remnant Movement, and Restructuring in West Germanic Roland Hinterhölzl

Scrambling, Remnant Movement, and Restructuring in West Germanic

Roland Hinterhölzl

1 2006

3

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

Copyright © 2006 by Oxford University Press, Inc. Published by Oxford University Press, Inc. 198 Madison Avenue, New York, New York 10016 www.oup.com Oxford is a registered trademark of Oxford University Press. All rights reserved. No part of this publication may be reproduced, stored in a retrieval system, or transmitted, in any form or by any means, electronic, mechanical, photocopying, recording, or otherwise, without the prior permission of Oxford University Press. Library of Congress Cataloging-in-Publication Data Hinterhölzl, Roland. Scrambling, remnant movement, and restructuring in west Germanic / Roland Hinterhölzl. p. cm. A revision of the author’s thesis (Ph.D.)—University of Southern California, 1999. Includes bibliographical references and index. ISBN-13 978-0-19-530820-4; 978-0-19-530821-1 (pbk.) ISBN 0-19-530820-4; 0-19-530821-2 (pbk.) 1. Germanic languages—Syntax. 2. Germanic languages— Verb. I. Title PD369.H56 2005 430'.045—dc22 2005055492

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

To Mara

This page intentionally left blank

ACKNOWLEDGMENTS

I thank Richard Kayne for his support in enabling me to publish this work in Oxford’s comparative syntax series and two anonymous reviewers for extensive comments and suggestions that helped me a great deal in working out my ideas. This book constitutes a much revised and extended version of my USC dissertation in 1999 on restructuring inﬁnitives. Chapters 3, 4, and 5 are based on this work. I thank Martin Prinzhorn and Hubert Haider in Vienna and Joseph Aoun, Barry Schein, and Jean Roger Vergnaud in Los Angeles for teaching me what linguistics in general and comparative syntax in particular are about. I am also grateful to Marga Reis for her interest in my work and for encouraging me to get it published. Since my return to Europe, I have been working on the proper characterization of scrambling (chapter 2), the ﬁne structure of verb clusters (chapter 6), and the complex relation between restructuring and VP-topicalization (chapter 7). A preliminary version of chapter 2 has been published in Hinterhölzl (2004). The materials of chapter 6 and chapter 7 have been presented at Tilburg University in November 2004, at the ZAS in Berlin in January 2005, at the Incontro di Grammatica Generativa in Rome in February 2005, and at the GLOW workshop in Geneva in April 2005. I thank these audiences for valuable comments and feedback. I would like to thank Sjef Barbiers, Hans Bennis, Hans Broekhuis, Hilda Koopman, and Jan-Wouter Zwart for their patience in checking and discussing Dutch data with me. All remaining errors are mine. I am especially grateful to Liliane Haegeman for her friendship and her invaluable support along the way.

This page intentionally left blank

CONTENTS

1.

Introduction 1.1 Basic Observations 1.2 The Basic Syntactic Phenomena of West Germanic 1.3 Coherent Inﬁnitives 1.4 Remnant Movement 1.5 Theoretical Assumptions

2.

Scrambling 2.1 2.2 2.3 2.4 2.5 2.6 2.7 2.8

3 3 6 14 21 28

33

The Grammatical Nature of Scrambling The Properties of Scrambling in German Scrambling and Optional Movement The Trigger Problem Coping with Optionality Feature Checking and Scope Addressing Vikner’s Problem Conclusions

34 35 44 46 50 56 61 63

3. Coherent Inﬁnitives in Dutch and West Flemish

64

3.1 Coherent Inﬁnitives in Dutch 3.2 Coherent Inﬁnitives in West Flemish 3.3 Conclusions

4. A Remnant Movement Account of Restructuring 4.1 The Verb-Final Position in Inﬁnitives 4.2 Evidence for Licensing Movement out of the VP

64 78 88

89 89 91

x

CONTENTS

4.3 An XP-Movement Account of Restructuring 4.4 Toward a Theory of Sentential Complementation

5. Coherent Inﬁnitives in German 5.1 5.2 5.3 5.4 5.5 5.6

Evidence for Biclausality Verbs That Take To-Inﬁnitives as Complements The Idiosyncractic Properties of Coherent To-Inﬁnitives A Biclausal Account of Coherent To-Inﬁnitives Explaining the Properties of Coherent To-Inﬁnitives Conclusions

6. Verbal Complexes and the Syntax of IPP-Complements 6.1 Left-Branching Verb Clusters in German 6.2 The Syntax of IPP-Constructions 6.3 Right-Branching Verb Clusters in German 6.4 Overview of the Structure of Verb Clusters

7. Open Issues: Extraposition, VP-Topicalization, and the Status of Gerunds 7.1 7.2 7.3 7.4 7.5

8.

Verbal Complexes and the Distribution of CP-Complements Topicalization of Verb Projections Toward a Movement Account of VP-Topicalization The Status of the Gerund Conclusions

Summary and Conclusions 8.1 8.2 8.3 8.4 8.5 8.6

Scrambling and Optionality A VO-Based Account of Verb Raising and Verb Projection Raising Coherent Inﬁnitives in German and the Issue of Monoclausality The IPP-Eﬀect and the Uniﬁed Account of Verb Clusters in West Germanic Extraposition, VP-Topicalization, and the Status of Gerunds The Connection between Scrambling, Remnant Movement, and Restructuring

101 117

129 131 136 142 147 152 159

160 162 165 175 179

185 186 190 198 208 211

213 215 215 221 224 227 229

Notes

233

References

243

Index

251

Scrambling, Remnant Movement, and Restructuring in West Germanic

This page intentionally left blank

1

Introduction

T

his book is dedicated to an exploration of the empirical properties and the theoretical implications of three salient phenomena of West Germanic, namely, remnant movement, scrambling, and restructuring. The following section provides a brief outline of the interdependence of these phenomena.

1.1 Basic observations 1.1.1

The connection between remnant movement and scrambling

The term remnant movement was coined by Den Besten and Webelhuth (1987) to account for a peculiarity of VP-fronting in German. Consider example (1). It seems that what has been fronted (into [Spec,CP]) is just the verbal head. In order to reconcile this fact with the well-known Verb Second Constraint, which requires that the ﬁnite verb in matrix clauses is preceded by one (and only one) maximal phrase, Den Besten and Webelhuth propose to analyze the structure in (1) parallel to the case in (2), where the entire VP has been fronted. (1)

gelesen hat Hans das Buch read has Hans the book

(2)

das Buch gelesen hat Hans the book read has Hans 3

4

SCRAMBLING , REMNANT MOVEMENT , AND RESTRUCTURING IN WEST GERMANIC

What makes (1) diﬀerent from (2) in their analysis is the fact that in (1) the direct object has scrambled out of the VP prior to VP-fronting. The subsequent operation of VP-fronting will then move whatever remains in the VP, the so-called remnant category, into [Spec,CP], as is illustrated in (3). Thus, it seems that remnant movement of the VP is dependent on scrambling, which evacuates the VP of its internal argument. Since English does not allow scrambling, the parallel structure to (1) in English is correctly ruled out. If the VP is fronted in English, the object has to move along with the verb (3b–c). (3)

a. [CP[VP tSCR gelesen] hat [IP Hans das BuchSCR t VP]] b. John promised to read something *. . . and read he did the book c. John promised to read the book . . . and read the book he did

1.1.2

The connection between remnant movement and restructuring

In German, VP-fronting is not restricted to participles but may also aﬀect inﬁnitives. Example (4a–b) shows that VP-fronting can apply to bare inﬁnitives as well as to toinﬁnitives. However, there is one restriction to this operation. VP-fronting may aﬀect an inﬁnitive only if the matrix verb belongs to the class of restructuring verbs. This is illustrated by the contrast in (4c–d), where the matrix verb try, a typical restructuring verb, licenses fronting of the embedded VP, while the verb deny, a typical representative of non-restructuring verbs, fails to license the respective operation. Thus, it seems that remnant VP-fronting is dependent not only on scrambling but also on restructuring. (4)

a. kaufen will er das Buch buy wants he the book b. zu kaufen versprach er das Buch to buy promised he the book c. zu kaufen hat er das Buch versucht to buy has he the book tried d. *zu kaufen hat er das Buch abgelehnt to buy has he the book refused

1.1.3

The connection between scrambling and restructuring

Scrambling in German may permute the arguments of the verb. For instance, in (5a) the direct object has scrambled across the subject. It is important to note that this operation is clause-bound. As (5b) shows, the embedded object may not scramble

INTRODUCTION

5

across the matrix subject. As (5c) shows, this also holds for embedded nonﬁnite clauses. However, in restructuring contexts, the embedded object can do exactly that, namely, scramble up to a position above the matrix subject, as is illustrated in (5d). (5)

a. weil den Max jeder kennt since the Max-ACC everyone-NOM knows b. *weil den Max Peter glaubt dass jeder kennt since the Max-ACC the Peter-NOM believes that everyone-NOM knows c. *weil den Max jeder zu kennen bedauert since the Max-ACC everyone-NOM to know regrets d. weil den Max jeder zu kennen glaubt since the Max-ACC everyone-NOM to know believes

Thus, scrambling has often been used as a diagnostic for detecting restructuring inﬁnitives. At this point we hit a bifurcation and two ways of explaining the contrast in (5c–d) seem to suggest themselves. First, one may hold that scrambling is always strictly clause-bound, implying that restructuring constructions are monoclausal structures (in this sense, scrambling is a diagnostics for the clause-mateness in inﬁnitival constructions). Under these assumptions, studying restructuring constructions would amount to deriving a theory of which (verbal) predicates may combine within a single clause (cf. Cinque [2001], Wurmbrand [2001]). Alternatively, one may assume that restructuring constructions are biclausal. Under this assumption, studying restructuring involves investigating the conditions under which an element may scramble from one clause into the next one up—in other words, under which conditions an element may undergo so-called long-distance scrambling. In this sense, scrambling is a diagnostics for the transparency of inﬁnitival constructions. Once the direct object of the embedded inﬁnitive has undergone scrambling, the remnant inﬁnitival phrase, however large it may be, probably just a VP in the monoclausal approach and possibly an entire CP in the biclausal approach, is free to undergo fronting, providing a possible account for the contrast in (4c–d), as is illustrated in (6). (6)

a. [hat er versucht [das Buch zu verkaufen]]

base structure

b. [hat er [das Buch]SCR versucht [t SCR zu verkaufen]] long-distance scrambling c. [[t SCR zu verkaufen] hat er das Buch versucht]

fronting of the inﬁnitival

In this book, I will carefully investigate the phenomenon that has been called scrambling (chapter 2) and argue that scrambling is a cover term for various operations with diﬀerent properties and triggers. It will be shown that one of these operations, so-called S-scrambling, is actually not clause-bound, but that all other scrambling operations are strictly clause-bound. Furthermore, I will argue on the basis of a thorough analysis of remnant movement and its restrictions (see section 1.4 later) that Den Besten

6

SCRAMBLING , REMNANT MOVEMENT , AND RESTRUCTURING IN WEST GERMANIC

and Webelhuth’s famous account of data like (1) is untenable, since scrambling cannot be taken to create the remnant category that is the input for VP-fronting. In (4c–d) earlier, I have illustrated that VP-fronting, an operation of remnant movement, is dependent on restructuring. In turn, I will argue that restructuring is dependent on remnant movement. The goal of this book is to argue that restructuring is based on remnant movement of large projections, both of the argument domain and of the licensing domain, of the inﬁnitival clause (chapter 4). More speciﬁcally, I will show that S-scrambling does not play a role in restructuring. Thus, accounts of restructuring that assume that the constituents of the embedded inﬁnitive undergo long-distance movement into the matrix clause are doomed. I will also show that what has been fronted in (4c) is not an inﬁnitival clause (out of which the direct object has been scrambled into the matrix clause) but constitutes just an inﬁnitival VP. These observations, which I consider the empirical core of the book, seem to support a monoclausal account of restructuring constructions. However, I will argue (in chapter 5) that only a biclausal approach can provide a comprehensive and coherent account of all cases of restructuring in German, making necessary an account of restructuring in terms of remnant movement that splits up the embedded clause in various parts and has them licensed in dedicated positions in the matrix clause. Before we take a closer look at restructuring, I have to introduce a number of phenomena and concepts that are essential for the description of the syntactic structure of West Germanic as well as for the understanding of the discussion of restructuring inﬁnitives.

1.2 The basic syntactic phenomena of West Germanic In this section, I provide an informal account of the basic notions and processes that govern the syntax of West Germanic. Though the diﬀerent phenomena are introduced and illustrated by German data, the relevant generalizations carry over to the other West Germanic languages and dialects. 1.2.1

Verb second

First, note that German is a so-called Verb Second language. This means that the ﬁnite verb in matrix clauses moves from its base position, which is generally taken to be the position it occupies in an embedded clause (7a), to a sentence-initial position that has generally been assumed to be C0 (Den Besten 1977/1983). This movement only aﬀects the ﬁnite part of a verb or verbal complex, leaving behind nonﬁnite verbs as in (7b) or verbal particles if they are separable (cf. [7c], which contains the particle verb mit-teilen). German is called Verb Second (V2) because in many sentence types the verb in this position must be preceded by another constituent, which is generally analyzed as occupying [Spec,CP] (7b). (7)

a. weil Hans der Maria das Buch gegeben hat since Hans to Mary the book given has

INTRODUCTION

7

b. *(Gestern) hat Hans der Maria das Buch gegeben [t] yesterday has Hans to Mary the book given

The V2-structure in (7b) neatly displays the three topological ﬁelds, the Vorfeld (preﬁeld), Mittelfeld (middle ﬁeld), and Nachfeld (postﬁeld), that have become standard terminology in describing the syntactic structure of German sentences. A constituent that precedes the ﬁnite verb in C0 occupies the Vorfeld or is said to be topicalized. Elements that occur between the ﬁnite verb and the nonﬁnite verb (or sometimes the verb-particle in case there is no auxiliary) are said to make up the middle ﬁeld of the German sentence. A constituent that occurs after the nonﬁnite verb (or a verb-particle) occupies the Nachfeld and is assumed to be extraposed. For instance the subject Hans is topicalized in (7c), the sequence der Maria gestern forms the middle ﬁeld of the sentence (7c), and the sentential complement dass sie krank ist is extraposed. (7)

c. Hans teilte der Maria gestern mit [t] dass sie krank ist Hans shared Mary-DAT yesterday with that she sick is ‘Hans shared with Mary (told Mary) that she is sick’

1.2.2

Scrambling and topicalization

Second, I want to introduce the notion of scrambling in an informal way. Chapter 2 provides an extensive discussion of the formal nature of scrambling; here it suﬃces to characterize scrambling as a movement process that allows for the re-arrangement of constituents in the middle ﬁeld, as is illustrated in (10) later. This notion of scrambling presupposes that there is an (unmarked) basic word order, from which alternative orders are derived via scrambling and that this base order is identiﬁable with independent criteria (cf. Höhle [1982], Lenerz [1977], for some discussion of these issues). One of the tests for identifying base orders is the naturalness of certain orders in out-of-the-blue contexts, as is illustrated in (8). (8)

Was ist passiert? Ich habe gehört What happened? I have heard a. dass der Hans die Maria getroﬀen hat that Hans-NOM Maria-ACC met has b. dass die Maria der Hans getroﬀen hat that Maria-ACC Hans-NOM met has

The order in (8a) is unmarked in this context and thus counts as base order. The order in (8b) is inappropriate in this context and thus counts as a marked word order that is derived from the basic word order. This test reveals that basic word orders in German are verb class–dependent (cf. Haider and Rosengren [1998]). This means that the base order cannot simply be determined by the Case properties of the arguments of the verb, say NOM > DAT > ACC, but is dependent on the speciﬁc (thematic) class a verb belongs to. These classes are illustrated in the following table, adopted from Haider and Rosengren (1998).

8

SCRAMBLING , REMNANT MOVEMENT , AND RESTRUCTURING IN WEST GERMANIC

(9) a. Nom > Acc: bedauern ‘regret’, interpretieren ‘interpret’ b. Acc > Nom: interessieren ‘interest’, imponieren ‘impress’ c. Nom > Dat: helfen ‘help’, gratulieren ‘congratulate’ d. Dat > Nom: gefallen ‘appeal/please’, fehlen ‘lack’ e. Nom > Dat > Acc: anvertrauen ‘entrust’, verbieten ‘forbid’ f. Nom > Acc > Dat: aussetzen ‘expose’, unterordnen ‘subordinate’

Returning to scrambling and assuming that (10a) exempliﬁes the basic word order, scrambling can be described as a process that moves a constituent (preferably an argument) across arguments and adverbs to the top of the middle ﬁeld. For instance, in (10b) the subject has been scrambled across both adverbs. In (10c), in addition to the scrambling in (10b), the direct object has been scrambled across the indirect object and the lower adverb, and so on. As should be clear from Example (10), the alternations I list only display a small subset of all the scrambling possibilities: In (10) I only gave the permutations that have the subject in the highest position in the middle ﬁeld, then there are the permutations that have the subject in second position in the middle ﬁeld and so on, not all of which, but most of which, yield felicitous sentences in one context or other. (10) a. weil vermutlich gestern Hans der Maria das Buch gegeben hat since presumably yesterday Hans to Maria the book given has b. weil Hans vermutlich gestern der Maria das Buch gegeben hat c. weil Hans vermutlich das Buch gestern der Maria gegeben hat d. weil Hans vermutlich gestern das Buch der Maria gegeben hat e. weil Hans vermutlich der Maria das Buch gestern gegeben hat f. weil Hans der Maria das Buch vermutlich gestern gegeben hat

For our purposes, it is irrelevant at this point what motivates scrambling or which alternations are most felicitous in which contexts. It is also irrelevant at this point whether scrambling is A- or A'-movement and how it is formally to be distinguished from topicalization. What is important for us here is that we assume that the alternations in (10) are derived by movement that we call scrambling and do not constitute diﬀerent patterns that can be base generated, as has been proposed repeatedly by diﬀerent scholars such as Fanselow (2001), Haider (1994), and Neeleman (1994a). To summarize, we characterize movement into the middle ﬁeld as scrambling and movement into [Spec,CP] as topicalization. It is important to note that this type of scrambling is clause-bound. Example (11) shows, while there is long-distance topicalization (11a), there is no process of long-distance scrambling in German (11b). In (11a), the embedded object has been moved into the topic position of the matrix clause and the result is ﬁne. Scram-

INTRODUCTION

9

bling of the embedded object into the middle ﬁeld of the matrix clause, however, results in ungrammaticality. (11) a. Den Mann hat Peter geglaubt, dass Maria t geküsst hat the man-ACC has Peter believed that Mary kissed has ‘It was the man who Peter believed that Mary kissed’ b. *Peter hat den Mann geglaubt, dass Maria t geküsst hat Peter has the man-ACC believed that Mary kissed has

Scrambling and topicalization also diﬀer in the following respect: Scrambling seems to be restricted to arguments, while almost any (maximal) category, with the exception of IP, can be topicalized. Predicative elements, like predicate nominals (12a), heads of small clauses (12c) and idiomatic expressions (12e) cannot be scrambled, while the same elements can be topicalized quite naturally (cf. [12b, d, f]). Even verb-particles when they are separable from the verb, as can be detected from their behavior under V2 (whether they go along with the verb or stay behind in the base position), may be topicalized if the contribution of the particle to the meaning of the complex verb is transparent, while scrambling of verb-particles results in strong ungrammaticality (cf. [13b–c]). The particle weg (‘away’) is a separable verb preﬁx as can be seen from the alternation in (13a): When the main verb is ﬁnite it undergoes V2 and leaves the particle behind. (12) a. ??Er hat einen Idioten gestern seinen Freund genannt he has an idiot-ACC yesterday his friend called ‘He called his friend an idiot yesterday’ b. Einen Idioten hat er gestern seinen Freund genannt an idiot has he yesterday his friend called c. ??Er hat grün gestern den Zaun gestrichen he has green yesterday the fence painted ‘He painted the fence green yesterday’ d. Grün hat er gestern den Zaun gestrichen green has he yesterday the fence painted e. *Sie ist ihm ins Wort noch nie gefallen she is him into-the word yet never fallen ‘So far she has never interrupted him’ f. Ins Wort ist sie ihm noch nie gefallen into-the word is she him yet never fallen (13) a. Er ist noch nie weggelaufen/Er lief noch nie weg he is not yet away-run/He ran not yet away

10

SCRAMBLING , REMNANT MOVEMENT , AND RESTRUCTURING IN WEST GERMANIC

b. Weg ist er gelaufen (und ist nie wieder zurückgekommen) away is he run (und is never again back-come) ‘He ran away (and has never come back again)’ c. ??Er ist weg noch nie gelaufen he is away not yet run ‘So far he has never run away’ d. *Gelaufen ist er noch nie weg run is he yet never away

It should be pointed out that the topicalizability of a separable preﬁx (particle) is a bit surprising. When the ﬁnite verb undergoes V2 to C0, [Spec,CP] has to be ﬁlled. X-bar theory prescribes that it be ﬁlled with a maximal projection. What seems to have been topicalized in (13b) is a head or even part of a head, if we analyze the otherwise inseparable string weg-laufen as a complex head. However, if we assume that a separable preﬁx starts out as an independent syntactic unit that heads its own projection that can either cliticize onto the verb by head movement or move as a maximal projection (when it receives some kind of contrastive stress), then the fact in (13b) can be brought in line with standard assumptions about topicalization. Finally, it should be noted that a particle verb cannot be topicalized leaving behind its separable preﬁx, as is shown in (13d). 1.2.3

Extraposition

It is important to note that the term extraposition, though we intend to use it as a pretheoretic notion to refer to constituents that occupy the postﬁeld, has been used to describe a movement operation in what I will call the standard theory of the phrase structure of the West Germanic languages. The standard theory assumes that these languages are head-ﬁnal and analyzes constituents to the right of the nonﬁnite verb as being extraposed, that is, moved to this position by right-adjunction to VP or IP from its base position that precedes the verb that governs it. Example (14a) shows the base structure, where a complex sentential complement precedes the matrix verb that governs it. In (14b), the sentential complement of the matrix verb has been extraposed with the lower inﬁnitival complement remaining in its (preverbal) base position. In (14c), both the complement of the matrix verb and the complement of the inﬁnitival have been extraposed, that is, right-adjoined to the immediately dominating VP/IP note. It is important to note that extraposition is clause-bound. In (14d), the lower inﬁnitival clause is moved across the dominating inﬁnitival clause and right-adjoined to the matrix VP/IP, resulting in ungrammaticality. Viewing extraposition as derived by movement, we may assume that this operation may not cross a CP-node.1 (14) a. dass der Lehrer [VP die Kinder [CP [CP die Aufgaben zu lösen] zu versuchen] bestärkt] that the teacher the children the problems to solve to try encourages ‘that the teacher encourages the children to try to solve the problems’

INTRODUCTION

11

b. dass der Lehrer [VP [VP die Kinder t bestärkt] [CP die Aufgaben zu lösen zu versuchen]] c. dass der Lehrer [VP [VP die Kinder t1 bestärkt] [CP1 [VP [VP t2 zu versuchen] [CP2 die Aufgaben zu lösen]]]] d. *dass der Lehrer [VP [VP die Kinder [CP t zu versuchen] bestärkt] [die Aufgaben zu lösen]]

1.2.4

Verb-ﬁnalness

As we have seen earlier, in embedded clauses the ﬁnite verb occurs in a sentenceﬁnal position (15a). Nonﬁnite verbs follow their complements (15b) and generally precede the ﬁnite verb in clause-ﬁnal position (15c). (15) a. weil Hans gestern das Buch las since Hans yesterday the book read b. Hans hat gestern das Buch gelesen Hans has yesterday the book read c. weil Hans gestern das Buch gelesen hat since Hans yesterday the book read has

When we compare this state of aﬀairs with the positioning of ﬁnite and nonﬁnite verbs in a VO language like English, there are two options to account for these diﬀerences. The ﬁrst option consists in assuming parametric variation in the head complement order and represents the standard approach to the syntax of the West Germanic OV languages. In this approach, the lexical and functional projections in the IPdomain are assumed to be head-ﬁnal. Thus, the embedded verb-ﬁnal position in (15a) is compatible either with the lack of V-movement (16a) or with rightward V-movement to a functional head within IP (16b). Furthermore, because of the right-headedness of the VP, an argument that precedes the verb may be analyzed as occupying its base position in the VP or as having scrambled out of the VP to a position in the middle ﬁeld (cf. [16b] and [16c]). A considerable amount of work in German linguistics, culminating in Diesing’s (1992) mapping hypothesis, was dedicated to the question of which factors determine whether scrambling out of the VP does apply or does not apply. (16) a. [CP weil [IP Hans gestern [VP das Buch las]]] b. [CP weil [IP Hans gestern [VP das Buch ti] lasi ]] c. [CP weil [IP Hans gestern [das Buch]j [VP tj ti] lasi]] since Hans yesterday the book read

The second option is to adopt the Universal Base Hypothesis (Kayne 1994), according to which all phrase structure is head-initial. In this approach it is assumed

12

SCRAMBLING , REMNANT MOVEMENT , AND RESTRUCTURING IN WEST GERMANIC

that complements in German like in English are base generated following the selecting head and move to a licensing position to the left of the selecting head in the course of the derivation (cf. Zwart [1993]). The derivation of (15a) is illustrated in (17). In this approach, licensing movement is taken to derive the unmarked or neutral word order (in this case, the direct object moves to an Agreement head to check its Case, as I will argue later), before scrambling may further aﬀect constituents in the middle ﬁeld to derive more specialized or marked word orders. Therefore, this account predicts that scrambling is to be distinguished from movement out of the VP. I will take up this issue in section 1.4 on remnant movement. (17) a. [CP weil [IP Hans gestern [VP las das Buch]]] since Hans yesterday read the book b. [CP weil [IP Hans gestern [das Buch]j [VP las tj]]] since Hans yesterday the book read

Note that in this approach, leftward movement must not only aﬀect nominal arguments but also all VP-internal (verbal and non-verbal) predicates. In the case at hand, the participle in (15b–c) must move leftward to derive the unmarked word order PART > AUX. A number of questions arise in this approach: First there is the question of where the ﬁnite verb in embedded clauses is located. It could be in its base position in the VP, as is assumed in (17b), or in a sentence medial I-position, as has been argued for by Haegeman (2001, 2002b). In the latter approach, movement of the ﬁnite verb to a medial Tense-head is followed by remnant movement of the VP to an even higher position in the middle ﬁeld to derive the basic order arguments > verb. Throughout this book, I will adopt the former approach and assume that the verb in embedded clauses stays in the VP and can maximally move to a low functional position, below T (called Aspect, which will be introduced and motivated later). Second, the question arises of what the positions are to which VP-internal arguments and predicates (and VP-remnants in the account of Haegeman) move. And most important, what are the motivations and triggers for all these movements? In the remainder of the book, I will try to provide satisfactory answers to these questions by investigating the properties of verbal complexes (chapter 3 and 4) as well as the distribution of arguments and adjuncts in the middle ﬁeld (chapter 2). The general procedure will be the following: I will ﬁrst adopt the standard approach to restructuring, scrambling, and remnant topicalization and point out the diﬃculties it faces in accounting for these phenomena. In exploring the alternative approach, I will provide empirical evidence for leftward movement of VP-internal material and develop an antisymmetric account of the syntactic core phenomena of the West Germanic OV languages. In the process of developing the Universal Base approach, I will often provide analyses of phenomena in both accounts, switching back and forth between them in order to evaluate their advantages and disadvantages. Thus, it is important

INTRODUCTION

13

for the reader to keep in mind the basic tenets of the two approaches as outlined earlier. 1.2.5

Comparing the standard approach and the UBH approach

To summarize, in the standard approach the West Germanic OV languages are analyzed as head-ﬁnal. The following processes are relevant for the description of a sentence in West Germanic: Verb Second moves the ﬁnite verb into C0 in matrix clauses (and certain complementizerless embedded clauses). Topicalization moves a constituent into the Vorfeld, which is identiﬁed as [Spec,CP]. Scrambling is an operation that allows for the rearrangement of syntactic constituents in the middle ﬁeld. In the standard approach, scrambling is analyzed as an operation that moves constituents out of the VP. Extraposition moves a constituent into the Nachfeld. The core candidates for extraposition are embedded clauses. These elements are analyzed as being right-adjoined to VP or IP. In the approach that endorses the UBH, the West Germanic OV languages are analyzed as head-initial languages, in which VP-internal material, both arguments and predicates, is moved leftward into licensing positions in the middle ﬁeld (but see Haegeman 2001 and 2002b for an interesting alternative). Scrambling in this approach must be distinguished from licensing movement. The latter moves material out of the VP to derive the unmarked word order, while the former aﬀects constituents in the unmarked word order to derive more specialized or marked word orders. Extraposition and rightward movement in general are disallowed. While in the standard approach the ﬁnite verb in embedded clauses may be taken to stay in the VP or to move to a clause ﬁnal I-position, I will be adopting an antisymmetric approach in which the ﬁnite verb stays low and does not move to the clause-medial I-position. (18)

Standard Approach CP Topicalization

C' IP

C IP

Extraposition IP I'

SU

I' VP

I V

Scrambling Vorfeld

Mittelfeld

Nachfeld

14 (19)

SCRAMBLING , REMNANT MOVEMENT , AND RESTRUCTURING IN WEST GERMANIC

UBH Approach CP C' C Scrambling

IP I' SU I Licensing

VP V

Having familiarized ourselves with the basic tenets of the syntactic structure of German, we are now able to discuss (and appreciate) the properties that distinguish restructuring from non-restructuring inﬁnitives.

1.3 Coherent infinitives The West Germanic SOV languages German, Dutch, Frisian, and their numerous dialects have two types of inﬁnitival complements. Like English, these languages have non-coherent,2 that is, sentential, inﬁnitival complements that behave like ﬁnite embedded clauses with respect to extraction. Like Italian, they also have so-called coherent or restructuring inﬁnitival complements that are transparent for several types of extraction processes. For instance, the direct object of the embedded verb can be moved into the matrix clause in restructuring inﬁnitives (20a) but not in sentential inﬁnitives (20b). (20) a. weil die Maria der Hans [t zu besuchen] versprach since the Maria-ACC the Hans-NOM to visit promised ‘since Hans promised to visit Maria’ b. *weil die Maria der Hans den Peter [t zu besuchen] bat since the Maria-ACC the Hans-NOM the Peter-ACC to visit asked ‘since Hans asked Peter to visit Maria’

There are two types of explanations in the literature for the transparency of restructuring inﬁnitives. Many scholars (cf. Haider [1993], Napoli [1981], Rochette [1988], Rosen [1989, 1990], Rutten [1991]) have assumed that restructuring inﬁnitives are monoclausal structures. In this account, the problem of transparency disappears trivially, though it reappears in the form of the question of under which circumstances two main verbs can project a single clause (see Cinque [2001] and Wurmbrand [2001] for an interesting new answer to this question). In biclausal accounts, their transpar-

INTRODUCTION

15

ency has been traditionally explained with the assumption that either restructuring inﬁnitives are not full CP-complements or Verb Raising (VR) opens the domain for long-distance scrambling (LDS). In this section, I want to discuss only the most crucial properties that distinguish coherent from incoherent inﬁnitives and reserve more language-speciﬁc properties of coherent inﬁnitives for discussion in later sections that are dedicated to an in-depth analysis of these constructions in the individual languages. Here I want to outline the core properties of coherent inﬁnitives, which will provide us with a ﬁrst circumscription of the problem of restructuring. The most salient properties of coherent inﬁnitival constructions are the following: First, coherent inﬁnitives, as opposed to other (sentential and non-sentential) arguments, show a very restricted distribution in the sentence. Second, as already illustrated, coherent inﬁnitives are transparent for several types of extraction processes. Third, the formation of a coherent inﬁnitival construction is typically, though not necessarily, associated with a morphological eﬀect that has become known as IPP (inﬁnitivus pro participio)-eﬀect.

1.3.1

The immobility of coherent inﬁnitives

I will ﬁrst discuss the restricted distribution of coherent inﬁnitives. Coherent inﬁnitives, as opposed to incoherent ones, cannot be extraposed. Subject-raising verbs like scheinen (‘seem’) obligatorily require a coherent inﬁnitival complement, the extraposition of which results in ungrammaticality (21a). (21) a. *dass Hans schien [sich zu rasieren] that Hans seemed himself to shave b. dass Hans versprach [sich zu rasieren] that Hans promised himself to shave

Subject control verbs like versprechen (‘promise’) may take a coherent or incoherent inﬁnitival complement. Extraposition of the incoherent inﬁnitival complement is ﬁne (21b). The fact that incoherent inﬁnitives may extrapose has been taken to show that these inﬁnitives are full CPs, since it is assumed that only full-ﬂedged sentential projections can be extraposed. Along the same lines, the fact that coherent inﬁnitives cannot be extraposed has been taken as evidence that these inﬁnitivals are smaller than CPs. They are often analyzed as IPs or sometimes even as VPs (cf. Broekhuis et al. [1995]). At this point it is important to note that incoherent inﬁnitives are not restricted to the extraposed position. Like other arguments, they can be scrambled into the top of the middle ﬁeld. Example (22) shows that they may occur in a relatively high position in the middle ﬁeld. In (22a), the inﬁnitival has been scrambled across the subject. In (22b), the inﬁnitival has been scrambled to a position just below the subject. However, sentences in which the inﬁnitive occupies a very low position in the middle ﬁeld are considerably degraded (22c).

16

SCRAMBLING , REMNANT MOVEMENT , AND RESTRUCTURING IN WEST GERMANIC

(22) a. weil [sie morgen zu besuchen] Hans gestern noch nicht versprechen wollte since her tomorrow to visit Hans yesterday yet not promise wanted ‘since Hans did not yet want to promise yesterday to visit her tomorrow’ b. weil Hans [sie morgen zu besuchen] gestern noch nicht verprechen wollte since Hans her tomorrow to visit yesterday yet not promise wanted c. ??weil Hans gestern noch nicht [sie morgen zu besuchen] versprechen wollte since Hans yesterday yet not her tomorrow to visit promise wanted d. *weil [sich zu rasieren] Hans schien since himself to shave Hans seemed

Coherent inﬁnitives, as opposed to incoherent inﬁnitives, cannot scramble to the front of the middle ﬁeld. The raising verb scheinen only allows a coherent complement, which cannot be scrambled at all (22d). 1.3.2

The transparency of coherent inﬁnitives

Let us now look at the second salient property of coherent inﬁnitives: transparency. Coherent inﬁnitives, as opposed to incoherent ones, allow for long-distance scrambling. The arguments of an embedded inﬁnitive can be scrambled into the middle ﬁeld of the matrix verb. For instance, the embedded object can be scrambled across the matrix subject (23a). It is important to note that this process of long-distance scrambling, though it occurs most naturally with weak pronouns and anaphors, is not restricted to them. Example (23b) shows that full DPs that are arguments of the embedded inﬁnitive can be scrambled across the matrix subject. Long-distance scrambling out of incoherent inﬁnitives results in ungrammaticality. In (23c), the embedded object phrase das Buch has been scrambled across the matrix subject out of an extraposed, hence incoherent, inﬁnitival complement that leads to ungrammaticality. (23) a. dass siei der Mann [ti zu besuchen] versprach that her the man-NOM to visit promised ‘that the man promised to visit her’ b. dass [der Maria]i [das Buch]j Hans gestern [ti tj zu geben] versprach that Maria-DAT the book-ACC Hans yesterday to give promised ‘that Hans promised yesterday to give the book to Mary’ c. *dass uns [das Buch]i Hans gestern bat [der Maria ti zu geben] that us the book Hans yesterday asked to Mary to give ‘that Hans asked us yesterday to give the book to Mary’

As in English, the scope of non-wh-quantiﬁers and operators is restricted by clausal boundaries in German. Consequently, the sentences in (24) and (25), where the matrix verb bedauern (‘regret’) selects only an incoherent inﬁnitival complement, may only have the narrow scope reading in (b) but not the readings with a

INTRODUCTION

17

wide scope of the quantiﬁer/operator represented in (c). Examples (24d) and (25d) illustrate again that incoherent inﬁnitival clauses are also opaque with respect to (overt) extraction. The direct object, be it a personal pronoun like in (24d) or a negative quantiﬁer as in (25d), may not be extracted from the inﬁnitival and scrambled above the matrix subject. (24) a. weil er [sie nicht geheiratet zu haben] bedauerte since he her not married to have resented b. since he resented not having married her c. since he did not resent having married her d. ??weil sie der Mann [ t geheiratet zu haben] bedauerte since her the man married to have regretted (25) a. weil er [niemanden geheiratet zu haben] bedauerte since he nobody married to have regretted b. since he regretted not having married anybody c. since for no x: he regretted having married x d. ??weil niemanden der Mann [t geheiratet zu haben] bedauerte since nobody the man married to have regretted

However, a non-wh-quantiﬁer/operator embedded in a coherent inﬁnitive may take scope over the matrix clause. Thus, the sentences in (26) and (27), in which the matrix verb wagen (‘dare’) allows for a coherent inﬁnitival complement, are ambiguous between the readings given in (b) and (c). In (26) and (27), I have somewhat impressionistically bracketed arguments that belong to the embedded verb within the inﬁnitive, to highlight the eﬀects of restructuring. In (26), these elements include the negative marker since it occurs between the inﬁnitive and its direct object argument. (26) a. weil er [sie nicht zu küssen] wagte since he her not to kiss dared b. since he dared to not kiss her c. since he did not dare to kiss her (27) a. weil er [ niemanden zu küssen] wagte since he nobody to kiss dared b. since he dared for no x: to kiss x c. since for no x: he dared to kiss x

Finally, coherent inﬁnitives, as opposed to incoherent inﬁnitives, allow for “long extraposition.” I have shown in (14) that extraposition is clause-bound. In (14d), repeated here as (28a) for convenience, the matrix verb bestärken (‘encourage’) only

18

SCRAMBLING , REMNANT MOVEMENT , AND RESTRUCTURING IN WEST GERMANIC

takes an incoherent inﬁnitival complement headed by versuchen (‘try’), extraposition out of which leads to ungrammaticality. The verb versuchen, however, optionally takes a coherent inﬁnitival complement. Hence, when it functions as matrix verb as in (28b), long extraposition becomes possible. (28) a. *dass der Lehrer [VP [VP die Kinder [CP t zu versuchen] bestärkt] [die Aufgaben zu lösen]] b. dass der Lehrer [VP [VP die Kinder [? t zu bestärken] versucht] [die Aufgaben zu lösen]]

1.3.3

The IPP-eﬀect in coherent inﬁnitives

The third salient property of coherent inﬁnitives is the IPP-eﬀect. As I said before, coherent inﬁnitives typically, though not necessarily, display this morphological eﬀect. For instance, in German, though not in Dutch, only bare inﬁnitives, but not toinﬁnitives, give rise to the IPP-eﬀect. The IPP-eﬀect occurs when a verb that selects a coherent inﬁnitive (the dependent inﬁnitive) is put into a perfect tense (present perfect or past perfect tense). In this case, the verb does not show up in its expected past participial form but is realized as bare inﬁnitive (the IPP-inﬁnitive). Example (29) illustrates the IPP-eﬀect in Dutch with the dependent inﬁnitive being a toinﬁnitive. Example (30) illustrates the IPP-eﬀect in High German, where the dependent inﬁnitive must be a bare inﬁnitive. (29) a. *dat Elsje hem een brief heeft gezien te schrijven that E him a letter has tried-PART to write b. dat Elsje hem een brief heeft zien te schrijven that E him a letter has try-INF to write (30) a. *dass sie Hans nicht küssen gewollt hat that her Hans not kiss want-PART has ‘that Hans has not wanted to kiss her’ b. dass sie Hans nicht hat küssen wollen that her Hans not has kiss want-INF

I will discuss the IPP-eﬀect in more detail in the diﬀerent sections devoted to the individual West Germanic languages. 1.3.4

Conclusions

To summarize, coherent inﬁnitives, compared to other (sentential) arguments, have a very restricted distribution: they can be neither scrambled nor extraposed; they are transparent for several types of extraction processes, including long-distance scrambling, quantiﬁer/operator movement, and long extraposition; and ﬁnally, they typically give rise to the mysterious IPP-eﬀect.

INTRODUCTION

1.3.5

19

The formation of verb clusters in coherent inﬁnitives

We have seen that while a coherent inﬁnitive as a whole cannot be scrambled, parts of it, namely, its arguments, can undergo movement into the middle ﬁeld of the selecting verb. It is important to note that the inﬁnitival head itself cannot be separated from the selecting verb. The contrast in (31) shows that while the embedded direct object can be scrambled across the matrix subject, the inﬁnitive itself cannot be scrambled across an adverb that modiﬁes the matrix verb. In (31a–b), I have scrambled the direct object of the inﬁnitive above the matrix subject to make sure that we are dealing here with a coherent construction. In this case, the adverb has to precede the inﬁnitive and can modify both the matrix verb and the dependent inﬁnitive, as can be expected from the general transparency of coherent inﬁnitives (31b). (31) a. *weil sie der Hans zu besuchen oft versprach since her the Hans to visit often promised ‘since Hans often promised to visit her’ b. weil sie der Hans oft zu besuchen versprach since her the Hans often to visit promised ‘since Hans (often) promised to (often) visit her’

Finally, let us look at the behavior of coherent inﬁnitives with respect to topicalization. First note that the head of the coherent inﬁnitive and the verb selecting it can be topicalized as if they formed a constituent, while if the inﬁnitive is incoherent its head cannot be topicalized with the verb selecting it. In (32a), versprechen (‘promise’), which optionally selects a coherent inﬁnitival, can be topicalized with its dependent inﬁnitive, whereas topicalizing the verb bestärken (‘encourage’) together with its dependent inﬁnitive results in ungrammaticality, since this verb only selects an incoherent inﬁnitival complement. (32) a. [zu besuchen versprochen] hat sie der Hans noch nie to visit promised has her the Hans-NOM yet never ‘So far Hans has never promised to visit her’ b. *[zu besuchen bestärkt] hat mich Hans seine Schwester noch nie to visit encouraged has me Hans-NOM his sister yet never ‘So far Hans has never encouraged me to visit his sister’

It should be noted, however, that although the selecting verb cannot be topicalized without its dependent inﬁnitive in a coherent construction (33a), the dependent inﬁnitive can be topicalized to the exclusion of its selecting head (33b). This contrast is reminiscent of the contrast that we observed in the behavior of particle verbs with respect to topicalization (cf. [13c–d] earlier): In a restricted way, it is possible to topicalize the particle without the verb it seems to be attached to, while it is impossible to topicalize the particle verb without its particle. Furthermore, it is important to note that topicalization of the dependent inﬁnitive alone voids the IPP-eﬀect, as is illustrated in (33c).

20

SCRAMBLING , REMNANT MOVEMENT , AND RESTRUCTURING IN WEST GERMANIC

Voiding of the IPP-eﬀect pertains only to perception verbs in German but occurs quite generally in Dutch and West Flemish (see chapter 7 for details). (33) a. *[sehen] wird sie Hans nicht kommen see will Hans her not come ‘Hans will not see her coming’ b. [kommen] wird sie Hans nicht sehen come will her Hans not see c. [kommen] hat sie Hans nicht *sehen / gesehen come has her Hans not see-INF / see-PART ‘Hans has not seen her coming’

The fact that the head of a coherent inﬁnitive cannot be separated from its selecting verb (see [31a–b]), as well as the fact that the head of a coherent inﬁnitival can be topicalized with the verb that selects it to the exclusion of the arguments of the inﬁnitive have been taken as direct evidence for the assumption that coherent inﬁnitival constructions involve the formation of a verbal complex by Verb Raising (VR). VR is a process of head movement that adjoins a dependent inﬁnitive to its selecting verb (Evers 1975). The formation of a complex head by either VR or reanalysis (Haegeman and van Riemsdijk 1986) has been taken as a causal explanation for the restrictive distribution and the transparency of coherent inﬁnitives. The former property of coherent inﬁnitives would follow trivially from the fact that the dependent inﬁnitive and its selecting verb form a complex head. The latter property of coherent inﬁnitives has been explained in the following way: Evers assumed that VR gives rise to a process of S-pruning. Van Riemsdijk and Haegeman assumed that reanalysis gives rise to a multidimensional representation: one dimension to represent the biclausal properties and one dimension to represent the monoclausal properties of coherent inﬁnitives. In more modern terminology but still exploiting the same idea, we may analyze VR as a process that incorporates the dependent inﬁnitive into the selecting verb. We may then assume with Baker (1988) that once the head of a projection has been incorporated, the entire projection, here the coherent inﬁnitival complement, becomes transparent with respect to the incorporator. VR has also been taken to be responsible for the IPP-eﬀect. If it should turn out to be correct that VR actually is the trigger for the IPP-eﬀect, then the data we have looked at so far suggest that we are dealing here with two independent factors for the following reason: We have seen that to-inﬁnitives in German may be transparent for several types of extraction processes, but I have also pointed out that they never give rise to an IPP-eﬀect. It thus seems that VR and the transparency of coherent constructions are independent factors and are not causally related to each other as has been assumed traditionally; VR may be a side eﬀect of the transparency of coherent inﬁnitives but not its trigger. Also, my initial presentation of the core data shows that the evidence for the formation of a complex head is not as direct or clear-cut as it seems at ﬁrst sight. The

INTRODUCTION

21

fact that the head of a coherent inﬁnitival cannot be separated from its selecting verb cannot only be explained by adjunction or incorporation but may simply follow from the inability of the head of a coherent inﬁnitive to scramble. I have pointed out that constituents that are interpreted as predicates generally resist scrambling. Of course, this alternative account presupposes an explanation of why and how coherent inﬁnitives are to be interpreted as part of the main predicate. Furthermore, the argument that the fact that the dependent inﬁnitive and its selecting verb can be topicalized together shows that these two heads form a constituent (32a) is considerably weakened by the observation that the dependent inﬁnitive can be also topicalized without its selecting head (33b). To summarize, the properties discussed earlier have been standardly interpreted as indicating that restructuring inﬁnitives involve (1) verbal complexes and (2) a monosentential middle ﬁeld. Monoclausal accounts assume that verbal complexes can be base generated. Haider (1993) assumes that verbal complexes are base generated head-adjunction structures. Wurmbrand (2001) argues, following Chomsky (1995b) in analyzing transitive verb phrases as involving two verbs—the verbal root and an abstract causative verb (small v)—that the verbs in a coherent construction share a single small v. Cinque (2001) argues that restructuring verbs in Italian are those verbs, in essence modal verbs and aspectuals, that can be assumed to be base generated in designated functional positions. All these accounts share the assumption that restructuring inﬁnitival constructions involve only one VP, from which assumption it follows that restructuring inﬁnitives have a monoclausal middle ﬁeld, that is, have only one IP. Biclausal approaches hold that each verb in a coherent construction projects its own clause (full or reduced) and assume special restructuring processes (S-pruning, reanalysis, Verb Raising) to account for the presumed monoclausal properties of coherent constructions.

1.4 Remnant movement As already noted, the term remnant movement was coined by Den Besten and Webelhuth (1987) to account for a peculiarity of verb-preposing in German and Dutch. Provided that only XPs can move into XP-positions, it follows that what has been moved into [Spec,CP] in (34) is not simply a verb but must be minimally a full VP. Thus, they propose to analyze (34a) parallel to cases of VP-preposing (cf. [34b]) in which the direct object has been scrambled out prior to VP-to-CP movement. The moved VP is called a remnant category since it contains, at least, the trace of the direct object as is indicated in (34c). (34) a. gelesen hat Hans das Buch read-PART has Hans the book ‘Hans has read the book’ b. [VP das Buch gelesen] hat Hans tVP c. [VP tSCR gelesen] hat Hans [das Buch]SCR

22

SCRAMBLING , REMNANT MOVEMENT , AND RESTRUCTURING IN WEST GERMANIC

In the following subsection, I discuss the essential properties of and the relevant restrictions on remnant movement. The latter will be shown to follow from the workings of Attract Closest. Then, I discuss the interaction between remnant movement and head movement. I argue that remnant categories created by head movement cannot undergo further movement and show how this restriction can be derived from Attract Closest as well. Finally, I show that the original account of (34a) by Den Besten and Webelhuth is ﬂawed by presenting data that show that the remnant VP in (34a) cannot be taken to have been created via scrambling. Instead, I argue that remnant VPs in German are created by licensing movement of VP-internal material into dedicated licensing positions in the lower middle ﬁeld. 1.4.1

Properties of remnant movement

Remnant movement seems to be exempted from the Proper Binding Condition (PBC), which requires that traces be bound. It is typical of remnant movement to create unbound traces as we have seen earlier. In (34c), the fronted VP contains the trace of the direct object, which is not c-commanded by its antecedent. Remnant movement typically also leads to so-called Anti-Freezing eﬀects. A Freezing eﬀect occurs if extraction of constituent a out of a constituent b takes place in a derived position of b. This is illustrated in (35a–b). In (35a), extraction of the wh-PP takes place from the base position of the direct object. In (35b), the direct object has been scrambled to a higher position from which extraction is excluded. (35) a. Worüberi hat keiner [ein Buch ti] gelesen where-about has nobody a book read b. *Worüberi hat [ein Buch ti] keiner gelesen where-about has a book nobody read ‘Which topic has nobody read a book about’ c. [Ein Buch ti] hat darüberi keiner gelesen a book has there-about nobody read ‘Nobody has read a book about that’

Example (35c) is a case of remnant movement. From a representational point of view, (35c) should be as bad as (35b), since the direct object in (35c) occurs in a derived position and contains a trace created by extraction. However, if we look at these cases from a derivational point of view, we immediately understand why (35b) is ungrammatical and (35c) is ﬁne. The diﬀerence follows from the Extension Condition on derivations (Strict Cyclicity). In (35c), it is possible to extract the PP out of the DP in its base position and then move the remnant DP, obeying the Extension Condition, to a higher position. This derivation, however, is not available in (35b). If the PP were extracted ﬁrst, movement of the remnant DP to a lower position would violate cyclicity. Thus, we see that obeying the Extension Condition voids—in a manner of speaking—Freezing eﬀects but necessarily leads to unbound traces. Consequently, we may assume that the PBC is not a derivational constraint.

INTRODUCTION

23

Furthermore, as Müller (1996, 1998) has discussed at length in his important work on incomplete category fronting, cases of remnant movement display unexpected asymmetries. If we restrict ourselves to the interaction between remnant movement and scrambling, it is interesting to note that remnant categories can be topicalized while scrambling may not aﬀect them, as is illustrated in (36). This asymmetry does not show up in cases of complete category fronting (37). (36) a. [zu lesen] hat das Buch keiner versucht to read has the book nobody tried-PART ‘Nobody has tried to read the book’ b. ??dass [zu lesen] das Buch keiner versucht hat that to read the book nobody tried has ‘that nobody has tried to read the book’ (37) a. [das Buch zu lesen] hat keiner versucht the book to read has nobody tried-PART ‘Nobody has tried to read the book’ b. dass [das Buch zu lesen] keiner versucht hat that the book to read nobody tried has ‘that nobody has tried to read the book’

Müller (1996) concludes from these and other similar observations that remnant XPs cannot undergo Y-movement if the antecedent of the unbound trace has also undergone Y-movement, where Y-movement ranges over scrambling, wh-movement, and topicalization. In other words, a remnant category cannot, say, undergo scrambling if the operation that created the remnant category was itself a scrambling operation. Müller derives this constraint on remnant movement from his Principle of Unambiguous Domination, which is motivated by the need of traces, so Müller assumes, to be unambiguously identiﬁable. There are several problems with Müller’s account though. First, there are cases where remnant categories can undergo scrambling, as we will see in chapter 4 (section 4.3.4). Second, various cases where remnant categories cannot scramble can be reduced to independent restrictions on the individual operations involved, as I will also argue in chapter 4. Third, Müller is assuming (following Den Besten and Webelhuth [1987]) that the remnant XPs in (36) are created by scrambling. In section 1.4.3, I will argue that remnant VPs are created by licensing movement of VP-internal arguments into dedicated positions in the middle ﬁeld, which is to be distinguished from scrambling. Fourth, Müller’s Principle of Unambiguous Domination lacks conceptual plausibility in a theory of movement in terms of copy and delete, that is, in a theory where there are no traces to be identiﬁed. From the point of view of feature checking, Müller’s observations indicate that for remnant movement to be possible two sets of features must be involved that cannot be checked in the same (type of) position. Thus, Müller’s generalization (to the extent that it is correct) can be derived from Attract Closest (Chomsky 1995b), as is

24

SCRAMBLING , REMNANT MOVEMENT , AND RESTRUCTURING IN WEST GERMANIC

illustrated in (38). In order for a (remnant) category A to undergo scrambling, extraction of B out of A must involve a type of movement other than scrambling. In (38), the head F cannot select category B as a target since category A, containing and dominating B, is a closer target that shares the scrambling feature with B (cf. Fukui [1997]).3 (38)

FSCR . . . [ASCR . . . [BSCR . . .]]

There is a class of cases of remnant movement that are illicit although they obey Attract Closest (or Unambiguous Domination, for that matter). These involve topicalization of a remnant category out of which a category has been extracted via whmovement as is illustrated in (39). Example (39a) is a case of topicalization of a clause across a wh-island, which leads to a mild, subjacency-like violation. Example (39b), where the wh-word is extracted from the embedded clause (creating a remnant category), however, is ungrammatical. The same contrast can be observed in English, as is shown in (40), taken from Pesetsky (2000). The relative grammaticality of (39c) is interesting in this respect. If the trace of the wh-word were contained in the fronted VP, as is standardly assumed, then (39c) should be on a par with (39b) and (40b). However, it is just as good as (39a). I will come back to this diﬀerence and show how it is to be explained in section 1.4.3. (39) a. ??[dass Fritz Peter liebt] weiß ich nicht wer gesagt hat that Fritz loves Peter know I not who has said ‘I don’t know who has said that Fritz loves Peter’ b. *[ dass Fritz t liebt] weiß ich nicht wen er gesagt hat [that Fritz loves t] know I not who he has said ‘I don’t know who he said that Fritz loves’ c. ??[geküsst] weiß ich nicht wen sie hat [kissed] know I not whom she has ‘I don’t know whom she has kissed’ (40) a. [give a book to John] I can guess who will b. *[give a book to t] I can guess who Mary will (I can guess who Mary will give a book to)

It is not clear why wh-movement diﬀers in this respect from other types of movement. In other words, it is not clear why the PBC nevertheless seems to be relevant for wh-movement. Pesetzky (2000) assumes that there is a (special) command restriction on wh-movement. Cecchetto (2001) argues that the illicit cases of remnant movement in (39b) and (40b) follow from the Phase Impenetrability Condition (cf. Chomsky [1998, 2001]). Contrary to Chomsky, Cecchetto proposes that the CP contains two escape hatches, one for wh-elements and one for D-linked elements like topics, and argues that elements that contain a wh-trace cannot make use of the escape hatch reserved for

INTRODUCTION

25

D-linked elements. Thus, (39b) and (40b) involve movement across a strong phase in his system, while the topics in (39a) and (40a) can move via the extra escape hatch for D-linked elements. This seems like a promising approach to me, though it remains unclear how Cecchetto can account for the relative grammaticality of (39c). I will introduce the Phase Impenetrability Condition in chapter 6 and leave the question of the best treatment of the data in (39) open here, concluding from the preceding discussion that as long as we steer clear of extraction via wh-movement, remnant movement is freely available if Attract Closest and Strict Cyclicity are obeyed. 1.4.2

Remnant movement and head movement

In this section, I would like to address a further restriction on remnant movement that has been discussed in the literature (cf. Haider [1990], Sabel [1996], Takano [2000]) and show how it can be reduced to the workings of Attract Closest as well. So far, I have only discussed cases in which a remnant category is created by extraction via XP-movement. In the sentences in (41), the remnant category (topicalized in [41a–b] and clefted in [41c]) has been created by extraction via head movement.4 (41) a. *[Ihr ein Buch t] gab Hans her a book gave Hans b. *[Het boek aan Marie t] gaf Jan waarschijnlijk The book to Mary gave John probably c. *It’s [a book to Mary] that John gave

All three cases involve remnant movement of the VP that contains a trace of the ﬁnite verb, which either has undergone V2 as in (41a–b) or has raised out of the lower VPshell in a Larsonian structure in (41c). Takano (2000) argues that the data in (41) fall under the generalization in (42). (42)

Remnant movement of a is impossible if the head of a has moved out of a

At this point the question arises what (42) can be derived from. It can be shown that the major restrictions on remnant movement follow from the mechanism of feature checking, more speciﬁcally from the restrictions on target selection. In (43), F and G are functional heads and XP is some category that contains Speciﬁer A and Complement B. I make the standard assumption that XP and its head X share all relevant feature (i.e., XP is a projection of X). Furthermore, I assume that head and complement do not share features via agreement, while Speciﬁer and head may share certain features via agreement. (43)

G . . . F . . . [XP A [X B]]

If the functional head F attracts a feature of X (for checking), I assume that due to Attract Closest, movement of the entire XP will ensue (since the XP shares the feature with X

26

SCRAMBLING , REMNANT MOVEMENT , AND RESTRUCTURING IN WEST GERMANIC

but is closer to F than X), unless the functional head F itself requires a head, that is, it is an aﬃx. If F attracts a feature of B and G a feature of X (not shared with B), we have a standard case of remnant movement. If X and B for any reason share the relevant feature with F, remnant movement is blocked, as has been discussed in section 1.4.1. If on the one hand, F attracts a feature of A that is not shared by X (a non-agreeing feature), then the Speciﬁer will extract from XP. If on the other hand, F attracts a feature that A and X agree in, then movement of XP will ensue (a case of pied-piping). Let us now look at the situation in which a remnant category is created by head movement. In this case F, being an aﬃx, attracts a feature of X and X undergoes head movement to F, creating a remnant XP. The issue now is why the functional head G cannot attract the remnant XP. We have to distinguish two cases: (1) G attracts a feature shared by X. In this case, G will select F or FP but not XP due to Attract Closest. (2) G attracts a feature within the constituents of XP not shared by X. In this case, either A or B may extract from XP, but no movement of the remnant category XP will be triggered. It thus follows that movement of a remnant category that has been created by head movement is impossible.5 1.4.3

Remnant movement and licensing movement

As I have pointed out earlier, in the original account of Den Besten and Webelhuth (1987) of data like (1), the operation that evacuates the VP before it undergoes topicalization is assumed to be scrambling. English does not allow for the topicalization of remnant VPs. In their account this simply follows from the fact that English does not allow for scrambling. Ingenious though this account at ﬁrst sight seems, it nevertheless turns out to be severely ﬂawed at closer inspection. It suﬀers from two inconsistencies, the extraction paradox and the evacuation paradox that will be discussed and illustrated later. To avoid these problems, I propose that the VP in German (and Dutch) is evacuated by licensing movement. In Hinterhölzl (1999), I argue that not only arguments but also VP-internal predicates move out of the VP to be licensed in speciﬁc positions in the middle ﬁeld as is illustrated in (44). (44)

[DPs [Neg0 [Focus0 [DPs [VP-adverbs [Pred0 [Asp0 [VPV ]]]]]]]]

In chapter 4, I will provide empirical arguments for the following licensing operations: Verb-particles are licensed in [Spec,AspP]. Small clause predicates, idiomatic expressions, and directional PPs are licensed in [Spec,PredP] above Asp0. Nominal arguments of the verb undergo Case-licensing movement to positions above VPadverbs. From there they may undergo scrambling to higher positions according to their semantic/discourse properties. Thus, movement of arguments out of the VP is licensing movement that applies to all DPs independently of whether they are deﬁnite or indeﬁnite and has to be distinguished from further movement that applies to DPs according to their semantic/discourse properties as is discussed in detail in chapter 2. The assumption of licensing movement of VP-internal material that is to be distinguished from scrambling proper can solve the two long-standing problems with

INTRODUCTION

27

the standard account of VP-topicalization in terms of scrambling and remnant movement, namely, the extraction paradox (1) and the evacuation paradox (2): 1. The assumption that there is licensing movement of VP-internal material to speciﬁc positions in the middle ﬁeld is corroborated by the existence of so-called string vacuous scrambling as is illustrated in (45). In (45), the underlined phrases seemingly occur in their canonical position (in [45b] the direct object follows the subject) but have to be assumed to have been scrambled out of VP in order to be exempted from VP-topicalization. (45) a. [VP t gerechnet] hat wie immer keiner damit counted has as always no one there-with ‘as usual nobody has reckoned with that’ b. [VP t gelesen] hat gestern der Fritz ein Buch darüber read-PART has yesterday the Friz a book there-about ‘Fritz has read a book about this yesterday’

As (46) shows, these presumed scrambling positions do not behave like regular scrambling positions since they do not exhibit any Freezing eﬀect (cf. [35] earlier) in that they allow for further extraction, which scrambles part of the phrase to a higher position. Thus, the data in (45) pose the following problem: If scrambling viewed as an operation that moves VP-internal material into the middle ﬁeld were a unitary operation, then why is it that scrambled phrases allow for further extraction in certain positions but not in others? (46) a. gerechnet hat da wie immer keiner mit counted has there as always noone with b. gelesen hat darüber gestern der Fritz ein Buch read has there-on yesterday the Fritz a book

2. Another problem for the standard approach is the fact that elements that resist scrambling can be left behind by VP-topicalization. This is illustrated for small clause predicates and indeﬁnite “w-words” in (47a) and (47b), respectively. I will show in chapter 2 that scrambling proper may not aﬀect (small clause) predicates. Furthermore, I will also demonstrate that indeﬁnite w-words may only scramble to extend their scope domain. This is not the case in (47b). If scrambling were the only operation that can evacuate material from the VP, then it is hard to understand how these elements can be left behind by VP-topicalization. (47) a. gegessen hat der Karl das Fleisch roh eaten has the Karl the meat raw ‘Karl has eaten the meat raw’ b. gelesen hat die Maria erst gestern was read-PART has the Maria only yesterday something ‘It’s only yesterday that Mary read something’

28

SCRAMBLING , REMNANT MOVEMENT , AND RESTRUCTURING IN WEST GERMANIC

A possible solution to these problems is the assumption of licensing movement (prior to scrambling) that moves VP-internal material, irrespective of its quantiﬁcational or referential potential, out of the VP. This assumption immediately solves the evacuation problem—elements that do not scramble nevertheless leave the VP to be licensed in speciﬁc positions in the lower middle ﬁeld—but also opens up the way toward a solution to the extraction problem. Given the distinction between licensing movement and scrambling, we may assume that licensing movement out of the VP does not give rise to a Freezing eﬀect and that this Freezing eﬀect is connected with the “Speciﬁcity eﬀect” of the semantically motivated scrambling operation. We know independently that extraction out of speciﬁc DPs is illicit. Since DPs that have scrambled across sentential adverbs or negation (crucially not ones that have scrambled across VP-adverbs) exhibit a Speciﬁcity eﬀect, the Freezing eﬀect associated with these scrambling operations follows without further ado. Finally, note that the trace within the VP in (39c) earlier, repeated here as (48a), does not behave like a Case-marked wh-trace; otherwise we would expect (39c) to be on a par with (39b) and (40b). However, (39c) gives rise only to a mild, subjacencylike violation and is on a par with (39a), as we have seen in section 1.4.1 earlier. This fact, too, follows straightforwardly, if we assume that there is licensing movement of DPs (and of VP-internal predicates) out of the VP that leaves A-movement traces in the VP such that the wh-word in the embedded clause can bind a wh-trace outside of the fronted VP within its clause, as is indicated in (48a). The analysis of (39c) in (48a) makes this case of remnant movement parallel to cases of movement of remnant categories that are created by movement of the subject to [Spec,TP] or another licensing position for subject, as is illustrated in (48b). Example (48a) thus only diﬀers from (48b) in that it involves extraction of the topicalized phrase out of a weak whisland, which accounts for its slightly marked grammatical status. (48) a. ??[ ti geküsst]j weiss ich nicht [weni sie hat ti [VP tj]] kissed I don’t know whom she has ‘I don’t know whom she has kissed’ b. [ﬁred t by the company] John indeed was

To summarize, I have argued that the assumption of movement of DPs and VPinternal predicates to speciﬁc licensing positions in the middle ﬁeld, as illustrated in (44) earlier, can provide a solution to the extraction problem as well as the evacuation problem and gives us an explanation for the otherwise rather mysterious contrast between (39b) and (39c).

1.5 Theoretical assumptions In the following, I will endorse the Minimalist Program (MP) in its programmatic sense. That is to say that the discussion and treatment of scrambling, remnant movement, and restructuring is cast in the spirit of MP, without endorsing all its tenets and principles to the letter. For instance, I do make use of agreement projections, though

INTRODUCTION

29

they are argued to be non-minimalistic in Chomsky (1995b). However, I do not make use of multiple speciﬁers though they are part of minimalist accounts of Case-licensing and other phenomena and abstract away completely from assumptions made within the bare phrase structure framework (Chomsky 1995a). An important minimalist notion that I consider and adopt in the description and treatment of scrambling, remnant movement, and restructuring is the concept of the phase. In particular, I propose that employing the notion of phases may be useful in approaching the problems posed by extraposition and VP-topicalization. Consequently, I will now outline the core ideas and concepts that I will adopt and later indicate in the account of individual phenomena where I diﬀer from standard treatment. In my view, the basic weakness and strength of MP lies in its programmatic character. I think MP cannot be simply adopted since it is not a ﬂeshed-out theory (by now) as, say, Government and Binding theory was. But I would like to endorse it in the way in which it approaches grammatical phenomena. The core idea lies in the way in which it conceptualizes grammatical representations and their well-formedness. The levels of representations are reduced to the two interface levels, PF and LF. As a consequence, every principle has to be formulated in a way that it can be reasonably taken to apply at one of the interfaces (as an output condition) or in such a general way that it can be taken to apply at every step in the derivation. Another important concept of the MP is that the grammaticality of a derivation may depend on the properties of another derivation. In particular, the MP invokes economy principles that compare derivations that involve the same lexical resources and select the most economical derivation. In GB-theory, operations such as move alpha apply freely and a single derivation that results from the free application of such operations is checked against the constraints formulated in the diﬀerent modules. In the MP, individual constituents have particular needs that motivate them to move and failure to meet these needs can lead to a derivation that is uninterpretable at one of the interfaces. In short, movement operations of constituents are motivated or triggered by the particular needs that they have, technically implemented by the types of features that they possess. This means, in turn, that every movement operation that is proposed in the account of a certain phenomenon needs to be justiﬁed by specifying the features, in the ideal case, the interface features, that motivates it. I will come back to this point, shortly. I consider this an important tenet of the MP, because it will prompt us to search for genuine explanations rather than for merely technical accounts of syntactic phenomena, insofar as we strive to provide derivations that are preferably triggered by (interpretable) interface properties. To be concrete and to specify the needs that a particular constituent may have, a DP argument holds a thematic relation to a predicate, has a certain grammatical function in the clause, has scope over other elements in the sentence, may be interpreted as a variable bound by a particular operator within a certain domain, may be the topic or the focus of the sentence, and so on, possibly all or several of these at the same time. These relations are encoded by various means in the languages of the world, among which we ﬁnd morphological marking, prosodic phrasing, word order, and intonational marking.

SCRAMBLING , REMNANT MOVEMENT , AND RESTRUCTURING IN WEST GERMANIC

30

It has been a leading idea in generative theory from the beginning that these diﬀerent types of marking are incorporated into the syntax or dependent on syntactic structure. In minimalist terminology this means that these relations are either part of the computational system itself or the result after a complex computation at one of the interfaces. A sentence can thus be viewed as the most economic derivation that connects the individual partial representations that encode the diﬀerent relations among the lexical items that have been chosen to express a piece of meaning, say, a proposition. A core ingredient in this system is the copy theory of movement. Chains formed by movement in the MP consist of a sequence of copies of the moved constituent. Where the chain is pronounced and where it is interpreted is decided by independent principles. This analysis of chains as a sequence of copies allows for reconstruction without any special operation (of putting lexical material back). In this manner, the copy of a DP argument within the VP may be said to designate its thematic relation, the copy within IP its grammatical function, and its copy within the CP-layer its discourse status in the sentence. Locality principles that constrain movement seem to follow from a least eﬀort strategy. This is captured in various economy conditions, most important among which is Shortest Move. The basic idea is that a constituent must not move beyond the ﬁrst position of the relevant type, an idea that was ﬁrst captured, though in diﬀerent terms, as Relativized Minimality by Rizzi (1990). Another important economy principle that I will make use of is the principle of last resort. Last resort allows a certain operation only if all other options are prohibited. Do-insertion in English questions and negative clauses is an example of a last-resort operation that may apply to save the derivation in case the verb itself cannot move to C. To summarize, here are the most important tenets of the MP that I endorse in this book: 1. Movement is triggered and needs justiﬁcation. 2. Grammaticality depends on a comparison of derivations. 3. Constraints apply only at the interfaces or throughout (uniformly at) the derivation. 4. Chains are analyzed as a sequence of copies, whose interpretation and pronunciation are determined by independent principles. 5. Derivations are constrained by economy conditions and phase based. 1.5.1

Narrow syntax and the role of Case

One idea or concept that I will not adopt is the assumption that the computation is driven by uninterpretable formal features predominantly or exclusively. Dealing with phenomena like scrambling and restructuring, I reject the approach that has become known as narrow syntax. In chapter 2, I will provide a comprehensive discussion of the empirical nature of scrambling phenomena in German and argue that scrambling is triggered (partially) by pragmatic features like familiarity/speciﬁcity. Another issue with which I disagree is the role of Case in the MP. There are actually two assumptions connected with Case that I think are mistaken: (1) The assumption that Case is not interpretable at the interfaces, in particular, that it is uninterpretable at LF. In investigating scrambling phenomena, I will argue that Case is essential for

INTRODUCTION

31

permuting arguments by marking their grammatical functions (see chapter 2). (2) The assumption that Case is invisible for the computational system. To evaluate this assumption, let us look at the role that Case plays in Icelandic and German. We have seen earlier that basic word order in German is dependent on the particular verb class. One such class is constituted by the so-called Dative-Nominative verbs, which are stative psych-verbs that assign the experiencer role to the Dative argument and the theme role to the Nominative argument. The experiencer is hierarchically higher than the theme argument and, consequently, the unmarked word order is DAT > NOM in German, as is illustrated in (49a). (49) a. weil dem Hans das Essen schmeckt since Hans-DAT the food-NOM tastes b. Þer hefur likað maturinn you-DAT have liked food-the-NOM c. [T

[DP-DAT

[DP-NOM]]]

The same class is found in Icelandic, where the two arguments are realized in a parallel fashion as Dative experiencer and Nominative theme (cf. [49b]). The diﬀerence between Icelandic and German, however, is the diﬀerent grammatical function of the arguments. In German, as is expected, the Nominative argument, though the lower argument, is the subject of the clause. Icelandic is special in that these Dative experiencers function as the subject of the sentence, as is shown by their binding properties and their behavior in control structures (cf. Zaenen, Maling, and Thrainsson [1985]). How can we account for the diﬀerence in syntactic behavior between Dative experiencers in German and Icelandic? If we identify the subject as the argument that moves to or is attracted by the T-head, then the Icelandic case is straightforward: T will simply attract the closest element with a D-feature, namely, the higher argument marked with Dative Case (cf. [49c]). In German, the case is more diﬃcult since T has to attract the hierarchically lower argument, to render the Nominative argument subject of the clause. I want to argue that one way to achieve this is to assume that T in German does not simply attract the closest element with a D-feature, but that the T-head can refer to the Case properties of a DP and will attract the closest argument marked with Nominative Case. This account of the diﬀerence between German and Icelandic implies that Case is visible in the computation in German. This diﬀerence between German and Icelandic is very instructive since it shows that the visibility of Case is not simply a matter of the presence (German) or absence (English) of morphological Case. Thus, contrary to narrow minimalist assumptions, my representations will include Case-positions and other positions that relate to the semantic contribution and the pragmatic role of the constituents in the clause. 1.5.2

The universal base hypothesis and triggers for movement

In the course of this study, I will adopt the UBH (Kayne 1994). Applying the UBH to a language like German, as I have discussed and illustrated in section 1.2, requires

32

SCRAMBLING , REMNANT MOVEMENT , AND RESTRUCTURING IN WEST GERMANIC

additional movement operations to derive the surface forms that in the traditional head-ﬁnal account are taken to be base generated. To derive OV-surface orders from VO-base structures, I will provide additional (see section 1.4.3) empirical evidence that VP-internal predicates and arguments move out of the VP (to appropriate licensing positions in the middle ﬁeld). To derive the monoclausal properties and verbal complexes of restructuring inﬁnitives, I will make substantial use of remnant movement of diverse categories that include VPs and TPs. It is an important goal of this study to not only describe the movements necessary in a Kaynean analysis to derive the surface properties of restructuring inﬁnitives but to also provide motivation for the complex derivations proposed at each step. Hence I will try to justify each movement operation by specifying its syntactic or interface-related triggers. Only to the extent that these triggers seem non-arbitrary and non-stipulative can the endeavor be considered successful that consists of combining the restrictive theory of phrase structure implied by the UBH with the restrictive theory of movement inspired by the MP. At several points of this study, I will evaluate the results of this endeavor by comparing the new analysis with the traditional account, discussing their strengths and weaknesses. In my opinion, this endeavor is successful overall. However, some recalcitrant problems remain, foremost the issue of extraposition and the topicalization of verbal projections (extended verb phrases). With these issues the goal of this study has to remain more modest: In chapter 7, I discuss these issues, pinning down the problems they pose for this approach, and outline paths toward their solution. The solution I provide for VP-topicalization is very promising, while I have to leave a full account of extraposition for further research.

2

Scrambling

T

his chapter provides a comprehensive discussion of both the empirical nature and the theoretical implications of scrambling. I argue that the phenomenon of scrambling can only be done justice to if it is recognized that scrambling is subject to both PF- and LF-interface conditions. In particular, I address the claim by Haider and Rosengren (1998), henceforth H&R, that trigger accounts are essentially inappropriate for the phenomenon of scrambling on account of its optional nature. Contrary to H&R, I argue that a trigger account is indeed feasible in a copy theory of movement in which both LF- and PF-conditions determine which copy is to be spelled out. Furthermore, I argue that one type of trigger for scrambling involves scope features of arguments. Scope features are relational syntactic properties of scope-taking elements, and as such they are inherently non-lexical. I propose an extension of the minimalist framework that allows for the introduction of non-lexical features in the course of the derivation to account for this aspect of scrambling. Finally, I show that scrambling in German and Dutch obeys a uniform minimality condition and argue that this minimality condition must not be reduced to, that is, must be kept distinct from, the Minimal Link Condition in order to uphold a nonoptional trigger account in contexts that display minimality eﬀects. The chapter is organized in the following way. Section 2.1 contains introductory remarks. Section 2.2 provides a discussion of the properties of scrambling in German. On the basis of these data, I argue in favor of an A-movement analysis of the phenomenon, discarding arguments having been put forth in favor of an A'-movement analysis. The section concludes with a discussion of how the pertinent A/A'-distinction can 33

34

SCRAMBLING , REMNANT MOVEMENT , AND RESTRUCTURING IN WEST GERMANIC

be drawn independently of the notion of L-relatedness. Section 2.3 introduces the account of H&R highlighting its virtues as well as its problems. Section 2.4 sketches and illustrates H&R’s argument that trigger accounts necessarily fail to do justice to the phenomenon of scrambling. In section 2.5, I provide a solution to the trigger problem. Outlining the distributional properties of deﬁnite and indeﬁnite, speciﬁc and nonspeciﬁc, focused and non-focused DPs with respect to the negative marker in German, I argue that scrambling is not optional and show how apparently irregular word order patterns can be explained by the interaction of output conditions that determine the Spell-out of movement copies. In section 2.6, I address the issue of how scope features that represent relational properties of constituents can be handled in a simple trigger account. I propose an extension of the standard minimalist framework and show that the computational complexity of the generalized transformation proposed is suﬃciently reduced by the standard economy conditions as to render it an attractive alternative to assuming optionality in syntactic computation. Section 2.7 discusses minimality eﬀects in scrambling operations and outlines a uniform account that constrains scrambling in both German and Dutch and discusses its implication for the status of the Minimal Link Condition.

2.1 The grammatical nature of scrambling The term scrambling was coined by Ross (1967) to refer to “the permutation of two adjacent constituents.” Ross focused on the clause-boundedness of the operation and argued that scrambling should be placed in the stylistic component of the grammar, since scrambling was formally so unlike other transformational rules. Along these lines, Williams (1984) proposed that scrambling applies between S-Structure and PF. However, it was soon noticed that scrambling has syntactic eﬀects that inﬂuence the grammaticality and the interpretation of the clause. Webelhuth (1985) pointed out that scrambling induces anti-crossover eﬀects, as is illustrated in (1). (1)

a. *weil seinei Lehrer jedeni fürchten since his teachers-NOM everyone-ACC fear b. weil jedeni seinei Lehrer fürchten since everyone-ACC his teachers-NOM fear ‘since his teachers are afraid of everyone’

Furthermore, it was noted that scrambling aﬀects the grammaticality of focused DPs (2) and wh-phrases in situ (3), in the sense that scrambling of these elements leads to ungrammaticality. In (2), the focused constituent appears in small capitals. (2)

a. *weil dem Studenten das buch der Professor geliehen hat since the student-DAT the book-ACC the professor-NOM lent has b. weil dem Studenten das Buch der professor geliehen hat

SCRAMBLING

(3)

35

a. Was hat gestern der Professor wem geliehen? what has yesterday the professor-NOM who-DAT lent b. *Was hat gestern wem der Professor geliehen? what has yesterday who-DAT the professor-NOM lent

Since scrambling is related to restrictions that take eﬀect at LF, it became clear in the hey-day of GB-theory that scrambling must be considered a syntactic operation. The following years saw a vivid and heated debate about the issue of how the syntactic operation of scrambling can be ﬁtted into the A/A'-dichotomy of movement operations in the Extended Standard Theory. Since scrambling does not ﬁt either movement type completely, the question of what type of movement operation scrambling is and whether scrambling should be described as the result of movement at all (rather than of base-generation) was far from being subject to general consensus for a long time. I cannot go into the details of this historic debate and refer the reader to two volumes that deal extensively with the question sketched earlier (cf., Corver and van Riemsdijk [1994], Grewendorf and Sternefeld [1990]). Instead I will provide a discussion of the descriptive properties of scrambling in German and explain why I consider the debate as having been decided in favor of A-movement.

2.2 The properties of scrambling in German Scrambling has been used as a cover term for diﬀerent operations that aﬀect word order in the German middle ﬁeld. It is important to distinguish two such operations in order to detect the constitutive properties of the operation of word order change that we are interested in here. Depending on whether the moved item bears stress or not, the movement operation displays quite signiﬁcantly diﬀerent properties. The operation of scrambling in which the moved item is stressed shows clear properties of A'-movement: It can aﬀect arguments as well as predicates and is not clausebound. This operation moves contrastive topics and comes with a special intonation, the so-called hat contour that comprises a fall-rise tone on the moved topic and a fall tone on the (contrastive) focus-element in the remainder of the clause. This operation has been called focus scrambling by Neeleman (1994a), I-topicalization by Jacobs (1997), and T-scrambling by H&R. Because the moved item bears stress, I shall use the term S-scrambling. This movement operation is illustrated in (4) and (5), in which stressed syllables are given in small capitals. In Dutch, contrary to German, an object cannot scramble over the subject. However, if the operation is accompanied with the special intonation of a contrastive topic, scrambling of the object across the subject is ﬁne (cf. [4b–c]). Example (5) depicts long-distance S-scrambling in German: The embedded object has been moved into the middle ﬁeld of the matrix clause. (4)

a. dat Jan de boeken niet koopt that Jan the books not buys b. *dat de boeken Jan niet koopt that the books Jan not buys

36

SCRAMBLING , REMNANT MOVEMENT , AND RESTRUCTURING IN WEST GERMANIC

c. dat zulke boeken selfs jan niet koopt that such books even Jan not buys (5)

Noch gestern haben diese Frau alle geglaubt, dass niemand einladen wird Still yesterday have this woman-ACC all believed that nobody invite will ‘Of this woman everyone believed still yesterday that no one would invite her’

Once the distinction is made between S-scrambling and scrambling operations in which the moved element bears no special stress, it can be shown that scrambling proper is restricted to arguments (6) and strictly clause-bound as well (7). In (6), the inﬁnitival predicate has been scrambled across negation, leading to ungrammaticality. In (7), long-distance scrambling of the embedded object results in ungrammaticality. (6)

a. weil jeder oft gewinnen möchte since everyone often win wants b. *weil jeder gewinnen oft möchte since everyone win often wants ‘since everyone often wants to win’

(7)

a. Gestern hat niemand geglaubt, dass er die Maria einladen wird yesterday has nobody believed that he the Maria invite will b. *Gestern hat niemand die Maria geglaubt, dass er einladen wird yesterday has nobody the Maria believed that he invite will ‘yesterday nobody believed that he will invite Maria’

Scrambling can create new binding possibilities. In (8a), the quantiﬁer cannot A-bind the pronoun simply for lack of c-command. However, if the direct object quantiﬁer is scrambled across the subject that contains the pronoun, the latter can be bound with no weak crossover eﬀect (henceforth WCO-eﬀect) being noticeable. (8)

a. *weil seinei Mutter jedeni liebt since his mother-NOM everyone-ACC loves b. weil jedeni seinei Mutter liebt since everyone-ACC his mother-NOM loves

Scrambling can lead to scope ambiguities. Stress in the former statement is on the modal can, since there seem to be two groups of speakers. For speakers of the ﬁrst group, scope is solely a matter of surface relations. For these speakers, including me, the scrambled structure in (9b) is as unambiguous as the base structure in (9a), though the scope relations have been inverted by scrambling. For speakers of the second group, the base structure is unambiguous as well, but the scrambled structure is ambiguous. In (9b), these speakers not only get the reading that results from surface scope (as speakers of the ﬁrst group do) but also get the reading that results from reconstructing the scrambled object into its base position.

SCRAMBLING

37

(9) a. weil [mindestens eine Frau] [fast jeden Mann] liebt since at least one woman-NOM almost every man-ACC loves b. weil [fast jeden Mann] [mindestens eine Frau] t liebt since almost every man-ACC at least one woman-NOM loves

It is important to note that also for the second group binding relations are strictly read oﬀ from surface relations. In German, scrambling may not only create new binding possibilities; it may also destroy binding possibilities, as the example adopted from H&R shows. Even for speakers of the second group, the unbound anaphor in (10b) cannot be saved by reconstructing it into its base-position. (10) a. dass der Mann die Bilder einander anglich that the man the pictures-ACC each other-DAT made-alike b. *dass der Mann einander die Bilder t anglich that the man each other-DAT the pictures-ACC made-alike c. einander hat der Mann die Bilder angeglichen each other-DAT has he the pictures-ACC made-alike

To sum up, the fact that scrambling (1) is restricted to arguments, (2) is clausebound, and (3) can create new binding possibilities speaks in favor of an analysis in terms of A-movement. Moreover, the fact that scrambling cannot be reconstructed for reasons of binding speaks against an analysis in terms of A'-movement, since as is shown in (10c), an anaphor that has been A'-moved into [Spec,CP] can be bound via reconstruction. Thus, there is little reason to assume that scrambling in German should not be A-movement. One argument that played an important role in the debate on the nature of scrambling was based on the observation that scrambling can license parasitic gaps. If parasitic gaps can only be licensed by A'-movement, as seems to be the case in English, then scrambling ought to be A'-movement, so the argument went. Neeleman (1994b) provides an excellent discussion of this argument and convincingly argues that the evidence that is given by examples of the type of (11a) has been overrated. It is suﬃcient here to present his strongest argument, namely, the observation that the so-called parasitic gap in (11a) can also be licensed by A-movement. In (11b), the presumed parasitic gap is licensed by a passive subject. This observation also holds for German (cf. [11c] and [11d]). Given Neeleman’s observation, the argument that comes from the licensing of (so-called) parasitic gaps is mute. (11) a. dat Jan het boek [zonder pg in te kijken] aanprijst that Jan this book [without at to look] oﬀers ‘that Jan oﬀers this book without looking at it’ b. dat [de boeken] door Jan [zonder pg in te kijken] angeprijst worden that the books by Jan without at to look oﬀered were ‘that the books were oﬀered by Jan without looking at them’

38

SCRAMBLING , REMNANT MOVEMENT , AND RESTRUCTURING IN WEST GERMANIC

c. dass Hans die Bücher [ohne pg zu lesen] weitergegeben hat that Hans the books without to read on-passed has d. dass die Bücher [ohne pg zu lesen] weitergegeben wurden that the books without to read on-passed were

In an inﬂuential article, Grewendorf and Sabel (1999) investigate scrambling in German and Japanese and argue that clause-internal scrambling in German, contrary to clause-internal scrambling in Japanese, must be A'-movement. Their claim is decisively based on two arguments/observations. First, scrambling of a potential binder in German, contrary to scrambling in Japanese, cannot license an anaphor contained in a DP. Second, they argue that the lack of WCO-eﬀects with scrambling in German should not be taken as evidence for an A-movement analysis, since clear instances of A'-movement in German can be found that do not give rise to a WCO-eﬀect, either. I will take up this issue in the following subsection. To illustrate their ﬁrst point, let us look at the data in (12). In (12a), the anaphor contained in the direct object is licensed by the c-commanding subject. In (12b), the anaphor contained in the subject cannot be licensed by the direct object, a potential antecedent, for lack of c-command. If scrambling were an instance of A-movement in German, so they argue, then the direct object in (12c) should be able to license the anaphor, contrary to fact. Since in the parallel case in Japanese the anaphor is licensed (12d–e), clause-internal scrambling in German must be an instance of A'-scrambling. (12) a. weil Peter i einen Freund von sich i eingeladen hat since Peter-NOM a friend of himself-ACC invited has b. *weil ein Freund von sich den Peter eingeladen hat since a friend of himself the Peter-ACC invited has c. ??weil den Peter ein Freund von sich eingeladen hat since the Peter-ACC a friend of himself-NOM einvited has d. ?*[IP otoga i i-no sensei-ga [VP karera i–o hihansita]] (koto) each other-GEN teacher-NOM them-ACC criticized e. ?[IP karera i–o [IP otoga i i-no sensei-ga [VP hihansita]]] (koto) they-ACC each other-GEN teacher-NOM criticized

Note, however, that the postnominal anaphor in German behaves like a longdistance anaphor in being strongly subject oriented, as is shown in (13). In (13), only the subject qualiﬁes as an antecedent for the anaphor within the complex noun phrase. Thus, I would like to contend that (12c) is ungrammatical not because scrambling is not an instance of A-movement in German but because an object does not qualify as an antecedent for the subject oriented anaphor in German. (13)

weil der Peter i den Hansj zu einem Freund von sich i/ ??j geschickt hat since the Peter-Nom den Hans-ACC to a friend of himself sent has

SCRAMBLING

39

Since the binding of anaphors is the only empirical argument they advance in favor of an A'-movement analysis and since clause-internal scrambling in German in their account would stand out compared to clause-internal scrambling in Hindi (which has been shown by Mahajan [1990] to be a clear instance of A-movement) and Japanese (which Grewendorf and Sabel themselves argue to be an instance of A-movement as well), I feel justiﬁed in concluding that the observation in (13) weakens their argument to a degree that it seems ill advised to base such a strong claim on the sole data of (12). 2.2.1

Vikner’s argument

Vikner (1994) puts forward an important comparative argument in favor of an A'movement analysis of scrambling in German. Vikner points out that scrambling in Dutch and object-shift in Scandinavian, contrary to scrambling in German, may not permute arguments. For instance, in Icelandic, a deﬁnite direct object cannot move across the indirect object, as is shown in (14). (14) a. Petur keypti bokina ekki Peter bought book-the not b. *Petur syndi bokina oft Mariu Peter showed book-the often Mary-DAT c. Peter zeigte das Buch oft der Maria Peter showed the book often the Maria-DAT

If object-shift is analyzed as A-movement into AgrO, then the contrast in (14) can be explained as a violation of (relativized) minimality (cf. Rizzi [1990]): The intervening indirect argument in an A-position would block object-shift of the direct object in (14b). If scrambling in German is A-movement as well, so Vikner argues, then it is surprising that no minimality eﬀect can be observed in this language (14c). Vikner concludes that free permutation of arguments in German calls for an account in terms of A'-movement: Not only would this account explain the lack of minimality eﬀects, but adjunction would also be the most elegant way to derive the great number of possible word orders in German. Example (15) shows for the case of three arguments that all possible permutations, namely, six, are grammatical. The permutations of the base order in (15a) are most easily derivable by (multiple) adjunction to VP and IP. (15) a. weil Peter der Maria das Buch zeigte since Peter the Maria-DAT the book-ACC showed b. weil Peter das Buch der Maria zeigte c. weil das Buch Peter der Maria zeigte d. weil der Maria Peter das Buch zeigte

40

SCRAMBLING , REMNANT MOVEMENT , AND RESTRUCTURING IN WEST GERMANIC

e. weil der Maria das Buch Peter zeigte f. weil das Buch der Maria Peter zeigte since Peter showed the book to Mary

Vikner then goes on to challenge the arguments that favor an analysis of scrambling in terms of A-movement. He notes that almost all the arguments are based on the lack of WCO-eﬀects in German scrambling. Given that wh-movement does trigger WCO-violations (16a), it is argued that scrambling and wh-movement cannot be of the same kind. However, so Vikner argues, it is not possible to have this kind of WCO-violation at all in German, and he concludes that the crucial diﬀerence would not seem to be between scrambling and wh-movement but between German and English. As is shown in (16b), not even (local) wh-movement triggers a WCO-violation in German. (16) a. *Whoi does hisi mother love? b. Weni liebt seinei Mutter nicht? who-ACC loves his mother-NOM not ‘who is such that his own mother does not love him’ c. Weni liebt [ti [seine Mutter nicht ti]]

The observation is correct, but the conclusion here is invalid. It is true that no WCO-eﬀects are observable in a single clause in German. However, when it comes to long-distance movement, we can detect an interesting contrast (cf. Frey [1993]). Long-distance wh-movement of the embedded object does give rise to a WCOeﬀect if the pronoun is contained in the matrix subject but does not if the pronoun is contained in the embedded subject, as is illustrated in (17). On the one hand, if A'-movement in German, as Vikner argues, were exceptionally not subject to WCO, then the contrast in (17) remains unexplainable. (17) a. Weni glaubt Peter dass seinei Mutter t nicht liebt who-ACC believes Peter that his mother not loves ‘Who does Peter believe is such that his own mother does not love him’ b. *Weni glaubt seinei Mutter dass der Peter t nicht liebt who-ACC believes his mother that the Peter not loves ‘Who does his own mother believe is such that Peter does not love him’

On the other hand, if we assume that the WCO-eﬀect can be circumvented by prior scrambling of the wh-word (cf. [16c]),1 then the contrast in (17) follows simply from the fact that scrambling is clause-bound. Given this assumption, scrambling can provide an A-binder for the pronoun in the embedded subject but not for the pronoun in the matrix subject. Consequently, wh-movement in (17b) only leads to an A'-bound pronoun in the matrix clause and therefore does give rise to a WCO-violation inasmuch as wh-movement does in English.

SCRAMBLING

41

However, the (well-motivated) assumption that German scrambling is A-movement gives us for free the property that scrambling, contrary to S-scrambling, is clausebound. Were scrambling indeed an operation of A-bar movement, then it would be hard to explain why long-distance scrambling, for instance in (7b) earlier, cannot make use of [Spec,CP] of the embedded clause as a licit escape hatch. If scrambling is treated as A-movement, then movement via the embedded [Spec,CP] falls prey to the uniformity condition on chains, an option that we may assume is open for the A'-movement operation of S-scrambling. At this point, a question concerning the A/A'-dichotomy arises, namely, how the distinction between A- and A'-positions is to be made. It seems ill advised to me to base the distinction on the L- or non-L-relatedness of the respective head. Due to V2, [Spec,CP] should count as L-related in German; nevertheless, movement to this position clearly and unambiguously has the properties of A'-movement. Furthermore, there are languages in which movement into a designated focus position, an operation that is standardly analyzed as A'-movement, is accompanied by verb movement. Thus, I would like to propose to make the nature of movement dependent on the type of feature that is checked in the target head, irrespective of L-relatedness. If the feature checked is an operator feature, [wh], [foc], or [neg] to name a few, then the movement operation will have the properties of A'-movement. If a non–operator feature is checked, for instance, [Speciﬁcity] or [Topicality] of an argument, then the movement operation will have properties of A-movement (see Grewendorf [2002] for a diﬀerent view). That is to say that A-movement will be extended to check other features than just Case. S-scrambling involves movement of a contrastive topic. A contrastive topic presupposes a set of alternatives and by introducing a variable that ranges over the members of the presupposed set will involve, despite its topicality, an operator feature. Therefore, S-scrambling counts as A'-movement. In a parallel fashion, the features checked by scrambling (proper) will thus have to be of the non– operator type. To conclude, Vikner’s refutation of the A-movement approach to German scrambling fails, but the comparative issue that he raises is important and must be taken seriously. I will take it up as a challenge to my approach. Assuming that German scrambling is A-movement, I have to ﬁnd an explanation for the fact that scrambling can freely permute arguments in German but is severely restricted in Dutch in a similar way as is object-shift (assuming that this operation is to be analyzed as A-movement as well) in Scandinavian. 2.2.2

Weak pronouns and scrambled DPs

Having argued that scrambling is A-movement, I have to address the question of what the landing positions and the triggers of this type of A-movement are (for a diﬀerent approach see Meinunger 2000). In this book, I argue that scrambling in German does not have a uniﬁed trigger but is initiated by two types of triggers, namely, speciﬁcity in the sense of Enç (1991)2 and scope (cf. Hinterhözl 2002a, 2004). In the following, I want to propose that scrambling of speciﬁc DPs is movement into the Speciﬁerposition of heads that license weak pronouns or clitics, as has been initially proposed by Sportiche (1996) (cf. also Cecchetto (2000), Haegeman [1996a, 1996b] for an

42

SCRAMBLING , REMNANT MOVEMENT , AND RESTRUCTURING IN WEST GERMANIC

implementation of this idea in diﬀerent languages). In German weak object pronouns can be licensed in two diﬀerent positions, one below the subject and one above it, as is shown in (18). Weak subject pronouns are licensed in a position above the position for weak object pronouns, as is shown in (19a–b). We thus arrive at the representation in (20) (licensing positions for weak pronouns are simply termed cl[itic]).3 (18) a. weil der Hans ihn gestern t getroﬀen hat since the Hans him yesterday met has b. weil ihn der Hans gestern t getroﬀen hat since him the Hans yesterday met has (19) a. weil er ihn gestern getroﬀen hat since he him yesterday met has b. *weil ihn er gestern getroﬀen hat since him he yesterday met has (20)

[C [Cl-S [Cl-O1[ Su [Cl-O2 [. . . ]]]]]]

Given the structure in (20), it can be shown that all the orders in (15) earlier can be derived without adjunction by assuming that scrambled DPs move into the Speciﬁers of heads that license weak pronouns. Example (15a) corresponds to the base order. In (15b), the direct object has moved into the Speciﬁer of the lower object licensing position, whereas the order in (15c) results from movement of the direct object into the higher licensing position for objects. Example (15d) involves movement of the indirect object into the higher object licensing position. Finally, in (15ef) both licensing positions above the subject are ﬁlled with the two objects in alternating orders.4 Summing up, the scrambling orders in (15) can be derived without adjunction, simply by using positions for which we have independent evidence. One argument that could be launched against this account of scrambling is based on the observation that weak pronouns are subject to conditions that are not observed by scrambled DPs and which are speciﬁed in (21). (21) a. Weak pronouns appear in the order NOM > ACC > DAT. b. Weak subject pronouns precede scrambled DPs.

Weak pronouns reach their licensing positions either by XP-movement only or by XP-movement (if necessary) and subsequent head movement. In the latter case, they form a cluster, and I would like to propose that condition (21a) pertains to clitic clusters, which are licensed in the highest licensing position for weak pronouns. Since scrambled DPs do not form clusters, they are not subject to condition (21a). Furthermore, if weak pronouns “cliticize” by XP-movement only, they can appear in various orders and diﬀerent positions, very much like scrambled full DPs. This later option is available in West Flemish (cf. Haegeman [1994]) and various German dialects (cf. Gärtner and Steinbach [2000]).

SCRAMBLING

43

To account for condition (21b), I will assume that weak subject pronouns always have to move to the highest licensing position for weak pronouns and that these licensing heads cannot simultaneously license an element in the head position and another element in their Speciﬁer. The proposal that one type of scrambling is movement into the Speciﬁers of heads that license weak pronouns is supported by cross-linguistic considerations. Alexiadou and Anagnostopoulou (1997) argue that there are some striking resemblances between scrambling in Germanic and clitic doubling constructions in Greek and Romance (cf. also Cecchetto [2000]). It is interesting to note that in those languages (or dialects) that allow for clitic doubling as in River Plate Spanish and Romanian a typical WCO-violation can be circumvented by the introduction of a clitic (data taken from Suñer [1988] and Dobrovie-Sorin [1990], respectively), as is indicated in (22) and (23). In River Plate Spanish (22), the WCO-eﬀect induced by QR at LF is obviated if the quantiﬁer phrase is doubled. Likewise, a WCO-eﬀect induced by the overt operation of wh-movement in Romanian (23) is obviated by clitic doubling of the wh-phrase. (22) a. *Sui madre quire a todosi their mother likes a everyone b. Sui madre los quire a todosi their mother them likes a everyone (23) a. *[A cuales de ellos] i no aguanta ni su i madre ‘a’ which ones of them not stand even their mother b. [A cuales de ellos] i no los aguanta ni su i madre ‘a’ which ones of them not them stand even their mother

We have seen earlier that overt scrambling in German can obviate WCO-eﬀects induced by QR and proposed that the lack of WCO-eﬀects in local wh-movement can be explained by assuming that the wh-object ﬁrst scrambles across the subject that contains the pronoun, providing an A-binder for the latter before wh-movement moves the wh-object into [Spec,CP]. Assuming that scrambling is movement into the Speciﬁers of clitic positions will allow me to provide a uniform explanation of the parallel eﬀects of scrambling in German and clitic doubling in Romance (and Greek) with respect to WCO, if we make the assumption, which seems natural, that clitic doubling involves movement of the doubled phrase into the Speciﬁer of a functional projection licensing clitics but spells out the lower copy, which is then only prosodically marked as being part of the background of the sentence (while in German this is indicated by the overt position of a discourse-anaphoric element via Spell-out of the higher copy). The data in (22) and (23) provide indirect evidence against Vikner’s and Grewendorf and Sabel’s (1999) position that WCO-eﬀects are not a good criterion for deciding whether scrambling is A- or A'-movement, since they clearly show that a WCO-eﬀect that is induced by an uncontroversial instance of A'-movement is

SCRAMBLING , REMNANT MOVEMENT , AND RESTRUCTURING IN WEST GERMANIC

44

circumvented by the introduction of clitics that are assumed to be base generated in or related to an uncontroversial instance of an A-position.

2.3 Scrambling and optional movement 2.3.1

The account of Haider and Rosengren (1998)

H&R argue that scrambling is a cover term for distinct and independent constructions whose common feature is word order variation and put forth a rather restrictive theory of scrambling, the most important properties of which can be summarized as follows: 1. Scrambling is restricted to OV languages in which arguments are identiﬁed with an independent, non-positional mechanism 2. Scrambling is restricted to the permutation of arguments and applies exclusively within the lexical domain of the selecting head 3. Scrambling is truly optional—in the sense that no trigger can be identiﬁed that would force obligatory scrambling—as such it is blindly generated in the syntax and exploited at the interface levels The ﬁrst claim amounts to saying that scrambling cannot occur in VO languages at all and is restricted to those OV languages that license their arguments nonpositionally. H&R assume that a language licenses its arguments non-positionally, if it has V-class dependent base orders. In chapter 1, we have seen that German has V-class dependent base orders (cf. section 1.2). Thus, German is an OV language that allows for scrambling in their account. Another language that has V-class dependent word orders is Icelandic. But since Icelandic is a VO language, it lacks scrambling according to H&R. The second claim needs some further explication. For H&R, it is a deﬁning feature of scrambling that it always involves “bypassing” of at least one non-empty (argument) base position. This assumption has two important consequences. First, scrambling is non–string vacuous. H&R are endorsing a representational view of grammar, in which economy principles select the minimal convergent structure, excluding representations like (24b). (24) a. dass der Peter die Maria besucht hat that the Peter the Maria visited has b. dass der Peteri die Mariaj ti tj besucht hat

Second, it has the following implication for adverbials: If adverbials do not have structurally unique base positions, as H&R assume, then the relative order of arguments and adverbials cannot be an instance of scrambling. For H&R, so-called scrambling of arguments to the left of adverbials is the result of alternative adverb placement. In a similar vein, object-shift in Icelandic is not an instance of scrambling, either, but should be explained in terms of alternative adverb placement.

SCRAMBLING

45

Furthermore, scrambling positions are restricted to left-adjoined positions in head-ﬁnal lexical projections. This restriction follows from Haider’s basic branching condition (BBC) and the condition that nodes attached to the projection line are licensed directionally. (25)

BBC: The branching node of the projection line is to the right of its sister node.

The BBC excludes left-branching structures but allows for structures with multiple adjunctions within the maximal projection of a head. Example (26a) shows the base structure of a ditransitive verb in a head-ﬁnal language. This structure is ﬁne, since it obeys the BBC and since all nodes are licensed in the parametrically given directionality of an OV language, namely, toward the left. Example (26a), however, could also be the base structure of a verb with two arguments and a trace. In this case A and C would be coindexed and the structure would be appropriate for A having scrambled across B. (26) a. [VP A [V’ B [V’ C V ]]] b. [VP A V [VP B [V’ V C ]]]

Example (26b) shows the base structure of a ditransitive verb in a VO language. Since the licensing direction is to the right in VO languages, complex VO-structures, contrary to complex OV-structures, must contain a V-chain to have all attached nodes licensed in the canonical direction. Adjoined positions in (26a) are potential argument identiﬁcation positions and hence count as A-positions, since they lie in the licensing direction of the head. Thus, scrambling in an OV language has radically diﬀerent properties from the functionally parallel operation in a VO language, so H&R conclude: Adjunction to a VO-phrase does NOT create a new argument identiﬁcation position. For H&R, it thus follows that scrambling in the strict sense as deﬁned earlier cannot occur in VO languages. 2.3.2

Problems of the account of H&R

The account of H&R raises a number of questions. First and most important, their account presupposes that OV and VO languages have radically diﬀerent base structures, that is, project their argument structures in very diﬀerent ways. Certainly the introduction of VP-shells was originally motivated by the c-command properties of the arguments of ditransitive verbs (cf. Larson [1988])—properties that are also captured by the representation that H&R provide for a ditransitive verb in an OV language—but the existence of VP-shells has meanwhile been given semantic motivation in the work of Hale and Keyser (1993). The assumption of abstract verb positions—adopted in Chomsky (1995b)—which allows for signiﬁcant generalizations about transitive and non-transitive verbs and their respective thematic and Caseassigning properties, if correct, should also hold for OV languages. However, the properties of abstract verb positions cannot be captured in H&R’s representation of the base structure of a verb in OV languages.

46

SCRAMBLING , REMNANT MOVEMENT , AND RESTRUCTURING IN WEST GERMANIC

Second, their account makes a number of predictions that, given what we know about these languages, seem at least problematic. The ﬁrst type of prediction involves free word order languages like Polish and Russian, which are standardly assumed to be VO languages. Either these languages have to be reanalyzed as OV languages or one has to assume that there are free word order phenomena that cannot be captured in terms of scrambling. The second type of prediction involves VO languages like English or the Mainland Scandinavian languages and amounts to saying that these languages should allow for diﬀerent word orders postverbally. For instance, it is not clear in H&R’s account what prevents the direct object from adjoining above the indirect object or moving into a higher Speciﬁer, in the VP-shell structure in English, giving rise to sentences like (27a). (27) a. *John gave the book Mary b. Hans hat gestern das Buch gelesen Hans has yesterday the book read c. Hans hat das Buch gestern gelesen

Third, the question arises whether the distinction between scrambling of an argument across another argument and scrambling of an argument across an adverb is real. The latter process in H&R’s account is dealt with by the mechanism of alternative adverb placement. Thus, the alternative orders in (27b–c) are assumed not to be the result of scrambling but to be the consequence of generating the adverb gestern ‘yesterday’ in diﬀerent base positions. This assumption makes the following prediction concerning quantiﬁed adjuncts and arguments: Both orders should be unambiguous. The scrambling approach predicts that the two orders should behave diﬀerently with respect to the degree of their ambiguity: For speakers of dialect 2, the scrambling order, in contradistinction to the base order, should be ambiguous. Frey (2000) and Frey and Pittner (1998, 1999) argue that adverbs in German do have syntactically determined base positions. Furthermore, they show that scrambling of arguments across adjuncts as well as of adjuncts across other adjuncts gives rise to ambiguous readings (examples taken from Frey [2000]). (28) a. weil an fast jedem Tag mindestens einer eine Wahlrede halten wird (unambiguous) since on almost every day at least one person a speech give will b. weil mindestens einer an fast jedem Tag eine Wahlrede halten wird (ambiguous) since at least one person on almost every day a speech give will

Given that scrambling of arguments across adjuncts has the same properties as scrambling of arguments across arguments, both operations should be given a uniform account, which is impossible in the approach of H&R.

2.4 The trigger problem H&R argue that scrambling may not be considered as being triggered by a feature that needs to be checked in a designated position. They point out that trigger accounts

SCRAMBLING

47

are often circular in the sense that they postulate features that apparently are only introduced into the theory to trigger scrambling. Moreover, H&R argue that accounts that introduce substantive, independently motivated features prove to be either too weak, too strong, or both. A trigger account is too weak if the trigger just involves DP-type features, like Case or a strong [D]-feature, since it does not cover scrambling of PPs and CPs in German. What is needed is substantive features that are independent of the syntactic category of an argument. A number of features that relate to the semantic or discourse properties of an argument have been proposed in the literature. Such accounts, so H&R argue, are too strong by necessity since they entail that if the respective feature is present scrambling is obligatory, and then they go on to show that scrambling is optional. First, they argue that the semantic and pragmatic eﬀects induced by scrambling cannot be taken to be the triggering factor of scrambling, since the interpretational eﬀect that is induced by scrambling is found in unscrambled structures as well. Scrambling seems to reduce, but not to replace, the interpretation potential of a phrase. As evidence they provide examples in which a generic interpretation (29a), a speciﬁc interpretation of an indeﬁnite (29b), and a speciﬁc deﬁnite interpretation (29c) are applicable to DPs in situ. For instance, in (29a) the deﬁnite DP object follows an indeﬁnite pronoun subject. So the generically interpretable DP is likely to be in situ. Analogous considerations apply to (29b–c). In (29b), the indeﬁnite is interpreted as speciﬁc, given that its reference is picked up by Maria, though it seems to occur in its base position following an indeﬁnite subject. The same holds for the deﬁnite DP her dress in (29c), which—though occurring in its base position—can have a speciﬁc interpretation referring to Maria’s dress. (29) a. dass wer die Pockenviren ausrotten sollte that someone the pockvirus exterminate should b. wenn wer eine rothaarige Frau sucht dann ist das Maria if someone a red-haired woman seeks then is it Maria c. dass er wem ihr Kleid gezeigt hat, hat Maria nicht gefallen that he someone her dress shown has, has Maria not pleased

Note that indeﬁnites may not occur in the domain of negation in German (30a–b). However, H&R argue that this case of alleged obligatory scrambling rests on a controversial premise, namely, that negation universally c-commands the whole VP. They argue that the relevant condition for German and Dutch is that negation only needs to c-command the ﬁnite verb in its base position. This condition is fulﬁlled if negation is adjoined to V below the base position of arguments rendering scrambling of indeﬁnite w-pronouns superﬂuous in order to derive the grammatical order in (30b). (30) a. *dass jemand nicht wen jagte that someone not somebody chased b. dass jemand wen nicht jagte that someone somebody not chased

48

SCRAMBLING , REMNANT MOVEMENT , AND RESTRUCTURING IN WEST GERMANIC

Note that if this idiosyncratic assumption about the base position of negation in German is not taken, the data in (29) are not less problematic. Even if we assume that indeﬁnite w-pronouns can scramble in principle, it seems hard to come up with a motivation that has the indeﬁnite argument move across the generically interpreted deﬁnite argument. Other triggers that have been proposed involve semantic-driven movement, that is, movement of weak DPs into strong DP positions, as in De Hoop (1992), or familiarity, as in the account of Corver and Delﬁtto (1997). But again, so H&R argue, these accounts prove to be too strong. If indeed scrambling was triggered by a speciﬁc interface feature, (31a–b) should diﬀer to the extent deﬁned by the absence or presence of that feature. However, (31a) can have the same interpretation as (31b), namely, that ‘Max in general admires prima ballerinas.’ Therefore, so H&R argue, the claim that a strong “generic” feature triggers scrambling cannot be correct. (31) a. dass ja Max Primaballerinas bewundert that PRT Max prima ballerinas admires b. dass ja Primaballerinas Max bewundert that PRT prima ballerinas Max admires

In this case, I ﬁnd the evidence less convincing. In my judgment, I get an existential reading of the bare plural if the direct object in (31a) receives nuclear stress and a generic reading if the verb is assigned nuclear stress. This latter fact would indicate that the direct object in this reading of (31a) does not occupy its base position. In this case, it also seems plausible that the deﬁnite subject DP Max has moved to a high position above the generically interpreted object. This analysis of (31a) presupposes that the modal particle ja, which is generally assumed to mark the VP-boundary (cf. Diesing [1992], Kratzer [1995]) can occupy a position high up in the clausal domain. Such an analysis, however, is not available to H&R since they exclude string vacuous scrambling quite generally, as we have seen in the previous section. In a similar vein, if scrambling were triggered by a discourse-type feature, like [+Familiar] as in Corver and Delﬁtto’s account, one would expect that the reading associated with scrambled constituents is absent in the base order. Example (32) provides a context in which the existence of a unique white ball is given or familiar. Nevertheless, scrambling is not obligatory. In fact, the unscrambled order is the preferred one. (32)

In diesem Sack sind zwei rote und eine weisse Kugel in this bag are two red and one white ball a. Wenn jemand die weisse Kugel zieht, hat er gewonnen if someone the white ball draws, has he won b. ?Wenn die weisse Kugel jemand zieht, hat er gewonnen

I do not think that the context in (32) is well chosen for two reasons. First, note that the indeﬁnite in (32) is interpreted as universally quantiﬁed (for every x, if x draws

SCRAMBLING

49

the white ball, x wins), so that it is far from clear that the indeﬁnite is in its base position. Second, the context itself may not be appropriate to trigger scrambling of a familiar discourse referent, as the example (32') with a pronoun shows. In (32'), even the pronoun is awkward in a scrambled position, possibly because the discourse referents have been introduced as a group (or sum) and the choice of one member over the other seems to require some kind of focal stress (selecting but not contrastive focus). (32') In diesem Zimmer beﬁnden sich ein Mann und eine Frau in this room there-are a man and a woman a. Wenn jemand ihn sprechen will, wählt er die 9 if someone him speak wants, dials he the 9 b. ??Wenn ihn jemand sprechen will, wählt er die 9

With the reservations indicated earlier, I agree with H&R’s observations. Also, I think that these observations are rather problematic for simpleminded trigger accounts. To summarize the discussion earlier, we have seen that deﬁnite and indeﬁnite speciﬁc DPs obligatorily scramble across negation while in other contexts, generic, deﬁnite, speciﬁc, and familiar DPs seem to be allowed to stay in their base position. So the picture we arrive at is rather puzzling. There is an instance of scrambling that is indisputably obligatory. This operation is scrambling for reasons of scope taking. In German, a quantiﬁed object has to scramble across a quantiﬁed subject to take scope over it, as is shown in (33). (33) a. weil jeder mindestens zwei Bücher gelesen hat (SU > OB only) since everyone-NOM at least two books-ACC read has b. weil mindestens zwei Bücher jeder gelesen hat (OB > SU preferred) since at least two books-ACC everyone-NOM read has

Note, however, that the property of obligatoriness does not make it easier to provide a satisfactory trigger account of at least this type of scrambling. The diﬃculty arises since it is quite inappropriate to assign the respective scope feature to any speciﬁc F-head in the clause. Scope features—if we introduce the features [w], [n], and [i] (for wide, narrow, and intermediate scope) for the sake of concreteness—by their very nature are not absolute properties. Scope properties are relative properties: A DP has wide scope only in relation to another DP. Thus, a checking account in terms of privative features seems inadequate for scope phenomena in principle. In the account of H&R, the problem does not arise, since scrambling is not considered as movement into a designated position to check the relevant feature. In their account, scrambling is the result of the syntactic mechanism of chain formation, applying blindly in the identiﬁcation domain of the selecting head, which can be exploited at the interface. In this approach, the presumed scope features [w], [i], and [n] could be treated as interpretable semantic features of scopal elements that do not require syntactic checking and have to obey the following interface condition at LF, which will ﬁlter out all derivations/representations that do not conform to it.

50

SCRAMBLING , REMNANT MOVEMENT , AND RESTRUCTURING IN WEST GERMANIC

(34) Scopal Well-formedness: a. A phrase with the feature [w] must c-command a phrase with the feature [n]. b. A phrase with the feature [i] must c-command a phrase with the feature [n] and must be c-commanded by a phrase with the feature [w].

Note that this conception of grammar is very much in the spirit of minimalism. It is based on a maximally narrow syntax in which only morphological and a small selected set of formal features drive the derivation, with semantic and pragmatic properties being utilized in the workings of sophisticated interface conditions. In short, the syntactic derivation is not geared by features that relate to semantic or pragmatic properties of constituents. This analysis seems to be a relatively simple and maximally elegant solution to the problem of relational features, but it comes with the prize of allowing for optional operations in the syntax.

2.5 Coping with optionality Taking H&R’s observations at face value, we are confronted with two problems: (1) There are data that require elements like w-indeﬁnites to scramble, though the accounts given so far would have them as non-scrambleable elements. (2) There are data that suggest that elements that should scramble, like speciﬁc DPs, occupy their base position. I will tackle the ﬁrst problem by proposing that next to [Speciﬁcity] there is another factor that triggers scrambling, namely, scope. H&R assume that w-indeﬁnites do not scramble. However, this assumption is wrong. That w-indeﬁnites can indeed scramble in German is shown in (35a–b). In (35a), the w-indeﬁnite is in its expected position following the indeﬁnite (negative) subject. (35a) is unambiguous, meaning ‘that nobody met anyone does not surprise me’. In (35b), the w-indeﬁnite has scrambled across the subject, yielding the interpretation ‘that there is someone that nobody met does surprise me’. (35) a. dass keiner wen getroﬀen hat, überrascht mich nicht that no one–NOM someone-ACC met has, surprises me not b. dass wen keiner getroﬀen hat, überrascht mich that someone-ACC no one–NOM met has, surprises me

To utter (35b) the speaker does not need to have a speciﬁc individual in mind: it suﬃces that he has some evidence that there is a person with the relevant property. Thus, the w-indeﬁnite is only speciﬁc in the sense that it is the DP with widest scope, but it is not speciﬁc in the sense that it is presupposed, known to the speaker, or in any other way anaphorically anchored in the context. This shows that scope properties of arguments are triggers of scrambling, and it conﬁrms my former observation that scrambling in this case is obligatory since the reading of (35b) is only available if the object is scrambled across the subject.

SCRAMBLING

51

Assuming that scope is a trigger for scrambling, of course, requires me to explain how scope requirements can be dealt with in a feature-checking mechanism. This issue will be taken up in section 2.6. In the following, I want to tackle the second problem, namely, the issue that deﬁnite DPs and the like seem to scramble optionally. That is, to address the observation that sometimes they appear to have moved and sometimes they appear to be in their base position. My answer to the optionality issue is that DPs that have the triggering property always undergo scrambling but that there are intervening factors that prevent the Spell-out of the scrambled DP in its checking position. To show that the distribution of scrambleable arguments is not optional but subject to speciﬁc conditions, I will investigate the distribution of arguments with respect to the negative marker. More speciﬁcally, I will address the assumption by H&R that the negative marker in German occupies a very low position in the clause. This assumption allows them to refrain from assuming obligatory scrambling of speciﬁc DPs to account for the unmarked word order in (36). That the negative marker in German occupies a much higher position in the clause is shown in the following section. (36)

2.5.1

weil der Hans die Maria nicht liebt since the Hans the Maria not loves

The syntax of the negative marker in German

The negative marker nicht (‘not’) obligatorily precedes manner adverbs as is shown in (37). Assuming as is standardly done that manner adverbs are adjoined to the VP (I will show in chapter 4 that they actually occupy a higher position in the clause), I conclude that the negative marker in German cannot occur VP-internally—adjoined to the verb as is assumed by H&R—but occupies a functional head position in the I-domain of the clause, as is the standard assumption for the syntactic representation of negation in various languages (cf. Haegeman and Zanuttini [1991]). (37) a. weil der Hans das Buch nicht sorgfältig gelesen hat since the Hans the book not carefully read has ‘since Hans has not read the book carefully’ b. ??weil der Hans das Buch sorgfältig nicht gelesen hat since the Hans the book carefully not read has

All deﬁnite nominal arguments precede the negative marker (38). If an indeﬁnite NP precedes the negative marker, it is interpreted as speciﬁc (39a). If an indeﬁnite NP follows the negative marker, the reading one gets most easily is the one in which the negative marker is interpreted as negating only the nominal argument that follows it (this holds for deﬁnite and indeﬁnite NPs), which receives a (negative) contrastive interpretation. Depending on whether it is the determiner or the noun that receives the non-neutral (contrastive) stress, the negative marker negates the (cardinality of the) determiner or the descriptive content of the NP in (39b).

52

SCRAMBLING , REMNANT MOVEMENT , AND RESTRUCTURING IN WEST GERMANIC

(38) a. weil der Hans das Buch nicht gelesen hat since the Hans the book not read has ‘since Hans did not read the book’ b. ??weil der Hans nicht das Buch gelesen hat5 since the Hans not the book read has (39) a. weil Hans ein Buch nicht gelesen hat (only speciﬁc interpretation) since Hans a book not read has ‘there is a book that Hans did not read’ b. weil Hans nicht ein Buch gelesen hat (sondern zwei; sondern ein Journal) since Hans not a book read has (but two; but a journal) ‘Hans did not read ONE book; he read TWO books; Hans did not read a BOOK; he read a JOURNAL’

It has been argued that in these cases the negative marker acts as “constituent negation.” Because the negative marker can be topicalized together with an argument of the verb, it is assumed that negation in German can simply be adjoined to an XP that it narrowly or exclusively negates (40a–b). (40) a. [?P Nicht der Hans] hat das Buch gelesen not the Hans has the book read ‘It was not Hans that read the book’ b. [DP nicht [DP der Hans]]

The nonspeciﬁc (existential) interpretation of an indeﬁnite NP in a negated German sentence is expressed with the determiner kein (41a). In this case, the negative marker is non-overt or fused (41b).6 Thus, we cannot determine in (41) whether a nonspeciﬁc indeﬁnite NP has to move across the negative marker. That a nonspeciﬁc indeﬁnite NP must at least move up to the negative marker is indicated by the behavior of negative existentials in Upper Austrian. This dialect exhibits, like other Bavarian dialects, the phenomenon of negative concord, which allows for the Spellout of the negative marker even in the presence of negative constituents. As (42) shows, the negative existential NP kein Buch (‘no book’) has to precede the negative marker net (‘not’). In the following, I will assume that a negative existential NP with a kein-determiner occupies [Spec,NegP]. (41) a. weil Hans kein Buch gelesen hat since Hans ‘kein’ book read has ‘since it’s not the case that Hans read a book’ b. weil Hans (*nicht) kein Buch (*nicht) gelesen hat since Hans not ‘kein’ book not read has

SCRAMBLING

53

(42) a. woei da Hans ka Buach net glesn hot since the Hans ‘kein’ book not read has ‘since it’s not the case that John read a book’ b. *woei da Hans net ka Buach glesn hot since the Hans not ‘kein’ book read has

That nonspeciﬁc indeﬁnite NPs may not only follow what may be analyzed as constituent negation but may also follow sentential negation can only be shown with more than one indeﬁnite NP. If a sentence contains more than one indeﬁnite NP, the highest nonspeciﬁc argument is spelled out with the kein-determiner, that is, checks the negative marker in [Spec,NegP]. This is shown in (43). Example (43a) is a sentence that contains three indeﬁnites. If all three DPs are interpreted nonspeciﬁcally, the negation of (43a) must be (43b). In (43c), where the indirect object is spelled out with a keindeterminer, the higher subject must be interpreted as speciﬁc if we exclude focus-aﬀected and quantiﬁcational readings. Example (43b) in conjunction with (42) and (43c) shows— if we analyze the kein-phrase as occupying [Spec,NegP]—that nonspeciﬁc indeﬁnites may follow negation. Given these conclusions, (43b) is analyzed as shown in (43d). (43) a. weil ein Mann einer Frau eine Blume schenkte since a man a woman-DAT a ﬂower gave b. weil kein Mann einer Frau eine Blume schenkte since ‘kein’ man a woman-DAT a ﬂower gave ‘since it is not the case that some man gave some ﬂower to some woman’ c. weil ein Mann keiner Frau eine Blume schenkte since a man ‘keiner’ Frau a ﬂower gave ‘since a (certain) man did not give some ﬂower to some woman’ d. [CP weil [TP [NegP kein Mann [einer Frau eine Blume [VP schenkte]]]]]

Also, quantiﬁed NPs may follow the negative marker without giving rise to a constituent negation-interpretation (44a–b). In (44a–b), the negative marker can act as sentence negation. As (44c) shows, the negative marker can be construed narrowly with the higher subject NP, which is a typical property of sentence negation but is unexpected of constituent negation. Hence it follows that quantiﬁed NPs need not move across negation. (44) a. weil der Hans nicht viele Bücher liest since the Hans not many books reads ‘since it is not the case that Hans reads many books’ b. weil der Hans nicht jede Frau anbetet since the Hans not every woman adores ‘since it is not the case that Hans adores every woman’

54

SCRAMBLING , REMNANT MOVEMENT , AND RESTRUCTURING IN WEST GERMANIC

c. Der HANS hat nicht viele Bücher gelesen, der PETER hat viele gelesen the Hans has not many books read, the Peter has many read ‘it was not Hans but Peter who read many books’

Let me summarize what I have observed so far. Speciﬁc NPs obligatorily move across the negative marker while nonspeciﬁc indeﬁnite NPs, unless they move into [Spec,NegP] to check sentential negation, remain below the negative marker. Deﬁnite NPs may only then remain below the negative marker if they receive a contrastive interpretation. Along the same lines, nonspeciﬁc indeﬁnite NPs may defy movement into [Spec,NegP] only if they receive a contrastive interpretation as illustrated in (39b) earlier. Finally, quantiﬁed NPs depending on their scope may stay below or move across the negative marker. A QP below the negative marker may have a speciﬁc or nonspeciﬁc interpretation. Example (45a) depicts a speciﬁc, that is, partitive, QP that occurs below the negative marker that acts as sentence negation. But if a QP scrambles higher than the negative marker, then it can, like indeﬁnites, only have a speciﬁc interpretation as is shown in (45b). (45) a. Der HANS hat nicht viele der Bücher gelesen, der PETER hat viele davon gelesen the Hans has not many of the books read, the Peter has many thereof read b. weil der Hans viele Bücher nicht gelesen hat (only speciﬁc interpretation) since the Hans many books not read has ‘since for many of the books it holds that Hans did not read them’

Following the proposal in section 2.2.2, I assume that movement across negation is triggered by speciﬁcity. To get rid of constituent negation, I will take up a proposal by Richard Kayne (p.c.) and assume that there is a focus-phrase just below negation into which contrastively focused elements move. The resultant structure is given in (46). I have not said anything so far about where and how arguments are Case-licensed in German. In chapter 4, I will argue that they are licensed outside of VP in Case-agreement projections, as is indicated in (46). The resultant structure is given in (46) (recall that according to [20] scrambled [speciﬁc] DPs can be licensed below or above the subject). (46)

[CP [IP Speciﬁcs (Su) T Speciﬁcs [Neg [ Focus AgrC [V]]]]]

Given (46), the regularities discussed earlier can be described in the following way: It seems that a strong focus-feature blocks the movement of deﬁnite NPs into the licensing positions of speciﬁcs as well as the movement of the highest nonspeciﬁc indeﬁnite into [Spec,NegP]. A speciﬁc QP may stay below negation if it is to be read with narrow scope, while a deﬁnite NP must (in the absence of any focus-feature) check its speciﬁcity feature. It is not evident how to properly express these regularities in a system of feature checking. One possibility is to assume that something like (47) holds. (47)

Once Case is checked, only the feature of a DP with the closest licensing head is checked overtly.

SCRAMBLING

55

The statement in (47) may be okay as a descriptive generalization, but it is unsatisfactory as a principle of grammar. Why should it be that the possibility of checking a certain feature is dependent on the presence or absence of certain other features? So (47) cannot be correct. However, (47) has the virtue of showing that the distribution of arguments and of deﬁnite DPs in particular is not optional at all, as claimed by H&R, but subject to speciﬁc restrictions. In the following section, I will provide an account in terms of conditions on the Spell-out of copies that allows us to get rid of the generalization in (47) and to solve the problem of optionality. 2.5.2

Conditions on spell-out

The solution to the problem posed by (47), namely, the fact that certain features can only be checked in the absence of other features (conditional checking), is to assume that feature checking is unconditional but to refrain from positing that the checked category is unconditionally spelled out in the position of the highest feature checked. Let us assume for the sake of discussion, as earlier, that speciﬁcity is the relevant feature that triggers movement of arguments across negation. Then, we may assume (1) that a speciﬁc DP (independently of other features) always moves to check its feature in a position above the negative marker and (2) that its Spell-out is determined by the conditions in (48). (48) a. contrastive DPs are spelled out in the focus position b. quantiﬁed DPs are spelled out in their scope position c. de-accented DPs are spelled out before accented DPs

Condition (48a) captures the data in (38) and (39) earlier, which shows that a deﬁnite DP unless contrastively stressed appears before negation. Condition (48b) captures the data in (44) and (45) earlier, namely, the fact that speciﬁc QPs are spelled out above negation, unless they are to be interpreted with narrow scope with respect to negation. The notion “scope position” that is used in (48b) will be deﬁned in section 2.6. While the conditions in (48a–b) are related to the LF-interface, condition (48c) is a condition related to the PF-interface. While the LF-related conditions are inviolable—we may assume that the heads that license contrastive focus und scope have a phonological EPP-feature—the PF-related condition is soft. This is illustrated in (49). For the working of condition (48c) I assume that backgrounded and discourseanaphoric DPs are deaccented. The question in (49) can be answered either with (49a) or with (49b). While (49a) is completely unmarked and the preferred option, (49b) is slightly marked but completely grammatical. (49) Q: Wem hat Otto das Buch gegeben? Who has Otto the book given? A: a. Otto hat das Buch dem PETER gegeben b. Otto hat dem PETER das Buch gegeben Otto has (the book) to Peter (the book) given

56

SCRAMBLING , REMNANT MOVEMENT , AND RESTRUCTURING IN WEST GERMANIC

The marked/unmarked status of the examples of (49) can be ascribed to the workings of an interface condition that determines the mapping between informationstructure and prosodic structure. I assume that in both answers the direct object has scrambled (across the indirect object) with the diﬀerence following from spelling out either the higher or the lower copy. (50)

Interface condition The phonological phrase that contains the focus (main accent) must be rightmost within its intonational phrase (cf. Chierchia [1986], Frascarelli [2000], Hayes and Lahiri [1991]).

As is illustrated in (51), the diﬀerent status of the answers follows from the prosodic constraint in (50). Example (51) shows the prosodic structure of both answers, where parentheses indicate phonological phrases and iP indicates an intonational phrase. We see that (51a) optimally fulﬁlls the prosodic condition in (50). While (51b) violates this prosodic condition. I propose that this is the reason why (51a) is preferred over (51b). Example (51b) is repaired by being assigned a stronger pitch accent, while in (51a) the assignment of normal sentence accent suﬃces to mark the focused constituent. Thus, (51b) is prosodically more marked than (51a), but speakers are free to use the more marked forms for their communicative purposes, whatever they are. (51) a. [iP (Otto) (hat das Buch) (dem PEter gegeben)] b. [iP (Otto) (hat dem PEter) (das Buch gegeben)]

In sum, (48c) is a statistical consequence of the workings of the interface condition in (50). Note that I managed to restrict optionality to the workings of condition (50) only. By this maneuver, optionality is conﬁned to a PF-interface condition that speciﬁes prosodic requirements on the linearization of phonological material. The syntactic computation, however, including the branch leading to the LF-interface, is deterministic throughout. Thus, contrary to optimality theoretic syntax, I assume that surface constraints have no place in the grammar itself. Alternatively, I assume that the grammar speciﬁes a limited set of options (one being the Spell-out of copies) that are ﬁxed by the child who has access to (next to general syntactic principles) vocabulary and conditions operative at the interface levels only, which are necessarily “surfacey.” In short, “surface constraints” are only relevant for the Spell-out of copies, not for the internal working of syntax that creates the copies.

2.6 Feature checking and scope In section 2.4, we have seen that scrambling for reasons of scope taking, though being non-optional, still poses a problem for trigger accounts because of the relational nature of scope. Also, since I showed that scrambling proper is an A-movement

SCRAMBLING

57

operation, we need a ﬂexible mechanism to replace adjunction. That is, we need to devise a mechanism that allows us to check one and the same feature in diﬀerent positions in diﬀerent occasions. In this section, I want to sketch a possible account of scope in terms of feature checking and explore its implications for the theory of grammar. For the sake of discussion, let us assume that the features introduced earlier, [w], [i], and [n], are suﬃcient to account for scope phenomena and, furthermore, that these features drive the derivation. One question that arises is whether these features are formal or purely semantic in nature. For sure, these features are non-lexical and thus cannot be part of the numeration. Take note of the fact that these features are not in any sense part of the lexical properties of a quantiﬁer. As such these are essentially diﬀerent from the scope features that Beghelli and Stowell (1996) abstract from speciﬁc (types of ) quantiﬁers to account for their scope properties. The scope features I am assuming are essentially independent of the properties of lexical items and are purely relational (or syntactic) in nature. Thus, it seems appropriate to assume that they cannot be assigned to a speciﬁc lexical or functional head in the numeration. 2.6.1

Relational features and the Inclusiveness Condition

Assuming that the Inclusiveness Condition (cf. Chomsky [1995b]) only holds for linguistic items in the numeration, I propose that non-lexical features can be assigned to any head in the course of the derivation. To restrict scrambling to the middle ﬁeld, I assume that scope features can be assigned to extended projections of the verb. The enrichment of an existing structure with a non-lexical feature is possible in two ways, as speciﬁed in (52). (52) Dynamic feature assignment a. Conservative extension: assign the feature to an existing structure (the head at the root) in the course of the derivation. b. Dynamic extension: assign the feature to (a copy of) a bare functional head and merge the head with the existing structure.

Economy of derivation guarantees that dynamic extension in (52b), which is more complex than conservative extension in (52a), will only apply in case the derivation that uses only conservative extension does not converge. Furthermore, fewest steps ensures that these features are not assigned repeatedly to a structure and shortest step requires that only the smallest extension that guarantees a convergent derivation will be taken. This is illustrated in (53). In (53) the assignment of the scope features to the corresponding DPs is left out for reasons of space, but I assume that DPs (and other scope-bearing elements)—according to the intentional purposes of the speaker— come equipped with scope features. In other words, the mechanism in (52) only concerns the assignment of formal scope features to functional heads, which check the semantic features of arguments and adjuncts corresponding to the reading that a speaker has in mind.

58

SCRAMBLING , REMNANT MOVEMENT , AND RESTRUCTURING IN WEST GERMANIC

(53) a. [CP weil . . . [jeder AgrNom[w] [mindestestens eine Frau AgrAcc[n] [VP liebt]]]] since everyone-NOM at least one woman-ACC loves b. [CP weil . . . [F[w] [jeder AgrN[n] [mindestestens eine Frau AgrA [VP liebt]]]]] c [CP weil [F[w] . . . [F[w] [jeder AgrN[n] [mindest. eine Frau AgrA [VP liebt]]]]]] d. [CP weil [T [F[w] [jeder AgrN[n] [mindestestens eine Frau AgrA [VP liebt]]]]]] e [CP weil [F[w] . . . [T [jeder AgrN[n] [mindest. eine Frau AgrA [VP liebt]]]]]]

In (53a), the scope features are assigned by operation (52a) to the respective Caseagreement heads. Assuming that the subject is to be read with wide scope with respect to the object, no extension of the derivation, that is, no scrambling, is necessary. In (53b–e), I assume that the object is to be read with wide scope with respect to the subject. The structures in (53b–e) show possible extensions of the derivation before the direct object moves to check its scope feature. In this case, the derivation (53b) wins out over the derivation in (53c) due to fewest steps. In the same vein, the derivation in (53d) will be selected over the derivation in (53e), since the movement of the object in (53d) to check its scope feature will be shorter than its movement in (53e) (shortest step). In short, the assignment of scope features has to obey the regular economy conditions. Furthermore, it has to meet the interface condition in (54). (54) Scope Filter a. A head assigned the feature [w] must c-command a head assigned the feature [n]. b. A head assigned the feature [i] must c-command a head assigned the feature [n] and be c-commanded by a head assigned the feature [w].

For the sake of better illustration, I will discuss the derivation of the sentences in (55). First, the arguments are merged in their thematic position in the VP. Then the Case-checking heads are merged and the arguments move into their Case-licensing positions in a parallel fashion. (55) a. weil jeder mindestens eine Frau liebt since everyone-NOM at least one woman-ACC loves b. weil mindestens eine Frau jeder liebt since at least one woman-ACC everyone-NOM loves

In (55a), where the subject is to be read with wide scope with respect to the object, the scope features can be directly assigned and checked in the Case-positions, with AgrAcc being assigned and checking [n] and AgrNom being assigned and checking [w]. In (55b), where the object is to be read with wide scope with respect to the subject, only one scope feature can be assigned and checked in the Case-position, since the assignment of [w] to AgrAcc and [n] to AgrNom would violate the Scope Filter in (54). Thus, the assignment of [n] to AgrNom, which according to (52) is more economic than not using any Case-position for the checking of scope features, forces

SCRAMBLING

59

the assignment of the remaining scope feature [w] to a higher head. Since the direct object in (55) is nonspeciﬁc—if it were speciﬁc the scope features could be assigned to the head that licenses weak pronouns (cf. section 2.2)—and since the object does not have any other features to check in the I-domain, the scope feature is assigned to a bare functional head, which is merged with the existing structure. Then, the direct object moves across the subject to check its scope feature. Finally, the complementizer is merged to complete the derivation of the clause. The assignment of a scope feature to a functional head deﬁnes the scope position of an argument. Since according to (48b) quantiﬁed DPs have to be spelled out in their scope positions, the copies of the two arguments can only be spelled out in the order given in (55), with the wide scope phrase preceding the narrow scope phrase. Along the same lines it follows that a quantiﬁed DP if it is to be read with narrow scope with respect to negation (cf. [44] and [45] earlier) must be spelled out below negation even if it has a speciﬁc (partitive) interpretation. A DP of this type will check its scope feature in its Case-position below negation, move across negation to check its [Speciﬁcity] feature, but be spelled out in its scope position below negation. The same considerations apply to contrastively focused deﬁnite and indeﬁnite speciﬁcs (cf. [38] and [39] earlier). I will now address the question of whether these scope features are to be considered as purely formal features. Remember that there are two diﬀerent dialects concerning the interpretation of quantiﬁed expressions. For speakers of the ﬁrst dialect scrambled structures are unambiguous. That is to say, the scrambled phrase is interpreted in its surface position. For speakers of the second dialect scrambled structures are ambiguous. That is to say, the scrambled phrase can be interpreted in its surface position or in its base position (I am leaving open the question whether it is the Thetaor the Case-position that is relevant here). That is to say that a sentence produced to express a certain meaning can be ambiguous, although the speaker may have just had one meaning in his mind. The scope features that I assumed here do not determine the semantic interpretation of scopal elements. Rather, they provide a syntactic limit to the interpretation of a scopal element. This property is more akin to the nature of a formal feature (that gears but does not [directly] determine the interpretation of the element it is assigned to). Let us call the position in which a formal scope feature is checked the scope position of a scopal element; then the interpretation of these elements in the two dialects can be characterized most simply as given in (56). (56)

Dialect 1: A scope-bearing element is interpreted in its scope position. Dialect 2: A scope-bearing element can be interpreted in its scope or in its base position.

Again, this characterization of the properties of scrambling in the two dialects highlights the fact that the scope features that I introduced are formal in nature rather than being purely semantic. Rather than being interpreted directly, they drive the derivation that provides the input for more general principles of interpretation. To summarize, if we want to refrain from positing an optional mechanism in the syntax like H&R’s mechanism of freely creating scrambling chains that can be

60

SCRAMBLING , REMNANT MOVEMENT , AND RESTRUCTURING IN WEST GERMANIC

exploited at the interfaces, we have to extend the computational system and allow for the introduction of non-lexical features in the course of the derivation. The enrichment mechanism I propose obeys cyclicity and is conservative in that it involves either the assignment of non-lexical (= relational) features to pre-existing structure or the introduction of a bare functional head with the feature in question that represents the smallest extension. 2.6.2

Optionality and the role of interface features

I will now brieﬂy address the question of whether I have achieved more than replacing one optional mechanism, that of chain formation in the system of H&R, with another optional mechanism, that one involving the assignment of scope features.7 At this point I must clarify a number of issues. Remember that in German if the speaker has in mind the meaning that for every woman there is a man who loves her, he must use a scrambling structure to express that meaning, as is illustrated again in (57). (57) a. weil ein Mann jede Frau liebt (* with the intended meaning) since a man-NOM every woman-ACC loves b. weil jede Frau ein Mann liebt (essential for the intended reading) since every woman-ACC a man-NOM loves

Crucially, I do not assume that the syntax will generate both options in (57) and that (57b) is then selected at the interface since it corresponds to the meaning the speaker had in mind. I assume that semantic scope features (according to the intentions of the speaker) can be assigned to arguments of a verb. In the syntax, these scope features need to be checked by corresponding formal features in the course of the derivation. If the scope relation corresponds to the hierarchy of the arguments (or adjuncts) involved, no scrambling is necessary. The feature assignment mechanism ensures that in this case scope features are checked in the Case-licensing positions of the arguments. Only if the scope relation does not correspond to the thematic hierarchy of arguments does scrambling apply. The economy condition on the mechanism of feature assignment ensures that only the smallest extension necessary is taken to arrive at the inverse scope representation. If we want to have a grammar without free adjunction, it seems to me this is the only way to handle the scope eﬀects of scrambling. However, there is a more principled issue behind the preceding question. H&R employ a speciﬁc perspective on grammar that is essentially GB-like: Diﬀerent syntactic structures are generated freely (move alpha), after which process the legitimate ones are quasi ﬁltered out by requirements of the modules. The diﬀerent ﬁlters in the account of H&R are then extended to include interface requirements. This perspective is widely spread and can be considered the standard method of handling so-called interface requirements. For instance, many accounts to focus interpretation assume that a formal feature [f ] can be assigned freely to syntactic constituents and that at the end of the derivation the representation that best corresponds to speciﬁc interface requirements is chosen. I have a diﬀerent perspective on grammar and assume that the interaction

SCRAMBLING

61

between syntax and the interfaces works diﬀerently. I assume that interface features can trigger syntactic operations. More speciﬁcally, I assume that if the speaker wants to focus a certain constituent, then a (semantic) focus feature is assigned to this constituent (in the numeration). This semantic feature must then be licensed by the speciﬁc means foreseen by the individual grammar, that is, either by a corresponding focus morpheme in the structure, by an intonational morpheme, or by movement to a certain position (an abstract morpheme). In a parallel fashion, if the speaker wants the direct object to be interpreted with wide scope with respect to the subject, then the respective scope features are assigned to these constituents in the numeration. And these scope features must then be licensed (checked) in the course of the derivation by corresponding formal features. As we have seen earlier, the assignment of formal scope features is heavily constrained by economy conditions rather than being a mechanism that may apply freely. Thus, we can conclude that optionality is not part of this account of scrambling, though the speaker is of course free to choose a passive sentence instead of the scrambling structure in (57) to achieve the same communicative eﬀect.

2.7 Addressing Vikner’s problem To remind ourselves, Vikner pointed at the following question: If scrambling in German and Dutch and object-shift in Scandinavian are all operations of A-movement, why is it that object movement in Dutch and Scandinavian is severely restricted but may permute arguments freely in German? The answer that I will give is that A-movement is subject to a minimality constraint that is circumvented by the Case properties of DP-arguments in German (cf. Haeberli 2002 for a diﬀerent solution). As a ﬁrst approximation to the minimality condition relevant here, consider (58). (58) Minimality a. Movement of a DP-argument across an overt categorically non-distinct argument is blocked. b. Case (that distinctively marks grammatical functions) can render DP-arguments categorically distinct.

Let us look at the case of a ditransitive verb with an indeﬁnite indirect object and a deﬁnite or speciﬁc direct object in a language like Dutch, in which DPs are non-distinct for lack of Case, as is illustrated in (59). In (59), the indirect object will not block the selection for agreement (or attraction) of the direct object by the head marked for [Speciﬁcity], since the latter is the closest category with the relevant feature. But it will block via minimality movement of the categorically non-distinct direct object. Note that the minimality condition is taken to only block category movement of the direct object. The agreement or attract relationship between the head marked for [Speciﬁcity] and the direct object is undisrupted by the minimality condition. To evade the problem that certain features seem to trigger movement only optionally, I

62

SCRAMBLING , REMNANT MOVEMENT , AND RESTRUCTURING IN WEST GERMANIC

assume that the [Speciﬁcity] feature is checked by either category movement or AGREE. Thus, we see that in order to solve the problem of optionality, it is important that minimality is not reduced to (or replaced by) the Minimal Link Condition (MLC). The two conditions are independent restrictions that have to be kept separate. More speciﬁcally, a minimality eﬀect only blocks (syntactic) movement of a category but will not disrupt an agreement relation that obeys the MLC. (59)

[[X [+spec] IO [–spec] DO [+spec] V]]

There is an alternative interpretation of the case illustrated in (59) that would allow one to conﬂate the MLC and minimality. The alternative account would assume that the DO in fact has moved across the IO to check its speciﬁcity feature but must be spelled out in its lower position due to an interface constraint that demands that A-movement must not obscure grammatical functions. This principle would state that in cases of A-movement copies must be spelled out in a way such that grammatical functions remain visible (in the sense that they can be read oﬀ directly from either word order or Case distinctions). Since it involves actual movement of a syntactic category, this alternative approach predicts that there may be syntactic eﬀects of scrambling even in superﬁcially non-scrambled word orders. In the absence of any evidence pro or contra to such an eﬀect, I will put this alternative to the side. Let us now see what happens in the same type of situation in German. Again, the direct object will be selected for Agreement by the attracting head. In this case, minimality will not block category movement of the direct object across the indirect object, since their Case morphology renders the two arguments categorically distinct. Addressing the qualiﬁcation in (58b), I may have to assume that only a certain type of Case can fulﬁll the function of rendering DPs categorically distinct. In Icelandic, Case can obviously not help circumvent the minimality condition on A-movement: Object-shift may not permute arguments in Icelandic. Either we adopt Holmberg’s (1999) account of the restrictions on object-shift in terms of a phonological minimality condition (any phonological material other than adjuncts blocks object-shift in his system) or we assume that only Case that distinctively marks grammatical functions in a language can render DPs categorically distinct. Note that in Icelandic, Case distinctions, though numerous and robust, do not have this property. Icelandic does have Dative subjects as well as Nominative objects. In German, a DP assigned Nominative, even if it is a deep object, always represents the subject of the clause and the subject of the clause can be no other phrase (except for clauses) than the DP marked Nominative. Is there any evidence for the assumption that Case can render DPs categorically distinct? First, note that also in German scrambling across a categorically non-distinct DP is illicit. In constructions where two arguments are marked with the same Case, as in ECM-constructions, scrambling is impossible. In (60), a case of restructuring, all three arguments occur in the matrix clause, as can be deduced from the fact that they all precede negation that can have scope over the matrix verb. Nevertheless, scrambling of the embedded object across the embedded subject is impossible. Example (60) cannot mean that ‘she saw Maria kiss him’.

SCRAMBLING

(60)

63

weil sie ihn die Maria nicht küssen sah since she-NOM him-ACC the Maria-ACC not kiss saw ‘since she did not see him kiss Maria’

Second, in Dutch, as I have noted earlier, the permutation of full DP-arguments is illicit. Dutch DPs, like English DPs, have lost their Case distinctions. However, pronouns that have preserved their Case distinctions enjoy a much freer distribution: A direct object pronoun can move across an indirect DP-object. In West Flemish, pronouns can even move across the subject and can appear in various orders, similar to the free distribution of DP-arguments in German (cf. Haegeman [1996b]).

2.8 Conclusions To summarize, in this chapter I have provided an account of scrambling in terms of feature-driven movement. Investigating the distribution of arguments with respect to the negative marker, I have shown that scrambling is not optional, as is claimed by H&R, but subject to speciﬁc conditions. Instead, I proposed an account of scrambling in which movement obtains obligatorily but the Spell-out of copies is subject to speciﬁc interface conditions. These conditions require, for instance, that a speciﬁc DP that is focused or receives narrow scope cannot be spelled out in its scrambling position. I have argued that there are two triggers for scrambling, namely [Speciﬁcity] and [Scope], and proposed an account in which these features are checked in a Spechead conﬁguration that renders superﬂuous the operation of adjunction in accounting for alternating word orders. Alternatively, I have proposed a mechanism of assigning features in the course of the derivation that is constrained by the standard economy conditions.

64

SCRAMBLING , REMNANT MOVEMENT , AND RESTRUCTURING IN WEST GERMANIC

3

Coherent Infinitives in Dutch and West Flemish

T

his chapter provides an initial discussion of the properties of coherent inﬁnitives in Dutch and West Flemish. On the basis of the Dutch data, I ﬁrst describe the traditional OV-based biclausal approach to restructuring inﬁnitives in terms of VR and LDS (cf. Den Besten and Rutten [1989], Rutten [1991]). Then I discuss the distribution of verb-particles in Dutch verb clusters and argue that some occurrences cannot be derived by incorporation in terms of head movement but must involve XPmovement. This observation will lead me to reject the assumption that VR is a process that only involves head movement. On the basis of West Flemish data, I will ﬁrst introduce the standard approach to Verb Projection Raising constructions in terms of extraposition of the (extended) inﬁnitival VP (VPR). Then I will investigate the distribution of IPP-complements in WF and argue that VPR cannot be given a satisfactory account in terms of “extraposition,” that is, in terms of rightward movement of an extended verb projection. At the end, I will conclude that both VR and VPR involve leftward movement (of an extended projection) of the verb, which will set the basis for a VO-based analysis of these constructions to be given in chapter 4.

3.1 Coherent infinitives in Dutch I will start the discussion of coherent constructions in the individual languages with Dutch. Since the pioneering work of Evers (1975), there has been a vivid tradition within Dutch linguistics of ongoing research on the topic, such that this phenom64

COHERENT INFINITIVES IN DUTCH AND WEST FLEMISH

65

enon in Dutch is very well described, which does, however, not imply that it is very well understood as well. The following discussion of the data is based on the work by Den Besten and Rutten (1989) and Rutten (1991). Inﬁnitival complements are divided into the three groups given in (1). Various main verbs can select om + te-inﬁnitivals as a (prepositional) object. The element om is generally taken to be the inﬁnitival complementizer. Thus, an example like (2a) is traditionally analyzed as displayed in (2b), with the inﬁnitival clause being extraposed. (1)

a. om + te-inﬁnitivals: Extraposition b. bare inﬁnitivals: Verb Raising (VR) c. te-inﬁnitivals: VR or extraposition (classiﬁcation following Evers [1975])

(2)

a. dat Jan besloot om een liedje te zingen that Jan decided a song to sing b. dat Jan tEXTR besloot [CP om [IP PRO een liedje te zingen]]EXTR

3.1.1

Bare inﬁnitives

Bare inﬁnitives are characterized by the lack of the inﬁnitival marker te (‘to’). According to Evers (1975), verbs that take bare inﬁnitives fall into three groups. The ﬁrst group consists of the Exceptional Case Marking (ECM)–verbs that license a lexical subject in the bare inﬁnitival. This group includes the verba sentiendi, the causative/permissive verb laten (‘to make/let’), and vinden (‘to consider’). The second group consists of root modals and some isolated cases like leren (‘to learn/teach’) and helpen (‘to help’), which are analyzed as Control verbs. The last group consists of epistemic modals and a subset of (semi-) aspectual verbs like the inchoative verb gaan (‘to go’), which are properly characterized as Subject Raising verbs. Example (3a) shows a bare inﬁnitival selected by the ECM-verb horen (‘to hear’). Here the matrix verb intervenes between the embedded inﬁnitival and its arguments. To accommodate this order with the head-ﬁnal character that he assumed for Dutch, Evers (1975) proposed a rule of VR that right-adjoins the dependent inﬁnitive to its selecting verb as depicted in (3b). (3)

a. dat ik Jan een liedje hoor zingen that I Jan a song hear sing ‘that I hear Jan sing a song’ b. dat ik [Jan een liedje tR ] hoor zingenR

While om+te-inﬁnitives are unequivocally to be described as non-coherent inﬁnitive constructions, bare inﬁnitives in Dutch display all the properties of a coherent construction that I showed to be characteristic in chapter 1. It should suﬃce for the sake of illustration to discuss just three of the properties that are characteristic of a coherent construction. Example (4a) shows that a bare inﬁnitival cannot be

66

SCRAMBLING , REMNANT MOVEMENT , AND RESTRUCTURING IN WEST GERMANIC

extraposed. Example (4b) shows that a bare inﬁnitival is transparent for certain extraction processes. In (4b), the embedded direct object has been scrambled into the middle ﬁeld of the matrix verb.1 Finally, (4c) shows that the negative operator niet can take scope within the embedded clause or within the matrix clause. Example (4c) can mean either ‘we saw that the crows were not ﬂying’ or ‘we didn’t see that crows were ﬂying’. (4)

a. *dat ik tEXTR hoorde [Jan een liedje zingen]EXTR that I heard Jan a song sing b. dat wij heti [Cecilia ti tR] hoorden vertellenR that we it Cecilia heard tell ‘that we heard Cecilia tell it’ c. dat wij de kraaien niet tR zagen vliegenR that we the crows not saw ﬂy ‘that we didn’t see the crows ﬂy’

Following is a list of those verbs that select bare inﬁnitives in Dutch. This list is adapted from Rutten (1991) and shall function as a reference list for comparing bare inﬁnitives in diﬀerent languages.2 (5)

blijven doen durven*3 gaan hebben hoeven* horen komen kunnen moeten mogen

stay, remain do, make dare go, will have need to hear come can must may

laten liggen* lopen staan* vinden voelen willen zien zijn zitten* zullen

let lie (down) walk stand ﬁnd feel want see be sit will

Before I discuss the behavior of te-inﬁnitives, I would like to address two phenomena that Rutten (1991) describes as exceptions to VR. Example (6a–b) illustrates that not only is it possible to have the order modal < inﬁnitive (as is expected if VR is obligatory) but the order inﬁnitive < modal also yields a grammatical sentence. (6)

a. dat Anna hem een CD wil geven b. dat Anna hem een CD geven wil that Anna him a CD wants give/give wants c. *Karel zei morgen niet komen te kunnen Karel said tomorrow not come to can d. Karel zei morgen niet tR te kunnen komenR ‘Karel said that he is not able to come tomorrow’

COHERENT INFINITIVES IN DUTCH AND WEST FLEMISH

67

Rutten gives the following descriptive generalization: VR is optional if only one inﬁnitive (without te) is embedded under a ﬁnite modal verb. Rutten notes that when the modal verb itself is nonﬁnite, VR becomes obligatory again (6c–d). Furthermore, he notes that when more than one inﬁnitive is embedded under a ﬁnite modal, VR must apply to all inﬁnitives (7). (7)

a. *dat Karel morgen wel niet komen durven zal that Kares tomorrow well not come dare will b. *dat Karel morgen wel niet komen tR zal durvenR c. dat Karel morgen wel niet tR tr zal [durven komenR]r d. *dat Karel morgen wel niet tR durven komenR zal e. *dat Karel morgen wel niet tR durven zal komenR ‘that Karel probably won’t dare to come tomorrow’

In (7a), none of the two inﬁnitives has been raised and the structure is ungrammatical. Example (7b) shows that it is not suﬃcient to only raise the higher inﬁnitive; since durven is itself a VR-verb, the lowest inﬁnitive has to be raised as well (7c). The ungrammaticality of (7b) can be made to follow from Rutten’s generalization. After all, the verb that embeds the inﬁnitive komen that failed to undergo VR is not a ﬁnite modal. What is more surprising given the preceding generalization is that (7d) is ungrammatical as well. Since the verb that selects it is nonﬁnite, komen has to undergo VR and we derive the sequence [durven komen] zal. Since the verb that selects durven is a ﬁnite modal verb, it is not clear why VR is obligatory in this conﬁguration as well. Obligatory VR in this case would follow from Rutten’s generalization if the complex verb durven komen, derived by head-to-head adjunction, does not count as a single unit.4 Finally, (7e) is ungrammatical presumably because the most deeply embedded inﬁnitive, komen, has been moved across two intervening heads in violation of the Head Movement Constraint. These observations can be summarized in the following generalization: Verbs in a verbal complex that comprises more than two verbs can only appear in the strict order V1 V2 V3 . . . Vn in Dutch, where a higher number indicates a deeper level of embedding. There is one exception to this rule: Participles that represent the most deeply embedded verb in a verbal sequence may precede the verbal complex properly ordered so (see discussion and examples [8–10] later). Participles seem to be exempted from obligatory VR independently of the context they are part of: A participle may fail to undergo VR when the selecting verb is ﬁnite (8) or nonﬁnite (9). A participle may stay in its base-generated position even when the selecting verb has to undergo VR (10). In (10), the inﬁnitival auxiliary worden (‘be’) selects a (passive) participle. The auxiliary, since it is not the only inﬁnitive embedded under the ﬁnite modal zal, has to undergo VR; nevertheless, the participle may fail to raise with its selecting verb. (8)

a. dat Anna hem een CD gegeven heeft b. dat Anna hem een CD heeft gegeven ‘that Anna has given him a CD’

68

SCRAMBLING , REMNANT MOVEMENT , AND RESTRUCTURING IN WEST GERMANIC

(9) a. Anna beweerde hem een CD gegeven te hebben Anna claimed him a CD given to have b. Anna beweerde hem een CD te hebben gegeven ‘Anna claimed to have given him a CD’ (10) a. *dat hij tot voorzitter gekozen worden kunnen zal that he (to) chairperson chosen become can will b. dat hij tot voorzitter gekozen zal kunnen worden c. dat hij tot voorzitter zal kunnen worden gekozen ‘that he will be able to be elected chairperson’

Rutten assumes that VR is the result of a formal licensing requirement (he assumes that verbs need “verbal Case” that is assigned to the right in Dutch) and that this requirement that holds for verbs in general does not apply to participles, since they are not real verbs but may be considered as having the category label [+V]. Now it is interesting to see what word orders are possible in a sentence that combines a participle, a ﬁnite modal, and a single inﬁnitive. We expect that both the participle and the inﬁnitive can precede the ﬁnal verb. However, this expectation is not borne out. Rutten notes that (11a) is very odd and that at least the inﬁnitive should be raised (11b). Raising of the participle is possible as well (11c). (11) a. *dat Anna hem wel een CD gegeven hebben kan that Anna him well a CD given have can b. dat Anna hem wel een CD gegeven kan hebben c. dat Anna hem wel een CD kan hebben gegeven ‘that Anna may very well have given him a CD’

It seems that Rutten’s rule for the distribution of inverted structures, that is, structures inverted with respect to the expected VR-order (cf. note 4), is too general. Also, it is not clear why a PF-rule like inversion should be restricted to the small class of modal verbs. In fact, Broekhuis and Den Besten (1989) point out that inversion is not restricted to modal verbs but possible with all verbs that select a bare inﬁnitive as long as the prosodic ﬁlter in (12) is observed. (12)

Prosodic Filter The embedded verb can only occur in inverted position if it carries primary stress.

Broekhuis and Den Besten cite contrasts such as the one given in (13) as evidence for the relevance of their prosodic account of inversion. They argue that since a nonspeciﬁc indeﬁnite object NP always carries primary stress, inversion is impossible if such an NP is present. They also observe that it is a characteristic feature of clauses that contain a small clause predicate that primary stress is contained within the small clause predicate, whereas the verb is assigned a secondary stress. They thus

COHERENT INFINITIVES IN DUTCH AND WEST FLEMISH

69

predict that verbs that select small clauses may not invert. This prediction seems to be borne out, as example (14) illustrates. (13) a. dat Karel het boek lezen wil that Karel the book read wants b. ?*dat Karel waarschijnlijk een boek lezen wil that Karel probably a book read wants (14) a. ?*dat Jan ziek zijn kan that Jan ill be may b. ?*dat ik hem boos worden zag that I him angry become saw c. ?*dat hij niet altijd soldaat blijven zal that he not always soldier remain will d. ?*dat hij aardig blijken zal that he kind appear will

The facts that Broekhuis and Den Besten (1989) point out are very interesting, but I will not pursue this matter any further here (cf. Koopman and Szabolcsi [2000] for further discussion of this issue). These facts point to an interesting interaction between syntax and prosody that deserves further investigation but is beyond the scope of this book. 3.1.2

Te-inﬁnitives and the Third Construction

Now let us look at te-inﬁnitivals. As I have already pointed out in the beginning of the previous section, te-inﬁnitivals can appear in both VR- and extraposition structures. There is a small class of verbs that select te-inﬁnitivals and only allow VR; that is to say, they do not allow extraposition. This class consists of Subject Raising verbs, semi-modals, and aspectual verbs and is given in (15). The vast majority of the verbs that select te-inﬁnitivals seem to allow VR and extraposition. However, a closer inspection of these verbs shows that there is only a small group of verbs that allow both VR and extraposition, the rest of these verbs are only compatible with extraposition, and what appear to be cases of VR are really instances of Remnant Extraposition, a type of construction that is also called the Third Construction (the ﬁrst being extraposition proper and the second being VR). (15)

blijken te dienen te hebben te hoeven te komen te lijken te liggen te

appear be obliged have to need to come to seem, appear to lie and

lopen te plegen te schijnen te staan te weten te zien te zitten te

walk and be used to seem to stand and know to see to it that sit and

70

SCRAMBLING , REMNANT MOVEMENT , AND RESTRUCTURING IN WEST GERMANIC

Evers (1975) noted that some verbs like proberen (‘try’) and beloven (‘promise’) can enter into both extraposition-type and VR-type constructions and thus called them optional verb-raisers. (16) a. dat Jan (het boek) probeerde (het boek) te lezen b. dat Jan (het boek) beloofde (het boek) te lezen that Jan (the book) tried/promised (the book) to read

It can be shown, however, that proberen and beloven are exponents of two classes of verbs that have quite diﬀerent properties. Proberen is an exponent of a small class of verbs that are optional verb-raisers in the sense of Evers. Beloven is an exponent of a large class of subject- and object-control verbs that only allow for the Third Construction. The Third Construction and VR-structures diﬀer with respect to the freedom of word order in the verbal sequence and with respect to the IPP-eﬀect. We have seen before that the order of verbs in a VR-structure is very strict. So, for instance, the order Auxiliary + VR-verb + Inﬁnitive may not be changed (17a–b). In the Third Construction, however, not only the order Aux + main verb + te-inﬁnitive but also the order main verb + Aux + te-inﬁnitive is possible. Besluiten (‘decide’) is a verb that belongs to the beloven-class (17c–d). (17) a. dat Marie geen boek heeft hoeven te lezen that Marie no book has need-IPP to read ‘that Marie did not need to read a book’ b. *dat Marie geen boek hoeven heeft te lezen c. dat Marie dat boek heeft besloten te lezen that Marie the book has decided to read ‘that Marie has decided to read the book’ d. dat Marie dat boek besloten heeft te lezen

The IPP-eﬀect also distinguishes VR-structures from the Third Construction. I have noted before that in compound tenses the main verb cannot take its participial form when it is embedded in a verb cluster of more than two verbs (cf. [18]). The verbs besluiten (‘decide’) and beloven (‘promise’), as opposed to proberen (‘try’), however, do not display the IPP-eﬀect in the relevant environment. (18) a. *dat Marie een boek heeft gewild lezen that Marie a book has wanted read b. dat Marie een boek heeft willen lezen that Marie a boek heeft want-IPP read (19) a. *dat Marie een boek heeft besluiten te lezen that Marie a book has decide-IPP to read

COHERENT INFINITIVES IN DUTCH AND WEST FLEMISH

71

b. *dat Marie een boek heeft beloven te lezen that Marie a book has promise-IPP to read c. dat Marie een boek heeft proberen te lezen that Marie a boek has try-IPP to read d. dat Marie een boek heeft geprobeerd te lezen that Marie a book has tried to read

It should be made explicit that Den Besten and Rutten (1989), though they do not oﬀer an explanation of the IPP-eﬀect, use it as a diagnostic for VR and I think correctly so. Before their seminal work on the Third Construction, the IPP-eﬀect was a mysterious phenomenon that appeared in the verb clusters of some verbs but not in those of others. Even worse, as the example of proberen shows (19c–d), one and the same verb may or may not display the IPP-eﬀect in what appears to be the very same structure. In their analysis, den Besten and Rutten suggest that the Third Construction involves extraposition plus scrambling (see also Broekhuis et al. [1995]). Example (19d), for instance, can be derived by extraposing the inﬁnitival complement and extracting the DP een boek from the extraposed clause and adjoining it to a projection of the matrix clause, that is, by long-distance scrambling. The resulting structure has been (later) called Remnant Extraposition, since the extraposed part consists of those elements that remain after scrambling. However, it must be noted that Den Besten and Rutten (1989) do not mean by this name that scrambling and extraposition are intrinsically ordered with respect to each other. Given this account and what has been said before about VR, the similar-looking sentences in (19c–d) have the following quite divergent structural analysis. (20) a. VR

dat Jan [PRO een boek tR tr] heeft [proberen [te lezen]R]r

b. R-EXT dat Jan een boekSCR tEXT heeft geprobeerd [PRO tSCR te lezen]EXT

Given this analysis, we may now assume that VR actually triggers the IPP-eﬀect or, in other words, that the IPP-eﬀect is a side eﬀect of the formation of a single verbal complex (by head movement). The analysis of the Third Construction as involving scrambling makes the prediction that elements like parts of idioms (21), small clause predicates (22), or verb-particles (23) that resist scrambling will necessarily be part of the extraposed clause. As the following examples show, this prediction is borne out. These examples should be compared with cases of VR where these elements may and in some cases must (24b) occur before the verbal complex (24). (21) a. dat Jan de prijs heeft besloten [in ontvangst te nemen] that Jan the award has decided in acceptance to take ‘that Jan has decided to accept the award’ b. *dat Jan de prijs in ontvangst heeft besloten [te nemen] (22) a. dat Jan het hekje heeft besloten [lekker groen te schilderen] that Jan the gate has decided nicely green to paint b. *dat Jan het hekje lekker groen heeft besloten [te schilderen]

72

SCRAMBLING , REMNANT MOVEMENT , AND RESTRUCTURING IN WEST GERMANIC

(23) a. dat Jan besloot [op te geven] that Jan decided up to give b. *dat Jan op besloot [te geven] (24) a. dat Jan de prijs in ontvangst is komen nemen that Jan the award in acceptance is come-IPP take ‘that Jan has come to accept the award’ b. dat Jan het hekje lekker groen had willen schilderen that Jan the gate nicely green has want-IPP paint ‘that Jan has wanted to paint the gate nicely green’ c. dat Jan op moest geven that Jan up must.Past give ‘that Jan had to give up’

So the IPP test and the test with respect to scrambling can be combined to yield a reliable criterion for determining whether a given verb is a verb-raiser or not. Rutten reports that the majority of the subject and object control verbs do not allow VR. Rutten points out that there seems to be a relation of mutual exclusion between the IPP-eﬀect and te, indicating that if te is present IPP is excluded. However, there are a number of exceptions to this generalization. The verbs of the proberen-class can take the IPP and te, although te can be omitted in many cases (25). The verbs that allow for the Third Construction but do not allow for VR, the beloven-class, are listed in (26). (25) Verb-raisers (the proberen-class) beginnen te begin menen te dreigen te threaten proberen te durven (te) dare trachten te helpen (te) help wagen (te) leren (te) teach, learn weigeren te

think try try dare refuse

(26) Third Construction Verbs (the beloven-class) (all take te) aanraden advise vergeten forget begeren desire verlangen long for beloven promise verleren forget how to do beogen aim at vermijden avoid besluiten decide verplichten oblige beweren claim, say vertellen tell denken think verwachten expect dwingen force verzuimen neglect eisen demand voorstellen propose geloven believe vrezen fear hopen hope wensen wish opdragen order zeggen say verbieden forbid

COHERENT INFINITIVES IN DUTCH AND WEST FLEMISH

3.1.3

73

Particles and verb raising

In this section, I will argue that the complex distribution of particles in the Dutch verb cluster cannot be properly explained in an account that assumes that verb clusters are derived by head movement. I will demonstrate that the central mechanism of this account, namely, long head movement, forestalls an account of why particle verbs and participles block VR. I will present data that show that long head movement should not be permissible in the grammar anyway, leading me to the conclusion that the distribution of particles can only be explained in terms of XP-movement. This in turn will lead me to the conclusion that verb clusters must provide XP-landing positions for particles, indicating that verb clusters cannot simply be head-adjunction structures. We have seen in the previous section that a particle may precede the verb cluster created by VR (24c). As (27a) shows, a particle may also move along with its selecting verb and become part of the verb cluster. That this property is not restricted to particles is shown in (27b). The head of a small clause predicate may also become a member of a verb cluster created by VR. However, if the small clause predicate is modiﬁed, the resulting complex cannot be part of a verb cluster (cf. [27c] and [24b]). This suggests that only elements that can incorporate into the selecting verb, that is, heads, can undergo VR to become a member of a verb cluster. (27) a. dat Jan de wedstrijd moest op geven that Jan the competition must.Past up give ‘that Jan had to give up the competion’ b. dat Jan het hekje had willen groen schilderen that Jan the gate had want-IPP green paint ‘that Jan had wanted to paint the gate green’ c. *dat Jan het hekje had willen lekker groen schilderen that Jan the gate had want-IPP nicely green paint

The discussion about the underlying position of particles in Dutch goes back to Koster (1975) and van Riemsdijk (1978). Koster argues that a verb and a particle constitute a compound verb. If the verb and the particle are separated, this is due to the eﬀect of Verb Movement (V2). I am following van Riemsdijk, who assumes that the particle is the head of a particle phrase, a PP in fact, in the complement domain of the verb (28a). The fact that the verb and the particle often act as a unit is accounted for in this approach by adopting a rule of Particle Incorporation (PI) that moves the particle to the verb. If PI is optional, then the particle may stay in its PP and will be stranded by VR yielding sentences like (28b). Or the particle may incorporate into the verb and subsequently undergo VR yielding (28c).5 (28) a. dat Jan [PP [P op]] geven moest

base structure

b. dat Jan [PP op] tR moest gevenR

VR

c. dat Jan [PP tPI] tR moest [opPI geven]R

PI + VR

74

SCRAMBLING , REMNANT MOVEMENT , AND RESTRUCTURING IN WEST GERMANIC

Bennis (1992) points out that while everybody who has ever discussed the particle construction in Dutch has noticed that in the case of VR the particle may either be left behind or go along with the verb, it has been overlooked that the particle may appear at several positions within the verb cluster, as is shown in (29). (29)

dat hij mij (weg) zou (weg) kunnen (weg) horen (weg) rijden that he me (away) would (away) can (away) hear (away) drive ‘that he would be able to hear me drive away’

Koopman (1995) notes that in her dialect the particle can only occur in the positions illustrated in (30): immediately preceding the selecting verb (30a), immediately preceding the verbal cluster (30b), and in a position immediately following the ﬁnite verb (30c). Koopman suggests that sentences like (30c) can be derived from sentences like (30b) by assuming optional movement of the ﬁnite verb to some higher functional head position. (30) a. dat ik Piet de doktor heb willen laten op bellen that I Piet the doctor have want -IPP let up call ‘that I wanted to let Peter call up the doctor’ b. dat ik Piet de doktor op heb willen laten bellen c. dat ik Piet de doktor heb op willen laten bellen d. *dat ik Piet de doktor heb willen op laten bellen

Also, Evers (1994) notes that particles can occur in more than one position in the verb cluster and provides the example given in (31a). Evers asserts that “the Dutch particle may take any of the positions marked above with the (!) sign, and in any of these positions it attracts the main stress.” (31)

Die lamp! Als Jasmijn hem op (!) had (!) willen (!) laten (!) poetsen, dan zou de geest verschenen zijn that lamp! If Jasmin him up had want let polish then would the ghost appeared be ‘That lamp! If Jasmin had been willing to let it be polished up, the ghost would have appeared’

Given the agreement between Bennis and Evers, I will assume that Koopman’s (30d) is grammatical in some dialect and is thus in need of explanation. Note that Koopman’s strategy of deriving (30c) from (30b) is unlikely to be applicable to (30d). In order to derive (30d) from (30b) two verbs would have to undergo leftward head movement to two diﬀerent head positions. While there may be some motivation for moving the ﬁnite auxiliary, it is hard to see why the inﬁnitive willen in (30d) should move after it has undergone VR and has been adjoined to its selecting verb and thereby presumably been formally licensed. It is clear that not all positions that the particle can occupy in (31) can be derived by PI or the lack thereof and by subsequent optional movement of the auxil-

COHERENT INFINITIVES IN DUTCH AND WEST FLEMISH

75

iary. More speciﬁcally, the order had willen op laten poetsen cannot be derived in any way discussed so far. The solution seems to be that a particle may undergo movement itself. Bennis (1992) proposes that particles may undergo long head movement. To derive the sequence in (32a) from the underlying structure in (32b), Bennis proposes that ﬁrst bellen undergoes VR and right-adjoins to the selecting verb willen. (32) a. dat ik de doktor heb op willen bellen that I the doctor have up want call b. dat ik de doktor op bellen willen heb (assumed base structure) that I the doctor up call want have

Then the particle is left-adjoined to this newly formed complex head by long head movement. Long head movement moves the particle across the trace of the selecting verb to the complex verb [willen bellen]. Normally the intervening head position would count as a minimality barrier for government of the trace of the particle (traces of heads must be antecedent governed and the intervening head would count as a closer governor and would by minimality act as a barrier for antecedent government of the trace). But, so Bennis argues, since the intervening head has incorporated into the head that the particle moves to, this position loses its status as a barrier. Long-distance head movement allows us to generate all the occurrences of the particle in (31) by employing either only VR or VR plus (long-distance) PI at each cycle in the derivation. For instance, (30c) can be derived by raising bellen to laten and [laten bellen] to willen before PI left-adjoins the particle to the complex head [willen laten bellen]. Evers reports the observation of Den Besten and Rutten (1989) that “the particle may not always move around in the row of the verbs” (Evers [1994], 11). Sometimes it has to be adjacent to the verb it is selected by. This is especially the case if the immediately preceding verb is itself a particle verb or if the immediately preceding verb is in past participle form. The following example (borrowed from Evers [1994]) contrasts the aspectuals [begin]V and [[aan]P vangen]V (‘to start’). (33) a. toen Jasmijn de lamp wou beginnen op te poetsen when Jasmijn the lamp wanted begin up to polish ‘when Jasmijn wanted to begin to polish up the lamp’ b. toen Jasmijn de lamp wou op beginnen te poetsen c. toen Jasmijn de lamp op wou beginnen te poetsen d. toen Jasmijn de lamp wou aan vangen op te poetsen when Jasmijn the lamp wanted prt start up to polish ‘when Jasmijn wanted to start to polish up the lamp’ e. *toen Jasmijn de lamp wou aan op vangen te poetsen f. *toen Jasmijn de lamp aan wou op vangen te poetsen g. *toen Jasmijn de lamp op wou aan vangen te poetsen

76

SCRAMBLING , REMNANT MOVEMENT , AND RESTRUCTURING IN WEST GERMANIC

In the variants in which start is spelled out with a particle verb, the lower particle has to stay close to its selecting verb. One may account for this by assuming that the higher particle blocks the movement of the lower one or by assuming that the higher particle blocks VR of the embedded verb and so prevents the extension of the movement domain for the lower particle. I assume that the second account is the correct one. If the higher particle blocks VR, then we know from section 3.1.2 that we must be dealing with a Third Construction in (33d–g). Since particles cannot be scrambled they have to stay close to their selecting verb. We also know from 3.1.2 that beginnen is an optional verb-raiser; this explains the contrast between (33a–c) on the one hand and (33d–g) on the other hand. What is missing in Den Besten and Rutten’s (1991) account of the Third Construction is an explanation for the fact that an optional verb-raiser when it appears in the past participle form like in (34) is only compatible with the Third Construction. (34) a. toen Jasmijn de lamp was begonnen op te poetsen when Jasmin the lamp was begun up to polish ‘when Jasmin had begun to polish up the lamp’ b. *toen Jasmijn de lamp op was begonnen te poetsen

In (34), the participle does not seem to allow a VR-construction; otherwise movement of the particle should be okay (compare [34b] and [33c]). The crucial diﬀerence in (34a–b) lies in the position of the particle op. Example (34a), where the particle stays close to its selecting head, can be analyzed as a case of Remnant Extraposition with the direct object de lamp having been scrambled into the matrix clause. Example (34b), however, cannot be analyzed as a case of Remnant Extraposition, since the particle occupies a position that precedes the verbal cluster, a position it cannot have reached by scrambling. Thus, (34b) has to be analyzed as a case of VR. From its ungrammaticality I conclude that participles block VR. Den Besten and Rutten (1989) simply stipulate that particle verbs and past participles cannot restructure their complements, meaning that they do not permit the head of their complement to undergo VR. Then, the question arises of what property particle verbs and past participles have in common that causes them to block VR. Particle verbs and past participles form a very disparate class. It is clear that this property cannot have anything to do with the peculiar properties of past participles, the tenets of their temporal interpretation, or the peculiarities of particle verbs, their compound semantics for instance. The answer must be simple and syntactic. What they have in common is the particle. In the West Germanic languages, with the exception of Low German and Frisian, verbs form their past participles by preﬁxing the stem with the particle ge and suﬃxing the stem with an inﬂectional ending. I would like to propose that verb-particles and the ge-preﬁx block VR in a similar manner.6 I will ﬁrst outline an account in terms of head movement, which will lead us to the conclusion that long head movement should be excluded from the grammar. This will leave us with the problem that many occurrences of verb-particles cannot be derived in the standard theory. I will later development an XP-movement account,

COHERENT INFINITIVES IN DUTCH AND WEST FLEMISH

77

in which it is derived that verb-particles and the ge-preﬁx block VR since they are inserted or licensed in the same syntactic position. I will brieﬂy sketch how this account works for particle verbs and will reserve for later the discussion of how this account can be applied to the ge-preﬁx in a way that derives the IPP-eﬀect. I propose that a particle verb and its dependent inﬁnitival enter into the underlying structure in (35). (35)

[VP [PP [InfP V2] P] V1]7

As earlier, I assume that the particle heads its own projection in the complement of the verb. The inﬁnitival itself (InfP) is a complement of the particle. In this conﬁguration, the particle intervenes between the selecting verb and the dependent inﬁnitive and will trivially block VR: Moving the dependent inﬁnitive across the particle violates the Head Movement Constraint. Note that we have to exclude the possibility that the dependent inﬁnitive ﬁrst incorporates into the particle and this complex head then incorporates into the higher verb. Within the minimalist framework that I have adopted it suﬃces to assume that the particle is not a licenser of the dependent inﬁnitive in order to rule out incorporation of the dependent inﬁnitive into the particle. Note, however, that if—in following Bennis (1992)—we allow for long head movement, then the dependent inﬁnitive can undergo VR in one swoop and adjoin to the selecting verb after the particle has itself incorporated into its selecting head. So if we allow for long head movement we have no way to derive the descriptive generalization that particle verbs disallow VR. That long head movement should be excluded in principle is also suggested by the following data. Den Dikken (1995) notes that particles block the incorporation of the heads of small clause predicates. As I have noted earlier, heads of small clause predicates can incorporate into their selecting verb, in which case they undergo VR together with their incorporator and become part of the verb cluster (36b). When the head of the small clause predicate fails to incorporate it will be found outside, that is, to the left of the entire verb cluster (36a). (36) a. dat ze de schuur rood hebben geschilderd that they the barn red have painted ‘that they have painted the barn red’ b. dat ze de schuur hebben rood geschilderd that they the barn have red painted

As the sentences in (37) show, incorporation of the head of the small clause predicate is impossible in the presence of a verb-particle. Den Dikken (1995) argues extensively and convincingly that sentences like (37) are best analyzed with the particle being the head of a small clause in the complement of the verb. The adjective rood in turn is the head of a small clause in the complement of the particle. Thus, the particle intervenes between the potential incorporee and the verb as incorporator. Again, if long head movement were available in the grammar of Dutch, it is not clear

78

SCRAMBLING , REMNANT MOVEMENT , AND RESTRUCTURING IN WEST GERMANIC

how to exclude (37d). While (37b) can be excluded in Bennis’s account, since the particle that precedes the verb cluster cannot be taken to have incorporated into the selecting verb and will thus block incorporation of the adjectival head, (37d), in contrast to (37c), should be grammatical since it can be analyzed as ﬁrst incorporating the particle to the selecting verb and then incorporating the lower adjectival head through head-adjunction to the resulting complex of particle and verb.8 (37) a. dat ze de schuur rood over hebben geschilderd that they the bard red prt have painted ‘that they painted the barn up red’ b. *dat ze de schuur over hebben rood geschilderd c. *dat ze de schuur hebben over rood geschilderd d. *dat ze de schuur hebben rood over geschilderd

Since I believe that the account sketched earlier of the fact that particle verbs do not allow for VR is correct—an account that receives independent support from the facts in (37)—I have to conclude that long head movement is not an option in grammar. Thus, I have to reject Bennis’s account of the distribution of particles in VRstructures and have to look for an alternative approach. The only remaining possibility is that particles undergo XP-movement (cf. Koopman [1995], who reaches the same conclusion on the basis of the assumption that Dutch is a head-initial language). We have seen that particles can undergo XPmovement when they receive some stress, as was the case with topicalization. As Evers pointed out, in (31) the particle carries stress independently of whether it occupies a high, medium, or very low position in the verb cluster (contrary to Cinque’s [1993] assumption that stress [mainly] follows from depth of embedding). Thus, it seems to be not entirely implausible that particles may undergo XP-movement within the verb cluster. If particles can undergo XP-movement in the verb cluster, then it follows that Dutch verb clusters cannot be head-adjunction structures. They must consist of maximal projections that can host landing sites for XP-moved particles. Furthermore, I will assume that particles undergo XP movement to be licensed in a Speciﬁer position of a functional head above VP (F1). I will also argue that it is the (extended) projection that contains the licensing position of the particle that is moved in cases of VR. In this account, the particle can then be assumed to move up within the verb cluster to [Spec,F1P] of the higher verb.9

3.2 Coherent infinitives in West Flemish In this section, I will discuss the syntax of Verb Projection Raising–constructions. In section 3.2.1, I introduce the standard account of this construction. I will then argue that the scope facts in VPR-constructions cannot be properly accounted for in terms of scrambling and extraposition. In section 3.2.2, I will discuss the distribution of inﬁnitives, participles, and IPP-complements and reach the conclusion that the distribu-

COHERENT INFINITIVES IN DUTCH AND WEST FLEMISH

79

tion of IPP-complements in West Flemish cannot be given a coherent explanation in terms of extraposition. Instead, I will demonstrate that all the facts discussed fall into place if one assumes that (1) complements are base generated to the right of the selecting head and (2) there is leftward head movement of the verb in West Germanic. West Flemish is a dialect of Dutch spoken in West Flanders, a province in Belgium. This dialect shows some striking dissimilarities to Standard Dutch. West Flemish allows for subject pro-drop and shows an elaborate system of Complementizer agreement with the subject, two phenomena that are probably related. See Haegeman (1992) for an extensive discussion of these phenomena. West Flemish has two types of coherent inﬁnitives: VR-constructions and Verb Projection Raising (VPR)–constructions. VPR also occurs in a number of German dialects, especially in Swiss German dialects. The following discussion of the data is drawn from the pioneering work of Liliane Haegeman on the structure of WF and on the syntax of VPR, especially Haegeman (1992, 1995a). 3.2.1

Verb Raising and Verb Projection Raising

VPR in West Flemish (WF) applies to those verbs that trigger VR in standard Dutch. In WF, not only the nonﬁnite verb embedded under a modal can be raised (38a), but also the verb plus its direct object (38b), or the verb plus the direct and indirect object, can be raised (38c). Den Besten and Edmondson (1983) have argued that these facts should be accounted for by assuming that in such a dialect not only V0 may be raised but also V’ or VP, whence the name Verb Projection Raising. (38) a. da Marie Jan nen boek tR wilt gevenR that Marie Jan a book wants give b. da Marie Jan tR wilt [nen boek geven]R that Marie Jan wants a book give c. da Marie tR wilt [Jan nen boek geven]R that Marie wants Jan a book give ‘that Marie wants to give Jan a book’

However, it is clear, as Haegeman (1995a) points out, that the inﬁnitival complement in (38b–c), which is loosely referred to as VP in the literature, is not to be equated with a bare VP-projection. VPR-constituents must contain the relevant functional projections to host clitics, adverbs, and sentential negation, as the examples in (39) taken from Haegeman (1995a) show. (39) a. da Valere wilt [ze morgen nen boek geven] that Valere wants her tomorrow a book give ‘that Valere wants to give her a book tomorrow’ b. da Valere durft [tegen niemand nie klapen] that Valere dares against no one not talk ‘that Valere dares to not talk to anyone’ c. da Valere tEXTR wilt [ze morgen nen boek geven]EXTR

80

SCRAMBLING , REMNANT MOVEMENT , AND RESTRUCTURING IN WEST GERMANIC

The VPR-complement in (39a) contains the clitic ze (‘her’). WF clitics occupy a position high in the clausal domain (Haegeman [1992]) analyzes them as being adjoined to IP). Example (39a) also contains the temporal adverb morgen (‘tomorrow’). Remember that adverbs may not occur in a verbal cluster created by VR. Similarly, the presence of negative constituents in (39b) suggests that VPR-complements contain the functional projection that is the locus of sentential negation and triggers the leftward movement typical of negative constituents (cf. Haegeman [1995b]). These data thus seem to indicate that VPR-structures are best described in the standard theory in terms of extraposition of the inﬁnitival complement, which is at least as big as an IP. A sentence like (39a) would thus be analyzed as illustrated in (39c). This is the analysis that Haegeman argues for in her book on West Flemish (1992). Note that this analysis is very similar to Den Besten and Rutten’s analysis of the Third Construction. Haegeman (1992) assumes that VPR equals scrambling plus extraposition. In (38c), the whole inﬁnitival complement has been extraposed, but since VPRcomplements are transparent, as I will show later, scrambling may aﬀect its constituents and move them into the matrix clause before extraposition applies.10 Like VR-structures, VPR-complements are transparent for several types of extraction processes. For instance, (40) shows that an object clitic that originates in the inﬁnitival complement can precede the subject of the higher clause,11 independently of whether we are dealing with a case of VR as in (40a) or with a case of VPR as in (40b). (40) a. da et Jan Valere deeg kuopen that it Jan Valere made buy b. da et Jan Valere deeg vu zen wuf kuopen that it Jan Valere made for his wife buy ‘that Jan made Valere buy it (for his wife)’

VPR-complements are also transparent for scope-taking elements but only if they are not aﬀected by VPR, that is, by extraposition in the analysis given earlier. Example (41a) is a case of VR, and (41b) and (41c) are cases of VPR. Both (41a) and (41b) are ambiguous between the readings given in (41d–e). Example (41c), however, where the adverbial is aﬀected by VPR, that is, where it is extraposed, has only the reading given in (41e). (41) a. da Jan Valere drie kiers da boek deeg lezen that Jan Valere three times that book made read b. da Jan Valere drie kiers deeg da boek lezen c. da Jan Valere deeg drie kiers da boek lezen d. ‘that on three occasions Jan made Valere read the book’ e. ‘that Jan made Valere read the book three times’

COHERENT INFINITIVES IN DUTCH AND WEST FLEMISH

81

Haegeman (1992, 112) states that a verb cluster created by VPR is opaque. The empirical generalization is the following: A scopal element that is outside of the verb cluster (i.e., that has undergone scrambling) can take its scope inside and outside of the verb cluster, whereas an element contained in the verb cluster may take its scope only within the verb cluster. Let us look at example (42). In (42a), the negative existential occurs outside of the verb cluster. Example (42a) is ambiguous between the reading Jan did not force Valere to read any book three times, with geenen boek taking wide scope, and the reading Jan forced Valere to not read any book three times, with geenen boek taking narrow scope. In (42b), the negative existential occurs inside of the verb cluster. What is crucial is that here the negative existential cannot take scope over the higher verb. Example (42b) is only compatible with the reading that Jan forced Valere to not read any book three times. (42) a. da Jan Valere geenen boek deeg drie keers lezen that Jan Valere no book made three times read b. da Jan Valere deeg drie keers geenen book lezen

This is almost a paradoxical situation. A VPR-complement must be transparent to permit the scrambling of the negative existential into the middle ﬁeld of the matrix clause in (42a) but must be opaque to prevent QR or scrambling of the negative existential in (42b). Haegeman (1992) proposes the following solution to the paradox: Scrambling applies before extraposition. When scrambling applies, the VPRcomplement is still in its base position, where it is L-marked by the governing verb and does not constitute a barrier for extraction processes. Hence the negative existential in (42a) may be scrambled into the matrix clause. After extraposition applied, the VPR-complement no longer occupies its base position; it has been adjoined to a maximal projection (VP or IP of the matrix verb). Haegeman proposes that in terms of Chomsky’s barrier framework, it could be argued that the extraposed complement is no longer L-marked by the matrix verb and will thus become a barrier for movement and government (Haegeman 1992, 203ﬀ). Haegeman (1992) takes the versatile nature of her account of the scope facts in VPR-structures as a strong argument in favor of her adjunction analysis. However, I will argue that her account has two shortcomings: a technical one and an empirical one. The technical problem with Haegeman’s account has been noted by Den Dikken (1994). Den Dikken points out that the trace left by an element extracted before extraposition applies still has to be licensed at LF. This means the trace is subject to the ECP and has to be antecedent governed. If the extraposed VPR-complement constituted a barrier, a trace created by scrambling-before-extraposition would not be able to fulﬁll the ECP.12 Maybe it is possible to overcome this technical problem by devising a theory that derives that for scrambling the trace in the base position of the inﬁnitival complement can be taken into account while for QR the trace in the extraposed position is relevant for meeting the ECP. However, there is also an empirical problem with Haegeman’s approach. First note that VPR, like VR but unlike Remnant Extraposition, gives rise to the IPP-eﬀect (43).

82

SCRAMBLING , REMNANT MOVEMENT , AND RESTRUCTURING IN WEST GERMANIC

(43) a. *da Jan ee gewild [Marie nen boek geven] that Jan has wanted Marie a book give b. da Jan ee willen [Marie nen boek geven] that Jan has want-IPP Marie a book give ‘that Jan has wanted to give Marie a book’

Second, note that Haegeman’s account to VPR does not diﬀer from Den Besten and Rutten’s (1989) account of Remnant Extraposition; both constructions involve scrambling plus extraposition. The diﬀerence between VPR and Remnant Extraposition with respect to the IPP-eﬀect must thus follow from an additional movement that applies to VPR-structures but is absent in the Third Construction. I, following Den Besten and Rutten, would like to link the IPP-eﬀect with VR. Under these assumptions, VPR would diﬀer from R-Extraposition by employing VR at LF. VR is a process of head movement that adjoins the dependent inﬁnitive to its selecting verb. Head movement is subject to the Head Movement Constraint, which, as Baker (1988) has shown, can be derived from the ECP. Within the barrier’s framework that Haegeman adopts in her account, the trace of a head-moved element must be antecedent governed. If, however, a VPR-complement that is extraposed constitutes a barrier, then the dependent inﬁnitive in (43b) cannot undergo head movement to adjoin to the matrix verb. This means that under Haegeman’s account of the scope facts in VPR-structures no unitary account of the IPP-eﬀect in VR and VPR is possible. It is conceivable that one could devise a theory that derives the eﬀect that extraposed VPR-complements are barriers for XP-movement at LF but not for head movement at LF, a notion of barrierhood that makes use of Rizzi’s (1990) concept of Relativized Minimality. However, there is evidence that an XP can be extracted out of an extraposed VPR-complement. Let us look at (44) (the example is taken from Haegeman [1995b]). Example (44a) involves an IPP-complement the properties of which I will discuss in more detail later. In the standard theory, (44a) must be derived from the base structure in (44b) by ﬁrst extraposing the most deeply embedded inﬁnitival clause to moeten and then extracting the CP da Jan siek was from the extraposed clause and adjoining it to the higher verb, the auxiliary oat in this case. Extraposing ﬁrst the CP to the auxiliary and then the clause that contained the CP to the lower verb would certainly violate cyclicity (or economy, in the minimalist framework).13 (44) a. da Valere moeten [aan Marie ti zeggen] oat [da Jan ziek was]i that Valere must to Maire say had that Jan sick was ‘that Valere had had to say to Marie that Jan sick was’ b. da Valere [aan Marie [CP da Jan ziek was] zeggen] moeten oat

I think that these arguments are suﬃcient to reject Haegeman’s (1992) account of the scope facts in VPR-structures. I think that these facts do not present an argument in favor of an adjunction analysis of VPR. As an alternative, I would like to consider the following two hypotheses, which make some interesting predictions. I assume that VPR-complements are transparent throughout the whole derivation. The

COHERENT INFINITIVES IN DUTCH AND WEST FLEMISH

83

ﬁrst hypothesis assumes that QR is not an option in a language that allows overt scrambling. We have seen for German in chapter 2 that a sentence like Some man loves every woman is not ambiguous if the arguments appear in the base order and that inverse scope is only possible if the object scrambles over the subject. However, it is less clear whether QR is also dispensable in a language like West Flemish, which has a much more ﬁxed word order. To sort this out would take an empirical investigation of quantiﬁer scope in West Flemish, which is beyond the scope of this book. There is a less radical alternative, however. One could assume that QR is possible in a scrambling language like West Flemish and investigate the hypothesis that an element that is not long-distance scrambled in a VPR-structure but could in principle be scrambled is necessarily focused. In a theory that assumes that a focused element is mapped into the nuclear scope of the sentence that contains it, an element that has undergone VPR will—though it could undergo QR in principle—be mapped into the nuclear scope of scopal elements in the matrix clause. An initial check of relevant data with Liliane Haegeman for West Flemish and Manuela Schönenberger for Swiss German revealed that the assumption that VPR-raised, scrambleable elements are focused seems to be correct. I have argued that it is not necessary to analyze VPR-structures in terms of rightadjunction (extraposition) to account for their scope properties. In the following section, I will try to show that an analysis in terms of adjunction not only is unnecessary but also actually fails to explain certain word order facts that can be naturally accounted for by assuming several types of leftward-movement within a VO-based approach. The data and most of the arguments are taken from Haegeman (1995a), who provides an excellent discussion of the issues concerning IPP-complements in WF. 3.2.2

Participles, inﬁnitives, and IPP-complements in WF

VR in WF diﬀers from VR in standard Dutch in two respects. In standard Dutch, a single inﬁnitive embedded under a ﬁnite modal verb may be inverted (cf. [45a–b]). As (46) shows, inversion of inﬁnitives is not possible in WF. In WF, an inﬁnitive always has to follow its selecting verb. This is also true for VPR-raised inﬁnitives, as is shown in (47). In an SOV approach, one would assume that the inﬁnitival complement must be extraposed (for whatever reason). (45) a. dat Jan dat boek wil kopen that Jan that book wants buy b. dat Jan dat boek kopen wil that Jan that book buy wants ‘that Jan wants to buy that book’ (46) a. da Valere dienen boek wilt kuopen that Valere that book wants buy b. *da Valere dienen boek kuopen wilt that Valere that book buy wants ‘that Valere wants to buy that book’

84

SCRAMBLING , REMNANT MOVEMENT , AND RESTRUCTURING IN WEST GERMANIC

(47) a. dan ze kosten [willen dienen boek kuopen] that they could want that book buy b. *dan ze [willen dienen boek kuopen] kosten that they want that book buy could ‘that they might want to buy that book’

Participles in standard Dutch may or may not undergo VR. Participles in WF, however, may never undergo VR in the sense of Rutten (1991), that is, right-adjoin to the selecting verb (48). (48) a. da Valere dienen boek gekocht eet that Valere that book bought has ‘that Valere has bought that boek’ b. *da Valere dienen boek eet gekocht that Valere that book has bought14

In contrast to inﬁnitival VPR-complements, IPP-complements, that is, complements that contain an inﬁnitive that stands for a participle, can both precede and follow the selecting (ﬁnite) verb (49). In the standard approach, one would account for this fact by assuming that extraposition of IPP-complements (as opposed to extraposition of inﬁnitival complements) is optional. (49) a. da Jan ee [willen [Marie nen boek geven]] that Jan has want-IPP Marie a book give b. da Jan [willen [Marie nen boek geven]] eet that Jan want-IPP Marie a book give has ‘that Jan has wanted to give Marie a book’

However, if the temporal auxiliary is nonﬁnite, the IPP-complement cannot be extraposed. Example (50) shows that the IPP-complement has to precede the nonﬁnite auxiliary when the auxiliary is part of a “te-inﬁnitival.” The distribution of an IPPcomplement is actually very restricted in this case: It has to follow te and it has to precede the inﬁnitival auxiliary. Example (51) shows that an IPP-complement has to precede its temporal auxiliary when the auxiliary represents a bare inﬁnitive. (50) a. Mee Jan te [willen [Valere dienen boek geven]] een with Jan to want-IPP Valere that book give have ‘with Jan having wanted to give Valere that book’ b. ??*Mee Jan te een [willen [Valere dienen boek geven]] (51) a. dan ze toch kosten [ [willen dienen boek kuopen] een] that they yet could want-IPP that boek buy have ‘that they very well could have wanted to buy that book’ b.

*dan ze toch kosten [een [willen dienen boek kuopen]]

COHERENT INFINITIVES IN DUTCH AND WEST FLEMISH

85

We have seen earlier that the IPP-complement of a ﬁnite verb may or may not be extraposed. However, an IPP-complement has to be extraposed when the ﬁnite verb incorporates the negative particle en. In WF, sentential negation is expressed by means of a negative constituent, which in ﬁnite sentences may cooccur with en (52a). The Spell-out of en on the V0 head is optional in ﬁnite sentences but illicit in nonﬁnite sentences. The head-status of en is indicated by the fact that when V moves to C (V2), en is moved along (52b). Example (52c–d) shows that when the IPPcomplement is not extraposed en cannot be spelled out on the ﬁnite verb. (52) a. da Valere ier nie (en) weunt that Valere here not (en) lives ‘that Valere does not live here’ b. Valere en-weunt ier nie Valere en-lives here not ‘Valere does not live here’ c. da Jan nooit en-eet willen an Valere nen boek geven that Jan never en has want-IPP to Valere a book give ‘that Jan has never wanted to give a book to Valere’ d. da Jan nooit willen an Valere nen boek geven (*en)-eet

Bringing the facts illustrated in (50), (51), and (52) into a single picture, it is hard to see what rule might govern extraposition in WF. Remember that the regularities are the following: On the one hand, an IPP-complement may or may not be extraposed if the selecting verb is ﬁnite; however, it must be extraposed when the selecting ﬁnite verb incorporates the negative particle en; on the other hand, it may not be extraposed at all when the selecting verb is nonﬁnite. Finally, an inﬁnitival complement must always be extraposed independently of the ﬁniteness of the selecting verb. To capture these regularities in a single rule that follows from other properties of West Flemish or from general principles of grammar seems almost impossible. Hence I will try to give an alternative explanation of these regularities based on the assumption that the West Germanic languages are VO languages rather than OV languages. I would like to point out that the account described here is Haegeman’s analysis, which I merely adopt to make my case against extraposition. In the subsequent discussion I will follow Haegeman (1995a) and adopt most of her generalizations and suggestions. In a VO-based approach all complements, that is, both inﬁnitival and IPP-complements in our case, start out to the right of the selecting verb. All I have to say about inﬁnitival complements is that they stay in their base position. IPP-complements seem to move leftward. Remember, as shown in (48) earlier, that participles, as opposed to inﬁnitives, have to precede the selecting auxiliary. Also note that the inﬁnitive in an IPP-complement stands for a participle, so there is good reason to assume that an IPP-complement is in fact a Participle Phrase (PartP). Thus, I would like to suggest that IPP-complements like participles move into the Speciﬁer of the selecting auxiliary to check the auxiliary’s subcategorization (the auxiliary have c-selects a PartP).15

86

SCRAMBLING , REMNANT MOVEMENT , AND RESTRUCTURING IN WEST GERMANIC

A te-inﬁnitive in WF consists of two independent heads. This is shown by the fact that the inﬁnitival marker can be separated from the verb by an IPP-complement (cf. [50]). Following Haegeman (1995a), I assume that the inﬁnitival marker heads a functional projection F1 directly above VP. So far I have developed the following picture: Inﬁnitives stay in their base position (53a). Participial clauses move into [Spec,VP] of the selecting auxiliary (53b). If we assume that nonﬁnite verbs, as opposed to ﬁnite ones, cannot move to F1, then we derive the fact that an IPP-complement may not be “extraposed” if the selecting verb is nonﬁnite. All we have to do to derive the fact that an IPP-complement may or may not be extraposed if the selecting verb is ﬁnite (and does not incorporate the negative marker en) is assume that ﬁnite verbs may optionally move to F1 (53c–d). This contrast between ﬁnite verbs and nonﬁnite verbs can be easily explained when one assumes that the degree of movement of a head correlates with the amount of inﬂectional material on that head that needs to be checked. Still the question arises why an IPP-complement can be “extraposed” while a simple participle, as is shown in (48b), cannot. The answer is that a participle moves even higher than F1. Example (54) shows that a participle cannot intervene between te and the auxiliary but must always precede the inﬁnitival marker.16 Finally, if we assume, following Haegeman’s (1995a) proposal, that (1) the negative particle en is licensed in F1 (or a higher functional head) and (2) that it has to incorporate into the verb, then we derive the fact that an IPP-complement has to be extraposed if the ﬁnite verb incorporates en (53e). If nonﬁnite verbs cannot move to F1, as we have assumed, then we also derive the fact that en can only be spelled out on a ﬁnite verb. (53)

F1

SPEC

a. (verb)

VERB

COMPLEMENT

(verb)

inﬁnitive

b. te

IPP

Aux

tIPP

c.

IPP

Auxﬁn

tIPP

d. Auxﬁn

IPP

tauxﬁn

tIPP

e. en-Aux

IPP

taux

tIPP

(54) a. Mee Valere dienen boek gewild te een (vu zenen verjoardag) with Valere that book wanted to have (for his birthday) ‘Valere having wanted that book for his birthday’ b. *Mee Valere dienen boek te gewild een (vu zenen verjoardag) c. ??*Mee Valere dienen boek t’ een gewild (vu zenen verjoardag)

It is clear that an account that combines extraposition with leftward head movement within an SOV approach is as good or elegant as the account just sketched. In this account, which presupposes that complements precede the selecting verb, one would assume that only inﬁnitival clauses are extraposed while participial clauses

COHERENT INFINITIVES IN DUTCH AND WEST FLEMISH

87

stay in their base position. The application or non-application of leftward head movement, according to the conditions outlined earlier, would then derive the correct distribution of IPP-complements. I have no empirical argument against such an account within an OV-based approach. I would just like to point to the peculiarity of the situation that such an OVbased approach presupposes and which consists in the fact that while all functional projections in the sentence seem to be head-initial, the verbal projection stands out with being head-ﬁnal. Zwart (1993) showed that there is no empirical argument that the verbal projection is head-ﬁnal. To my knowledge, there is also no empirical argument that it is not head-ﬁnal. All we are left with are conceptual considerations that lead us to investigate what a VO-based approach can reveal to us about VR, VPR, and other coherent inﬁnitival constructions.17 (55)

participle

te IPP

verb

inﬁnitival complement

In (55), I display the distribution of participles, inﬁnitives, and IPP-complements with respect to the selecting verb and the functional head F1, the position of which is indicated by the inﬁnitival marker te in (55). This pattern, again, clearly shows that IPP-inﬁnitives pattern with participles in that they move into an extended projection of the selecting auxiliary. It also shows that the additional movement of participles must be triggered by the participial morphology that IPP-inﬁnitives lack. I want to close this section with a phenomenon that was thought to distinguish VR and VPR. Like in standard Dutch, VR in WF can give rise to auxiliary switch. WF has the two aspectual auxiliaries een (‘to have’) and zyn (‘to be’), where zyn is used for unaccusative verbs and modals select een (cf. [56]). (56) a. da Valere no Gent gegoan is/*eet that Valere to Gent gone is/has b. da Valere dienen boek gewild eet/*is that Valere that book wanted has/is

As we know, modals trigger VR in Dutch and WF. If the verb-raised verb is an unaccusative verb, modals can appear with both auxiliaries in the perfect tense (57a). This is called auxiliary switch (auxiliary switch also occurs in Italian restructuring constructions). Based on the contrast between (57a) and (57b), Haegeman (1994) concludes that auxiliary switch is restricted to VR-structures and is not possible if the complement of the modal undergoes VPR. However, Haegeman (p.c.) overlooked that the auxiliary-switched case becomes grammatical when the auxiliary follows the verbal cluster as in (58a) and that the same contrast holds in VR-cases (58b). (57) a. da Valere nie no t schule eet/is willen goan that Valere not to school has/is want-IPP go ‘that Valere did not want to go to school’ b. da Valere nie eet/??is willen [no t schule goan] that Valere not has/is want-IPP to school go

88

SCRAMBLING , REMNANT MOVEMENT , AND RESTRUCTURING IN WEST GERMANIC

(58) a. da Valere nie [willen no t schule goan] is/??eet b. da Valere nie no t schule willen goan is/??eet that Valere not to school want-IPP is/has

I conclude from the data in (57) and (58) that the phenomenon of auxiliary switch does not distinguish between VR and VPR. Both constructions allow auxiliary switch, which seems to be subject to some PF-constraint: In both VR- and VPR-structures the auxiliary that is selected by the adjacent verb is strongly preferred over its alternative form.18

3.3 Conclusions In this chapter, I have discussed the properties of coherent inﬁnitives in Dutch and West Flemish. I introduced the standard OV-based biclausal approach to coherent inﬁnites in Dutch in terms of VR and LDS and the standard approach to Verb Projection Raising in terms of extraposition of the (extended) inﬁnitival VP. Then I outlined a number of problems and shortcomings of the traditional approach. In particular, in section 3.1 I concluded that the behavior of verb-particles shows that the Dutch verb cluster cannot be analyzed as a head-adjunction structure. In section 3.2.1, I argued that the scope facts in VPR-complements cannot be properly accounted for in terms of scrambling and extraposition. In section 3.2.2, we saw that the distribution of IPP-complements in West Flemish cannot be given a coherent explanation in terms of extraposition and concluded with Haegeman (1995a) that all the facts, that is, the distribution of inﬁnitives, participles, and IPP-complements, fall nicely in place if we assume that (1) complements are base generated to the right of the selecting head and (2) there is leftward head movement of the verb in West Germanic. The latter two assumptions follow from Kayne’s (1994) antisymmetric hypothesis. In the following, I will adopt the antisymmetric hypothesis and assume that all phrase structure is right-branching and that movement is only to the left. These assumptions exclude rightward movement of phrases (extraposition) as well as rightward movement of heads (VR) and will force us to provide new explanations for the distribution of syntactic elements in OV languages.

4

A Remnant Movement Account of Restructuring

B

efore we take a closer look at the properties of coherent inﬁnitives in German, I want to provide an alternative account of Verb Raising and Verb Projection Raising in a VO-based approach (cf. Hinterhölzl 1997, 1999, 2002b). Let me start with a discussion of how the verb-ﬁnal pattern in inﬁnitives can be derived in the traditional OV-based account and the antisymmetric VO-based account.

4.1 The verb-final position in infinitives As I have already outlined in chapter 1, the traditional approach, allowing for parametric variation in the head-complement order, assumes that the functional and lexical projections in the IP-domain are head-ﬁnal in the West Germanic OV languages. Therefore, I concluded that the embedded verb-ﬁnal position in (1a) is compatible either with the lack of V-movement (1b) or with rightward V-movement to a functional head within IP (1c). Furthermore, because of the right-headedness of the VP, an argument that precedes the verb may be analyzed as occupying its base position in the VP or as having scrambled out of the VP to a position in the middle ﬁeld (cf. [1c] and [1d]). (1)

a. weil Hans gestern das Buch las since Hans yesterday the book read-Past b. [CP weil [IP Hans gestern [VP das Buch las]]] 89

90

SCRAMBLING , REMNANT MOVEMENT , AND RESTRUCTURING IN WEST GERMANIC

c. [CP weil [IP Hans gestern [VP das Buch ti] lasi]] d. [CP weil [IP Hans gestern [das Buch]j [VP tj ti] lasi]]

A slight complication for the traditional approach is posed by the position that the inﬁnitival marker occupies in inﬁnitival clauses, as is illustrated in (2a). The sequence zu lesen (‘to read’) looks very much like a head-initial right-branching structure, in which the inﬁnitival marker that heads a functional position takes the inﬁnitival VP as its complement to the right. To accommodate this order with the particular assumptions about the word order in the IP-domain within the standard approach, it is assumed either that the inﬁnitival marker in the West Germanic SOV languages is not an independent head but rather a verbal aﬃx (Haider 1993), as illustrated in (2b), or that the inﬁnitival verb undergoes rightward head movement to (right-)adjoin to the head-ﬁnal inﬁnitival marker in the IP-domain, as illustrated in (2c). (2)

a. ohne gestern das Buch zu lesen without yesterday the book to read-INF b. [CP ohne [IP PRO gestern [VP das Buch [V zu-lesen]]]] c. [CP ohne [IP PRO gestern [VP das Buch ti] [I zu+leseni]]]

In the antisymmetric approach (Kayne [1994], Zwart [1993]), it is assumed, as I have noted earlier, that all structure is head-initial. In this approach, the embedded verb-ﬁnal order is derived by some sort of feature-driven leftward movement of VP-internal material. This is illustrated for arguments of the verb in (3a) and for VP-internal predicates in (3b). While the movement of DPs into AgrPs is in accordance with recent minimalist assumptions about Case-licensing of arguments, the movement of APs, PPs, and other VP-internal constituents required within the antisymmetric approach appears stipulative in the absence of any empirical evidence. (3)

a. [weil Hans gestern [das Buch]i [VP las ti]] since Hans yesterday the book read-PAST b. [weil Hans gestern [AP krank]i [VP war ti]] since Hans yesterday sick was

The crucial empirical evidence that supports the assumptions necessary within the antisymmetric approach and at the same time disqualiﬁes assumptions necessary within the traditional approach comes from the position of the inﬁnitival marker in nonﬁnite IPP-constructions in West Flemish (4a), taken from Haegeman (1995a), and Afrikaans (4b), taken from Donaldson (1993).1 (4)

a. mee Valere te [willen [dienen boek kuopen]] een with Valere to want that book buy have ‘with Valere having wanted to buy that book’

A REMNANT MOVEMENT ACCOUNT OF RESTRUCTURING

91

b. Die banke moes oop gewees het, om dit gister te [kan betaal] het the bank should open been have it yesterday to can buy have ‘the bank should have been open to have been able to buy it yesterday’

Since material, given in brackets in (4), can intervene between the inﬁnitival marker and the corresponding inﬁnitival verb, it follows that the inﬁnitival marker in the West Germanic SOV languages cannot be analyzed as a verbal aﬃx but, like in English, has to be analyzed as occupying a functional position within the IPdomain. That this functional position is not head-ﬁnal and that the sequence te+V cannot be accounted for by assuming rightward head movement of the verb within the traditional approach also follows directly from (4a). In (4a), the constituent that has been moved to the right of the inﬁnitival marker cannot possibly be a head since it contains the DP that book.2 I thus assume that the inﬁnitival marker occupies a functional head to the left of VP, which Haegeman (1995a) named F1 and which I will argue is the head of an Aspect Phrase. Though in German and Dutch the inﬁnitival marker cannot be separated from the verb, the minimal assumption seems to be (until we ﬁnd evidence to the contrary) that the inﬁnitival marker in German and Dutch, too, occupies F1, to be called Asp0 in the remainder of this book. The position of the inﬁnitival marker can thus be used to show that all VP-internal constituents, including APs, verbparticles, and PPs, move out of the VP. In the following section, I make us of the inﬁnitival marker to show that there is licensing movement out of the VP in German. Furthermore, I outline the structure of the lower middle ﬁeld in German, assuming that the generalizations extrapolated from German data carry over to Dutch and West Flemish.

4.2 Evidence for licensing movement out of the VP In this section, I investigate the structure of the German lower middle ﬁeld from the perspective of a VO-based approach. The most important results of this investigation are the following generalizations: (1) Nominal complements of the verb always have to leave the VP before Spell-out (independently of whether they are deﬁnite or indeﬁnite) and are licensed in functional projections above the position of manner adverbs. (2) Also, small clauses, idioms, and directional PPs have to move out of the VP and are licensed in a position below manner adverbs. (3) Full sentential complements are not licensed in their base position within VP, either. They undergo short movement and are licensed in a functional projection directly below the position to which the verb in embedded clauses moves. 4.2.1

The licensing of nominal arguments

That arguments move out of the VP obligatorily can already be shown with the help of manner adverbs. Adverbs like sorgfältig (‘carefully’), genau (‘precisely’, ‘exactly’), gut (‘well’), schlecht (‘badly’), and so forth show that both deﬁnite and

92

SCRAMBLING , REMNANT MOVEMENT , AND RESTRUCTURING IN WEST GERMANIC

indeﬁnite DPs have to leave the VP (cf. Brugger and Poletto [1995], Haiden [1995], Hinterhölzl [1999]). Since manner adverbs are usually analyzed as being adjoined to VP (I will later show that they actually occupy a higher position in the tree), a DP that precedes a manner adverb must have moved out of the VP. Examples (5) and (6) show that in sentences with neutral intonation, the direct object precedes the manner adverb. As already discussed in chapter 1, I assume that sentences with neutral intonation correspond to the unmarked word order, from which sentences with marked word order (often accompanied with a special intonation pattern) are derived by additional movement. These considerations are important since in a scrambling language like German alternative orders are often equally good but can—due to the assumptions made earlier—be treated as marked or secondary.3 (5)

a. weil Hans das Buch/ein Buch sorgfältig gelesen hat since Hans the book/a book carefully read has ‘since Hans has read the/a book carefully’ b. ??weil Hans sorgfältig das Buch/ein Buch gelesen hat4 since Hans carefully the book/a book read has

(6)

a. weil Hans den Plan/einen Plan genau ausführte since Hans the plan/a plan exactly executed ‘since Hans executed the/a plan exactly’ b. ??weil Hans genau den Plan/einen Plan ausführte since Hans exactly the plan/a plan executed

The negative marker nicht (‘not’) obligatorily precedes manner adverbs as is shown in (7). As I have shown in chapter 2, DPs may undergo further movement across the negative marker depending on their semantic and pragmatic properties. What is important at this point is that movement of DPs across manner adverbs is obligatory and independent of the semantic and pragmatic conditions that characterize scrambling. Thus, I would like to propose that the Case-licensing positions in German occur between manner adverbs and sentential negation and that movement of DPs across manner adverbs is licensing movement into these designated Casepositions, as is illustrated in (8). (7)

a. weil der Hans das Buch nicht sorgfältig gelesen hat since the Hans the book not carefully read has ‘since Hans has not read the book carefully’ b. ??weil der Hans das Buch sorgfältig nicht gelesen hat since the Hans the book carefully not read has

(8)

[CP [TP T [Scrambled DPs [Neg [Focus [Agr1 [Agr2 [Agr3 [Manner . . . [VPV]]]]]]]]]]

In (8), I assume that Case is licensed in designated Agreement positions. These positions are not assigned to any ﬁxed Case-values since German has V-class depen-

A REMNANT MOVEMENT ACCOUNT OF RESTRUCTURING

93

dent unmarked orders, as we have seen in chapter 2. Remember that in German the unmarked order of arguments cannot be determined by Case properties alone but is dependent on the thematic class a verb belongs to. The diﬀerent thematic classes with their corresponding unmarked orders in terms of Case sequences are illustrated again in (9). These unmarked orders can be derived if we assume that arguments move into their Case-licensing positions preserving their hierarchical relations as deﬁned by the thematic structure of the corresponding verb class and if we assume that the assumptions in (10) hold. (9) a. Nom > Acc : bedauern (‘regret’), interpretieren (‘interpret’) b. Acc > Nom : interessieren (‘interest’), imponieren (‘impress’) c. Nom > Dat : helfen (‘help’), gratulieren (‘congratulate’) d. Dat > Nom : gefallen (‘appeal/please’), fehlen (‘lack’) e. Nom > Dat >Acc : anvertrauen (‘entrust’), verbieten (‘forbid’) f. Nom > Acc > Dat: aussetzen (‘expose’), unterordnen (‘subordinate’) (10) a. (1) Agr1 is only projected if a little V (v) is present (2) Agr2 is only projected if the verb assigns lexical Case b. (1) Agr3 checks Accusative if v is present (2) Agr2 checks lexical Case (usually Dative) (3) Nominative is a default Case

The statement in (10b3) is meant to say that in the absence of any speciﬁc requirements, which are stated in (10b1, 2), any Agr-head can assign Nominative. The unmarked orders in (9) then follow from the assumptions in (10), since arguments must move into Case-licensing positions by preserving the hierarchical order given by the thematic properties of the verb. That is to say that an agent has to be licensed in a higher Case-position than, say, the theme argument and an experiencer has to be licensed in a higher position than the theme, and so on. Example (9a) then illustrates the typical basic order of a transitive verb with the agent licensed with Nominative in Agr1 and the object licensed with Accusative in Agr3. Examples (9b and d) show the typical basic order of intransitive verbs where the higher experiencer is licensed with lexical Accusative or lexical Dative in Agr2, while the lower argument is licensed with Nominative in Agr3 (since Agr3 cannot assign Accusative in this case). Example (9c) represents the case of a transitive verb with lexical Case, which is checked in Agr2 with the higher argument obligatorily moving to the higher Agr1 to be licensed with Nominative. Example (9e) represents the typical basic order of a ditransitive verb with the agent licensed with Nominative in Agr1, the recipient licensed with lexical Dative in Agr2, and the theme licensed with Accusative in Agr3. Example (9f) represents a subclass of ditransitive verbs that is problematic for the system expounded here, since the Dative is licensed not higher but lower than the Accusative, that is, below Agr3. I would like to propose that this subclass does not represent an exception to the rules

94

SCRAMBLING , REMNANT MOVEMENT , AND RESTRUCTURING IN WEST GERMANIC

speciﬁed in (9–10) earlier. Instead, I assume that the Dative is assigned by the separable particles that make up the verbs of this class, proposing that the Dative in this case is checked in the Speciﬁer of PredP, a functional position below the Agr-heads that I will introduce later. 4.2.2

Attract Closest and the issue of parallel movement

At this point, the question arises of how it is achieved that arguments undergo Caselicensing movement by preserving the hierarchical order in the VP, since the Caselicensing heads that I have proposed earlier do not have the ability to attract speciﬁc Case-forms.5 Let me address this issue and the reviewer’s criticism expressed in the previous note by ﬁrst pointing out that the system of Case-licensing discussed here is designed to account for a fair amount of idiosyncratic variation of Caselicensing properties of German verb classes. Therefore, unless we want to put all the information into the lexicon, a syntactic account of these facts is necessarily complex. In particular, any account must say something about how it is possible that Nominative is assigned to the lower argument with unaccusative verbs. Moreover, since I am subscribing to a framework that does not allow for multiple Speciﬁers, the descriptive generalization that structural Accusative is bound to (a certain type of) little v, and which is embodied as a second merge of the object in [Spec,vP] within minimalist treatments, has to be captured in a more indirect way. To return to the issue of how we can account for the fact that arguments when undergoing Case-licensing movement must preserve the hierarchical relationships established in the vP, the reviewer is right in pointing out that there is currently no principle of grammar that enforces crossing movements in the process of licensing arguments. In minimalism, we only have the principle Attract Closest (AC). AC, in the absence of any inherent Case-feature on the Agr-head, would have the subject being moved to the lowest Case-checking head instead of attracting the object of a transitive verb. The alternative that the direct object is licensed before the subject is merged is also not viable within the account of restructuring that I am developing here. Crossing movements (or parallel movement) should follow from a strict version of the Cycle Condition that has it that any rule applies in the smallest domain ﬁrst. Attract Closest as it is formulated is too coarse a principle. First, it must be taken to hold at the end of the phase. Second, it must be taken to compute not the eﬃciency of single operations but the eﬃciency of all operations within a phase (the economy of the sum of movements that follow from a certain selection). Attract Closest as it stands yields the right results in a case where we have one probe and two possible matching targets, as is illustrated in (11a), but fails in a situation in which we have two potential probes and two matching targets, as is illustrated in (11b). (11) a. C [wh] b. Agr1

Su[wh]

V

Obj [wh]

Agr2

Su

V

Obj

A REMNANT MOVEMENT ACCOUNT OF RESTRUCTURING

95

In (11a), we want the wh-feature of C to be checked by the subject, which follows from AC. In (11b), however, we do not want the subject to check Agr2, which property does not follow from AC. To account for both cases, Attract Closest can then be replaced by Cyclic Attraction, which applies if a probe has two alternative suitable targets within its phase, as is deﬁned in (12). (12) a. Cyclic Attraction A probe selects a suitable target in the smallest cyclic domain unless an alternative selection (of a target in a bigger domain) is more economic. b. Economy The selection of an alternative target in a larger domain is more economic if this selection leads to shorter movements in the overall derivation within the phase that contains the probe.

Let us see how these principles work. In (11a) Cyclic Attraction selects the object; alternative selection selects the subject. Since the latter selection is more economic than the former (fewer nodes are crossed by the subject than by the object), the subject wins out and we derive the eﬀect of simple AC. In the situation of (11b), however, Cyclic Attraction will select movement of the object to Agr2 (and movement of the subject to Agr1). Note that the alternative selection that involves movement of the subject to Agr2 and movement of the object to Agr1 is not more economic since the overall computation involves the same number of nodes being crossed in subject and object movement together. Thus, movement implied by alternative selection is not possible and the derivation having the object move to Agr2 will be selected as more economic when the phase is evaluated. Maybe this principle could be formulated more elegantly, but it will derive the eﬀect that (all other things being equal) multiple licensing movements preserve preexisting orders. 4.2.3

The licensing of predicates

In a VO-based approach, we have to assume that the non-verbal predicates in (13) have been moved leftward from a position to the right of the verb. I assume that these non-verbal predicates, together with the “direct object” of the verb, form a small clause, which in turn forms the complement of the verb as is illustrated in (13'). (13) a. weil Hans das Haus gelb färbte since Hans the house yellow painted b. weil Hans ihn einen Idioten nannte since Hans him an idiot called c. weil Hans Maria intelligent ﬁndet since Hans Maria intelligent ﬁnds since Hans considers Maria intelligent

96

SCRAMBLING , REMNANT MOVEMENT , AND RESTRUCTURING IN WEST GERMANIC

(13') a. weil Hans [VP färbte [SC das Haus gelb]] b. weil Hans [VP nannte [SC ihn einen Idioten]] c. weil Hans [VP ﬁndet [SC Maria intelligent]]

I assume that these small clauses form a complex predicate with the selecting verb, with the result that they share the licensing domain of the selecting verb. Following Zwart (1993) and Koster (1995), I assume that these small clauses undergo XP-movement into the Speciﬁer of a Predicate Phrase (PredP) that dominates the VP (cf. also Bowers [1993] for the role of the Predicate Phrase). Example (13") shows that this Predicate Phrase occupies a position below the position of manner adverbs. In (13"), I used the particle-verb an-streichen (‘paint’) in order to make sure that the head of the small clause predicate gelb (‘yellow’) has not incorporated into the verb and actually occupies a position in PredP; recall from the discussion in 3.2.3 that particles block the incorporation of small clause heads into the verb. Example (13"a) also shows that the subject of the small clause den Zaun is moved out of the domain of the manner adverb. Presumably it moves, like the nominal arguments of verbs, to its Case-licensing position below the negative marker and from there it moves like other speciﬁcs (if not contrastively focused) to a scrambling position above the negative marker, as is indicated in (13"c). (13") a. weil Hans den Zaun sorgfältig gelb angestrichen hat since Hans the fence carefully yellow up-painted has b. ??weil Hans den Zaun gelb sorgfältig angestrichen hat since Hans the fence yellow carefully up-painted has c. weil Hans den Zauni nicht ti [ti gelb] angestrichen hat since Hans the fence not yellow up-painted has

That VP-internal predicates (including small clauses, idioms, and directional PPs) indeed occupy a position in the middle ﬁeld, namely PredP, rather than remain within a right-headed VP, can be shown with the help of the inﬁnitival marker. The data in (14) show that an adjectival small clause predicate cannot remain within the VP or, for that matter, incorporate into the verb, as is often assumed (cf. Neeleman [1994b]), but has to move out of the VP to a licensing position above the inﬁnitival marker. Example (14) shows that PredP dominates the AspP in German (14a–b) and in West Flemish (14c–d). The contrast in (14c–d) is even more signiﬁcant since we have seen in (4a) earlier that the inﬁnitival marker in West Flemish can in principle be separated from the verb. We have argued above that small clauses are licensed in PredP. Example (14) shows that the head of a small clause predicate obligatorily precedes the inﬁnitival marker that occupies Asp0; hence PredP must dominate AspP. (14) a. ohne die Tür grün zu färben without the door green to paint

A REMNANT MOVEMENT ACCOUNT OF RESTRUCTURING

97

b. *ohne die Tür zu grün färben without the door to green paint ‘without painting the door green’ c. K goan proberen van die deure groen te verwen (Haegeman 1995a, 23) I go try of the door green to paint d. *K goan proberen van die deure te groen verwen I go try of the door to green paint ‘I will try to paint the door green’

The test with manner adverbs indicates that idiomatic expressions and directional PPs are licensed in PredP, too. Since these elements can only occur between manner adverbs and the inﬁnitival marker, they must occupy PredP in (15). (15) a. um es ihr schnell zur Verfügung zu stellen in order it her quickly to-Agr availability to put b. *um es ihr zur Verfügung schnell zu stellen in order it her to-Agr availability quickly to put ‘in order to make it available for her quickly’ c. um die Milch vorsichtig in den Kühlschrank zu stellen in order the milk carefully into the refrigerator to put d. *um die Milch in den Kühlschrank vorsichtig zu stellen in order the milk into the refrigerator carefully to put ‘in order to put the milk into the frige carefully’

So far we have arrived at the following picture of the lower middle ﬁeld in German: Nominal arguments (including prepositional arguments) are Case-licensed in Agr-heads, as illustrated in (8), above the position of manner adverbs. The negative marker dominates manner adverbs. Small clauses, idiomatic expressions, and directional PPs are licensed in PredP directly below the position of VP-adverbs. This Predicate Phrase dominates AspP, which hosts the inﬁnitival marker and in the Speciﬁer of which verb-particles are licensed. AspP itself immediately dominates VP (16). The schema in (16) does not display the complete structure of the lower middle ﬁeld. It just indicates a few landmark positions that I have chosen as criterial for diﬀerentiating licensing movement out of the VP from additional movement that scrambles arguments higher up in the structure. I assume that the licensing positions of arguments and their related scrambling positions are interspersed with functional projections that license adverbs (cf. Frey and Pittner [1998] for some discussion). (16)

[Scrambled DPs [Neg [DPs Agr [Manner adverbs [Predicates Pred0 [Particles Asp [VP]]]]]]]

98

SCRAMBLING , REMNANT MOVEMENT , AND RESTRUCTURING IN WEST GERMANIC

4.2.4 The licensing of sentential arguments Let us now look at the behavior of sentential complements (CPs). CPs in German appear invariably to the right of the verb that selects them. Given that the German verb in embedded clauses does not move, the CPs in (17a–b) seem to be in their base position, as is indicated by the analysis of (17b) in (17c). Although there is clear evidence that the German embedded verb does not move to Tense—as we have seen earlier, it cannot move across negation or even VP-adverbs—it is still possible that this verb undergoes some type of short local movement that moves it across a sentential complement that has itself moved into the Speciﬁer of a functional projection for purposes of licensing, as is illustrated in (17d) for the sentence in (17b). (17) a. weil Hans der Maria nicht sagte, dass Peter krank ist since Hans Maria-DAT not told that Peter sick is ‘since Hans did not tell Maria that Peter is sick’ b. ohne der Maria zu sagen, dass Peter krank ist without Maria-DAT to say that Peter sick is ‘without telling Maria that Peter is sick’ c. ohne der Maria [Asp zu [VP sagen dass Peter krank ist]] d. ohne der Maria [Asp zu [F2 sagen [F3 CP [VP tV tCP]]]]

From the licensing movement of adjectives (cf. [13] and [14] earlier) it follows that CP-complements cannot remain within the VP. Example (18a) shows the only possible order between adjective and CP-complement in an inﬁnitival clause. The adjectival phrase undergoes licensing movement into PredP above the inﬁnitival marker. As (18b) shows, pied-piping of the CP-complement leads to ungrammaticality. It follows then that the CP-complement has to move out of the VP before the adjectival phrase moves to PredP, in order to derive (18a) from the underlying structure in (18c). (18) a. ohne froh zu sein, dass der Hans nicht kam without happy to be that the Hans not came ‘without being happy that Hans did not come’ b. *ohne [froh, dass der Hans nicht kam] zu sein without happy that the Hans not came to be c. [CP ohne . . . [AspP zu [VP sein [AdjP froh [CP]]]]]

In the standard theory, cases like (18) have been accounted for by the operation of extraposition that right-adjoins CP-complements to the local VP or IP. Since, following Kayne (1994), I want to refrain from positing right-adjunction, I propose that the CP-complements are licensed in [Spec,F3P] directly above VP. This entails that the verb in German moves up (at least) to F2, a functional projection that dominates the licensing projection for CPs. We thus arrive at the following structure of

A REMNANT MOVEMENT ACCOUNT OF RESTRUCTURING

99

the German sentence, a structure that I assume also holds for Dutch and West Flemish. In (19), S-DPs stands for “scrambled DPs”, C-DPs for “Case-licensed DPs” in their base order, and S-Advs for “sentential adverbs” like oft (‘often’). For the sake of simplicity, I assume that the domain of S-DPs corresponds to the traditional TP. Similarly to (16), the structure of (19) is intended to indicate, for instance, that arguments are licensed in the domain of event-related adverbs, while scrambling moves them into the domain of sentential adverbs, with the border marks being deﬁned by PredP and negation. In particular, (19) is not intended to show that arguments are Case-licensed as a uniform block above the positions of all event-related adverbs, that is, above time, place, and manner adverbs. (19) [S-DPs [S-Advs [Neg [C-DPs [VP-Advs [Pred0 [AspP zu [F2P V [F3P CP [VP ]]]]]]]]]]

4.2.5

The rationale for licensing movement

In the previous section, I remarked that the assumption of licensing movement of APs, PPs, and particles seems stipulative in the absence of any empirical evidence. Using the inﬁnitival marker, I have shown that there is empirical evidence for leftward movement of all VP-internal material. At this point two questions arise. First, the question indeed remains what the motivation of the movement of these constituents is. There are two notions of licensing relevant here: formal licensing and functional licensing. Formal licensing can be made sense of if we follow Chomsky (1995b) and assume that only thematic properties are satisﬁed (through merge) in the VP. Then all other properties of the verb, its morphological properties, the speciﬁc lexical requirements it imposes on its complements (what has generally been called c-selection), must be satisﬁed outside of the VP, in the extended projection of the verb. For instance, lexical Case in the GB-framework was assumed to be assigned to the argument in the VP together with the assignment of a thematic role. In MP, it is assumed that no feature checking is possible in the VP. Instead, I want to propose that speciﬁc morphological or lexical properties are satisﬁed through the projection of a corresponding functional head within the extended projections of the verb. In the case at hand, this requirement is satisﬁed through the projection of a functional head that can be checked by a DP with lexical Case (we can assume that if a verb with lexical Dative is selected the numeration also contains the functional head Agr2[Dat]). The parallel motivation applies to PP-complements. They have to move out of the VP to satisfy the selection of the verb for a speciﬁc preposition in the Spec-head relation with a functional head that represents this selectional property of the verb. In other words, the extended projections of the verb are a complex function of the semantic and lexical properties of the verb. We have seen in (10) earlier that the types of Case-checking heads that a verb projects are a function of its semantic (absence or presence of little v) and lexical (absence and presence of lexical case) properties. In conclusion, c-selection can be thought of as being satisﬁed through projection. This approach can thus be characterized as the complete syntactization of the idea behind the Projection Principle, namely, the requirement that semantic and lexical properties are represented in the syntax through projection (c-selection via upward projection).

100

SCRAMBLING , REMNANT MOVEMENT , AND RESTRUCTURING IN WEST GERMANIC

What is meant by functional licensing? To answer this question, I will point out some observations of a general kind and some observations that are implied by the previously described distribution of elements in the German middle ﬁeld. A secondary purpose of these licensing movements seems to be to determine which ones of the formal complements of the verb function as arguments or as (co-)predicates of the verb. The choice between these two options is not always so trivial as it may seem. DPs are the prime candidates to function as arguments. But idiomatic DPs and non-referential DPs such as in Klavier spielen (‘to play the piano’) and in einen Lügner nennen (‘to call someone a liar’) are licensed as predicates in PredP. Also, small clauses are licensed in PredP. But if Moro (1997) in his treatment of predicate raising is right, then the choice is not even predetermined with small clauses, since the so-called subject of the small clause can function both as subject or as predicate in the main clause. Along the same lines, PPs can be licensed as arguments or as predicates. The preceding investigation shows that directional PPs are licensed as predicates, while other PP-complements of the verb as in warten auf Maria (‘wait for Mary’) are licensed as arguments. CP-complements are typically licensed as arguments. However, the basic account that I will provide for coherent inﬁnitives in the following section is that they are licensed as copredicates, not as arguments, implying that since predicates are transparent for extraction, the functional domain of the verb above PredP is accessible for the arguments of both predicates. We have also seen that verb-particles are licensed in [Spec,AspP]. This can be made sense of in terms of functional licensing since particles typically modify the aspectual interpretation of the verb (phrase). As for the licensing of arguments, the functional structure above VP seems crucial since it determines the grammatical status of each verbal argument. Assuming that the thematic structure of active and passive verbs is identical, passive verbs can be taken to assign a diﬀerent argument status to their complements (the agent is licensed as an implicit argument). In a similar vein, assuming that the thematic structures of verbs in ergative and non-ergative languages are identical (consider, in this respect, languages with split-ergativity), we can assume that the functional structure above VP assigns a diﬀerent grammatical status (oblique or structural) to the arguments of ergative and non-ergative verbs. Although these observations cannot be worked out in more detail here, I hope they give the reader some idea of what I have in mind when I speak of functional licensing. Also, since licensing movement as described previously preserves the hierarchy of VP-internal material, a simpler analysis that moves the verb out of the VP and then moves the remnant VP in front of the verb, as proposed by Haegeman (2001, 2002b), would seem preferable. However, since the licensing positions of VP-internal material are interspersed with functional heads that arguably are VPexternal, like Aspect, heads that license adverbs (cf. Cinque [1999]) and negation, licensing movement out of the VP could at best be in several chunks. Such an account, however, would not conform to the previously sketched idea of grammatical licensing.

A REMNANT MOVEMENT ACCOUNT OF RESTRUCTURING

101

4.3 An XP-movement account of restructuring In this section, I introduce a novel VO-based approach to the analysis of Verb Raising (VR) and Verb Projection Raising (VPR) constructions. I will ﬁrst describe the analysis of these constructions in the standard approach and show that the basic assumptions that this analysis rests on are untenable. The basic tenets of the standard analysis are Verb Raising (VR), an operation of head movement that is taken to account for the formation of verb clusters, and long-distance scrambling (LDS), an operation that moves arguments of the inﬁnitival verb into the matrix domain in order to account for the so-called clauseunion phenomenon. I will ﬁrst demonstrate with the help of verb-particles in Dutch that verb cluster formation involves movement of the (extended) inﬁnitival VP. Second, I will argue with the help of adverbs and other elements that resist scrambling that the distribution and interpretation of elements that belong to the dependent inﬁnitive (its arguments and adverbs that modify it) cannot be accounted for with the assumption that these elements can scramble individually into the domain of the selecting verb but calls for the movement of a larger constituent into the domain of the selecting verb that pied-pipes these elements. Dispensing with LDS will allow me to drop the assumption that the complements of restructuring verbs are not full CPs. In order to allow for LDS, which is illicit out of non-restructuring inﬁnitives and out of tensed embedded clauses, it has been assumed within the standard approach that the complements of restructuring verbs are IPs or bare VPs. This consequence has the advantage that we henceforth may assume that all sentential complements are CPs. Instead of VR and LDS, I will argue that restructuring involves remnant movement of larger parts of the inﬁnitival. More speciﬁcally, I propose that restructuring breaks down into movement of the inﬁnitival AspP and the inﬁnitival TP into designated positions in the matrix clause. I demonstrate that these movements are cyclic (operating in the same fashion in every clause) and occur for licensing reasons. Assuming that the complementizer is essential for checking the requirements that the verb imposes on the embedded clause, I show how these movements can be derived from a general theory of sentential complementation. Finally, I will argue that an OV-based approach to the syntax of the West Germanic languages has to assume the same types of movements that are predicted to occur within a VO-based approach but that the VO-based approach is to be preferred over an OV-based approach since it allows one to dispense with an operation of rightward movement, namely, Extraposition. 4.3.1

The OV-based account and its problems

In this section, I will brieﬂy sketch again the traditional OV-based analysis of VRconstructions. I will then point to empirical and conceptual problems of this analysis that lead me to the conclusion that VR-constructions crucially involve XP-movement rather than VR in the sense of Evers (1975).

102

SCRAMBLING , REMNANT MOVEMENT , AND RESTRUCTURING IN WEST GERMANIC

To remind us of the traditional OV-based analysis of coherent inﬁnitives, let us look again at an example of a bare inﬁnitival selected by an ECM-verb in Dutch. In (20), the matrix verb intervenes between the embedded inﬁnitive and its arguments. To accommodate this order with the head-ﬁnal character that he assumed for Dutch, Evers (1975) proposed a rule of VR that right-adjoins the dependent inﬁnitive to its selecting verb as depicted in (20b). (20) a. dat ik Jan een liedje hoor zingen that I Jan a song hear sing ‘that I hear Jan sing a song’ b. dat ik [Jan een liedje tR ] hoor zingenR

In an OV-based account, verb cluster formation in coherent inﬁnitives is accounted for by the rule of VR that right-adjoins dependent inﬁnitives to their selecting verb in Dutch but left-adjoins them in German. The transparency of these constructions is simply accounted for by the assumption that bare inﬁnitives are VP-complements (cf. Broekhuis et al. [1995]). Remember that in Dutch a particle may precede the verb cluster created by VR, as is illustrated in (21a), or, as is shown in (21b), it may also move along with its selecting verb and become part of the verb cluster. (21) a. dat Jan Marie op wil bellen that Jan Marie up wants call b. dat Jan Marie wil op bellen that Jan Marie wants up call ‘that Jan wants to call up Marie’

Following van Riemsdijk (1978), I assume that the particle is the head of a particle phrase, a PP in fact, in the complement domain of the verb (22a). The fact that the verb and the particle often act as a unit is accounted for in this approach by adopting a rule of Particle Incorporation (PI) that moves the particle to the verb. If PI is optional, then the particle may stay in its PP and will be stranded by VR, yielding the analysis depicted in (22b) for sentences like (21a); or the particle may incorporate into the verb and subsequently undergo VR, yielding the analysis depicted in (22c) for sentences like (21b). (22) a. dat Jan Marie [PP [P op]] bellen wil (base structure) b. dat Jan Marie [PP op] tR wil bellenR c. dat Jan Marie [PP tPI] tR wil [opPI bellen]R

First, note that while this analysis accounts for the behavior of particles in VRconstructions, it fails to explain why VR viewed as a process of head movement may pied-pipe or strand particles, while Verb Second must strand particles, as is shown

A REMNANT MOVEMENT ACCOUNT OF RESTRUCTURING

103

in (23). If Verb Second is an operation of head movement that moves the ﬁnite verb into C0 in root clauses, as is standardly assumed (cf. Den Besten [1974/1983]), then it is not clear why Verb Second may not, very much like VR, pied-pipe particles. If Verb Second is to be characterized as XP-movement (cf. Hallman [1997]), then VR that can aﬀect even a bigger chunk of structure, the verb plus a particle, should be analyzed as XP-movement as well. (23) a. *Jan opbelde Marie Jan up-called Marie b. Jan belde Marie op Jan called Marie up ‘Jan called up Marie’

Second, note that particles can never be taken to incorporate into the verb in teinﬁnitives: The inﬁnitival marker te always intervenes between the particle and the selecting verb (24). (24) a. *zonder Marie te op bellen without Marie to up call ‘without calling up Marie’ b. zonder Marie op te bellen without Marie up to call

Remember that in the traditional OV account, where the VP precedes its selecting head I0, it is assumed that the inﬁnitival verb undergoes head movement and right-adjoins to the inﬁnitival marker that occupies I0, in order to derive the sequence te + inﬁnitive. In the beginning of this chapter (cf. [4]) I presented data from West Flemish and Afrikaans that cast serious doubt on the head-adjunction analysis of te-inﬁnitives. If the head-adjunction analysis of te-inﬁnitives is given up, it must be assumed that particles incorporate into Asp0, the position of the inﬁnitival marker, rather than into the verb. Movement of a particle and an inﬁnitival verb as in (22c) must then always involve Asp0. This movement can only then be assumed to be head movement in an OV-based approach if it is assumed that the inﬁnitive moves up to Asp0. Note that this latter movement would constitute leftward verb movement within an OV-based approach. This consequence is incompatible with the basic assumptions of an OV-based system. Given that the functional heads within the IP-domain occur to the left (cf. Koopman [1995], Zwart [1993]), as the position of the inﬁnitival marker shows, the VP would be the only left-branching projection within an OV-based system. If the inﬁnitive does not move up to Asp0, then movement of the particle and the inﬁnitival verb in (22c) must involve Asp0 plus the verb in its base position in the VP and we have a clear case of XP-movement. But even if the head-adjunction analysis of te-inﬁnitives were tenable, there would still be problems for the OV-based approach. The important point is that the sequence op te bellen cannot have an analysis in which it is assumed that the

104

SCRAMBLING , REMNANT MOVEMENT , AND RESTRUCTURING IN WEST GERMANIC

verb is right-adjoined to te and the particle is left-adjoined to te. Example (25a) shows the base structure and example (25b) shows the desired resulting structure. (25) a. [IP PRO [VP [PP op] V] te] b. [IP PRO [VP [PP ti] tj] opI + te + Vj]

That is to say that regardless of whether we assume that the particle has incorporated into the selecting verb, this sequence cannot be analyzed as a single complex head. If the particle incorporates into the selecting verb (which then excorporates to adjoin to the inﬁnitival marker), it cannot excorporate (out of a head-adjunction structure in which it does not constitute the head) in order to left-adjoin to the inﬁnitival marker. If it does not incorporate into the verb, it cannot incorporate to the higher inﬁnitival marker, since this would require that it cross an intervening head in violation of the HMC.6 All the preceding discussion implies that the VR-structure in (26) cannot have been derived by an operation of head movement that right-adjoins a (complex) inﬁnitival head to its selecting verb. I therefore conclude that VR involves XP-movement of a VP or some bigger projection. Furthermore, I assume that particles are never licensed by incorporation. I assume that they undergo XP-movement to be licensed in [Spec,AspP] (or [Spec,PredP] if modiﬁed). This assumption naturally explains why particles must be stranded by the operation of Verb Second, if it is assumed that Verb Second is head movement, while they may be pied-piped by VR, an operation that involves movement of a larger maximal phrase. (26) dat Jan Marie tR probeerde [op te bellen]R that Jan Marie tried up to call ‘that Jan tried to call up Marie’

An OV-based approach is certainly not committed to a head-movement analysis but requires the assumption of rightward movement, which is illicit under Kayne’s LCA. Another problem of the head movement account of coherent inﬁnitives is that it treats some clausal complements as CPs and others as TPs or VPs. It would be nice to have a theory of coherent inﬁnitives that allows one to assume that all sentential complements are CPs. In the following section, I will propose a VO-based account that assumes that both coherent and non-coherent inﬁnitival complements are full CPs. 4.3.2

A VO-based account of verb raising

The main problem that a VO-based account of coherent inﬁnitives faces is the question of how to account for the distribution and interpretation of nominal arguments, adverbs, and sentential complements in this construction. If we look at a typical case of VR in Dutch (cf. [27a–b]), then we realize that the nominal arguments of the inﬁnitive and adverbs and adverbials that modify it precede the selecting verb wilde, while the inﬁnitive itself and a sentential complement of the inﬁnitive (27c) follow the selecting verb. In (27) constituents that belong to

A REMNANT MOVEMENT ACCOUNT OF RESTRUCTURING

105

the embedded inﬁnitival are given in brackets. In a VO-based account, we have to assume that a coherent inﬁnitive like (27a) is derived from an underlying structure of the type given in (27b). (27) a. dat Jan [Marie het boek morgen] wilde [geven] that Jan Marie-DAT the book tomorrow wanted give ‘that Jan wanted to give Marie the book tomorrow’ b. dat Jan wilde [PRO Marie het boek morgen geven] that Jan wanted Marie the book tomorrow give c. dat Jan [Marie morgen] wilde [vertellen dat Piet ziek is] that Jan Marie tomorrow wanted tell that Piet sick is ‘that Jan wanted to say to Marie tomorrow that Piet is sick’

The simplest possibility of relating the structure in (27a) with the underlying structure in (27b) is to assume that the bracketed constituents that precede the matrix verb have been moved individually via scrambling from the embedded clause into the matrix clause. However, a closer look at this assumption reveals that scrambling (alone) is not a solution for our problem at hand. In the following section, I will illustrate why. 4.3.2.1

Against scrambling

In this section, I present two arguments that constituents of the inﬁnitive cannot be moved via scrambling into the domain of the matrix clause in VR-constructions. (1) Verb particles, small clause predicates, and idiomatic expressions cannot scramble but can precede the verb that selects the inﬁnitival complement in VR-constructions. This is illustrated for small clause predicates in (28). Example (28a) shows the basic order in which the small clause predicate follows the small clause subject and precedes the ﬁnite verb. Example (28b), in which the small clause predicate has been scrambled (across the adverb), is ungrammatical. Thus, it is implausible that the small clause predicate has been moved via scrambling into the matrix clause in (28c). (28) a. dat Jan de schuur gisteren rood schilderde that Jan the barn yesterday red painted ‘that Jan painted the barn red yesterday’ b. ??dat Jan de schuur rood gisteren schilderde that Jan the barn red yesterday painted c. dat Jan de schuur rood wil schilderen that Jan the barn red wants paint ‘that Jan wants to paint the barn red’

When I claimed earlier that verb-particles, small clause predicates, and idiomatic expressions cannot scramble, that was not entirely correct. What I meant was

106

SCRAMBLING , REMNANT MOVEMENT , AND RESTRUCTURING IN WEST GERMANIC

that they cannot undergo scrambling proper (cf. chapter 2 for a more extensive discussion). However, there is another type of scrambling that both arguments and nonarguments may undergo, as is discussed in detail in chapter 2. To remind us of the basic facts, let us look again at the situation in Dutch. In Dutch, like in German, arguments can scramble across adverbs. Unlike in German, however, this is permissive as long as the hierarchy of arguments is preserved. Scrambling that permutes the arguments in a clause is illicit. For instance, as shown in (29), the object may not scramble across the subject. In (29a), both deﬁnite DPs have scrambled across the negative marker preserving their respective hierarchy, with the subject preceding the object. Example (29b), where the object has scrambled across the subject, is ungrammatical. There is, however, a type of scrambling that Neeleman (1994b) calls focus-scrambling (and that I termed S-scrambling in chapter 2) that may do exactly that, namely, move the object in front of the subject, as illustrated in (29c). (29) a. dat Jan de boeken niet koopt dat Jan the books not buys b. *dat de boeken Jan niet koopt that the books Jan not buys c. dat zulke boeken zelfs Jan niet koopt that such books even Jan not buys

Recall from chapter 2 that the type of scrambling illustrated in (29c) comes with a special intonation pattern and involves a focused constituent, zelfs Jan in (29c), that is contrasted with the moved element that is best described as a (contrastive) topic. Remember also that this type of scrambling is not restricted to arguments but may also involve predicates (30). (30)

dat rood zelfs Jan de hekje niet verfd that red even Jan the fence not paints

What is important, however, in the case of the alleged scrambling of the adjectival predicate rood in (27c) is that no special intonation pattern or any focused constituent is necessary here. So that we can safely exclude also S-scrambling (or focus-scrambling) as a possible explanation for cases like (28c). (2) It is generally assumed that adverbs do not scramble. That is to say that if two adverbs, ADV1 and ADV2, can occur both in the order ADV1 > ADV2 and in the order ADV2 > ADV1, it is assumed that at least one of them can be base generated in a higher and in a lower position. This assumption is conﬁrmed by the observation that in English, which is a non-scrambling language, adverbs of the same type may occur in diﬀerent positions. Furthermore, Cinque (1999) in his extensive work on the relation between the functional hierarchy and the positions of adverbs argues convincingly that cases in which two adverbs can occur in two diﬀerent orders involve movement of a larger phrase that contains one adverb over the other adverb.

A REMNANT MOVEMENT ACCOUNT OF RESTRUCTURING

107

In addition, there is also direct empirical evidence for the assumption that adverbs at least cannot undergo long-distance scrambling (LDS), which comes from cases of Remnant Extraposition. Example (31) is a case of Remnant Extraposition and shows that while arguments can undergo LDS, adverbs and adverbials cannot. Here the DP het boek has undergone LDS into the matrix clause, but the adverbial om drie uur (‘at three o’clock’) must have been base generated in the matrix clause, since it cannot be interpreted as modifying the embedded inﬁnitive. In VRconstructions, however, adverbs and adverbials that precede the matrix verb are always ambiguous between a matrix clause construal and an embedded clause construal. (31)

Jan heeft het boeki om drie uur geweigerd ti weg te brengen Jan has the book at three o’clock refused away to take ‘Jan has refused at three o’clock to take away the book’

4.3.2.2

The alternative account

In this section, I will outline an account of VR in a VO-based approach that dispenses with rightward movement as well as with unmotivated scrambling operations. This approach is based on the generalizations about the basic clausal structure that I outlined in section 4.2 earlier (cf. [19]) and on the assumption that coherent inﬁnitives are full CP-complements. Let us look again at a simple case of VR in Dutch to remind us of the problem at hand. If (32a) is derived from the underlying structure given in (32b), then it seems that while the inﬁnitive may stay in the embedded clause, everything else, namely, the arguments of the embedded verb and the adverb that modiﬁes it, must move into the matrix clause. (32) a. dat Jan het boek vaak lang wil lezen that Jan the book often long wants read ‘that Jan often wants to read the book for a long time’ b. [dat vaak [VP Jan wil [CP . . . [lang [AspP 0 [VP PRO lezen het boek]]]]]]

I have argued that this movement cannot be scrambling of the individual constituents. In particular, I have argued in the previous section that adverbs cannot be taken to undergo scrambling. It is thus natural to assume that adverbs end up in the domain of the matrix clause by being pied-piped by the movement of a larger constituent. In addition, it must be ensured that movement of this larger constituent is to a position below all matrix adverbs, since (32a) cannot mean ‘for a long time, Jan wanted to read the book often’. Furthermore, I argued in section 4.3.1 that the operation of VR itself cannot be taken to be head movement but must be analyzed as involving XP-movement of a larger constituent. I thus propose that the embedded AspP, which is essentially the VP plus the potentially empty inﬁnitival marker, is moved into [Spec,CP] of the inﬁnitival, after which process the remaining TP of the embedded clause is moved into [Spec,PredP] of the matrix verb. After the embedded TP and the embedded AspP have undergone XP-movement, the inﬁnitival Tense-head adjoins to the functional

108

SCRAMBLING , REMNANT MOVEMENT , AND RESTRUCTURING IN WEST GERMANIC

head that licenses the controller of PRO to ensure the identiﬁcation of the inﬁnitival subject (to be discussed in chapter 5). The inﬁnitival verb undergoes additional XP-movement into a Speciﬁer position of the selecting verb, accounting for the so-called IPP-eﬀect. This movement applies in both German and Dutch verb clusters, which, however, crucially diﬀer in the way they spell out this movement chain: While German spells out the higher copy, the lower copy is spelled out in Dutch. In section 4.2, I have argued that nominal arguments (and VP-internal predicates) leave the VP in German, for reasons of licensing before Spell-out. The same holds for Dutch. The structure that results from this step of operation is given in (32c). In the next step, AspP that has been emptied up to the verb is moved into [Spec,CP] of the inﬁnitival (and then moved on into a position that precedes the matrix verb but is spelled out in [Spec,CP]). The result of this operation is shown in (32d). In the following step, the remaining TP of the inﬁnitival is moved into [Spec,PredP] of the matrix verb. The resulting structure is given in (32e). In the ﬁnal step, both the matrix subject and the embedded direct object scramble to positions above the sentential adverb vaak. This last step is optional. Hence both dat vaak Jan het boek lang wil lezen and dat Jan vaak het boek lang wil lezen are ﬁne sentences in Dutch. If we replace the adverb often with sentence negation in (32), then scrambling of both arguments becomes obligatory (in the absence of any contrastive focus) as we expect (cf. the discussion of scrambling in chapter 2). Scrambling of embedded arguments is enabled by extraction of the TP out of the embedded CP. (32) Step 1: licensing movement in the embedded clause c. [dat vaak Jan [VP wil [CP [TP PRO het boek lang [AspP [VP lezen]]]]]] Step 2: AspP moves into [Spec,CP] in the embedded clause d. [dat vaak Jan wil [CP [AspP lezen] [TP PRO het boek lang tAspP]]] Step 3: TP moves into [Spec,PredP] in the matrix clause e. [dat vaak Jan [PredP [TP PRO het boek lang t AspP] wil [CP lezen tTP]]] Step 4: scrambling of the matrix subject and the embedded object f. [dat Jani het boekj vaak ti [PredP [TP PRO tj lang] wil lezen]]

This account has several advantages. (1) It provides a simple and natural explanation of the position and interpretation of adverbs in VR-structures. An adverb that modiﬁes the verb in a coherent inﬁnitive is moved along with the remainder (after AspP-movement) of the inﬁnitival to a position below the adverbs in the matrix clause. An adverb that precedes a verb cluster is ambiguous between modifying the embedded verb or the matrix verb because it can be analyzed as occupying the embedded TP or as occupying the matrix TP. No scrambling of adverbs has to be assumed. (2) It provides us with a simple explanation for the transparency of coherent inﬁnitives. Due to TP-movement, arguments of the embedded clause (including quantiﬁers) may freely undergo A-movement into the matrix clause. For instance, (33a) is ambiguous between the readings speciﬁed in (33b–c), because the negative quantiﬁer niemand (‘nobody’) can be analyzed as occupying the matrix or the embedded TP, as is illustrated in (34).

A REMNANT MOVEMENT ACCOUNT OF RESTRUCTURING

109

(33) a. weil Hans niemanden besuchen will since Hans nobody visit wants b. ‘since Hans does not want to visit anyone’ c. ‘since Hans wants to visit nobody’ (34) a. weil [TP1 Hans niemandeni [TP2 ti] besuchen will] b. weil [TP1 Hans] [TP2 niemanden] besuchen will]

I have shown in chapter 2 that scrambling proper, being A-movement, is clausebound in German. Long-distance scrambling has properties of A-bar movement (cf. Mahajan [1990]). I assume that scrambling is conﬁned to the local CP. A DP may only move out of the local CP if it undergoes A-bar movement. What is special with coherent inﬁnitives is that TP extraction out of the embedded CP “merges” the scrambling domains of matrix clause and embedded clause. In coherent inﬁnitives the CPcomplement is not licensed as an argument but like small clauses as a co-predicate of the matrix verb by moving the embedded TP into [Spec,PredP] of the latter. If it is correct that predicates allow for extraction via A-movement (while arguments generally allow only for extraction via A'-movement), then it follows that the arguments of the matrix verb and the embedded inﬁnitive (can) share the functional domain above PredP in the matrix clause. This is my account of the so-called clause-union phenomenon, namely, the observation that in coherent inﬁnitives the arguments of the embedded inﬁnitival “seem” to be licensed in the matrix clause. It has been suggested that in coherent inﬁnitives VR-verb and dependent inﬁnitive form a complex predicate whose argument structure is derived by merging the argument structures of the combining predicates. Under these assumptions, coherent inﬁnitives are analyzed as monoclausal structures. I argue that the clause-union phenomenon can be simply accounted for by assuming that due to TP-movement part of the functional domain of the matrix clause can be shared by two (independent) predicates that project their own clause. That arguments of coherent inﬁnitives can stay in the embedded clause and need not obligatorily move into the matrix clause is shown in (35a), where the adverb öfter (‘more often’) can have narrow scope with respect to the matrix verb. In this reading the adverb must occupy a position in the embedded clause, implying that the argument eine Frau (‘a woman’) that follows it is contained in the embedded clause as well. This interpretation of the facts in (35a) is corroborated by the observation that in the narrow scope interpretation of the adverb in (35b) the negative quantiﬁer that follows it is no longer ambiguous, that is to say, it cannot take scope over the matrix verb. (35) a. weil Hans öfter eine Frau besuchen will since Hans more-often a woman visit wants ‘since Hans wants to more often meet some woman’ b. weil Hans öfter niemanden besuchen will since Hans more-often nobody visit wants ‘since Hans wants to more often visit nobody’

110

SCRAMBLING , REMNANT MOVEMENT , AND RESTRUCTURING IN WEST GERMANIC

(3) It gives us for free the right-branching structure of Dutch verb clusters without making use of right-adjunction. I assume that verb-particles in Dutch are not licensed via incorporation but by XP-movement to [Spec,AspP] or to [Spec,PredP] if modiﬁed. Thus, the cases in which a to-inﬁnitive has been raised with its particle that are so problematic for the standard theory (cf. [26] earlier) fall in place nicely, as the analysis in (36a) shows. (36) a. dat Jan [Marie]TP probeerde [CP [AspP [PP op] te [VP bellen tPP]] tTP] that Jan Marie tried up to call ‘that Jan tried to call up Marie’ b. *dat ze de schuur hebben rood over geschildert that they the barn have red over painted c. *dat ze de schuur hebben rood willen over schilderen that they the barn have red want-IPP over paint ‘that they have wanted to paint up the barn red’

Also, the distribution of Dutch particles can be accounted for by assuming that they can move to the higher [Spec,AspP] within the verb cluster. This does not hold for particles and small clause predicate heads if they are modiﬁed, since material that is licensed in PredP will be moved by TP-movement into the higher PredP and will thus always end up in front of the matrix verb. This will derive the ungrammaticality of the example (37d) in the previous chapter (repeated here as [36b]) and the generalization that particles prevent the incorporation of other predicate heads. The examples in (36b) and (36c) are ungrammatical because the small clause predicate contained within the particle phrase will be moved along with the particle in its licensing movement to the local [Spec,AspP]. In this position, the small clause predicate, though unmodiﬁed, can only be licensed in [Spec,PredP] and will thus always end up in front of the matrix verb. The alternative derivation, where the small clause predicate extracts from the particle phrase to be licensed in [Spec,AspP] followed by licensing movement of the particle phrase to PredP, can be ruled out as an illicit case of remnant movement, given that the containing phrase, the particle phrase, arguably checks the same feature as the small clause predicate (the Aspect-head in Dutch attracts particles and small clause predicates alike). As we have seen in the introductory section on remnant movement, this case can be excluded as a violation of Attract Closest or, in more standard terms, as a violation of the A-over-A condition. (4) While at ﬁrst sight it seems that TP-movement into the domain of the selecting verb is an artefact of a VO-based analysis, it can be shown that an OV-based account needs to assume TP-movement as well. Given that both accounts have to assume the same type of movements, we may then say that the VO-based approach is to be preferred over the OV-based approach since the former, but not the latter, can do without rightward movement. At ﬁrst sight it seems that an OV-based approach that is endowed with our insights is to be preferred. Looking at a two-member verb cluster, we might assume that within an OV-based approach right-adjunction of the embedded AspP to the

A REMNANT MOVEMENT ACCOUNT OF RESTRUCTURING

111

matrix verb (to explain verb cluster formation) and head movement of the inﬁnitival Tense-head (to explain the transparency of the inﬁnitival complement) are suﬃcient. Note that the latter assumption is probably suﬃcient to allow for the local scrambling of the arguments of the dependent inﬁnitive into the matrix domain while adverbs and other elements that do not scramble are already in the right position, preceding the selecting verb and lower than the other material in the clause of the selecting verb. However, once we look at bigger verb clusters we see that TP-movement is essential. Let us look at a three-member verb cluster in which the deepest embedded verb is modiﬁed by an adverb. The underlying structure within an OV-based account of such a construction is given in (37). The only possible surface order of the verb cluster in such a construction in Dutch is V1 V2 V3. In order to derive this cluster, we would assume that AspP of V3 right-adjoins to V2 and that AspP of V2 right-adjoins to V1. Note, however, that the second step would necessarily pied-pipe the adverb in the inﬁnitival TP of V3. Thus, it is essential that the lowest TP is moved out of the domain of V2 before verb cluster formation aﬀects AspP of V2. I conclude that an OV-based account has to assume the same cyclic movement operations per clause, namely, movement of AspP and movement of TP, that I was compelled to assume within a VO-based approach. (37)

[CP1 . . . [AspP [CP2 . . . [AspP [CP3 [TP3 Adv [AspP V3]]] V2 ]] V1]]

To summarize, TP-movement is independent of whether a VO-based or an OVbased approach is taken. Empirically, the existence of TP-movement is implied not only by the interpretation of quantiﬁers in VPR-constructions, as I will argue in the next section, but also by the distribution and interpretation of adverbs in VR-constructions. (5) It provides a simple and elegant solution to the distribution of CP-complements in VR-structures without making use of the operation of extraposition. CP-complements are licensed within AspP in Dutch and German. Thus, they are pied-piped by AspP-movement into [Spec,CP] of the inﬁnitival. This yields the eﬀect that while all other arguments of a coherent inﬁnitival show up in positions to the left of the matrix verb, the CP-complement stays with the verb that selects it and appears in a position to the right of the matrix verb (38). (38) dat Jan aan Marie morgen wilde [AspP 0 [F2P zeggen [dat Piet ziek is]]] that Jan to Marie tomorrow wanted say that Piet sick is ‘that Jan wanted to say to Marie tomorrow that Piet is sick’

(6) It provides us with a simple explanation for the diﬀerence in word order between German and Dutch verb clusters. In German, the dependent inﬁnitive generally precedes the VR-verb. After AspP-movement into [Spec,CP] the inﬁnitive undergoes additional movement that brings it to a left-adjacent position to the matrix verb, leaving behind its CP-complement to the right of the matrix verb (compare [38] with its German counterpart in [39]). In Dutch, contrary to German, the lower

SCRAMBLING , REMNANT MOVEMENT , AND RESTRUCTURING IN WEST GERMANIC

112

copy of the inﬁnitival will be spelled out. Whether this additional movement of the inﬁnitive is to be characterized as XP-movement or head movement will be discussed in chapter 6. There I will also provide the motivation for the additional movement of the inﬁnitive. (39)

dass Hans der Maria morgen [sagen]i wollte [AspP ti dass Peter krank ist] that Hans Maria-DAT tomorrow say wanted that Peter sick is

(7) It provides us with a very straightforward explanation for the so-called IPPeﬀect, which will be given in chapter 6. 4.3.3

A VO-based account of verb projection raising

In this section, I will discuss the syntax of VPR-complements in a VO-based approach. I will ﬁrst brieﬂy describe again the syntax of this construction in terms of the basic assumptions of the standard approach. Then I will discuss two problems concerning the proper account of scopal relations within VPR-constructions and show that they are intrinsically related to the assumption that VPR-constructions involve extraposition and scrambling. I will show that once these assumptions are given up, the problems disappear. Finally, I will demonstrate that the scopal facts in VPR-constructions fall out naturally if we extend the alternative account of VR-constructions to VPR-constructions. 4.3.3.1

The OV-based account and its problems

VPR in West Flemish, a Dutch dialect spoken in the West Flanders province of Belgium, applies to those verbs that trigger VR in standard Dutch. Recall from section 3.2 that in this dialect not only the nonﬁnite verb embedded under a modal (40a) but also the verb plus its direct object (40b) or the verb plus the direct and the indirect object (40c) can be raised. Den Besten and Edmondson (1983) have argued that these facts should be accounted for by assuming that in such a dialect not only V0 may be raised but also V' or VP, whence the name Verb Projection Raising (VPR). (40) a. da Marie Jan nen boek tR wilt [geven]R that Maria Jan a book wants give b. da Marie Jan tR wilt [nen boek geven]R that Marie Jan wants a book give c. da Marie tR wilt [Jan nen boek geven]R that Marie wants Jan a book give ‘that Marie wants to give Jan a book’

In the standard theory, the VPR-structures in (40) are described in terms of extraposition of the inﬁnitival clause, which, as I discussed in chapter 3 (section 3.2), must be at least as big as an IP/TP. Under these assumptions, a sentence like (41a) is analyzed as illustrated in (41b) (cf. Haegeman [1992]).

A REMNANT MOVEMENT ACCOUNT OF RESTRUCTURING

113

(41) a. da Valere wilt [ze morgen nen boek geven] that Valere wants her tomorrow a book give ‘that Valere wants to give her a book tomorrow’ b. da Valere tEXTR wilt [IP ze morgen nen boek geven]EXTR

Since elements of the inﬁnitival clause can appear in the matrix clause (cf. [40a– b]), it has to be assumed that VPR-complements are transparent and that scrambling may aﬀect their constituents and move them into the matrix clause (before extraposition applies). This operation of scrambling since it moves elements out of the local TP has to be characterized as A' movement (or LD scrambling). Recall from chapter 3 (section 3.2) that VPR-complements are also transparent for scope-taking elements but only if they are not aﬀected by VPR, that is, by extraposition in the standard analysis given earlier. The empirical generalization is the following: A scopal element that is outside of the verb cluster (i.e., that has undergone scrambling) can take its scope inside or outside of the verb cluster, whereas an element contained in the verb cluster may take its scope only within the verb cluster. Let us look at the examples in (42). In (42a), the negative existential occurs outside of the verb cluster. Example (42a) is ambiguous between the reading Jan did not force Valere to eat meat yesterday, with geen vlees taking wide scope, and the reading Jan forced Valere to not eat any meat yesterday, with geen vlees taking narrow scope. In (42b), the negative existential occurs inside of the verb cluster. What is crucial is that here the negative existential cannot take scope over the higher verb. Example (42b) is only compatible with the reading that Jan forced Valere to not eat any meat yesterday. (42) a. da Jan Valere geen vlees deeg gisteren eten that Jan Valere no meat made yesterday eat b. da Jan Valere gisteren deeg geen vlees eten

This is almost a paradoxical situation. A VPR-complement must be transparent to permit the LD scrambling of the negative existential into the middle ﬁeld of the matrix clause in (42a) but must be opaque to prevent LD scrambling or QR (another A' movement operation that in particular is thought to apply at LF) of the negative existential in (42b). In the next section, we will see that this paradox disappears once we apply the alternative account of VR to VPR-structures. Before we do that, let me point out another problem intrinsic to the scrambling analysis within the standard account. I noted earlier that an element that has undergone “scrambling” into the matrix clause may take wide or narrow scope with respect to the matrix verb. It is important to note that such an element in its narrow scope reading always takes wide scope with respect to material that has undergone “VPR.” So, for instance, (43a) in its narrow scope reading cannot mean that ‘Jan made Valere three times read two books (possibly diﬀerent ones at each occasion)’. Since in the standard approach the narrow scope reading is derived by reconstructing the scrambled element into its base position, it is hard to see how the preceding reading can be excluded.7 In other words, it is not clear what excludes a derivation of (43a), as illustrated in (43b).

SCRAMBLING , REMNANT MOVEMENT , AND RESTRUCTURING IN WEST GERMANIC

114

(43) a. da Jan Valere twee boeken deeg drie keers lezen that Jan Valere two books made three times read b. da Jan Valere [twee boeken]i tExtr deeg [drie keers ti lezen]Extr

4.3.3.2

The alternative account

In this section, I will argue that VR- and VPR-constructions are basically the same, diﬀering only in the amount of structure that can be pied-piped by AspP-movement into [Spec,CP] of the inﬁnitival. There is evidence that VPR-structures also involve movement of the inﬁnitival TP and movement of the inﬁnitival (extended) VP (= AspP). This evidence comes from the behavior of adverbs and adverbials in VPR. The adverb that occurs in the matrix domain in (44) can be interpreted as modifying the dependent inﬁnitive. From my basic assumption that adverbs cannot undergo scrambling it follows that the adverb in (44) must have been pied-piped by the movement of a larger constituent. I have argued in section 4.3.2 that this larger constituent in VR-structures is the TP of the inﬁnitival clause. Hence we can safely conclude that the adverb in (44) provides direct evidence for TP-movement in VPR-structures. The VPR-structure in (44) diﬀers from a typical VR-structure only in the amount of structure that undergoes movement into [Spec,CP] of the inﬁnitival. In (44), a constituent larger than AspP has been moved into [Spec,CP]. The amount of structure that can be pied-piped by AspP-movement into [Spec,CP] seems to be variable, as is indicated by the diﬀerent size of the bracketed material in (40). (44) a. da Valere drie keers deeg da boek lezen that Jan Valere three times made that book read ‘that on three occasions Jan made Valere read that book’ ‘that Jan made Valere read the book three times’ b. da Jan Valere deeg drie keers da boek lezen that Jan Valere made three times that book read ‘that Jan made Valere read that book three times’

In this account the transparency/opacity paradox that I discussed in the previous section disappears. We can now dismiss this paradox as an artefact of a false scrambling analysis. Remember that the problem that the scrambling and extraposition analysis of structures like (44a) posed was the question of why it is that the adverb can undergo LD scrambling, that is, A' movement, into the matrix domain in sentences like (44a) before Spell-out, while it cannot undergo scrambling or QR into the matrix domain at LF if it is aﬀected by VPR, as is illustrated in (44b). Since in the alternative account the adverb in (44a) does not scramble into the matrix domain but is moved there via TP-movement, the question of whether VPR-complements are transparent does not even arise. Moreover, this alternative account also explains why material that appears in the matrix domain in its narrow scope reading with respect to the matrix verb necessarily has to take scope over all the material in the

A REMNANT MOVEMENT ACCOUNT OF RESTRUCTURING

115

embedded clause, as I pointed out in my discussion of (43) in the previous section. As we will see later, this fact follows naturally from the mechanics of the system. In section 4.2, I argued that what was traditionally called scrambling in the West Germanic languages actually breaks down into two types of movement operations: an operation that obligatorily moves all DPs out of the VP into their licensing positions above manner adverbs and scrambling proper, which moves DPs to positions above frequency adverbs like often or n-times. In chapter 2, I discussed how scrambling aﬀects the interpretation of the involved DP in the following way: A DP that is scrambled across such an adverb has necessarily wide scope with respect to that adverb but has necessarily narrow scope with respect to that adverb if it fails to undergo scrambling. If we assume that it is possible to pied-pipe the phrase that corresponds to the domain that contains these adverbs with AspP-movement into [Spec,CP], then it follows without stipulation that the DP in (43) necessarily has wide scope over the adverb in the VPR-complement: In more simple terms, in order for an element to move into the domain of the selecting verb via TP-movement, it has to take scope over the elements that stay behind and are pied-piped by movement of AspP, as is illustrated in (45). In (45), PPD stands for pied-piped domain, that is, the domain that has been pied-piped by movement of AspP. (45)

da Jan Valere [PredP [VP deeg [CP [TP [twee boeken]i [PPD drie keers ti lezen]]]]]

4.3.3.3

Conclusions

To summarize, I have provided a VO-based approach to VR that allows us to assume that all inﬁnitival complements are CPs and does not employ rightward movement or unmotivated scrambling operations (of adverbs and predicates). Instead it employs movement of the inﬁnitival AspP into [Spec,CP] of the inﬁnitival clause and movement of the inﬁnitival TP into [Spec,PredP] of the selecting verb. These movements are systematic and cyclic; that is, they take place in every inﬁnitival clause in the same fashion. These movements occur for licensing purposes, as will be argued in section 4.4 later. The remaining movements of arguments are to be characterized as scrambling proper and occur for the same reasons that motivate scrambling of arguments in simple clauses. Finally, we have seen that VPR-structures and their (scopal) properties can be accounted for in the same manner, if it is assumed that in VPR-structures movement of the inﬁnitival VP can optionally pied-pipe larger phrases. The main argument that I provided for the account sketched earlier was that the distribution of DP-arguments, adverbs, and predicates in restructuring inﬁnitives cannot be accounted for in terms of scrambling. At this point, the question arises of whether VR-constructions could not be derived in a simpler way in a VO-based approach in terms of licensing movement—of those elements that we argued resist scrambling but precede the verb cluster. Such an account would assume that VRconstructions are monoclausal, implying that the licensing positions for VP-internal material of all the verbs in a verb cluster occur above the VP of the highest verb (cf. Zwart [1993]). I have two comments to make about this alternative. First, such an account is possible in principle but would forestall a uniﬁed account of VR- and VPR-

116

SCRAMBLING , REMNANT MOVEMENT , AND RESTRUCTURING IN WEST GERMANIC

constructions. In such an approach it must be assumed that the licensing positions for VP-internal material of the inﬁnitive optionally can be projected “upstairs” or “downstairs.” Second, within the monoclausal construal the question arises of how the licensing positions of the dependent inﬁnitive(s) are ordered with respect to the licensing positions of the matrix verb (and with respect to each other). In chapter 5, I will show that in VR-constructions the licensing positions of the dependent inﬁnitive occur lower than the licensing projections of the matrix verb, implying that there are two licensing domains present in these constructions, not just one. Furthermore, I will show in chapter 5 that the binding properties of coherent inﬁnitives also imply the presence of more than one licensing domain. In conclusion, an alternative monoclausal VO-based approach that employs only licensing movement of VP-internal material is both theoretically and empirically inferior to the account that I have provided earlier. 4.3.4

Implications of this account for the theory of remnant movement

In this account of restructuring, embedded inﬁnitives can no longer be analyzed as (possibly referential) remnant categories that contain scrambling traces. Thus, the illicit cases of remnant scrambling that I discussed in chapter 1 (section 1.4) can no longer be ruled out by Attract Closest. In this account, embedded inﬁnitives are simply inﬁnitival VPs that contain traces of licensing movement, among which are traces left behind by Case movement. In particular, I have to ﬁnd another explanation for the fact that these inﬁnitival VPs may topicalize but cannot undergo scrambling in the matrix clause as is illustrated again in (46). (46) a. [zu lesen] hat das Buch keiner versucht to read has the book nobody tried-PART ‘Nobody has tried to read the book’ b. ??dass [zu lesen] das Buch keiner versucht hat that to read the book nobody tried has ‘that nobody has tried to read the book’

My explanation of the contrast is rather simple. These inﬁnitival VPs are predicates and as such resist scrambling. This explanation is based on the assumption that [Spec,CP], the so-called topicalization position, is a multi-purpose position that can host focused phrases, which are not necessarily referential, as well as discourse topics and sentence topics. While discourse topics are necessarily referential (token topics), sentence topics can be predicates or so-called type topics. In chapter 2, I have argued that there are two types of triggers for scrambling in German: scope and speciﬁcity/familiarity. Thus, the assumption is natural that only quantiﬁcational expressions (for reasons of scope taking) and token topics (due to their speciﬁcity features) may scramble. In this approach, scrambling of predicates including VPs is ruled out under the assumption that the VP denotes an event-type (with the TP denoting an event-token). That the correct generalization is indeed un-

A REMNANT MOVEMENT ACCOUNT OF RESTRUCTURING

117

availability of predicates for scrambling is shown in (47). Example (47) shows that scrambling of a predicate is equally bad whether it is the entire predicate as in (47a) or only a remnant as in (47b) that undergoes scrambling. (47) a. ??dass [die Maria geliebt] Hans hat that the Maria loved Hans has b. ??dass [geliebt] Hans die Maria hat that loved-PART Hans the Maria has ‘that Hans has loved Maria’

This observation is corroborated by the following facts. Recall that S-scrambling may also apply to predicates, as is illustrated again by the contrast between (48a) and (48b). Example (48a) involves illicit scrambling of the adjectival predicate across the position of the direct object. Example (48b) involves S-scrambling of a contrastive topic to the top of the middle ﬁeld and is ﬁne. As (48c) shows, this type of scrambling is also possible for a remnant VP. (48) a. ??dass Hans grün den Zaun strich that Hans green the fence painted ‘that Hans painted the fence green’ b. dass so grün nur der Hans den Zaun streichen würde that so green only Hans the fence paint would ‘that only Hans could paint the fence so green’ c. dass so geliebt die Maria nur er hat that so loved the Maria only he has ‘that only he loved Maria to such a degree’

4.4 Toward a theory of sentential complementation In this section, I propose an account of sentential complementation (see also Boscovic 1993) that will allow us to derive the movement operations argued for in section 4.3. The basic idea is rather simple. I will argue that the complementizer is responsible for licensing both V and T in an embedded clause and if the complementizer is deﬁcient, both V and T have to be licensed in the matrix clause. But let us ﬁrst see how sentential complementation works in clauses with a non-deﬁcient complementizer. 4.4.1

Basic assumptions about sentential complementation

I assume that there is a licensing dependency between the matrix verb, the complementizer, and the morphological and temporal properties of the embedded verb. The complementizer acts as a placeholder for the c-selectional requirements of the matrix verb. A matrix verb may select a ﬁnite or nonﬁnite indicative or subjunctive complement.

118

SCRAMBLING , REMNANT MOVEMENT , AND RESTRUCTURING IN WEST GERMANIC

In addition, I assume that the complementizer is essential for rendering a sentential complement (a TP) into an argument. I propose that the complementizer nominalizes the clause and in this way qualiﬁes it as an argument of the higher verb. Given these assumptions, the complementizer dass (‘that’) in German can be assigned the morphological features [+ ﬁnite, + indicative] and the categorial feature [+N]. Following recent work on the split C-domain (cf. Rizzi [1997] and Haegeman [1996c]), I assume that the C-domain is made up of various functional projections. I assume that the complementizer is inserted in FinP, where it licenses the ﬁniteness of the clause and moves through a Mood Phrase where it licenses the clause as indicative or subjunctive (cf. Roussou [2000]), to Force, constituting the highest head in the C-domain. This is illustrated in (49), where it is assumed that the traditional CP corresponds to ForceP (CP = ForceP). Then the entire CP is moved into [Spec,F3P] of the selecting verb, as I have assumed earlier, to check the features of the complementizer against the c-selectional features of the matrix verb. (49)

[CP

Force [MoodP

M

[FinP

F

[TP . . .]]]]

dass

As I have stated earlier, I assume that restructuring verbs s-select a proposition. The canonical syntactic representation of a proposition is a CP. The c-selectional features of a CP-complement that I discussed earlier are satisﬁed or checked by licensing movement of the CP, given that C is speciﬁed (or valued) for these features. If the complementizer is deﬁcient, that is to say, is not speciﬁed for these features, then the relevant features will be checked by those parts of the embedded clause that are dependent on the complementizer by movement into their licensing positions in the matrix clause. In the following sections, I will argue that these parts are the inﬁnitival AspP and the inﬁnitival TP. This I assume is the reason that the inﬁnitival clause is split up in coherent constructions. To summarize, the preceding approach is based on the assumption that both V and T are licensed by C in embedded clauses. If the licenser is deﬁcient, then the licensee will move on to the next higher licensing position. This case can be compared to the Inﬂ-head in a raising construction that fails to license the embedded subject. The embedded subject will then undergo (further) licensing movement to a Case-checking position in the matrix clause. In the following, I will present some arguments to show that V and T in an embedded clause are indeed dependent on C. First, I will develop that the complementizer is essential for rendering a sentential complement (a TP) into an argument. I will argue that the complementizer plays a crucial role in linking the embedded TP to the matrix event time. (cf. Den Besten [1983], Enç [1987], Guéron and Hoekstra [1988], among others). In section 4.4.2, I discuss the temporal interpretation of inﬁnitives and argue that the event denoted by an inﬁnitive is not anchored with respect to a local Tense-predicate but is directly dependent in its interpretation on the matrix verb. This dependency between matrix verb and the inﬁnitival Tense-head is mediated by the complementizer. In section 4.4.3, I argue that the complementizer in general serves as a placeholder for the c-selectional properties of the matrix verb and thus becomes essential for licensing the inﬁnitival verb as well. Having established that there are two dependency relations, one between C and

A REMNANT MOVEMENT ACCOUNT OF RESTRUCTURING

119

T and one between C and V, I will discuss two ways in which these dependencies might be encoded syntactically. In section 4.4.4, I will sketch an account that assumes that these dependencies are established in terms of head movement of V and T to the licensing head C. There I will argue that T-C embodies a control relationship that when disrupted leads to overt raising of the inﬁnitival AspP and the inﬁnitival TP. In section 4.4.6, I will sketch an account that assumes that these dependencies are established in terms of XP-movement of the inﬁnitival AspP and the inﬁnitival TP into designated licensing positions in the C-layer. This will be the account that I adopt at the end, since it can best handle the complex data posed by VP-topicalization and CP-extraposition. 4.4.2

The temporal interpretation of inﬁnitives

In the following discussion of the temporal properties of inﬁnitives, I follow Reichenbach (1947) in assuming that Tense deﬁnes a relation between speaking time and reference time and Aspect a relation between reference time and event time. In syntactic terms, this means that the interpretation/licensing of the Aspect Phrase depends on the interpretation/licensing of the Tense Phrase. This will become important when I talk about the temporal licensing of restructuring inﬁnitives later. Since the seminal work by Enç (1986, 1987) on the temporal interpretation of linguistic expressions, we know that verbs must be temporally anchored. Since then much research has been conducted on this topic, which reveals that it is the role of Tense to anchor the event denoted by the verb with respect to the “speaking time” in matrix clauses and with respect to the speaking time or the “event time” of the matrix verb in ﬁnite complement clauses and relative clauses (see especially the work of Hornstein [1990], Stowell [1993], and Zagona [1990] for a detailed discussion of these issues). Inﬁnitival clauses do not contain a Tense-predicate. Formally, this is immediately obvious from the fact that inﬁnitives are not inﬂected for Tense. But also from an interpretative point of view, it is clear that inﬁnitives do not contain the abstract Tense-predicates (+/– past) that have been posited for the temporal interpretation of matrix clauses and ﬁnite embedded clauses. The temporal properties of inﬁnitives cannot be accounted for by employing these predicates. On the contrary, it seems that the temporal interpretation of an embedded inﬁnitive is highly dependent on semantic properties of the matrix verb. Some verbs that select to-inﬁnitives require a simultaneous interpretation of matrix event and embedded event, like the English verb try, while some verbs that select to-inﬁnitives require a posterior interpretation of the embedded event with respect to the matrix event. The English verb promise is an example for this class of verbs. Most verbs that select bare inﬁnitives require a simultaneous interpretation of the matrix event and the embedded event, while modal verbs that also select bare inﬁnitives often require a future interpretation of the embedded event with respect to the matrix event.8 Thus, it seems that the temporal interpretation of an inﬁnitival complement does not correlate with its syntactic property of being a bare inﬁnitive or a to-inﬁnitive. However, if we look at the semantics of the German verbs versprechen (‘promise’) and wollen (‘want’), the former selecting a to-inﬁnitive, the latter selecting a bare inﬁnitive, it becomes clear why their inﬁnitival complements are temporally interpreted alike: One promises to do something in the future as much as one can want to do something in the future.

120

SCRAMBLING , REMNANT MOVEMENT , AND RESTRUCTURING IN WEST GERMANIC

Pesetsky (1992) notes that under a non-habitual reading, eventive predicates can be embedded under control predicates but not under ECM-predicates and cites the contrast in (50) (cf. also Martin [2001]). This contrast follows, so he argues, if eventive predicates contain a temporal argument that must be bound by Tense. He follows Stowell (1982), who proposes that ECM-inﬁnitives, in contrast to control inﬁnitives, are not speciﬁed for Tense. Stowell (1982) argues that the Tense in control inﬁnitives denotes a possible future and that it lacks the feature [+/–past] and notes that the time frame of ECM-inﬁnitives is the same as the matrix time frame. (50) a. John tried to bring the beer b. *John believed Peter to bring the beer

Though this particular observation in (50) is valid, the conclusions that Pesetsky (1992) draws from it are too strong. First note that ECM-verbs that select bare inﬁnitives can embed eventive predicates (51a). Furthermore, there are ECM-verbs that select to-inﬁnitives that also allow eventive predicates, like expect and want. In the case of these verbs the time frame of the matrix also diﬀers from the time frame of the inﬁnitive, as is illustrated in (51b–c). (51) a. I saw John bring the beer b. I expected John to bring the beer c. I wanted John to come to the party tomorrow

Second, it is not clear what the Tense of control inﬁnitives should be. Gamon (1992) argues that control inﬁnitives have to be divided into three diﬀerent groups according to their temporal properties. He discusses the temporal properties of the control verbs remember, try, and promise in English. He notes that in the case of remember and promise the event of the embedded clause follows the event of the matrix clause (e2 > e1), while in the case of try the two events are contemporaneous (e2 = e1). Only remember and promise, as opposed to try, allow their complements to have a shifted (future) reading (52) and to contain a perfect predicate (53).9 (52) a. At six o’clock, John remembered/promised to take his medicine at seven o’clock b. *At six o’clock, John tried to take his medicine at seven o’clock (53) a. ?John remembered to have done his work (e2 < e1) b. John promised to have done his work (e2 > e1) c. *John tried to have done his work

Gamon notes, however, that the perfect complements of remember and promise diﬀer in their temporal interpretion as is indicated in (53): The perfective complement in (53b) has a future-perfect reading, while the perfective complement in (53a) can only have a past-shifted reading with respect to the matrix event. Although Gamon’s observations are basically correct, I think that he confounds two meanings

A REMNANT MOVEMENT ACCOUNT OF RESTRUCTURING

121

of remember in (52) and (53). One can remember a fact or a certain state of aﬀairs or one can remember one’s obligation or intention to do something. In the latter meaning the inﬁnitival complement has modal force and its temporal properties are exactly like the temporal properties of the inﬁnitival complement of promise in German. The inﬁnitival complement of remember in its ﬁrst (non-modal) reading behaves like the inﬁnitival complement of the ECM-verb believe, which allows for a perfect complement that can only have a past-shifted reading (54a). (54) a. I believed John to have brought the beer b. I believed John to be bringing the beer

I argue that according to their temporal properties, inﬁnitival complements fall into two categories: the promise-category and the try-category, with the properties as discussed earlier. The ECM-verb believe and the verb remember in its non-modal meaning belong to the try-category. The basic interpretation of their inﬁnitival complements is a simultaneous reading with the matrix event, as is shown for the ECMverb believe in (54b). Try, like many ECM-verbs, does not allow a perfective inﬁnitival complement. I believe that this fact does not need to be accounted for in syntactic terms but directly follows from what it means to try to do something. The attempt and the attempted event necessarily occur simultaneously. Similarly, it follows from the meanings of these verbs that the ECM-verb believe allows for a perfective complement while the control verb remember in the relevant sense requires a perfective complement since one cannot remember what has not yet happened. That the ECMverb believe seems to select a stative complement (cf. [50b] and [54b]) I consider an idiosyncratic property of this verb. According to Pesetsky’s (1992) conjecture about Tense in inﬁnitives, the complement of a control verb should not behave like the complement of an ECM-verb. However, if the temporal properties of inﬁnitival complements directly follow from the semantics of the verbs that select them, as I assume, then the parallelism in the interpretation of (53a) and (54a) is not surprising. Again, you may remember that a certain event happened in the past and you may also believe that a certain event happened in the past. In this book, I cannot undertake the interesting task of further exploring the exact nature of the interaction between certain semantic properties of verbs and the temporal properties of their inﬁnitival complements. What I have tried to argue for is that the temporal properties of inﬁnitival complements follow from semantic properties rather than from syntactic distinctions like control inﬁnitive versus ECM inﬁnitive or to-inﬁnitive versus bare inﬁnitive. To summarize, I conclude that the event denoted by an inﬁnitive is not anchored by Tense but directly dependent on the matrix verb. I assume that verbs may select two diﬀerent types of inﬁnitival complements, the try-category and the promise-category. Though an event denoted by an inﬁnitive is not directly dependent on a Tensepredicate, it is clear that inﬁnitival complements do contain a syntactic Tense-head. This head is relevant for licensing PRO, clitics, and temporal adverbs. This head, being semantically empty must be identiﬁed somehow in order to meet the Condition of Full Interpretation. In section 4.4.4, I propose that the temporal dependency of an inﬁnitive on the matrix verb is syntactically encoded by a control relation that is also responsible

122

SCRAMBLING , REMNANT MOVEMENT , AND RESTRUCTURING IN WEST GERMANIC

for the identiﬁcation of the inﬁnitival Tense-head. Before I do that I want to discuss another condition of the matrix verb that seems to be encoded in C. 4.4.3

Morphological selection

A complementizer not only serves as a placeholder for properties like ﬁniteness or modality that can be taken to be selected by the matrix verb but also represents features that deﬁne the morphological type of the selected inﬁnitive that cannot be derived from the semantic properties of the matrix verb. For instance, whether a verb selects a bare inﬁnitive or a to-inﬁnitive is arbitrary and cannot be determined on semantic grounds. In German, the verb wünschen (‘wish’) selects a to-inﬁnitive and the verb wollen (‘want’) selects a bare inﬁnitive. I do not see how this can be made to follow from a semantic distinction between the two verbs, since there does not seem to be any. So complementizers also represent next to the ﬁniteness feature and the modal feature discussed earlier, features pertaining to the morphological category of the inﬁnitival selected by the matrix verb. I propose that movement of the inﬁnitival VP to the selecting verb is related to a deﬁcient complementizer that lacks this type of morphological feature. 4.4.4

An X0-movement account of sentential complementation

In this section, I propose that the temporal dependency of an inﬁnitive on the matrix verb is syntactically encoded by a control relation, which ensures that a temporal index is assigned to the inﬁnitival Tense-head. This control relation is mediated by C0 in the following way: I propose that control is a (local) coindexing relation between heads (see also chapter 5 for further discussion of control in inﬁnitives). This mechanism ensures that the matrix event time is assigned to C0 in the inﬁnitival clause. I have argued earlier that a matrix verb may select two diﬀerent types of inﬁnitival complements: the try-category and the promise-category. I propose that this is syntactically encoded by allowing the matrix verb to select one out of two diﬀerent types of complementizers: A verb may select a complementizer with a [+modal] or with a [–modal] feature. A complementizer with the feature [+modal] will assign a temporal index to T0 that follows the matrix event time assigned to C0 by control, yielding the interpretation that the embedded event follows the matrix event, disregarding the aspectual information in the inﬁnitival clause (e2 > e1). A complementizer with the feature [–modal] will assign a temporal index to T0 that overlaps with the temporal index assigned to C0 by control, yielding the interpretation that matrix event and embedded event occur (partially) simultaneously (e2 = e1). The assignment of a temporal index by the complementizer to the inﬁnitival Tense-head requires that the inﬁnitival Tense-head enters into a checking relation with the complementizer. I propose that this checking relation is normally established via head movement of T to C. The event denoted by the inﬁnitive is then anchored to the index assigned to T0 in the inﬁnitival complement (by a dependency that involves the Tense-head and the Aspect-head). 4.4.5

The licensing of VR- and VPR-complements

I propose that in VR- and VPR-inﬁnitives the control relation is disrupted due to a defective complementizer and that the temporal dependency between matrix verb and inﬁnitive is ﬁxed by a movement relation. If the inﬁnitival complementizer can-

A REMNANT MOVEMENT ACCOUNT OF RESTRUCTURING

123

not transmit a temporal index to the inﬁnitival Tense-head due to its deﬁciency, the inﬁnitival T0 becomes inert and the event denoted by the inﬁnitive must be directly anchored to the matrix verb by moving the inﬁnitival VP (its Aspect Phrase) into [Spec,CP] of the inﬁnitive and further up into the licensing domain of the selecting verb. In the licensing domain of the matrix verb, the inﬁnitival Aspect Phrase is then assigned a temporal index. Since with inﬁnitives the temporal relation between matrix verb and embedded verb follows from the semantic properties of the selecting verb, as I have argued in section 4.4.2, it is suﬃcient that the inﬁnitival AspP simply enters into a checking relation with the selecting verb. To summarize, the inﬁnitival event, denoted by the inﬁnitival AspP, is normally temporally located via the inﬁnitival TP. In restructuring inﬁnitives, due to the deﬁciency of the complementizer, the inﬁnitival TP fails to specify a temporal index for the local AspP. Hence the inﬁnitival AspP undergoes movement into the domain of the selecting verb to be linked to the matrix event time directly. The rationale behind this scenario is the following idea: Next to the control relation in the nominal domain, there is also a control relation in the temporal domain in inﬁnitival clauses. If the control relation is disrupted due to a deﬁcient complementizer, the control relation is ﬁxed by raising the temporal argument of the inﬁnitival clause, that is, the inﬁnitival AspP, into a checking position of the temporal licenser, that is, the selecting verb. The nature of this functional head in the V-domain that is responsible for temporal licensing of inﬁnitival verbs will be discussed in chapter 6, where the formation and structure of verb clusters is dealt with in detail. Movement of the inﬁnitival VP into [Spec,CP] is necessary in order to license the deﬁcient complementizer. I propose that complementizers have either an [N] (most embedded clauses) or a [V] feature (matrix clauses and embedded V2–clauses). The [N] feature is checked by the insertion of a complementizer. If the complementizer is absent or deﬁcient, C must be licensed with a [V] feature via verb movement. If the inﬁnitival TP is not assigned a temporal index, it fails to denote an eventtoken (with the VP denoting an event-type) and thus will not qualify as an argument of the selecting verb and will have to restructure as predicate by moving into [Spec,PredP] of the selecting verb. As I have argued at the end of section 4.3, VPR diﬀers from VR only in the amount of structure that can be pied-piped by movement of the inﬁnitival AspP into [Spec,CP] of the inﬁnitival clause. One ﬁnal comment is in order: If the abstract dependencies between C and T, and C and V, were syntactically encoded via head movement, then it would follow from considerations of locality that there must be head movement of V to T in German at some point in the derivation and that there is no direct relationship between V and C in embedded clauses. 4.4.6

An XP-movement account of sentential complementation

The XP-movement account, however, assumes that there is such a direct relationship between V and C (unmediated by T). This relation involves the morphological properties of the dependent verb selected by the matrix verb. In section 4.4.2, I argued that it is an idiosyncratic feature of the matrix verb whether it selects for a to-inﬁnitive or a bare inﬁnitive. I want to generalize this ob-

124

SCRAMBLING , REMNANT MOVEMENT , AND RESTRUCTURING IN WEST GERMANIC

servation and assume, following Bech (1955/1983), that the matrix verb may select for the status of its sentential complement. Bech distinguished three statuses of the nonﬁnite verb. The ﬁrst status is the bare inﬁnitive. The second status is the toinﬁnitive, and the third status represents the participle of the verb. We can subsume all diﬀerent forms of the verb in one paradigm if we treat ﬁnite verbs as representing zero-status. In this manner, Rizzi’s (1997) ﬁniteness-feature in C can be related to a lexical entity, namely, the paradigm of the diﬀerent morphological categories of the verb. Bech conceived of the diﬀerent morphological categories of the verb as verbal “Cases.” If Cases need to be licensed, then also the speciﬁc verbal form selected by the higher verb needs to be licensed. I propose, following Rizzi (1997), that this inﬂectional category should receive independent syntactic status. I also propose, generalizing Rizzi’s FinP, that the verb form, its status, is checked by movement of the extended VP (the Aspect Phrase) into the Status Phrase in the C-layer of the clause, as illustrated in (55). (55) a. [CP C b. [CP C

[SP S(tatus)

[TP T [AspP V]]]]

[SP AspP

S(tatus)

checking of the selected status →

[TP T tAspP]]]

In section 4.4.2, I have argued that the temporal interpretation of an inﬁnitival clause depends on the modality of the complementizer selected by the matrix verb. I have assumed that there are [+/– modal] complementizers that can modify the temporal interpretation of the inﬁnitival Tense-head. Now I would like to propose that the modal feature of complementizers is projected in a separate functional head, Mood, in the C-layer (cf. Roussou [2000]). Furthermore, I assume that the modality of a complement is checked by TPmovement into [Spec,MoodP], as illustrated in (56). TP- and AspP-movement could be taken to occur in embedded as well as in matrix clauses, modulo V2, that is, head movement of the verb to an empty C head. In matrix clauses, Mood and Status are not determined by a higher verb but by the (speech-act intentions of the) speaker and the direct dependencies that exist between Mood and Status (cf. the imperative mood requiring speciﬁc ﬁnite/nonﬁnite forms in the languages of the world). (56) a. [CP C [MP Mood [SP S(tatus) [TP T [AspP V]]]]]

Status Checking →

b. [CP C [MP Mood [SP AspP S(tatus) [TP T tAspP]]]] Temporal Anchoring → c. [CP C [MP TP Mood [SP AspP S(tatus) tTP]]]

As outlined earlier, the complementizer is inserted into the head position of the Status Phrase (the former FinP) and checks the morphological form of the dependent verb, its ﬁnite or nonﬁnite status. Then it moves on to the head of the Mood Phrase to check the modal category of the embedded clause. Depending on the value of the Moodhead, the embedded TP in its Speciﬁer will be assigned a value for the reference time that either equals or follows the matrix event time. Finally, the complementizer moves into the highest head position in the C-domain, possibly to check the force of the clause. For illustration consider (57) below in conjunction with (49) earlier.

A REMNANT MOVEMENT ACCOUNT OF RESTRUCTURING

(57)

125

licensing movements in an embedded clause

MoodP Force StatP M TP S

AspP

dass

Under these assumptions, there is no stage in the derivation in which the verb would move to T. We could assume that the relationship between V and T is established indirectly. In embedded clauses, movement of the complementizer creates a chain between SP and MP, the Speciﬁers of which are occupied by AspP and TP, respectively. In matrix clauses, the ﬁnite verb can be taken to move from [Spec,SP] to the Mood-head, establishing a Spec-head relation between the ﬁnite verb and the TP. If this approach is correct, then Verb Second can be characterized as a very local operation that involves movement of M to C that pied-pipes the ﬁnite verb and the V2-property can be assumed to follow from a requirement that the highest Speciﬁer in the C-domain, [Spec,ForceP] must be ﬁlled with lexical material (phonological EPP), as is illustrated in (58). However, I will not venture into providing a full-ﬂedged account of V2 in a split C-domain. Restricting myself to the syntax of nonﬁnite verbs, I will leave this issue for further research. (58)

licensing movements in a V2-clause

F TP M AspP S Vfin e

Alternatively, we may assume that the inﬁnitival AspP moves through TP to check Tense on its way to [Spec,SP]. It remains to be seen whether this presumed type of VP-movement is the same as the type of VP-preposing that Kayne (1998) proposed for English in order to account for the distribution and interpretation of negative phrases. Hroarsdottir (2001) argues that verb movement to a clause-medial

126

SCRAMBLING , REMNANT MOVEMENT , AND RESTRUCTURING IN WEST GERMANIC

position was in fact VP-movement (to T or to a functional head just below T) throughout the history of Icelandic. Also, Haegeman (2001, 2002b) argues on the basis of the position of the negative marker in WF that the verb moves to Tense in West Germanic. In her account, though, the presumed movement operation is head movement. Given that rich morphology implies that German has overt V to T (cf. Vikner [1995]) and assuming, for the time being without argumentation, that this operation is XP-movement, I have to slightly change our assumptions about the licensing position of the inﬁnitival subject. Under these assumptions, I analyze PRO as occupying the Speciﬁer of an Agreement Phrase directly above TP rather than [Spec,TP] itself. The Tense-head would move to this Agreement-head to license PRO, and this Agreement Phrase would then move to [Spec,MP] in the CP-layer. In this account, the very same movements that I argued constitute restructuring, namely, movement of TP and AspP, already occur in the simple clause. Restructuring verbs then diﬀer from non-restructuring verbs only in that they fail to license the inﬁnitival TP and the inﬁnitival AspP in the embedded clause. The reason again is a deﬁcient complementizer that is not (fully) valued for the features checked in MP and SP. The deﬁciency of the complementizer can be thought of in two ways. First, it can be taken to fail to license the syntactic heads M(ood) and S(tatus), implying that the inﬁnitival TP and the inﬁnitival AspP cannot be licensed in the embedded clause (cf. Haegeman [2002a], and Starke and Cardinaletti [1999]) for diﬀerent applications of this idea). In this account, the inﬁnitival AspP ﬁrst moves into [Spec,ForceP] in order to license the deﬁcient complementizer and then into a licensing position of matrix verb to check its status, while the inﬁnitival TP, not denoting an event-token, moves into PredP of the matrix verb to be licensed as a predicate. Alternatively, if the comparison with a defective Inﬂ-head in raising constructions is indeed relevant, the syntactic heads Mood and Status can be taken to be projected but remain partially inert due to the failure of the complementizer to value them completely (cf. the essays in Mauck and Mittelstaedt [2002]). In this scenario, the inﬁnitival TP and AspP would just pass through these positions, which, with a non-defective complementizer, also constitute the ﬁnal Spell-out positions of TP and AspP in embedded clauses. This is illustrated in (59). (59)

PredP

Pred

FP(= functional position in the V-domain) CP = ForceP V C M TP S

0[defective]

AspP

A REMNANT MOVEMENT ACCOUNT OF RESTRUCTURING

127

We may assume that the Status-head just checks the nonﬁnite status of the embedded Aspect Phrase but fails to check whether it is a to-inﬁnitive, a bare inﬁnitive, or a participle, while the Mood-head can be taken to check whether the inﬁnitival clause is subjunctive or indicative but fails to assign a temporal index to the inﬁnitival TP.10 Thus, the inﬁnitival TP and the inﬁnitival AspP move on into their licensing positions in the matrix clause, as is illustrated in (59). In both versions, independently of whether we assume that the Mood Phrase and Status Phrase are projected or not, the defective CP cannot be taken to constitute a strong phase, since otherwise the inﬁnitival TP, not being in the left edge of the phase head, that is, [Spec,CP], could not move to [Spec,PredP] of the matrix verb. I will introduce the phase condition and its relevance for verb cluster formation and VPtopicalization in chapter 6. In the following, I will adopt the XP-movement account in the version where both MP and SP are projected, since as I will discuss later, it will allow us to provide a relatively simple account of CP-extraposition.

4.4.7

Generalized licensing

I have outlined an account of sentential complementation from which the movements necessary in restructuring contexts, namely, AspP- and TP-movement, fall out naturally. In this account, the movements of AspP and TP are not special movements anymore that just occur in restructuring contexts but are regular licensing movements that occur in a parallel fashion in every clause in German. Furthermore, the account sketched earlier allows us to treat the movements that occurr in restructuring constructions in a parallel way to the movement of subjects in raising constructions: The functional structure of the embedded clause is deﬁcient in a way that forces elements that depend on it to move to respective licensing positions in the matrix clause. All the relevant structures and movements are illustrated again in (60a–c).

(60) a. licensing movements in an embedded clause

C M TP S

dass

AspP

128

SCRAMBLING , REMNANT MOVEMENT , AND RESTRUCTURING IN WEST GERMANIC

b. licensing movements in a V2-clause

F TP M AspP S Vfin e

c. licensing movements in a restructuring clause

Pred

V C M TP S

0[defective]

AspP

5

Coherent Infinitives in German

B

ecause of their peculiar properties, coherent inﬁnitives have been extensively studied within modern German linguistics. The list of German linguists who have contributed descriptions and analyses of coherent inﬁnitives is rather long (cf. Bierwisch [1990], Fanselow [1989], Geilfuß [1991], Grewendorf [1987], Haegeman and van Riemsdijk [1986], Haider [1991], Höhle [1989], Lötscher [1978], Kroch and Santorini [1991], Olsen [1981], Reis [1976], Rosengren [1992], Stechow [1990], Sternefeld [1989]). As Rosengren (1992) points out, despite all this research on coherent constructions, German linguists so far have failed to provide an intuitively convincing and a theoretically consistent description and deﬁnition of what coherent verbs are. Recent advances in theory, in particular the proliferation of functional categories (cf. Cinque [1999]), have allowed us to take a complete new perspective on restructuring verbs, opening up the possibility of treating them as occupying functional heads in the clause (cf. especially Cinque [2001] for an interesting proposal that restructuring verbs in Italian are “functional” verbs). Wurmbrand (2004) draws an important distinction between lexical and functional restructuring verbs, arguing that restructuring in German “exploits” lexical restructuring verbs, since the order restrictions between several restructuring verbs, which are typical for functional restructuring verbs in Italian, do not obtain in German. For many years now the debate has been centered on the question of whether coherent constructions are monosentential structures that involve complex verbs or bisentential structures that involve an embedded inﬁnitive. Recently, Wurmbrand (2001) proposed a new account of restructuring, building largely on Haider’s (1991) 129

130

SCRAMBLING , REMNANT MOVEMENT , AND RESTRUCTURING IN WEST GERMANIC

empirical arguments for the monoclausal nature of coherent to-inﬁnitives in German. Within the proponents of the biclausal approach the question was put forward of whether the embedded inﬁnitive in coherent constructions is a CP, an IP, or a mere VP (cf. Sabel [1996] and Roberts [1997] for a more recent treatment of restructuring inﬁnitives in this paradigm). In the previous chapter, I argued in favor of a biclausal approach that assumes that inﬁnitival complements even in coherent constructions are full CPs. In this chapter, I will provide two empirical arguments against a monoclausal account of restructuring. (1) I will show in 5.1.1 that adverbs in restructuring contexts can co-occur in an order that is impossible in simple sentences, as is schematized in (1). (1)

a. [CP Adv1 Adv2 V2 V1] b. *[CP Adv1 Adv2 V1/V2] c. [CP Adv2 Adv1 V1/V2]

In (1a), the ﬁrst adverb (Adv1) modiﬁes the matrix verb (V1), while the second adverb (Adv2) modiﬁes the embedded verb, which has been restructured (V2). As (1b) indicates, there are cases in which two adverbs can modify neither the matrix verb nor the embedded verb in the given order. In these cases, a grammatical monoclausal sentence is only possible if the order of the two adverbs is reversed, as shown in (1c). This disparity between the possible orders of adverbs in simple clauses and in restructuring contexts follows straightforwardly in my account, which assumes that the middle ﬁeld in structuring inﬁnitives comprises more than one TP, that is, more than one licensing domain for adverbs. (2) I show that the binding properties in restructuring inﬁnitives diﬀer from the binding properties of simple clauses. In section 5.1.2, I will demonstrate that the binding properties of coherent inﬁnitives of ECM-verbs diﬀer from the binding properties of ditransitive verbs. I will argue that the binding properties of ECM-inﬁnitives follow directly if it is assumed that ECM-inﬁnitives comprise more than one TP. In section 5.2, I will compare my approach with Haider’s (1986, 1987, 1991) monoclausal account in terms of complex verbs and take issue with his claim that a biclausal approach fails to account for the essential properties of to-inﬁnitives in German. Coherent to-inﬁnitives in German allow for the so-called long passive, as is illustrated in (2), and have rather idiosyncratic binding properties. (2)

Der Zaun wurde zu reparieren versprochen the fence was to repair promised ‘Someone promised to repair the fence’

In, for example, an object control structure without restructuring, the matrix subject can be coreferent with an embedded object pronoun (3a). The fact that this option disappears in its restructuring variant (3b), together with the availability of long passives, has been taken as evidence for the monoclausality of restructuring toinﬁnitives. Note that (3b) becomes ungrammatical if the embedded object pronoun is replaced with an anaphor (4a). The fact that the Accusative anaphor in (4a) cannot

COHERENT INFINITIVES IN GERMAN

131

be bound by the Nominative subject Hans is problematic for the monoclausal approach, since this binding relation is, of course, available in a clearly monoclausal structure projected by a triadic verb, as is shown in (4b). I will show in section 5.2 that the binding properties of both pronouns and anaphors can be accounted for straightforwardly in a biclausal approach. (3)

a. weil Hansi ihrj geraten hat, [PROj ihni/k zu waschen] since Hans her recommended has him to wash b. weil Hansi ihn??i/k ihr zu waschen geraten hat since Hans him her to consult recommended has ‘since he recommended to her to wash him’

(4)

a. *weil sichi/j ihrj der Hansi zu waschen geraten hat since herself/himself her the Hans to wash recommended has ‘since Hans recommended to her to wash herself/himself’ b. weil sichi ihr der Hansi vorgestellt hat since himself her the Hans introduced has ‘since Hans introduced himself to her’

To summarize, I will demonstrate in this chapter that, on the one hand, a biclausal approach can account for the properties of the inﬁnitival complements of modals and ECM-verbs as well as for the rather idiosyncratic properties of to-inﬁnitives in coherent constructions and argue that, on the other hand, Haider’s treatment of coherent to-inﬁnitives is incapable of accounting for the properties of coherent constructions of modals and ECM-verbs.

5.1 Evidence for biclausality 5.1.1

Modal verbs

In this section, I want to argue that inﬁnitival constructions of modal verbs have biclausal properties. First, note that coherent constructions of modal verbs allow for two distinct time references. This is illustrated in (5). (5)

a. weil sie der Hans gestern morgen besuchen wollte since her the Hans yesterday tomorrow visit wanted b. weil der Hans gestern sie morgen besuchen wollte since the Hans yesterday her tomorrow visit wanted ‘since Hans wanted yesterday to visit her tomorrow’

If there are two adverbial modiﬁcations, some speakers prefer the order in (5b), where the pronoun has not scrambled into the matrix clause, over the order in (5a). But this preference is immaterial for my argument here.1 In general, a simple clause allows for only one independent time reference.2 Time adverbials are standardly assumed

132

SCRAMBLING , REMNANT MOVEMENT , AND RESTRUCTURING IN WEST GERMANIC

to modify the reference time in a Reichenbachian system of Tense. The reference time is located with respect to the speaking time by a temporal predicate in TP. Thus, it seems that two independent time modiﬁcations require two independent TPs. It has been observed that in certain complex tenses a time adverbial can modify two time points in the temporal structure. For instance, in the past perfect a time adverbial may specify the event time (6b) or the reference time (6c). However, as (6d) shows, it is not possible to specify the event time and the reference time with a separate time adverbial.3 Compare in this respect (6d) with (7). Example (7) contains two time adverbials, each of which speciﬁes the temporal location of a distinct event denoted by an independent clause. (6)

a. He had visited her at two o’clock b. the visiting took place at two o’clock c. at two o’clock he was in a state of having visited her earlier d. *At four o’clock he had visited her at two o’clock

(7)

Um 2 Uhr hat er sie um 4 Uhr besuchen wollen at two o’clock has he her at four o’clock visit want-IPP ‘At two o’clock he wanted to visit her at four o’clock’

Another argument in favor of a biclausal analysis is the observation that coherent constructions of modals can host two adverbials that cannot co-occur in a single clause with the same meaning or in the same order. As (8) shows, two adverbs can occur in an order in a coherent inﬁnitival construction (8a) that is unavailable in the simple clauses made up of either of the two verbs that appear in the coherent construction (8b–c). As (8d) shows, these adverbs can only co-occur in a single clause in the opposite order. (8)

a. weil Peter mich schon lange heute besuchen wollte since Peter me already for-a-long-time visit wanted ‘Already for a long time has Peter wanted to visit me today’ b. *weil mich Peter schon lange heute besucht hat since me Peter already for-a-long-time today visited has c. *weil das Peter schon lange heute wollte since that Peter already for-a-long-time today wanted d. weil mich Peter heute schon lange besucht hat since me Peter today already for-a-long-time visited has

It is important to note that the preceding tests for biclausality can be combined with other diagnostics of transparency that will be introduced in section 5.2. Example (9) shows that multiple time reference is possible in sentences with a topicalized inﬁnitive (9a) and long extraposition, which are two of the diagnostics held relevant for coherent inﬁnitives.

COHERENT INFINITIVES IN GERMAN

(9)

133

a. besuchen hat sie der Hans schon gestern morgen wollen visit has her the Hans already yesterday tomorrow want-IPP ‘as for visiting her, this is something that John has already yesterday wanted to do tomorrow’ b. weil er die Tür gestern morgen öﬀnen wollte, die zur Gruft führt since he the door yesterday tomorrow open wanted that leads to the grave

5.1.2

ECM-verbs

It is often observed that the vast majority of restructuring verbs that take bare inﬁnitival complements do not allow for distinct time references. These are the modals können (‘can’), sollen (‘shall’), müssen (‘must’), dürfen (‘may’), the ECM-verbs hören (‘hear’), sehen (‘see’), fühlen (‘feel’), and the causative/permissive verb lassen (‘let’). For instance, Wurmbrand (2001) takes this property, namely, the possibility of only a single temporal reference, as a deﬁning characteristic of restructuring verbs. From this semantic observation Wurmbrand believes to be allowed to directly deduce the syntactic structure of the complements of restructuring verbs. In her analysis, the complements of restructuring verbs lack a CP- and a TP-layer. Restructuring verbs in her system only take Aspect Phrases as complements and the modal verb wollen is treated as an exception. My take on this particular view is the following: The observation is correct, but the conclusion drawn from it is wrong. That the vast majority of these verbs do not allow for two (or several) distinct temporal references is a lexical semantic property of these restructuring verbs and should not be made a syntactic characteristic of restructuring itself. To draw a comparison, PRO in contexts of obligatory control is anaphoric. But this observation has not led people to assume that PRO is missing in these contexts. However, if Tense is anaphoric, people tend to take this as evidence that the embedded clause lacks a TP. In my opinion, this conclusion is completely unwarranted. I will later argue that the temporal interpretation of inﬁnitives is established via control, which is a lexical property of the selecting verb. That in “see X come” or in “have X come” matrix and embedded event obligatorily overlap follows from what it means to directly perceive an event happening and from what it means to directly cause an event to happen. I will later give a list of syntactic characteristics that deﬁne all restructuring verbs that take bare inﬁnitival complements, including wollen (‘want’), plus all restructuring verbs that take to-inﬁnitives as complements that do or do not allow for more than one distinct temporal reference. In the following, I want to investigate the binding properties of ECM-inﬁnitives. This brief investigation will reveal that coherent constructions of ECM-verbs embody two (distinct) binding domains. In the following examples, the order of arguments, if not indicated otherwise by coindexed traces, is always matrix subject > ECM-subject > ECM-object. Example (10) shows that in a typical coherent construction all arguments, the matrix subject as well as the embedded subject and object, can appear in the matrix clause. Since the negation in (10a–b) can have matrix scope (the most natural reading),

134

SCRAMBLING , REMNANT MOVEMENT , AND RESTRUCTURING IN WEST GERMANIC

the embedded subject and object must have moved into the matrix TP, after restructuring has moved the embedded TP into [Spec,PredP] of the matrix verb (10c). I will take this characteristic as the deﬁning property of a coherent construction. (10) a. weil er sie ihn nicht waschen ließ since he her him not wash let-Past b. weil er sie ihn nicht waschen sah since he her him not wash saw c. [CP weil [TP1 er siei ihnj [nicht [PredP [TP2 ti tj] waschen ließ/sah]]]]

Note that the arguments in a coherent ECM-construction behave very diﬀerently from the arguments of a triadic verb. With triadic verbs both object pronouns have to be disjoint from the subject, as expected (11c). In coherent constructions with ECMverbs, the embedded subject must be disjoint from the matrix subject, while the embedded object can be coreferent with the matrix subject (11a–b). Example (12) shows that the matrix subject cannot license an anaphoric embedded object (12a) but can license an embedded anaphoric subject (12b). An embedded anaphoric object can, however, be licensed by the embedded subject (12c). (11) a. weil eri ihn*i/j sie nicht waschen ließ since he him her not wash let-Past b. weil eri sie ihni/j nicht waschen ließ since he her him not wash let-Past c. weil der Hansi ihn*i/j ihm*i/k vorstellte since the Hans him-ACC him-DAT introduced (12) a. *weil eri sie sichi nicht waschen sah since he her himself not wash saw b. weil eri sichi sie nicht waschen sah since he himself her not wash saw c. weil er siei sichi nicht waschen sah since he her herself not wash saw

The examples in (11) and (12) clearly show that there are two distinct binding domains in coherent inﬁnitives of ECM-verbs, which are constituted by the embedded TP and the matrix TP. The binding properties of the DPs in (11) and (12) can be given a coherent explanation if we assume that the ECM-subject is Case-licensed in the matrix TP, but the object is Case-licensed in the embedded TP. The facts of pronominal binding, that is, the facts in (12), follow from the following principle.4 (13)

Pronouns and anaphors are interpreted in their Case-licensing positions (i.e., they must be reconstructed into their Case-licensing position before the Binding Theory applies at LF).

COHERENT INFINITIVES IN GERMAN

135

The respective LFs of (12a–b) are given in (14a–b) after reconstruction has applied. If the pronominal embedded subject in (12a) could reconstruct into its base position in the embedded TP, (12a) should be grammatical. However, if the embedded object in (12b) could not reconstruct into its Case-position in the embedded TP, it could not be coreferent with the matrix subject. After reconstruction of the anaphors into their Case-positions has applied, it follows that they must be bound by the subject of their respective TP in (12). (14) a. *[CP weil [TP1 eri ihni [nicht [PredP [TP2 ti siej] waschen liess/sah]]]] b. [CP weil [TP1 erj siei [nicht [PredP [TP2 ti ihnj] waschen liess/sah]]]]

Grewendorf (1988) points out that the distribution of weak pronouns in ECMconstructions diﬀers considerably from their distribution in other coherent constructions. We have seen that in coherent constructions of modal verbs an embedded object pronoun can move in front of the matrix subject. In ECM-constructions, the embedded subject can cliticize to the matrix subject but not the embedded object. Example (15c) can only express the meaning of (15a), not the meaning of (15b). (15) a. weil der Vater ihn den Lehrer grüssen ließ since the father him the teacher greet let-Past ‘since the father had him greet the teacher’ b. weil der Vater den Lehrer ihn grüssen ließ since the father the teacher him greet let-Past ‘since the father had the teacher greet him’ c. weil ihni/*j der Vater ti den Lehrer tj grüssen ließ since him the father the teacher greet let-Past

Grewendorf (1988) further observes that only an object pronoun es (‘it’), no other pronoun, can move in front of the embedded subject but not in front of the matrix subject (cf. [16b] versus [16c]). However, I do not agree with Grewendorf’s judgment in (16c). An object pronoun es can quite naturally cliticize in front of the matrix subject, as shown in (16d). (16) a. weil die Mutter den Vater es (das Baby) wickeln ließ since the mother the father (it) (the baby) wrap let-Past b. weil die Mutter es den Vater wickeln ließ c. *weil es die Mutter den Vater wickeln ließ d. weil es die Mutter den Vater lesen/sehen ließ since it the mother the father read/see let-Past

It seems that the word order in ECM-constructions is rather frozen, probably because of the Case syncretism. The embedded subject and object are marked with

136

SCRAMBLING , REMNANT MOVEMENT , AND RESTRUCTURING IN WEST GERMANIC

Accusative in the vast majority of the cases (only if the embedded verb governs a lexical Case, for instance, the Dative, can the embedded subject and object be distinguished formally). In my account of scrambling given in chapter 2, it follows from the Minimality Condition that the ECM-subject blocks movement of embedded object pronouns. However, the ECM-subject can move in front of the matrix subject, because the Case distinction between those arguments circumvents a violation of minimality, as is argued in chapter 2. Thus, ECM-verbs display a hallmark of what Haider (1991) considers as constitutive of coherent constructions, as we will see in the following section: They allow both of their arguments to move into the matrix domain and the argument that is Case-distinct from the matrix subject can scramble across the matrix subject. Thus, I argue that the embedded object could in principle move across the matrix subject in ECM-constructions but is prevented from doing so by the embedded subject due to minimality. In this account, we can assume that the clitic pronoun es can escape a violation of minimality by undergoing head movement across the non-distinct embedded subject. To summarize, I have shown that coherent inﬁnitivals of ECM-verbs comprise two binding domains. If the constitution of a binding domain involves the notion “structural subject” and if the notion “structural subject” is identiﬁed with [Spec,TP], then the data discussed earlier could be taken to show that coherent constructions of ECM-verbs are biclausal. Since small clauses also constitute separate binding domains, the preceding take would force us to assume that small clauses, too, are minimally TPs. While it is certainly possible to analyze small clauses as full CPs that restructure—after all, they are licensed in [Spec,PredP] of the selecting verb—it would go beyond the scope of this book to decide whether small clauses really are restructuring clauses or are better analyzed as lexical complements. Thus, I conclude this section with the weaker claim that the binding properties of ECM-inﬁnitives receive a simple and straightforward explanation if we assume that ECM-inﬁnitives contain (at least) two TPs. Furthermore, I have argued that in order to explain the binding and licensing properties of pronouns and anaphors in coherent constructions one needs to assume the principle in (13).

5.2 Verbs that take to-infinitives as complements I will center the discussion of the properties of coherent constructions with toinﬁnitives on Haider’s seminal work on coherent inﬁnitives. In a number of publications, Haider (1986, 1987, 1990, 1991) put forward the claim that restructuring (coherent) inﬁnitives in German display a number of syntactic peculiarities that cannot be properly accounted for by approaches that assume that inﬁnitival complements must be sentential and that derive coherent inﬁnitives from sentential ones (Haider 1991, 1). In the latter publication, he argues that there are a number of facts, little understood and mostly overlooked in the literature on restructuring, that decisively speak for a monosentential analysis of coherent inﬁnitives. I want to take up this challenge and have a good look at these peculiarities. Before I do this, I would like to list the properties that Haider considers to be characteristic

COHERENT INFINITIVES IN GERMAN

137

for coherent to-inﬁnitives. Haider claims that these properties deﬁne a monosentential structure and tries to derive them from his theory of base generated verbal complexes. I will later show that most of these descriptive properties also hold of the other coherent constructions, including those that clearly have biclausal properties, that is, of coherent ECM-constructions and of coherent constructions of modal verbs, including wollen. (17) Properties of restructuring to-inﬁnitives (Haider 1991, 2) a. verbal complex b. “monosentential” middle ﬁeld c. restricted to a subset of control verbs

The properties in (18) are taken to follow from the characteristics of verbal complex-formation (17a). I have already discussed the properties (18a–c) in chapter 1. The property in (18d) will be discussed in detail in the following chapter on the properties of verbal complexes and property. (18e) will be addressed later. To refresh our memories, the properties (18a–d) are illustrated in (19). Statement (18b) is meant to say that, for instance, negation in (19b) can modify either the embedded verb or the matrix verb. Example (19c) shows that the inﬁnitive in a coherent construction cannot scramble across an adverb. (18) a. The verbal complex can be topicalized. b. The scope of a scopal element aﬀects all verbs. c. The verbal complex cannot be split up. d. The ﬁnite verb can be inverted with IPP-complements. e. (Marginal) long passive is available. (19) a. [zu traktieren verstanden] hätte er die Mägde schon (Bech 1955/1983, 121) to disturb understood would-have he the maids very well b. dass er das nicht zu behalten vermochte that he that not to retain able-was ‘that he was unable to retain that’ ‘that he was able to not retain that’ c. *dass sich Max zu konzentrieren kaum vermochte that himself Max to concentrate hardly able-was d. dass er sie nicht erst hat zu überreden versuchen müssen that he her not ﬁrst has to convince try must-IPP

Important are the properties that Haider considers as indicative of a monosentential middle ﬁeld. These are listed in (20).

138

SCRAMBLING , REMNANT MOVEMENT , AND RESTRUCTURING IN WEST GERMANIC

(20) a. Cliticization in front of the matrix subject is possible. b. No scrambling of the embedded inﬁnitival is possible. c. Long extraposition is possible. d. Only one domain of negation is available.

Haider considers the property in (20a), which is illustrated in (21a), to be essential for a coherent, that is, for him, monosentential structure. We have seen that in ECM-constructions the embedded object can generally not cliticize in front of the matrix subject, but that the embedded subject can do so. The properties given in (20a– c) I have already discussed in chapter 1. They are illustrated once again in (21a–c). (21) a. weil ihn ihr der Hans zu konsultieren geraten hat since him her the Hans to consult recommended has ‘since Hans recommended her to consult him’ b. *weil [ihn zu kennen] Hans nicht scheint since him to know Hans not seems ‘since Hans does not seem to know him’ c. weil er die Tür ti zu öﬀnen zu versuchen wagte [die zur Gruft führt]i since he the door to open to try dared which to the crypt leads ‘since he dared to try to open the door that leads to the crypt’

The criterion in (20d) is in need of some explication. Haider acknowledges that negative expressions are generally ambiguous in coherent constructions but denies that coherent constructions contain more than a single domain of negation. So, for instance, the negative elements in (22) are ambiguous between a matrix scope (‘not dare’) and an embedded scope (‘not annoy’) construal. But for Haider, the sentences in (22) contain only one domain of negation, with the ambiguity arising from the fact of whether we are dealing with sentence negation or constituent negation. He claims it is impossible to have two separate clausal domains within a coherent construction and cites examples like (23c), where he claims that the two combined negations only cancel each other but cannot scope in diﬀerent domains. Hence the low acceptability of (23c). (22) a. weil er damit niemanden zu belästigen wagte since he there-with nobody to annoy dared ‘since he didn’t dare to annoy anybody with it’ ‘since he dared to annoy nobody with it’ b. weil er sie nicht zu belästigen wagte since he her not to annoy dared ‘since he did not dare to annoy her’ ‘since he dared not to annoy her’

COHERENT INFINITIVES IN GERMAN

139

(23) a. weil das kein vernünftiger Mensch leugnen kann since that no reasonable person deny can ‘since no reasonable person can deny that’ b. weil das ein vernünftiger Mensch nicht leugnen kann since that a reasonable person not deny can ‘since a reasonable person cannot deny that’ c. ??weil das kein vernünftiger Mensch nicht leugnen kann since that no reasonable person not deny can

I do not agree with Haider’s judgment. Example (23c) already improves, if we simply change the word order to that in (23d). Example (23d) can mean that ‘a reasonable person must deny that’, indicating that the two negations can scope in distinct domains (not [possible that not P] = necessary that P). I agree that this reading is hard to get, but it is deﬁnitely available. With some carefulness better examples can be constructed. Example (23e) can have the reading that it is possible that a man does not love women. Example (23f) can mean that it is not possible that a man does not love any women. This reading is even more readily available in (23g). It may be objected that with changing the word order as in (23d) one cannot be sure that we are really dealing with a coherent construction in (23d–g). This objection is irrelevant since modal verbs, other than control verbs, do not allow for a non-coherent construction, as is evidenced by the fact the presumptive sentential complement of a modal cannot be extraposed, as is illustrated in (23h). (23) d. ?weil kein vernünftiger Mensch das nicht leugnen kann since no reasonable person that not deny can e. weil ein Mann keine Frauen lieben kann since a man no women love can f. weil kein Mann keine Frauen lieben kann since no man no women love can g. weil kein Mann Frauen nicht lieben kann since no Mann women not love can h. *weil kein Mann kann [Frauen nicht lieben] since no man can women not love

Next, Haider argues that in non-coherent constructions each clause can be negated separately, leading to easily distinguishable, distinct interpretations (24a–c). In the coherent construction in (24d), however, he argues, the distinct interpretations evoked by more than one negation not only disappear, but the structure also becomes unacceptable, in a similar fashion as is the case with stranded constituent negation in a simple clause (25). Once again, the verbal complex in (24d) has been topicalized to ensure the sentence is constructed coherently.

140

SCRAMBLING , REMNANT MOVEMENT , AND RESTRUCTURING IN WEST GERMANIC

(24) a. Hans hat nicht gelobt, sie zu besuchen Hans has not vowed to visit her b. Hans hat gelobt, sie nicht zu besuchen Hans has vowed not to visit her c. Hans hat nicht gelobt, sie nicht zu besuchen Hans has not vowed not to visit her d. *zu besuchen gelobt hat er sie nicht nicht to visit vowed has he her not not (25) a. Ich habe sie nicht nicht geküsst I have her not not kissed b. *Geküsst habe ich sie nicht nicht Kissed have I her not not

I am not sure whether (24d) in conjunction with (25b) really shows what Haider wants it to show. Example (25b) can also be taken to show that it is simply bad in any structure, hence also in (24d), to strand a double negation. Also, note that the relevant reading is available in the coherent construction without topicalization (26a). Though (26a) is not the most elegant German sentence and is a bit hard to process, the reading is clearly available with an intonational break (#) between the two negative markers. Furthermore, I would like to point out with respect to (24d) that, in general, a stranded adverb cannot modify an embedded topicalized verb. The frequency adverb oft (‘often’) in (26b) can only have matrix scope, giving rise to the (sole) interpretation that ‘he often promised to read the book’. (26) a. ?weil er sie nicht (#) nicht zu besuchen gelobte since he her not not to visit vowed b. [[zu lesen] versprochen] hat er das Buch oft (only matrix scope) to read promised has he the book often c. [zu lesen] versprach er das Buch oft (ambiguous) to read tried he the book often

The contrast between (26b) and (26c) in conjunction with (26a) shows that the unavailability of the relevant reading in (24d) is a property of topicalized verb phrases rather than an inherent property of the coherent construction itself. Given the observation in (27), it follows that both negative markers must be construed with the matrix verb in (24d), explaining why the grammatical status of (24d) is on a par with (25b). (27)

Observation An adverb stranded in the middle ﬁeld cannot modify a verb that is embedded in a topicalized verb cluster

COHERENT INFINITIVES IN GERMAN

141

Of course, (27) is just an empirical generalization and should be derived either from the theory of coherent to-inﬁnitives or from a restrictive theory of VP-topicalization. One way to explain (27) in terms of restrictions on topicalizability, pointed out to me by an anonymous reviewer, is to assume that PredP may be aﬀected by topicalization rather than the local Aspect Phrase. If the entire PredP is topicalized in a restructuring inﬁnitive, it contains the adverb that modiﬁes the dependent inﬁnitive, while an adverb that modiﬁes the matrix verb would be stranded by topicalization. This assumption makes the prediction that an adverb that modiﬁes a dependent inﬁnitive can be topicalized with it. As (28) shows, this prediction is indeed borne out. (28) a. oft zu lesen versucht hat er das Buch nicht often to read tried has he the book not b. ?nicht zu besuchen gelobt hat er sie gerade nicht not to visit vowed has he her indeed not c. nicht zu besuchen gelobt hat er keine der Frauen not to visit vowed has he none of the women ‘for none of these women he did vow to not visit them’

The examples in (28) are rare and hard to process, since they require a very speciﬁc context, but are deﬁnitely grammatical and thus do show (cf. especially [28bc]) that there are two domains of negation in coherent to-inﬁnitive. More research is to be done on the restrictions of topicalization of coherent inﬁnitives. For this reason, I will disregard criterion (20d). Finally, let us look at property (17c), or rather, the question of which verbs allow for a coherent construction with to-inﬁnitives. First of all, note that all Subject Raising verbs require a coherent inﬁnitival complement. With control verbs, the coherent construction is always optional. Bech (1955/1983) deﬁnes the class of verbs that can take either coherent or noncoherent inﬁnitival complements as those subject control verbs that do not have a nominal argument of their own (their sole internal argument is the inﬁnitive). However, as Haider (1991, 250) points out, Bech’s deﬁnition of the class of restructuring verbs speciﬁes neither the necessary nor the suﬃcient conditions for coherency. Haider notes that, on the one hand, there are verbs that satisfy Bech’s deﬁnition but do not allow coherent inﬁnitival complements (29). On the other hand, there are subject control verbs and object control verbs with nominal arguments that quite readily allow for a coherent construction (30). The verbs in (29) all take factive complements. The verbs in (30) all take Dative DPs as additional arguments (next to the inﬁnitival complement). Haider argues that control verbs with an additional direct nominal argument, an Accusative DP, do not allow for coherent constructions. I believe that Haider’s observation is correct and adopt (31) as a working hypothesis. If we exclude verbs that take factive complements, the deﬁnition in (31) is particularly interesting if it turns out to be tenable, since it allows for a purely syntactic characterization of restructuring verbs with to-inﬁnitives.

142

SCRAMBLING , REMNANT MOVEMENT , AND RESTRUCTURING IN WEST GERMANIC

(29) a. *weil sie der Hans noch nie geheiratet zu haben bedauert hat since her the Hans yet never married to have regretted has b. *weil sie der Hans nicht zu besuchen fortfuhr since her the Hans not to visit continued c. *weil sie der Hans nicht wiedersehen zu wollen vorgab since her the Hans not again-see to want pretended (30) a. weil ihn uns jemand nicht auszukosten erlaubte (object control) since him us somebody not to enjoy allowed b. weil ihn mir jemand zu konsultieren geraten hat (object control) since him me somebody to consult recommmended has (31)

With the exception of verbs that take factive complements, control verbs without a direct (Accusative) object allow for a coherent construction.

5.3 The idiosyncratic properties of coherent to-infinitives Coherent to-inﬁnitives diﬀer from their non-coherent counterparts in terms of Caseassignment and in their binding properties. Höhle (1978) observed that coherent toinﬁnitives allow for the so-called long passive. In the long passive, the object of the embedded verb appears as the subject of the passivized matrix verb. In the following examples, taken from Haider (1991, 8), the nonﬁnite verbs have been topicalized (property [18a] earlier) to ensure that we are dealing with a coherent construction. (32)

Zu reparieren versucht wurde der Wagen/* den Wagen schon dreimal to repair tried was the car-NOM/* the car-ACC already three times ‘Someone tried already three times to repair the car’

Haider (1991) notes that the surprising property of this construction is not so much the (marginal) acceptability of the Nominative but the complete unacceptability of the variant with Accusative. He argues that the Passivization of the matrix verb should not have any impact on the Case-licensing of the arguments of the embedded verb, as is the case in the non-coherent variant of (32), exempliﬁed in (33). (33)

weil schon dreimal versucht wurde, den Wagen/* der Wagen zu reparieren since already three times tried was, the car-ACC/* the car-NOM to repair

Haider calls the property illustrated in (32) “Case-conversion” and argues that it should not be considered an exceptional peripheral phenomenon, possibly due to a peculiarity of Passivization in German, but constitutes a core property of coherent constructions. Assuming that coherent constructions are monosentential, he reasons that coherent constructions with an unaccusative matrix verb should also display Caseconversion. As (34) shows, his prediction is borne out. Gelingen is an unaccusative

COHERENT INFINITIVES IN GERMAN

143

verb whose experiencer is marked with Dative Case and whose theme-argument appears as Nominative subject (34a). Its theme-argument can also be realized as an inﬁnitival complement, which is non-coherently constructed in (34b). In the coherent version in (34c), the embedded argument cannot be realized as an Accusative DP but must be spelled out as a Nominative phrase. (34) a. Der Brief ist mir gelungen the letter-NOM is me succeeded ‘I did well with (writing) the letter’ b. es ist mir auf Anhieb gelungen, den Brief/* der Brief zu entziﬀern it is me immediately succeeded the letter-ACC/the letter-NOM to decipher c. ?zu entziﬀern gelungen ist mir der Brief/* den Brief auf Anhieb to decipher succeeded is me the letter-NOM/* the letter-ACC immediately ‘I succeeded immediately in deciphering the letter’

Another peculiarity of coherent to-inﬁnitives concerns the interpretation of pronouns and the licensing of anaphors. In a non-coherent inﬁnitival construction of an object control verb, the matrix subject can be coreferent with the embedded object (35a). This option disappears in its coherent variant. In (35b), the embedded object pronoun has cliticized in front of the matrix subject, indicating that we are dealing with a coherent construction. If the embedded object pronoun is replaced with an anaphor, the result is ungrammatical (36). (35) a. weil der Hansi ihrj geraten hat, [PROj ihni/k zu konsultieren] since her the Hans recommended has him to consult b. weil ihn??i/k ihr der Hansi zu konsultieren geraten hat since him her the Hans to consult recommended has (36) a. weil der Hans ihri geraten hat [PROi sichi zu waschen] since he her recommmended has PRO herself to wash b. weil (?? sich) ihr (* sich) der Hans zu waschen geraten hat since (herself) her (herself) the Hans to wash recommended has

Haider (1991, 9) also observes that the coherent construction with an unaccusative matrix verb is ungrammatical if the direct argument of the inﬁnitive is an anaphor. Compare the non-coherent construction in (37a) with its coherent counterpart in (37b). This follows if the object of the to-inﬁnitive in a coherent structure with an unaccusative matrix verb must be licensed as a Nominative DP: German does not have Nominative anaphors. (37) a. weil es Maxi nicht gelungen ist, [PROi sichi zu befreien] since it-NOM Max-DAT not succeeded is himself to free b. *zu befreien gelungen ist sich Max nicht to free succeeded is himself Max-DAT not

144

SCRAMBLING , REMNANT MOVEMENT , AND RESTRUCTURING IN WEST GERMANIC

For Haider these peculiarities of coherent to-inﬁnitives are indicative of the monoclausality of coherent to-inﬁnitives. He assumes that the arguments in a coherent to-inﬁnitival are arguments of a single (complex) verb. If coherent constructions contain only one (complex) verb, then it follows that an internal argument will be realized as a Nominative DP if this verb is passivized (long passive) or inherently unaccusative. If coherent constructions contain only one (complex) verb, then it also follows (trivially) that object pronouns must be disjoint from the subject (cf. [35]). And ﬁnally, if coherent constructions contain only one (complex) verb, then it follows that they must be monosentential and hence cannot contain PRO. From the latter fact and from the fact that Dative arguments cannot license non-oblique anaphors in German (probably a language-speciﬁc property) it follows that direct object anaphors cannot be licensed in coherent constructions with object control verbs (cf. [36]). Haider (1991) proposes that not only simple heads can project phrases but also complex heads, that is, base-generated adjunction structures of heads. If the complex head is verbal, then the argument structures of two verbs are uniﬁed in that the argument structure of the adjoined verb replaces the internal argument of the “matrix” verb. In his system, the argument structure of a verb is represented by a Lambdaexpression, in which Lambda operators bind variables in a predicate formula that represents the conceptual structure of the verb. The uniﬁcation of argument structures in complex heads corresponds to the operation of functional composition in categorial grammar. In this process, the external argument of the adjoined verb is identiﬁed with an argument position of the “matrix” verb as determined by its control properties. After identiﬁcation, one Lambda operator binds two variables in the newly formed predicate formula, the variable corresponding to the controller and the variable corresponding to the controllee, such that one syntactic DP can be evaluated in two positions in the conceptual representation after functional application has applied. I do not want to describe this account in its details but hope that the preceding exposition suﬃces to show that a very speciﬁc mechanism of argument identiﬁcation that is geared by the control properties of the matrix verb is needed to account for the idiosyncratic properties of coherent to-inﬁnitives. These properties are thus stipulated rather than explained. We may wonder why it is that the external argument of the dependent inﬁnitive is identiﬁed with the controller rather than simply always with the external argument of the matrix verb. Be that as it may, such stipulations may be needed if there is no biclausal analysis of coherent to-inﬁnitives. In the next section, I will argue that there is a relatively elegant biclausal solution to the problems posed by the idiosyncratic properties of coherent to-inﬁnitives. Before I do that, I want to have another look at coherent ECM-inﬁnitives. It should be obvious that Haider’s account of coherent to-inﬁnitives is inapplicable to coherent ECMinﬁnitives, since I have shown in section 5.1.2 that ECM-inﬁnitives comprise two binding domains. It is not clear to me how to diﬀerentiate in Haider’s system the binding properties of a complex ECM-verb and a simplex triadic verb. Furthermore, I have shown in section 5.1.1 that coherent constructions of modals clearly have biclausal properties. In the following, I will show that coherent to-inﬁnitives, ECM-inﬁnitives, and coherent inﬁnitives of modal verbs all share the properties that Haider considers to

COHERENT INFINITIVES IN GERMAN

145

be characteristic for a monoclausal structure. The properties in (18), repeated here as (38) for convenience, which Haider considers indicative for the presence of a verbal complex, are illustrated for ECM-verbs in (39) and for modal verbs in (40). (38) a. The verbal complex can be topicalized. b. The scope of a scopal element aﬀects all verbs. c. The verbal complex cannot be split up. d. The ﬁnite verb can be inverted with IPP-complements. e. A (marginal) long passive is available. (39) a. lesen lassen/sehen hat mich der Hans das Buch nicht (cf. [19a]) read let/see-IPP has me the Hans not ‘Hans has not seen/let me read the book’ b. weil mich der Hans das Buch nicht lesen liess/sah (cf. [19b]) since me the Hans the book not read let-Past/saw ‘since Hans did not see/let me read the book’ ‘since Hans saw/let me not read the book’ c. *weil der Hans mich das Buch lesen oft sah (cf. [19c]) since the Hans me the book read often saw d. weil der Hans mich das Buch nicht hat lesen lassen (cf. [19d]) since the Hans me the book not has read let-IPP

The test with long passives is not directly applicable to ECM-verbs that display the IPP-eﬀect in coherent constructions, since IPP-inﬁnitives are incompatible with a passive meaning. Those verbs that display the IPP-eﬀect only optionally allow a slightly marked long passive with the participle, as is indicated in (39e). (39) e. ?Er wurde die Maria küssen gesehen (cf. [32]) he-NOM was the Maria-ACC kiss see-PART ‘He was seen kissing Maria’ (40) a. schlafen wollen hat der Hans nicht sleep want-IPP has the Hans not ‘Hans has not wanted to sleep’ b. weil der Hans nicht schlafen wollte since the Hans not sleep wanted ‘since Hans did not want to sleep’ ‘since Hans wanted not to sleep’ c. *weil der Hans schlafen oft wollte since the Hans sleep often wanted

146

SCRAMBLING , REMNANT MOVEMENT , AND RESTRUCTURING IN WEST GERMANIC

d. weil der Hans nicht hat schlafen wollen since the Hans not has sleep want-IPP e. Not applicable, since modals do not passivize

The properties in (20), repeated here as (41) for convenience, which Haider considers indicative for the presence of a monosentential middle ﬁeld, are illustrated for ECM-verbs in (42) and for modals in (43). Property (20/41a) holds generally only for ECM-subjects but not for ECM-objects, as we have seen in section 5.1. Property (20/41d) is not illustrated because I have disputed its validity earlier. (41) a. Cliticization in front of the matrix subject is possible. b. No scrambling of the embedded inﬁnitival is available. c. “Long extraposition” is possible. d. Only one domain of negation is available. (42) a. weil mich der Hans sie nicht küssen liess since me the Hans her not kiss let-Past ‘since Hans did not allow me to kiss her’ b. *weil [mich sie küssen] der Hans nicht liess since me her kiss the Hans not let-Past c. weil er uns sagen ti lassen müssen wird [was zu tun ist]i since he us say let must shall what to do is ‘since he will have to allow us to say what is to be done’ (43) a. weil sie der Hans nicht küssen wollte since her the Hans not kiss wanted b. *weil [sie küssen] der Hans nicht wollte since her kiss the Hans not wanted c. weil er uns sagen ti will [was zu tun ist]i since he us tell wants what to do is ‘since he wants to tell us what is to be done’

To summarize, I have shown that ECM-inﬁnitives, coherent to-inﬁnitives, and inﬁnitives selected by modals share all the properties that are taken to be deﬁning characteristics of restructuring. For example, to take a criterion for verb cluster formation, ECM-verbs, modal verbs, and verbs that take coherent to-inﬁnitives do not diﬀer in their ability to topicalize their dependent inﬁnitive. Moreover, to take the important criterion of moving an embedded argument across the matrix subject, the three verb classes do not diﬀer in this property, either. Verbs that take coherent to-inﬁnitives and modal verbs alike allow the embedded object to move across the matrix subject. With ECM-verbs, only the embedded sub-

COHERENT INFINITIVES IN GERMAN

147

ject can scramble across the matrix subject, while the embedded object cannot do so. This diﬀerence, however, was argued to be immaterial for the criterion in question. The criterion is relevant insofar as it provides a “safe” test for whether embedded arguments can occur in the matrix clause (monoclausal middle ﬁeld). I have shown with an independent test that also with ECM-verbs the embedded arguments occur in the matrix clause: In section 5.1.2 I have shown that the embedded subject as well as the embedded object precede matrix negation. Then I provided a reason for why the embedded subject and the embedded object diﬀer in their scrambling abilities in the matrix clause, the reason being that the embedded object is blocked by the embedded subject from undergoing further scrambling due to a minimality violation caused by Case-syncretism. Since ECM-inﬁnitives, inﬁnitives selected by modals, and coherent to-inﬁnitives share all these properties, they are in need of a unitary account, and since I provided good arguments that the former two are biclausal, this uniform account can only be biclausal. Before I move on to a biclausal account of coherent inﬁnitives, let me address a crucial diﬀerence between coherent to-inﬁnitives and coherent constructions formed with bare inﬁnitives. Note that the long passives with ECM-verbs and with to-inﬁnitives are not strictly parallel. In the former case it is the embedded subject that is raised to the matrix subject position, while in the latter case it is the embedded object that is raised to the matrix subject position.5 This diﬀerence follows from the Case-licensing properties of coherent bare inﬁnitives and to-inﬁnitives. Bare inﬁnitives Case-license their direct object, while (coherent) to-inﬁnitives fail to do so. We will see in the following sections that this diﬀerence in Case-licensing follows from the [-verbal] character of the inﬁnitival marker in coherent to-inﬁnitives. Thus, this diﬀerence between long passives with ECM-verbs and verbs that take coherent to-inﬁnitives follows from an independent property within the biclausal approach that I am going to develop in the next section.

5.4 A biclausal account of coherent to-infinitives It is evident that the idiosyncratic properties of to-inﬁnitives sketched earlier, namely, the availability of the long passive, the pattern of pronominal binding, and the inability of licensing embedded anaphors with object control verbs, at ﬁrst sight all speak against a biclausal structure of coherent to-inﬁnitives. The binding facts in (34) and (35) seem to directly militate against the presence of PRO. If PRO were present in coherent to-inﬁnitives, one might ask, how can it be that an embedded direct object pronoun must shy away from being coreferent with the matrix subject and, more important, how can it be that an anaphoric argument of the inﬁnitive fails to be licensed? With long passives, the argument is more indirect but seems to lead to the very same conclusion. Haider (1991) argues that coherent to-inﬁnitives must be monosentential since Passivization of the matrix verb should not aﬀect the licensing of the arguments of the embedded verbs. Note that this conclusion is not entirely warranted, since we know of cases where an embedded argument becomes the Nominative

148

SCRAMBLING , REMNANT MOVEMENT , AND RESTRUCTURING IN WEST GERMANIC

subject of the passivized matrix verb, namely, in passives with ECM-verbs in English, as is illustrated in (44). It is generally assumed that in ECM-constructions the embedded subject moves into the higher clause to be assigned Case by the matrix verb. If the matrix verb is passivized, as in (44), then the embedded subject will be realized as matrix Nominative subject. (44)

Johni was (ti) expected [ti to come to the party]

If we take (44) as a model for the biclausal approach to long passives in coherent to-inﬁnitives, then we have to assume that in this construction it is the embedded object that moves into the matrix clause for reasons of Case-assignment. But how is Case-licensing movement of the object out of the embedded clause possible, one might ask, if an embedded subject, that is, PRO, is present? All this seems to indicate that PRO is missing in coherent to-inﬁnitives. And this conclusion would militate decisively against a biclausal approach. My take on this is that the earlier conclusion that PRO is missing in coherent toinﬁnitives is too fast and unwarranted. I think that all that the preceding data show is that PRO, if present, is not in a position to block movement of the embedded object or to license an embedded anaphor, and so on. At the end of this section, I will provide empirical evidence to show that PRO is present in coherent to-inﬁnitives. I need exactly one assumption to explain all the properties of coherent to-inﬁnitives within my alternative biclausal approach. This assumption is given in (45) and will be motivated later. (45)

A to-inﬁnitive in a coherent context cannot license (structural) Case.

Let us look at the derivation of the coherent sentence in (46b) from its non-coherent version in (46a). Given (45), the embedded pronoun/anaphor must be Case-licensed in the matrix clause. Haider is certainly right that a standard biclausal approach of restructuring in terms of VR and long-distance scrambling of individual DPs cannot derive the coherent version of a sentence like (46) with the stipulation in (45). Scrambling of the direct object in (46) into the matrix clause should be blocked, despite the availability of head movement of the embedded verb, by the presence of PRO. However, I have argued that arguments do not move individually into the matrix clause but rather are pied-piped by the movement of two larger constituents, an extended verb projection and the inﬁnitival TP, into the matrix clause. (46) a. weil der Hansi ihrj empfahl [PROj ihni /sichj zu waschen] since the Hans her recommended him/herself to wash b. weil ihn??i /sich*j ihrj der Hansi zu waschen empfahl since him/herself der Hans to wash recommended

The derivation proceeds in the following way: The embedded object pronoun/ anaphor contained within the Agreement Phrase is pied-piped by movement of the AspP via the C-domain of the inﬁnitival into [Spec,AspP] of the matrix verb, while

COHERENT INFINITIVES IN GERMAN

149

PRO contained within the embedded TP is moved into [Spec,PredP] of the matrix verb. Note that in this ﬁrst step the embedded object does not move across PRO. In the second step, the embedded object moves out of AspP to its Case-licensing position in the matrix clause. Movement of the embedded object is again unhampered by PRO, since PRO in the matrix clause is contained in a larger phrase, namely, the embedded TP. (47) [CP Hans ihr empfahl [CP [TP PRO [AgrP ihn/sich zu waschen]]]] Step 1: i) AgrP (the extended AspP) moves via [Spec,StatP] into [Spec,CP] a) [CP Hans ihr empfahl [CP [AgrP ihn/sich zu waschen] [TP PRO tAgrP]]] ii) AgrP moves to [Spec,AspP] in the matrix clause b) [CP Hans ihr [AspP [AgrP ihn/sich zu waschen] empfahl [CP tAgrP [TP PRO tAgrP]]]] iii) TP moves to [Spec,PredP] in the matrix clause c) [CP Hans ihr[PredP [TP PRO tAgrP] [AspP[AgrP ihn/sich zu waschen] empfahl [CP tAgrP]]]] Step 2: the embedded direct object moves into its Case-position without crossing PRO d) [CP Hans ihr [AgrO ihn/sichk [PredP [TP PRO tAgrP][AspP [AgrP tk zu waschen] empfahl]]]]

In an additional step the embedded object pronoun/anaphor may cliticize (like the matrix pronoun) in front of the matrix subject. This is essentially the derivation of a coherent to-inﬁnitive. It should be obvious that the derivation is essentially the same, diﬀering only in Step (47d), if the matrix verb is passivized. In this case, the embedded object will be assigned Nominative Case in an Agr-position in the matrix clause. Note that in the account described here it is crucial that the object is moved via movement of the Aspect Phrase, while PRO is moved via movement of the inﬁnitival Tense Phrase. It is a separate issue and ultimately an empirical question whether the direct object moves to AgrO in the absence of Case-licensing in the embedded clause. In (47), I have assumed it does. Then it is crucial that AgrOP is pied-piped by movement of the Aspect Phrase. However, if it turns out that the direct object cannot be taken to move to AgrOP in the embedded clause in this case, then no problem arises, either, since in any case it would be moved by the containing AspP into the matrix domain. The issue that is at stake here is the question of whether AgrO can be taken to have an EPP-feature in the absence of Case-licensing, as is generally assumed for AgrS in cases of Subject Raising. In (47), I have assumed object movement to AgrO in the embedded clause also to make this analysis parallel to my treatment of movement of TP and AspP to the respective Speciﬁers in the C-domain in restructuring contexts, as discussed at the end of chapter 4. The underlying assumption here is that licensing projections can be defective but are still present in the derivation. The empirical argument for this approach is that we need an A-position in the embedded clause to account for scope reconstruction in coherent inﬁnitives,

150

SCRAMBLING , REMNANT MOVEMENT , AND RESTRUCTURING IN WEST GERMANIC

as will be discussed later. If strong quantiﬁers cannot be taken to reconstruct into their Theta-position (cf. Lechner [1998]), there must be another argument position in the embedded clause to account for narrow scope readings of strong quantiﬁers in coherent inﬁnitives. This position must be AgrO in (47) earlier. Let us now look at the binding properties of the embedded object. From the binding properties of pronouns and anaphors in ECM-constructions I derived the principle in (13), repeated here as (48). (48)

Pronouns and anaphors are interpreted in their Case-licensing positions (i.e., they must be reconstructed into their Case-licensing position before the Binding Theory applies at LF).

Because of (48), the embedded object pronoun/anaphor cannot be reconstructed back into the embedded clause, although the moved TP and AspP may very well be reconstructed. Following (48) the embedded object pronoun/anaphor is interpreted in the position it occupies in (47d) earlier, that is, its Case-position in the matrix clause. It should be obvious from the structure in (47d) that the embedded object pronoun must be disjoint from both DP arguments of the matrix verb, in particular, from the matrix subject (cf. [46b]), and that the embedded object anaphor cannot be bound by PRO. It cannot be bound by PRO, neither in the matrix clause nor in the embedded clause. In the matrix clause, it cannot be bound by PRO because PRO being contained within TP in [Spec,PredP] does not c-command the anaphor. In the embedded clause, it cannot be bound by PRO, where PRO would be in a c-commanding position after reconstruction has applied to TP and AspP, simply because the anaphor fails to reconstruct together with AspP. The anaphor could be bound by the indirect argument of the matrix verb that controls PRO, but since this argument is marked with Dative Case and since Dative DPs in German (idiosyncratically) cannot bind Accusative anaphors, an embedded object anaphor is ungrammatical in a coherent to-inﬁnitive of an object control verb (I will come back to the issue of why the anaphor cannot be licensed with a diﬀerent indexing in [47d] later). While pronouns and anaphors seemingly have to be interpreted in their Caselicensing positions (at least for the purposes of the Binding Theory), there is independent evidence that quantiﬁcational expressions and presumably also names can be reconstructed into non-Case-positions, which comes from Subject Raising verbs in English. It is well-known that (49a) can have the interpretation given in (49b). This interpretation is only possible if the subject QP can reconstruct from its Caseposition in the matrix clause to its subject or base position in the inﬁnitival clause. (49) a. Someone seems to love everyone b. It seems that for everyone there is someone who loves him c. seems [every y [some x [x loves y]]]

The negative quantiﬁer in the coherent to-inﬁnitive in (50a) is ambiguous in the usual manner as given in (50b–c). The interpretation in (50c), where the negative

COHERENT INFINITIVES IN GERMAN

151

quantiﬁer is interpreted in the embedded clause, follows if it can reconstruct in the same fashion as the existential quantiﬁer in (49a). (50) a. weil niemanden der Hans seiner Frau zu konsultieren riet since nobody the Hans his wife-DAT to consult recommended b. for no person x [Hans recommended his wife to consult x] c. Hans recommended his wife to consult no person x

We see that by adopting (45) and (48), an empirical generalization derived from the binding properties of pronouns and anaphors in ECM-constructions, we do not lose anything while gaining a lot that was thought to be unexplainable in a biclausal approach. That is to say, by adopting (45) and (48) I can explain both the idiosyncratic properties of coherent to-inﬁnitives (their binding properties and their passivizability) as well as the properties typical of all coherent structures, namely, the ambiguity of operators and quantiﬁed expressions with respect to the matrix verb. All we have to assume is that pronouns and anaphors, on the one hand, behave diﬀerently with respect to reconstructability from quantiﬁers and names, on the other hand. At ﬁrst sight, it seems that the reconstructability of a syntactic element follows from economy principles. For a pronoun or anaphor, lacking both descriptive and quantiﬁcational content, it does not matter whether it is interpreted in the matrix clause or in the embedded clause (after reconstruction). In both cases it is interpreted de re. However, for a quantiﬁed expression or a name, it matters whether it is interpreted inside (de dicto) or outside (de re) of the scope of an intensional verb like recommend. From this we may conclude that reconstruction is possible if it gives rise to a distinct interpretation. However, this view of reconstruction is wrong, since reconstruction of the pronoun in (47d) would also give rise to a distinct interpretation, not available without reconstruction, namely, the coreferent interpretation with the matrix subject. It thus seems that something like (51) holds. (51)

A DP can be reconstructed if reconstruction does not alter its binding properties as determined by its Case-licensing properties (cf. note 4 earlier)

For a name or a quantiﬁer it does not matter whether it is licensed in the embedded clause or in a higher clause as far as its binding properties are concerned. They must be free in every domain. Since the binding domain of names and quantiﬁers is not determined as the minimal TP/DP in which they are Case-licensed, reconstruction may freely apply to them. For pronouns and anaphors, due to their locality condition, it really matters where they are Case-licensed: The binding domain for pronouns and anaphors is the minimal TP/DP in which they are Case-licensed and (for anaphors) which contains an accessible SUBJECT. This fact seems to be at the core of the apparent diﬀerence in reconstructability of Principle C-expressions and elements viable to BT as determined by Case-licensing. The principle in (51) is rephrased in (52). (52)

Reconstruction must preserve the binding domain of an element as determined by its Case-licensing properties

152

SCRAMBLING , REMNANT MOVEMENT , AND RESTRUCTURING IN WEST GERMANIC

5.5 Explaining the properties of coherent to-infinitives It is important to note that the PRO subject in coherent to-inﬁnitives has referential properties. In fact, its reference is supplied by the controlling argument of the matrix verb, that is, the subject with a subject control verb and the object with an object control verb, as is indicated in (53). (53) a. weil das Buch Hansi (PROi) zu lesen versprach since the book Hans to read promised b. weil das Buch der Hans der Mariai (PROi) zu lesen empfahl since the book Hans-NOM Maria-DAT to read recommended

In German, this control relation is preserved under Passivization with both subject and object control verbs, where with subject control verbs PRO is interpreted as coreferential with the passive implicit argument, as is illustrated in (54). The very same interpretations we ﬁnd in long passives, as is illustrated in (55). (54) a. weil von Hansi (Agensi) versprochen wurde, PROi das Buch zu lesen since by Hans promised was the book to read b. weil der Mariai (Agens) empfohlen wurde, PROi das Buch zu lesen since Maria-DAT recommended was the book to read (55) a. weil (von Hans) das Buch zu lesen versprochen wurde since (by Hans) the book-NOM to read promised was b. weil der Maria das Buch zu lesen empfohlen wurde since Maria-DAT the book-NOM to read recommended was

5.5.1

A note on passivization in German

I do not assume that Passivization detransitivizes the underlying verb.6 In the previous chapter, I have argued that the projection of argument-licensing heads is a function of the lexical and semantic properties of the verb. Thus, the presence of v implies that both Agr1 and Agr3 are projected (cf. [8] and [10] in section 4.2.1 of the previous chapter). However, the Case-assigning properties of Agr3 depend not only on the presence of v but also on the aspectual properties of the verb in many languages (cf. the assignment of Partitive Case or Accusative Case in Finnish). I assume that Agr3 assigns Accusative if there is a local Agreement relation between Agr3, little v, and the relevant Aspect-head (cf. Kratzer’s [1996] Voice-head). In this scenario, we can assume that in passive sentences a passive morpheme is inserted in the head of the Aspect Phrase (in the form of the participle morphology) that prevents agreement between Agr3 and v. Therefore, the direct object cannot be assigned Accusative in Agr3. In this case, the direct object is licensed with default Nominative Case.7 The agent argument is licensed as implicit argument in Agr1 (as empty impersonal pronoun, cf. Hinterhölzl [1995]). In this position, the agent argument is capable of exerting control of an inﬁnitival subject (see [60] later).

COHERENT INFINITIVES IN GERMAN

153

While in German also intransitive verbs quite generally passivize and form the so-called impersonal passive, these impersonal passives diﬀer in whether they allow for a referential interpretation of the implicit argument. Many intransitive verbs only allow for an existential or generic interpretation of the implicit argument, as is illustrated in (56). (56) a. weil hier getanzt wird (in general, habitually and in high numbers) since here danced gets ‘since a lot of dancing is going on here’ b. *weil hier von John Travolta getanzt wird since here by John Travolta danced gets c. weil in Afrika weiter gehungert und gestorben wird since in Africa still hungered and died gets d. *weil von Hans gehungert wird since by Hans hungered gets

Whether the ungrammaticality of the a-examples in (57)–(59) should be explained with the assumption that a bound implicit argument cannot serve as a controller for the embedded PRO argument I will leave open here. While the impersonal passive is ﬁne if no control relation is involved (cf. the b examples), examples (57)–(59) show that the status of long passives is on a par with the (control) status of impersonal sentential passives. (57) a. ?*weil (von Hans) geglaubt wurde, die Maria zu lieben since (by Hans) believed was to love Maria b. weil allgemein geglaubt wurde, dass Hans die Maria liebt c. ?*weil (von Hans) die Maria zu lieben geglaubt wurde (58) a. ??weil (von Hans) gehoﬀt wurde die Maria zu treﬀen since (by Hans) hoped was to meet Maria b. weil gehoﬀt wurde, dass der Hans die Maria triﬀt c. ??weil (von Hans) die Maria zu treﬀen gehoﬀt wurde (59) a. ?weil (von Hans) vergessen wurde, die Maria einzuladen since (by Hans) forgotten was to invite Maria b. weil vergessen wurde, dass der Hans die Maria einladen soll c. ?weil (von Hans) die Maria einzuladen vergessen wurde

Summing up, the examples in (57–59) show that the grammaticality of a long passive depends on the grammatical status of the sentential impersonal construction, implying that the long passive is to be derived from a sentential construction via restructuring, as argued for earlier.

154

5.5.2

SCRAMBLING , REMNANT MOVEMENT , AND RESTRUCTURING IN WEST GERMANIC

Control and the licensing of anaphors in coherent to-inﬁnitives

Now the question arises how the relevant control relation is established in a coherent to-inﬁnitive. To answer this question I must ﬁrst explicate how control works in sentential inﬁnitives. I assume that control involves a co-indexing relation between the relevant Agr-head of the matrix verb and the Tense-head of the embedded inﬁnitive that is mediated by the complementizer, that is to say, that the semantics of the matrix verb determines which of its agreement heads is co-indexed with the complementizer that is itself co-indexed with the embedded T-head. The coreferential interpretation of the controller and PRO is then simply an eﬀect of Spec-head agreement, as is illustrated in (60). (60)

with x being an index variable and i being the referential index transferred to PRO [AgrP Controlleri Agrx [VP V [CP Cx [TP PROi Tx . . .]]]

In restructuring inﬁnitives the complementizer is defective and thus fails to transmit a referential index to the embedded Tense-head, leaving the PRO argument without an interpretation. In this case the missing control relation is ﬁxed in that the embedded T-head undergoes head movement to the respective Agreement-head after restructuring. In other words, after the embedded TP has been moved into PredP of the matrix verb, the Tense-head moves to the relevant Agreement-head as determined by the control properties of the matrix verb. The coreferential interpretation of controller and PRO again is a consequence of Spec-head agreement. With this implementation of control in coherent inﬁnitives in mind, I would like to take up the issue of why the anaphor in (47d) cannot be licensed by another DP, in particular by the subject, within its local domain. Example (47d) is repeated here as (61). (61)

[CP Hans ihr [AgrO sichk [PredP [TP PRO tAgrP][AspP [AgrP tk zu waschen] empfahl]]]]

The question arises why the structure in (61) cannot be interpreted as meaning ‘John recommended to her to wash him, John’. It seems that the licensing domain of the anaphor in (61) is restricted to the domain of the indirect object. What we see at play here is the remnant vestiges of PRO. It cannot bind its local anaphor, but it holds it within the orbit of its antecedent. But how can PRO achieve this eﬀect? Earlier I have argued that after the inﬁnitival TP has been moved into [Spec,PredP] of the selecting verb, its Tense-head undergoes head movement to the relevant Agreement-head to identify PRO. In the case of (61), an object control verb, we have to assume that the inﬁnitival Tense-head moves to AgrIO , as is indicated in (62). If we then assume, following Chomsky (1993), that an anaphor is licensed by moving to the local Tense-head with its binding domain being restricted to the local Speciﬁer, then it follows that the subject that occurs higher in the structure is not a possible binder for the anaphor in (61). Note that for this account to work it is crucial that the embedded T that is adjoined to AgrIO c-commands the anaphor. Only then will it count as a closer licenser than the matrix T-head. It is thus crucial that we adopt Kayne’s (1994) deﬁnition of antisymmetric c-command, in which a head c-commands

COHERENT INFINITIVES IN GERMAN

155

out of its adjoined position. Thus, in order to explain the binding eﬀects in coherent to-inﬁnitives, Chomsky’s (1993) proposal about anaphoric licensing must be combined with Kayne’s (1994) treatment of head movement. (62)

[CP Hans [AgrIO ihr [Ti-AgrIO [AgrDO sich [PredP [TP PRO ti . . .] . . . ]]]]] \________/

Note that the explanation of the binding properties of the anaphor in (61) as given in (62) is strong support for Chomsky’s (1993) proposal that anaphors are licensed by undergoing (local) movement to Tense. No other theory of anaphoric licensing can explain so neatly why the binding domains of anaphors and pronouns may diﬀer in this manner. Thus, I will assume the following pre-theoretic deﬁnitions of the binding properties of pronouns and anaphors. (63) a. After reconstruction, an anaphor is licensed by moving (and adjoining) to the closest Tense-head. b. After reconstruction, a pronoun is licensed if it is unbound within its local TP.

Since adjunction of the embedded Tense-head to AgrIO in (62) does not make AgrIOP a Tense Phrase, a pronoun in place of the anaphor in (61) will still have to be disjoint from the matrix subject. It should be obvious that the restriction on the antecedent of the anaphor in (62) is a huge problem for Haider’s account of coherent inﬁnitives in terms of complex verbs as well as for any monoclausal account. In fact, this restriction may prove to be detrimental to this whole approach. I don’t see how Haider can distinguish, in a non-stipulative manner, between a complex verb of the sort of zu waschen-empfehlen and a triadic verb like vorstellen, where, the subject can, of course, bind a direct object anaphor (64). Similar problems arise for Wurmbrand’s (2001) account. (64)

weil der Hansi sichi ihr vorstellte since the Hans himself her introduced ‘since Hans introduced himself to her’

I conclude that the idiosyncratic binding properties of coherent to-inﬁnitives that Haider has brought to the attention of the linguistic community (though he does not discuss the restriction shown in [61]) do not support a monoclausal account of coherent to-inﬁnitives. While at ﬁrst sight the binding facts seem to indicate that PRO is absent in coherent to-inﬁnitives, they, if viewed with scrutiny, provide in their totality, including the restriction discussed in (61), evidence for the presence of PRO in coherent to-inﬁnitives. 5.5.3

Case-assignment in coherent to-inﬁnitives

What remains to be done is to motivate the assumption in (45) that this analysis hinges on. The question that arises is why to-inﬁnitives in coherent constructions cannot Case-license their direct objects, while to-inﬁnitives in a sentential construction do

156

SCRAMBLING , REMNANT MOVEMENT , AND RESTRUCTURING IN WEST GERMANIC

not fail to Case-license them. Moreover, this property cannot be ascribed to restructuring itself, since bare inﬁnitives in restructuring contexts do Case-license their direct object. The relevant property must be ascribed to the syntax of zu in coherent constructions. In fact, there are other occurrences of zu in which the corresponding inﬁnitive fails to Case-license its direct object, as is illustrated in (65b–d). (65) a. weil das Buch der Hans zu lesen versprach since the book the Hans to read promised ‘since Hans promised to read the book’ b. Das Buch ist zu lesen the book is to read ‘the book must/can be read’ c. Das Buch ist leicht/schnell zu lesen the book is easy/fast to read d. *ein [schnell das Buch zu lesender] Mann a quickly the book to read-Agr man ‘a man who must read the book quickly’ e. ein [das Buch schnell lesender] Mann a the book quickly read-Agr Mann ‘a man who reads the book quickly’

What the to-inﬁnitives in (65a–d) have in common is that they fail to Caselicense their direct object. I have assumed earlier that the direct object of coherent toinﬁnitives is Case-licensed in the matrix clause (65a). In the modal inﬁnitive in (65b) and in the so-called tough-construction in (65c), the direct object of the to-inﬁnitive is realized as Nominative subject of the auxiliary. The ungrammaticality in (65d) follows if we assume that the to-inﬁnitive fails to Case-license an argument that is assigned Accusative by the bare inﬁnitive constructed in a parallel fashion in (65e). In this section, I want to propose that coherent to-inﬁnitives (65a) and the inﬁnitives in so-called modal inﬁnitives (65b) and tough-constructions (65c) are nominal inﬁnitives. In this context, it is interesting to note that to-inﬁnitives have replaced bare inﬁnitives in more and more environments in the history of German. According to Demske (2001), modal verbs, ECM-verbs, and raising verbs were restricted to bare inﬁnitives while control verbs allowed both forms in Old High German. Originally, inﬁnitive and so-called to-inﬁnitive were quite diﬀerent both formally and functionally. The zu in the to-inﬁnitive goes back to a preposition that took a nominalized predicate as a complement. Zu in this context required a Dative complement that was still formally realized in Old and Middle High German times, as is illustrated for OHG in (66a). Also, the predecessor of the modern modal inﬁnitive was built with a nominalized verb carrying Dative Case, as is illustrated in (66b). (66) a. Sie gerotun al bi manne inan zi rinanne (Otfried II.14.24) sie begehrten alle, ihn zu berühren ‘they all desired to touch him’

COHERENT INFINITIVES IN GERMAN

157

b. Dhar ist auh in dhemu gotes nemin fater zi ﬁrstandanne (Isidor 4.14) da ist auch in dem Namen Gottes der Vater zu verstehen ‘in this (case) one has to understand the father also in the name of god’

In the German traditional literature the Gerundium, as the nominalized inﬁnitive is called, and the inﬁnitive are distinguished, even though some scholars argue that the inﬁnitive should be analyzed as a frozen Accusative form of a nominalized verb as well.8 At least formally, gerundium and inﬁnitive were distinct in their past, since only the gerundium displayed gemination of the ﬁnal n, a process that is probably due to the loss of the phoneme j. After nominalized verbs lost their Case-endings, the morphological distinction between inﬁnitive and gerundium slowly eroded and they became completely homophonous in Early New High German. In the following, I use the term gerund to refer to nominalized inﬁnitives in German.9 Gerunds in OHG could be used as both nouns and verbs. In their verbal use, gerunds probably could license Accusative Case, as can be seen in (66), in which an argument that belongs to the gerund is realized as the Accusative pronoun inan (‘him’). The element zu in (66) can thus be analyzed as a prepositional complementizer that selected a nominal complement, the gerund, that in turn selected a TP-predicate, as is illustrated in (67), with the gerund head attracting the VP and TP moving into a higher functional projection, possibly MoodP, in the C-domain (cf. Kayne [1999]). Later, when inﬁnitive and gerund became indistinguishable morphologically, the zu was probably reanalyzed as an inﬁnitival marker and to-inﬁnitives replaced more and more occurrences of bare inﬁnitives in the complements of control verbs. (67) a. [ b. [TP

zi

[

G

[TP

[VP ]]]]

base structure

zi

[VP

G]

[tTP

[tVP

surface structure

]]]

However, I would like to propose that nominalized inﬁnitives have not completely disappeared from the language. As stated earlier, I propose that the inability of toinﬁnitives to license structural Case in (65a–c) is connected with the fact that these to-inﬁnitives are in fact nominal verbs. In this account, the gerund is analyzed as a zero phrasal aﬃx. The Case properties of nominal verbs are taken to depend on the level of the clause structure at which the aﬃx attaches to syntactically, that is to say that they depend on whether the nominal-aﬃx attaches to the VP or to a higher projection (AgrOP, TP, etc.). I have argued that due to the deﬁciency of the complementizer the embedded TP is not temporally linked to the matrix predicate. Let us assume that the inﬁnitival Tense-head in this case fails to transfer a temporal index to the local Aspect-head. Next, I assume that the inﬁnitival marker zu that occupies the Aspect Phrase has the selectional property in (68). Since the gerund aﬃx must attach low, it follows that a “to-inﬁnitive” cannot license structural Case in this context. (68)

The inﬁnitival marker zu that is not assigned a temporal index selects a gerund.

Because of the reanalysis of the prepositional complementizer as inﬁnitival marker that occupies the head of AspP, a synchronic coherent to-inﬁnitive will have

158

SCRAMBLING , REMNANT MOVEMENT , AND RESTRUCTURING IN WEST GERMANIC

the structure in (69). The gerund is inserted as a zero-aﬃx whose selectional properties are satisﬁed via XP-movement of the inﬁnitival verb into its Speciﬁer. The embedded TP and AspP cannot be licensed in their own C-domain and undergo restructuring movement into their licensing positions in the matrix clause in the known fashion. (69)

[CP

[TP

[AspP zu [GP

0

[F2P

V

]]]]]

However, the inﬁnitival marker in a non-coherent construction is temporally linked and selects an inﬁnitive that as a verbal category does not fail to license structural Accusative. The gerund as a category with nominal properties fails to license structural Case but, as the modal inﬁnitive in (65c) shows, seems to be capable of licensing adverbial modiﬁcation. Zu lesen in (65c) fails to assign Accusative Case to its argument (which is realized with Nominative Case) but can be modiﬁed by the adverb schnell (‘fast’). More work on the syntax of nominalized verbs is necessary to shade light on this mixture of verbal and nominal properties that they do display. The standard way of accounting for the mixed properties of gerunds is to divide the properties according to the attachment site of the gerund. Structure below the attachment site of the gerund has verbal properties, and structures above it have nominal properties. The fact that the to-inﬁnitives in (65) can license adverbs speaks in favor of a high attachment site of the gerund. In this scenario, we cannot make the presumed nominal properties of gerunds responsible for the inability of licensing structural Case. There is, in fact, a simpler solution that assumes that zu is directly (rather than indirectly via selection of a gerund) involved in blocking Case-assignment to the direct object. From the discussion of Case-assignment in passives we know that the presence of a little v itself is not suﬃcient to license Accusative Case (rather it implies the projection of Agr1). Instead, I argued that the Case-assigning properties of Agr3 depend on the aspectual properties of the verb, that is to say, on the properties of material in the Aspect-head. Along these lines, it is plausible that zu in the Aspecthead blocks the Agreement relation between the Agr3 head and little v when it is not temporally linked. I will evaluate these two options when I discuss the syntax of gerunds in more detail in chapter 7. To conclude, the Case-assigning properties of coherent to-inﬁnitives diﬀer from the Case-assigning properties of a sentential to-inﬁnitive because in one case we are dealing with an inﬁnitival marker plus inﬁnitive and in the other case we are dealing with a preposition-like element that selects a gerund. I think this proposal contains enough immediate plausibility as to render this account of coherent to-inﬁnitives an interesting new contribution to the discussion of the proper treatment of coherent inﬁnitives. Having motivated the assumption in (45), it becomes clear why the class of verbs that take coherent to-inﬁnitives is restricted in the way it is. The matrix predicate must provide a controller for the embedded PRO-Subject as well as a Case-licenser for the embedded direct object. Thus, only subject control verbs where the object position is free and object control verbs with a Dative controller can license coherent to-inﬁnitives (cf. [31] earlier).

COHERENT INFINITIVES IN GERMAN

5.6

159

Conclusions

To summarize the chapter, I have shown that a biclausal approach can account for all the idiosyncratic properties of coherent to-inﬁnitives in a rather elegant manner. I have also shown that explaining the totality of the binding facts within coherent toinﬁnitives requires (1) the presence of the full functional structure on the part of the inﬁnitival complement—in particular, it must be assumed that the inﬁnitival complement contains a TP and PRO—and (2) a fairly sophisticated theory of restructuring, as I have put it forward in terms of movement of the inﬁnitival AspP and of the inﬁnitival TP and head movement of the inﬁnitival Tense-head, as well as a sophisticated theory of reconstruction. To explain the Case-licensing properties of coherent to-inﬁnitives I have proposed that these inﬁnitives are in fact nominalized verbs that fail to license structural Case.

160

SCRAMBLING , REMNANT MOVEMENT , AND RESTRUCTURING IN WEST GERMANIC

6

Verbal Complexes and the Syntax of IPP-Complements

I

n chapter 3 (section 3.2), I concluded that verb-particles are licensed by undergoing XP-movement into [Spec,AspP] rather than by head movement that adjoins them to the selecting verb. In chapter 4, I argued that VR-constructions in German and Dutch must be described as involving movement of the inﬁnitival AspP and TP into the C-domain of the inﬁnitival clause followed by additional movement of these constituents into licensing positions in the matrix clause. In particular, I have proposed that the inﬁnitival TP moves into [Spec,PredP] of the higher verb and that the inﬁnitival AspP moves into a licensing position of the higher verb below PredP that I have not identiﬁed yet. The purpose of this ﬁnal step in the licensing of dependent verbs is to check the (exact) subcategorization of the selecting verb and to temporally localize the embedded event with respect to the matrix event. In non-restructuring clauses the embedded event is localized via the local TP, which in turn is assigned a reference time dependent on the matrix event time, as I have argued in chapter 4 (section 4.4.2). In restructuring clauses, the embedded TP is not temporally linked to the matrix event time and the embedded event is then directly localized with respect to the matrix event by movement of the inﬁnitival Aspect Phrase into a checking position of the matrix verb. For this local relationship, any of the functional positions in the V-domain of the selecting verb, illustrated in (1) later, seems appropriate. In sum, further movement of the embedded Aspect Phrase serves two purposes:

160

VERBAL COMPLEXES AND THE SYNTAX OF IPP - COMPLEMENTS

161

(1) checking of the subcategorization of the matrix verb and (2) temporal linking with the matrix verb. Another issue that is in need of further explication is the question of whether the ﬁnal step in the licensing of dependent verbs in restructuring contexts also constitutes XP-movement. In this chapter, I will take a closer look at the formation of verb clusters and investigate the positions that are relevant for licensing nonﬁnite verbs. In chapter 4, I have established that all VP-internal material, including CP-complements, has to move out of the VP to be licensed in speciﬁc licensing positions. On the basis of to-inﬁnitives, I have concluded that CP-complements are licensed in [Spec,F3P] below F2, the position I had assumed that the verb in embedded clauses moves to. To remind us of the pertinent structures, let us have a look again at example (19) of chapter 4, the relevant part of which is repeated for convenience in (1a). (1)

a. [AspP

zu [F2P

b. [AspP (V2) zu [F2P (V2) c. [AspP

te [F2P

V

[F3P

CP

[VP . . .]]]]

V1 [F3P

CP

[VP . . .]]]]

V1 [F3P

V2

[VP . . .]]]]

Assuming that the CP in [Spec,F3P] is the CP-complement of a restructuring verb, movement of the dependent verb to the left of the selecting verb will derive a left-branching verb cluster if it is spelled out in the higher copy, as is typical for German (1b), or yield a right-branching verb cluster if it is spelled out in the lower copy, as is typical for Dutch (1c). Given that the matrix verb is spelled out in the head position of F2, the dependent inﬁnitive could in principle be taken to be licensed in [Spec,F2P] or in [Spec,AspP] in German. In this chapter, we will look at diﬀerent types of data from the West Germanic languages that can help us settle the issue of what the licensing positions of the diﬀerent nonﬁnite verbs in restructuring contexts are. I will present arguments to indicate that F2P is responsible for temporal linking, while the subcategorization of the matrix verb can be checked either in [Spec, F2P] or in [Spec,AspP]. While the surface data (in German) suggest that dependent verbs in restructuring contexts are licensed in [Spec,AspP] of the selecting verb, there is evidence from the syntax of IPP-complements in West Flemish and Afrikaans that (at least) participles move through (and sometimes remain there) the Speciﬁer of F2P below AspP. With the help of Frisian data, I will establish that the West Germanic dialects have two types of inﬁnitives, one being directly licensed in AspP, the other, like participles, moving through [Spec,F2P] below AspP, thereby giving rise to the IPPeﬀect. Finally, I will discuss the derivation and distribution of the diﬀerent types of verb clusters in German, Dutch, and West Flemish. But ﬁrst, I would like to establish that movement of the dependent inﬁnitive into the domain of the licensing verb is indeed XP-movement, as I have simply assumed so far, by showing that a headmovement account cannot explain the properties of German (left-branching) verb clusters.

162

SCRAMBLING , REMNANT MOVEMENT , AND RESTRUCTURING IN WEST GERMANIC

6.1 Left-branching verb clusters in German In this section, I will argue on the basis of the syntax of verb clusters in German that left-branching verbal complexes, like right-branching verbal complexes, are created by XP-movement. I will discuss two phenomena that strongly indicate that an XPmovement account is to be preferred over a head-movement account. Hence I will adopt an XP-movement account and conclude this section with a note on the distribution of CP-complements in left-branching verb clusters. The ﬁrst phenomenon concerns the behavior of particles in German verb clusters. Particles in German have a very restricted distribution in the verb cluster: They may only occupy a position immediately preceding the selecting verb. Remember that Dutch particles enjoy a rather free distribution in their verb clusters (cf. [2a–b]). (2)

a. (Dutch)

dat hij mij (weg) zou (weg) kunnen (weg) horen (weg) rijden that he me (away) would (away) can (away) hear (away) drive ‘that he would be able to hear me drive away’

b. (German) dass er mich (*an) wird (*an) haben (an) rufen (*an) wollen that he me (up) will (up) have (up) call (up) want-IPP ‘that he may very well have wanted to call me up’ c. (German) *dass er mich wird haben rufeni wollen [CP [F1P [XP an] ti]] that he me may very well have call want up

I have accounted for the distribution of particles in the Dutch verb cluster by assuming that in Dutch particles are licensed in [Spec,AspP]. Thus, they will be piedpiped by movement of AspP into the C-domain in restructuring contexts, making them part of the verb cluster. Within the verb cluster they may move via XP-movement into [Spec,AspP] of the higher verb. The same holds for small clause predicates in Dutch provided that they are unmodiﬁed. If they are modiﬁed, small clause predicates are licensed in [Spec,PredP] of the selecting verb and will thus be pied-piped by TP-movement in restructuring contexts, which lands them in front of the matrix verb, that is, in front of the entire verb cluster. That the particle may not follow its selecting verb German shares with Dutch, but contrary to Dutch, the particle in German may also not climb up in the verb cluster, as indicated in (2b). Let us assume for the sake of the argument that left-branching verb clusters in German are derived by head movement of the dependent inﬁnitive. In this approach, we could assume that the dependent verb left-adjoins to the selecting verb in the head position of F2P or of the Aspect Phrase. It is not clear in this account what prevents the particle in (1b) from climbing up in the verb cluster by moving it into the Speciﬁer of the Aspect Phrase of the selecting verb, as they arguably do in Dutch. Assuming for the moment that German right-branching verb clusters are formed like their Dutch counterparts by spelling out the lower copy of the dependent verb, the ungrammatical cases of (2b) cannot be excluded. Also, this account cannot derive the only grammatical order in (2b). What it derives is (2c), which is ungrammatical.

VERBAL COMPLEXES AND THE SYNTAX OF IPP - COMPLEMENTS

163

In order to derive (2b) from (2c) the particle would have to undergo movement on its own. We arrive at a contradiction. With (2b), I concluded that the particle in German, unlike its Dutch counterpart, cannot undergo movement separately of the inﬁnitival verb. This additional movement of the particle can be avoided if we assume either that particles in German are licensed in PredP or that in (2c), instead of moving only the verb as head, the whole AspP is moved from [Spec,SP] in the C-domain into [Spec,AspP] of the higher verb. The ﬁrst option is untenable. If particles are licensed in PredP, then they are not aﬀected by AspP-movement into the C-domain but are pied-piped by TP-movement into PredP of the higher verb. This implies that an auxiliary that precedes the particle must have moved high up (beyond PredP) in the tree. Furthermore, we expect that an adverb that precedes the auxiliary in this (high) position should only have matrix scope, because adverbs that modify the embedded verb are contained in PredP of the matrix verb. As the ambiguity of the adverb in (3) shows, this expectation is not borne out. (3)

weil Hans die Marie oft hat anrufen wollen since Hans the Maria often has up-call want-IPP ‘since Hans has often wanted to call up Maria’ ‘since Hans has wanted to often call up Maria’

This leaves us with the second option, which also accounts for the restricted distribution of particles in German verb clusters. If dependent inﬁnitives in German are licensed by movement into [Spec,AspP] of the selecting verb, then it is clear that this Speciﬁer is not available for further movement of the particle of the dependent verb as it is in Standard Dutch. This, in turn, implies that Dutch inﬁnitives cannot be assumed to be licensed in [Spec,AspP] and must thus be taken to be licensed in [Spec,F2P] below. I will come back to this issue at the end of this chapter. The second phenomenon concerns to-inﬁnitives. I have argued extensively in chapter 4 that the sequence to + inﬁnitival verb cannot be analyzed as a head-adjunction structure. Thus, movement of the to-inﬁnitive in (4a) in front of the matrix verb must involve XP-movement. So at best we could have a mixed system: To-inﬁnitives undergo additional XP-movement, but bare inﬁnitives undergo additional head movement. (4)

a. weil Hans das Buch [AspP zu [F2P lesen]] versprach since Hans the book to read promised b. *ohne das Buch [AspP zu [leseni wollen [CP ti]]] without the book to read want c. ohne das Buch [lesen]i [AspP zu [F2P wollen [CP ti]]] without the book read to want ‘without wanting to read the book’

However, this mixed system leads to problems whenever a verb cluster comprises a bare inﬁnitive and a to-inﬁnitive. First of all, the case of a bare inﬁnitive selected by a

164

SCRAMBLING , REMNANT MOVEMENT , AND RESTRUCTURING IN WEST GERMANIC

to-inﬁnitive shows that bare inﬁnitives cannot be taken to head-adjoin to the selecting verb. If this were possible, then (4b) should be grammatical, contrary to fact. As (4c) shows, the bare inﬁnitive, since it precedes the inﬁnitival marker, can only be taken to head-adjoin to the inﬁnitival marker of the selecting verb. Note, however, that the dependent inﬁnitive in (4c) cannot have reached its surface position via head movement, since it cannot skip the intervening verb. The position of the dependent inﬁnitive in (4c) also suggests that bare inﬁnitives are licensed in [Spec,AspP] of the selecting verb. A problem similar to the problem illustrated in (2c) arises if a to-inﬁnitive is embedded under a verb that itself is embedded under a modal verb that selects a bare inﬁnitive. In this scenario, the to-inﬁnitive would in the ﬁrst cycle undergo XPmovement to a position in front of the selecting verb. The resulting verb cluster would be moved in the second cycle into the C-domain of the clause of the bare inﬁnitive (below the modal). If bare inﬁnitives were licensed by undergoing X0-movement to the selecting verb, then we would expect that the to-inﬁnitive selected by the verb in bare inﬁnitival form would be stranded to right of the matrix verb, as is shown in (5a). (5)

a. weil Hans versprecheni wollte [CP [AspP [zu kommen] ti]] since Hans promise wanted to come b. weil Hans [zu kommen versprechen] wollte since Hans to come promise wanted ‘since Hans wanted to promise to come’

Example (5a), contrary to (2c), is not ungrammatical but appears to be diﬀerent from the standard construction given in (5b). The order of verbs in (5a) is reminiscent of cases of Remnant Extraposition in Dutch. As Kroch and Santorini (1990) point out, the verb sequence in cases of Remnant Extraposition may be interrupted by sentence adverbs like nicht ‘not’ or doch ‘after all’. This is illustrated in (5c–d). Example(5a) may thus have an entirely diﬀerent derivation from (5b). Whether sentences of the type of (5c–d) can be analyzed as Cases of VPR I will have to leave open here.1 (5)

c. weil ich seinen neuesten Roman beschlossen habe, nicht zu lesen since I his newest novel decided have not to read ‘since I have decided not to read his most recent novel’ d. weil ich seinen neuesten Roman beschlossen habe, doch zu lesen since I his newest novel decided have to read after all ‘since I have decided to read his most recent novel after all’

The important point here is that (5b), the standard VR-construction, cannot be derived under the preceding account without postulating an additional movement operation for the to-inﬁnitive, which seems unmotivated. I will thus assume that left-branching verb clusters are derived via XP-movement of the dependent inﬁnitive into [Spec,AspP] of the higher verb. In the following, I will brieﬂy outline the consequence of this account for the distribution of CP-

VERBAL COMPLEXES AND THE SYNTAX OF IPP - COMPLEMENTS

165

complements in V(P)R-constructions. I have argued that CP-complements, contrary to other arguments of the verb, are licensed within AspP of the selecting verb. As such they will be pied-piped by movement of AspP into the C-domain of the inﬁnitival clause. If the entire phrase, as I have argued is moved into [Spec,AspP] of the selecting verb, we derive structures of the type of (6a), which are ungrammatical. In order to derive the grammatical (6b) we have to assume that CP-complements move out of the containing AspP at some point in the derivation. In the standard SOV approach this fact is accounted for by assuming that CP-complements are extraposed, that is, right-adjoined to VP or TP. So far, I have not addressed the issue of how extraposition can be accounted for in the antisymmetric approach. I will take up this issue in chapter 7, which deals with the complex interaction between verb cluster formation and extraposition. (6)

a. *weil Hans [[AspP zu sagen [CP dass er krank war]] versuchte] since Hans to say that he sick was tried ‘since Hans tried to say that he was sick’ b. weil Hans [[AspP zu sagen] versuchte] [dass er krank war] since Hans to say tried that he sick was

6.2 The syntax of IPP-constructions As I mentioned in chapter 1 (section 1.3.3), the (IPP)-eﬀect occurs when a restructuring verb, for example, a modal verb, that selects an inﬁnitival complement (the dependent inﬁnitive) is used in a perfect tense. In this case, the selecting verb does not show up in its participial form but is realized as a bare inﬁnitive (the IPPinﬁnitive). This is illustrated for Dutch in (7) and German in (8), where the inﬁnitival forms willen/wollen replace the participial forms gewild/gewollt of the modal verb want. (7)

a. *dat Elsje hem een brief heeft gewild schrijven that E him a letter has wanted-PART write b. dat Elsje hem een brief heeft willen schrijven that E him a letter has want-INF write ‘that E has wanted to write him a letter’

(8)

a. *dass Else ihm einen Brief schreiben gewollt hat that E him a letter write wanted-PART has ‘that E has wanted to write him a letter’ b. *dass Else ihm einen Brief schreiben wollen hat that E him a letter write want-INF has c. dass Else ihm einen Brief hat schreiben wollen that E him a letter has write want-INF

166

SCRAMBLING , REMNANT MOVEMENT , AND RESTRUCTURING IN WEST GERMANIC

As the contrast between (7b) and (8b) shows, in German it is not suﬃcient to simply replace the participle with a bare inﬁnitive, as it is in Dutch. In order to yield a grammatical sentence in German, the auxiliary has to invert with the cluster comprised of the dependent inﬁnitive and the IPP-inﬁnitive (8c). I will give an account of inversion in German in this context when I talk about the internal syntax of IPP-complements. The interesting issue that the IPP-eﬀect raises is the question of whether IPPinﬁnitives are real inﬁnitives or hidden participles of some sort. Most notoriously, Jakob Grimm (1899/1969, 195) put forth the hypothesis that the IPP-inﬁnitive is a preﬁxless participle. I will adopt the hidden participle account for the following reasons: First, based on the distribution of participles, inﬁnitives, and IPP-complements in West Flemish and Afrikaans, I will argue in section 6.2.1 that IPP-complements behave like participles and unlike inﬁnitives. Second, the hidden participle account allows us to assume that for the purpose of checking the subcategorization of the auxiliary (which selects for a participial phrase) and for the purpose of temporal interpretation the IPP-inﬁnitive counts as a participle. 6.2.1

The external syntax of IPP-complements

In chapter 3 (section 3.2), I discussed in detail the distribution of participles, inﬁnitives, and IPP-complements in West Flemish, concluding that IPP-complements behave like participles rather than inﬁnitival complements. To remind us of the essential regularities in West Flemish, (9) summarizes the distribution of participles, inﬁnitives, and IPP-complements with respect to the selecting verb and the inﬁnitival marker te. (9)

participle

te

IPP

verb

inﬁnitival complement

To minimize the diﬀerence between participles and IPP-complements (hidden participles) I propose that participles are not moved in one swoop from their base position to the right of the selecting verb to their surface position but that they, like IPP-complements, ﬁrst undergo XP-movement to [Spec,F2] of the selecting auxiliary (to check its subcategorization) and subsequently undergo further XP-movement to AspP. The latter (additional) movement of participles must be triggered by the participial morphology that IPP-inﬁnitives lack. That participles indeed undergo this complex two-step movement in West Flemish and the other West Germanic languages is supported by the behavior of participles in Afrikaans. In this language, the intermediate step of the complex movement of participles, unevidenced in all other West Germanic languages, is displayed. As is illustrated in (10), the participle is spelled out between the inﬁnitival marker and the inﬁnitival verb. Example (10) also provides the ultimate conﬁrmation for my hypothesis that IPP-inﬁnitives are hidden participles: In Afrikaans, participles and IPPinﬁnitives have exactly the same distribution. Compare (10) and (4b) from chapter 4 (section 4.1), repeated here as (11). (10)

Jy behoort die lig af te geskagel het You ought the light oﬀ to turned have ‘You should have turned the light oﬀ’

VERBAL COMPLEXES AND THE SYNTAX OF IPP - COMPLEMENTS

(11)

167

Die banke moes oop gewees het, om dit gister te [kan betaal] het the bank should open been have it yesterday to can-IPP buy have ‘the bank should have been open to have been able to buy it yesterday’

6.2.2

The internal syntax of IPP-complements

In this section, I provide an account of the IPP-eﬀect, which I have left unexplained so far. I will also explain why participles and IPP-inﬁnitives pattern exactly alike in Afrikaans but have a slightly diﬀerent distribution in the other West Germanic languages, as was illustrated earlier for the case of West Flemish. The IPP-eﬀect occurs in restructuring contexts. One important feature of restructuring is the formation of verb clusters. I propose that the formation of verb clusters is motivated by the need of the dependent inﬁnitive, due to a defective complementizer, to check the subcategorization of the selecting verb. In section 6.1 earlier, I have discussed evidence that dependent inﬁnitives are licensed in [Spec,AspP] of the selecting verb in German. Here I will assume that inﬁnitives, like participles, move through [Spec,F2P] to [Spec,AspP] of the selecting verb. This assumption will be motivated in 6.2.3 on the basis of Frisian data. Following Bech (1955/1983), I assume that a verb selects for the status of its nonﬁnite complement. That is, it determines whether the dependent nonﬁnite verb is a participle, a bare inﬁnitive, or a to-inﬁnitive. To explain the IPP-eﬀect, I will make use of the particular structure of participles in West Germanic. Note that the languages and dialects in which the participle is formed without the participial preﬁx ge, namely, Frisian and Low German, do not display an IPP-eﬀect (cf. also vanden Wyngaerd [1994, 1996]). In the following I will show how the IPP-eﬀect can be reduced to a structural incompatibility between the participial preﬁx and the inﬁnitive dependent on the restructuring verb. In the West Germanic languages that display the IPP-eﬀect, the participle is formed by aﬃxation of the preﬁx ge and the suﬃx t/d. I follow Halle and Marantz (1993) in assuming that inﬂected forms are (partially) derived in the syntax. More speciﬁcally, I propose that the participial preﬁx ge is inserted in [Spec,F2P] of the participial phrase. The verb in the participial phrase will ﬁrst move to F2, to check its preﬁx, and then up to the Aspect-head to merge with the suﬃx that contains the temporal interpretation. In the ﬁnal step the preﬁx left-adjoins to the complex of verb and suﬃx to form the participle before Spell-out. This is illustrated in (12). The participial preﬁx originally was an optional derivational morpheme separable from the verb that has become grammaticalized as an obligatory marker of the participle. In this account the preﬁx is base generated as an XP in the Speciﬁer of a functional head. Since the preﬁx is both a maximal and a minimal projection it can incorporate into the inﬂected verb in the higher head position. (12)

[AspP -t [F2P [ge] [F2 [VP V]]]]

If the verb in the participle phrase is a restructuring verb, then the dependent inﬁnitive will move into [Spec,F2P] in the course of the derivation. It follows that a verb in participial form and a bare inﬁnitive selected by such a verb rule each other

168

SCRAMBLING , REMNANT MOVEMENT , AND RESTRUCTURING IN WEST GERMANIC

out. In this case the participial preﬁx is blocked by the dependent inﬁnitive, that is to say, it cannot be inserted. The question now arises of why the pertinent verb is realized as an inﬁnitive and not as a preﬁxless participle.To address this question it is useful to consider also the diachronic development of participles in general and that of restructuring verbs in particular. In a synchronic account, one could assume that the blocking of the preﬁx leads to a violation of a selectional requirement, namely, of the requirement that a participle consist of a suﬃx and a preﬁx. The grammaticalization of the preﬁx, which initially was selected by the (aspectual properties of the) verb stem such that an imperfective verb would form its participle with the preﬁx but perfective verbs without it, can be accounted for by assuming that currently the preﬁx is selected by the participial suﬃx, which requires to combine with a verb stem and a preﬁx. In this account we can assume that the violation of this requirement is avoided in that (1) no phonological material is inserted in the Aspect-head that would require a preﬁx, that is, assume that the suﬃx is dropped, and (2) the verb remains in F2 and is spelled out with the default morphology of a bare inﬁnitive. Instead a zero-morpheme is inserted in the head of AspP that contains the formal feature [+participle] and a semantic feature [+ PAST] (or the condition that event time precede reference time in a Reichenbachian system). This is illustrated in (13). (13)

[AspP 0 [F2P [dependent inﬁnitive] IPP-inﬁnitivei [VP ti]]]

From a diachronic perspective, it is important to note that modal verbs, which form a large portion of today’s VR-verbs that trigger the IPP-eﬀect, belonged to the class of strong verbs in Old High German and became weak verbs rather late, when compared to other verbs, at the end of the Middle High German period. In this scenario it is tempting to analyze IPP-inﬁnitives as related to preﬁxless participles. Note that the participial ending of strong verbs is en, which is identical to the inﬁnitival ending. Thus, with some verbs, depending on the ablaut-class of the stem, inﬁnitive and preﬁxless participle were probably indistinguishable, as is illustrated in (14). (14) a. sehen-sah-(ge) sehen b. lassen-ließ-(ge) lassen

(inﬁnitive, past tense, and participle of see) (inﬁnitive, past tense, and participle of let)

c. Inﬁnitive and past participle of the modals in OHG können (‘can’)-cunnan-(gi)cunnan dürfen (‘may’)- durfan-(gi)dorfan sollen (‘shall’)-sculan-(gi)scolan mögen (‘want’)-magan-(gi)magan müssen (‘must’)-muozan-(gi)muozan

However, it is not known whether the strong forms of modals were still productive when the periphrastic perfect, which gave rise to the modern IPP-construction, had become fully grammaticalized in MHG. In some dialects, including my own, strong participles have been preserved, as is illustrated in (15). In (15a), the position of the auxiliary suggests that the form wollen should not be analyzed as an IPP-inﬁnitive but

VERBAL COMPLEXES AND THE SYNTAX OF IPP - COMPLEMENTS

169

as a preﬁxless strong participle, an analysis that is corroborated by the appearance of the same form in a nominal context that does not license IPP-inﬁnitives. (15) a. weil er das Buch lesen wollen hat since he the book read want-Inf?/Part? has b. weil er einen Kuchen wollen hat since he a cake want-Part has

If strong participles were still productive—if only in dialects and the spoken language—at the time of the introduction of the IPP-construction, then we could assume that when the new participial forms that ended in -t were introduced, possibly ﬁrst as optional variants, replacing the strong form in more and more contexts, preﬁxless participles in the IPP-context may have become too isolated to be recognized as participles and hence were reanalyzed as inﬁnitives. Note, however, that this scenario, though plausible, must remain a mere speculation, for several reasons. First of all, the constructed participial forms in (14) are not attested in texts. Only three participial forms of the class of preterito-presentia, which modal verbs belonged to, are attested in OHG texts: giwizzan (‘known’), vercunnan (‘been able’), and gitorran (‘dared’). Moreover, periphrastic perfects were not yet grammaticalized in OHG times, implying that the source-construction for an IPP-inﬁnitive did not exist in OHG. In MHG times, the periphrastic perfect had fully grammaticalized, but modals in this period lacked a participle according to MHG grammars. It is not clear whether this statement means that they still lacked the new weak form of the participle that developed late with modals or whether modals lacked a participle altogether. In the former case, a development as sketched in the scenario described earlier can be assumed. Note that even in this scenario, IPP-inﬁnitives today cannot be interpreted as (leftover) preﬁxless strong participles since the modals can and must in IPP-contexts are marked with the umlaut that is conﬁned to the morphological forms derived from the present stem (cf. können and müssen instead of konnen and mussen). In the latter case, we must assume that the inﬁnitive was used as a suppletive form, since no participial form whatsoever was available. More research on the development and usage of modals in the MHG period is in order to shed some light on the development of IPP-inﬁnitives, a type of research that is beyond the scope of this book. Nevertheless, the diachronic dimension is important since in both scenarios it can provide a reasonably good explanation of why the participle in restructuring contexts was replaced with an inﬁnitive: At the time, the construction can be assumed to have developed modals either lacked participles altogether or displayed participles that looked like inﬁnitives. Independently of its diachronic conditioning, IPP-inﬁnitives synchronically must be assigned a structure like (13), with the verb being spelled out with inﬁnitival morphology in F2 and the presence of an additional morpheme in the Aspect-head that guarantees the correct temporal interpretation of the IPP-inﬁnitive, since synchronically modals do possess participles that are formally quite diﬀerent from IPP-inﬁnitives. Given the structure in (13), we can assume that further movement of true participles is triggered by the need to move the semantic feature of the participle up to the

170

SCRAMBLING , REMNANT MOVEMENT , AND RESTRUCTURING IN WEST GERMANIC

Aspect-head of the auxiliary to be linked with the tense of the auxiliary. These assumptions derive the two-step movement process of participles. First the participle phrase moves into [Spec,F2P] of the auxiliary to check its status (its subcategory) with the auxiliary. Then it moves up into [Spec,AspP] to be linked with the tense of the auxiliary. The latter movement remains invisible in the case of an IPP-inﬁnitive. Let us look at a concrete derivation. In the derivation of the West Flemish example in (16a), the participle phrase will ﬁrst move into [Spec,F2P] below the inﬁnitival marker of the selecting verb een. Then movement of the AspP of the participle to the AspP of the selecting verb will move the participle in front of the inﬁnitival marker. In the case of an IPP-inﬁnitive, the latter step will only aﬀect a phonologically empty morpheme, leaving the IPP-inﬁnitive that is spelled out in F2P in the participial phrase behind. Stranding of the IPPinﬁnitive is only possible if the empty morpheme can be taken to undergo head movement (while the participle undergoes XP-movement) into the Aspect Phrase. The diﬀerence follows if we assume that only heads that contain phonological material can induce movement that pied-pipes the entire phrase. This stipulation is supported by the observation that [Spec,AspP] of the auxiliary is available for other material to move in, as will be shown later. (16) a. mee Valere dienen book [AspP gekocht te [F2P [AspP gekocht [F2P]] een]] with Valere that book bought to have b. mee Valere dienen book [AspP 0 te [F2P [AspP 0 [F2P willen kuopen]] een]] with Valere that book to want-IPP buy have

Returning to obligatory inversion with the IPP-complement of the auxiliary in German (cf. [8b–c]), I propose that ﬁnite and nonﬁnite verbs in German, contrary to West Flemish, always move up to the head of the Aspect Phrase, as is illustrated in (16c).2 If the IPP-inﬁnitive is stranded in [Spec,F2P] of the auxiliary, as I have argued earlier, it follows that the auxiliary precedes the inﬁnitives in German (16c). Example (16c) also clearly demonstrates a case of leftward V-movement in German. (16) c. dass Else ihm den Brief [AspP hati [F2P [AspP 0 [[schreiben] wollen]] ti [VP ti]]] that Else him a letter has write want-IPP d. dass Else ihm den Brief [AspP [AspP schreiben]i [Asp hat [F2P ti wollen]]] that Else him the letter write has want-IPP

The analysis in (16c) is supported by data that indicate that the Speciﬁer of the Aspect Phrase of the auxiliary is indeed available for other elements in this case. Bavarian dialects display the famous V3 V1 V2 order with IPP-inﬁnitives, which can be analyzed, parallel to the movement of particles in Dutch, as movement of the dependent inﬁnitive into the Speciﬁer of the higher AspP, as is illustrated in (16d). A similar behavior can be seen in Afrikaans, which ﬁlls the highest AspP with the particle that belongs to the suﬃxless participle that stays behind in [Spec,F2P], as we have seen in (10) earlier.

VERBAL COMPLEXES AND THE SYNTAX OF IPP - COMPLEMENTS

171

Why then do IPP-inﬁnitives and participles behave alike in Afrikaans while they diﬀer in their distribution in the other West Germanic languages? Note that verbs in Afrikaans have lost all their endings. In particular, participles, while retaining the gepreﬁx, have lost their d/t-suﬃx. It stands to reason that participles in Afrikaans, like IPP-inﬁnitives in general, contain an empty morpheme in the Aspect-head, movement of which will fail to pied-pipe the participle in F2P below. Thus, I have assimilated my account of the IPP-eﬀect in West Germanic to an independent fact in one of the West Germanic languages, namely, Afrikaans. In both cases, we ﬁnd the morphology of the participle to be defective. In the case of IPP-inﬁnitives, this is caused by the blocking of the preﬁx part of the participial morphology. In the case of the participle in Afrikaans, this is due to the general loss of verbal endings in this language. We also have now an explanation for why it is that verbal elements in West Germanic do not normally appear between the inﬁnitival marker and the inﬁnitival verb (cf. in German and Dutch). The explanation is that only nonﬁnite verbs with defective morphology will remain there, while all others will just move through [Spec,F2P] on their way to the Aspect-head of the selecting verb. To conclude, the behavior of participles in Afrikaans provides good independent evidence for this account of IPP-inﬁnitives in West Germanic. 6.2.3

What Frisian may teach us about the IPP-eﬀect

It is a peculiarity of perception verbs in Standard German that they can optionally trigger an IPP-eﬀect, as is illustrated in (17a–b). This case is to be distinguished from the case of the presence and absence of the IPP-eﬀect with the permissive/causative verb lassen (‘let’, ‘make’). Example (17c) exhibiting the IPP-eﬀect is clearly distinct in meaning from (17d). While (17d) can only mean ‘he allowed her to continue to sing’ (presupposing that she was already singing), (17c) can mean ‘he caused her to start to sing’ (presupposing that she did not sing already). In (17a–b), however, no similar or for that matter, any other diﬀerence in meaning can be observed between the a and b examples to the best of my knowledge. (17) a. weil Hans die Maria nicht hat kommen hören/sehen since Hans the Maria not has come hear/see-IPP ‘Hans did not hear/see her come’ b. weil Hans die Maria nicht kommen gesehen/gehört hat since Hans the Maria not come seen/heard has c. weil Hans die Maria hat singen lassen since Hans the Maria has sing let-IPP ‘Hans caused Maria to sing’ d. ?weil Hans die Maria singen gelassen hat since Hans the Maria sing let-PART has

The optionality in (17) is problematic for any account of the IPP-eﬀect. However, I will provide another argument (see chapter 5 for an initial argument derived

172

SCRAMBLING , REMNANT MOVEMENT , AND RESTRUCTURING IN WEST GERMANIC

from the syntax of coherent to-inﬁnitives) that German has two types of inﬁnitives, which are accidentally homophonous and only one of which triggers the IPP-eﬀect. Contrary to German, Frisian also formally distinguishes two inﬁnitives: the regular inﬁnitive ending in e, called the Nammefoarm, and the so-called Doelfoarm, ending in en. About Doelfoarms Tiersma (1985) makes the following statement in his Frisian Reference Grammar: “These forms may function as verbs, in which event they are very similar to inﬁnitives; or as nouns, in which case they are gerundives.”3 Now, it is interesting to note that the group of verbs that select the Doelfoarm in Frisian largely overlaps with restructuring verbs that do not trigger an IPP-eﬀect in German. According to Wolf (1997), the two inﬁnitives are in complementary distribution in Standard Frisian. Doelfoarms occur, among other places, after the inﬁnitival marker te, in the complement of perception verbs, and in the complement of bliuwe (‘remain’). Nammefoarms (the regular inﬁnitives) occur in most other positions, for example, in the complement of litte (‘let’). Example (18) is taken from Wolf (1997). (18) a. Ik kin har der rinnen sjen b. *Ik kin har der rinne sjen I can her there run see c. *Ik sil har mar restich lezen litte d. Ik sil har mar restich leze litte I will her just calmly read let

Wolf (1997) also notes that in Dutch both inﬁnitives are used as well, but that they can freely alternate. Compare (18) with (19). (19) a. Ik kan haar daar zien lope(n) I can her there see run b. Ik zal haar maar rustig laten leze(n) I will her just calmly let read

This suggests that the (morphological) distinction between Doelfoarm and Nammefoarm has been lost in Dutch. Since the IPP-eﬀect is so pervasive with VR-verbs in Standard Dutch, it stands to reason that these speakers analyze both phonetic forms as Nammefoarms (the two endings are analyzed as free allomorphs of the morpheme that represents the Nammefoarm). In Standard German, there is only one inﬁnitival form that ends in (e)n. We may assume, however, that the ending (e)n is a homomorph that represents two diﬀerent inﬁnitival morphemes, one that has the feature [+D] (for Doelfoarm) and one that has the feature [–D] (for Nammefoarm). Then we can assume that perception verbs in Standard German are special in that they can select both types of inﬁnitives.4 Lassen (‘let’) in one of its meanings, namely, ‘allow to remain’, selects the German version of the Doelfoarm, and in its other meaning, namely, ‘allow to begin’, it selects the German version of the Nammefoarm. Lassen in its ﬁrst meaning can thus be analyzed as the transitive version of bleiben (‘remain’), which, in German, like in Frisian,

VERBAL COMPLEXES AND THE SYNTAX OF IPP - COMPLEMENTS

173

selects a Doelfoarm. That lassen in its causative meaning only selects a Nammefoarm is consistent with this analysis, since to cause something means bringing about something that was not there before (making something start). If this analysis is correct, then we could assume that Nammefoarms are checked in [Spec,F2P] and Doelfoarms are checked in [Spec,AspP] of the selecting verb. In the ﬁrst case, the dependent inﬁnitive would trigger an IPP-eﬀect (by interfering with the ge-preﬁx in [Spec,F2]). In the latter case, the dependent inﬁnitive would not trigger an IPP-eﬀect. This account then raises the question of why IPP-inﬁnitives are never found with particle verbs. Under this analysis, IPP-inﬁnitives and verb-particles should not be incompatible: The inﬁnitive is licensed in [Spec,F2P] and the particle is licensed in [Spec,AspP] of the selecting verb. In this account, I would have to either stipulate that particle verbs only select Doelfoarms or assume that Nammefoarms also have to move into [Spec,AspP] as well, in order to check some feature. The latter assumption is necessary anyway, since a Nammefoarm selected by a modal always has to precede the inﬁnitival marker in the Aspect-head, as I noted in section 6.1 earlier and is illustrated again in (20). (20) a. *ohne das Buch [AspP zu [F2P [lesen] wollen]] without the book to read want b. ohne das Buch [AspP [lesen] zu [F2P wollen]] without the book read to want ‘without wanting to read the book’

Is there any evidence that Nammefoarms indeed move through [Spec,F2P] rather than moving directly into [Spec,AspP]? De Haan (1992) notes that there is a variety of Frisian, called Dutchiﬁed Frisian or Interference Frisian, henceforth IF, in which the distinction between Nammefoarm and Doelform is in the process of being lost. IF is mostly spoken by young people and seems to be the result of the heavy inﬂuence of the Dutch language on the Frisian-speaking minority in the Dutch province of Frisia (cf. De Haan [1996] and Wolf [1997] for some discussion of the considerable changes in the linguistic situation in Frisia in the past decades). The interesting fact about the recent changes in IF is that concomitant with the loss of the morphological distinction between inﬁnitives, inﬁnitives in IF may5 fail to precede the selecting verb as they do in German and Standard Frisian (SF). Compare the sentences in (21), taken from De Haan (1996). (21) a. SF sunder ferfelend weze te wollen without boring be[–D] to want [+D] ‘without wanting to be boring’ b. IF

sunder ferfelend te wolle_ weze without boring to want be

The inﬁnitives in (21) diﬀer in both their syntactic positioning and their morphological marking. While the inﬁnitives in SF must be analyzed as marked for

174

SCRAMBLING , REMNANT MOVEMENT , AND RESTRUCTURING IN WEST GERMANIC

[+/– D], all the inﬁnitives in IF can be analyzed either as Nammefoarms or simply as unmarked for the feature [+/–D]. De Haan points out that speakers of IF not only produce structures like (21b), which is essentially the Standard Dutch order in a verbal complex that comprises a to-inﬁnitive, but also structures as in (21c), where the dependent inﬁnitive appears between the inﬁnitival marker and the inﬁnitival verb. (21) c. IF

sunder syn auto te meitsje litte without his car to repair let ‘without letting repair his car’

I interpret the occurrence of verbal complexes like (21c) in the following way: Since the order of verbs in (21c) is possible neither in SF nor in Standard Dutch, speakers of IF who produce verbal complexes of the type of (21c) must follow a principle of UG, namely, that inﬁnitives move through [Spec,F2P] at some point in the derivation. Speakers who produce verbal complexes of the type in (21b) and speakers who produce verbal complexes of the type in (21c) only seem to diﬀer in their interpretation of the morphological breakdown of inﬁnitival forms, that is, in their interpretation of whether there is any morphological feature (a non-binary feature) left that needs to be checked overtly. Wolf (1997) in his very interesting statistical investigation of the recent changes in the verbal complex in Frisian notes that structures of the type of (21c) are rather rare compared to structures of the type of (21b). The frequency of structures like (21c) is only between 3 and 10 percent. Assuming that speakers do not make arbitrary mistakes, I treat the occurence of sentences like (21c), despite their low frequency, as signiﬁcant and attribute their low frequency to sociolinguistic factors that lead speakers to settle predominantly for the Standard Frisian or the Standard Dutch order. Given this evidence from IF, I propose that those inﬁnitives in German that correspond to the Nammefoarm in Frisian move through [Spec,F2P] on their way to [Spec,AspP]. This raises the question of which property of these inﬁnitives in German requires this double checking in both [Spec,F2P] and [Spec,AspP]. We can assume that both Nammefoarms and Doelfoarms have to check the subcategorization of the selecting verb. Thus, we should expect that there is a diﬀerence in temporal linking between the two forms. Enç (1986) has argued convincingly that verbs need to be linked with Tense, while nominals are interpreted independently of Tense (this was one of her major arguments against treating Tense as a [sentential] operator).6 If it turns out that it is correct to interpret Nammefoarms as verbal inﬁnitives and Doelfoarms as nominal inﬁnitives (or gerunds), then we can assume that F2P is responsible for the temporal linking of dependent nonﬁnite verbs. Nammefoarms as verbal inﬁnitives can then be taken to move into [Spec,F2P] to be temporally anchored to the matrix event time and move on into [Spec,AspP] to check the subcategorization of the selecting verb, while Doelfoarms as nominalized inﬁnitives move directly into [Spec,AspP] to the check the subcategorization of the matrix verb. This treatment of one type of inﬁnitives as nominalized verbs that do not need (the same type of) temporal anchoring as verbs do can help us understand the diﬀerence in meaning in (17c–d) earlier. I have noted that the singing event that is designated by the nominal inﬁnitive in (17d) is interpreted temporally independently of

VERBAL COMPLEXES AND THE SYNTAX OF IPP - COMPLEMENTS

175

the matrix event. This would suggest that in principle a diﬀerence in meaning is also present in (17a–b), but that this diﬀerence is leveled out by the semantics of a verb of direct perception. I will have to leave this issue for further research. One advantage of this analysis of the checking positions in German verb clusters is that it helps us to consolidate the analysis of coherent to-inﬁnitives given in chapter 5. Recall that I assumed that an inﬁnitival marker zu that is not temporally anchored selects a nominalized inﬁnitive (the gerund). In this analysis, the coherent to-inﬁnitive is directly moved into [Spec,AspP] during restructuring and remains temporally unlinked throughout the derivation (preserving the selectional properties of zu). In sum, the distinction between the Nammefoarm and the Doelfoarm not only can help us explain the presence or the absence of the IPP-eﬀect but also sheds some light on the role of the diﬀerent licensing positions in German verb clusters. In the following chapter I will therefore assume that this distinction is relevant also in (modern) German and will refer to the Nammefoarm and the Doelfoarm as (verbal) inﬁnitive and nominal inﬁnitive or gerund, respectively. To return to the appearance of the IPP-eﬀect, the distinction between the two types of inﬁnitives seems to me to be the best means for accounting for the considerable dialectal variation in the occurrence of the IPP-eﬀect. I have mentioned earlier that Dutch VR-verbs are pervasive in displaying the IPP-eﬀect. However, the situation is diﬀerent in the various dialects. There are dialects where the same verb that displays the IPP-eﬀect in the standard does not display it. Given the distinction between Nammefoarm and Doelfoarm, we may then simply account for this divergence by assuming that these dialects are more conservative in preserving the Doelfoarm.

6.3 Right-branching verb clusters in German While the prevalent order in German verb clusters is the one that results from a strictly left-branching structure, that is, V4 V3 V2 V1 for a four-member cluster, with V1 being the matrix verb, which is illustrated in (22a), right-branching structures are possible as long as the two most deeply embedded verbs show a left-branching structure. This is illustrated in (22b), which shows a verb cluster of the type V1 V2 V4 V3. (22) a. weil er den Text [[[[lesen] können] müssen] wird] since he the text read can must will b. weil er den Text [wird [müssen [lesen können]]] since he the text will must read can ‘since he will have to be able to read the text’

The formation of right-branching verb clusters is also subject to the following condition: Once a right-branching structure is introduced at one level—as is possible as soon as we have a three-member verb cluster as in (23a)—the structure has to be right-branching also at the next level. This is illustrated by the ungrammaticality of the verb cluster in (23b), which shows a left-right-branching order.

176

SCRAMBLING , REMNANT MOVEMENT , AND RESTRUCTURING IN WEST GERMANIC

(23) a. weil er den Text muss lesen können since he the text must read can ‘since he must be able to read the text’ b. ??weil er den Text [[müssen [lesen können]] wird] since he the text must read can will

Given the (restricted) possibility of having right-branching verb clusters in German, it is striking that right-branching verb clusters cannot be topicalized (24a), with the exception of a right-branching cluster that ensues from an IPP-inﬁnitive, as is given in (24b), which only yields a mildly deviant result. Contrary to the situation in German, right-branching verb clusters prevail in Dutch, as we have seen in chapter 3 (section 3.1), and these verb clusters can be topicalized as they are (24c). (24) a. ?*[müssen lesen können] wird er den Text must read can will he the text b. ?[haben lesen wollen] wird er den Text have read want-IPP will he the text ‘he will have wanted to read the text’ c. [hebben willen lezen] zal hij het boek have want-IPP read will he the book ‘he will have wanted to read the book’

The data in (24) suggest that right-branching verb clusters in Dutch diﬀer from rightbranching verb clusters in German. The variation in order illustrated in (22) seems to be of a stylistic nature or, more precisely, of a prosodic nature. Also, the condition that a right-branching order triggers a right-branching order at the next level up seems to be of a prosodic kind in Germanic. Koopman and Szabolcsi (2000) provide a systematic syntactic account of why right-branching verb clusters trigger right-branching in the higher cycle in Hungarian verb clusters. In Germanic, however, such a condition can be overruled if the syntactic derivation requires such an order as is the case with IPPconstructions in West Flemish and Afrikaans, which display the order V2 V3 V1. One way to account for the soft nature of this condition is to assume that (23b) only violates a prosodic condition. Violation of this condition yields a (prosodically) marked structure that is grammatical if there is no alternative derivation but counts as ungrammatical if there is an alternative derivation or Spell-out option available. This condition can be formulated as a constraint on the mapping between syntactic and prosodic constituents, as given in (25a). Furthermore, to prevent verb clusters that are entirely right-branching in Standard German (the situation is diﬀerent in the dialects), the condition in (25b) can be assumed to hold. (25) a. A right-headed phonological phrase in a verb cluster must sit on a right branch with respect to the non-head. b. The most deeply embedded phonological phrase in a verb cluster must be left-headed.

VERBAL COMPLEXES AND THE SYNTAX OF IPP - COMPLEMENTS

177

I have argued earlier that (coherent) inﬁnitives in German ﬁrst move to [Spec,F2P] and then to [Spec,AspP] of the selecting verb, while the selecting verb, whether ﬁnite or not, always moves to the head position of the Aspect Phrase. Given the conditions in (25), we can then assume that German allows dependent inﬁnitives to be spelled out either in [Spec,F2P] or in [Spec,AspP]. Thus, (23a) is analyzed as given in (26a), with the Spelled-out copy in bold letters. The condition in (25a) then ensures that a right-branching verb cluster like (26a) must be spelled out in the lower Speciﬁer included in the right branch of the selecting verb in the next cycle up. The parallel structures of the phonological phrases are given in (26c–d). (26) a. weil er den Text [AspP [lesen können] [muss [F2P[lesen können]]] b. weil er den Text [AspP [müssen lesen können] [wird [F2P [müssen lesen können]]] c. (muss ((lesen) können)) d. (wird (müssen ((lesen) können))) versus * ((müssen ((lesen) können)) wird)

This account in terms of variable Spell-out is supported by historical ﬁndings. Ebert (1980) observes that the order within the verbal complex is almost entirely free in early Modern German, subject only to prosodic constraints that favored sequences that alternated weak and strong elements.7 The current state of aﬀairs can then be seen as the result of a development in which certain prosodic patterns where favored while alternative patterns became increasingly marked. As for the diﬀerence in topicalizability of right-branching verb clusters in German and Dutch, I want to propose that it is the Aspect Phrase, not the VP itself, as proposed in Chomsky (1998, 2001), that constitutes a (strong) phase, implying that extraction out of a verb cluster is only possible via [Spec,AspP], with that Speciﬁer constituting the left edge of the phase. In Chomsky’s theory of phases, the access to the lexicon is a onetime selection of a lexical array LA. LA enters the derivation in diﬀerent steps. In each step a subarray of LA is put in active memory. The syntactic object that is formed when a subarray is exhausted is called a phase. Furthermore, Chomsky assumes that VPs and CPs, but crucially not IPs, are strong phases. A derivation by phases involves a cyclic Spellout of (sub)structures, the point of which is determined by (27). Computation is strictly local and constrained by the Phase Impenetrability Condition as given in (28). (27)

Evaluation for a phase is done at the level of the next highest strong phase.

(28)

Phase Impenetrability Condition (PIC) The complement of a strong phase a is not accessible to operations at the level of the next highest strong phase b, but only the head and the edge of a are.

The term evaluation in (27) is meant to indicate the moment in which Spell-out takes place. After Spell-out only the entire phase or its edges are still accessible for further computation. Given (28), the assumption that the Aspect Phrase constitutes a strong phase will constrain VP-topicalization in the following way.

178

SCRAMBLING , REMNANT MOVEMENT , AND RESTRUCTURING IN WEST GERMANIC

Since a right-branching verb cluster in German is derived via Spell-out in the lower [Spec,F2P], the escape hatch in the Aspect Phrase for a right-branching verb cluster is blocked by its own copy (cf. [26b]). A topicalized right-branching verb cluster, as in (24a), can only be derived if the verb cluster (müssen lesen können) is spelled out in [Spec,AspP] of the matrix verb. Then it can be extracted from the matrix Aspect Phrase and be topicalized without inducing a syntactic violation. However, (24a) is rendered ungrammatical since it adduces a prosodic violation in the course of the derivation. If Spell-out obtains cyclically as determined in (27) earlier, the derivation of (24a) involves a violation of the prosodic condition in (25a) in the cycle of the matrix Aspect Phrase, although it does not violate the condition at the end of the derivation. Since there is an alternative derivation that does not violate the prosodic ﬁlter, namely, the derivation that spells out the cluster lesen können in the higher Speciﬁer in the lower cycle, (24a) is ungrammatical. In the case of the right-branching cluster that involves an IPP-inﬁnitive in (24b), however, an alternative derivation is not available, since the IPP-inﬁnitive only moves to [Spec,F2P] in the lower cycle, as I have argued in the previous section. Since IPPinﬁnitives cannot be optionally spelled out in the lower or higher Speciﬁer, there is no alternative derivation that does not violate the prosodic condition, and since the derivation of (24b) does not involve a syntactic violation, besides the prosodic violation, the result is marked but grammatical. In Dutch, dependent inﬁnitives are licensed in [Spec,F2P] alone (see later discussion) such that the local [Spec,AspP] is (1) available for particles to climb and (2) free as an escape hatch for topicalization, which will be discussed in detail in the following chapter. To sum up, I have provided an account of rather subtle diﬀerences in (23) and (24) that rests on three assumptions: (1) the Aspect Phrase (not the VP) constitutes a phase, (2) Spell-out is cyclic (rather than ensuing at the end of the entire derivation), and (3) there are interface constraints (like the mapping rules between syntactic structure and prosodic structure in [25]) whose violation only leads to ungrammaticality under certain conditions. Finally, right-branching verbal complexes in German provide evidence that Standard German (not only some of its dialects) allows for VPR. First, note that entirely left-branching verb clusters are ambiguous between a VR- and VPR-analysis. Thus the sentence in (29a) can be analyzed as being either the result of VR (29b) or the result of VPR (29c). (29) a. weil er das Buch lesen wollte since he the book read wanted b. [CP weil [er [PredP das Buch [AspP [lesen] wollte]]]] c. [CP weil [er [PredP [AspP [das Buch lesen] wollte]]]]

In the previous section, I have argued that in IPP-constructions the IPP-inﬁnitive stays in [Spec, F2P] below the position to which the temporal auxiliary moves to. In this position, the auxiliary precedes quite naturally manner adverbs and material licensed

VERBAL COMPLEXES AND THE SYNTAX OF IPP - COMPLEMENTS

179

in PredP belonging to the dependent inﬁnitive, as is illustrated in (30a–b). Example (30c) shows that even deﬁnite DPs can undergo VPR in German; (30c) is taken from Haider (1991). (30) a. weil er das Buch hat langsam lesen wollen since he the book has slowly read want-IPP ‘since he wanted to read the book slowly’ b. weil er ihr das Buch hat zur Verfügung stellen wollen since he her the book has to availability put want-IPP ‘since he wanted to make the book available for her’ c. weil er für ihn nicht hat die Firma am Leben halten wollen since he for him not has the company on life keep want-IPP ‘since he did not want to keep the company alive for him’

6.4 Overview of the structure of verb clusters In this section, I want to summarize the structural representations of the diﬀerent verb clusters in German, Dutch, and West Flemish. In the following examples, characters in boldface represent copies of constituents that are spelled out, whereas characters in italic represent copies in checking positions that are not spelled out. Let us start with German. 6.4.1

Verb clusters in German

In German, both ﬁnite and nonﬁnite verbs move into the head position of the Aspect Phrase. A dependent inﬁnitive undergoes checking movement into [Spec,F2P] to be temporally linked and subsequently into [Spec,AspP] to check the subcategorization of the selecting verb. In a left-branching verb cluster the dependent inﬁnitive is also spelled out in [Spec,AspP]. In a right-branching verb cluster, however, the inﬁnitive is spelled out in [Spec,F2P] below. This is illustrated in (31a–b). (31) a. left-branching dependent inﬁnitive in German: lesen wollen wird

AspP1 AspP2

lesen

F2P

wird

AspP3 wollen

AspP2

tAspP3

wollen lesen

180

SCRAMBLING , REMNANT MOVEMENT , AND RESTRUCTURING IN WEST GERMANIC

b. right-branching dependent inﬁnitive in German: wird lesen wollen

AspP1 AspP2 F2P

wird

AspP3 wollen

lesen

AspP2

tAspP3

wollen lesen c. verb cluster that involves a gerund: lesen gesehen hat AspP1 AspP2 F2P

hat

AspP3 lesen-0

gesehen

tAspP2

Contrary to a dependent inﬁnitive, a gerund is directly moved into [Spec,AspP] failing to induce an IPP-eﬀect, as is illustrated in (31c). An IPP-inﬁnitive, like other inﬁnitives, is licensed in [Spec,F2P] of the selecting verb but does not move on into [Spec,AspP]. Only the empty participial morpheme undergoes head movement to the head of AspP. This is illustrated in (32). Since [Spec,AspP] of the temporal auxiliary remains empty, the ﬁnite verb appears in front of the other verbs of the verb cluster in Standard German (Inversion). In Bavarian dialects, this position is regularly ﬁlled with the dependent inﬁnitive (Zwischenstellung), as is illustrated in (32b). (32) a. external and internal structure of an IPP-inWnitive: hat lesen wollen AspP1

F2P

hat AspP2

lesen

Asp 0 [+ participle] [+ past]

F2P t wollen

F3P

VERBAL COMPLEXES AND THE SYNTAX OF IPP - COMPLEMENTS

181

b. IPP-inﬁnitive in Bavarian: lesen hat wollen (Zwischenstellung)

AspP1

F2P

hat

lesen

AspP2 wollen lesen

6.4.2

Verb clusters in Dutch

For Dutch, we must assume that F2P is responsible for the temporal linking as well as for checking the subcategorization with inﬁnitives, since particles can raise in the verb cluster. Since Dutch verb clusters are predominantly right-branching, I assume that dependent inﬁnitives undergo licensing movement into [Spec,F2P] of the selecting verb but are spelled out in the licensing position of the containing CP, that is, in [Spec,F3P] of the selecting verb. The Speciﬁer of the Aspect Phrase in a Dutch verb cluster is empty and can thus be used for the preposing of verb particles, as is illustrated in (33). (33) a. right-branching dependent inﬁnitive in Dutch: (zal) willen uitlezen AspP1 F2P

Asp

F3P

uit lezen willen

uit lezen

b. particle-raising: (zal) uit willen lezen AspP1

uit

F2P

Asp

F3P

uit lezen willen

uit lezen

A dependent inﬁnitive in an IPP-construction is spelled out in [Spec,F3P], while the IPP-inﬁnitive is spelled out in the head position in F2, since the head of the Aspect Phrase, like in German, is occupied by the empty participial morpheme, as is

182

SCRAMBLING , REMNANT MOVEMENT , AND RESTRUCTURING IN WEST GERMANIC

illustrated in (34a). Assuming that Dutch nonﬁnite verbs move at least as high as the head position of F2P, dependent inﬁnitives and IPP-inﬁnitives are analyzed parallel to the position of the inﬁnitive in te-inﬁnitives (cf. [34a–b]). (34) a. verb cluster that contains an IPP-inﬁnitive in Dutch: heeft willen lezen AspP1

Asp heeft

F2P AspP2 F2P 0 lezen

F3P willen lezen

b. verb cluster that contains a te-inﬁnitive: te willen lezen AspP F2P te lezen

F3P willen lezen

In conclusion, Dutch dependent inﬁnitives diﬀer in two respects from their German counterparts. First, they are not licensed in the highest Speciﬁer position in the V-domain, allowing for particles to climb in the verb clusters. Second, given that Dutch nonﬁnite verbs like their West Flemish counterparts do not move to the Aspecthead, a dependent inﬁnitive in Dutch is not spelled out in [Spec,F2P] but in [Spec,F3P] below. This has important consequences as far as the possible topicalization of parts of the verb cluster is concerned, as we will see in chapter 7. 6.4.3

Verb clusters in West Flemish

In West Flemish, as in Dutch, dependent inﬁnitives are checked in [Spec,F2P] of the selecting verb but must be taken to be spelled out in [Spec,F3P], since, as we have already seen in chapter 3, nonﬁnite verbs in West Flemish fail to move up to the head of the Aspect Phrase. Only IPP-inﬁnitives are checked and spelled out in [Spec,F2P], while participles in West Flemish move through F2P and are licensed and spelled out in [Spec,AspP], as is illustrated in (35).

VERBAL COMPLEXES AND THE SYNTAX OF IPP - COMPLEMENTS

183

(35) a. verb cluster that contains a dependent inﬁnitive in WF: te willen kuopen

AspP F2P te kuopen

F3P willen kuopen

b. verb cluster with an IPP-inﬁnitive in WF: te willen dienen boek kuopen een AspP1 F2P te AspP2 een F2P

0 dienen boek kuopen

F3P willen dienen boek kuopen

c. verb cluster that contains a participle in WF: gekocht te een AspP

gekocht

F2P

te gekocht

een

F3P

West Flemish shows that participles clearly have a diﬀerent distribution from inﬁnitives in the verb cluster. The partial parallelism in the distribution with IPPinﬁnitives indicates that participial verbs move into [Spec,F2P] for temporal linking and for checking the subcategorization of the selecting verb. Contrary to inﬁnitives, however, participial phrases move on into [Spec,Asp] of the selecting verb to make the temporal information contained in the participial morphology available (visible) for the rest of the derivation. This is an eﬀect of the phase condition discussed earlier. 6.4.4

Conclusions

To summarize, German, on the one hand, and Dutch and West Flemish, on the other hand, diﬀer in the amount of movement of nonﬁnite verbs in the clause. While in German ﬁnite and nonﬁnite verbs always move to the highest head in the VP-phase,

184

SCRAMBLING , REMNANT MOVEMENT , AND RESTRUCTURING IN WEST GERMANIC

that is, the head of the Aspect Phrase, only ﬁnite verbs in Dutch and West Flemish move to this position, with nonﬁnite verbs only moving up to the head of F2P. This has important consequences for the analysis of verb clusters. A dependent nonﬁnite verb in a left-branching verb cluster in German must be analyzed as occupying [Spec,AspP] preceding the selecting verb in Asp0. A dependent nonﬁnite verb in a right-branching verb cluster in Dutch or West Flemish must be analyzed as occupying [Spec,F3P] following the selecting verb in F20. Independently of these surface facts, we have seen evidence that indicates that the licensing position that dependent nonﬁnite verbs in the three languages share is [Spec,F2P], where I argued that dependent verbs are temporally linked to the matrix event. The three languages, however, diﬀer in the positions in which they check the subcategorization of the selecting verb as well as in the positions in which they spell out dependent nonﬁnite verbs.

7

Open Issues: Extraposition, VP-Topicalization, and the Status of Gerunds

In this chapter, I would like to address questions that are left open by the account that I have developed in this book. Furthermore, I would like to tie up some loose ends and bundle certain observations, foremost about the role of gerunds, that are scattered in diﬀerent chapters throughout the book. The ﬁrst issue pertains to extraposition, a notoriously diﬃcult topic within antisymmetric approaches. To be clear from the outset, I am not even trying to devise a theory of extraposition within a universal base approach. Such a theory would involve ﬁnding out the triggers and licensing positions of the diﬀerent types of extraposition and is well beyond the scope of this book. I will only address one case of extraposition, namely, that of CP-complements, that is problematic due to the account of sentential complementation developed in chapter 4. Because of their licensing requirements, CP-complements become part of the verb cluster (they are licensed by movement into [Spec,F3P] of the selecting verb) and due to the PIC become inaccessible for further computation. Clearly, this is an unwanted result, since CPcomplements (1) can be topicalized and (2) must be extraposed from left-branching verb clusters. The account that I provide is a technical solution to this particular problem and does not claim to be a solution to extraposition in general. The second issue concerns the topicalization of verb projections in coherent inﬁnitives. As we have seen in chapter 5, topicalization of the verb cluster or the dependent inﬁnitive alone is one criterion for detecting a coherent construction. However, the test of topicalization also shows that large parts of the embedded clause can also be topicalized, even with verbs that restructure obligatorily. This is a hard problem, which to my knowledge has not been given a satisfactory explanation so 185

186

SCRAMBLING , REMNANT MOVEMENT , AND RESTRUCTURING IN WEST GERMANIC

far. I will outline an account that takes advantage of the availability of the two kinds of inﬁnitives in West Germanic. I argue that the two forms are in partial competition with each other, with the gerund functioning as a means of last resort. Though some questions of mostly technical execution remain, this account paves the way to a general solution to this problem. Finally, I will address the observations that I made about gerunds in diﬀerent places in the last two chapters and try to tie them together to a coherent account of the structure and the role of gerunds in West Germanic.

7.1 Verbal complexes and the distribution of CP-complements As pointed out at the end of section 6.1, left-branching verb clusters pose a problem for my account of the licensing of CP-complements. In chapter 4, I have argued that CP-complements are licensed in [Spec,F3P], just above VP, by checking the features of the complementizer against the subcategorization properties of the selecting verb. Since the German verb moves up to the Aspect-head and the entire Aspect Phrase in front of the selecting verb in the formation of a German verbal complex, CP-complements are incorrectly predicted to show up within verb clusters, as is illustrated again in (1a). Furthermore, it is not clear how the correct distribution of CP-complements to the right of the entire verb cluster is to be derived in an antisymmetric approach (1b) (see also Lattewitz [1997]). (1)

a. *weil Hans [[AspP zu sagen [CP dass er krank war]] versuchte] since Hans to say that he sick was tried ‘since Hans tried to say that he was sick’ b. weil Hans [[AspP zu sagen] versuchte] [dass er krank war] since Hans to say tried that he sick was

Note, however, that the assumption that CP-complements are licensed inside the Aspect Phrase of the selecting verb not only raises problems with respect to leftbranching verb clusters. Given our assumption that the Aspect Phrase constitutes a strong phase, the question arises how CP-complements, being deeply embedded within the verb cluster, can ever be moved out of the matrix Aspect Phrase in the ﬁnal cycle. Consider, for instance, that CP-complements can be topicalized in the matrix clause, that is to say, can be moved into [Spec,CP] of the matrix clause after restructuring has applied (2a). (2)

a. [dass er krank war] hat Hans [AspP [zu sagen t] versucht] that he sick was has Hans to tell tried ‘Hans has tried to say that he was sick’ b. [zu sagen dass er krank war] hat Hans mehrmals versucht to say that he sick was has Hans several times tried

OPEN ISSUES

187

Thus, the account of CP-licensing given in chapter 4 cannot be complete, not even for right-branching verb clusters. In right-branching verb clusters the distribution of verbs and CP-complements correctly follows from the fact that the dependent inﬁnitive with its CP-complement is spelled out in a position to the right of the selecting verb, but extractability remains a problem also for right-branching verb clusters. One way to solve this problem would be to assume that CP-complements move out of the local Aspect Phrase before the latter moves into the Aspect Phrase of the higher verb. This raises the question of what the motivation of CP-complements could be to undergo additional movement after having been formally licensed. In the standard SOV approach to the syntax of West Germanic, it is assumed that CP-complements must be extraposed, that is, right-adjoined to the local IP/TP. To my knowledge, no satisfactory motivation has been proposed for this kind of movement. Furthermore, there is the issue that in cases of VP-topicalization the CP-complement can be piedpiped by the movement of the selecting verb, seemingly implying that the CPcomplement may also stay within the Aspect Phrase of the selecting verb (assuming for the time being that VP-topicalization involves movement of the Aspect Phrase), as is illustrated in (2b). I have no genuine insight to oﬀer on the nature of extraposition and its motivation. I will outline an account of extraposition in the standard theory that explains the data in (2). Then I will provide a parallel account in the antisymmetric approach. This account, like the standard account, merely constitutes a technical solution to the problem posed by the data given in (2). Thus, the purpose of this section is not to provide a novel account of extraposition but simply to sketch an account that is at least as good as the standard approach to extraposition. The standard account that I will adopt and modify is the treatment of extraposition by Büring and Hartmann (1997) (henceforth B&H). Among other issues, this article is concerned with the fact that verbs can both pied-pipe and strand their CP-complement in cases of VP-topicalization. Let us have a look at a typical example. (3)

a. [zeigen dass die Erde eine Scheibe ist] möchte er show that the earth a disk is wants he b. [zeigen] möchte er [dass die Erde eine Scheibe ist] show wants he that the earth a disk is ‘He wants to be able to show that the earth is a disk’

To account for cases like (3), they propose a dynamic account of VP-topicalization that assumes that there is no particular or unique position for extraposition. Piedpiping of the CP-complement in (3a) requires that the CP is adjoined to the VP, while stranding of the CP-complement in (3b) requires that the CP is adjoined higher than the VP, say, IP. They argue that the variability of adjunction sites can be derived from the following trigger condition on extraposition. Government in (4) is deﬁned, following van Riemsdijk and Williams (1981, 291), as given in (5). (4)

Finite sentences may not be governed by V or I.

188 (5)

SCRAMBLING , REMNANT MOVEMENT , AND RESTRUCTURING IN WEST GERMANIC

X governs Y iﬀ (a) X-c-commands Y and (b) XP is the smallest maximal projection that contains Y.

According to them, this ﬁlter is reminiscent of Stowell’s Case-Resistance Principle (cf. Stowell [1981]) and rules out ﬁnite sentences in their base position. In eﬀect, the clause has to ﬂee from the government domain of V0 and I0. According to B&H, this can be achieved in three ways. Either the clause is extraposed, that is to say adjoined to IP, or it can be topicalized alone or together with the VP. In the last case, adjunction to VP is suﬃcient in order to fulﬁll (4), since a clause adjoined to VP in [Spec,CP] is outside of the government domain of I0 and V0. This is a rather elegant account that derives the diﬀerent adjunction sites from the general freedom of move alpha. Extraposition as an instance of move alpha may in principle target any position provided that the resulting structure fulﬁlls the condition in (4). To translate this account into a minimalist framework is not a trivial issue. In minimalism, movement is assumed to be triggered and non-optional, in the sense that movement cannot freely target either category A or B or C. Note also that in a derivational approach we face a look-ahead problem with respect to the target of extraposition: At the point of the derivation where it is to be decided whether extraposition is to target the VP of the selecting verb or some other XP, it is not known whether VP-topicalization does apply or not. Thus, a derivational approach to extraposition must assume that there is only one extraposition site and derive the variable distribution of CP-complements in VP-topicalization in other ways. Also, there is a technical diﬃculty with B&H’s approach: In cases where the CP-complement is stranded by VP-topicalization, as illustrated in (3b), the CP is governed by the ﬁnite verb when extraposed to the matrix IP and is governed by (the trace of) I0 when it is extraposed to the VP projected by the matrix verb. B&H must assume that if I0 moves to C0 the trace of I0 does not count as a governor while it is standardly assumed that I-to-C movement would extend the government domain of C0 (cf. Baker [1988]). However, if the trace of I0 does not count as a governor, then the matrix IP protects the CPcomplement adjoined to the VP projected by the matrix verb from government by the verb in C0. In minimalism we cannot make use of the restricted notion of locality as deﬁned by government. Replacing government with C-command, we cannot assume that it is the verb that triggers extraposition, since all material to the right of the ﬁnite verb in C0 is in the C-command domain of the verb in V2–clauses. Thus, I propose to replace (4) with the condition in (6). (6)

A Tense-head may be neither in the checking domain nor in the scope (deﬁned by C-command) of an Aspect- or Tense–head.

The deﬁnition in (6) is certainly a stipulation, but it is not more stipulative than (4) and does the same job as (4), as we will see later. I propose that the condition in (6) is met by moving the CP-complement, after it has been formally licensed in [Spec,F3P] of the selecting verb, into the Speciﬁer of a functional head above TP, which constitutes the unique invariant extraposition site in this account. The surface

OPEN ISSUES

189

property that extraposed material appears to the right of the rest of the clause is achieved via licensing movement of both AspP and TP around it, as I have proposed in chapter 4 and as is illustrated again in (7). (7)

a. [CP C [MoodP M [StatP S [TP [AspP V CP]]]]] b. [CP C[MoodP M [StatP S [CP [TP T [AspP V t]]]]]]

extraposition

c. [CP C[MoodP M [StatP [AspP V t] S [CP [TP T]]]]]

licensing of the VP

d. [CP C[MoodP [TP T] M [StatP [AspP V t] S [CP]]]]

licensing of the TP

At the end of the derivation neither Tense nor Aspect c-command Tense within the “extraposed” CP. Note that in restructuring contexts every verb in the verb cluster, not only the selecting verb, can pied-pipe or strand a CP-complement when undergoing VP-topicalization, as is illustrated in (8). (8)

a. Zeigen [dass die Erde eine Scheibe ist] wird er schon können wollen show that the earth a disk is will he well can want ‘He may well want to be able to show that the earth is a disk’ b. Zeigen können [dass die Erde eine Scheibe ist] wird er schon wollen c. Zeigen können wollen [dass die Erde eine Scheibe ist] wird er schon d. Zeigen können wollen wird er schon [dass die Erde eine Scheibe ist]

At the end of chapter 4, I proposed that in restructuring inﬁnitives the licensing projections of the VP and TP in the C-domain are present but inert, so that the embedded TP and (extended) VP will just move through these positions on their way to the licensing positions in the next cycle. The variable distribution of CP-complements in (8) then comes about in the following way: In each derivational cycle, movement of the Aspect Phrase into the higher Aspect Phrase can either pied-pipe or strand the CP-complement contained in the Speciﬁer below. Starting with (7d), which represents the structure at the end of the ﬁrst cycle, pied-piping the CP complement involves movement of the entire StatP into the Aspect Phrase of the next cycle. In the case in which the CP-complement is stranded, it will be induced by the Tense-head in the next cycle to move to the Speciﬁer position above TP of the higher verb. In restucturing inﬁnitives this movement (across domains) is possible since the CP of a defective complementizer does not constitute a strong phase, as I assumed in chapter 6, and movement of the CP to its “scope” position in the next cycle only involves movement out of the strong phase of the Aspect Phrase into the next higher weak phase. This movement is allowed by the PIC, since only movement from a strong phase a into another strong phase b (or beyond) must proceed via the left-edge of a. In nonrestructuring contexts, this movement will be prohibited, since the containing CP constitutes a phase, allowing extraction only via its left edge, that is, [Spec,CP]. Thus, in non-restructuring contexts, “extraposition” can only apply locally, deriving the Right Roof Constraint from the Phase Impenetrability Condition. If a CP-complement is pied-piped by movement of the local AspP into the next higher AspP, the

190

SCRAMBLING , REMNANT MOVEMENT , AND RESTRUCTURING IN WEST GERMANIC

derivation will only then converge, if the lower AspP is extracted from the higher AspP via VP-topicalization, since the CP-complement being contained too deeply in the matrix AspP cannot extract by its own, leading to a violation of the condition in (6). In this account, (8a) is derived in that the CP-complement is pied-piped by movement of the AspP in the ﬁrst cycle (into the second cycle). Example (8b) is derived in that the CP-complement is pied-piped by movement of the AspP of the second cycle into the next cycle, and so on. No pied-piping applies in the case of (8d), which involves CP-movement to the Speciﬁer above the higher TP in every cycle. Summarizing, in this approach variability comes into play via pied-piping or the lack of it. We can then assume that pied-piping is possible (and necessary) if the AspP and the CP-complement share a feature. In the case of VP-topicalization, this shared feature can be assumed to be a discourse-related feature like TOP(ic) or FOC(us), which is checked by movement into [Spec,CP] in the matrix clause. Since exactly one phrase can be moved into [Spec,CP], lack of pied-piping would leave the discourse-related feature of either the AspP or the CP-complement unchecked, causing the derivation to crash. I have provided an account of extraposition of CP-complements in a UBH approach. It is a partial solution to a more general problem. The trigger that I proposed for this case of extraposition is not very insightful and needs to be improved. The account just serves as a “patch” till a better and more general solution to extraposition is found.

7.2 Topicalization of verb projections The preposing of verb(phrase)s supports the account that I have argued for in this book, namely, that restructuring in general and verb cluster formation in particular involve XP-movement rather than head movement. For example, note that if the dependent inﬁnitive in (9a) were incorporated (via head movement) into the selecting perception verb, then it is not clear how the dependent inﬁnitive could have been fronted to the exclusion of the perception verb. Since excorporation is excluded, the dependent inﬁnitive can only have been fronted by ﬁrst moving out the perception verb. (9)

a. weil Hans die Maria kommen gesehen hat since Hans the Maria come seen has ‘since Hans saw Maria come’ b. kommen hat Hans die Maria gesehen come has Hans the Maria seen ‘it was coming what Hans saw Maria do’ ‘as for coming, it was Hans who saw Maria do it’

On the other hand, if the dependent inﬁnitive has been moved via XP-movement to the position preceding the perception verb in (9a), then no additional movement of the perception verb itself has to be assumed in order to derive (9b): The dependent

OPEN ISSUES

191

inﬁnitive in [Spec,AspP] of the perception verb just moves on into [Spec,CP] of the matrix clause. To be more precise, (9b) shows that movement of the Speciﬁer in the Speciﬁer of the left edge of the matrix VP into the next higher strong phase of CP is allowed. Earlier I assumed that AspP and CP constitute the strong phases within a cycle. Having proposed that the CP of the dependent inﬁnitive does not constitute a strong phase in restructuring contexts, I may assume that the phase of the Aspect Phrase “closes” as soon is it is moved into the next higher phase up, that is, into the Aspect Phrase of the selecting verb. In other words, all material that has not been moved into the left edge, that is, [Spec,AspP], in the previous cycle is inaccessible for further computation after verb cluster formation in each cycle. This is the reason that I proposed that CP-complements must be “extraposed” in each cycle before the local Aspect Phrase moves into the higher Aspect Phrase. In this way, I derive that every phrase in the Speciﬁer of an Aspect Phrase can be topicalized at the point of the derivation that constitues the end of verb cluster formation, which is illustrated in (10a). In (10a), the property of being in the left edge of a phase is transitively extended to the speciﬁers of all Aspect Phrases contained in the Aspect Phrase of the matrix. Thus, (10b) is derived by moving the left edge (= the Speciﬁer) of the highest Aspect Phrase into [Spec,CP], while (10c) is derived by moving the left edge within the left edge of the highest Aspect Phrase into [Spec,CP], and so on. (10) a. weil er einen Roman [AspP [[[lesen] können] wollen] wird] since he a novel read can want will ‘since he will want to be able to read a novel’ b. lesen können wollen wird er einen Roman c. lesen können wird er einen Roman wollen d. lesen wird er einen Roman können wollen

Thus, the variable topicalizablility of verbs in the German verb cluster follows from the fact that they are licensed as XPs in the left edge of the VP-phase (the Aspect Phrase) in each cycle. This raises the question of how dependent inﬁnitives can be topicalized in Dutch and West Flemish, which according to our analysis are not spelled out in the left edge of the Aspect Phrase. Before I address this question, I will discuss two long-standing problems posed by VP-topicalization. 7.2.1

Problems posed by topicalization data

The ﬁrst problem is that topicalization of the dependent inﬁnitive either alone or together with (some of) its arguments leads to a bleeding of the IPP-eﬀect. This generalization holds to my knowledge without exception in Dutch and West Flemish. In German, perception verbs pattern with the behavior of Dutch and West Flemish V(P)R-verbs in this context, while modals and the permissive/causative verb lassen display the IPP-eﬀect even when the dependent inﬁnitive is topicalized. This is illustrated for modals by the contrast between the West Flemish example in (11a)

192

SCRAMBLING , REMNANT MOVEMENT , AND RESTRUCTURING IN WEST GERMANIC

and its Standard German pendant in (11b).1 Even those speakers of Standard German who have retained the participial form of modals in constructions like (11c) prefer the inﬁnitive over the participle in (11b). This observation is important, since one might believe that due to the gradual loss of the participial forms of modals (cf. ??gemusst < müssen [‘must’] and *gesollt < sollen [‘shall’]) the inﬁnitival forms in (11b) are simply suppletive for the participial forms, which are becoming rarer and rarer in Standard German. The fact that speakers can still use the participle in sentences like (11c) but prefer the inﬁnitive in structures like (11b) strongly suggests that we are dealing with a real case of IPP-eﬀect, meaning that we are dealing with a situation in which the participle is blocked in the syntax, rather than with one in which it is replaced by the inﬁnitive because it is lost in the lexicon. (11) a. in nen bank werken ee se niet gewild/*willen in a bank work has she not wanted/want-IPP b. in einer Bank arbeiten hat sie nicht wollen/*gewollt in a bank work has she not want-IPP/wanted c. das hat sie nicht gewollt/gekonnt that has she not wanted/can-PART

The bleeding of the IPP-eﬀect in the context of VP-topicalization is problematic for my account of it, since I explained the IPP-eﬀect as a syntactic, mechanic eﬀect of the licensing of nonﬁnite verbs in restructuring contexts. The data in (11a) can be taken to indicate that the IPP-eﬀect is a shallow morphological “Agreement” eﬀect that arises between two adjacent nonﬁnite verbs. While the data in (11b) show that the IPP-eﬀect in German proves to be more “stable” under non-adjacency, it raises an additional problem, namely, the issue of how to account for the crosslinguistic variation in the bleeding of the IPP-eﬀect between German on the one hand and Dutch and West Flemish on the other hand, and also for the intralinguistic variation between modal verbs and perception verbs in German. The second problem concerns the behavior of the nominal complements of the dependent inﬁnitive. The dependent inﬁnitive can be topicalized together with one of its arguments to the exclusion of the VR-verb (12a). In (12a), we face the problem that the preposed phrase is not a constituent anymore after VR has applied. Remember that the arguments of the dependent inﬁnitive undergo TP-movement into [Spec,PredP] of the selecting verb, while the inﬁnitive itself is moved into the Speciﬁer of the AspP of the selecting verb below PredP. In order to derive (12a) from this resultant structure (the output of restructuring), we have to assume that the modal wollen (‘want’) undergoes additional movement for which there does not seem to be any motivation but to evacuate a larger projection for topicalization. We can describe this state of aﬀairs by assuming that the modal undergoes evacuating movement. (12) a. ein Haus kaufen hat er ihr wollen a house buy has he her want-IPP ‘buy a house that is what he wanted to do for her’

OPEN ISSUES

193

b. ??ein Buch zu geben hat er seiner Frau versucht a book to give has he his wife-DAT tried c. zu geben hat er seiner Frau ein Buch versucht to give has he his wife a book tried

At this point, it is important to note that there is a diﬀerence between bare inﬁnitives and to-inﬁnitives. Split topicalization with to-inﬁnitives leads to decreased acceptability. In (12b), a to-inﬁnitive has been topicalized together with its direct object while the indirect object to his wife is left behind in the IP of the main clause. This is what I mean by split topicalization. With to-inﬁnitives, either the inﬁnitive with all its arguments can be topicalized (in which case we probably deal with a non-coherent construction) or the inﬁnitive alone can be topicalized, as is illustrated in (12c). The alternative to accounting for cases like (12a) in terms of evacuating movement of the selecting modal verb is to assume that (12a) results from a VPRstructure. In this analysis, the minimal XP that contains the indeﬁnite object would be pied-piped by movement of the AspP of the inﬁnitival into [Spec,AspP] of the selecting modal. From there, the constituent [ein Haus kaufen] can be easily extracted and moved into [Spec,CP]. The diﬀerence between bare inﬁnitives and to-inﬁnitives with respect to split topicalization one would then reduce to the fact that verbs that select to-inﬁnitives do not allow for VPR in German (in West Flemish some verbs do if the inﬁnitival marker is deleted). In the previous section, we have seen evidence from right-branching verbal complexes that indicates that VPR is indeed possible in Standard German. The problem with this alternative is that examples like (12a) can also be found in Dutch, which clearly does not allow for VPR. Example (13) is taken from Zwart (1993). (13) a. [Snel Marie gekust] heeft Jan niet quickly Mary kissed has Jan not ‘Kiss Mary quickly is not what Jan did’ b. [Snel Marie kussen] wil Jan niet quickly Mary kiss wants Jan not ‘Kiss Mary quickly is not what Jan wants’

If we want to provide a unitary account of verb(phrase)-topicalization in German and Dutch, then the account sketched earlier does not seem to be viable in the light of data like (13).2 Zwart (1993) himself argues on the basis of data like (13) and following an idea by Haider (1990) that so-called preposed verb phrases are basegenerated in a position outside of [Spec,CP]. I will discuss this alternative in the following section. 7.2.2

The base generation approach to VP-topicalization

In this section, I would like to investigate whether the base generation approach fares any better with respect to the problems that the account in terms of XP-preposing is confronted with, as I outlined in the previous section.

194

SCRAMBLING , REMNANT MOVEMENT , AND RESTRUCTURING IN WEST GERMANIC

Haider (1990) and Zwart (1993) propose that topicalized verb-projections are base-generated in a position outside the matrix CP. In this analysis, the preposed verb (phrase) is not reconstructed. What is reconstructed is the possibly empty d-word that occupies [Spec,CP] and is somehow identiﬁed with the (so-called) preposed phrase. An analysis of the sentence in (12a) under this approach is given in (14). (14)

[ein Haus kaufen] [CP dasi\0i hat er ti wollen]

Before we take a closer look at the theoretical implications of this account, let us ﬁrst see how it fares with respect to the problems I discussed earlier. The ﬁrst problem concerned the bleeding of the IPP-eﬀect. The d-word in Zwart’s analysis has nominal features though its (semantic) content stands for a verb. Nominal complements of participles never induce an IPP-eﬀect. This explains the absence of the IPPeﬀect in Dutch and West Flemish in cases where the dependent inﬁnitive has been topicalized. But it fails to explain the presence of the IPP-eﬀect with modals and the causative verb lassen in German. To put it diﬀerently, this account is forced to assume that the inﬁnitive in (14) earlier is a suppletive form for the participle rather than a “real” IPP-case. I have argued earlier that this is unlikely. The second problem consisted in the fact that the preposed phrase, when containing a complement of the dependent inﬁnitive, does not represent a constituent anymore after restructuring. This problem disappears (trivially) in a base-generation approach but reappears in a diﬀerent form, namely, in the question of how material that is left behind in cases of split topicalization is related to preposed material. To be more concrete, the question arises in this approach of how, for instance, a CPcomplement stranded by VP-topicalization (as is illustrated in [8d] in section 7.1) is related with its selecting verb in its base-generated position above CP. This issue will be dealt with in more detail later. Finally, addressing a problem discussed in the previous chapter, it is not clear at all how a base generation account can explain why right-branching verbal complexes can be topicalized in Dutch but are mostly (with the exception of IPP-inﬁnitives) ungrammatical in German. 7.2.3

Problems of the base generation approach

In this section, I will try to make the following argument: First I will point out several technical diﬃculties for the base generation approach. Then I will argue that these diﬃculties force us to assume that the d-word is an exact copy of the preposed phrase. Hence if we have to reconstruct an exact copy of the preposed phrase we might as well assume that the preposed phrase is reconstructed. Hence it should be assumed that the preposed phrase has been moved to its surface position from a clause-internal position rather than having been base generated there. The ﬁrst technical issue concerns Case checking. This approach raises the question of how the Case of a DP is checked when the DP is preposed and how it is checked when the DP remains behind. With regard to the ﬁrst part of the question, let us look at a case such as (15a). It is generally assumed that participles cannot assign or check Accusative Case. Nevertheless, (15a) is fully acceptable. In approaches to passive

OPEN ISSUES

195

that try to derive the diﬀerent properties of active and passive clauses from a unique participle (cf. Baker, Johnson and Roberts [1989], Roberts [1985]), it is assumed that the auxiliary haben (‘have’) assigns Case to the direct object of an active participle. Hence the d-word must contain a copy of the DP ein Buch (‘a book’), which is licensed (after reconstruction) by the auxiliary and which in turn licenses, probably via identiﬁcation, the copy of this DP in the base generated phrase. With regard to the second part of the question, let us look at a case like (15b). While it is reasonable to assume that the auxiliary have can check Accusative Case, it is unlikely that it can also check Dative Case. So the question arises as to how the Dative Case of the pronoun ihr (‘her’) is checked in (15b).3 The answer seems to be that the d-word contains a copy of the relevant feature of the verb schenken (‘give’) that serves to license the Dative argument that remains within IP in (15b). Along a similar line, it has to be assumed that the d-word also contains the relevant feature of the verb lesen (‘read’), in order to guarantee that the right auxiliary (haben rather than sein) is selected in structures like (15a–b). (15) a. [ein Buch gelesen] das hat er nicht a book read-PART that has he not b. ?[ein Buch geschenkt] das hat er ihr noch nie a book given that has he her yet never

The second technical issue concerns the question of how PRO in the basegenerated constituent is interpreted. In a sentence like (16), PRO would be assigned an arbitrary interpretation according to standard assumptions, since it does not have a c-commanding controller. However, its interpretation is that of a PRO controlled by the matrix subject. To accomodate this fact with a base generation approach, we have to assume that the d-word is a copy of the base generated phrase, whose member PRO is identiﬁed by the matrix subject after reconstruction has applied to the d-word. (16)

[ PROarb? ein Buch lesen] dasi will er ti

The third technical issue concerns temporal features. Example (17) shows that it is relevant for the licensing of adverbials like noch nie (‘so far never’) that the preposed constituent contains a verb and also what type of verb it contains. The contrast between (17a) and (17b) shows again that we cannot assume that in reconstructing the d-word only its nominal features are reconstructed. If that were the case, then (17a) should be as good or as bad as (17b), where we replaced the d-word with another nominal constituent, the DP einen Apfel (‘an apple’). The contrast between (17a) and (17c) shows that it is relevant that (1) the preposed phrase contains a verb and (2) this verb has the right temporal feature, namely, one that licenses a past interval presupposed by the adverbial. The participle in the preposed phrase in (17a) has this feature, which the inﬁnitive in (17c) lacks. This information has to be available in the IP within which the adverb is interpreted. Hence the d-word, being the only constituent that is reconstructed under the base-generation approach, must contain this information.

196

SCRAMBLING , REMNANT MOVEMENT , AND RESTRUCTURING IN WEST GERMANIC

(17) a. [ein Buch gelesen] dasi hat er noch nie ti a book read-PART that has he yet never b. *er hat noch nie [einen Apfel] he has so far never an apple c. *[ein Buch lesen] das wird/hat er noch nie a book read-INF that will/has he yet never

The fourth technical issue involves thematic licensing. It is not clear how in this approach thematic restrictions are handled. If thematic roles are relations in the VP, then this account is untenable, since, for instance, in (18a–b) the pronoun ihr (‘her’) is not related with any position in a VP. (18) a. ?ein Buch geschenkt das hat er ihr noch nie a book given that has he her yet never b. ein Buch geschenkt hat er ihr noch nie a book given has he her yet never

The same argument holds for the “stranded” CP-complement in (19a): There is no position in the VP that the CP-complement could be related with. That thematic restrictions need to be checked also in topicalization structures is shown in (19b). Example (19b) can only be ruled out as a violation of thematic restrictions within Zwart’s account. It cannot be ruled out as a Case violation, for instance. Remember (cf. note 3 earlier) that Case can be freely checked in Agr-positions of auxiliaries and VR-verbs under Zwart’s assumptions. (19) a. [fragen müssen] wird er wohl [ob wir einverstanden sind] ask must will he well whether we agree b. *[ein Buch gelesen] das hat er ihr noch nie a book read-PART that has he her yet never

If, however, thematic roles are features, then we have another instance for the argument that a whole lot of features of the preposed phrase have to be reconstructed. Thus, instead of identifying the preposed phrase with the d-word and reconstructing only the d-word, it is simpler to assume that the preposed phrase (with or separately of the d-word) reconstructs. 7.2.4

Diﬀerences between topicalization and d-word left dislocation

In Haider’s and Zwart’s accounts the preposing of verb projections is identiﬁed with d-word left dislocation where the d-word is phonetically empty. Example (20a) is a standard case of d-word left dislocation and (20b) is a standard case of topicalization. According to the base generation approach, (20b) is analyzed as given in (20c) and should thus not diﬀer in its core properties from (20a). I will argue, however, that

OPEN ISSUES

197

there are major diﬀerences between the two constructions, which do not seem to be reducible to the fact of whether the d-word is overt or covert. (20) a. [einen Roman gelesen] das hat er noch nie a novel read that has he yet never b. einen Roman gelesen hat er noch nie a novel read has he yet never c. [einen Roman gelesen] [d-word 0] hat er noch nie

Fronting of only the (nonﬁnite) verb leads to ungrammaticality in cases of d-word left dislocation, while topicalization of only the verb is ﬁne (cf. [21]). Split topicalization leads to reduced acceptability if the fronting (of the partial constituent) involves a d-word. No such eﬀect is found in cases of fronting without d-word ([22], cf. also [18] earlier). (21) a. *gelesen das hat er einen Roman noch nie b. gelesen hat er einen Roman noch nie read-PART (that) has he a novel yet never (22) a. ??einen Roman geschenkt das hat er seiner Frau noch nie b. einen Roman geschenkt hat er seiner Frau noch nie a novel given (that) has he to-his wife yet never c. *geglaubt das hat er nicht an Gott d. geglaubt hat er nicht an Gott believed (that) has he not in God

Furthermore, there are reconstruction eﬀects that appear in cases of topicalization but are absent in cases of d-word left dislocation. This is illustrated in (23). In (23a), negation can only have narrow scope with respect to the modal (as is expected), meaning more or less ‘what he wanted to do was not reading a book’. Example (23b) can have the same reading. But it has an additional reading in which the negation takes scope over the modal, meaning ‘what he did not want to do was to read a book’. Very likely related to this diﬀerence in reconstructability of the preposed phrase is the phenomenon that in Bavarian dialects, which allow for negative concord (NC), NC is possible in cases of topicalization but missing in cases of d-word left dislocation. So (24a) can mean that ‘he did not want to read a book’ while (24b) construed with a d-word can only mean that ‘he did not want to read no book’. (23) a. kein Buch lesen das wollte er (modal > neg) b. kein Buch lesen wollte er (neg > modal) no book read (that) wanted he

198

SCRAMBLING , REMNANT MOVEMENT , AND RESTRUCTURING IN WEST GERMANIC

(24) a. ka Buach lesn woit’a net (NC) b. ka Buach lesn das woit’a net no book read (that) wanted-he not

To summarize, in the previous section I have shown that the base-generation approach that has been proposed by Haider (1990) and adopted by Zwart (1993) is untenable. In this section, I have shown that the properties of VP-topicalization also diﬀer signiﬁcantly from the properties of its presumed source structure, namely, d-word left dislocation. Thus, I conclude that whatever the correct account of d-word left dislocation may be, VP-topicalization cannot be explained by a base generation approach.

7.3 Toward a movement account of VP-topicalization So we are back to square one. In the following I will sketch a movement account of VP-topicalization that is partially speculative but derives support from the coherence of the overall picture it yields as well as from the insight into the architecture of grammar it provides. I will start with developing a solution to the exceptional behavior of perception verbs with respect to the bleeding of the IPP-eﬀect in German. Before we look at concrete data, let me point out that I am only concerned with the syntactic conditions of extracting VPs out of verbal complexes and not with the discourse pragmatic conditions that license or motivate VP-topicalization.4 The question of what the relevant discourse conditions of VP-topicalization are will not be addressed here at all and would merit a separate (empirical) investigation that is beyond the scope of this book. 7.3.1

VP-topicalization in German

Remember that perception verbs in German only optionally display the IPP-eﬀect (25a–b) and that this IPP-eﬀect is “bled” (contrary to the behavior of modals in German) in VP-topicalization contexts (25c). (25) a. weil Hans die Maria kommen gesehen hat since Hans the Maria come seen has b. weil Hans die Maria hat kommen sehen since Hans the Maria has come see-IPP c. kommen hat Hans die Maria nicht gesehen/ *sehen come has Hans the Maria not seen/ see-IPP

In chapter 6, I argued that the optionality of the IPP-eﬀect in (25a–b) can be explained best if we assume that perception verbs select two types of inﬁnitives: one type that is licensed in [Spec,AspP] without inducing an IPP-eﬀect, as in (25a), and the other type, which is licensed in [Spec,F2P], where it induces the IPP-eﬀect, before it is moved to [Spec,AspP] as well. Furthermore, I have proposed to analyze the

OPEN ISSUES

199

ﬁrst type of inﬁnitive as a nominalized inﬁnitive (called gerund) and the second type of inﬁnitive as a verbal inﬁnitive (simply called inﬁnitive). As will become clear in what follows, this analysis is supported by VP-topicalization data. At the end of verb cluster formation, before VP-topicalization applies, the sentences in (25a–b) have the following structures: The inﬁnitive of (25a) is in the (transitive) left edge of the Aspect Phrase of the auxiliary and can be extracted for topicalization without further ado. The inﬁnitive of (25b), however, occupying the lower Speciﬁer, is not accessible for further computation. Extraction of the inﬁnitive in (25b) would only be possible if it had moved to [Spec,AspP] in the previous cycle. (26) a. weil Hans die Maria [AspP [[kommen] gesehen] [Asp’ hat]] b. weil Hans die Maria [AspP [Asp’ hat [kommen sehen]]] c. weil Hans das Buch nicht hat lesen wollen since Hans the book not has read want-IPP d. lesen hat Hans das Buch nicht wollen lesen hat Hans the book not want-IPP e. weil Hans das Buch nicht lesen hat wollen (Bavarian) since Hans the book not read has want-IPP

That the latter movement must be possible in principle is evidenced by modal verbs in VP-topicalization contexts in German. Given that (26d) is derived via extraction from the verb cluster in (26c), the dependent inﬁnitive, due to the presence of the IPP-eﬀect, must have been moved from the lower Speciﬁer into [Spec,AspP] of the auxiliary before extraction took place. This movement of the dependent inﬁnitive from the lower Speciﬁer into the left edge, which is evidenced in dialects of Bavarian (cf. [26e]), is clearly non-triggered movement. That is to say that it cannot be taken to be due to any licensing requirement, since all the verbs in the cluster have been licensed at the end of cluster formation. Thus, this movement must be considered as a last-resort operation that is only allowed if no other option is available to prevent the derivation from crashing.5 Assuming that the dependent inﬁnitive has a focus- or topic-feature that needs to be checked in [Spec,CP], the derivation would crash without the application of the last-resort operation since with modal verbs, contrary to perception verbs, there is no other option in the grammar. However, with perception verbs there is another option. I have said earlier that perception verbs may optionally select an inﬁnitive or a gerund. When taking the lexical choice of the gerund, topicalization of the dependent gerund can proceed without a last-resort operation and must thus be considered to be less costly than the parallel derivation with an inﬁnitive. A note is in order here concerning the rationale of competition. In the MP, it is assumed that derivations with diﬀerent lexical choices, that is, diﬀerent numerations, do not compete with each other. Here we are dealing with one and the same lexical item, a perception verb that, however, has two subcategorization frames, selecting an inﬁnitive or a gerund. However, the distinction between inﬁnitive and gerund in this case is a purely formal one and does not involve a distinction in either meaning or style. Since normally lexical choices give

200

SCRAMBLING , REMNANT MOVEMENT , AND RESTRUCTURING IN WEST GERMANIC

rise to a diﬀerence in meaning (or style), we can assume that even derivations with diﬀering lexical choices can compete as long as the lexical choice does not imply a diﬀerence in meaning (the role of stylistic diﬀerences in cases of competition needs further empirical investigation, but it seems to me that a stylistic eﬀect is enough to prevent competition between two derivations that involve lexical choices that in all other respects are completely identical). As for the second problem concerning topicalization, namely, topicalization of the dependent inﬁnitive with one or more of its arguments, I propose to assume the simplest solution that is possible in German. Instead of assuming that there is evacuating movement, I propose that these cases fall out simply from the fact that even Standard German allows for VPR. With these ramiﬁcations, topicalization of verb-projections in German proves to be unproblematic. But the burden of explanation is shifted onto topicalization of verb-projections in Dutch. Before I start to tackle this problem, I will address the issue of the surprising scope eﬀects with VP-topicalization in German. In discussing the diﬀerences between topicalization and d-word left dislocation in German I found that a topicalized negative phrase can give rise to a wide scope interpretation in the Standard and to negative concord (NC) in Bavarian. The data are illustrated again in (27). (27) a. [kein Buch lesen] wollte er no book read wanted he ‘he did not want to read a book’ b. [koa Buach lesn] woit a net (NC) no book read wanted he not ‘he did not want to read a book’ c. [ein Buch lesen] wollte er nicht a book read wanted he not ‘he did not want to read a book’

Assuming that topicalized verb projections are derived via movement, we may assume that the topicalized phrase can reconstruct into its extraction site, that is, [Spec,AspP] in the matrix clause. In this position the negative determiner kein is accessible for further computation. In particular, we may assume that it can enter into an Agreement relation with matrix negation, giving rise to a wide scope interpretation in (27a) and to negative concord in (27b). Alternatively, we may assume that the negative DP kein Buch moves into [Spec,NegP] in the matrix clause, to license sentence negation/negative concord—before topicalization aﬀects the containing verb projection—but is spelled out in [Spec,AspP]. Then topicalization applies and the resulting structure is interpreted like the parallel clause in (27c). 7.3.2

VP-topicalization in Dutch

Before we look at VP-topicalization, recall how we have come to analyze complex verb clusters in Dutch. For Dutch, I have assumed that the dependent inﬁnitive moves

OPEN ISSUES

201

into [Spec,F2P] of the matrix verb but is spelled out in [Spec,CP] of the embedded clause, which is itself licensed in [Spec,F3P] (the formal licensing position of CPcomplements) of the selecting verb, as is illustrated for the dependent inﬁnitive lezen in (28a). The selecting nonﬁnite modal willen in (28a) is thus analyzed parallel to the position of the nonﬁnite verb of a te-inﬁnitive, as in (28b) and parallel to nonﬁnite verbs in West Flemish, which I showed fail to move up to the head of the Aspect Phrase. (28) a. dat Jan het boek [AspP zal [F2P lezen willen [F3P lezen]]] that Jan this book will want read b. om het boek [AspP te [F2P willen [F3P lezen]]] in order this book to want read

Let us assume we want to topicalize the dependent inﬁnitive lezen out of the resulting verb cluster in (28a). Then due to the PIC, the dependent inﬁnitive should move to [Spec,AspP] of zal in the previous cycle. Though this movement must be considered as a last-resort operation, it should be freely available given the lack of an alternative derivation. However, there is a crucial diﬀerence between a Dutch verb cluster as analyzed in (28a) and a German verb cluster. On the one hand, the dependent inﬁnitive in (28a) would have to move across its own un-spelled-out copy, something that we may assume is excluded in principle. On the other hand, we expect that the containing F2P willen lezen can move to the local escape hatch and thus can be topicalized without problems. However, these predictions are not borne out. The data in (29) show that topicalization in Dutch may be as liberal as topicalization in German. (29) a. [willen lezen] zal hij het boek niet want read will he the book not b. [lezen] zal hij het boek wel willen read will he the book well want c. (*)?[hebben willen lezen] zal hij het boek niet have want read will he the book not d. (*)?[willen lezen] zal hij het boek niet hebben want read will he the book not have e. [het/een boek willen lezen] zal hij niet the/a book want read will he not f. [het/een boek lezen] zal hij niet willen the/a book read will he not want

While most speakers judge (29c) and (29d) as more marked than the rest but consider them still acceptable, there does not seem to be a diﬀerence between topicalization of part of the nonﬁnite verbs and topicalization of the entire nonﬁnite verb cluster (cf. [a–b, c–d, and e–f]). Also note that so far we have no derivational base for the sentences in (29e-f), given that Dutch does not allow for VPR. Before I address

202

SCRAMBLING , REMNANT MOVEMENT , AND RESTRUCTURING IN WEST GERMANIC

this issue, let me point out that there is an interesting gap in the paradigm that may point to a possible solution of the problem. When we look at IPP-inﬁnitives, we ﬁnd the restriction borne out that I hypothesized earlier. Incidentally, there seem to be two groups of speakers, who diﬀer in an interesting way in their judgments. For speakers of both groups topicalization of the dependent inﬁnitive out of a verb cluster that contains an IPP-inﬁnitive is impossible (cf. the a examples in [30] and [31]). However, speakers of group B have an alternative strategy that employs a participle that is for some reason unavailable to the speakers of group A (cf. the b examples in [30] and [31]), a strategy, however, that becomes available also for speakers of group A if topicalization does involve not only the dependent inﬁnitive but the inﬁnitive plus its direct object (cf. [30c]). No problem arises for speakers of both groups if the complete nonﬁnite verb complex is topicalized, as is illustrated in (30d). Finally, note that speakers who can use the alternative strategy in (31a) can freely rephrase the topicalization as d-word left dislocation (31c). (30) Group A a. **lezen heeft hij het boek niet willen read has he the book not want-IPP b. *lezen heeft hij het boek niet gewild read has he the book not wanted c. een boek lezen heeft hij niet gewild a book read has he not wanted d. willen lezen heeft hij het boek niet (also B) want-IPP read heeft he the book not (31) Group B a. *lezen heeft hij het boek niet willen read has he the book not want-IPP b. lezen heeft hij het boek niet gewild read has he the book not wanted c. lezen dat heeft hij het boek niet gewild read that has he the book not wanted

How can we account for these data? Let me ﬁrst address the speakers of group A. Seemingly, speakers of this group cannot extract a dependent inﬁnitive out of an IPP-inﬁnitive cluster but manage to do that out of a true inﬁnitive cluster (cf. ([30a] versus [29b]). From the fact that they have no alternative strategy available in (30a–b) we can relatively safely conclude that the topicalization in (29b) really does involve subextraction of the dependent inﬁnitive out of the previously formed verb cluster rather than some alternative derivation. We may assume that speakers of group A only use/can use the alternative strategy if topicalization aﬀects a verb plus one of its arguments. The behavior of speakers of group A follows if they analyze structures like (28a) as is indicated in (32a), with nonﬁnite verbs moving into the Aspect-head

203

OPEN ISSUES

like in German. However, IPP-inﬁnitives, as I have proposed in chapter 6, exceptionally stay in F2 with the local Aspect-head being occupied by a zero-morpheme. Thus, a verb cluster with an IPP-inﬁnitive must be analyzed also for the speakers of group A as given in (32b). In this structure, the dependent inﬁnitive lezen cannot be extracted from the cluster by moving across its own copy. Since these speakers do not have an alternative strategy available in this case, there is a gap in the paradigm. (32) a. dat Jan het boek [AspP zal [F2P [AspP willen [F2P lezen]]]] A b. dat Jan het boek [AspP heeft [F2P [AspP 0 [F2P lezen willen [F3P lezen]]]]] A + B c. dat Jan het boek [AspP zal [F2P [AspP [F2P lezen willen [F3P lezen]]]]]

B

Finally, let me address the question of what the alternative strategy is that saves the topicalization of verb plus argument for speakers of both groups (cf. [30c]). This strategy is also available for speakers of group B in case of topicalization of a dependent inﬁnitive. Furthermore, I note that this strategy induces the presence of a participle instead of the IPP-inﬁnitive and seemingly involves a nominalized inﬁnitive as is suggested by the optional presence of the d-word. I would like to propose that the derivation of (30/31c) and of the parallel structures in (29e–f) involves a gerund. I analyze the gerund as a phrasal aﬃx that morphologically selects for an inﬁnitive. This means that the gerund can be attached to any verbal projection provided that the latter can satisfy its morphological requirement.6 Assuming that this requirement is satisﬁed through adjacency, we can analyze the gerund as a functional head that attracts the phrase that it attaches to into its Speciﬁer, as is illustrated in (33). Let us assume we are at the point of the derivation where we have Case-licensed the direct object in ein Buch lesen/ een boek lezen. In a German coherent inﬁnitive we can simply VP-raise this constituent, which will move after a number of steps into [Spec,AspP] of the matrix verb, from which position it can be subextracted for topicalization. In Dutch, VPR is not available. Let us assume that as a last resort a functional head that contains the gerund aﬃx can be inserted that attracts the complement into its Speciﬁer and is XP-moved in the very same fashion as a VPR-constituent into a Speciﬁer of the selecting verb. However, contrary to a plain inﬁnitive, it will not move into [Spec,F2P] but directly into [Spec,AspP] of the selecting verb. Thus, it will fail to induce an IPP-eﬀect and be eligible for subextraction to be topicalized in the ﬁnal step of the derivation. (33) a. [0 G [AgrAcc een boek [AspP lezen [VP]]]] Insertion of a functional head b. [GP [AgrAcc een boek [AspP lezen [VP]]] 0] Adjacency c. the aﬃx is fused with the adjacent inﬁnitive to fulﬁll its morphological subcategorization

The question now arises as to why the gerund in Dutch can only be used in VPtopicalization contexts. We can assume that modals and other verbs that take bare inﬁnitival complements do not select a gerund. They select an inﬁnitive. Thus, the

204

SCRAMBLING , REMNANT MOVEMENT , AND RESTRUCTURING IN WEST GERMANIC

gerund cannot compete with the inﬁnitive and is only available as a means of last resort. Since the derivation with the gerund involves extra steps (the insertion of a functional head, attraction of the complement of the gerund into its Speciﬁer), it will only be licit if the more economical derivation with the inﬁnitive crashes. This in turn raises the question of how the subcategorization of the selecting verb can be satisﬁed by a gerund. At this point the answer to this question, though plausible, must remain somewhat speculative. There is a selectional chain between the selecting verb and the complement of the gerund. The selecting verb selects an inﬁnitive and a lastresort operation inserts an element, the gerund, that itself selects an inﬁnitive. To summarize, we have seen that speakers of group A can use the gerund only if an extended verb phrase is topicalized. This is consistent with a treatment of the gerund as a means of last resort: An inﬁnitive cannot give rise to VPR in Dutch and hence the (extended) inﬁnitive is replaced, actually is augmented, with a gerund. Topicalization of the inﬁnitive alone does not and cannot employ the gerund, as is indicated by the ungrammaticality of (30b). This seems to imply that (1) the gerund is only employed in the topicalization of extended VPs and (2) the topicalized bare VP in (29b) is a real inﬁnitive and does not involve a gerund. How can we characterize the judgments of speakers of group B in this account? These speakers allow the replacement of an IPP-inﬁnitive by a participle in case of the topicalization of the bare dependent VP. Within my account this situation can be analyzed in two ways. Either these speakers have extended the insertion of the gerund as a last-resort operation from non-derivable extended VPs to non-extractable bare VPs, which then shows up in the context of an IPP-inﬁnitive, or they have used a last-resort operation also in this case, since bare VPs cannot be extracted at all in their dialect, so that the gerund is used in all cases of topicalization. This would then be the case if they quite generally analyze the Dutch verb cluster parallel to the West Flemish verb cluster, where only ﬁnite verbs move up into the local AspP. This analysis is indicated in (32c). In this structure, extraction of the dependent inﬁnitive lezen is blocked by its own copy and topicalization of any verbal projection must thus always make use of the gerund. Now the question arises of whether there is any evidence for this analysis beyond its initial appeal. At this point, the proposal that is implied by my rigid account of verb clusters is a bit speculative. Let me stress, though, that some speculation is allowed or even in order in this case, since so far there is no convincing account of VP-topicalization anyway. Whether this approach is correct I have to leave for further research. What I can do here is simply point out that if this approach is adopted, then some loose ends can be tied up and some observations that I made in the last chapters fall into place nicely, thereby giving rise to a rather coherent picture of restructuring inﬁnitives in German, Dutch, and West Flemish. The ﬁrst observation is that topicalization of extended VPs seems to involve two constructions in German but only one in Dutch. We have seen earlier that topicalization and d-word left dislocation for some German speakers have diﬀerent properties, while according to Zwart (1993) they cannot be distinguished in Dutch. As far as German is concerned, there are speakers who do accept cases of split d-word left dislocation more readily.7 And it remains to be seen whether all Dutch speakers agree with Zwart’s judgments. Apart from these empirical issues, which would deserve

OPEN ISSUES

205

further investigation, I can make the following generalizations. In German, due to the availability of VPR, topicalization may involve an extended verbal projection or a gerund. Thus, topicalization without a d-word in German can be analyzed as the topicalization of a VPR-constituent. Assuming that the d-word indicates the presence of a nominalized constituent, topicalization with the d-word involves the gerund, yielding two constructions with the diﬀerences noted earlier. In Dutch, however, due to the unavailability of VPR, the topicalization of an extended VP must involve a gerund independently of the presence of a d-word. That is why topicalization and d-word left dislocation can be equated in Dutch or, as proposed by Jan-Wouter Zwart, can be analyzed as involving covert and overt variants of the d-word. The second observation concerns the fact that d-word left dislocation in German and Dutch, at least for one group of German speakers, behaves slightly diﬀerently. As we have seen in the previous section, split d-word left dislocation is degraded for some German speakers but is judged completely grammatical in Dutch. This is a fact that, given the natural assumption that d-word left dislocation in German and Dutch involve the same construction, is hard to explain. In my account, this diﬀerence can be relegated to the diﬀerent status of the use of the gerund in German and Dutch. In Dutch, use of the gerund is the only way of topicalizing an extended verb projection. In German, there is a more economical alternative as long as topicalization does not involve the entire nonﬁnite complement of a selecting verb, namely, topicalization of a VPR-constituent. In other words, the gerund as a means of last resort may only apply to the entire inﬁnitival TP, where there is no alternative derivation in terms of VPR. The degraded character of split d-word left dislocation in German can then be assumed to follow from being blocked by a more economic alternative derivation that involves VPR, while in Dutch, since there is no alternative derivation in terms of VPR, split d-word topicalization is judged grammatical. For those German speakers who accept cases of split d-word left dislocation, we have to assume that their use of the gerund is not blocked by their use of VPR. This could be the case if the competition between the two forms is somehow suspended, either because the two forms have slightly diﬀerent functions (d-word left dislocation can only be used if the topicalized constituent is given in the discourse; no such restriction applies to [bare] topicalization) or possibly because these forms are treated as stylistically marked alternatives. More research on grammatical variants that are semantically indistinct is necessary to evaluate these options. Summing up, the preceding explanation of the diﬀerence between d-word left dislocation in German and Dutch can be called a functional explanation, since it considers the status or the role of the gerund in the entire grammatical system. In Dutch, use of the gerund is without alternative and thus unconditionally grammatical. In German, the use of the gerund is restricted by the availability of a more economical derivation in terms of VPR for one group of speakers and by functional or stylistic factors for the other group of speakers. The third observation concerns the fact that topicalized extended verb-projections in German and West Flemish also diﬀer considerably in the two languages. These facts will be discussed in the following section.

206

7.3.3

SCRAMBLING , REMNANT MOVEMENT , AND RESTRUCTURING IN WEST GERMANIC

VP-topicalization in West Flemish

West Flemish, contrary to Dutch, does allow for VPR and thus supplies a good testing case for my account. First, consider the observations about VPR in West Flemish that I made in chapter 3. VPR-constituents always have narrow scope (they cannot undergo QR at LF to take inverse scope) and, if negated, cannot give rise to negative concord. These properties fall out nicely from our analysis of the structure of West Flemish verb clusters: Being deeply contained in the verb cluster, operators in VPRconstituents are not eligible for further computation. It is interesting to observe that these operators in VPR-constituents behave diﬀerently under VP-topicalization in German and West Flemish. In other words, it is interesting to note that the lack or the presence of the IPP-eﬀect in these constructions is paired with an interpretive diﬀerence, as is illustrated in (34). Example (34a) in German is ambiguous, allowing for a wide scope interpretation of negation over the modal verb want, while in the parallel West Flemish structure in (34b) negation can only take narrow scope with respect to the modal verb. (34) a. kein Buch lesen hat er wollen (neg > modal possible) no book read has he want-IPP b. geen boek lezen eet Jan gewild (only modal > neg) no book read has Jan wanted c. kein Buch lesen das hat er gewollt (only modal > neg) no book read that has he wanted

This fact is surprising if we assume that the topicalized VP in West Flemish is derived in a similar fashion as it is in German, namely, via VPR and subextraction out of the verb cluster. If subextraction of a VPR-constituent (via [Spec,AspP] of the matrix verb) were possible in West Flemish, then we could not explain why the negative operator cannot give rise to the same reading via agreement with a local and ccommanding negative head after reconstruction into [Spec,AspP], as I argued is the case in German. The answer is that VPR-constituents in West Flemish cannot extract from the verb cluster. The last-resort operation of moving an inﬁnitival VP into the local [Spec,AspP], necessary within a right-branching verb cluster, is blocked. Remember that I have argued that dependent inﬁnitives in West Flemish are spelled out in [Spec,F3] of the licensing verb. Further movement of the dependent inﬁnitive is thus blocked by its own copy in [Spec,F2] of the selecting verb as is illustrated in (35). (35)

dat Jan [AspP zal [F2P [AspP [F2P [een boek lezen] willen [F3P een boek lezen]]]]]

What is important in this context is not only that (34a) and (34b) diﬀer in their interpretation but also the fact that VP-topicalization in West Flemish has the same interpretational properties as d-word left-dislocation in German, as is indicated by the parallelism between (34b) and (34c). Assuming that the alternative derivation in West Flemish involves a gerund, the derivation of (34b) proceeds as follows: The gerund will directly move into [Spec,AspP] of the selecting verb and thus fail to give

OPEN ISSUES

207

rise to an IPP-eﬀect. From there it can undergo last-resort movement to [Spec,AspP] of the auxiliary, the escape hatch for subsequent movement into [Spec,CP] of the matrix clause. To explain the interpretative diﬀerence between (34a) and (34b–c) we have to assume that the gerund blocks agreement between the negative operator and negation under reconstruction. For German, this fact is explained straightforwardly. Since I argued that in German the gerund attaches to the entire TP, the gerund can be taken to contain a more local negative head for the negative operator, barring any interaction with a higher, less local head. For West Flemish we can also assume that the gerund can only attach to the entire inﬁnitival TP, though I will have to leave this issue open for further research that clears the status of split d-word left dislocation in this language. 7.3.4

Conclusions

To summarize, if we assume that topicalized extended VPs in West Flemish simply involve VPR and extraction out of the verb cluster, then we cannot explain the interpretive diﬀerences in VP-topicalization between German and West Flemish. Again, where German has two constructions with diﬀerent interpretations, West Flemish has only one, which has the properties of d-word left dislocation and arguably involves the gerund. In West Flemish, like in Dutch, the derivational source of the alternative construction is lacking, despite the availability of VPR. This shows that the presence or absence of VPR is not crucial for the availability of topicalized extended VPs. What is crucial is the licensing position of VPR-constituents. These are diﬀerent in German and West Flemish. The crucial diﬀerence between German on the one hand and Dutch and West Flemish on the other is the (branching) structure of the verb cluster. Whereas in German both inﬁnitives and gerunds are licensed in the left edge of the verb cluster and can thus be extracted for VP-topicalization, Dutch and West Flemish have to resort to the gerund alone, either because VPR is not available at all, as in Dutch, or because the VPR-constituent is not extractable from the verb cluster, as in West Flemish. In addition, we have seen that the presence or absence of the IPP-eﬀect in VP-topicalization correlates with syntactic and interpretative properties of the topicalised VP, providing additional support for my account of the IPP-eﬀect as a structural eﬀect in the licensing of dependent verbal projections. Finally, I have shown that the availability and the interpretational properties of VPtopicalization can be derived from the ﬁne structure of the verb cluster (together with the assumption that the Aspect Phrase constitutes a strong phase). Also, I have argued that the availability of the gerund is constrained by competition with the inﬁnitive, explaining why certain structures only show up in VPtopicalization contexts, that is, in contexts where the inﬁnitive does not provide a derivational source for topicalization. This account of VP-topicalization supports and strengthens my general approach of deriving coherent inﬁnitives from a sentential source structure. VP-topicalization in Dutch shows that coherent inﬁnitives in this language—even though Dutch only allows for VR—do license their arguments in a separate domain from the one of the matrix verb. An alternative account that derives topicalization patterns and VRpatterns from diﬀerent sources would also have to explain why the diﬀerent source

208

SCRAMBLING , REMNANT MOVEMENT , AND RESTRUCTURING IN WEST GERMANIC

patterns are restricted in the way they are in Dutch (diﬀerently from German and West Flemish). Besides comprising a double base (projection of arguments in the matrix clause and in the embedded clause), such an account must provide speciﬁc restrictions as to why the topicalization pattern (O Vinﬁn) does not show up as VPR in sentences without VP-preposing in Dutch.

7.4 The status of the gerund In the last three chapters, I have been talking about diﬀerent occurrences of gerunds. I assumed that some occurrences of gerunds are selected and that sometimes they can be used as a means of last resort. The common assumption that I made about it is that the gerund is a phrasal aﬃx that morphologically selects for an inﬁnitive, which requirement is satisﬁed via adjacency. All the other properties of the gerund should follow from the place in the clause it is inserted and from the restrictions, including last resort, that govern its insertion. Let me now address in turn the diﬀerent occurrences of the gerund. 7.4.1

The selected gerund (the Doelfoarm)

I propose that a gerund that is selected by the matrix verb is inserted in the head of the Status Phrase. In this position, it probably serves to nominalize the complement of the matrix verb, very much like a complementizer does. Since the Status Phrase attracts the local Aspect Phrase, the morphological requirement of gerunds is fulﬁlled through the standard licensing movement of Aspect Phrases in this position. The direct object can be assigned structural Case in the usual manner in the local TP. Contrary to a complementizer, the gerund cannot license the TP in the local Mood Phrase and the clause undergoes restructuring with the TP moving into PredP and the Status Phrase (since the gerund aﬃx cannot be separated from the inﬁnitive within AspP) moves into the Aspect Phrase in the matrix clause via [Spec,CP] in order to license the clause with a [+N] feature. This is illustrated in (36). (36)

7.4.2

[CP

C

[MP

M

[StatP

G

[TP

[AspP

]]]]]

The gerund in to-inﬁnitives

In chapter 5, I argued that in a coherent to-inﬁnitive the gerund is selected by the inﬁnitival marker zu in the Aspect-head of the inﬁnitival clause. Thus, I assumed that the gerund is inserted directly below the Aspect Phrase and attracts its complement, that is, F2P, into its Speciﬁer to fulﬁll its morphological requirement. This is illustrated in (37). The inﬁnitival clause, containing a defective complementizer, then restructures in the usual manner. (37)

[CP

[TP

[AspP zu [GP G

[F2P V-en [VP

]]]]]]

OPEN ISSUES

7.4.3

209

The gerund as a means of last resort

In the case of topicalization of an extended verb-projection in Dutch, the gerund is not selected by the matrix verb but is inserted in the course of the derivation. For instance, the gerund can be inserted on top of Agr3P in order to allow for topicalization of inﬁnitive and direct object, as is illustrated again in (38). Then restructuring applies to the constituents in the embedded clause and the gerund is moved in the known fashion into the left edge of the matrix Aspect Phrase, in which position it is free to be extracted for topicalization. (38)

[CP

[TP

[GP

G [Agr3P een boek

[AspP

[F2P

lezen [VP]]]]]]]

Let me now address the question of the special status of gerunds in the grammar. In particular, I would like to know what the restrictions are that govern the attachment of the gerund. Also, the question arises as to whether there are other elements in the grammar that can function in a similar way. There is an analysis of VPR that suggests that inﬁnitives can have the same properties. My standard description of VPR is one of a process in which movement of the Aspect Phrase of the inﬁnitival clause pied-pipes additional structure, for instance, an AgrOP that contains the direct object of the inﬁnitival verb. Note that this informal way of talking about the process is actually misleading. According to the standard notion, a head X or a Speciﬁer of X can pied-pipe other material by inducing the entire XP to undergo movement. But an Aspect Phrase cannot pied-pipe in the technical sense the entire AgrOP. Rather, in the other way round, the direct object in AgrOP could pied-pipe the Aspect Phrase contained in it. However, in my account of restructuring, movement of the Aspect Phrase (the inﬁnitival VP) serves to check a feature of the verb and not a feature of the direct object. There are two ways in which the verb could induce pied-piping of the entire AgrOP in the technical sense. (1) If the inﬁnitive undergoes head movement to AgrO but is spelled out in the Aspect-head, it can be taken to pied-pipe the entire AgrOP when it undergoes further licensing movement into the C-domain or into the Aspect Phrase of the higher verb. (2) If the inﬁnitive can also function as a phrasal aﬃx, it can be inserted on top of AgrOP and attract the latter into its Speciﬁer. Further movement of the inﬁnitive would then necessarily pied-pipe the entire ArgOP. The ﬁrst option can be excluded. Dutch does not allow for VPR. Given the ﬁrst option, we have to assume that Dutch diﬀers with respect to head movement of the verb in this way from both German and West Flemish. This is unlikely. Or, at least, I know of no argument or evidence in this direction. Thus, we are left with the second option, which could render the status of the gerund in the grammar less idiosyncratic. Dutch can be described as a language in which the inﬁnitival ending cannot function as a phrasal aﬃx, explaining why Dutch does not have VPR. Therefore, the gerund is used as a means of last resort in cases of topicalization of an extended verb-projection. More work on the nature of VPR is needed to evaluate this hypothesis.

SCRAMBLING , REMNANT MOVEMENT , AND RESTRUCTURING IN WEST GERMANIC

210

7.4.4

The gerund as a clausal aﬃx

Let me address now the ﬁrst issue, namely, the question of what the restrictions are that govern the attachment of the gerund. The preceding discussion has shown that the relevant restriction seems to be that the gerund can attach to any extended projection of the verb that can satisfy its morphological selection. This is consistent with its analysis as a (nominal) phrasal aﬃx. The question, though, arises of whether it is possible to have a uniﬁed analysis of the occurrences of the gerund illustrated in (36) to (38) earlier. One possibility is to assume (1) that the gerund aﬃx is always inserted in the head of the Status Phrase in the C-domain, as given in (36) earlier, and (2) that the gerund can attract any extended projection of an inﬁnitive into its Speciﬁer in order to fulﬁll its morphological requirement. This uniﬁes the three occurrences of the gerund discussed earlier and has the following advantage: All three occurrences of gerunds can be taken to be nominalized clauses. If we assume that the gerund is in fact a nominal phrasal aﬃx, this analysis can explain why in all three kinds of nominal inﬁnitives adverbs and, with one restriction, Case are licensed. Since the nominal aﬃx is introduced high in the structure, verbal properties (Case) and verbal modiﬁers are licensed within the inﬁnitival TP. This consequence is of special importance for the analysis of coherent to-inﬁnitives. If the gerund is indeed a nominal aﬃx, then it is hard to explain in the analysis of (37) how a coherent inﬁnitive can license event-related adverbs, since the gerund aﬃx is taken to attach very low in the structure. In the analysis of the gerund in (36), this property follows without ado. But (36) raises the question of how the selectional property of the inﬁnitival marker can be satisﬁed in this account. In chapter 5, I proposed that an inﬁnitival marker that is not temporally linked selects for a gerund and assumed that selection means syntactic selection. In the revised account this requirement must be treated as morphological selection. The gerund aﬃx is inserted in a higher position in the structure but merges with the inﬁnitival ending after a suitable extended projection has been moved into its Speciﬁer. The relevant concept that we need, to provide a comprehensive treatment of the gerund in coherent to-inﬁnitives, is one that comes from the framework of Distributed Morphology (DM). In this approach two morphemes can be merged in Morphological Form (MF), that is, after Spell-out, under the condition of adjacency (cf. Halle and Marantz [1993]). This operation is called fusion in DM. Therefore, the nominal aﬃx can be inserted in the head of the Status Phrase fulﬁlling its morphological selection after a suitable inﬁnitival projection has been moved into its Speciﬁer, as is illustrated in (39). (39) [CP

[MP [StatP

[Agr3P ein Buch [AspP zu [F2P lesen ]]]

G

[TP tAgr3P]]]]

FUSION

After Spell-out, the sister of zu in the morphological word is a gerund (Inﬁnitive + G). It is important to note that this analysis is not such an ad hoc solution as it may seem. This analysis of a coherent to-inﬁnitive represents a special case of a classical bracketing paradox. Let us consider another typical example of the latter in (40). Contrary to the morphological bracketing structure indicated in (40b), which is called

OPEN ISSUES

211

for by the restriction that -ier can only attach to maximally bisyllabic adjectives, the scope of the degree morpheme must comprise preﬁx and adjectival stem, since unhappier means ‘being more unhappy’. The paradox can be resolved if it is assumed that -ier is a phrasal aﬃx that fuses with the adjectival stem after its syntactic complement has moved into its Speciﬁer, as is illustrated in (40c, d). (40) a. unhappier b. [un [happy + ier]] c. [DegP -ier

[ModP un

d. [DegP [ModP [un

[AdjP happy]]]

[AdjP happy]] –ier] tModP]

How can we account for the fact that a coherent to-inﬁnitive does not license structural Case? We have to assume that the culprit is the inﬁnitival marker zu. I propose that the inﬁnitival marker that is not temporally linked is a preposition-like element that has the categorial feature [-V]. This [-V] head blocks agreement between Agr3 and little v, which is necessary to assign Accusative Case, as I have argued in chapter 5. These revised assumptions about the syntax of coherent to-inﬁnitives are given in (41). (41)

The inﬁnitival marker zu that is not temporally linked has the categorial feature [-V] and morphologically selects for a gerund.

With these modiﬁcations, we can derive all the properties of coherent to-inﬁnitives. (1) They can license adverbs since the nominalizing phrasal aﬃx attaches high in the structure. (2) They fail to license structural Accusative due to the presence of the inﬁnitival marker with the categorical feature [-V]. (3) The selectional property of the inﬁnitival marker is fulﬁlled after Spell-out in MF. 7.4.5

Conclusions

To summarize, we arrive at a uniform characterization of the diﬀerent occurrences of gerunds as nominalized clauses. In each case the gerund is introduced in the C-domain, more precisely, in the head position of the Status Phrase. More research, especially empirical work on the syntax of nominalized inﬁnitives, is needed to evaluate this hypothesis. Also, more diachronic research would be useful to throw light on the development of these inﬁnitives. However, I have developed an account of nominalized inﬁnitives that provides a coherent account of their essential properties.

7.5 Conclusions In this chapter I have addressed three issues that were left open by my account of restructuring. These were (1) the issue of how to deal with extraposition in a UBH approach, (2) the question of the proper account of VP-topicalization, and (3) the issue of the proper characterization of gerunds in restructuring contexts.

212

SCRAMBLING , REMNANT MOVEMENT , AND RESTRUCTURING IN WEST GERMANIC

With respect to the ﬁrst issue, many questions remain. With respect to the second and the third issue, I have succeeded in providing a comprehensive account that solves some long-standing problems: The description and treatment of the VP-topicalization facts that fall out from my account of restructuring inﬁnitives go a long way in solving the hard problems discussed in section 7.2. The analysis of coherent to-inﬁnitives as gerunds (nominal inﬁnitives) seems promising and accounts for both the similarities with other occurrences of this nominal aﬃx in Doelfoarms, as is evidenced by the presence or absence of the IPP-eﬀect, and the property that distinguishes coherent to-inﬁnitives from the latter, namely, the ability of assigning structural Case. The coherence of the overall picture that this account yields, for instance, the fact that the properties and restrictions of VP-topicalization follow from the ﬁne structure of the diﬀerent verb clusters in West Germanic, provides good evidence that the general approach to restructuring that I have taken in this book, namely, to derive the properties of the diﬀerent types of coherent inﬁnitives from a common sentential source structure through complex derivations, is on the right track.

8

Summary and Conclusions

In this book, I have discussed three salient syntactic phenomena in West Germanic, namely, scrambling, remnant movement, and restructuring, and investigated in detail their interdependence. It was shown that the original account of remnant topicalization in (1a) in terms of prior VP-evacuation via scrambling, as has been proposed by Den Besten and Webelhuth (1989) and as is illustrated in (1b), is mistaken. Scrambling does not feed remnant movement, neither within the clause, as in (1), nor across clauses in cases of restructuring. What has been topicalized in (2a) is not the inﬁnitival clause the direct object of which has been scrambled out, as is illustrated in (2b), but only the inﬁnitival VP: a remnant category that is created by licensing movement and restructuring operations. (1)

a. gelesen hat Hans das Buch read has Hans the book b. [CP[VP tSCR gelesen] hat [IP Hans das BuchSCR t VP]]

(2)

a. zu kaufen hat er das Buch versucht to buy has he the book tried b. [[t SCR zu verkaufen] hat er das Buch versucht]

I have shown that within the clause it is licensing movement (emptying the VP up to the verb) that constitutes the basis for remnant movement of verbal projections, while in cases of restructuring it is remnant movement that feeds apparent scrambling across clauses. 213

214

SCRAMBLING , REMNANT MOVEMENT , AND RESTRUCTURING IN WEST GERMANIC

Licensing movement out of the VP also provides the basis for the antisymmetric account to the syntax of the West Germanic OV languages that I have developed and argued for in this book. All syntactic phenomena are derived within a purely right-branching clause structure via leftward movement into dedicated Speciﬁers of functional heads, rendering superﬂuous the operations of rightward movement and adjunction as well as the assumption of multiple Speciﬁers. It is argued that licensing movement not only aﬀects the constituents of the VP but also applies to the major constituents of the clause, that is, to AspP and TP, which are taken to move into speciﬁc licensing positions in the C-domain. Furthermore, I have argued that restructuring involves movement of the inﬁnitival AspP and TP into dedicated licensing positions in the matrix clause. In this approach, restructuring has the same rationale as subject movement has in cases of (subject) raising constructions: A constituent that fails to be licensed in the embedded clause undergoes further movement to be licensed in the matrix domain. Thus, restructuring falls out as a special case within a theory of generalized licensing that obtains when the licensing heads in the C-domain are defective. Constituents are licensed by checking/validating their features in dedicated functional projections. Along these lines, I have tried to combine Kaynean assumptions about phrase structure with minimalist assumptions about movement and feature checking. In particular, I have adopted the assumption that movements are triggered and that derivations are phase based (Chomsky 2001). The results of the empirical investigation of the three syntactic phenomena of West Germanic within this combined approach that have interesting implications for the construction of the theory are summarized as follows: 1. The restrictions on VP-topicalization follow from the Phase Impenetrability Condition if cyclic Spell-out is assumed (chapter 7). 2. Despite the massive use of remnant movement in this approach, interpretational eﬀects in coherent to-inﬁnitives show that head movement cannot be dispensed with and must be deﬁned in terms of antisymmetric c-command (chapter 5). 3. The principle of Attract Closest can only handle a subset of movement operations and must be replaced with a principle that guarantees strict cyclicity in a derivation (chapter 4). 4. The phenomenon of scrambling implies that if we want to exclude optional movement within a derivational framework, not only a small set of formal features can be taken to underlie a syntactic derivation, as is assumed in narrow syntax, but interface features must be imported into the syntax and be taken to drive derivations as well (chapter 2). 5. The ﬂexibility of adjunction operations can be replaced with a system that allows for feature assignment in the course of the derivation (chapter 2). The most important empirical results are discussed and summarized in the individual subsections dedicated to the diﬀerent chapters that follow.

SUMMARY AND CONCLUSIONS

215

8.1 Scrambling and optionality Chapter 2 provides a comprehensive discussion of scrambling operations in West Germanic. I ﬁrst argue that two types of scrambling operations must be distinguished. I conclude that the scrambling operation in which the scrambled element is stressed has clear properties of A'-movement: It is not clause-bound and may aﬀect arguments and predicates alike. Then, I focus on the properties of the type of scrambling operation in which the scrambled element does not receive a special stress or is completely unstressed. I argue that this type of scrambling operation, called scrambling proper, is clausebound and needs to be captured as an A-movement operation. I identify two types of triggers for scrambling proper, namely, speciﬁcity, deﬁned in pragmatic terms as being familiar to speaker and hearer in the discourse situation, and scope, deﬁned in terms of relational features. I argue that both types of features are checked by A-movement into the Speciﬁers of functional heads. While speciﬁcity is proposed to be checked in the Speciﬁer positions of heads that license weak pronouns, the checking of the relational scope features, in the absence of free adjunction, requires an extension of the minimalist framework that allows for the introduction of non-lexical features in the course of the derivation. Finally, I address the claim by H&R that scrambling is essentially optional and therefore deﬁes any account as triggered movement operation. I argue that a trigger account is indeed feasible in a copy theory of movement in which both LF- and PF-conditions determine which copy is to be spelled out.

8.2 A VO-based account of verb raising and verb projection raising Chapter 3 and chapter 4 introduce Verb Raising and Verb Projection Raising constructions in Dutch and West Flemish. First, I provide an empirical argument for a VO-based approach to the syntax of the West Germanic OV languages. With the help of the inﬁnitival marker it is shown that all VP-internal material is moved out of the VP into dedicated licensing positions in the middle ﬁeld. Then, I address the question of how VR and VPR are to be accounted for in a VO-based approach. 8.2.1

Verb raising

As is illustrated in (3), a typical case of VR in Dutch, nominal arguments of the inﬁnitive and adverbs and adverbials that modify it precede the selecting verb, while the inﬁnitive itself and a sentential complement of the inﬁnitive (3c) follow the selecting verb. In (3), constituents that belong to the embedded clause are given in brackets. Within a VO-based approach, we have to assume that (3a) and (3c) are derived from an underlying structure of the type given in (3b). (3)

a. dat Jan [Marie het boek morgen] wilde [geven] that Jan Marie-DAT the book tomorrow wanted give ‘that Jan wanted to give Marie the book tomorrow’

SCRAMBLING , REMNANT MOVEMENT , AND RESTRUCTURING IN WEST GERMANIC

216

b. dat Jan wilde [PRO Marie het boek morgen geven] that Jan wanted Marie the book tomorrow give c. dat Jan [Marie morgen] wilde [vertellen dat Piet ziek is] that Jan Marie tomorrow wanted tell that Piet sick is ‘that Jan wanted to say to Marie tomorrow that Piet is sick’

I present two arguments that show that the simplest possibility of relating (3a, c) with (3b), namely, by scrambling embedded material into the matrix clause, is untenable. (1) Verb-particles, small clause predicates, and idiomatic expressions cannot scramble, as has been shown in chapter 2, but precede the matrix verb in VR constructions. (2) Adverbs cannot undergo long-distance scrambling. Furthermore, I argue that an alternative approach in which it is assumed that constituents of the embedded clause undergo licensing movement into the matrix domain, rather than scrambling—as has been proposed by Zwart (1993) and then recast in a more advanced minimalist system by Wurmbrand (2001)—is untenable as well. There are two arguments against such an approach to VR-constructions. (1) It would forestall a uniﬁed account of VR- and VPR-constructions and (2) Coherent inﬁnitives comprise more than one licensing domain for their arguments and modiﬁers, as I argued in chapter 5. I then describe in detail my account of VR in terms of remnant movement of the main phases of the inﬁnitival clause into dedicated licensing positions in the matrix clause that dispenses with rightward movement as well as with unmotivated scrambling operations. The approach is based on the generalizations about the basic clausal structure outlined in section 4.2 (cf. [19]) and on the assumption that coherent inﬁnitives are CP-complements, rather than reduced clauses. In this approach, arguments and adjuncts are not moved individually into the matrix clause but are pied-piped by the movement of a larger constituent. This larger constituent is argued to be the inﬁnitival TP, which is moved to a position below all adverbs in the matrix clause. The basic tenets of restructuring are given in the following section. 8.2.2

Restructuring

1. movement of the embedded Aspect Phrase into [Spec,StatP] of the embedded clause 2. movement of the remaining TP of the embedded clause into [Spec,PredP] of the matrix verb via [Spec,MoodP] in the embedded clause 3. movement of the inﬁnitival Tense-head to the functional head that licenses the controller of PRO to ensure the identiﬁcation of the inﬁnitival subject (as is discussed in detail in chapter 5) 4. additional XP-movement of AspP into a Speciﬁer of the selecting verb to account for the formation of verb clusters and the appearance of the IPP-eﬀect (see later discussion and chapter 6 for details) The complex derivation of a VR-construction in Dutch is illustrated in (4). Given my assumptions, (4a) is derived from the source structure in (4b). In the ﬁrst step, the arguments leave the VP to be licensed in the embedded clause, as is illustrated in (4c).

SUMMARY AND CONCLUSIONS

217

In the next step, AspP that has been emptied up to the verb is moved into [Spec,StatP] of the inﬁnitival (and then moved on into a position that precedes the matrix verb but is spelled out in the inﬁnitival clause. The result of this operation is shown in (4d). In the following step, the remaining TP of the inﬁnitival is moved via [Spec,MoodP] of the embedded clause to [Spec,PredP] of the matrix verb. The resulting structure is given in (4e). The reason for these licensing movements into the matrix domain is a deﬁcient complementizer that cannot fully value the licensing projections of AspP and TP in the inﬁnitival C-domain. In particular, it is assumed that StatP fails to check the subcategorization of the matrix verb; thus the inﬁnitival AspP moves into a checking position in the V-domain of the selecting verb, giving rise to the formation of verb clusters. The MoodP is unable to temporally link the inﬁnitival TP, which, not denoting an event-token, fails to qualify as an argument of matrix verb and thus “restructures” as a predicate by moving into the PredP of the matrix verb. In the ﬁnal step, both the matrix subject and the embedded direct object scramble to positions above the sentential adverb vaak. This last step is optional. Hence both dat vaak Jan het boek lang wil lezen and dat Jan vaak het boek lang wil lezen are ﬁne sentences in Dutch. If we replace the adverb often with sentence negation in (4), then scrambling of both arguments becomes obligatory (in the absence of any contrastive focus) as we expect (cf. the discussion of scrambling in chapter 2). Scrambling of embedded arguments is enabled by extraction of the TP out of the embedded CP. Here we see one eﬀect of the diﬀerentiation between licensing movement and scrambling. The arguments of the coherent inﬁnitive are licensed in the embedded clause but can undergo scrambling according to their referential or quantiﬁcational properties in the domain of the matrix clause. (4)

a. dat Jan het boek vaak lang wil lezen that Jan the book often long wants read ‘that Jan often wants to read the book for a long time’ b. [dat vaak [VP Jan wil [CP . . . [lang [AspP 0 [VP PRO lezen het boek]]]]]] Step 1: licensing movement in the embedded clause c. [dat vaak Jan [VP wil [CP [TP PRO het boek lang [AspP [VP lezen]]]]]] Step 2: AspP moves into [Spec,StatP] in the embedded clause d. [dat vaak Jan wil [CP [AspP lezen] [TP PRO het boek lang tAspP]]] Step 3: TP moves into [Spec,PredP] in the matrix clause e. [dat vaak Jan [PredP [TP PRO het boek lang t AspP] wil [CP lezen tTP]]] Step 4: scrambling of the matrix subject and the embedded object f. [dat Jani het boekj vaak ti [PredP [TP PRO tj lang] wil lezen]]

8.2.3

Verb projection raising

In cases of VPR, the verb cluster may contain arguments and adverbs that belong to the embedded inﬁnitival, as is illustrated in (5). What is peculiar about this partition of arguments is the following observation: A scopal element that is outside of the verb cluster can take its scope inside or outside of the verb cluster, but an element

218

SCRAMBLING , REMNANT MOVEMENT , AND RESTRUCTURING IN WEST GERMANIC

that is part of the verb cluster may take its scope only inside of the verb cluster. It is important to note that an element outside of the verb cluster in its narrow scope reading always takes wide scope with respect to material that has undergone “VPR.” So, for instance, (6a) cannot mean that Jan made Valere three times read two books (possibly diﬀerent ones at each occasion). (5)

a. da Marie Jan nen boek tR wilt [geven]R that Maria Jan a book wants give b. da Marie Jan tR wilt [nen boek geven]R that Marie Jan wants a book give c. da Marie tR wilt [Jan nen boek geven]R that Marie wants Jan a book give ‘that Marie wants to give Jan a book’

In my approach, the VPR-structure in (6a) diﬀers from a typical VR-structure only in the amount of structure that is moved by the Aspect Phrase into the C-domain of the inﬁnitival. The interpretational eﬀect described in (6a) then simply follows from the mechanics of the system. In the inﬁnitival clause, arguments cannot only undergo licensing movement but can also undergo scrambling, for instance, in order to take scope over an adjunct. In chapter 2, we have seen that a DP that is scrambled across an adverb has necessarily wide scope with respect to such an adverb but has necessarily narrow scope with respect to such an adverb if it fails to scramble. If we assume that it is possible to pied-pipe the phrase that corresponds to the domain that contains these adverbs with AspP-movement into the C-domain, then it follows without stipulation that the DP in (6a) necessarily has wide scope over the adverb in the VPR-complement: In more simple terms, in order for an element to move into the domain of the selecting verb via TP-movement, it has to take scope over the elements that stay behind and are pied-piped by movement of AspP, as is illustrated in (6b). In (6b), PPD stands for pied-piped domain, that is, the domain that has been pied-piped by movement of AspP. (6)

a. da Jan Valere twee boeken deeg drie keers lezen that Jan Valere two books made three times read b. da Jan Valere [PredP [VP deeg [CP [TP [twee boeken]i [PPD drie keers ti lezen]]]]]

8.2.4

An account of sentential complementation

Finally, I propose an account of sentential complementation that allows us to derive the movement operations argued for earlier. The basic idea is that the local C-domain is responsible for licensing the AspP and TP in the clause, with the complementizer acting as a placeholder for the selectional requirements of the selecting verb. In this approach, movement of AspP and TP, which we argued to make up restructuring, occur in every clause. Following recent work on the split C-domain (cf. Rizzi [1997]), I assume that the C-domain is made up of various functional projections. I assume that the com-

SUMMARY AND CONCLUSIONS

219

plementizer is inserted in StatP, where it licenses the ﬁniteness of the clause and moves through MoodP, where the tense of the clause is linked to the speaking time with matrix clauses and to the matrix event time with embedded clauses, to Force, constituting the highest head in the C-domain. This is illustrated in (7), where it is assumed that the traditional CP corresponds to ForceP (CP = ForceP). (7)

[CP

Force [MoodP

M

[StatP

S

[TP . . .]]]]

dass

In an embedded clause with a non-deﬁcient complementizer, the complementizer will value the heads Stat0 and Mood0 such that the embedded AspP and TP can be licensed in the respective Speciﬁer, as is illustrated in (8). In this approach, the ﬁnite verb in matrix clauses only undergoes local movement within the C-domain: After the Aspect Phrase that contains the ﬁnite verb has been moved into [Spec,StatP], the latter extracts and moves via the head of MoodP to the highest head in the C-domain, as is illustrated in (9). In the case of a restructuring verb, the complementizer is deﬁcient and fails to license the embedded AspP and TP, which move on into dedicated licensing positions in the matrix clause, as is illustrated in (10). (8)

licensing movements in an embedded clause MoodP Force StatP M TP S

dass

(9)

licensing movements in a main clause

F TP M AspP S Vfin e

AspP

220 (10)

SCRAMBLING , REMNANT MOVEMENT , AND RESTRUCTURING IN WEST GERMANIC

licensing movements in a coherent inﬁnitive PredP

Pred

FP (= functional position in the V-domain) CP = ForceP V C M TP S

AspP

0[defective]

Assuming that a biclausal analysis is appropriate, let me take up at this point the question of whether the embedded clause should be a full CP, as is proposed earlier, or something smaller. In other words, the question arises of how we can distinguish between an analysis of coherent inﬁnitives with a defective C-layer or a missing C-layer. If we assume that coherent inﬁnitives are TPs in the latter approach, the same kind of licensing movements into the matrix clause without the intermediate steps in the C-domain in the embedded clause could be assumed. I would like to discuss this point again in order to render some assumptions that I make in diﬀerent parts of this book more explicit. On the empirical side, I argue for the presence of MoodP and StatP in restructuring inﬁnitives in order to be able to handle extraposition. In the absence of rightward movement, TP- and AspP-movement must be taken to apply in every clause in coherent inﬁnitives (cf. the discussion of the data in [8] in chapter 7). In chapter 4, I argue that the C-domain, though defective, is not completely inert. The complementizer in coherent constructions is not fully valued or underspeciﬁed. I argue that it selects for a nonﬁnite complement, thereby excluding ﬁnite clauses from restructuring in German, and for a subjunctive complement, thereby excluding factive complements from restructuring contexts. On more conceptual grounds the rationale of this approach is that these verbs simply s-select for propositions and that the canonical syntactic representation of a proposition is uniformly taken to be a CP. In other words, the diﬀerence between restructuring and non-restructuring verbs is not that the latter take propositions and the former something smaller, say, event-descriptions or something, but that they diﬀer solely in the way they license their propositional complements. The basic question about coherent inﬁnitives has always been the issue of whether these inﬁnitives are full or reduced clauses. I have no new insight to oﬀer that could decide this question. My analysis solely provides a technical solution that allows us to treat coherent inﬁnitives as full CPs, provided that we accept the notion of deﬁciency as it is understood earlier.

SUMMARY AND CONCLUSIONS

221

8.3 Coherent infinitives in German and the issue of monoclausality In chapter 5, I provide empirical arguments to show that coherent bare inﬁnitives in German are biclausal. The empirical arguments come from two diﬀerent domains, the order of adjuncts and the binding properties in coherent inﬁnitives, but point into the same direction, namely, that more than one licensing domain is present in coherent inﬁnitives. 8.3.1

The order of adjuncts

Adjuncts in coherent inﬁnitives with a selecting modal verb can appear in an order that is impossible in monoclausal structures. As is illustrated in (11), an aspectual adverb that modiﬁes the matrix verb precedes a temporal adverb that modiﬁes the inﬁnitive in a coherent inﬁnitival construction. Since temporal adjuncts occur in a structurally higher position than aspectual adjuncts (cf. Cinque [1999]), this order is ungrammatical in a single clause. These data speak against Cinque’s (2001) analysis of modal verbs as functional restructuring verbs as well as Wurmbrand’s (2001) monoclausal analysis of modal verbs as lexical restructuring verbs. In my approach these data are unproblematic, since coherent inﬁnitives are taken to comprise a separate licensing domain for each verb and adjuncts that modify the inﬁnitive are moved, via TP-movement to PredP, into a position below all adjuncts that modify the matrix verb. (11) a. weil Peter mich schon lange heute besuchen wollte since Peter me already for-a-long-time visit wanted ‘already for a long time Peter has wanted to visit me today’ b. *weil mich Peter schon lange heute besucht hat since me Peter already for-a-long-time today visited has c. *weil das Peter schon lange heute wollte since that Peter already for-a-long-time today wanted d. weil mich Peter heute schon lange besucht hat since me Peter today already for-a-long-time visited has

8.3.2

The binding properties of ECM-inﬁnitives

The binding properties of coherent inﬁnitives embedded under ECM-verbs reveal that these inﬁnitivals comprise two (distinct) binding domains. Assuming that the binding domain within the clause is constituted by the TP by default, I take these data as indicating that coherent inﬁnitives must be at least as big as TPs. As is indicated in (12), while in a single clause both objects must be disjoint from the subject, the embedded object in a coherent inﬁnitive can be coreferent with the matrix subject. Since the negation in (12) can have matrix scope (the most natural reading), the embedded subject and object must have moved into the matrix TP, after restructuring has moved the embedded TP into [Spec,PredP] of the matrix verb, as is illustrated in (12d).

222

SCRAMBLING , REMNANT MOVEMENT , AND RESTRUCTURING IN WEST GERMANIC

(12) a. weil eri ihn*i/j sie nicht waschen liess since he him her not wash let-Past b. weil eri sie ihni/j nicht waschen liess since he her him not wash let-Past c. weil der Hansi ihn*i/j ihm*i/k vorstellte since the Hans him-ACC him-Dat introduced d. [CP weil [TP1 er siei ihnj [nicht [PredP [TP2 ti tj] waschen liess/sah]]]]

Assuming that the embedded subject is Case-licensed in the matrix clause and the embedded object is Case-licensed in the embedded clause, the binding properties of ECM-inﬁnitives follow from the principle in (13). These data therefore also provide evidence against the accounts of Wurmbrand (2001) and Zwart (1993), who assume that in VR-constructions the arguments of the inﬁnitive must be licensed in the domain of the matrix verb. (13)

8.3.3

Pronouns and anaphors are interpreted in their Case-licensing positions (i.e., they must be reconstructed into their Case-licensing position before the Binding Theory applies at LF).

Coherent to-inﬁnitives

Coherent to-inﬁnitives in German have ﬁgured prominently in the debate of whether coherent inﬁnitives are to be analyzed as monoclausal or biclausal structures. Coherent to-inﬁnitives allow for the so-called long passive and both Haider (1991) and Wurmbrand (2001) take this fact as conclusive evidence for the monoclausality of these inﬁnitives. As is illustrated in (14), in a long passive what would be the object in the embedded clause is realized as the Nominative subject of the matrix verb. (14)

Der Zaun wurde zu reparieren versprochen the fence was to repair promised ‘Someone promised to repair the fence’

According to Haider (1991), this indicates that the object of the inﬁnitive is Caselicensed by the matrix verb (with the object receiving Nominative Case if the matrix verb is passivized). That an embedded argument is Case-licensed by the matrix verb is familiar to us from ECM-verbs. In an ECM-construction, the subject as the highest argument can undergo Case-licensing movement into the matrix domain. But the same analysis cannot be applied to coherent to-inﬁnitives, since the embedded object cannot be taken to undergo Case-licensing movement into the matrix domain across the embedded subject, that is, PRO. Therefore, a biclausal analysis of coherent to-inﬁnitives seems to be impossible. However, such an analysis is possible in my account of restructuring that is based on remnant movement of parts of the inﬁnitival clause. The derivation proceeds in

SUMMARY AND CONCLUSIONS

223

the following way: The embedded object pronoun contained within the Agreement Phrase is pied-piped by movement of the AspP via the C-domain of the inﬁnitival into [Spec,AspP] of the matrix verb, while PRO contained within the embedded TP is moved into [Spec,PredP] of the matrix verb. Note that in this ﬁrst step the embedded object does not move across PRO. In the second step, the embedded object moves out of AspP to its Case-licensing position in the matrix clause. Movement of the embedded object is again unhampered by PRO, since PRO in the matrix clause is contained in a larger phrase, namely, the embedded TP. The derivation for a typical case like weil Hans ihr ihn zu waschen empfahl (‘since Hans recommended to her to wash him’) is given in (15). (15)

[CP Hans ihr empfahl [CP [TP PRO [AgrP ihn zu waschen]]]] Hans her recommended him/self to wash Step 1: AgrP (the extendend AspP) moves via [Spec,StatP] into [Spec,CP] a) [CP Hans ihr empfahl [CP [AgrP ihn zu waschen] [TP PRO tAgrP]]] Step 2: AgrP moves to [Spec,AspP] in the matrix clause b) [CP Hans ihr [AspP [AgrP ihn zu waschen] empfahl [CP tAgrP [TP PRO tAgrP]]] Step 3: TP moves to [Spec,PredP] in the matrix clause c) [CP Hans ihr [PredP [TP PRO tAgrP] [AspP [AgrP ihn zu waschen] empfahl [CP tAgrP]]]] Step 4: The embedded direct object moves into its Case-position without crossing PRO d) [CP Hans ihr [AgrO ihnk [PredP [TP PRO tAgrP][AspP [AgrP tk zu waschen] empfahl]]]]

This shows that a biclausal analysis of coherent to-inﬁnitives is possible, if an independent explanation for why the embedded object cannot be Case-licensed within the inﬁnitival clause is provided. I argue that coherent to-inﬁnitives are nominalized inﬁnitives (also called gerunds in this book), in which the categorial status of the inﬁnitival marker is responsible for blocking assignment of structural Case to the object. Then I provide two arguments that indicate that a biclausal analysis of coherent to-inﬁnitives is not only possible but also necessary. First, I show that the binding properties of anaphors in coherent to-inﬁnitives, illustrated in (15e) later, are problematic for a monoclausal approach, while they follow naturally from the tenets of my biclausal approach. Second, I take Haider’s (1991) tests and criteria for monoclausal inﬁnitives (derived from the properties of coherent to-inﬁnitives) and show that these properties also hold of the other coherent inﬁnitives, including those that clearly have biclausal properties. Since a monoclausal analysis of coherent ECM-inﬁnitives and of coherent modal inﬁnitives is impossible, the monoclausal analysis of coherent to-inﬁnitives is to be discarded on grounds of parsimony. (15) e. *weil sichi/j ihrj der Hansi zu waschen empfahl since herself/himself her the Hans to wash recommended ‘since Hans recommended to her to wash herself/himself’

224

SCRAMBLING , REMNANT MOVEMENT , AND RESTRUCTURING IN WEST GERMANIC

8.4 The IPP-effect and the unified account of verb clusters in West Germanic In chapter 6, I propose an account of the IPP-eﬀect and provide a uniform format for the analysis of left- and right-branching verb clusters in German, Dutch, and West Flemish. In chapter 4, I argue that the inﬁnitival TP and AspP cannot be licensed in the embedded C-domain and move into dedicated licensing positions in the matrix clause. While the inﬁnitival TP is licensed in [Spec,PredP] of the matrix clause, the inﬁnitival AspP that contains the dependent verb moves into a licensing position below PredP that needs to be identiﬁed. The purpose of the ﬁnal step in the licensing movement of the dependent verb is twofold: (1) the subcategorization of the selecting verb needs to be checked. Following Bech (1955/1985), I assume that a verb selects for the status of its nonﬁnite complement. That is, it determines whether the dependent nonﬁnite verb is a participle, a bare inﬁnitive, or a toinﬁnitive. (2) Following recent work on Tense that requires verbs to be temporally anchored, I propose that dependent verbs must be linked to the matrix event time. Nominal categories, including nominalized inﬁnitives, are exempted from this formal licensing requirement. Chapter 6 discusses various types of evidence for determining which functional positions in the V-domain serve which licensing function. Example (16) illustrates the functional positions in the V-domain that are argued for in this book. (16)

8.4.1

Skeleton of functional positions in the V-domain [AspP (zu) [F2P [F3P [VP . . . ]]]]

Accounting for the IPP-eﬀect

In section 6.2, I argue that IPP-inﬁnitives are hidden participles and that the IPPeﬀect reduces to a structural incompatibility between the participial preﬁx and the inﬁnitive dependent on the restructuring verb, on account of the fact that the languages and dialects in which the participle is formed without a preﬁx, namely, Frisian and Low German, do not display an IPP-eﬀect. In the West Germanic languages that display the IPP-eﬀect, the participle is formed by aﬃxation of the preﬁx ge and the suﬃx t/d. I follow Halle and Marantz (1993) in assuming that inﬂected forms are (partially) derived in the syntax. More speciﬁcally, I propose that the participial preﬁx ge is inserted in [Spec,F2P] of the participial phrase. The verb in the participial phrase will ﬁrst move to F2, to check its preﬁx, and then up to the Aspect-head to merge with the suﬃx that contains the temporal interpretation. In the ﬁnal step the preﬁx left-adjoins to the complex of verb and suﬃx, to form the participle before Spell-out. This is illustrated in (17). (17)

[AspP -t [F2P [ge] [F2 [VP V]]]]

If the verb in the participle phrase is a restructuring verb, then the dependent inﬁnitive will move into [Spec,F2P] for licensing purposes. It follows that a verb in

SUMMARY AND CONCLUSIONS

225

participial form and a bare inﬁnitive selected by such a verb rule each other out. In this case the participial preﬁx is blocked by the dependent inﬁnitive, that is to say, it cannot be inserted. Since the preﬁx is selected by the participial suﬃx, the latter is dropped and the verb remains in F2 and is spelled out with the default morphology of a bare inﬁnitive. Instead a zero-morpheme is inserted in the head of AspP that contains the formal feature [+ participle] and a semantic feature [+ past] to guarantee the correct interpretation of the hidden participle phrase. This is illustrated in (18). IPP-inﬁnitives in German are obligatorily right-branching, since movement of the zero-morpheme into the Aspect Phrase of the selecting auxiliary strands the IPPinﬁnitive in [F2P] below. (18)

8.4.2

[AspP 0 [F2P [dependent inﬁnitive] IPP-inﬁnitivei [VP ti ]]]

A uniﬁed analysis of verb clusters in West Germanic

While the surface data (in German) suggest that dependent verbs in restructuring contexts are licensed in [Spec,AspP] of the selecting verb, I provide evidence from the syntax of IPP-complements in West Flemish and Afrikaans that participles and verbal inﬁnitives move through the Speciﬁer of F2P below AspP (and sometimes remain there). With the help of Frisian data, I establish that the West Germanic dialects have two types of inﬁnitives, one being directly licensed in AspP, the other, like participles, moving through [Spec,F2P] below AspP, thereby giving rise to the IPP-eﬀect. Linking this observation with the fact that coherent to-inﬁnitives, which I proposed to analyze as nominal inﬁnitives in chapter 5, never give rise to an IPPeﬀect, I propose that F2 is responsible for temporal linking of dependent verbs, while the subcategorization of the matrix verb can be checked either in [Spec,AspP] or in [Spec,F2P]. The latter choice seems to depend on the amount of head movement of the verb in the V-domain in a language. While German nonﬁnite verbs always move into the highest head in the V-domain, that is, AspP, nonﬁnite verbs in Dutch and West Flemish only move as far as F2. This has important consequences for the analysis of verb clusters in the three languages. A dependent nonﬁnite verb in a left-branching verb cluster in German must be analyzed as occupying [Spec,AspP] preceding the selecting verb in Asp0. A dependent nonﬁnite verb in a right-branching verb cluster in Dutch or West Flemish must be analyzed as occupying [Spec,F3P] following the selecting verb in F2. (19) a. Right-branching verb clusters in Dutch and Westﬂemish: temporal linking and subcategorization checking in [Spec,F2P], the selecting verb remains in F2 Spell-out in [Spec,F3P], the left edge of the V-domain remains empty [AspP (te) [F2P [V2] V1 [F3P [V2] [VP . . . ]]]] b. Left-branching verb clusters in German: temporal linking in [Spec,F2P], subcategorization checking in [Spec,AspP] Spell-out in [Spec,AspP], the selecting verb moves to the highest V-position [AspP [V2] V1 [F2P [V2] [F3P [VP . . . ]]]]

226

SCRAMBLING , REMNANT MOVEMENT , AND RESTRUCTURING IN WEST GERMANIC

c. Right-branching verb clusters in German same as above but the dependent inﬁnitive is spelled out in [Spec,F2P] the left edge of the V-domain is occupied [AspP [V2] V1 [F2P [V2] [F3P [VP . . . ]]]] d. Right-branching verb cluster projected by an IPP-inﬁnitive in German [AspP Aux [F2P [V3 V2] [F3P [VP . . . ]]]] the zero-morpheme adjoins to the auxiliary in the highest head position in the V-domain; thus [Spec,AspP] remains an available escape hatch

8.4.3

Extraction from the V-domain and the Phase Impenetrability Condition

Based on the preceding analysis of verb clusters in West Germanic, I show that the diﬀerent status of topicalized right-branching verb clusters in German and Dutch/West Flemish follows from the prosodic condition in (21) and the Phase Impenetrability Condition (PIC). While right-branching verb clusters in Dutch can be topicalized, topicalization of right-branching verb clusters, with the exception of clusters that comprise IPP-inﬁnitives, leads to ungrammaticality in German, as is illustrated in (20). (20) a. ?*[müssen lesen können] wird er den Text must read can will he the text b. ?[haben lesen wollen] wird er den Text have read want-IPP will he the text ‘he will have wanted to read the text’ c. [moeten kunnen lezen] zal hij het boek must can read will he the book ‘he will have to be able to read the book’ (21)

A right-headed phonological phrase in a verb cluster must sit on a right branch with respect to the non-head.

Extraction out of a verb cluster must proceed via the left edge of the V-domain, that is, [Spec,AspP]. Extraction of a right-branching verb cluster will thus lead to a violation of the prosodic constraint in (21). Violation of this constraint will lead to ungrammaticality if there is another Spell-out option as is the case in (20a), which could have been spelled out as lesen können müssen, but only to a marked grammatical result if there is no other Spell-out option, as is the case in (20b), since IPPinﬁnitives are obligatorily right-branching. This account presupposes that Spell-out options in the V-domain are ﬁxed before the derivation reaches the C-domain, since no such prosodic condition is at work in the C-domain in German. To sum up, the account of the rather subtle diﬀerences in (20) is based on three assumptions: (1) The Aspect Phrase (not the VP) constitutes a phase, (2) Spell-out is cyclic (rather than ensuing at the end of the entire derivation), and (3) there are interface constraints (like the mapping rule between syntactic

SUMMARY AND CONCLUSIONS

227

structure and prosodic structure in [21]), whose violation leads to ungrammaticality under certain conditions.

8.5 Extraposition, VP-topicalization, and the status of gerunds Chapter 7 addresses a number of open questions. The ﬁrst issue pertains to extraposition, a notoriously diﬃcult topic within antisymmetric approaches. Without trying to devise a comprehensive account of extraposition, section 7.1 addresses a technical problem that is brought about by the account of sentential complementation developed in chapter 4. Because of their licensing requirements, CP-complements become part of the verb cluster (they are licensed in [Spec,F3P] of the selecting verb and due to the PIC become inaccessible for further computation. Clearly, this result is unwanted since CP-complements (1) can be topicalized and (2) must be “extraposed” from leftbranching verb clusters. 8.5.1

Extraposition from verb clusters

I provide a technical solution to this problem, which treats extraposition as leftward movement into a high Speciﬁer in the clause that follows from the condition in (22), which is a rendition of Büring and Hartmann’s (1997) account, and makes use of the proposal that both TP and AspP undergo licensing movement into the C-domain. A case of extraposition is illustrated in (23). At the end of the derivation neither Tense nor Aspect c-command the Tense within the “extraposed” CP, as demanded in (22). From its “scope” position in (23) the CP-complement, being outside of the verb cluster, can be topicalized on its own as well as be stranded or pied-piped by topicalization of the verb cluster, that is, by movement of the AspP into [Spec,CP]. (22)

A Tense-head may be neither in the checking domain nor in the scope (deﬁned by C-command) of an Aspect- or Tense-head.

(23) a. [CP C [MoodP M [StatP S [TP [AspP V CP]]]]] b. [CP C[MoodP M [StatP S [CP [TP T [AspP V t]]]]]] extraposition c. [CP C[MoodP M [StatP [AspP V t] S [CP [TP T]]]]]

licensing of the AspP

d. [CP C[MoodP [TP T] M [StatP [AspP V t] S [CP]]]]

licensing of the TP

Another issue concerns the topicalization of verb-projections in coherent inﬁnitives. As we have seen in chapter 5, topicalization of the verb cluster or the dependent inﬁnitive alone is one criterion for detecting a coherent construction. However, the test of topicalization also shows that large parts of the embedded clause can also be topicalized, even with verbs that restructure obligatorily. This is a hard problem, which to my knowledge has not been given a satisfactory explanation so far. In section 7.2, I outline an account that takes advantage of the availability of the two kinds

228

SCRAMBLING , REMNANT MOVEMENT , AND RESTRUCTURING IN WEST GERMANIC

of inﬁnitives in West Germanic. I argue that the two forms are in partial competition with each other, with the gerund functioning as a means of last resort. Though some questions of a mostly technical nature remain, this account paves the way to a general solution to this problem. 8.5.2

VP-topicalization

VP-topicalization data raise two kinds of problems for the account that I have developed in this book. First, VP-topicalization leads to a bleeding of the IPP-eﬀect in Dutch and West Flemish, while in German the IPP-eﬀect is only voided with perception verbs. Second, the dependent inﬁnitive can be topicalized together with one of its arguments to the exclusion of the VR-verb. This is unexpected since verb and argument do not form a constituent anymore in my account after restructuring has applied. The solution that I provide makes use of the diﬀerent ﬁne structure of verb clusters in German and Dutch/West Flemish, as it is outlined in chapter 6, the PIC, and the availability of the nominalized inﬁnitive (the gerund) as a means of last resort. In German, allowing for VPR, dependent inﬁnitive plus argument embedded within an IPP-inﬁnitive can be extracted out of the verb cluster due to last-resort movement to the left edge of the V-domain. Thus, VP-topicalization generally does not lead to a bleeding of the IPP-eﬀect in German. Only in the case of perception verbs, movement of last resort of the inﬁnitive that invokes the IPP-eﬀect is blocked by a more economic derivation that involves the gerund, which generally fails to induce an IPP-eﬀect, since it is licensed directly in the left edge of the V-domain. In Dutch and West Flemish, however, movement of the dependent inﬁnitive to the left edge of the V-domain from [Spec,F3P] is blocked by its own copy in [Spec,F2P]. Therefore, the gerund, which as a phrasal aﬃx can attach to any extended projection of an inﬁnitive, is inserted in the course of the derivation as a means of last resort. This gerund, on the one hand, will not induce an IPP-eﬀect in cases of VPtopicalization, since it is licensed directly in the left edge of the V-domain. On the other hand, it is blocked by the more economic derivation that involves the (verbal) inﬁnitive in restructuring constructions without VP-topicalization in Dutch (and West Flemish), since it is not selected by the restructuring verb, explaining why Dutch verb clusters may only contain verbs (and verb particles) when untopicalized but may contain arguments and adjuncts when topicalized. Summing up, the diﬀerences in VP-topicalization between German, on the one hand and Dutch and West Flemish, on the other hand, follow from the ﬁne structure of the verb clusters in these languages and the Phase Impenetrability Condition. 8.5.3

A uniﬁed analysis of the gerund

In chapters 6 and 7, I discuss diﬀerent occurrences of nominalized inﬁnitives. I assume that some occurrences of gerunds are selected (cf. the so-called Doelfoarms) and that sometimes they can be used as a means of last resort. Furthermore, I propose that coherent to-inﬁnitives involve a nominalized inﬁnitive as well. The common assumption that I made about these diﬀerent occurrences is that the gerund is a

SUMMARY AND CONCLUSIONS

229

phrasal aﬃx of nominal nature that morphologically selects for an inﬁnitive and that this requirement is satisﬁed via adjacency. Section 7.4 proposes a uniﬁed account of the diﬀerent occurrences of the gerund. The gerund is treated as a phrasal aﬃx that always nominalizes full clauses. I propose that parallel to complementizers, this nominal aﬃx is inserted in the head position of the Status Phrase above TP. In this analysis, gerunds can be taken to be nominalized clauses that license adverbs and Case. This is especially important for the analysis of coherent to-inﬁnitives in German. Remember that the latter can license adverbs but fail to license structural Case. Now, the licensing of adverbs follows since the nominal aﬃx attaches above the inﬁnitival TP while the failure to license structural Case is relegated to the categorial nature of the inﬁnitival marker in coherent to-inﬁnitives. This analysis of coherent to-inﬁnitives highlights the importance of the role of the Status Phrase for sentential complementation. The Status Phrase is not only responsible for checking the ﬁniteness of the verb but also the place where the complementizer and nominalizing aﬃxes are inserted. These elements are essential for qualifying an embedded proposition as an argument of the selecting verb by nominalizing the embedded clause.

8.6 The connection between scrambling, remnant movement, and restructuring In this section, I would like to discuss how the analysis of VP-topicalization and the connection between scrambling, remnant movement, and restructuring that I sketched in chapter 1 are accounted for in the approach that is developed in this book. 8.6.1

The analysis of remnant topicalization

A typical case of VP-topicalization (also simply called remnant topicalization) is given in (24a). Its analysis in the standard account following Den Besten and Webelhuth (1987) is illustrated in (24b). In this account, arguments and adjuncts are scrambled into the matrix clause, while the inﬁnitival clause—however big it is supposed to be—that contains the remnant inﬁnitive is moved into [Spec,CP] of the matrix clause. In this account, (long-distance) scrambling permitted by the speciﬁc properties of a restructuring inﬁnitive feeds remnant topicalization. In my account, only the inﬁnitival AspP, a remnant category created by the clausal split triggered by restructuring, is topicalized. The topicalized AspP is a remnant category that only contains traces of licensing movement. Arguments and adjuncts of the inﬁnitival are not scrambled into the matrix clause but arrive there via remnant movement of the containing inﬁnitival TP, as is illustrated in (24c). Additional scrambling of a constituent can then obtain according to its referential and quantiﬁcational properties, either within the inﬁnitival TP, as is illustrated in (24c), or into the matrix TP. In this account, scrambling does not feed remnant topicalization. Remnant categories are solely created by standard licensing movements and licensing movements induced by restructuring.

230

SCRAMBLING , REMNANT MOVEMENT , AND RESTRUCTURING IN WEST GERMANIC

(24) a. besuchen will Hans die Maria morgen visit wants Hans-NOM the Maria-ACC tomorrow ‘Hans wants to visit Maria tomorrow’ b. [tScr tScr besuchen] will Hans die MariaScr morgenScr c. [AspP besuchen [VP tCase]] will [TP1 Hans [TP2 die MariaScr morgen tScr]]

As I have noted in chapter 6, bare inﬁnitives can be topicalized together with their direct object, while to-inﬁnitives fail to do so. This diﬀerence follows from the diﬀerent licensing status of direct objects in bare inﬁnitives and to-inﬁnitives. 8.6.2

Diﬀerences between bare inﬁnitives and to-inﬁnitives

For the sake of illustration, let us compare the derivations of the minimal pairlike sentences in (25). (25) a. die Maria besuchen wollte Hans morgen the Maria-ACC visit wanted Hans-NOM tomorrow ‘as for visiting Mary John wanted to do it tomorrow’ b. ??die Maria zu besuchen wünschte Hans morgen the Maria-ACC to visit wished Hans-NOM tomorrow

With the bare inﬁnitive in (25a), the direct object is Case-licensed by the inﬁnitive and both VR and VPR can apply to the inﬁnitival complement. If VPR applies, the direct object is pied-piped by movement of the Aspect Phrase into the matrix clause and can thus be topicalized with the dependent inﬁnitive by extracting out of the verb cluster created by restructuring. The essential movements of the derivation of (25a) are given in (26). (26)

CP1 AspP1 VP1 V' V1

CP2 AgrP

AspP2

die Maria besuchen

In (25b), however, the direct object, due to the presence of the inﬁnitival marker and its categorial nature in restructuring contexts, cannot be licensed in the embedded clause. The direct object is pied-piped by movement of the inﬁnitival AspP into the matrix clause but has to undergo Case-licensing movement into the matrix TP

SUMMARY AND CONCLUSIONS

231

after verb cluster formation. Thus, “VP-topicalization,” that is, movement of the inﬁnitival Aspect Phrase into the matrix [Spec,CP], cannot aﬀect the direct object, deriving that only verbs can be topicalized with coherent to-inﬁnitives. The essential movements of the derivation of (25b) are given in (27). (27)

CP1 TP1 AgrP1 AspP1 AgrP2

VP1 V' V1

die Maria

CP2 AgrP2 AspP2

8.6.3

Conclusions

The complex interplay between restructuring, remnant movement, and scrambling that is evidenced in cases of remnant topicalization is the result of licensing movements of various types that occur in coherent as well as in non-coherent clauses. What is special about restructuring inﬁnitives is that the main constituents of the inﬁnitival that are otherwise licensed in the embedded C-domain are licensed in the matrix clause. The movements that ensue from these licensing requirements lead to the formation of verb clusters and are responsible for the general transparency of coherent inﬁnitives. Verb cluster formation in turn is the basis of remnant topicalization, as we have seen earlier. Scrambling, however, while not playing any role in restructuring and verb cluster formation itself, is the operation that is responsible for the socalled “matrix clause-interpretation” of arguments that belong to the inﬁnitive and applies after the licensing movements of the inﬁnitival TP and AspP have “restructured” the clause. Moreover, I have shown that the facts of remnant topicalization provide convincing evidence for the uniﬁed biclausal analysis that I have proposed for coherent bare inﬁnitives and to-inﬁnitives. Finally, I have argued in this book that all movement operations involved in the complex derivations that lead up to remnant topicalization can be given a coherent account within a phase-based minimalist framework that only employs leftward movement into unique Speciﬁers of dedicated functional positions that are motivated by feature-checking requirements of the main constituents of the clause.

This page intentionally left blank

NOTES

Chapter 1 1. In (14), I assume that adjunction is to the VP. Whether this is correct or whether adjunction has to apply to the local IP is immaterial to my purposes here. 2. The term kohärenter Inﬁnitiv (‘coherent inﬁnitive’) was introduced by Bech (1955), who, to my knowledge, was the ﬁrst to describe in a fairly sophisticated and formal way the properties of this type of inﬁnitivals. To honor Bech’s pioneering work, the term coherent is standardly used within the German linguistic literature on the subject. Within the Romance tradition of work on the subject (cf. Burzio [1986], Rizzi [1978, 1982]) restructuring inﬁnitives became the standard term of reference. I will use both terms interchangeably in this book. 3. An anonymous reviewer points out that the relevant restriction on remnant movement, namely that the remnant creating and the remnant moving operations cannot be of the same type, falls out as a special case of a general restriction derivable from Williams’s (2003) recent work on representation theory and that Attract Closest seems insuﬃcient to characterize all cases of illicit remnant movement not ﬁltered out by strict cyclicity. The reviewer is correct in this position, and I show in chapter 4 that certain cases of illicit remnant scrambling are independently ruled out as cases of illicit scrambling of predicates. The point of (38) is to show that Müller’s restrictions on remnant movement can be interpreted as cases of an A-over-A eﬀect, which can be cast in the MP as a violation of Attract Closest. The reviewer is also correct in his point that the application of Attract Closest must be limited in a way as to allow, for instance, for the parallel movement of subject and object out of the VP into licensing positions in the IP. This issue will be dealt with in chapter 4. 4. Note, however, that Haegeman (2003) claims that parallel cases are marginally possible in West Flemish (cf. [i]). 233

234 (i)

NOTES TO CHAPTERS

1–2

a. ?Nen boek no Gent stiert Valere niet a book to Gent sends Valere not b. ?Nen boek uit leest Valere nooit a book out reads Valere never

5. Pearson (2000) assumes that if the head of an XP is extracted, even non-agreeing features of the Speciﬁer of XP may percolate up to XP. Under these assumptions, remnant movement of the entire XP is possible again. Pearson argues that the trace of the head is featureless and may therefore not give rise to the projection of features up to the maximum XP. While this seems plausible in an approach where traces are left behind by movement, it seems counterintuitive in a copy theory of movement, where the antecedent and its “trace” are copies of each other. Note that if such an approach is adopted as in Haegeman’s (2001, 2002b) account of the SOV order in embedded clauses in German and Dutch, another explanation has to be given for the data in (41).

Chapter 2 1. The careful reader will have noticed that this assumption is in conﬂict with the observation I made earlier in (3), namely, that w-words resist scrambling. To avoid this conﬂict, we have to assume that scrambling of w-words is grammatical but subject to the following PF condition: A w-word must be spelled out in a stressed position (in a multiple question in German, the w-word must be stressed; otherwise it is interpreted as an indeﬁnite pronoun). Since scrambling moves a constituent into an unstressed position, scrambling of a w-word will only be licit if the w-word is not spelled out in the scrambling position. 2. A DP is speciﬁc if it denotes a member of a set of individuals introduced in the previous discourse. It has been pointed out that names and generic expressions can scramble even in the absence of a discourse antecedent. Thus, the feature [Familiarity] has been proposed that encompasses discourse-antecedence and membership in the common ground (cf. Corver and Delﬁtto [1997]). I will leave the empirical question open of whether one type of trigger of scrambling is to be characterized with the notion [Familiarity] rather than [Speciﬁcity]. 3. That there are two licensing positions for object clitics (one below and one above the subject) is a relatively conservative assumption. Instead, one could assume that there is only one position for licensing object clitics, which is above the subject, and that the subject itself has moved into a higher position in (18a). Since this alternative proposal is neutral with respect to the main argument defended in this section, I will not pursue this issue any further. 4. Whether all scrambling orders, including those with several adverbs present, can be derived in this manner is subject to empirical investigation. More speciﬁcally, it remains to be seen whether these clitic-licensing heads occupy ﬁxed positions in the tree or whether they can be introduced at various points in the course of the derivation. For how this latter idea can be implemented—albeit for the purposes of checking scopal properties—see section 2.6. 5. Example (38b) is perfect if the negative marker is interpreted as constituent negation (see [39] later). 6. The determiner kein has been analyzed as created by fusing a determiner with existential force with negation (see Kratzer [1995]). 7. An anonymous reviewer points out that these scopal features make no semantic contribution of their own and that it is rather questionable whether they solve the syntactic problem that led to their introduction, arguing that the assignment of scopal features in the course of the derivation is also an optional syntactic operation in the sense that any of [w], [n], or [i] can be assigned to any DP at any point of the derivation as long as the interface conditions end up being respected.

NOTES TO CHAPTER

3

235

Chapter 3 1. It should be noted that weak pronouns and anaphors in Dutch always cliticize to a position below the subject. In German, as we have seen earlier, they can also move to a position above the subject. It is interesting to note that this diﬀerence in pronoun placement between Dutch and German coincides with a diﬀerence in scrambling. Dutch does not allow scrambling across the subject. Unlike in German, a direct or indirect object cannot be scrambled across the subject (see chapter 2 for more discussion). However, the correlation breaks down in West Flemish. West Flemish does not allow scrambling across the subject, while weak pronouns and anaphors may appear in front of the subject. 2. Rutten does not list here helpen (‘to help’) and leren (‘to learn/teach’), which Broekhuis, Den Besten, and Rutten (1995) explicitly mention as verbs that select bare inﬁnitives. Instead, Rutten (1991) enlists them as verbs that select te-inﬁnitivals and may enter into a VR-structure or into the Third Construction, which I will discuss later. Hans Bennis (p.c.) informs me that the presence of the inﬁnitival marker is optional with these verbs. 3. Hans Bennis (p.c.) informs me that the aspectual verbs liggen, staan, and zitten as well as the semi-modals durven and hoeven require the inﬁnitival marker in present tense but obligatorily drop it in perfect tense. 4. Rutten (1991) actually assumes that VR is always obligatory, but that structures that result from VR are subject to a late inversion rule that applies after VR (possibly in the phonological component, since it has no LF-eﬀect) and inverts structures of the form [a b] just in case “a” is a ﬁnite modal verb. Such a PF-rule may then be assumed to be sensitive to the internal (morphological) structure of “b.” This PF-rule must then be taken to be triggered only by modal verbs, since the causative laten never triggers inversion. 5. It should be noted that there is also an alternative analysis of the alternation in (28) that assumes that PI is obligatory. Under this assumption, we may assume that either the complex head undergoes VR, yielding (28c), or only the verbal head undergoes VR by excorporating out of the complex head, yielding (28b). 6. The account that I will provide eventually is one where ge-preﬁx and inﬁnitives, on the one hand, and particles and their complements, namely gerunds (nominalized inﬁnitives), on the other hand, compete for the same licensing positions in the extended projections of the selecting verb (see chapter 4 and chapter 6 for the details). 7. The argument holds independently of whether we assume that Dutch is basically an OV or a VO language. What is important is the hierarchical order between particle, dependent inﬁnitive, and selecting verb. 8. I am grateful to the anonymous reviewer who pointed out to me that the ungrammaticality of (37b–c) can be derived by the long head movement account, but that (37d) cannot be ruled out without further stipulations. 9. The same reviewer also points out that (37d) would also seem to be a problem for the account that I am going to develop. I will come back to this example in the following chapter and show that it and similar cases cannot be derived in the XP-movement account to be given. 10. In Haegeman’s (1992) account, contrary to Den Besten and Rutten’s (1989) assumptions, scrambling has to apply before extraposition. Haegeman assumes that extraposition as right-adjunction makes the extraposed inﬁnitival complement a barrier for extraction. 11. In this respect WF clitics behave like German clitics rather than clitics in standard Dutch. Remember that clitics in Dutch move to a position below the subject. 12. Obviously, reconstructing the extraposed inﬁnitival into its base position and checking the ECP after reconstruction would solve this problem. However, it is then not clear why QR cannot aﬀect a scopal element within a reconstructed inﬁnitival complement.

236

NOTES TO CHAPTERS

3–4

13. One might argue that the CP could be extraposed, that is, right-adjoined to the XP, probably IP, headed by zeggen, before the inﬁnitival is extraposed to the clause headed by moeten. Then the CP, since it is not included in the extraposed inﬁnitival, could move on further and right-adjoin to the temporal auxiliary. However, there is a problem with this account as well. Note that the inﬁnitival headed by zeggen is an argument and that adjunction to arguments is impossible within the barriers framework. Furthermore, note that if this type of adjunction were allowed exceptionally, any QP contained in a VPR-complement that has been extraposed could move out of the extraposed clause at LF without crossing a single barrier. 14. Note that the pendant of (48b) in standard Dutch is grammatical (i). (i)

dat Jan dat boek heeft gelezen that Jan that book has read

15. Here the question arises of why an inﬁnitival clause does not have to move as well in order to check the subcategorization of its selecting verb. One might argue that checking in this case is necessary since a given verb may select a bare or a to-inﬁnitive. For the time being and for reasons of simplicity I will assume that inﬁnitives, as opposed to participial clauses, host a complementizer that can check the subcategorization of the selecting verb by undergoing head movement that adjoins the complementizer to the selecting verb at LF. 16. Presumably the participle, like IPP-complements, ﬁrst undergoes XP-movement into [Spec,VP] of the auxiliary and then undergoes additional head movement that must be caused by the inﬂectional properties of participles (maybe its aspectual morpheme needs to be linked with the local tense head). 17. I use the word reveal in this context because, as we will see later, adopting a VObased approach will force me to assume a lot of movements, movements that need to be justiﬁed within the Minimalist Program by deﬁning the properties that motivate them. Finding the properties that motivate movement is tantamount, if I interpret the spirit behind the minimalist framework correctly, to ﬁnding the properties that deﬁne these constructions. 18. In the original example—Haegeman (1994; [28a]), our (57a)—Haegeman did not indicate that is is actually strongly disfavored with respect to eet.

Chapter 4 1. For an analysis of verb clusters in Afrikaans see Robbers (1997). 2. The only thing left to the traditional SOV approach is assuming that in (4) the entire VP has been right-adjoined to IP, which hosts the inﬁnitival marker. However, remember that I concluded in the last section of chapter 3 that the distribution of inﬁnitives, participles, and IPP-complements in West Flemish cannot be properly explained by a single rule of extraposition. We saw that several conditions on extraposition would be needed. Accounting for (4a) in terms of extraposition would only increase the stipulative character of extraposition in West Flemish. 3. An anonymous reviewer tells me that the data in (5) are surprising, pointing out that in WF one would get sentences like (i). Note, however, that both nicely and quickly also allow for a higher subject-oriented or aspectual reading. Example (ib) in German has the interpretation that ‘I reacted quickly in putting on my other clothes (I was fast in starting to put on the other clothes).’ See also note 4 and Cinque (1999) for additional discussion. (i)

a. da-tje schuone zen soepe eet that he nicely his soup eats b. dan-k zeere men andere kleren andegen that I quickly my other clothes on put

NOTES TO CHAPTERS

4–5

237

4. Often, as in (5b), the order manner adverb < nominal argument yields a perfect sentence. This is always then the case when the manner adverb is eligible for an alternative interpretation. So, for instance, (5b) is perfect under the interpretation ‘it was careful of Hans to read the book’, where the adverb is interpreted as subject oriented rather than as pure manner adverb. Also, (6b) is perfect under the interpretation that ‘Hans executed exactly one/this plan’, where the adverb is construed as modifying the determiner. 5. An anonymous reviewer correctly points out that the postulated Agreement heads that attract DPs out of the VP do not have inherent Case properties of their own: Which case is licensed in a given position in this approach is largely determined by VP-internal properties and a fairly complex set of conditions that relate Case-heads to each other. According to this reviewer, this makes the Case-licensing Agr-heads look like an ad hoc device, especially given that arguments must move into Case-licensing positions preserving their hierarchical order, while it is not clear which mechanism within a minimalist grammar would enforce this parallel movement. 6. The same argument against Particle Incorporation is made in Den Besten and Broekhuis (1992), who reach the same conclusion as here, namely, that VR may not be analyzed as only involving head movement. That particles cannot be taken to incorporate into the verb and that verb clusters that contain particles can therefore not be analyzed as head-adjunction structures is also shown by the behavior of particles in multi-member verb clusters, as is discussed in chapter 3 (section 3.1.3). Recall that particles in Dutch can occupy various positions in the verb cluster (cf. Bennis [1992]). For instance, in (ii) the particle must have reached its surface position via XP-movement (head movement would violate the HMC), entailing that the containing structure cannot be a head-adjunction structure. (ii)

dat hij mij zou kunnen [weg]i horen ti rijden that he me would can away hear ride ‘that he would be able to hear me drive away’

7. Of course, we could assume that reconstruction targets an intermediate position. Such an intermediate position that is dominated by the matrix verb but itself c-commands all the material in the embedded clause and that is also typically targeted by A' moved elements would be [Spec,CP]. Note, however, that as soon as we make the CP-layer available, we lose the ability to distinguish between coherent and non-coherent inﬁnitival complements within the standard approach. 8. Dutch also has some aspectual verbs that select bare inﬁnitives that require a projective interpretation on part of the inﬁnitive. 9. It is interesting to note that the gerunds that can be used instead of the inﬁnitive with remember and try do have the same temporal readings as the inﬁnitives in (52) (cf. [iii] later). I thank an anonymous reviewer for pointing this out to me. See chapter 6 for more discussion on the relation between inﬁnitives and gerunds. (iii)

a. John remembered doing his work/having done his work b. John tried using a key to cut the paper

10. Because of similar transparency eﬀects in restructuring inﬁnitives and ﬁnite subjunctive clauses, I would like to propose that restructuring inﬁnitives are analyzed as [+subjunctive], [–ﬁnite] clauses.

Chapter 5 1. This preference may be just signiﬁcant for a strategy of keeping the two adverbials apart: In (1a), there is an intonational break between the two adjacent adverbials. 2. Two time references in a single clause are only possible if one can be taken to further specify the other as exempliﬁed in (ia) (cf. Brugger [1998] for a discussion of these issues).

238

NOTES TO CHAPTER

5

That this is a more general property that holds of adverbs that can set up the “frame” for an assertion is suggested by (ib). (i)

a. Yesterday he met her at two o’clock b. In the park, she was sitting on a bench

3. The ambiguity of (6a), probably, results from the fact that the participle morphology can be interpreted as a temporal or as an aspectual morpheme. In the former case, the participle will denote an event that occurs prior to a point of time speciﬁed by the matrix TP. In the latter case, the participle denotes a state that results from the completion of the event denoted by the verb and that is located in time by the matrix TP. 4. The principle in (13) is intended to translate the standard deﬁnition of the binding domain in the theory of Government and Binding (cf. Chomsky [1981]) into a system that lacks the notion of government. The standard deﬁnition had it that the binding domain of x is the minimal TP/DP that contains x, the governor of x, and (for anaphors) a SUBJECT accessible to x. Here we deﬁne the binding domain of x as the minimal TP/DP in which x is Caselicensed and (for anaphors) which contains a SUBJECT accessible to x. 5. An anonymous reviewer argues that because of this diﬀerence there is, strictly speaking, no argument that long passives (in coherent to-inﬁnitives) should be derived from biclausal structures. However, in section 5.5.1. I show that the availability of a long passive depends on the availability of an impersonal passive with a sentential inﬁnitival complement. 6. Certain tests like the attachment of intentional adverbs and availability of control into a ﬁnal clause imply that both little v and the Agent argument are present in passive sentences in German. As an illustration, consider the well-known examples in (ii). (ii)

a. Das Schiﬀ wurde absichtlich versenkt the boat was sunk deliberately b. *Das Schiﬀ sank absichtlich the boat sank deliberately c. Das Schiﬀ wurde versenkt um die Versicherungssumme zu kassieren the boat was sunk to collect the insurance fee d. *Das Schiﬀ sank um die Versicherungssumme zu kassieren the boat sank to collect the insurance fee

7. I cannot address here the important question of how Nominative assignment is sanctioned in this case. Note that Nominative Case assignment is normally only possible in tensed clauses. However, there is ample evidence that passive subjects are licensed in a low position, corresponding to the position of the direct object. This is indicated in particular by word order facts and topicalization data. Example (iiia), in which the Dative object precedes the Nominative subject, displays unmarked word order vis-à-vis (iiib). Example (iiic), in which the passive subject is topicalized with the selecting participle, is okay, while (iiid), in which the agentive subject is topicalized with the selecting participle, is completely ungrammatical. (iii)

a. weil dem Mann das Fahrrad gestohlen wurde since the man-Dat the bike-NOM stolen was b. ?weil das Fahrrad dem Mann gestohlen wurde since a bike:NOM the man-DAT stolen was c. ein Fahrrad gestohlen wurde ihm noch nie a bike-NOM stolen was him yet never d. *ein Mann gestohlen hat das Fahrrad noch nie a man-NOM stolen has the bike-AKK yet never

NOTES TO CHAPTERS

5–6

239

8. The German terms are in fact borrowed from Latin grammarians who also distinguished between the Gerundium and the inﬁnitive (cf [iv]). Thus, when I use the English term “gerund” as a shorthand expression for nominalized inﬁnitives I refer to the German tradition, rather than to the English/Romance tradition, where the same term is used to designate a nominalized participle. (iv) Cetero censeo Carthaginem delendam esse moreover I-believe Carthago-AKK destroy-GER be-INF ‘Moreover I believe that Carthago must be destroyed’ 9. The use of the term gerund to signify a nominalized inﬁnitive, though corresponding to the German tradition, may seem unfortunate, since gerund in the English tradition signiﬁes a category that is derived from the participle. However, these terms have in common that they stand for a nominalized category that is derived from a verbal form via phrasal aﬃxation.

Chapter 6 1. Remnant Extraposition and the (standard) VR-construction diﬀer with respect to whether they allow for the so-called long passive (cf. chapter 5 for a discussion of the long passive in to-inﬁnitives in German). Example (ia) is a long passive in a VR-construction. Example (ib–c) show that a long passive is not possible with cases of Remnant Extraposition. (i) a. weil sein neuester Roman zu lesen beschlossen wurde since his newest novel-NOM to read decided became ‘since it was decided to read his newest novel’ b. ??weil sein neuester Roman beschlossen wurde zu lesen since his newest novel-NOM decided became to read c. *weil sein neuester Roman beschlossen wurde nicht zu lesen since his newest novel-NOM decided became not to read 2. Note that in German even nonﬁnite auxiliaries have to invert with IPP-complements (ii), while the opposite holds in West Flemish (iii). This diﬀerence between West Flemish and German follows, if we assume, as I have argued in chapter 3, that in West Flemish only ﬁnite verbs can move to the head of AspP. (ii) a. Else wird ihm einen Brief haben schreiben wollen Else will him a letter have write want-IPP b. *Else wird ihm einen Brief schreiben wollen haben Else will him a letter write want-IPP have ‘Else will have wanted to write him a letter’ (iii)

a. dan-ze kosten willen dienen boek kuopen een that they could want-IPP that book buy have b. *dan-ze kosten een willen dienen boek kuopen that they could have want-IPP that book buy ‘that they could have wanted to buy that book’

3. Thus, it stands to reason whether we are dealing with two types of inﬁnitives here or the Nammefoarm should be analyzed as an inﬁnitive and the Doelfoarm as a gerund. Since gerunds have both verbal and nominal properties this category would ﬁt quite well with Tiersma’s description. Tiersma himself calls the Doelfoarm gerundive, a label that does not seem appropriate since it is traditionally used to denote verbal adjectives with modal force.

240

NOTES TO CHAPTERS

6–7

4. In a similar fashion to ECM-verbs in English that can combine with an inﬁnitive or a gerund. (iv) a. He saw her dance b. He saw her dancing 5. Speakers of IF also produce, modulo the morphological marking, the Standard Frisian order in which the dependent inﬁnitive precedes the selecting verb. 6. If nominals like verbs needed to be bound by Tense, then sentences like (v) should be contradictory, since someone who is ﬂeeing cannot be at the same time in prison. (v)

The fugitive will be set in prison again

7. The source of this variation described by Ebert may have been the availability of two types of inﬁnitives, the Doelform and the Nammeform (or due to the availability of the gerund and the inﬁnitive). If we assume that they originally had a diﬀerent distribution in the verb cluster, with the Nammeform being licensed in [Spec,F2P] and the Doelform being licensed in [Spec,AspP], diﬀerent orders would arise according to the selectional properties of the higher verb. Doelforms would yield the order V2 V1, whereas Nammeforms would yield the order V1 V2. When the morphological distinction between the two forms was lost, alternating patterns probably were reanalyzed as free variants subject only to prosodic conditions. More research on the diachronic development is necessary to evaluate this scenario.

Chapter 7 1. It should be noted, though, that for some German speakers, especially speakers of originally Franconian dialects, the participle is preferred over the inﬁnitive even in constructions like (11b). 2. However, a closer investigation of verb phrase preposing may reveal that this construction has diﬀerent properties in Dutch and German. For instance, for Zwart (1993 and p.c.) split topicalizations with to-inﬁnitives are fully acceptable. 3. That it is necessary to assume that a Case feature of the preposed verb is copied onto the d-word and reconstructed with it and so made available within the IP to license a remaining argument of this verb is not so straightforward, since, for instance, Zwart (1993) assumes that arguments of the dependent inﬁnitive are licensed in Agr-projections of the selecting verb. These Agr-projections are freely available with auxiliaries and VR-verbs. (cf. also the discussion on thematic restrictions later). 4. An anonymous reviewer points out that Cinque (1990) shows that fronting of predicates is signiﬁcantly facilitated by negation. I do not think that negation per se is at issue here. Rather, what seems to facilitate VP-fronting is stressing the polarity of the clause (cf. [i] later). The sentences in (ia–b) seem to be cases of verum focus that stress the truth value of the clause with respect to a presupposed constituent represented by the topicalized phrase. However, the questions of whether this is the correct analysis of (i) and whether there are other discourse conditions that motivate or facilitate VP-topicalization are not issues here. (i)

a. das Buch gelesen hat er nicht the book read has he not b. das Buch gelesen hat er wohl/doch the book read has he indeed

5. This movement of last resort into the left edge of VP can be seen as parallel to movement into the local [Spec,CP] in long-distance wh-movement. Assuming that only the matrix CP has a wh-feature to check, successive cyclic movement into intervening [Spec,CP] posi-

NOTES TO CHAPTER

7

241

tions cannot be due to feature checking and must thus be analyzed as a last-resort operation allowed to escape the PIC. 6. Alternatively, we may assume that gerunds always attach to the entire TP. To allow for split topicalization we would have to assume that gerunds are transparent for the extraction of arguments. Here I cannot decide between the two analyses and will leave the issue for further research. However, there is one piece of evidence that favors the account that a gerund head can attach to any verbal projection. In Dutch, not only arguments can be split oﬀ from the selecting verb but also adverbs. In (ii), due to the past tense of the matrix verb, the adverb tomorrow must be taken to modify the topicalized (extended) VP. Since adverbs cannot scramble, this is good evidence for a clausal split that is typical of coherent constructions. Therefore, in what follows I will go with the ﬁrst option, which is spelled out in more detail later. A special thanks goes to Hans Broekhuis for help with topicalization data in Dutch. (ii)

een boek lezen dat wou hij morgen a book read that wanted he tomorrow

7. I am grateful to the audience of my presentation at the ZAS in February 2005 for pointing this out to me. These speakers essentially have Dutch-like judgments.

This page intentionally left blank

REFERENCES

Alexiadou, A., and E. Anagnostopoulou (1997) “Toward a Uniform Account of Scrambling and Clitic Doubling,” in W. Abraham and E. van Gelderen (eds.), German: Syntactic Problems—Problematic Syntax. Max Niemeyer, Tübingen, 142–161. Baker, M. (1988) Incorporation. University of Chicago Press, Chicago. Baker, M., K. Johnson, and I. Roberts (1989) “Passive Arguments Raised.” Linguistic Inquiry 20, 219–251. Bech, G. (1955/1983) Studien über das Deutsche Verbum Inﬁnitum. Max Niemeyer, Tübingen. Beghelli, F., and T. Stowell (1996) “Distributivity and Negation,” in A. Szabolcsi (ed.), Ways of Scope Taking. Kluwer, Dordrecht, 71–107. Bennis, H. (1992) “Long Head Movement: The Position of Particles in the Verbal Cluster in Dutch,” in R. Bok-Bennema and R. van Hout (eds.), Linguistics in the Netherlands 1992. Benjamins, Amsterdam, 37–47. Bierwisch, M. (1990) “Verb Cluster Formation as a Morphological Process,” in J. Booij and J. van Marle (eds.), Yearbook of Morphology 3. Foris, Dordrecht, 173–200. Boscovic, Z. (1993) The Syntax of Non-Finite Complementation. MIT Press, Cambridge, Mass. Bowers, J. (1993) “The Syntax of Predication.” Linguistic Inquiry 24, 591–655. Broekhuis, H., and H. Den Besten (1989) “Woordvolgorde in de werkwoordlijke eindreeks.” GLOT 12, 79–137. Broekhuis, H., H. den Besten, K. Hoekstra, and J. Rutten (1995) “Inﬁnitival Complementation in Dutch: On Remnant Extraposition.” The Linguistic Review 12, 93–122. Brugger, G. (1998) “Time Points.” Talk given at the UCLA Workshop on Tense and Aspect, May 1998. Brugger, G., and C. Poletto (1995) “On Negation in German and Bavarian.” Rivista di Grammatica Generativa, 111–159. Büring, D., and K. Hartmann (1997) “Doing the Right Thing.” The Linguistic Review 14, 1–42. 243

244

REFERENCES

Burzio, L. (1986) Italian Syntax. Reidel, Dordrecht. Cecchetto, C. (2000) “Doubling structures and reconstruction.” Probus 12, 1, 1–34. Cecchetto, C. (2001) “Proper Binding Condition Eﬀects Are Phase Impenetrability Condition Eﬀects,” in K. Minjoo and U. Strauss (eds.), Proceedings of NELS 31. GLSA, University of Massachusetts, Amherst. Chierchia, G. (1986) “Length, Syllabiﬁcation and the Phonological Cycle in Italian.” Journal of Italian Linguistics 8, 5–34. Chomsky, N. (1981) Lectures on Government and Binding. Foris, Dordrecht, 137–155. Chomsky, N. (1993) “A Minimalist Program for Linguistic Theory,” in K. Hale and J. Keyser (eds.), The View from Building 20. MIT Press, Cambridge, Mass., 1–52. Chomsky, N. (1995a) “Bare Phrase Structure,” in G. Webelhuth (ed.), Government and Binding Theory and the Minimalist Program. Blackwell, Cambridge, Mass., 383–483. Chomsky, N. (1995b) The Minimalist Program. MIT Press, Cambridge, Mass. Chomsky, N. (1998) “Minimalist Explorations.” Ms. MIT, Cambridge, Mass. Chomsky, N. (2001) “Derivation by Phase,” in M. Kenstowicz (ed.), Ken Hale: A Life in Language. MIT Press, Cambridge, Mass., 1–52. Cinque, G. (1990) Types of A'-dependencies. MIT Press, Cambridge, Mass. Cinque, G. (1993) “A Null Theory of Phrase and Compound Stress.” Linguistic Inquiry 24, 2, 239–298. Cinque, G. (1999) Adverbs and Functional Heads: A Crosslinguistic Perspective. Oxford University Press, Oxford. Cinque, G. (2001) “Restructuring and the Order of Aspectual and Root Modal Heads,” in G. Cinque and G. Salvi (eds.), Current Studies in Italian Syntax: Essays Oﬀered to Lorenzo Renzi. Elsevier, Oxford. 137–157. Corver, N., and D. Delﬁtto (1997) “Feature Primitives and the Syntax of Speciﬁcity.” GLOW Newsletter 34, 18–19. Corver, N., and H. van Riemsdijk (eds.) (1994) Studies on Scrambling: Movement and NonMovement Approaches to Free Word-Order Phenomena. Mouton de Gruyter, Berlin. De Haan, G. J. (1992) “The Verbal Complex in Frisian.” Us Wurk 41, 59–92. De Haan, G. J. (1996) “Recent Changes in the Verbal Complex of Frisian,” in A. Petersen and H. F. Nielsen (eds.), A Frisian and Germanic Miscellany. Odense University Press, Bredsted, 171–184. De Hoop, H. (1992) Case Conﬁguration and Noun Phrase Interpretation. Doctoral Dissertation, University of Groningen. Published 1996 by Garland Press, New York. Demske, U. (2001) “Zur Distribution von Inﬁnitivkomplementen im Althochdeutschen.” Linguistische Berichte 9, 61–86. Den Besten, H. (1977/1983) “On the Interaction of Root Transformation and Lexical Deletive Rules,” in W. Abraham (ed.), On the Formal Syntax of the West Germania. Benjamins, Amsterdam, 47–131. Den Besten, H., and H. Broekhuis (1992) “Verb Projection Raising in het Nederlands.” Spektator 21, 21–34. Den Besten, H., and J. Edmondson (1983) “The Verbal Complex in Continental West Germanic,” in W. Abraham (ed.), On the Formal Syntax of the West Germania. Benjamins, Amsterdam, 155–216. Den Besten, H., and J. Rutten (1989) “On Verb Raising, Extraposition and Free Word Order in Dutch,” in D. Jaspers (ed.), Sentential Complementation and the Lexicon. Linguisitic Models 13. Foris, Dordrecht, 41–56. Den Besten, H., and G. Webelhuth (1987) “Remnant Topicalization and the Constituent Structure of the VP in the Germanic SOV Languages (Abstract).” GLOW Newsletter 18, 15–16.

REFERENCES

245

Den Dikken, M. (1994) “Minimalist Verb Projection Raising,” in J.-W. Zwart and W. Abraham (eds.), Groninger Arbeiten zur Germanistischen Linguistik Groninger. 37, 71–88. Den Dikken, M. (1995) Particles: On the Syntax of Verb-Particle, Triadic and Causative Constructions. Oxford University Press, Oxford. Diesing, M. (1992) Indeﬁnites. Linguistic Inquiry Monograph 20. MIT Press, Cambridge, Mass. Dobrovie-Sorin, C. (1990) “Clitic Doubling, Wh-movement and Quantiﬁcation in Romanian.” Linguistic Inquiry 21, 351–397. Donaldson, B. (1993) A Grammar of Afrikaans. Mouton de Gruyter, Berlin. Ebert, R. P. (1980) “Social and Stylistic Variation in Early New High German Word Order.” Beiträge zur Geschichte der deutschen Sprache 102, 357–398. Enç, M. (1986) “Towards a Referential Analysis of Temporal Expressions.” Linguistics and Philosophy 9, 405–426. Enç, M. (1987) “Anchoring Conditions for Tense.” Linguistic Inquiry 18, 4, 633–657. Enç, M. (1991) “The Semantics of Speciﬁcity.” Linguistic Inquiry 22, 1–25. Evers, A. (1975) The Transformational Cycle in Dutch and German. Unpublished doctoral dissertation. Utrecht University. Evers, A. (1994) “A Note on Particle Stress.” TABU 47, 14–40. Fanselow, G. (1989) “Coherent Inﬁnitives in German: Restructuring vs. IP-Complementation,” in C. Bhatt, E. Löbel, and C. Schmidt (eds.), Syntactic Phrase Structure Phenomena. Benjamins, Amsterdam, 1–16. Fanselow, G. (1993) “The Return of the Base Generators.” Groninger Arbeiten zur Germanistischen Linguistik 36, 1–74. Fanselow, G. (2001) “Features, ϑ-roles, and Free Constituent Order.” Linguistic Inquiry 32, 3, 405–437. Frascarelli, M. (2000) The Syntax-Phonology Interface in Focus and Topic Constructions in Italian. Kluwer, Dordrecht. Frey, W. (1993) Syntaktische Bedingungen für die Interpretation. Studia Grammatika 35. Akademie Verlag, Berlin. Frey, W. (2000) “Über die syntaktische Position der Satztopiks im Deutschen.” ZAS Papers in Linguistics 20, 137–172. Frey, W., and K. Pittner (1998) “Zur Positionierung der Adjunkte im deutschen Mittelfeld.” Linguistische Berichte 176, 489–534. Frey, W., and K. Pittner (1999) “Adverbialpositionen im deutsch-englischen Vergleich,” in M. Doherty (ed.), Sprachspeziﬁsche Aspekte der Informationsverteilung. Akademie Verlag, Berlin, 14–40. Fukui, N. (1997) “Attract and the A-over-A Principle.” UCI Working Papers in Linguistics 3, 51–67. Gamon, M. (1992) “Lexical Selection of Temporal Properties in Inﬁnitival Complements.” University of Washington Working Papers, 43–75. Gärtner, H. M., and M. Steinbach (2000) “What Do Reduced Pronominals Reveal about the Syntax of Dutch and German?” Linguistics in Potsdam 9, 7–62. Geilfuß, J. (1991) “Verb- und Verbphrasensyntax.” Arbeitspapiere des SFB 340, 11. Grewendorf, G. (1987) “Kohärenz und Restrukturierung,” in B. Asbach-Schnitker and J. Roggenhofer (eds.), Neuere Forschungen zur Wortbildung und Histographie der Linguistik. Narr, Tübingen, 122–144. Grewendorf, G. (1988) Aspekte der deutschen Syntax: Eine Rektions- und Bindungsanalyse. Narr, Tübingen. Grewendorf, G. (2002) “Left Dislocation as Movement,” in S. Mauck and J. Mittelstaedt (eds.), Georgetown University Working Papers in Theoretical Linguistics 2, 31–81.

246

REFERENCES

Grewendorf, G., and J. Sabel (1999) “Scrambling in German and Japanese.” Natural Language and Linguistic Theory 17, 1–65. Grewendorf, G., and W. Sternefeld (eds.) (1990) Scrambling and Barriers. Benjamins, Amsterdam. Grimm, J. (1899/1969) Deutsche Grammatik. Gütersloh, Göttingen. Guéron, J., and T. Hoekstra (1988) “T-chains and the Constituent Structure of Auxiliaries,” in A. Cardinaletti, G. Cinque, and G. Giusti (eds.), Constituent Structure. Foris, Dordrecht, 35–99. Haeberli, E. (2002) Features, Categories and the Syntax of A-positions. Kluwer, Dordrecht. Haegeman, L. (1992) Theory and Description in Generative Syntax: A Case Study in West Flemish. Cambridge University Press, Cambridge. Haegeman, L. (1994) “Verb Raising as Verb Projection Raising: Some Empirical Problems.” Linguistic Inquiry 25, 509–522. Haegeman, L. (1995a) “IPP Constructions and Finite V-movement in West Flemish.” GenGenP 6, 29–49. Haegeman, L. (1995b) The Syntax of Negation. Cambridge University Press, Cambridge. Haegeman, L. (1996a) “Object Clitics in West Flemish,” in A. Halpern and A. Zwicky (eds.), Second Position Clitics and Related Phenomena. CSLI Press, Stanford, 135–164. Haegeman, L. (1996b) “The Typology of Syntactic Positions: L-Relatedness and the A/A'Distinction,” in W. Abraham, S. D. Epstein, H. Thrainsson, and J.-W. Zwart. (eds.), Minimal Ideas. Benjamins, Amsterdam, 141–165. Haegeman, L. (1996c) “Verb Second, the Split CP and Initial Null Subjects in Early Dutch ﬁnite clauses.” GenGenP 2, 133–175. Haegeman, L. (2001) “Remnant Movement and OV Order,” in P. Svenonius (ed.), The Derivation of OV and VO. Benjamins, Amsterdam, 69–96. Haegeman, L. (2002a) “Anchoring to Speaker, Adverbial Clauses and the Structure of CP,” in S. Mauck and J. Mittelstaedt (eds.), Georgetown University Working Papers in Theoretical Linguistics 2, 117–180. Haegeman, L (2002b) “Antisymmetry and Verb-Final Order in West Flemish.” The Journal of Comparative Germanic Linguistics 3, 207–232. Haegeman, L. (2003) “West Flemish Negation and the Derivation of SOV Order in West Germanic.” Nordic Journal of Linguistics 25, 154–189. Haegeman, L., and H. van Riemsdijk (1986) “Verb Projection Raising, Scope, and the Typology of Rules Aﬀecting Verbs.” Linguistic Inquiry 17, 417–466. Haegeman, L., and R. Zanuttini (1991) “Negative Heads and the Neg Criterion.” The Linguistic Review 8, 233–251. Haiden, M. (1995) “On the Semantic Evaluation of NP-Syntax.” Diplomarbeit. University of Vienna. Haider, H. (1986) “Fehlende Argumente: Vom Passiv zu kohärenten Inﬁnitiven.” Linguistische Berichte 101, 3–33. Haider, H. (1987) “Nicht-sententiale Inﬁnitive.” Groninger Arbeiten zur Germanistischen Linguistik 28, 73–114. Haider, H. (1990) “Topicalization and Other Puzzles of German Syntax,” in G. Grewendorf and W. Sternefeld (eds.), Scrambling and Barriers. Benjamins, Amsterdam, 93–112. Haider, H. (1991) “Fakultativ kohärente Inﬁnitkonstruktionen im Deutschen.“ Arbeitspapiere des SFB 340, 17. Haider, H. (1993) Deutsche Syntax, generativ: Vorstudien zur Theorie einer projektiven Grammatik. Narr, Tübingen. Haider, H. (1994) “Scrambling—Economy and Directionality.” Talk given at 1994 Tokyo Scrambling Festival, University of Tokyo.

REFERENCES

247

Haider, H., and I. Rosengren (1998) “Scrambling.” Sprache and Pragmatik 49, 1–104. Hale, K., and J. Keyser (1993) “On Argument Structure and the Lexical Expression of Syntactic Relations,” in K. Hale and J. Keyser (eds.), The View from Building 20. MIT Press, Cambridge, Mass., 53–110. Halle, M., and A. Marantz. (1993) “Distributed Morphology and the Pieces of Inﬂection,” in K. Hale and J. Keyser (eds.), The View from Building 20. MIT Press, Cambridge, Mass., 111–176. Hallman, P. (1997) “Reiterative Syntax,” in J. Black and V. Motopanyane (eds.), Clitics, Pronouns and Movement. Benjamins, Amsterdam, 87–131. Hayes, B., and A. Lahiri (1991) “Bengali Intonational Phonology.” Natural Language and Linguistic Theory 9, 47–96. Hinterhölzl, R. (1995) “The Passive Implicit Argument and the Impersonal Pronoun Man in German,” in V. Samiian and J. Schaeﬀer (eds.), Proceedings of WECOL XXIV. CSLI, Stanford, Calif., 261–278. Hinterhölzl, R. (1997) “Inﬁnitival Tense, Verb Raising and Verb Projection Raising,” in K. Kusumoto (ed.), NELS 27. GLSA, Amherst, Mass., 187–201. Hinterhölzl, R. (1999) Restructuring Inﬁnitives and the Theory of Complementation. Unpublished doctoral dissertation. University of Southern California. Hinterhölzl, R (2000) “Licensing Movement and Stranding in the West Germanic OV Languages,” in P. Svenonius (ed.), The Derivation of OV and VO. Benjamins, Amsterdam, 293–326. Hinterhölzl, R. (2002a) “Parametric Variation and Scrambling in English,” in W. Abraham and J-W. Zwart (eds.), Studies in Comparative Germanic Syntax. Benjamins, Amsterdam, 131–151. Hinterhölzl, R (2002b) “Remnant Movement and Partial Deletion,” in A. Alexiadou, E. Anagnostopoulou, S. Barbiers, and H-M. Gaertner (eds.), Dimensions of Movement. Benjamins, Amsterdam, 126–150. Hinterhölzl, R. (2004) “Scrambling, Optionality and Non-Lexical Triggers,” in A. Breitbarth and H. van Riemsdijk (eds.), Triggers. Mouton de Gruyter, Berlin, 173–203. Höhle, T. N. (1978) Lexikalistische Syntax: Die Aktiv-Passiv-Relation und andere Inﬁnitivkonstruktionen im Deutschen. Niemeyer, Tübingen. Höhle, T. N. (1982) “Explikationen für normale Betonung und für normale Wortstellung,” in W. Abraham (ed.), Satzglieder im Deutschen. Narr, Tübingen, 75–153. Höhle, T. N. (1989) “Reanalyse—Der Verbalkomplex in germanischen and romanischen Sprachen: Deskriptive und explanative Deutung.” SFB: Sprachtheoretische Grundlagen für die Computerlinguistik, 93–117. Holmberg, A. (1999) “Remarks on Holmberg’s Generalization.” Studia Linguistica 53, 1–39. Hornstein, N. (1990) As Time Goes By. MIT Press, Cambridge, Mass. Hroarsdottir, Th. (2001) Word Order Change in Icelandic: From OV to VO. Benjamins, Amsterdam. Jacobs, J. (1997) “I-Topikalisierung.” Linguistische Berichte 168, 91–133. Kayne, R. (1994) The Antisymmetry of Syntax. Linguistic Inquiry Monograph 25. MIT Press, Cambridge, Mass. Kayne, R. (1998) “Overt versus Covert Movement.” Syntax 1, 128–191. Kayne, R. (1999) “Prepositional Complementizers as Attractors.” Probus 11, 39–73. Koopman, H. (1995) “The Position of Incorporated Heads in the Verbal Complex.” TABU 25, 4, 174–178. Koopman, H., and A. Szabolcsi (2000) Verbal Complexes. MIT Press, Cambridge, Mass. Koster, J. (1975) “Dutch as an SOV Language.” Linguistic Analysis 1, 111–136.

248

REFERENCES

Koster, J. (1995) “Predicate Incorporation and Word Order in Dutch,” in G. Cinque, J. Koster, J.-Y., Pollock, L. Rizzi, and R. Zanuttini (eds.), Paths Towards Universal Grammar. Georgetown University Press, Washington D.C., 255–277. Kratzer, A. (1995) “Stage-level and Individual-level Predicates,” in G. Carlson and F. J. Pelletier (eds.), The Generic Book. University of Chicago Press, Chicago, 125–175. Kratzer, A. (1996) “Severing the External Argument from Its Verb,” in J. Rooryck and L. Zaring (eds.), Phrase Structure and the Lexicon. Kluwer, Dordrecht, 109–137. Kroch, A., and B. Santorini (1990) “Remnant Extraposition in German.” Ms. University of Pennsylvania, Philadelphia. Kroch, A., and B. Santorini (1991) “The Derived Structure of the Germanic Verb Raising Construction,” in R. Freidin (ed.), Principles and Parameters in Comparative Grammar. MIT Press, Cambridge, Mass., 269–338. Larson, R. K. (1988) “On the Double Object Construction.” Linguistic Inquiry 19, 335–391. Lattewitz, K. (1997) Adjacency in Dutch and German. Groningen Dissertations in Linguistics 19. University of Groningen. Lechner, W. (1998) “Two Kinds of Reconstruction.” Studia Linguistica 52, 3, 276–310. Lenerz, J. (1977) Zur Abfolge nominaler Satzglieder im Deutschen. Studien zur deutschen Grammatik 5. Narr, Tübingen. Lötscher, A. (1978) “Zur Verbstellung im Zürichdeutschen und in anderen Varianten des Deutschen.” Zeitschrift für Dialektologie und Linguistik 45, 1–29. Mahajan, A. (1990) The A/A-bar Distinction and Movement Theory. Unpublished doctoral dissertation. MIT. Martin, R. (2001) “Null Case and the Distribution of PRO.” Linguistic Inquiry 32, 141–166. Mauck, S and J. Mittelstaedt (eds.) (2002) Georgetown University Working Papers in Theoretical Linguistics, 2. Georgetown University Press, Washington, D.C. Meinunger, A. (2000) Syntactic Aspects of Topic and Comment. Benjamins, Amsterdam. Moro, A. (1997) The Raising of Predicates. Cambridge University Press, Cambridge. Müller, G. (1996) “A Constraint on Remnant Movement.” Natural Language and Linguistic Theory 14, 355–407. Müller, G. (1998) Incomplete Category Fronting: A Derivational Approach to Remnant Movement in German. Kluwer, Dordrecht. Napoli, D. (1981) “Semantic Interpretation vs. Lexical Governance: Clitic Climbing in Italian.” Language 57, 841–887. Neeleman, A. D. (1994a): Complex Predicates. Unpublished doctoral dissertation. Utrecht University. Neeleman, A. D. (1994b) “Scrambling as a D-Structure Phenomenon,” in N. Corver and H. van Riemsdijk (eds.), Studies on Scrambling: Movement and Non-Movement Approaches to Free Word-Order Phenomena. Mouton de Gruyter, Berlin, 387–429. Olsen, S. (1981) Problems of Seem/Scheinen Constructions and their Implications for the Theory of Predicate Sentential Complementation. Linguistische Arbeiten 96. Niemeyer, Tübingen. Pearson, M. (2000) “Two Types of VO Languages,” in P. Svenonius (ed.), The Derivation of OV and VO. Benjamins, Amsterdam, 327–363. Pesetsky, D. (1992) “Zero Syntax II: An essay on Inﬁnitives.” Ms. MIT, Cambridge, Mass. Pesetsky, D. (2000) Phrasal Movement and Its Kin. MIT Press, Cambridge, Mass. Reichenbach, H. (1947) Elements of Symbolic Logic. Macmillan, London. Reis, M. (1976) “Reﬂexivierung in deutschen A.C.I.-Konstruktionen: Ein transformationsgrammatisches Dilemma.” Papiere zur Linguistik 9, 5–82. Riemsdijk, H. van (1978) A Case Study in Syntactic Markedness. Foris, Dordrecht. Riemsdijk, H. van, and E. Williams (1981) “NP-structure.” The Linguistic Review 1, 171–217.

REFERENCES

249

Rizzi, L. (1978) “A Restructuring Rule in Italian Syntax,” in S. J. Keyser (ed.), Recent Transformational Studies in European Languages. MIT, Cambridge, Mass. 113–158. Rizzi, L. (1982) Issues in Italian Syntax. Foris, Dordrecht. Rizzi, L. (1990) Relativized Minimality. MIT Press, Cambridge, Mass. Rizzi, L. (1997) “The Fine Structure of the Left Periphery,” in L. Haegeman (ed.), Elements of Grammar. Kluwer, Dordrecht, 289–330. Robbers, K. (1997) Non-ﬁnite Inﬁnitival Complements in Afrikaans: A Comparative Approach. Holland Academic Graphics, Den Haag. Roberts, I. (1985) The Representation of Implicit and Dethematized Subjects. Ph.D. dissertation, University of Southern California. Published 1987 by Foris, Dordrecht. Roberts, I. (1997) “Restructuring, Head Movement and Locality.” Linguistic Inquiry 48, 423– 460. Rochette, A. (1988) Semantic and Syntactic Aspects of Romance Sentential Complementation. Unpublished doctoral dissertation. MIT. Rosen, S. T. (1989) Argument Structure and Complex Predicates. Unpublished doctoral dissertation. Brandeis University. Rosen, S. T. (1990) “Restructuring Verbs Are Light Verbs.” Proceedings of WCCFL 9, 477– 491. Rosengren, I. (1992) “Zum Problem der kohärenten Verben im Deutschen,” in P. Suchsland (ed.), Biologische and soziale Grundlagen der Sprachfähigkeit. Niemeyer, Tübingen, 265–297. Ross, J. R. (1967) Constraints on Variables in Syntax. Unpublished doctoral dissertation. MIT. Roussou, A. (2000) “On the Left Periphery: Modal Particles and Complementizers.” Journal of Greek Linguistics 1, 65–94. Rutten, J. (1991) Inﬁnitival Complements and Auxiliaries. Unpublished doctoral dissertation. University of Amsterdam. Sabel, Joachim (1996) Restrukturierung und Lokalität. Akademie Verlag, Berlin. Sportiche, D. (1996) “Clitic Projections,” in J. Rooryck, and L. Zaring (eds.), Phrase Structure and the Lexicon. Kluwer, Dordrecht, 213–276. Starke, M., and A. Cardinaletti (1999) “The Typology of Structural Deﬁciency,” in H. van Riemsdijk (ed.), Clitics in the Languages of Europe. Mouton de Gruyter, Berlin. 145– 233. Stechow, A. V. (1990) “Status Government and Coherence in German,” in G. Grewendorf and W. Sternefeld (eds.), Scrambling and Barriers. Benjamins, Amsterdam, 143–198. Sternefeld, W. (1989) “Beheaded Barriers.” Arbeitspapier Fachgruppe Sprachwissenschaft, Nr. 14, Universität Konstanz. Stowell, T. (1981) Elements of Phrase Structure. Unpublished doctoral dissertations, MIT. Stowell, T. (1982) “The Tense of Inﬁnitives.” Linguistic Inquiry 13, 561–570. Stowell, T. (1993) “The Phrase Structure of Tense,” in J. Rooryck and L. Zaring (eds.), Phrase Structure and the Lexicon. Kluwer, Dordrecht, 271–286. Suñer, Margharita (1988) “The Role of AGR(eement) in Clitic Doubled Constructions.” Natural Language and Linguistic Theory 6, 391–434. Takano, Y. (2000) “Illicit Remnant Movement: An Argument for Feature-Driven Movement.” Linguistic Inquiry 31, 141–156. Tiersma, P. M. (1985) Frisian Reference Grammar. Foris, Dordrecht. Vikner, Sten (1994) “Scandinavian Object Shift and West Germanic Scrambling,” in N. Corver and H. van Riemsdijk (eds.), Studies on Scrambling: Movement and Non-movement Approaches to Free Word-Order Phenomena. Mouton de Gruyter, Berlin, 487–517. Vikner, S. (1995) “V-to-I Movement and Inﬂection for Person in All Tenses.” Working Papers in Scandinavian Syntax 55, 1–27.

250

REFERENCES

Webelhuth, G. (1985) “German Is Conﬁgurational.” The Linguistic Review 4, 203–246. Williams, E. (1984) “Grammatical Relations.” Linguistic Inquiry 15, 639–673. Williams, E. (2003) Representation Theory. MIT Press, Cambridge, Mass. Wolf, H. (1997) “Default-morphological Forms in Inverted Verbal Complexes in Frisian.” Groninger Arbeiten zur Germanistischen Linguistik 40, 249–256. Wurmbrand, S. (2001) Inﬁnitives: Restructuring and Clause Structure. Mouton de Gruyter, Berlin. Wurmbrand, S. (2004) “Two Types of Restructuring—Lexical versus Functional.” Lingua 114, 991–1015. Wyngaerd, G. vanden (1994) “IPP and the Structure of Participles.” Groninger Arbeiten zur Germanistischen Linguistik 37, 265–276. Wyngaerd, G. vanden (1996) “Participles and Bare Argument Structure,” in W. Abraham, S. D. Epstein, H. Thrainsson, and J.-W. Zwart (eds.), Minimal Ideas. Benjamins, Amsterdam, 347–363. Zaenen, A., J. Maling, and H. Thrainsson (1985) “Case and Grammatical Functions: The Icelandic Passive.” Natural Language and Linguistic Theory 3, 441–483. Zagona, K. (1990) “Times as Temporal Argument Structure.” Ms. University of Washington. Zwart, J.-W. (1993) Dutch Syntax. Unpublished doctoral Dissertation. University of Groningen.

INDEX

Adjacency, 203, 208, 210, 231 Adjunction, 10, 21, 39, 42, 45, 57, 60, 63, 67, 73, 78, 81–83, 88, 98, 144, 155, 187–188, 214–215 head-adjunction, 21, 67, 73, 78, 88, 103–104, 163 right-adjunction, 10, 83, 98, 110 Agreement position, 92 A'-movement, 8, 33, 35, 37–41, 43, 109, 215 Quantifier Raising (QR), 43, 81, 83, 113– 114, 206 wh-movement, 23–25, 40, 43 Antisymmetry, 12–13, 85, 90, 92, 154, 167, 188–189, 216, 229 A-position. See Argument position A'-position, 41 Argument position, 144, 150 Aspect Phrase (AspP), 91, 119, 123–124, 127, 133, 141, 149, 152, 157, 160, 162, 170, 177–179, 181–182, 184, 186–189, 191, 199, 201, 207–209, 216, 218–219, 225– 226, 230–231 AspP-movement, 108, 111, 114–115, 124, 163, 218, 220 Attract Closest, 3, 22–26, 110, 116, 214 and parallel movement, 94–95 and the strict cycle condition, 94 Base generation, 8, 12, 21, 32, 35, 44, 67, 106, 137, 144, 193–196, 198 Basic Branching Condition, 45

Binding theory, 29, 134, 150, 222 Bracketing paradox, 210–211 Case checking, 194 Case position, 31, 58, 135, 149–150, 223. See also Agreement position Case-Resistance Principle, 188 C-command, 22, 36, 38, 45, 47, 50, 58, 150, 154, 188–189, 195, 206, 214, 227 Clitic pronoun, 43, 136. See also Weak pronoun Coherent infinitive, 14–21, 88–89, 100–105, 107–109, 112, 116, 185, 203, 207, 210, 220–223, 227, 231 in Dutch 64–65 in Frisian, 171–175 in German 129–132, 136, 141, 144, 147, 149–150, 154, 158 in West Flemish 78–79 see also Restructuring infinitive Copy theory 23, 30, 33, 43, 56–57, 108, 112, 161–162, 177–178, 194–195, 201–204, 206, 215, 228 Cyclic Attraction, 95 Doelfoarm, 172–175, 208, 212, 228. See also Gerund D-word, 194–197, 203–207 left dislocation, 198, 200, 202 ECM, 62, 65, 102, 120–121, 130–131, 133– 138, 144–148, 150–151, 156, 221–223

251

252

INDEX

Economy principle, 29–30, 44, 151 fewest steps, 57–58 last resort, 30, 186, 199, 201, 203–209, 228 and optional movement, 44, 74, 214 shortest move, 30 Empty Category Principle (ECP), 81–82 EPP-feature, 55, 125, 149 Extension Condition, 22 Extraposition, 10, 13, 15, 17–18, 29, 32, 64–65, 68–70, 78–86, 88, 111–114, 132, 138, 146, 165, 185, 187–189 in the antisymmetric VO account, 214, 220, 227 and CP-complements, 119, 127, 186–190 and the Phase Impenetrability Condition, 189 in the standard OV account, 101, 187 and verb clusters, 186–190, 227–228 Feature checking, 23, 25, 51, 54–57, 99, 214, 231 and scope, 51, 56–57 (see also Relational features) Focus, 26, 29, 34–35, 41, 53, 56, 60–63, 83, 92, 106, 108, 116, 119 contrastive, 49, 54–55, 59, 96, 108, 217 scrambling, 35 Freezing Effect, 22, 27–28 Fusion, 210 Gerund, 157–159, 174–175, 180, 185–186, 199, 202–209, 223, 227–228 as clausal affix, 210–211 and the infinitival marker, 211 as last resort, 186, 204, 207, 209 selected, 208–209 (see also Doelfoarm) in to-infinitives, 208–209, 212 Gerundium. See Gerund Head complement order, 89 Head movement, 10, 20, 22, 25–26, 42, 67, 73– 79, 82, 123, 126, 154–155, 209, 225 in verb clusters, 162–165 Head Movement Constraint (HMC), 67, 77, 82 Inclusiveness Condition, 57 Incorporation, 21, 64, 77–78, 96, 104, 110 of particle 73–75, 102 Infinitival marker, 65, 86–87, 96–99, 103–104, 107, 147, 157–158, 164, 166, 170–175, 193, 208, 210–211, 215, 223, 229–230 in IPP-constructions, 90–91 as a sentence final head, 90 as verbal affix, 90 Interface condition, 33, 49–50, 56, 58, 63 Interference Frisian, 173–175 IPP-complement, 64, 78–79, 82–88, 137, 145, 160–161, 166–167, 170, 217–218, 225 external syntax, 166–167 hidden participle hypothesis, 85, 166 internal syntax, 167–171

and negative particle, 85 and right-branching verb clusters, 170 and scope ambiguities, 80–81, 114–115 in te-infinitives, 84 in West Flemish, 83–88 IPP-effect, 15, 18–20, 70–72, 77, 81–82, 108, 112, 145, 161, 165–168, 180, 191–192, 194, 198–199, 203, 206–207, 212, 216, 224–225, 228 and infinitives in Frisian, 171–175 IPP-infinitive, 18, 87, 145, 165–171, 173, 176, 178, 180–183, 194, 202–204, 224–226, 228 as suffixless participle, 170 I-topicalization, 35 Last Resort, 30, 186, 199, 201, 203–209, 228 Licensing movement, 12–13, 22, 26, 28, 99– 101, 108, 110, 115, 125, 127–128, 148, 181, 209, 213, 216–220, 222, 224, 229, 231 and Case, 92, 94–95, 118, 222, 230 and functional licensing, 100 of nominal arguments, 91–95 of predicates, 95–97 of sentential complements, 98–101, 118 of the TP, 189, 227 of the VP/AspP, 189, 208, 227 out of the VP, 23, 26–28, 91–101, 116, 214 Linear Correspondance Axiom (LCA), 104 L-marking, 81 Long Distance Scrambling (LDS), 5, 15–16, 18, 36, 41, 64, 83, 88, 104, 107, 109, 148, 216 Long Passive, 130, 137, 142, 144–145, 147– 148, 152–153, 222 L-relatedness, 34, 41 Mapping Hypothesis, 11 Minimalist Program (MP), 28–30, 32, 99, 199 Minimality, 30, 33–34, 39, 61–62, 75, 82, 136, 147 Minimal Link Condition (MLC), 33–34, 62 Mood Phrase (MP), 118, 124–127 Morphological Form (MF), 124, 169, 210–211 Move Alpha, 29, 60, 188 Nammefoarm, 172–175 and the IPP-effect, 172 see also Verbal infinitive Negation, 28, 36, 47–49, 51, 55, 59, 62, 98– 100, 133, 137–141, 146–147, 197, 200, 206–207, 221 constituent negation, 52–54, 138–139 and contrastive interpretation, 55 negative concord (NC), 52, 197–198, 200, 206 the negative determiner ‘kein’, 52–53, 200, 206 the negative marker, 17, 34, 51–55, 63, 86, 92, 96–97, 106, 126, 140

INDEX

sentential negation, 53–54, 79–80, 85, 92, 108, 200, 217 Nominal infinitive, 156, 174–175, 210, 212, 225. See also Gerund Numeration, 57, 61, 99, 199 Opacity, 114 Operator, 16–18, 29, 66, 144, 151, 174, 206– 207 operator feature, 41 operator movement, 18 Parametric Variation, 11, 89 Parasitic Gap, 37 Participle, 4, 12, 67–68, 73, 75–76, 78, 83–88, 124, 127, 145, 152, 161, 166–171, 180, 182–183, 192, 194–195, 202–204 and IPP, 168 phrase, 85, 167, 170, 224–225 prefix ge-, 76–77, 167, 171 prefixless, 166, 168–169 Particle. See verb particle Passive, the, 37, 100, 145, 148, 152–153, 158, 194–195 Phase, 25, 29–30, 94–95, 127, 177–178, 183, 186, 189, 191, 207, 214, 216, 225, 231 Phase Impenetrability Condition (PIC), 24–25, 177, 185, 189, 201, 214, 226–228 Phonetic Form (PF), 29, 33–35, 56, 68, 88, 172, 215 Phonological Phrase, 56, 176–177, 226 Pied-piping, 26, 98, 101–104, 107, 111, 114– 115, 123, 125, 148–149, 162–165, 170– 171, 187, 189–190, 193, 209, 216, 218, 223, 227, 230 Predicate Phrase (PredP), 26, 94, 96–100, 104, 107–110, 115, 123, 126–127, 134, 136, 141, 149–150, 154, 160, 162–163, 179, 192, 208, 216–217, 220–221, 223– 224 Preterito-presentia, 169 Projection Principle, 99 Proper Binding Condition (PBC), 22, 24 Prosodic constraint, 56, 177, 226 Quantifier, 16–18, 36, 43, 57, 83, 108–111, 150–151 Quantifier Raising (QR), 43, 81, 83, 113–114, 206 Relational feature, 50, 57, 215 Relativized Minimality, 30, 82 Remnant extraposition, 69, 71, 76, 81–82, 107, 164. See also te-infinitives, and the third construction Remnant movement, 3–6, 12, 21–24, 32, 101, 110, 116, 214, 222 and head movement, 25–26 and licensing movement, 26, 89, 216 and scrambling, 27–29, 213, 229, 231 Remnant topicalization, 12, 213, 231

253

Restructuring, 3–6, 12, 14–15, 17, 21, 28–30, 32, 62, 64, 87, 101, 118–119, 123, 126– 127, 130, 134, 136–137, 141, 146, 148– 149, 153–154, 156, 158–162, 167, 175, 186, 189–192, 194, 208–209, 211, 213– 214, 216, 218, 220, 222, 229–231 and extraposition, 111–112 as head movement, 101–104 and long distance scrambling, 105–107 and remnant movement, 89, 94, 107–117 and transparency, 108–109 as XP movement, 101, 115–116 Restructuring infinitive, 5–6, 14–15, 21, 32, 64, 101, 115, 119, 123, 130, 141, 154, 189, 204, 212, 220, 229, 231 and adjuncts, 131–133, 221 biclausal analysis, 131–136 and binding properties, 133–136, 150, 221– 223 and long passive, 152–153, 222 monoclausal analysis, 221, 223 and reconstruction, 151 and scrambling, 133–136 Restructuring verb, 4, 21, 101, 118, 126, 129, 133, 141, 161, 165, 167–168, 172, 219– 221, 224, 228 control verbs, 15, 65, 70, 72, 120–121, 131– 133, 137, 139, 141–144, 147, 150, 152, 154, 156–158 ECM-verbs, 65, 102, 120–121, 130–131, 133–136, 144–148, 156, 221–222 functional, 129 lexical, 129 subject raising verbs, 65, 141 Right Roof Constraint, 189 Scope, 16–17, 27, 29, 33, 36, 41, 49–51, 54–60, 62–63, 78, 80–83, 88, 109, 113–116, 137– 140, 145, 149–151, 163, 188–189, 197, 200, 206, 211, 215, 217–218, 221, 227 feature, 33–34, 45, 57, 61, 215 filter, 58 position, 55, 59 Scrambling, 3–9, 11–16, 18, 21–24, 26–30, 33– 51, 56–63, 71–72, 76, 78, 80–83, 88, 92, 96–97, 99, 101, 105–109, 111–117, 136, 138, 146, 149, 158, 213–218, 229, 231 as alternative adverb placement, 44, 46 as A-movement, 109 as A'-movement, 38, 109 as base generation, 8, 35 clause boundedness of, 109–115 and licensing movement, 12–13, 22–23, 28, 97, 115, 216–218, 229, 231 and object shift, 39, 41, 44, 61–62 and optionality, 50–51, 55–56, 60–62, 171, 215 and permutation, 8, 34, 39, 44, 63 semantic/pragmatic effects of, 47 string vacuous scrambling, 27, 44, 48 and topicalization, 7–9, 13, 23, 26

254

INDEX

Scrambling of adjuncts, 46, 106, 108, 114–115 Scrambling of arguments, 44, 46, 115 Scrambling of predicates, 115–116 Scrambling trigger, 33, 51 familiarity, 30, 48, 116 scope (see Scope, feature) specificity, 28, 30, 41, 50, 54, 57, 59, 61–63, 116, 215 Sentential complementation, 101, 117–119, 122–128, 185, 218, 227, 229 Small Clause, 9, 26–27, 68–69, 71, 73, 77, 91, 95–97, 100, 105, 109–110, 136, 162, 216 Spell-Out, 34, 43, 51–52, 55–56, 63, 85, 91, 108, 114, 126, 167, 176–178, 184, 210– 211, 214, 224–226 Split C-domain, 118, 125, 218 S-scrambling, 5–6, 35–36, 41, 106–107 Status Phrase (StatP), 124, 127, 189, 208, 210– 211, 216–217, 219–220, 223, 229 te-infinitive, 65–66, 86, 103, 182, 201 and the IPP-effect, 71 and the third construction, 69–70 Temporal anchoring. See Temporal interpretation Temporal interpretation, 118–119, 124, 133, 166–167, 169, 174, 176, 218–220, 224 tense and control, 120–123 Tense Phrase (TP), 28, 99, 101, 107, 109, 112, 114–119, 123–127, 133–135, 149–150, 154, 157–160, 189–190, 208, 214, 216– 221, 223–224, 229–231 to-infinitive (coherent), 4, 18, 20, 110, 119– 124, 127, 131, 136, 141, 146–147, 149, 157–159, 161, 163–164, 167, 172, 174– 175, 193, 208, 210–212, 214, 222, 224– 225, 228–230 biclausal analysis, 147–159, 222–223 and binding, 133, 142–143, 154–155 and case assignment, 142, 148, 155–159 and control, 141–142, 150, 152, 154 monoclausal analysis, 130, 137, 144, 221 and negation, 138–141, 150–151 and scrambling, 138 and topicalization, 140–141, 193, 231 Topicalization of verb projections. See VPtopicalization TP-movement, 108–111, 114–115, 127, 162– 163, 192, 208, 221 Trace, 21–25, 28, 45, 75, 81–82, 116, 133, 188, 229 Transparency, 5, 14, 16, 19–20, 102, 108, 111, 114, 132, 231 T-scrambling, 35 Unambiguous Domination, 23–24 Universal Base Hypothesis (UBH), 11, 13–14, 31–32 Verbal complex. See Verb cluster Verbal infinitive, 174, 199, 225

Verb Cluster, 19, 64, 71, 73–74, 77–78, 81, 88, 101–102, 108, 110–111, 113, 115, 123, 127, 140, 146, 161–165, 167, 175–187, 189–191, 199–204, 206–207, 212, 216– 218, 224–228, 230–231 and CP-complements, 165 in Dutch, 181–182 and extraction, 226–227 in German, 179–181 left branching, 161–165, 225–227 and remnant extraposition, 164 right branching, 161, 175–179, 225–226 and Spell-out, 177 and temporal licensing, 161 and to-infinitives, 163–164 and verb projection raising, 174, 178–179 in West Flemish, 182–183 Verb particle, 26, 76–77, 88, 97, 105, 160, 181, 216, 228 as head (see Incorporation) and participial prefix, 75–78 particle climbing, 74–78, 162 in verb clusters, 162–163 Verb Projection Raising (VPR), 64, 78–79, 80– 84, 87–89, 101, 111–116, 122–123, 164, 178–179, 193, 200–201, 203–209, 215– 218, 228, 230 and auxiliary switch, 87–88 and extraposition, 80–81 and pied-piping, 111–115 Verb Raising (VR), 15, 20, 56, 74–79, 81–84, 87–89, 101–105, 107, 109, 111–116, 122– 123, 148, 160, 164, 168, 172, 175, 178, 192, 196, 207, 215–216, 218, 222, 228, 230 and auxiliary switch, 87 as formal licensing, 68 as head movement, 20, 73, 78 and the IPP-effect, 20, 70–71, 77, 82, 172, 175, 191–192, 224–228 and inversion, 68, 83, 166, 170, 180 as XP-movement, 64, 73, 101, 103–104, 107– 108, 162–164 see also Restructuring, as head movement VP-movement, 125–126 VP-topicalization, 27, 29, 32, 119, 127, 141, 177, 185, 187–194, 198–200, 203–207, 211–212, 214, 227–229, 231 and the Aspect Phrase, 198, 207 and the base generation approach, 193–196 and d-word left dislocation, 196–198 and the IPP-effect, 191–192, 198–199 and nominal infinitives, 203–205 and the phase condition, 190–191, 201 and reconstruction effects, 197–198 and verb clusters, 192–193 and Verb Projection Raising, 198–200, 206–207 WCO-effect, 36, 38, 40, 43 Weak pronoun, 16, 41, 43, 59, 135, 215 Zero-morpheme, 168, 203, 225–226

Scrambling, Remnant Movement, and Restructuring in West Germanic (Oxford Studies in Comparative Syntax)

Verb Movement and Expletive Subjects in the Germanic Languages (Oxford Studies in Comparative Syntax)

Movement and Silence (Oxford Studies in Comparative Syntax)

Portuguese Syntax: New Comparative Studies (Oxford Studies in Comparative Syntax)

Studies in Comparative Germanic Syntax: Proceedings from the 15th Workshop on Comparative Germanic Syntax

The Syntax of Anaphora (Oxford Studies in Comparative Syntax)

The Syntax of Anaphora (Oxford Studies in Comparative Syntax)

The Syntax of Anaphora (Oxford Studies in Comparative Syntax)

Parameters and Functional Heads: Essays in Comparative Syntax (Oxford Studies in Comparative Syntax)

Comparative Studies in Romanian Syntax

Subjects, Expletives, and the EPP (Oxford Studies in Comparative Syntax)

Parameters and Universals (Oxford Studies in Comparative Syntax)

Comparisons and Contrasts (Oxford Studies in Comparative Syntax)

Restructuring and Functional Heads: The Cartography of Syntactic Structures Volume 4 (Oxford Studies in Comparative Syntax)

Clause Structure and Word Order in Hebrew and Arabic: An Essay in Comparative Semitic Syntax (Oxford Studies in Comparative Syntax)

Clause Structure and Word Order in Hebrew and Arabic: An Essay in Comparative Semitic Syntax (Oxford Studies in Comparative Syntax)

The Grammar of Q: Q-Particles, Wh-Movement, and Pied-Piping (Oxford Studies in Comparative Syntax)

Parameters of Slavic Morphosyntax (Oxford Studies in Comparative Syntax)

Advances in Comparative Germanic Syntax (Linguistik Aktuell Linguistics Today)

Particles: On the Syntax of Verb-Particle, Triadic and Causative Constructions (Oxford Studies in Comparative Syntax)

Particles: On the Syntax of Verb-Particle, Triadic and Causative Constructions (Oxford Studies in Comparative Syntax)

Negation and Clausal Structure: A Comparative Study of Romance Languages (Oxford Studies in Comparative Syntax)

Narrow Syntax and Phonological Form: Scrambling in the Germanic languages (Linguistik Aktuell Linguistics Today)

Subject Inversion in Romance and the Theory of Universal Grammar (Oxford Studies in Comparative Syntax)

The Syntax of Ellipsis: Evidence from Dutch Dialects (Oxford Studies in Comparative Syntax)

Adverbs and Functional Heads: A Cross-Linguistic Perspective (Oxford Studies in Comparative Syntax)

Principles and Parameters of Syntactic Saturation (Oxford Studies in Comparative Syntax)

Principles and Parameters of Syntactic Saturation (Oxford Studies in Comparative Syntax)

A Unified Theory of Verbal and Nominal Projections (Oxford Studies in Comparative Syntax)

The Oxford Handbook of Comparative Syntax (Oxford Handbooks)

Theoretical Comparative Syntax: Studies in Macroparameters (Routledge Leading Linguists)

Scrambling, Remnant Movement, and Restructuring in West Germanic (Oxford Studies in Comparative Syntax)

Verb Movement and Expletive Subjects in the Germanic Languages (Oxford Studies in Comparative Syntax)

Movement and Silence (Oxford Studies in Comparative Syntax)

Portuguese Syntax: New Comparative Studies (Oxford Studies in Comparative Syntax)

Studies in Comparative Germanic Syntax: Proceedings from the 15th Workshop on Comparative Germanic Syntax

The Syntax of Anaphora (Oxford Studies in Comparative Syntax)

The Syntax of Anaphora (Oxford Studies in Comparative Syntax)

The Syntax of Anaphora (Oxford Studies in Comparative Syntax)

Parameters and Functional Heads: Essays in Comparative Syntax (Oxford Studies in Comparative Syntax)

Comparative Studies in Romanian Syntax

Subjects, Expletives, and the EPP (Oxford Studies in Comparative Syntax)

Parameters and Universals (Oxford Studies in Comparative Syntax)

Comparisons and Contrasts (Oxford Studies in Comparative Syntax)

Restructuring and Functional Heads: The Cartography of Syntactic Structures Volume 4 (Oxford Studies in Comparative Syntax)

Clause Structure and Word Order in Hebrew and Arabic: An Essay in Comparative Semitic Syntax (Oxford Studies in Comparative Syntax)

Clause Structure and Word Order in Hebrew and Arabic: An Essay in Comparative Semitic Syntax (Oxford Studies in Comparative Syntax)

The Grammar of Q: Q-Particles, Wh-Movement, and Pied-Piping (Oxford Studies in Comparative Syntax)

Parameters of Slavic Morphosyntax (Oxford Studies in Comparative Syntax)

Advances in Comparative Germanic Syntax (Linguistik Aktuell Linguistics Today)

Particles: On the Syntax of Verb-Particle, Triadic and Causative Constructions (Oxford Studies in Comparative Syntax)

Particles: On the Syntax of Verb-Particle, Triadic and Causative Constructions (Oxford Studies in Comparative Syntax)

Negation and Clausal Structure: A Comparative Study of Romance Languages (Oxford Studies in Comparative Syntax)

Narrow Syntax and Phonological Form: Scrambling in the Germanic languages (Linguistik Aktuell Linguistics Today)

Subject Inversion in Romance and the Theory of Universal Grammar (Oxford Studies in Comparative Syntax)

The Syntax of Ellipsis: Evidence from Dutch Dialects (Oxford Studies in Comparative Syntax)

Adverbs and Functional Heads: A Cross-Linguistic Perspective (Oxford Studies in Comparative Syntax)

Principles and Parameters of Syntactic Saturation (Oxford Studies in Comparative Syntax)

Principles and Parameters of Syntactic Saturation (Oxford Studies in Comparative Syntax)

A Unified Theory of Verbal and Nominal Projections (Oxford Studies in Comparative Syntax)

The Oxford Handbook of Comparative Syntax (Oxford Handbooks)

Theoretical Comparative Syntax: Studies in Macroparameters (Routledge Leading Linguists)

Recommend Documents