Clause Structure and Adjuncts in Austronesian Languages 9783110922974, 9783110190052

Citation preview

Clause Structure and Adjuncts in Austronesian Languages

W G DE

Studies in Generative Grammar 87

Editors

Henk van Riemsdijk Jan Köster Harry van der Hulst

Mouton de Gruyter Berlin · New York

Clause Structure and Adjuncts Austronesian Languages

edited by

Hans-Martin Gärtner Paul Law Joachim Sabel

Mouton de Gruyter Berlin · New York

Mouton de Gruyter (formerly Mouton, The Hague) is a Division of Walter de Gruyter G m b H & Co. KG, Berlin. The series Studies in Generative Grammar was formerly published by Foris Publications Holland.

® Printed on acid-free paper which falls within the guidelines of the ANSI to ensure permanence and durability.

Library of Congress Cataloging-in-Publication

Data

Clause structure and adjuncts in Austronesian languages / edited by Hans-Martin Gärtner, Paul Law, Joachim Sabel. p. cm. — (Studies in generative grammar ; 87) Includes bibliographical references and index. ISBN-13: 978-3-11-019005-2 (cloth : alk. paper) ISBN-10: 3-11-019005-2 (cloth : alk. paper) 1. Austronesian languages - Clauses. 2. Austronesian languages — Word order. 3. Philippines — Languages — Clauses. 4. Grammar, Comparative and general - Syntax. I. Gärtner, Hans-Martin. II. Law, Paul S. III. Sabel, Joachim, 1962- . IV. Series. PL5035.C53 2006 499'.2—dc22 2006007370

Bibliographic information published by Die Deutsche

Bibliothek

Die Deutsche Bibliothek lists this publication in the Deutsche Nationalbibliografie; detailed bibliographic data is available in the Internet at .

ISBN-13: 978-3-11-019005-2 ISBN-10: 3-11-019005-2 ISSN 0167-4331 © Copyright 2006 by Walter de Gruyter G m b H & Co. KG, D-10785 Berlin. All rights reserved, including those of translation into foreign languages. No part of this book may be reproduced in any form or by any means, electronic or mechanical, including photocopy, recording, or any information storage and retrieval system, without permission in writing from the publisher. Cover design: Christopher Schneider, Berlin. Printed in Germany.

Contents Preface Hans-Martin Gärtner, Paul Law, and Joachim Sabel Clause Structure and Adjuncts in Austronesian Languages: A Critical Introductory Survey Hans-Martin Gärtner, Paul Law, and Joachim Sabel The Guest Playing Host: Adverbial Modifiers as Matrix Verbs in Kavalan Henry Yungli Chang Seediq - Adverbial Heads in a Formosan Language Arthur Holmer Patterns of Phrasal Movement: The Niuean DP Arsalan Kahnemuyipour, and Diane Massam Rigidity versus Relativity in Adverbial Syntax: Evidence from Tagalog Daniel Kaufman The Cleft Structure of Malagasy ^-Questions Eric Potsdam Three Systems of Remnant Movement II and Extraction from Specifier Position Craig Thiersch Voice Morphology in Malagasy as Clitic Left Dislocation or Through the Looking Glass: Malagasy in Wonderland

vii

1

43

83

125

151

195

233

281

Lisa deMena Travis List of Contributors

319

Subject Index

321

Index of Adverbials, Adjectives, and Particles

327

Preface Hans-Martin Gärtner, Paul Law, and Joachim Sabel

Modern linguistic theory is mostly developed from studying more familiar languages like English, German or Italian. It is thus natural to bring less familiar languages like those in the Austronesian language family to bear. In this way one can see to what extent various hypotheses regarding the grammatical structure of natural language hold cross-linguistically. Given that Austronesian languages are superficially very different from most European languages, they are a good testing ground for theories that are based on well-known languages like English, German or Italian. Clause structure has long been a major focus of attention in the formal analysis of Germanic and Romance languages, but the same rigorous approach to clause structure in Austronesian languages has a much shorter history. Given the superficial differences between the two language families - predicate-first vs. non-predicate-first declarative sentences, the restriction on extraction, etc. - attempts to capture what they have in common are few and far between. A central question is whether the strong claim can be upheld that clause structure may be universal, or whether we have to adopt a weaker position, i.e. that languages vary with respect to clause structure. Within approaches to understand the clause structure of natural languages, the syntax of adjuncts - adverb(ial)s as well as attributive adjectives - has only recently been the locus of intense and fruitful investigation of cross-linguistic phrase structure. Cinque's (1999) assumption that the universal order of adverbials is related to a universal hierachy of functional projections is a prominent approach in this respect. As for studies on adverbials in some Austronesian languages, they are relatively rare. To the extent that one finds it, they are mostly from the morphological point of view, e.g. Li's (2003) work on the Formosan language Thao. Three notable exceptions are the description of the positioning of some adverbs in Tagalog by Schachter and Otanes (1972), and the two formal analyses of adverbs in Malagasy (Rackowski 1998; Rackowski & Travis 2000). For these reasons we believe that it is time to devote some serious attention to the syntactic properties of adjuncts and bring them to bear on

viii

Hans-Martin Gärtner, Paul Law, and Joachim Säbel

clause structure. The papers collected for this volume, most of which contain empirical data that are little known, are a contribution toward this goal. They are concerned with the theoretical issues of clause structure and the syntax of adjuncts in Austronesian languages. Most of the papers stemmed from a special session on the same topics that was part of the 11th Austronesian Formal Linguistics Association (AFLA XI) conference held in ZAS (Zentrum für Allgemeine Sprachwissenschaft / Center for General Linguistics, Typology, and Universals Research), Berlin, on April 25, 2004. We gratefully acknowledge the financial support from the DFG (Deutsche Forschungs Gemeinschaft / German Science Foundation). Finally, we thank Mechthild Bernhard and Paul David Doherty for their invaluable assistance in turning the manuscript into a book.

The contributions GÄRTNER, Hans-Martin, Paul LAW, and Joachim SABEL Clause Structure and Adjuncts in Austronesian Languages: A Critical Introductory Surrey In this chapter, we first lay out the three well-known properties of the Philippine-type languages, namely, voice, predicate-first and the restriction on extraction, as these have a bearing on clause structure and the syntax of adjuncts. More specifically, we discuss three formal approaches to the relation between voice and the structural position of the arguments, and the various issues that arise in each of them. We consider how variants of Guilfoyle, Hung & Travis's (1992) analysis of sentence structure may account for the predicate-first property, bringing them to bear on Kayne's (1994) Antisymmetry Hypothesis, and how the well-known restriction on extraction in Austronesian languages may shed some light on the structural position of the extracted argument, and on the extraction of adjuncts. We then turn to the positioning of adjuncts in Austronesian languages, bring it to bear on recent proposals concerning the way in which adjuncts come to be in the position they are, i.e. whether they occur in a fixed hierarchical order of functional categories (Cinque 1999) or whether their positioning is a result of the syntactic correlates of the semantic relations (Ernst 2002). As it turns out, the facts in Austronesian languages considered in the remaining chapters have a direct bearing on the two

Preface

ix

views. Lastly, we discuss the issue of adjunct extraction. The issue is of special interest, since from the perspective of linguistic theory based on familiar languages like English and German it is not obvious why extraction of adjuncts is not subject to the stringent restriction on extraction of arguments.

CHANG, Henry Yungli The Guest Playing Host: Adverbial Modifiers as Matrix Verbs in Kavalan This chapter suggests that inflected adverbs in Kavalan have the syntax of serial verb construction (SVC). They are of two types. One type comprising manner, iterative / time-related and frequency adverbs taking non-active focus morphology has the property that the cooccurring thematic verb (the verb assigning a thematic role to the argument) must have active focus morphology, the same restriction on the thematic verb appearing after a phasal / aspectual verb like siangatu ('begin') in one type of SVC. The other type of adverbs to which frequency adverbs taking active focus morphology belong lacks the property of the first type with respect to the restriction on the co-occurring thematic verb. This latter type resembles the other type of SVC, in which an adjectival predicate cooccurs with a thematic verb with either active focus or non-active focus morphology. It is suggested that inflected adverbs form complex predicates with the following verbs, and the two jointly assign Case and a theta-role to the arguments. Chang speculates that the morphological restriction on the following verbs and the lack thereof is due to the extent to which the adverbs are closely bound to the verbs with respect to event integration on a scale that also applies to the different types of complements, finite vs non-finite with an infinitival marker vs non-finite without an infinitival marker in English (Givon 1980).

HOLMER, Arthur

Seediq - Adverbial Heads in a Formosan Language In this chapter it is shown that some adverbs in Seediq inflect morphologically like verbs in that they attract clitic pronouns, take voice morphology and appear in a special connegative morphology when occur-

χ

Hans-Martin Gärtner, Paul Law, and Joachim Säbel

ring after the negation ini. There are also some other adverbs that do not have these properties. The first type of adverbs occurs preverbally, and blocks head-movement, while the latter type appears postverbally and does not block head-movement. Despite their differences, it is claimed that postverbal adverbs too are heads like preverbal adverbs, appealing to the fact that these cannot be expanded as phrases or be replaced by phrases. Moreover, postverbal adverbs express the meaning of a complementizer or aspect which are projected as heads in other languages. To derive the correct word order in a theory like Cinque's where adverbs are base-generated in fixed positions, it must be assumed that postverbal adverbs trigger predicate-fronting, while preverbal ones do not. Holmer's analysis thus falls under the same class of accounts such as that proposed by Rackowski & Travis (2000) for Malagasy according to which the order of postverbal adverbs appearing in the mirror-image of the Cinque Hierarchy is derived by intraposition, i.e. fronting of the predicate to a specifier position.

KAHNEMUYIPOUR, Arsalan, and Diane MASSAM Patterns of Phrasal Movement: The Niuean DP This chapter considers the order of adjectival adjuncts in noun phrases in Niuean, a Polynesian language of the Tongic subgroup with VSO word order. It compares the order of constituents in clauses and the allowable orderings of elements in the noun phrase, including certain variations in the placement of numerals and the genitive possessor. Cinque (1996) accounts for Greenberg's typological generalization (Universal 20) concerning the order of elements within DP by assuming that N-final is the base-generated order. The authors adopt this view, taking Ν to be base-generated at the end of DP in Niuean. The orders in which other lexical items within DP follow Ν are derived via XP-movement as argued in Cinque (1996), although the proposed analysis is different in technical details. It is in line with the Rackowski & Travis (2000) account of the Niuean verbal clause. The chapter provides a roll-up complement-tospecifier analysis of Niuean DPs that derives the correct word orders and accounts for the variation in the position and properties of possessors and for the variation in the position of numerals. It is proposed that two factors can affect this pattern of phrasal movement. First, filled specifiers make

Preface

xi

movement to specifier impossible. Secondly, the data confirm a movement constraint observed by Rackowski & Travis (2000), which states that purely relational functional projections, which have no semantic content, are invisible to movement and cannot themselves move. It is argued that the attested orders are a consequence of universal constraints, rather than of typological stipulations.

KAUFMAN, Daniel

Rigidity versus Relativity in Adverbial Syntax: Evidence from Tagalog This chapter compares Cinque's (1999) theory of adverbs with that of Ernst's (2002) in accounting for the syntax and scope of Tagalog adverbials. It examines extensively the correlation between position and interpretation of multiple adverbs in Tagalog. The data validates Ernst's claims that scope is essentially concentric in relation to the verb, with outer adverbs scoping over inner adverbs regardless of left-right directionality. This state of affairs can only be derived at some cost from a fixed underlying hierarchy such as posited in Cinque (1999). It also shows that if scope is determined by base position and an adverb with wide scope is generated higher than its narrow scope counterpart then the theory both under- and over-generates interpretations. The data presented here appear to endorse a theory in which adverbs may be freely left and right-adjoined with scope being read off the surface structure. Taking a certain class of Tagalog adverbs to be VP-internal explains why these are marked with nang [ng\ just as internal arguments are. In addition, it also accounts for their tendency of having narrow scope regardless of linear order. The effects of prosody are briefly discussed as well as the problems of determining the scope of second-position clitics. Finally, the question is raised whether a verb-initial language could ever display the Cinque order of adverbs in the post-verbal domain. The fact that such a case is so far unattested further strengthens the position that a scope principle rather than a templatic hierarchy is responsible for the positioning of adverbs.

xii

Hans-Martin Gärtner, Paul Law, and Joachim Sabel

POTSDAM, Eric The Cleft Structure of Malagasy

Wh-Questions

This chapter considers the structural analysis of w/z-questions in the predicate-initial language Malagasy. In addition to a vv/i-in-situ strategy, Malagasy allows the wh-phrase to appear at the front of the clause: iza no nihomehy? (who+PRT+laugh; 'Who laughed?')· The author argues that whquestions in Malagasy do not involve w/z-movement but are pseudo-clefts in which the wh-phrase is the predicate and the remaining material is a headless relative clause in subject position involving empty operator movement. The example above is thus more literally translated as "The one who laughed is who?" The paper concludes that Malagasy is a pure wh-insitu language. The arguments in favor of the pseudo-cleft analysis and against a w/z-movement analysis come from: (i) Predicate properties of the initial w/z-phrase. Certain particles immediately precede predicates as well as w/z-phrases and certain others follow predicates as well as w/z-phrases in VOS clauses. The cleft analysis of w/z-questions correctly predicts the distribution of these particles in whquestions because the w/z-phrase too is a predicate. (ii) Syntactic and semantic parallels with a focus construction that Paul (2001) independently analyzed as a pseudo-cleft. It is shown that Malagasy has a focus construction that is formally similar to w/z-questions. (iii) The analysis of certain multiple w/z-questions from Sabel (2003). This argument for the cleft analysis comes from a consideration of multiple w/z-questions, in which there are two w/z-phrases at the front of the clause. It is argued that these sentences have properties different from sentences involving multiple w/z-movement as found in multiple w/z-fronting languages such as Bulgarian. The final section deals with a central analytical detail within the pseudo-cleft analysis, i.e. the treatment of the particle no. It is argued that no is the relative complementizer of the headless relative clause in subject position.

Preface

xiii

THIERSCH, Craig

Two Systems of Remnant Movement II and Extraction from Specifier Position This chapter takes as its starting point the observation that in the wake of Kayne's Antisymmetry Hypothesis and Linear Correspondence Axiom (LCA), there has been much fruitful research attempting to adjust syntactic analyses to those permitted by Kayne's restrictive system. In doing so, analyses which at first seem counter-intuitive may turn out to provide solutions to old problems. In the light of Kayne's (1994) 'antisymmetric program', remnant movement has become an important topic. Since then, certain word order phenomena have been reanalyzed as remnant movement phenomena. Three cases in point are discussed. The first case concerns the analysis of adverb order in Malagasy. Assuming Kayne's (1994) theory of antisymmetry, the universal order of elements within maximal projections is specifier < head < complement, i.e. SVO. To derive the unmarked word order of verb-initial languages such as Malagasy within the LCA framework, a rule of predicate-fronting has been proposed by Rackowski & Travis (2000), and Pearson (2001), among others. Rackowski & Travis argue that repeated ("roll-up") remnant movement in Malagasy can also account for the fact that adverbs in Malagasy show the mirror order with respect to the hypothesized universal order of adverbials proposed in Cinque (1999). The second case of remnant movement discussed here concerns the analysis of Hungarian and Dutch verbal clusters in Koopman & Szabolcsi (2000). Finally, these approaches are compared with the system in Müller (2004) who reanalyzes verb movement as remnant movement. Of central concern for the above mentioned analyses of remnant movement are conditions on the strict cyclicity of derivations as well as the nonextraction from specifiers a.k.a. "freezing condition". It is shown that each of the systems analyzed have to resort to rather ad hoc mechanisms in order to comply with these constraints.

xiv

Hans-Martin Gärtner, Paul Law, and Joachim Sabel

TRAVIS, Lisa deMena Voice Morphology in Malagasy as Clitic Left Dislocation or Through the Looking Glass: Malagasy in Wonderland This chapter represents an attempt to fit Malagasy into a language typology that disallows DP movement in predicate-intial languages. The starting point is the assumption that languages have two types of features that force displacement, D and V. Languages may vary on which one triggers XP movement and which one triggers head movement. English is a V°, DP movement language while Malagasy is a language that moves VPs to satisfy a predicate feature and D°s to satisfy a nominal feature. The task is then undertaken to find an analysis of the voice system in a Western Malayo-Polynesian language like Malagasy that does not involve DP movement. Combining work by Haegeman (2004), Keenan (2000), Pearson (2001), Sells (1998), and Sportiche (1996), the author explores the possibility that the voice system in Malagasy has more in common with the clitic left dislocated structures of Italian than the active / passive alternations of a language like English or the topicalized structures of Germanic languages. It is argued that voice morphology is more accurately viewed as verbal clitic morphology, and subjects, which are generated external to the VP, have their function identified by this morphology. The subject is in a predication relation with the rest of the clause. This relation is set up via pro in argument position that is identified by the verbal clitic. "Voice" alternations are analyzed as different clitic left dislocated structures. Apparent w/z-movement also involves no movement but, as proposed by Paul (2001), a pseudo-cleft structure. However, in contrast to Paul (2001) and others, the analysis of the pseudo-cleft construction does not involve empty operator movement to Spec CP but relies crucially on the subject / predicate relation. The final portion of the discussion situates adjuncts and prepositional phrases within this view of the subject / predicate structure of Malagasy.

Preface

xv

References Cinque, Guglielmo 1996 The Antisymmetric Programme: Theoretical and Typological Implications. Journal of Linguistics, 32: 4 4 7 ^ 6 4 . 1999 Adverbs and Functional Heads: A cross-linguistic perspective. Oxford Studies in Comparative Syntax. Oxford: Oxford University Press. Ernst, Thomas 2002 The Syntax of Adjuncts. Cambridge: Cambridge University Press. Givon, Talmy 1980 The Binding Hierarchy and the Typology of Complements. Studies in Language 4: 333-377. Guilfoyle, Eithne, Henrietta Hung, and Lisa Travis 1992 Spec of IP and Spec of VP: Two Subjects in Austronesian Languages. Natural Language and Linguistic Theory 10: 375-414. Haegeman, Liliane 2004 Adverbial Clauses, the left periphery and implications for the syntax of topics. Paper presented at Georgetown University Round Table (GURT), Georgetown University. Kayne, Richard 1994 The Antisymmetry of Syntax. Linguistic Inquiry Monographs 25. Cambridge, MA: MIT Press. Keenan, Edward 2000 Morphology is Structure: A Malagasy Test Case. In Formal Issues in Austronesian Linguistics, Ileana Paul, Vivianne Phillips, and Lisa Travis (eds.), 27—47. Dordrecht: Kluwer. Koopman, Hilda, and Anna Szabolcsi 2000 Verbal Complexes. Current studies in linguistics 34. Cambridge, MA: MIT Press. Li, Paul Jen-kuei 2003 Verbs or adverbs in Thao? Paper presented at the second Workshop on Formosan Languages, Academia Sinica, Taipei, Nov. 1-2. Müller, Gereon 2004 Verb Second as vP First. Journal of Comparative Germanic Linguistics 7: 179-234. Paul, Ileana 2001 Concealed pseudo-clefts. Lingua 111: 707-727. Pearson, Matthew 2001 The Clause Structure of Malagasy: A Minimalist Approach. Doctoral dissertation, UCLA.

xvi

Hans-Martin Gärtner, Paul Law, and Joachim Sabel

Rackowski, Andrea 1998 Malagasy Adverbs. In The Structure of Malagasy, Volume II, Ileana Paul (ed.), 11-33. Los Angeles: UCLA Occasional Papers in Linguistics. Rackowski, Andrea, and Lisa Travis 2000 V-initial Languages: X or XP Movement and Adverb Placement. In The Syntax of Verb Initial Languages, Andrew Carnie, and Eithne Guilfoyle (eds.), 117-142. Oxford: Oxford University Press. Sabel, Joachim 2003 Malagasy as an optional multiple wh-fronting language. In Multiple Wh-Fronting, Cedric Boeckx, and Kleanthes Grohmann (eds.), 229254. Amsterdam: John Benjamins. Schachter, Paul, and Fe Otanes 1972 Tagalog Reference Grammar. Berkeley CA: University of California Press. Sells, Peter 1998 The Functions of Voice Markers in Philippine Languages. In Morphology and its Relation to Phonology and Syntax. Steven Lapointe, Diane K. Brentari, and Patrick M. Farrell (eds.), 111-137. Stanford, CA: CSLI Publications. Sportiche, Dominique 1996 Clitic Constructions. In Phrase Structure and the Lexicon. John Rooryck, and Laurie Zaring (eds.), 213-276. Dordrecht: Kluwer.

Clause Structure and Adjuncts in Austronesian Languages: A Critical Introductory Survey* Hans-Martin Gärtner, Paul Law, and Joachim Sabel

Syntactic analyses of Austronesian languages have predominantly been concerned with three phenomena. First, and perhaps most widely known, there is the controversy about how to view Philippine-type voice systems. These are typically symmetrical in the sense that what resembles passivization does not lead to demotion, i.e. oblique status of one of the arguments involved. This symmetry is closely related to the difficulty of determining the grammatical function of "subject." Thus, although voice morphology correlates with (the semantic role of) a single designated argument, which we will call "trigger" (argument) henceforth, standard subject properties are distributed between this trigger and an actor argument when the two do not coincide. Secondly, Austronesian languages tend to have head initial word order, which often results in verb initial or predicate initial clause structure. Thirdly, there exists a condition on unbounded dependencies for arguments, disallowing extraction of anything other than the trigger. While questions surrounding these issues are clearly far from settled, the volume we are presenting and discussing here is intended to shift perspectives and reflect on the status of adjuncts in Austronesian languages as well as the repercussions this has on analyzing Austronesian clause structure. The most obvious motivation for this shift is that much less is known about adjuncts in Austronesian languages. Secondly, studying the syntax of adjuncts in other languages has regularly been a catalyst in developing more fine-grained theories of phrase structure and locality. Thirdly, recent controversies about the nature of adverb placement, i.e. whether or not it is governed by a universal hierarchy of functional projections, has made a survey of less well documented language types such as Austronesian languages more urgent, not the least because an initial study of Malagasy adverb order indicated an interesting kind of confirmation of the formalist / universalist hypothesis. We will now proceed as follows. Section 1 provides a rough sketch of the three "big questions" of Austronesian syntax mentioned above. Section

2

Hans-Martin Gärtner, Paul Law, and Joachim Sabel

2 will then be devoted to adjuncts and briefly document some of the most influential syntactic studies of adjuncts and their consequences for syntactic theorizing. This will be interlaced with discussion of Austronesian adjuncts, providing an overview of some currently available insights and raising a number of research questions that we feel have to be tackled in future research. As we go along, we will introduce and critically reflect on the seven studies comprising the main chapters of this volume and point out their contribution to the bigger picture as we see it. Summaries of the individual chapters can be found in the preface to this volume.

1. Three major issues in Austronesian syntax It is hardly controversial that the questions of how to analyze Philippinetype voice systems, head intial clause structure, and the "trigger-only" condition on unbounded dependencies have taken center stage in analyses of Austronesian syntax.1 It is equally clear that an appraisal of the syntactic behavior of adjuncts in Austronesian languages only makes sense against the background of these issues. It is thus useful to give at least a rough sketch of each of them in turn.

1.1. Philippine-type voice The specifics of Philippine-type voice systems can best be introduced by way of an example. Thus, consider the following sentences from Tagalog (Schachter 1993: 1419).2 (1)

a. Mag-aabot ang babae ng laruan sa bata. AT-will.hand Τ woman TH toy D child 'The woman will hand a toy to a/the child.' b. Iaabot ng babae ang laruan sa bata. THT.will.hand A woman Τ toy D child 'A/The woman will hand the toy to a/the child.' c. Aabutan ng babae ng laruan ang bata. DT.will.hand A woman TH toy Τ child 'A/The woman will hand a toy to the child.'

Clause Structure and Adjuncts in Austronesian Languages

3

As is well-known, alternating verbal morphology roughly correlates with (the thematic macro-role of) the ang-marked argument, which we will call "trigger", following Schachter's (1993) suggestion for avoiding the controversial term "subject". Thus, actor-trigger morphology (AT) on the verb in (la) designates ang babae as trigger, as evidenced by the trigger marker ang (T). Theme-trigger (THT) and directional trigger (DT) morphology single out ang laruan and ang bata as triggers in (lb) and (lc) respectively. One of the reasons for researchers being reluctant to identify the trigger with the (surface) subject is the fact that it can be reflexivized and bound from within the local clause. This is illustrated for Toba Batak in (2) (Sternefeld 1995: 56),3 where NAT stands for non-actor trigger morphology. (2)

Di-ida

si John

NAT-saw A J.

diri-na. T.himself

'John saw himself.' As Johnson (1977: 688) rightly pointed out already, facts like (2) and other "distributed" subject properties can be handled "within any theory that has grammatical stages", i.e. theories that rely on D-structure, S-structure and Logical Form like Government and Binding Theory (GB) (Chomsky 1981) or intial and final strata like Relational Grammar (RG) (Perlmutter 1980). What has remained controversial is whether the notion of subject should be dropped from the toolkit of universal grammar completely, as suggested for example by Schachter (1976) and assumed within the framework of Role and Reference Grammar (RRG) (Van Valin 1993), as well as by Williams (1984). Alternatively, the grammatical function of subject has either been taken to be defined phrase structurally (Chomsky 1965) or multifactorially (Keenan 1976b), or to figure as a primitive of the theory. The latter has been advocated in RG and Lexical Functional Grammar (LFG) (Bresnan 2001). Unsurprisingly, there are various formal analyses of the trigger, among which one can discern the following three main groups. Group 1 assimilates the trigger to a surface subject bearing nominative case4 and relegates binding to thematic structure, following Bell (1976). This is implemented LFG-style in Kroeger (1993) or GB-style by Guilfoyle, Hung & Travis (1992). Let us have a look at the latter very explicit and influential "GHT-system."

4

Hans-Martin Gärtner, Paul Law, and Joachim Sabel

By base generating all arguments VP-internally, hierarchically ordered in accordance with the thematic hierarchy, as argued for on the basis of his "Uniformity of Theta Assignment Hypothesis" (UTAH) by Baker (1988), this system allows binding principles to apply at D-structure and predicts actor arguments to bind reflexivized theme arguments as in (2).5 At the same time, trigger morphology on the verb is taken to block in situ case assignment to the corresponding argument, which is therefore forced to leave VP and acquire (nominative) case in Spec,IP. Thus, AT-morphology leads to an actor trigger and THT-morphology forces the theme argument to assume trigger function. Importantly, case absorption on the theme does not lead to oblique status of the actor. This modification of standard GB assumptions about passivization suffices to essentially capture the symmetrical nature of Philippine-type voice.6 Group 2 takes the trigger to correspond to the absolutive argument in an ergative system. Recent proponents of this approach are Maclachlan & Nakamura (1997), Wegmüller (1998), and Aldridge (2004). Under such a perspective, the THT variant of (1), i.e. (lb), is the unmarked transitive clause, not a passive, while the AT structure in (la) is taken to correspond to an antipassive. Arguments for this have to do with the overall lesser markedness of theme trigger morphology, which is mirrored in the higher frequency of THT clauses over AT clauses in texts (see Aldridge 2004: 59 for references). Also there exists an indefiniteness requirement on theme arguments in (la), which appears to be a wide-spread feature of ergative languages (cf. Wegmüller 1998: 81). At first sight, the analyses of group 2 resemble those of group 1 to a large extent. Thus, binding facts are again relegated to the uniformly projected VP. This is then supplemented by case marking mechanisms. These, however, show an asymmetry that the GHTsystem lacks. Thus, Wegmüller (1998: 247) manipulates the base position of the theme argument, generating it as specifier vs. adjunct of a lower VP

Clause Structure and Adjuncts in Austronesian

Languages

5

in THT vs. AT structures respectively. This allows her to exempt the (adjoined) theme argument from being assigned absolutive in antipassive structures.7 Aldridge (2004) distributes the assignment of absolutive among two heads. Thus, v°, the head of v(oice)P, assigns absolutive to the internal (theme) argument in transitive THT structures, while 1° (= T° in her system) assigns absolutive to the external (actor) argument in antipassive AT structures. Aldridge goes on to argue for a more symmetric ergative analysis of languages like Seediq, where T° alone figures as assigner of absolutive case. She also assumes Malagasy to be an ergative system whose antipassive has been reanalyzed as an active clause.8 One important consequence of this kind of approach is that the notion "ergative" must be taken as an envelope within which different grammar types emerge.9 According to group 3, finally, the trigger is a more peripheral category in an Α-bar position, roughly comparable to fronted XPs in Germanic V2 languages. Recent proponents of this idea are Richards (2000), Sells (2000), and Pearson (2001, 2005). Thus, Richards (2000: 107) extends the GHT structure to the one given in (4). (4)

[CP NPT [ C° [IP _ [ 1° [VP NPA [ V° NPT„ ] ] ] ] ] ]

According to him, the major difference between V2 languages and e.g. Tagalog lies in the point at which Spec,CP is filled: S-structure in the former and LF in the latter case. This approach dissociates trigger-choice from clausal transitivity, coming closer to more traditional conceptions of triggers as somehow information-structurally determined.10 At the same time, no commitment has to be made as to whether the trigger is either exclusively a topic or a focus (cf. Sells 2000: 122). Such an exclusive decision would be highly problematic as shown by Kroeger (1993). Instead, trigger choice acquires the well known text-sensitive elusiveness of fronting in languages like German.11 Additional assumptions are necessary, of course, to induce the correct trigger morphology on the verb. Pearson (2005: section 3) suggests that these morphemes are licensed on the verb when it goes into a specifierhead relation with an Α-bar operator within a projection where case is checked. For AT morphology the required projection would be TP, for THT some counterpart of v(oice)P. This mechanism clearly reinvokes earlier "w/z-agreement" approaches to Austronesian voice in Chamorro (Chung 1994), Palauan (Georgopoulos 1985), and Selayarese (Finer 1997).12

6

Hans-Martin Gärtner, Paul Law, and Joachim Sabel

A closely related variant of a group 3 analysis is defended by TRAVIS,13 who assimilates triggers to Α-bar binders in "clitic left-dislocation" (CLLD) structures (cf. Cinque 1990). TRAVIS reanalyzes trigger morphology as "verbal clitics" heading functional projections the specifier of which is occupied by an empty pronominal DP, i.e. pro.14 This is closely related to assumptions made by Sells (2000), who pointed out that empty trigger-related pro in a language like Tagalog may have overt counterpart clitics like ya in Kapampangan, as shown in (5).15 (5)

King=tindahan ya sinali ng=mangga ing=lalaki. OBL=store 3SG.N bought GEN=mango Ί - m a n 'At the store the man bought a mango.'

Now, crucially, TRAVIS takes the relation between the trigger and pro to be a binding relation like the one involved in CLLD and the relation between the trigger and its sister constituent containing pro as an instance of predication. This is part of a larger theoretical move toward eliminating all NP/DP-movement from verb initial Austronesian languages (see below). As noted by TRAVIS herself, more work is needed to account for the general constraint against multiple triggers. Appeal to the formal tools licensing CLLD alone does not suffice, given that "[i]n CLLD there is no (theoretical) limit to the number of "left-dislocated" phrases" (Cinque 1990: 58). The same issue arises wrt overt clitic constructions in Kapampangan, which in addition to the trigger can license a topicalized actor, as shown in (6) (Sells 2000: 124).16 (6)

ing=lalaki sell na NOM=man bought 3.SG.A 'The man bought the mango.'

ya 3.SG.N

ing=mangga. T=mango

1.2. Head initiality With this background on voice phenomena, 17 let us now turn to the second major issue of Austronesian syntax, namely, head initial structures. Thus, in terms of default word order, many Austronesian languages must be described as verb - or predicate initial. This is illustrated for Toba Batak in (7a) (Wouk 1986: 395) and Makassar in (7b) (Jukes 2005: 662).

Clause Structure and Adjuncts in Austronesian Languages (7)

a. Manga-lompa AT. cook 'Ria is cooking b. Bambangi=i hot=3.ABS

dengke fish fish.' allo-a.

7

si Ria. ART Ria

day-DET

'The day is hot.' In Formosan and Oceanic languages predicates are often preceded by auxiliary-like "preverbs" or TMA-markers. This is shown for Atayal in (8a) (Aldridge 2004: 80) and Tongan in (8b) (Otsuka 2005: 69). (8)

a. Musa' ASP

-maku'

pma-n

hiya.'

lSG.GEN wash-PT 3SG.NOM

Ί am going to wash him.' b. Na'e tangi a' Sione. PST cry ABS Sione ' Sione cried.' The main controversy among syntacticians concerns the question as to whether verb / predicate initial structures are base generated, as argued for by adherents of group 1 such as Holmer (1996) for Seediq and Keenan (2000) for Malagasy, or derived. Among those who assume the latter, group 2 advocate leftward head (X°-) movement, as do GHT (1992) for Cebuano, Indonesian, Malagasy, and Tagalog as well as Otsuka (2005) for Tongan. By contrast, group 3 defend application of leftward phrasal (XP-) movement for languages like Malagasy, Niuean, Palauan, and Seediq (cf. Aldridge 2004, Massam 2000, 2001, Rackowski and Travis 2000, Pensalfini 1995, Pearson 2000). Now, a group 1 approach would simply regularize the GHT structure in (3) somewhat further and, as shown in (9) (see next page), generate all specifiers to the right (all heads being on the left already) (cf. Holmer 1996: 108). This puts all NP arguments on the right side of the verb and also properly respects their hierarchy for binding. 1° would be the appropriate site for the auxiliary-like elements encountered in (8). However, (9) wrongly predicts NPTH < NPA to be a default order in voices other than AT and THT, contrary to what is usually the case, as (10) from Malagasy shows (Rackowski and Travis 2000: 136) (see next page).

8

Hans-Martin Gärtner, Paul Law, and Joachim Sabel

(10) Nividian 'ny renin-dRakoto lamba aho. PST.CT.buy A mother-Rakoto clothes T.lSG Ί was bought clothes by Rakoto's mother.' By contrast, Keenan (2000) sacrifices a c-command based account of binding and allows arguments to be introduced into VP in different orders depending on trigger morphology. He provides a number of constituency tests, showing among other things that non-AT verbs can be taken to combine with actors directly18 and create constituents able to undergo coordination with V°+NPTH units. (11) exemplifies this phenomenon for Toba Batak, as discussed in Sternefeld (1995: 52). (11)

Diantuk

si John jala

manipak si Bob

NAT.hit A J. and AT.kick THB. 'Fred was hit by John and kicked Bob.'

si Fred. Τ F.

The group 2 alternative of GHT keeps the structure in (3) and assumes obligatory V°-to-I° movement. This properly captures default linear order and binding. For (11), GHT can appeal to I'-coordination. Example (12a) (Keenan 2000: 43) (see next page), however, shows that an additional landing site for objects hierarchically higher than I' but below the surface trigger position would be needed in the GHT-system. (12b) in addition illustrates that V' or a lower VP excluding the agent, which GHT assume, does not seem to provide a constituent ready for coordination (Keenan 2000: 43) (see next page).

Clause Structure and Adjuncts in Austronesian Languages (12)

9

a. Nividy sy namaky boky roa Rabe. PST.AT.buy and PST.AT.read book two T.Rabe 'Rabe bought and read two books.' b. *Nividianana Rabe ilay satroka sy/aryRakoto ny boky Rasoa PST.CT.buy A.R. that hat and A.R. THbook T.R. 'Rasoa was bought a hat for by Rabe and a book for by Rakoto.'

It must be counted as an additional weakness of the group 2 analysis, at least in its GHT variant, that the V°-to-I°-shift across NPA is motivated only highly theory-internally, i.e. dictated by the way VPs have to be built according to Baker's UTAH, the VP-internal subject hypothesis and X-bar theory. Little evidence has been provided to motivate it independently.19 TRAVIS even suggests that this kind of V°-movement across NPA may take place on a pre-syntactic level called L(exical)-syntax (cf. Hale and Keyser 2002). However, the following paradigm from Tagalog might actually be interpreted as the right kind of data in favor of V°-movement. 20 (13)

a. Maingat binasa ngbata ang libro. carefully THT.read A child Τ book 'The/A child read the book carefully.' b. Maingat ng bata binasa ang libro. c. Maingat bumasa ng libro ang bata. carefully AT:read TH book Τ child 'The child read a book carefully.' d. * Maingat ng libro bumasa ang bata.

(13d) indicates that shifting an NP across the finite verb is not a general option. Instead, (13a) and (13b) could instantiate the GHT structure with and without V°-to-I° respectively. Of course, the position of the adverb would have to be established (see below). Following Kayne's (1994) influential work on the "antisymmetry" of syntax, which reduces the possible orders of X-bar theory to specifier < head < complement orders exclusively, group 3 analyses regularize the GHT-system in the other direction, as shown in (14) (see next page). While this had been taken to be a parametric choice by GHT to distinguish e.g. Malagasy "VOS", (3), from Indonesian "SVO", (14), group 3 adherents assume parametrization to concern the presence or absence of the additional VP-movement step, shown in (15) (see next page).

10

Hans-Martin Gärtner, Paul Law, and Joachim Säbel

(14)

(15)

Obviously, the desired outcome, i.e. a verb initial structure, only results for actor trigger configurations. In other cases, some kind of group-2-style V°shift across NPA to some additional landing site still has to be assumed, at least in L-syntax, if one follows TRAVIS. Group 3 adherents tend to point out that (15) most naturally also captures predicate intial structures in verbless clauses like (7b). Therefore, group 3 analyses are often called "predicate fronting" analyses, such as the one presented for Niuean by Massam (2000). As far as word order is concerned, this point is not very compelling though, given that under the GHT analysis in (3), intransitive structures like (7b) would be trigger final, and thus predicate initial, as well. Now, given the close correspondence between predicate fronting and "VOS" patterns, one might expect a group 2 approach to "VSO" languages. However, group 3 proponents have suggested that predicate fronting is adequate in these cases as well.21 What is required in addition is that all the arguments vacate the predicate before it fronts as a so-called "remnant" category. This was originally proposed by den Besten & Webelhuth (1990) to account for structures like (16) in German involving

Clause Structure and Adjuncts in Austronesian Languages

11

contrastive topicalization of a participle and scrambling of the direct object das (16)

Buch. [vp ti gelesen

]j

hat

Hans

read has H. 'Hans didn't read the book.'

das

Buchj nicht

the

book not

tj

As evidenced by the contributions to this volume, the controversy over group 2 and group 3 approaches to Austronesian clause structure is still ongoing. TRAVIS raises the stakes and explores a formal parametrization according to which predicate fronting languages have no NP/DP movement at all. This yields an inverse to V°+DP-movement systems like English. Equally, KAHNEMUYIPOUR & MASSAM make full use of remnant movement in analyzing different word order patterns in the Niuean DP. THIERSCH provides a largely impartial assessment of the various syntactic mechanisms surrounding remnant movement and their application to the problems at hand.22 As discussed in detail there, among the tools to rein in the descriptive power of remnant movement is the constraint against extraction from specifiers, a.k.a. "freezing" (Culicover and Wexler 1977). Its application has been closely studied wrt the "trigger-only" constraint in Austronesian languages, to which we now turn.

1.3. Unbounded dependencies and the "trigger-only" constraint Austronesian languages are also famous for the fact that unbounded dependencies (i.e. relative and question formation) involving arguments usually disallow extraction of anything other than the trigger. This has often been interpreted as a "subject-only" restriction, which clashes with typical conceptions of locality built to account for the extractability of direct objects in languages like English and German.23 Although descriptively the Austronesian pattern nicely fills the extreme (most restrictive) position in Keenan & Comrie's (1977) accessibility hierarchy, formalizing the constraint any further has not been without problems. In line with what has been indicated about freezing in section 1.2, various researchers have sought to exploit configuration (15) for an explanation. Thus, as observed by Pensalfini (1995) (cf. Rackowski and Travis 2000: 124-125), if specifiers are islands for extraction, VP movement "freezes" everything but the trigger in its place inside VP. However,

12

Hans-Martin Gärtner, Paul Law, and Joachim Sabel

as recently discussed in detail by Chung (2005), among the "VOS" languages only Seediq clearly confirms this prediction for VP. For "VSO" patterns like the ones allowed in Tagalog and Chamorro, Ο may be outside of VP as well. Consider the following word order permutations from Schachter & Otanes (1972: 83). (17)

a. Nagbigay AT.gave 'The man b. Nagbigay c. Nagbigay d. Nagbigay e. Nagbigay f. Nagbigay

nglibro sa babae ang TH book D woman Τ gave the woman a book.' ng libro ang lalaki sa babae. sa babae ng libro ang lalaki. sa babae ang lalaki ng libro. ang lalaki ng libro sa babae. ang lalaki sa babae ng libro.

lalaki. man

In the GHT-system ang lalaki is in Spec,VP in (18e)/(18f) and in Spec,IP otherwise. This requires an additional position to the right of Spec,IP, and thus outside of VP, for sa babae and ng libro in (18b) and (18d) respectively. Nevertheless, non-trigger arguments apparently cannot use such a position to circumvent the freezing condition, i.e. the "trigger-only" constraint still applies, as is shown in (18) (Kroeger 1993: 211). (18)

a. Sino ang nagnakaw ng kotse mo? who Τ PERF.AT.steal TH car your 'Who stole your car?' b. * Sino ang ninakaw ang kotse mo? who τ PERF.THT.steal τ car your

Recipients like sa babae can undergo "adjunct fronting" though (Kroeger 1993: 44), an issue we return to in section 2. At the same time, it is also well known that only trigger clauses are transparent for long extraction (see e.g. Kroeger 1993: 215-217). But, clearly, such a clause would itself be in a specifier (Spec,IP) and is thus incorrectly predicted to be an island for extraction within the freezing account. 24 A second approach emerges from the group 3 analysis of the trigger as Α-bar operator (Richards 2000, Pearson 2005). Given the observation that in Germanic V2 languages, w/z-movement and fronting of other constituents into Spec,CP are mutually exclusive, it can be assumed for the

Clause Structure and Adjuncts in Austronesian Languages

13

putative Austronesian counterpart too that trigger choice and choice of whor rel-extractee must coincide. Technically, this uniqueness is usually linked to the assumption that CP possesses only a single specifier, an assumption that has been discarded in Split-CP frameworks like Rizzi (1997) or approaches that allow multiple specifiers (Chomsky 1995). As mentioned above, following the earlier w/z-agreement analyses of Chung (1994) and Georgopoulos (1985), unique specifier-head agreement can be required for licensing the desired trigger-only configuration in the Austronesian case. Extending this account to the transparency of trigger clauses is not as natural though. There are two approaches that capture both configurations in terms of at least intuitively symmetrical assumptions. First, Nakamura (1998) applies a minimal link metric to extractions. This essentially works because links from and out of a trigger in Spec,IP are bound to be shorter than those from more deeply embedded constituents, as long as the trigger is structurally most prominent. Secondly, Sabel (2002, in preparations), building on Keenan's variant of a group 1 approach to head intiality, suggests that triggers are base generated as sisters of V°. It is then possible to apply the classical means of locality theory developed in Chomsky (1986), i.e. Lmarking under local government, to account for extractability and transparency of triggers. It is interesting to note that the latter analysis assimilates the trigger to direct objects and is thus closer to the ergative approach to Austronesian voice systems discussed in section 1.1. There are a couple of challenges to these approaches too, as might be expected. First, the trigger-only restriction must be parametrized, as pointed out by Klamer (2002) citing among others work on Indonesian (Cole and Hermon 1998) and Javanese. Next, there are triggerless constructions, such as the "recent past" construction in Tagalog, which seem to allow extraction of any NP. As long as NPs are hierarchically ordered inside VP, this is not automatically predicted in terms of minimal links.25 Finally, there are additional complications involving adjunct extraction. These will be dealt with in the following section.

2. Issues in the syntax of adjuncts and the view from Austronesian As has been indicated earlier, it can be argued that progress on the three major issues of Austronesian syntax demands that more attention be paid to the syntax of adjuncts. This is in spite of the fact that adjuncts come in

14

Hans-Martin Gärtner, Paul Law, and Joachim Sabel

such an awe-inspiring number and formal as well as semantic variety. 26 Let us begin by mentioning a potential embarrassment, which, at second glance may turn out to be of particular further interest. Thus, one of the main exponents of adjuncts, namely, adverbs, seems to be scarcely represented in many Austronesian languages. By this we don't mean the familiar difficulty of establishing a morphosyntactic category "adverb" as distinct from the category "adjective." Although this is an issue as well, e.g. for the analysis of Tagalog, it is more problematic to find that Formosan languages express a lot of meanings that are expressed by adverbs in Indoeuropean languages in terms of auxiliary-like preverbs. Indeed, this seems to be a familiar feature also of Oceanic languages, where "[...] verb phrase modification is often expressed by verb serialization [...] rather than adverbially" (Lynch, Ross, and Crowley 2002a: 87) ,27 This is precisely the line of analysis pursued by CHANG for the Formosan language Kavalan. 28 Crucial evidence for the verbal nature of "pseudo-adverbs", as we are going to call these items henceforth, comes from trigger morphology. In fact, an item like slowly (do) in Kavalan functions like a finite auxiliary in English in that it bears the "essential" morphology, while the semantically main verb goes into default AT form (CHANG). (19)

a. Paqanas=iku t(em)ayta slow. AT^l SG.T see(AT) Ί read a book slowly.' b. Paqanas-an~ku t(em)ayta slow-THT=lSG.A see(AT) Ί read the book slowly.'

tu sulal. OBL book ya sulal. Τ book

Tsou differs from this pattern in that the inflected pseudo-adverbs simply agree with the inflection of the main verb (cf. Szakos 1994). CHANG suggests that the formal tool to be explored wrt to Kavalan is complex predicate formation. 29 HOLMER's study of Seediq, dealing with similar facts, can be taken to be complementary in that additional efforts are made to account for the particular placement of pseudo-adverbs (see below).

2.1. Adjunct placement as formal classification and structural indicator In order to be able to talk about the placement of adjuncts in more theoryneutral terms it is useful to give a rough chart of their whereabouts. In this

Clause Structure and Adjuncts in Austronesian Languages

15

we follow KAUFMAN'S seminal study of adverb orders in Tagalog as well as the classificatory strategy pursued by Bonami, Godard & KampersManhe (2004). We thus assume four adjunct zones distributed evenly around the main verb, as illustrated in (20). (20)

[...α...β...\...γ...

δ...]

Two of these zones are preverbal. a signals a domain of fronted adjuncts whereas β signals a zone closely in front of the verb, γ and δ are supposed to provide mirror-image counterparts in the postverbal domain. One can think of a as roughly comprising the target domains for Tagalog inversion operations, ay-inversion and "emphatic inversion" (cf. Schachter and Otanes 1972) as well as Malagasy t//a-topicalization. β can be taken to host the Formosan pseudo-adverbs. The left edge of β would thus be constituted by the highest auxiliary, TMA-marker, or negation. 30 For Tagalog it is possible to map two well known formal adjunct classes onto a and β . Thus, those adverbs that combine with V° through a linking morpheme na/ng can be taken to dwell in β while those preverbal ones that disallow the linker are in a. Degree adverbs like bahagya ('slightly') belong to the former while temporal adverbs like kahapon ('yesterday') belong to the latter class as shown in (21) (Wegmüller 1998: 196-197). (21)

a. Bahagya-ng nagbago ang ministro slightly-LK PERF.AT.change Τ minister ng kaniyang palagay. TH his.LK opinion 'The minister changed his opinion slightly.' b. Kahapon niya sinulat ang liham kay yesterday 3SG.A PERF.THT.write Τ letter D 'Yesterday, she wrote the letter to Maria.'

Maria. M.

Both bahagya and kahapon can attract clitics. For the latter this is actually shown in (21b), where niya intervenes between kahapon and the finite verb. In the former case there would still be a linker, but it would follow the clitic instead and keep immediately preceding the verb. (23a) below is an example, a will eventually have to be split up in order to allow for fronted adjuncts that don't attract clitics (see below). Crucially, in Tagalog a mirror image of (21) can be mapped onto γ and δ. Thus, close to the verb

16

Hans-Martin Gärtner, Paul Law, and Joachim Sabel

there can be adverbs like mabilis ('fast') which require a linker on their left, while kahapon could drift further rightward, again without a linker. What is different though is that γ can be rather freely interspersed with NP arguments and it is still unclear what would count as its right boundary. For Malagasy, the boundary can, with some hesitation, be taken to be the trigger position. At this stage it may be useful to briefly consider a Greenbergian perspective on our rough sketch of adjunct zones. Head initiality is usually taken to imply that nothing but TMA-marking particles and auxiliaries precedes the main verb clause-internally (cf. Carnie and Guilfoyle 2000: 10). Dryer's (1992) survey, which accommodates a considerable number of Austronesian languages, is compatible with this prediction and we can interpret the Formosan pseudo-adverbs in zone β to confirm such a view as well. Thus, in Tsou only clitics like cu ('already') may enter inside β, while preverbal conjunctive, evaluative, and temporal adjuncts must be treated as peripheral elements in a preceding the highest auxiliary. This is shown for 'oc 'ocic 'o ('fortunately') in (22). (22)

a. Oc'ocic'o i-ta atavey-a fortunately NAT.REA-3SG finally-NAT mevcongu ta pasuya 'e paicu. marry.Ν AT OBL P. TP. 'Fortunately, Paicu finally got married to Pasuya.' b. ?I-ta 'oc'ocic'o atavey-a NAT.REA-3SG fortunately

mevcongu

ta

m a r r y .NAT

OBL

c. ?I-ta

'e paicu.

P.

Τ

atavey-a

NAT.REA-3SG

finally-NAT

pasuya

finally-NAT

P.

'oc'ocic'o fortunately

mevcongu

ta

pasuya

'e paicu.

marry.NAT

OBL

P.

TP.

That (22b)/(22c) aren't fully unacceptable is presumably due to an alternative parenthetical construal of the evaluative adverb. Dryer (1992: 93) further finds a positive correlation between VO-orders and postverbal manner adverbs. For Tagalog, this is immediately challenged by the alternation in (23) (Wegmüller 1998: 197), where we find maganda ('beautifully') in β or γ.

Clause Structure and Adjuncts in Austronesian Languages (23)

a. Maganda beautifully

siya-ng 3SG.T-LK

kumakanta. AT.INCOMPL.sing

'He is singing beautifully.' b. Kumakanta siya-ng AT.INCOMPL.sing

17

3SG.T-LK

maganda beautifully

One way of reconciling (23) with the Greenbergian picture would be to analyze maganda as an adjective and kumakanta as nominal / participial, given "that (noun,adjective) is not a correlation pair" (Dryer 1992: 96).31 The chart in (20) also sketchily represents an important fact about Tagalog, confirmed by KAUFMAN. (24)

"Generally, greater distance from V is interpreted as wider scope" (Ernst 2002: 17)

Like in English this can be interpreted as saying that V° is in or close to its base position. (24) has been shown to hold for Malagasy too by one of the by now most influential studies of Austronesian adverb placement, namely, by Rackowski & Travis (2000) (henceforth R&T). 32 Thus, where Cinque (1999) has diagnosed the continuously descending pattern of adverb types in (25a), ranging from speech act markers [1] via aspectual adverbs like already [4] to manner adverbs like well [10] for Italian, R&T (2000: 121) have found the closely corresponding sequence in (25b) for Malagasy, with items like matetika ('generally') [2], efa ('already') [4], foana ('always') [8], and tsara ('well')[10] (cf. THIERSCH). (25)

a. 1 < 2 < 3 < 4 < 5 < 6 < 7 < 8 < 9 < 1 0 < (V°) b. 2 < 3 < 4 < 5 < (3) V° 10 < 9 < 8 English: SUBJ: [Mod Argl] PRED: Action, Arg2 -> Seediq: PRED: Mod SUBJ: [Argl [REL Action Arg2]]

In analogy, applying the same analysis to the manner adverbials, we would get a situation such as that in (22). (22)

Underlying representation: Argl Arg2 Action -> English: SUBJ: Argl PRED: Action, Arg2, Manner -> Seediq: PRED: Manner SUBJ: Argl, Arg2, Action

Manner

One natural consequence of this analysis would be that, assuming that extraction universally takes place leftwards, a language with subject extraction would automatically be subject-initial, whereas a language with predicate extraction would automatically be verb-initial. However, while the situation appears to be exactly as in (21) for clauses where certain quantifiers and modifiers of NPs are realized as predicates, this analysis does not carry over in its entirety to adverbial main predicates as in (22). Whereas the remainder of the clause is beyond doubt a grammatical subject in (21) it cannot be a subject in (22).26 Instead, it would seem to be some kind of clausal complement. What the reason is for this difference is difficult to determine, and while the parallels may have a diachronic origin which could be linked with verb-initial order, it is not clear that they are valid synchronically.27 The reader will note the parallels between this analysis and the received antisymmetric analysis of VOS order (cf. e.g. Pearson 1998, Massam 2000), designed to accommodate VOS order to Kaynes's (1994) Antisymmetry hypothesis. Under this analysis, VOS order is derived by

Adverbial Heads in a Formosan language

99

predicate raising of VP to SpecTP (23a), whereas subject-initial order is derived by raising the subject to SpecTP (23b). (23)

TP

a.

SUBJj

T'

VPi OBJ

TP

b.

Τ

vP

vP

SUB J

ti

VP V

OBJ

However, we should also note the synchronic incompatibility of the two views: the raising mechanism which selects an adverbial predicate from the proposition would be the same mechanism which creates V-initial order. Nevertheless, adverbial predicate constructions and VOS order are not in complementary distribution. Thus, adopting the analysis shown above would require us to derive VOS order by means of a different mechanism. It would be preferable to find a solution which addresses word order and the behavior of adverbial predicates at the same time. In the following section, we shall outline an alternative analysis which can allow us to relate the facts in Seediq with the Cinque adverbial hierarchy in a manner compatible with a predicate raising account of VOS word order.

3.2. Adverbs as functional heads Given Cinque's analysis of adverbs as functional projections above vP, the behavior of Seediq adverbs is not only expected: it also fills what appears to be an unexpected gap in the typology of adverb realization. This is clear when we compare adverbs with the category of negation. It is generally accepted today that negation is located in a functional projection NegP above vP. It is also generally accepted that the negation itself can either be a realization of the head Neg° or of a phrasal category in SpecNegP, or, indeed both. That the negation nicht is phrasal in German seems clear, in that it does not block V->C movement (24a,b) and in that it can be replaced by a phrase with no concomitant change in syntactic properties (24c).28

100

Arthur Holmer

(24)

a. . . . dass er den Lehrer nicht gesehen hat. that he the teacher NEG seen has ' . . .that he didn't see the teacher.' b. Er hat den Lehrer nicht gesehen. he has the teacher not seen 'He didn't see the teacher.' c. Er hat den Lehrer [ nie im Leben ] gesehen. he has the teacher never in.the life seen 'He has never ever seen the teacher.'

In contrast, the Finnish negation verb is clearly a head (25a, b), which bears agreement morphology which would otherwise have been realized on the verb (25c, d).29 This morphology can never be realized on both (25e). Further, phrasal negations such as never are rendered in Finnish by combinations of the negation verb with adverbial negative polarity items (25 f). (25)

a. Mind e-n puhu suome-a. lSG NEG-lSG speak.CNEG Finnish-PART Ί don't speak Finnish.' b. Sind e-t puhu suome-a. 2SG NEG-2SG speak.CNEG Finnish-PART 'You don't speak Finnish.' c. Mind puhu-n suome-a. lSG speak-lSG Finnish-PART Ί speak Finnish.' d. Sind puhu-t suome-a. 2SG speak-2SG Finnish-PART 'You speak Finnish.' e. *Mind e-n puhu-n suome-a. lSG NEG-lSG speak-lSG Finnish-PART f. Mind e-n koskaan puhu suome-a. lSG NEG-lSG ever speak.CNEG Finnish-PART Ί never speak Finnish.'

If we suppose that adverbs are a functional category, we would expect, in analogy with negation, that both the head and the specifier realization should be attested in the world's languages, and, seen in this light, the

Adverbial Heads in a Formosan language

101

Formosan data represents a clear instance of the former option, corresponding to the Finnish realization of negation.

4. Adverbial heads and adverb ordering 4.1. Background This is not the first analysis which makes use of the concept of adverbial heads in a Cinque model. Although the present discussion takes an earlier proposal (Rackowski and Travis 2000; henceforth R&T) as its starting point, the conclusions we will reach are quite different. To make this point, it is first necessary to review briefly the central points of the R&T analysis. R&T represents an attempt to apply Cinque's (1999) adverb hierarchy and Kayne's (1994) Antisymmetry to the system of Malagasy adverbs. According to the Cinque analysis, which is based on a large and both genetically and typologically heterogeneous sample of languages, adverbs of various kinds are ranked in a universal hierarchy of scope. In typical SVO languages, this ranking is directly reflected in linear ordering. A subset of this hierarchy, quoted from R&T (2000: 121) is shown in (26). (26) speech act 1 general

completely 9 well 10

vP

Given Antisymmetry, the hierarchical ordering suggested by Cinque should, all things being equal, be mirrored by a universal linear ordering. The facts in the VOS language Malagasy, which is the topic of R&T, are unexpected in this light. Following the numbering used in (26), the linear order found in Malagasy is that in (27).

102 (27)

Arthur Holmer 2 - 3 - 4 - 5 - VERB -10 - 9 - 8 - 7 - 6 - 1 - SUBJ

Basically, the data in Malagasy conforms linearly to the Cinque hierarchy for preverbal adverbs, whereas the reverse order holds for all postverbal adverbs. More or less the same order is found in Seediq, too. Again following the same numbering, examples (28a-e) illustrate adverb ordering in Seediq.30 (28)

a. ye=su INT=2SG

ini

ekan

hlama?

NEG

eat.AF.CNEG hlama

1 3 V 'Don't you eat hlama [steamed rice with honey]?' b. M-uuyas ruru kiya klaali heya. AF-sing

stream

there always

3SG.NOM

V 8 S 'It (the frog) always sings by the stream.' c. ini=daha ba mhmet-i NEG=3PL.GEN

really

3 m-angal so

6 10 lukus bege=daha.

AF-take

cloth give=3PL.GEN

like

at.random-PF.CNEG

V S '(They) don't just take clothes and such which they (people suffering from gout) have given away.'31 d. Ini=ku k-qeni32 na. NEG=lSG.NOM CNEG-thirsty

yet

3

5

V

Ί am not thirsty (yet).'

e. m-usa AF-go

m-ekan

seedaq kiya gaga cghunn

kiya

AF-eat

person that be

there PRF QUOT

hang

di

sa.

V 4 1 'they (the crows) go and eat the person that has hanged himself, so it is said' The general validity of this order can be further confirmed by a sample of ungrammatical examples with different orders.33

Adverbial Heads in a Formosan language (29)

a. *Ini=su

ye

ekan /m-ekan

103

hlama?

NEG=2SG.NOM INT AF.CNEG.eat / AF-eat hlama 3 I V b. * Wada34 Hori sa di. went Puli QUOT PERF V 1 4 c. * t r m e x - u n = d a h a ini ekan ucikqaun. on.its.own-PF=3PL.GEN NEG AF.CNEG.eat chili 10 3 V

The ordering in Seediq can be summarized as in (30), still referring to the numbering in (26). The data is not entirely comparable to that given for Malagasy in R&T, but the basic picture is still the same as in Malagasy: preverbal adverbs appear in Cinque order, and postverbal adverbs appear in reverse Cinque order. One noticeable difference between Seediq and Malagasy is that all postverbal adverbs are presubject in Malagasy, whereas some of them are postsubject in Seediq, the reverse Cinque ordering still being the same as in Malagasy, however. (30)

1 -3-6-10-VERB-8-SUBJ-5/4-1

How is this reversed order to be derived? 35 R&T suggest an account based on a generalized system of predicate raising termed intraposition, where a large portion of the structure surfaces in reversed linear order to the right of VP (31) (see next page). An obvious question at this stage would concern the motivation for movement - R&T's suggestion is that predicate raising is not a feature of a single head (such as T) but rather of every level which is in some sense an expansion of VP (e.g. AspP, TP, various AdvP's). It might in fact be argued that it extends to the clause as a whole. R&T (2000: 130) and Travis (2004, this volume) argue quite plausibly, following Pearson (1998), that the major typological distinction should be that between argument-raising languages and predicate-raising languages, and that the mechanisms involved in movement in the two language types are inherently different. In a predicate raising language, at every relevant level, a head X will trigger raising of its complement to SpecXP (32) (see next page). 36

104

Arthur Holmer

Adv4P

(31)

Adv4°

Adv5P 5 Adv5°

Adv6P Adv6'

Adv7Pn Adv8Pi Adv9P k

VPj

Adv8' Adv7° \ 7 Adv8° tk 8

Adv9'

AdvlOP AdvlO'

Δ

Adv9°

Adv7' Adv6° 6

tm

t,

tj

9 AdvlO° 10

ti

XP

(32)

X'

YP; ZPj

γ,

X

o

Δ Generalized predicate raising is not restricted in its function to adverb ordering. Regardless of how we wish to derive adverb order, VOS languages are, on the surface, extremely head-initial in other ways as well. In fact, as Aldridge (2002) and Holmer (2005) have shown, recursive instances of predicate raising are required simply to allow for the order of the arguments, verbal heads and post-subject particles.37 Viewed in this light, the predicate raising mechanism as such is not a problem. Rather, the more serious question would instead seem to be what

Adverbial Heads in a Formosan language

105

prevents the process from continuing throughout the whole structure: why are not all Malagasy adverbs postverbal with reverse Cinque order? The predicate raising mechanism illustrated in (31) and (32) operates around heads, and this leads R&T (2000: 122) to suggest that preverbal adverbs are not heads, but are phrasal, and are located in the specifier positions themselves. The crucial consequence of this is that the specifier position is blocked, thus effectively preventing further predicate raising. Given that the entire analysis crucially rests on the assumption that postverbal adverbial elements are heads and preverbal adverbial elements are phrases (or at any rate located in specifier positions), it would be an advantage if some independent evidence for the X / XP status of the elements could be unearthed. Such evidence is hard to come by in Malagasy. However, as we have suggested in section 2.2, Seediq is a language where morphosyntactic evidence for the headhood of adverbials is readily forthcoming. The problem at this stage should be obvious: the only adverbs for which we have morphosyntacic evidence of head status are preverbal, not postverbal, in other words exactly the opposite of what would be expected given R&T's suggestion.

4.2. The status of Seediq adverbs We have noted that the order of adverbs in Malagasy and Seediq is quite similar (cf. examples 27-30). The only major difference between Seediq and Malagasy could be expressed in intraposition terms such that the raised predicate in Malagasy never contains the subject, whereas in Seediq the subject may be included in the raised structure (perhaps making the term "predicate raising" a misnomer, but with no crucial difference in the mechanism involved). Given these similarities, there is no a priori reason to assume that adverb ordering is derived differently in Seediq and Malagasy - rather, we would expect Rackowski and Travis' analysis to carry over directly to Seediq. Since it does not, an alternative analysis accounting for the presence vs. absence of predicate raising must be found, at least for Seediq, and having found such a solution, we would instead expect that it carries over to Malagasy. 38 So far we have only examined manner adverbs in detail. An interesting problem in this context remains how to deal with other types of adverbs. Here there seems to be a certain amount of variation. One clear instance of

106

Arthur Holmer

a head is the negator ini, which triggers connegative morphology (33a).39 At the same time, at least one preverbal element is clearly not a head according to the criteria mentioned in section 2.1, namely ba 'indeed', given that it can intervene linearly between the negator ini and the main verb without blocking the assignation of connegative morphology (33b). (33)

a. Ini=mu qta-i ka quyu kiya. NEG=3SG.GEN see-PF.CNEGNOM snake that Ί didn't see that snake.' b. Kiya ini=daha ba tleng-i thus NEG=3PL.GEN indeed touch-PF.CNEG ka seedaq m-n-cghuun. NOM person AF-PST-hang 'they certainly don't touch people who have hanged themselves.'

If we choose to view clitic attraction as a head characteristic (cf. section 5), this does not really change the picture. One element with an adverbial meaning, tena 'already' always attracts clitics (34a)40, whereas another, ncuin 'sometimes' is subject to idiolectal variation (34b, c). (34)

a. Tena=ku m-n-ekan ido. already=lSG.NOM AF-PST-eat rice 'I've already eaten.' b. Ncuin sa-an=daha m-ita, sa-adis=daha timu. sometimes go-LF=3PL.GEN AF-see lF-bring=3PL.GEN salt 'Sometimes they go and see it (the cow), and bring it some salt.' c. Ncuin=ku m-usa Taihoku. sometimes=lSG.NOM AF-go Taipei 'Sometimes I go to Taipei.'

Clearly, then, preverbal position is not associated exclusively with either heads or non-heads. For postverbal elements with adverbial meanings, the situation is less clear. We have no direct morphosyntactic evidence for or against head status for any of the postverbal adverbs in Seediq. It may be argued that a phrasal element can presumably be expanded in a more complex string, something which is normally not the case for a head (this contrast holds for German nicht41 and English not, which can be independently shown to be a phrase, and a head, respectively). This kind of test is difficult to apply, but would probably place final particles among the

Adverbial Heads in a Formosan language

107

heads - final particles cannot be expanded in any way or replaced by expanded phrases of similar meaning. Further, final particles generally express meanings corresponding to functional heads such as C or Asp: this is unexpected behavior for phrasal elements. For these reasons, it is argued in Holmer (2005) that final particles are to be analyzed as heads. Another possible test for head status would be movement: an element which can appear dislocated in a phrasal position is phrasal in nature, whereas it is a head if it can appear dislocated in a head position. If an element cannot appear dislocated, the evidence is inconclusive, as is the case for final particles, which can never appear in any other position in the clause. Certain elements can move around, however. The adverb ncuin 'sometimes', for which we noted idiolectal variation with respect to head status (cf. examples (34b, c) above) can also occur postverbally (35a). More importantly, the adverb klaali 'always', which usually occurs postverbally (35b), can also be fronted to a position preceding the host of cliticization (35c). This initial position is only open to phrasal categories, but not to any elements which can be independently recognized as heads. 42 (35)

a. M-usa=ku Taihoku ncuin. AF-go=l SG.NOM Taipei sometimes Ί sometimes go to Taipei.' b. m-narux—ku tunux klaali. AF-ill=l SG.NOM head always Ί always have headaches.' b. Klaali m-usa=ku Taihoku ali always AF-go=l SG.NOM Taipei day Ί usually go to Taipei on weekends.'

gdreg-an. rest-LF

Thus, to summarize our conclusions so far, elements which in Seediq correspond to adverbs may be either heads or phrases, regardless of whether they occur preverbally or postverbally. All combinations are attested. It follows that the X / XP difference is nothing which we can harness fruitfully to derive preverbal or postverbal linear order. Rather, it appears to be the case that those adverbial elements which have a reasonably fixed position in Seediq are those which are unequivocally heads, whereas those which can be dislocated are phrasal. This implies that the crucial ordering problem we are dealing with here (that distinguishing preverbal from post-

108

Arthur Holmer

verbal adverbs with fixed positions) only touches on the properties of the heads themselves.

5. The curious behavior of Seediq heads The pattern we have seen so far suggests that we are dealing primarily with two kinds of heads: postverbal heads which are involved in predicate raising of the type proposed by R&T, and preverbal heads which are not. What is it that causes these differences? In Malagasy, preverbal adverbs are generally (with one exception) structurally higher than postverbal adverbs, suggesting an analysis where some feature at a given level in the structure may block further predicate raising in the rest of the clause. Unfortunately, this is not the case in Seediq, where final particles often represent more or less the same type of syntactic category as preverbal elements: there are both final particles and preverbal elements with tense / aspect connotations, and there are both clause-initial and clause-final subordinators. It would be completely impossible to group these in any way so that final particles are uniformly either above or below preverbal elements, barring splitting both INFL and C and jumbling the components of both domains together. Such a procedure would be syntactically unmotivated. Therefore, our solution must somehow reflect a property of each individual head. Some heads are capable of triggering predicate raising, and some are not. It is this difference which we must capture. To do this, we must first look at another property of Seediq heads: head movement.

5.1. Head movement in Seediq Under the assumption that postverbal adverbs (including final particles) are heads, another problem arises, namely how it can be that these heads, if they are heads, do not block head movement according to the Head Movement Constraint (Travis 1984). Evidence for head movement (V->C) in Seediq derives from the behavior of the clitic pronouns in Seediq. Clause-level pronominal clitics attach to the highest head in the clause. This can be a subordinator (36a) or an interrogative particle (36b), but it can also be a T/A-marker (36c), a negator (36d), a manner adverb (36e) or the main verb (36f).

Adverbial Heads in a Formosan language

(36)

109

a. Ani—ku naq m-enaq Taihoku, even.if=lSG.NOM just AF-stay Taipei ini=ku kela kari mukan uri. NEG=lSG.NOM AF.CNEG.know language Taiwanese also 'Even though I live in Taipei, I don't know Taiwanese.' b. Ye=su m-n-imah sino ciga? INT=2SG.NOM

AF-PST-drink

wine yesterday

'Did you drink wine yesterday?' c. Wada=mu qta-un ka huling=su. PST^lSG.GEN see-PF Ί saw your dog.'

d. Ini=ku

NOM dog=2SG.GEN

kela

rengo

NEG=lSG.NOM know.AF.CNEG speak

kari

seediq.

language person

Ί can't speak Seediq.' e. skret-an=daha m-ekuy quwaq salo. tight-LF=3PL.GEN AF-tie mouth pot 'They tie the mouth of the pot tightly.' f. M-n-ekan=ku ido ciga. AF-PST-eat=lSG.NOM rice Ί ate rice yesterday.'

yesterday

It might be argued that these clitics are simply Wackernagel clitics or 2nd position clitics, a case of phonological cliticization to the first word in the clausal domain, as appears to be the case in Tagalog. However, this is clearly not the case in Seediq, where cliticization is a syntactic process which is sensitive to the status of the host. Thus, cliticization may not take place to conjunctions (37a, b), nor may it take place to w/z-phrases (37c). On the other hand, a wÄ-word may attract clitics if and only if it is a syntactic head, e.g. the w/z-verb hmuwa / hwaun 'to do what' (37d). Further, no phrasal adverbials of any kind may attract clitics (37e). It is crucial to note that, in contemporary Seediq, the phrasal adverb ini huwa 'it is OK' is not clause-external, as evidenced from the fact that the clitic can climb past it to an interrogative particle (37f)43. (37)

a Anisa=(*ku) but=(* 1 SG.NOM)

ini=ku

ekan

tmakn.

NEG=lSG.NOM eat.AF.CNEGtobacco

'But I don't smoke.'

110

Arthur Holmer b. ... ma=(*mu)

ini=mu

qta-i

a n d = ( * l SG.GEN) NEG=lSG.GEN '...and I d i d n ' t see that snake.'

c. Ima=(*su) who=(*2SG.GEN)

kiya.

see-PF.CNEG snake that

qta-an=su

ciga?

see-LF=2SG.GEN

yesterday

'Who did you see yesterday?' d. Hwa-un=ta seediq so nii do.what-PF=l PL.INCL person like this 'What shall we do with a thief like this?' e. Ini=(*su) huwa=(*su) NEG=(*2SG)

quyu

me-eguy? AF-steal

do.what.AF.CNEG=(*2SG)

m-ekan=su

tmaku

hini.

AF-eat=2SG.NOM

tobacco

here

'It's OK if you smoke here.' f. Ye=ku ini huwa m-ekan INT=lSG.NOM NEG do.what.AF.CNEG AF-eat 'Is it OK if I smoke here?'

tmaku hini? tobacco here

If cliticization in Seediq is syntactic rather than phonological, in that it is sensitive to the X°/XP status of its host, the null hypothesis must be that the clitics attach to a single discrete head rather than to the linearly first head position which happens to contain phonological material.44 Since the clitics can be realized on a subordinator, it is natural to assume that this discrete head must be C°, and that head movement ensures that a C° which is not filled by a subordinator is lexically filled by movement instead (paralleling Holmberg and Platzack's (1995) analysis of V2 languages).45 A minimal pair in this respect is given in (38). (38)

a. Netun=su m-imah=(*su) sino, bsukan=su dhenu. if=2SG.NOM AF-drink wine drunk=2SG.NOM consequently 'If you drink wine you will get drunk.' b. M-imah=su sino. AF-drink=2SG.NOM wine 'You drink wine.'

Other evidence of head movement is hard to come by, given the extremely head-initial nature of the language (there are no intervening specifier positions across which head movement can take place visibly). However, one possible case might be the behavior of ba 'indeed'. In (39a), ba is

Adverbial Heads in a Formosan language

111

located to the left of the manner adverb, whereas in (39b) it is located to the right of the manner adverb. These two examples can be captured under a head movement analysis assuming that ba is not a head and that the adverb in (39b) moves past it because the position to the left of ba is not occupied by any other element. The same would hold for (39c) with the predicative adjective cheya moving past ba. However, it should be noted that the positioning could also be explained by merging ba, for scope reasons, in different positions in (39a), (39b) and (39c). (39)

a. ini

ba

mhmet-i

m-angal pala

NEG really at.random-PF.CNEG AF-take 'they certainly d o n ' t just take clothes . . .'

...

cloth

b. bleq-un=daha ba snetun ma rees-un=daha. well-PF=3PL.GEN really follow and bury-PF=3PL.GEN 'they observe (the law) meticulously and bury (them).' c. Cheya ba puq-un46 trabus. tasty really eat-PF peanuts 'Peanuts are really tasty.' Another piece of evidence for head movement, discussed by Aldridge (2004: 199-201) concerns the unification of V with T, which is possible in affirmative contexts (40a), but impossible in negated contexts (40b, c), where instead a periphrastic construction must be used (40d).47 According to Aldridge (2004: 199-201) this is because the presence of the negation blocks movement from V to T. (40)

Pawan. a. M-n-ekanv tv ido ka AF-PST-eat rice NOM Pawan 'Pawan ate rice.' ido ka Pawan. b. *Ini m-n-ekan NEG AF-PST-eat rice NOM Pawan c. *M-n-ekanv ini tv ido ka Pawan. rice NOM Pawan AF-PST-eat NEG ido ka Pawan. d. Wada ini ekan PST NEG ACT.CNEG.eat rice NOM Pawan 'Pawan did not eat rice.'

However, neither of the latter two points serve as evidence for movement high enough in the structure to raise the verb past the final particles, so the

112

Arthur Holmer

problem of blocking only arises if we assume that cliticization (or agreement) facts exemplified in (38) are evidence of verb movement to C°. Head movement of this type is not blocked or in any way affected by the presence of final particles. Assuming that both kinds of elements are heads, this is tantamount to a violation of Travis' (1984) Head Movement Constraint.

5.2. Two kinds of heads Apparent violations of the Head Movement Constraint are not rare among the world's languages. Possibly the most publicized phenomenon of this type is Long Head Movement (cf. Borsley, Rivero, and Stephens 1996 and related work). Carnie, Harley, and Pyatt (2000) discuss similar facts in Old Irish and suggest, following Borsley, Rivero, and Stephens (1996), that heads, in analogy with phrasal positions, can also be classified in terms of the A / A' distinction. Just as a filled A position does not block A'-movement of a phrase, neither does a filled A head position block movement to an A'-head position. The distinction between Α-heads and A'-heads cannot be applied for our purposes here.48 Nevertheless, it is suggested, following Holmer (2005), that the main intuition of the idea, namely that heads are not a homogeneous group, but rather that they can be divided into two classes, is directly applicable to account for the non-interaction of verb movement with final particles. According to this analysis, syntactic heads in Seediq belong to one of the two classes, which, for lack of a more suitable term at present, we can refer to as X-heads and Y-heads. The properties of the two types of heads are given in (41). (41) Xheads:

Y heads:

-preverbal - undergo / block head-raising - do not trigger PRED-raising -postverbal - irrelevant for head-raising - trigger PRED-raising

It is important to note that the properties of X- and Y-heads listed above are in no way coincidental. Let us first address the question of raising. In Holmer (2005) it is suggested that heads can either partake in head move-

Adverbial Heads in a Formosan language

113

ment or in predicate raising, but not both - each type of head has its distinctive raising pattern. A head which triggers predicate raising is completely invisible to head movement processes: it cannot itself headmove, it cannot be a landing-site for head movement, nor can it block head movement. A head which does any of these things cannot trigger predicate movement. This proposal is supported empirically in Seediq (cf. Holmer 2005). Hence we can automatically relate linear position with the behavior with respect to head movement. Under an antisymmetric analysis, all heads above VP are underlyingly preverbal. Y-heads become postverbal during the course of the derivation, precisely by virtue of their triggering predicate raising (i.e. forcing the constitutent which includes the verb to move leftwards past them). X-heads, for exactly the same reason, do not. It follows, assuming that each head type is only involved in one kind of movement, that preverbal heads must be visible for head-movement processes and postverbal heads must be invisible for head-movement processes. This correlation is exactly what we find in Seediq. In Holmer (2005) the distinction between X-heads and Y-heads was harnessed to account for the fact that final particles (Y-heads) do not block verb movement.49 However, one further automatic consequence of this distinction will be harnessed here. Heads which do not trigger predicate raising will, at the surface, appear in the underlying order. If they are adverbs, they will conform to the Cinque hierarchy. Heads which do trigger predicate raising will, if they are adverbs, manifest reverse Cinque order. If we follow R&T in analyzing postverbal adverbs as heads, and given the clear morphological evidence that preverbal adverbs in Seediq are also heads, the distinction between preverbal and postverbal heads can easily be reduced to a distinction between X-heads (preverbal) and Y-heads (postverbal). Final particles are simply a special instance of Y-heads, hierarchically so high in the structure that the raised predicate includes the position containing the grammatical subject, whereas postverbal (but presubject) adverbs are located lower in the structure (as we indeed would expect, given their relation to the Cinque hierarchy), so that the raised predicate does not include the subject.50 The proposed distinction thus allows us to apply the central points of R&T's predicate raising analysis to Seediq postverbal adverbs (especially final particles), and to reconcile relevant parts of this analysis with morphological evidence in Seediq of head status for preverbal adverbs. In the narrowest sense, it only claims to account for data in Seediq. However,

114

Arthur Holmer

as an adaptation of the R&T analysis it may be applicable in Malagasy as well, allowing both languages a unified treatment. This further application to Malagasy and other VOS languages is only possible under the assumption that the distinction between X-heads and Yheads can be maintained without morphological desinences.51 If this is the case, then the distinction between X-heads and Y-heads may be relevant for a much wider range of VOS languages than hitherto assumed (possibly even an absolute majority of existing VOS languages). In fact, given that the X/Y distinction (or some comparable distinction) is the only possible method of reconciling head-movement with generalized predicate-raising (in the sense of Pearson 1998) in the same language, it may even be a typical property of VOS languages in general.

5.3. On heads and verbs Verbal morphology is, in generative syntax, a property not of a given verb but of the clause itself. It is realized on the verb (in some languages) because the verb enters a head-movement or checking relationship with the syntactic head responsible for the relevant morphology. In other languages, verbs and e.g. tense-markers are realized as separate constituents. In yet other languages, subordination is realized, not as an independent complementizer, but as an affix to the verb. The same holds for negation and presumably all other functional categories in the clause. For the verb and a given morphosyntactic category to be able to unify, head movement (overt or covert) is necessary. This implies that only Xheads can be part of the verbal system of morphology. In a given language, if an element is realized as a verbal affix (or as a verb), it must be an Xhead, not a Y-head. However, the converse is not necessarily true: while verbal elements and verbal morphological categories represent a subset of X-heads, a language may very well have X-heads which are neither verbs nor verbal morphosyntactic categories (e.g. subordinators in most European languages are X-heads without being realized as verb morphology). If a category behaves as a verb, this may of course be due to the trivial fact that it is an instance of V and therefore undergoes the same processes as other verbs do. However, it may also be the case that it is an X-head which is located above V and is the closest available candidate for the verbal morphology concerned. This distinction must be kept in mind. Thus,

Adverbial Heads in a Formosan language

115

the Finnish negative verb (cf. 25 above) is a verb in one very relevant sense (namely that it can bear agreement morphology), but this can be captured neatly by analysing it as the realization of the head Neg° rather than an negative adverb in SpecNegP (such as English never), instead of having to stipulate that it is of category V. In a similar vein, as far as adverbial verbs / heads in Formosan languages such as Seediq are concerned, they are verbs in the same sense. However, the question remains whether this is a lexical primitive or simply the consequence of other attested properties: being an X-head and having lexical content. Applying the distinction between X- and Y-heads allows us to account for the data in Seediq and Malagasy in a unified fashion which is in line with Cinque's cross-linguistic study of adverb ordering. A purely lexical approach, where adverbial verbs are simply verbs, does not capture this generalization (nor does it, in all fairness, attempt to do so). If both approaches cover the facts equally satisfactorily, the approach which gives us the greatest cross-linguistic leverage is preferable. However, it remains a topic for future research to examine in more detail possible areas where the two analyses make different predictions, and thereby to determine empirically which approach is more adequate.52

6.

Conclusion

We have presented a set of data concerning the realization of adverbial meanings in Seediq and other Formosan languages, and noted that a whole set of these, in particular those dealing with manner, duration and frequency, are generally realized as verbs. Two alternative models for relating these facts to other typological features of Formosan languages are given, one which connects the existence of manner adverbial verbs to the mechanisms underlying VOS word order, while the other connects them with the Cinque adverb hierarchy, suggesting a revision of one important point in Rackowski and Travis' (2000) analysis of adverb order in Malagasy. We have argued that the problems of adverb ordering in Malagasy and other typologically similar VOS languages can be solved by means of a classification of heads into two major categories for which we have morphosyntactic evidence in at least some Austronesian languages, and that it is not necessary to make use of a morphosyntactically unmotivated distinction between XP vs. X° status for adverbs to block

116

Arthur Holmer

intraposition in either Seediq or Malagasy in order to derive the correct linear order.

Notes *

1. 2. 3.

4. 5.

This chapter is a revised and extended version of Intraposition and Formosan adverbial heads, a paper I presented at AFLA XI at ZAS, Berlin, in April 2004. The material presented here is based on fieldwork conducted in Taiwan in 1993, 1995 and 1998. I gratefully acknowledge financial support from the Swedish Research Council for the Humanities and Social Sciences, The Lundberg Ido Foundation, The Bank of Sweden Tercentenary Fund and Vetenskapssocieteten, which made this research possible. I am also grateful to the AFLA XI audience for valuable comments. The chapter has benefited from comments from and discussions with Edith Aldridge, Henry Yungli Chang, Sandra Chung, Mark Donohue, Sheila Dooley, Hans-Martin Gärtner, David Gil, Nelleke Goudswaard, Paul Law, Joachim Sabel, Peter Sells, Lisa Travis, Dylan Wei-Tien Tsai and Jordan Zlatev, as well as two anonymous reviewers. Naturally, my deepest vote of thanks goes to my principal Seediq consultant and friend Temi Nawi Tseng of Puli, Taiwan, who not only has been tireless in answering my questions, but also has kindly allowed me to make use of a corpus of stories collected by her. Needless to say, any mistakes are mine and mine alone. In this chapter, the following abbreviations are used: AF = Actor Focus; Arg = argument; ART = article; BF = Beneficiary Focus; CNEG = connegative; DEM = demonstrative; EXCL = exclusive; FUT = future; GEN = genitive; IMP = imperative; INCL = inclusive; IF = Instrument Focus; INT = interrogative; LLG=ligature; LOC=locative; LF = Locative Focus; Mod = modifier; NEG = negation; NOM = nominative; OBJ = object; PART = partitive; PF = Patient Focus; PRED = predicate; PRF = perfective; PST = past; QUOT = quotative; SUBJ = subject; UF = Undergoer Focus. Cf. the varying treatment of adverbial negation such as German nicht as an element in SpecNegP or as an adverb adjoined to vP. The nominative marker ka is optional. Note that mapa has an irregular PF, paanun. Note also the rather opaque morphophonemic alternation of the stem /kmekul/ - /knkel/, which depends on the stress pattern, which is affected in turn by affixation. I.e. the Patient which is realized in NOM in (2a) is the Patient of "carry", not of "able". Analyzing mapa ka kulu hlama as a clausal subject would imply that it should be possible to precede it with a nominative marker, which is quite ungrammatical: *Knkelun mu ka mapa (ka) kulu hlama.

Adverbial Heads in a Formosan language 6.

117

Trivially, of course, every verb in Seediq must be realized with some kind of focus marking. However, as we have shown in (2a-e), only the first instance of focus marking in the clause is distinctive and reflects the relation between the NOM argument and its position in the argument structure of the main verb. Other verbs in the same clause are obligatorily realized in AF which cannot contrast with any other focus in the same position. 7. As Holmer (1999) has noted, the focus of the verb does not directly reflect thematic roles such as Agent and Patient, but rather a more complex relationship which takes into account the grammatical status of the argument with respect to the argument structure of the verb. Aldridge (2004: 350) also comments on the lack of a one-to-one correspondence between focus affixes and thematic roles. One obvious example of this is that mhuqin 'die.AF' takes a NOM Patient, whereas smebuc 'beat.AF' takes a NOM Agent. 8. This form is labelled connegative here. However, it does not occur with all negations, only with ini 'NEG' and iya 'PROHIBITIVE' (in the latter case the use of an imperative form with a negative imperative construction is not necessarily a case of connegativity). Nor is its use restricted to negations: besides being used as an imperative, it is also used after the auxiliary asi 'must; just have to; spontaneously', which is not to be confused with the homophonous subordinated asi ' i f . 9. The temporal interpretations of negated vs. affirmative forms of the various focus categories are subject to rather complex interaction patterns (e.g. negated PF is generally interpreted as perfect, whereas affirmative PF is generally interpreted as future), but this need not concern us here. In all cases the free translation given reflects either the temporal interpretation in the context from which the example is taken, or the most neutral interpretation in isolation (if an example is elicited). 10. Incidentally, Aldridge (2004: 199-201) also uses this kind of data as evidence for verb movement in Seediq. Briefly, given that negation intervenes between periphrastic T/A marking and the verb, and given that it precludes the possibility of morphological T/A marking on the verb itself, the null assumption must be that T/A morphology is merged in Τ (or some functional level above Neg) and that the verb moves to Τ in cases when it bears overt T/A morphology, while this movement is blocked when the negation intervenes. 11. The form pnuqi, combining as it does, connegative with T/A marking, represents an ungrammatical combination of features. 12. The reader is reminded that the default AF morphology of the main verb here is crucially not distinctive. Therefore, the nominative subject in this and comparable examples is not to be interpreted as the Actor of the subordinate verb, but as the nominative subject of the clause as a whole.

118

Arthur Holmer

13. The form burux is a connegative; the indicative form is murux. 14. To be quite exact, the k- prefix is a verbalizer which transforms adjectival stems into full verbs which can receive non-AF morphology. However, when no further morphology other than the k- is added, the net result is a verbal stem which serves as a connegative. The corresponding affirmative form would be mtengi 'full'. 15. The verb pure 'to cook' belongs to a class of verbs (the vast majority with stems beginning in labial stops) which do not form AF by -w-affixation. 16. Or possibly more accurately, to the mapping between the NOM subject of the clause and its role in the argument structure of the main verb. 17. Why the main verb in Tsou bears distinctive focus morphology while the main verb in Seediq does not is also an interesting question, but outside the scope of this study. It should be addressed in a more complete survey of the detailed behavior of the phenomenon in the different Formosan languages. 18. In Paiwan (cf. Egli 1990: 158) the same word order variation obtains as in Tagalog, but with verbal morphology on the adverb. 19. I thank a reviewer for drawing my attention to this point. 20. This is the beginning of a much longer sequence which tells of how the speaker got a fish bone stuck in his throat. 21. An alternative view would be that tteun tmekan together form a complex predicate. Under such a view we would still have to address the problem of the focus difference between the two verbs. I thank a reviewer for drawing my attention to this alternative. 22. Possibly less conceivable as a gloss, but certainly not to be excluded on these grounds, would be a treatment of pcngaun daha mekan (beras baso) in (10a) as 'They sometimes-do the baso grain such that they eat / by eating'. 23. Cf. Chang (2004, this volume) for some other categories which are realized as lexical verbs in Kavalan. 24. A reviewer points out that (17b) can in fact be followed by he walks slowly. This is, of course, true. A negation can refer to anything within the clause, given suitable intonation. However, a rather marked intonation would be required to force the reading that the verb is negated without explicitly adding he walks slowly. For more discussion and further examples cf. Givon (2001: 381-382). 25. It is clear that all of the bracketed constituents are noun phrases: in (18a), the form tnqian is LF, which is regularly used for nominalization (e.g. k-n-pah-an (< meepah 'work') = work (already done)); further, it can also be preceded by the nominative marker ka. In (18b) the adjective mngihur could not possibly have a 3SG possessor (na), only the noun 'bitterness' can. In (18c), kari is a noun. The same holds for the examples in (19) which can be preceded by ka,

Adverbial Heads in a Formosan language

26. 27.

28. 29.

30.

31.

32.

33.

34.

35.

119

and which clearly contain relative clauses. I thank a reviewer for drawing my attention to the importance of clarifying this point. Cf. the fact that it cannot receive nominative marking with ka. I thank a reviewer for drawing my attention to this fact. If they are diachronically valid, they may well still be synchronically valid in some Formosan languages. To address this question, a detailed study of each language which displays the phenomenon would have to be undertaken. One such study is Chang (this volume) on Kavalan. Cf. English not, which prevents unification of V with tense and agreement, while phrasal never does not: *He not works. / He never works. The only context where a verb in connegative is lacking is in the past tense, where the negation verb in Mind en puhunut suomea. Ί did not speak Finnish.' contrasts with the auxiliary ole- 'to be' in Mind olen puhunut suomea. Ί spoke Finnish'. I thank a reviewer for drawing my attention to the importance of this issue. A reviewer comments on the fact that most of the adverbs cited in (28) do not bear verb morphology. As we shall see in section 5.2, the proposed analysis automatically excludes verb morphology on all postverbal adverbs, as well as on any preverbal adverb which is located higher than Τ in the structure. Note that the negation primarily negates the adverbial head mhmeti 'at random.PF.CNEG', and not ba 'just', which is rather an emphatic post-modifier to ini. The adverb ba cannot bear connegative morphology because it is not a verbal element. We shall presently see evidence suggesting that it may not be a head at all. Note that k- functions here as a connegative marker, although it is strictly speaking an adjective verbalizer (cf. footnote 11). The non-connegative form is mqeni 'thirsty'. Of the adverbs shown above, both ba 'really' and klaali 'always' (and similar adverbs) can occur in varying positions, ba due to scopal factors, and adverbs like klaali being capable of fronting. Note that wada here is a past tense verb meaning 'went', which is clearly homonymous with the pure past tense marker, although it is not clear which is derived from which. A further issue, raised by a reviewer, concerns the mechanisms required for producing an order pattern such as that in (30), where type 10 is preverbal and type 8 is postverbal. However, this is not really a problem: once we have a mechanism for deriving reverse Cinque order for postverbal adverbs, the positions of preverbal adverbs (crucially, provided that they appear in Cinque order) can be derived if the mechanism simply fails to apply in a certain context. What prevents this mechanism from applying is the topic of this section.

120

Arthur Holmer

36. Although, as pointed out by a reviewer, they may ultimately be triggered by the same type of features. 37. Aldridge (2004: 229-231) suggests an alternative account for VP-raising and TP-raising that does not rely on generalized predicate raising. Instead, she makes use of a filter called the Stranded DP Constraint: "A DP cannot be spelled out in the leftmost position in a phase edge" - whenever a DP would be spelled out in such a position, the remainder of the phase moves leftwards past it. Given that this only applies to phases (i.e. CP or vP) it does not account for adverb ordering or for the ordering of final particles, two questions which Aldridge does not address. 38. As a reviewer points out, this conclusion is actually not logically necessary. It is quite possible that similar orders in Seediq and Malagasy are derived by entirely different mechanisms. However, the null hypothesis should be that both systems are derived in the same way, until evidence to the contrary is found. 39. Assuming, of course, that the negation is to be classified as an adverbial element. Here I am solely concerned with the implications of the Seediq data for the R&T analysis, and since that analysis includes the negation as one of the elements that it accounts for, I treat negation here on a par with adverbials, well aware that this treatment may not be semantically motivated in other contexts. 40. The reason why we cannot recognize the headhood of tena on the basis of morphological marking is simply that it located to the left of (i.e. structurally higher than) the category of tense, so it is not involved in any morphological processes typical to verbs. 41. Cf. example (24c) above. 42. No head may precede the head which serves as a host for pronominal cliticization, unless it is clearly located within a fronted phrase, such as the complex adverbial ini huwa 'it's OK', cf. section 5 for further examples, cf. also Holmer (1996: 58-60). 43. Although it is likely that it derives etymologically from a matrix clause taking the remainder of the utterance as its complement. 44. There is no shortage of cross-linguistic analogies with syntactically sensitive cliticization. To mention just two: in Spanish, object clitics are proclitic to tensed verbs but enclitic to non-tensed verbs, and in French, object clitics are proclitic to verbs. 45. In fact, it is entirely unclear how an alternative account would work. Verbinitial order can be derived by raising the entire TP to the outer specifier of CP (cf. Aldridge 2004: 207-209). As long as no complementizer is present, the clitic could conceivably be clause-initial by virtue of having reached Τ within the raised constituent. However, if a complementizer is present, it must

Adverbial Heads in a Formosan language

46. 47. 48.

49.

50.

51. 52.

121

precede TP, and the clitic, which must cliticize to the complementizer in this case, would either have to leave the raised predicate to join the backbone of the clause again, or would have to leave TP before TP is fronted. It is unclear what mechanisms would be involved here, or indeed if this is possible at all. Note that this is not a problem as long as cliticization is purely phonological attachment to the first word, but the contrast between (36) and (37) shows that cliticization is syntactically sensitive to head-hood in Seediq. "Tough constructions" in Seediq require the verb to be realized in PF. However, the main predicate is clearly the predicative adjective cheya 'tasty'. Examples (40b) and (40d) are quoted from Aldridge (2004: 199) but the glosses have been adapted to those used in the present chapter. cf. Holmer (2005) for a discussion. There is no obvious connection between either class of heads proposed here and typical A- or A'-properties. Since it is intuitively unappealing to propose a distinction which is orthogonal to another distinction already proposed (in essence suggesting that there are 4 types of heads instead of 2), a natural question to be posed in this context is whether the two distinctions can be connected in any way. This question is, however, outside the scope of this chapter. It should be noted that the problem of blocking has nothing to do with adverbs or with antisymmetry. It simply concerns the non-interaction of verb movement with final particles assumed to be heads. Antisymmetry and the X/Y distinction suggested here gives us the leverage to explain why the nonblocking heads are necessarily postverbal. For a longer discussion cf. Holmer (2005). The difference between Seediq and Malagasy might be the result of the subject surfacing structurally higher in Malagasy than in Seediq. It would be an interesting task to seek independent evidence for such an assumption. I.e. without the overt realization of adverbs with verbal morphology. Even if we adopt the lexical approach, this does not allow us to dispense with the X/Y distinction, since we still have to account for the fact that postverbal heads do not block verb movement while verbs, negators, and periphrastic T/A-markers as well as some preverbal adverbs (such as tena 'already') do. Further, recall that it is not the case that Y-heads (e.g. aspect particles, evidentiality particles, and some subordinators) are uniformly structurally higher or lower than X-heads (e.g. the verb, some adverbs, negators, T/A markers, and some subordinators), so we cannot explain the lack of blocking by simply stipulating that a constituent including all the X-heads raises past all the Y-heads to derive the correct order, nor can we account for the reverse Cinque order of the postverbal elements. I thank Sandra Chung (p.c.) and a reviewer for drawing my attention to the fact that these points required further clarification.

122

Arthur Holmer

References Aldridge, Edith 2002 Nominalization and WH-movement in Seediq and Tagalog. Language and Linguistics 3 (2): 3 9 3 ^ 2 6 , Taipei: Academia Sinica. 2004 Ergativity and word order in Austronesian languages. Ph. D. thesis, Cornell University. Borsley, Robert, Maria-Luisa Rivero, and Janig Stephens 1996 Long Head Movement in Breton. In The Syntax of the Celtic Languages, Ian Roberts, and Robert Borsley (eds.), 53-74. Cambridge: CUP. Carnie, Andrew, Heidi Harley, and Elizabeth Pyatt 2000 VSO order as raising out of IP? Some evidence from Old Irish. In Andrew Carnie, and Eithne Guilfoyle (eds.), 39-59. Carnie, Andrew, and Eithne Guilfoyle (eds.) 2000 The Syntax of Verb Initial Languages. Oxford: OUP. Chang, Henry Yung-Li 2004 The guest playing host: Modifiers as matrix verbs in Kavalan. Paper presented at AFLA XL, Zentrum für allgemeine Sprachwissenschaft, Berlin, April 2004. Cinque, Guglielmo 1999 Adverbs and Functional Heads. A Cross-linguistic Perspective. Oxford: OUP. Egli, Hans 1990 Paiwangrammatik. Wiesbaden: Harassowitz. Givon, Talmy 2001 Syntax. Vol I. Amsterdam: John Benjamins. Holmberg, Anders, and Christer Platzack 1995 The Role of Inflection in Scandinavian Syntax. Oxford: OUP. Holmer, Arthur 1996 A parametric Grammar of Seediq. Lund: Lund University Press. 1999 Structural implications of the function of Instrument Focus in Seediq. In Selected Papers from the Eighth International Conference on Austronesian Linguistics, Elizabeth Zeitoun, and Paul Jen-Kuei Li (eds.), 423-453. Taipei: Academia Sinica. 2002 The encoding of adverbs of manner in the Formosan language Seediq. Paper presented at International Symposium on Linguistics and Speech and Hearing Sciences, Kuala Lumpur, October 2002.

Adverbial Heads in a Formosan language 2005

123

Seediq - antisymmetry and final particles in a Formosan VOS language. In Verb First: Papers on the Syntax of Verb Initial Languages, Andrew Carnie, Sheila Dooley, and Heidi Harley (eds.), 175-201. Amsterdam: John Benjamins. Huang, Lillian 1993 A study of Atayal syntax. Taipei: Crane Publishing Co. Jeng, Heng-Hsiung 1977 Topic and Focus in Bunun. (Inst of History and Philology, Academia Sinica Special Publication no. 72). Taipei: Academia Sinica. Kayne, Richard 1994 The Antisymmetry of Syntax. Cambridge, MA: MIT Press. Massam, Diane 2000 VSO and VOS. Aspects of Niuean word order. In Andrew Carnie, and Eithne Guilfoyle (eds.), 97-116. Pearson, Matthew 1998 Predicate raising and VOS order in Malagasy. UCLA Occasional Papers in Linguistics 20: 94-110. Rackowski, Andrea, and Lisa Travis 2000 V-initial languages: X or XP movement and adverbial placement. In Andrew Carnie, and Eithne Guilfoyle (eds.), 117-141. Schachter, Paul, and Fe Otanes 1972 Tagalog Reference Grammar. Berkeley: University of California Press. Szakos, Joszef 1994 Die Sprache der Cou. Untersuchungen zur Synchronic einer austronesischen Sprache auf Taiwan. (2. Teil: Texte und lexikalische Analyse). Ph. D. thesis, Friedrich-Wilhelms-Universität, Bonn. Travis, Lisa 1984 Parameters and effects of word order variation. Ph. D. thesis, MIT. 2004 VP-, D°- movement languages: Malagasy in wonderland. Paper presented at AFLA 11, Zentrum für allgemeine Sprachwissenchaft, Berlin, April 2004.

*

Patterns of Phrasal Movement: The Niuean DP Arsalan Kahnemuyipour and Diane Massam

This chapter examines substantive noun phrases in Niuean, a Polynesian language of the Tongic subgroup with VSO word order. We describe the allowable orderings of elements in the Niuean noun phrase, which include certain variations in the placement of numerals and the genitive possessor. We then provide a roll-up complement-to-specifier analysis of Niuean DPs that derives the correct word orders and accounts for the variation in position and properties of possessors and for the variation in position of numerals. We propose that two factors can affect this pattern of phrasal movement. First, filled specifiers make movement to specifier impossible. Secondly, we confirm a movement constraint observed by Rackowski and Travis (2000), which states that purely relational functional projections, which have no semantic content, are invisible to movement and cannot themselves move.

1. Introduction This chapter examines substantive noun phrases in Niuean, a Polynesian language of the Tongic subgroup with VSO word order, isolating morphology, and an ergative case system. We describe the allowable orderings of elements in the Niuean noun phrase, which include certain variations in the placement of numerals and the genitive possessor, then we provide a phrasal movement analysis for these variations, treating first the variation in the position of the possessor, then the variation in the position of the numeral. Parallels are drawn between the derivation of nominal and sentential word order. There has been a large quantity of work, both recent and traditional, attempting to understand why certain orders of elements in clauses seem to be universally ruled out. To account for this, some linguists have posited that there is a universal order of elements and that allowable variations on this order are derived by various movement patterns (e.g. Cinque 1996, 1999, 2002, Belletti 2004, Rizzi 1997, 2003). This position allows for the

126

Arsalan Kahnemuyipour and Diane Massam

theory to rule out ungrammatical orders by universal constraints on movement, rather than by typological stipulations. Our exploration of the Niuean DP takes place in light of this type of work, as well as similar work specifically on DPs such as Cinque (2000, 2005), Coddington (2003), den Dikken (2003), Pearce (2002, 2003) and Shlonsky (2004). Given that Niuean DPs generally have a N-initial order, similarly to the V-initial order of clauses, our analysis also addresses the relation between nominal structure and sentential structure, finding striking parallels between the two phrase-types.1 We observe that the required derivations involve consistent complement-to-specifier movement patterns, with two variable factors. First, in the case of filled specifiers, the movement pattern is affected in that movement to specifier is impossible, and second, we confirm a movement constraint observed by Rackowski and Travis (2000), which states that purely relational functional projections such as Agreement, which have no semantic content, are invisible to movement.

2. Description of the Niuean substantive DP The Niuean DP is described in Seiter (1980), and in Massam and Sperlich (2000). In a DP without a possessor or numeral, the order of elements is as shown in (1). (1)

Surface Order of Elements without Possessors and Numerals2 Case+P/C a. e

#

Adjs

Dem

manu kula fulufuluola e: bird red beautiful DEM 'those beautiful red birds' (Field Notes.01) b. e kau kaiha ABS.C group thieves 'a group of thieves' (Seiter 1980.100a) c. e falu a tagata ABS.C some person 'some people' (Sperlich 1997.67) d. e taha tagata e. a Moka ABS.C one person ABS.P Moka 'a person' (Field Notes.01) 'Moka' (Field Notes.97) ABS.C

tau

Noun

PL

Patterns of Phrasal Movement

127

The first element in the nominal clause is a portmanteau morpheme, which indicates the case of the DP as well as whether it is common or proper (where proper includes pronominal). In (la-d) this particle is e (absolutive common), whereas in (le) it is a (absolutive proper).3 This is followed by an optional marker for number, which also has classifier-like properties, as can be seen in (la) and (lb) where a different plural marker appears depending on the nature of the noun or group. Other plural classifiers include lafu for a family group, atu for a row, and na: for a pair.4 This marker can also have article-like properties, in that an indefinite plural NP can appear with falu a, as in (lc) and in the case of a singular indefinite NP, the marker taha or ha can appear in this position, as in (Id). Seiter 1980 notes that it is not clear if the a appearing with falu should be considered a ligature item or not, and here we leave it without a gloss. We will refer to this complex morpheme simply as number (#) in this chapter.5 This morpheme is followed by the head noun, which is in turn optionally followed by one or more adjectives and a demonstrative as in (la), (le) shows a proper DP. The order of elements in the DP is fixed. We next examine DPs with possessors, while DPs with numerals will be discussed in Section 5. For possessors, there are two possible orders, as shown in (2).6 The first order finds the genitive case marked possessor in pre-nominal position. In this order, there is a ligature item a appearing between the possessor and the (#) noun, as in (2a). The second order finds the genitive marked possessor at the end of the entire DP (after the demonstrative if there is one), as in (2b). (2)

Surface Orders of Elements with Possessors (without Numerals) a. Case+P/C b. Case+P/C a', e ABS.C 'Sione's b'. e ABS.C 'Sione's

Poss Lig

# Noun

Adjs

Dem

# Noun

Adjs

Dem

ha Sione a leo GEN.Ρ Sione LIG voice voice' (Seiter 1980.92b) leo ha Sione voice GEN.Ρ Sione voice / voice of Sione' (Field Notes.97)

Poss

128

Arsalan Kahnemuyipour and Diane Massam

The pre-nominal possessive construction has two particular properties distinct from the properties of the construction with the possessor at the end of the phrase. First, the pre-nominal possessor gives a defmiteness reading to the DP as a whole, similarly to the situation in Hebrew and Arabic (see, e.g. Ritter 1988, Shlonsky 1988, Borer 1999), as shown in (3) (Sperlich 1997: 103). (3a) has a definite reading, whereas (3b), like nonpossessed Niuean DPs, can be definite or indefinite. 7 (3)

a. ko e haana a fale. PRED his LIG house 'It's his house.' [definite] b. ko e fale haana PRED house his 'It's his house. / a house of his'

The second property of the pre-nominal possessor construction is that the pre-nominal possessor must be proper as in (2a'), or pronominal as in (3a). It is ungrammatical to have a common pre-nominal possessor (4a) (Field Notes.01), although such a possessor is fine in final position, as shown in (4b) (Field Notes.01). (4)

a. *Ko e he faiaoga a pepa PRED GEN.C teacher LIG book 'the teacher's book' (intended meaning) b. Ko e pepa he faiaoga PRED book GEN.C teacher 'the book of the teacher'

These properties are summarized in (5). (5)

Property 1: The pre-nominal possessor gives a defmiteness reading to the DP as a whole (3 a). Property 2: The pre-nominal possessor must be proper or pronominal (2a', 3a).

Given the facts described above, we will next address the following two questions: How do we derive the order of elements? How do we account for the two positions (and corresponding properties) of the possessor?

Patterns of Phrasal Movement

3.

129

Setting the stage

Let us first turn to the question of the order of elements. One logical possibility is to assume that Ν is base-generated in the same place it surfaces in (2a), i.e. between the # and the Adjectives. Given the impossibility of Ν taking Adjectives and Demonstratives as complements, the only way to have this option is to assume a combination of right and left branching, contra Kayne's (1994) antisymmetric system, which disallows left branching complementation universally. Note that this would violate even a weaker version of an antisymmetric system, which would allow crosslinguistic variation in branching direction, but not different directions of branching within a single language or within a single phrasal category. We thus take Ν in (2a) to be base-generated at the end of the phrase. Next to be determined is the merge order of the other elements in the nominal phrase. In order to address this question, let us look at some typological generalizations and see how they are accounted for by assuming a universal order of elements. The usual order of elements in the Noun Phrase was perhaps first observed by Greenberg (1966).8 Greenberg's observation is given in (6).9 (6)

Universal 20 (Greenberg 1966: 111) "When any or all of the items - demonstrative, numeral, and descriptive adjective - precede the noun, they are always found in that order. If they follow, the order is either the same or its exact opposite."

The generalization in (6) is partially summarized in (7), as proposed in Cinque (1996). Numerals are included, but we leave them aside until Section 5. (7)

a. Dem - Num - A - Ν = Base Ordering b. *A - Num - Dem - Ν = Impossible c. Ν - Dem - Num - A = Noun Movement d. Ν - A - Num - Dem = Successive XP raising Cinque (1996) (see also Cinque 2000, 2005)

Niuean

Cinque (1996) accounts for the ordering restrictions in (7) in the following manner. (7a) is the basic merge order. No movements are necessary to derive this order. The order in (7c) is the result of N-movement through

130

Arsalan Kahnemuyipour and Diane Massam

phonologically null head positions.10 Note that in this case, the lexical items Num, Dem, and A are assumed to be in specifier position. The mirror-image order in (7d) is the order found in Niuean. Cinque argues that this order is the result of successive XP-movement, the details of which will be thoroughly outlined below. Crucially, if the XP-movement is successive and local, and necessarily originates with and includes N, (7b) is impossible, since in this case A, Num, and Dem would have had to move independently of N. Our approach is different in technical details from that of Cinque, but the idea is the same. One technical difference, for instance, is that Cinque places the pre-nominal modifiers in Specifier positions, whereas we consider them to be Heads. Our analysis is in line with the Rackowski and Travis (2000) account of the Niuean verbal clause in these respects. See also Shlonsky (2004), who considers that some elements are heads, while others are in specifier position. Another related fact is the order of descriptive adjectives. It has been suggested that there is a universal order of descriptive adjectives (Laenzlinger 2000, 2005, Scott 1998, Sproat & Shih 1991), given in (8). (8)

Proposed Universal Order of Adjectives Quantification > Quality > Size > Shape > Colour > Nationality

If in a language like Niuean there is successive XP-movement to derive the mirror-image order in nominal phrases, one would expect the adjectives to appear in reverse order as well as the functional elements. This prediction is borne out in all the examples we found in texts. We saw an example of this in (la) with "color" and "quality" in the opposite order. The same phenomenon is observed in (9a) (Nelisi 1995: 6) for "color" and "size" and in (9b) (de Sousa 2001: 50) for "size" and oti 'all', which we take to be a quantiflcational adjective. (9)

a. e letio kula ABS.C radio red 'the little red radio' b. e tau koloa ABS.C PL store 'all those small stores

tote little ikiiki small[PL] of mine'

oti ia haaku... all DEM my

Patterns of Phrasal Movement

131

In the next section, we discuss the details of how the inverse order is derived in Niuean.

4. Deriving inverse order In this section we will outline how the order of elements in the Niuean noun phrase can be derived. The merge order we assume for heads is given in (10).11 This order is based on a body of work on ordering of categories within the noun phrase. (For examples, see, Ghomeshi and Ritter 1996, Megerdoomian 2002, Pearce 2002, Ritter 1991, 1993, Schoorlemmer 1998, Travis 1992, among others). (10) DP Merge Order Κ

D

Poss

Dem

A

#

Ν

The merge order of Κ and D in (10) follows standard assumptions in generative theory (see; for K, Bittner and Hale 1996 and for D, Abney 1987). Note that D is phonologically null in Niuean. Our assumption is that D contains features for the proper-common dichotomy, since these features make reference to uniqueness of identity and thus are clearly related in function to determiners. The morphology associated with these features shows up in K, rather than in D, possibly due to head or feature movement. With respect to the Possessor (Poss) head, it has been suggested in the literature that there are two positions across languages, one lower position much closer to the noun, which is utilized, for instance, by Semitic languages and one higher one, which we suggest is the one used in Niuean. English possibly uses both positions, e.g. 'John's damaged car door'. Schoorlemmer (1998) outlines these two possibilities and a principal property she attributes to the languages that use the high position coincides with a key property of Niuean, namely the fact that if the possessor is definite the entire phrase becomes definite. We will return to this point below. For the merge position of Demonstratives (Dem) and Adjectives (A), we are following Cinque as outlined above in (7). For the low merge position of Number (#), we are following Ritter (1991, 1993). (But see note 5·)

132

Arsalan Kahnemuyipour and Diane Massam

The order (2a), with the pre-nominal possessor, is derived in a manner illustrated in (11) which involves successive "intraposition" or roll-up movement of complements to the empty specifiers in their shared maximal projection.12 As shown in (11), the #P moves to the spec of AP, then the whole AP moves to the spec of DemP. DemP cannot move to the already filled spec of PossP and finally, in accordance with Shortest Move, PossP (and not DemP which is further away from D) moves to the spec of DP. DP does not move to specifier of KP13, hence K, which is not represented in (11), appears at the far left. In (11a) we see the overview of movement patterns, while ( l i b ) gives the Merge tree (a hypothetical tree that never exists in the Minimalist system, in which Merge and Move are interwoven), and (1 lc) (see next page) the surface tree. (11) Pre-nominal possessor derivation (2a) a. Overview of Movements [DP D [POSSP DPposs Poss

[DemP

Dem [AP

A [#P # [NP Ν ]

b. Merge Positions (hypothetical tree) DP

D

PossP DPposs

Poss a

DemP

Dem

AP

A

#P #

NP

The question arises as to why NP does not move to the specifier of #P. This can be answered easily if # is in specifier position of #P, in which

Patterns of Phrasal Movement

133

case it would block the movement of NP. This claim is supported by the fact that falu usually appears with an optional particle a, which is arguably the head of #P, and perhaps by the fact that taha has an alternative form ha, suggesting that this number marker is formed of two parts, with ta in specifier and ha in head position14. Again, in accordance with Shortest Move, #P (not NP) will move to specifier of AP. (11)

c. Surface Tree DP

The pre-nominal order of the possessor has two properties as discussed above and summarized in (5). The first of these is that pre-nominal possessors give a defmiteness reading to the noun phrase as a whole. According to Schoorlemmer (1998), in languages with the high PossP, Poss is a potential carrier of a value for defmiteness. We posit that a in Niuean is one such element. The Poss head, which is home to a, an element with semantic content, gives the whole DP the definite reading. The second property of the pre-nominal possessor construction is that the possessor must be proper and cannot be a common noun. For this, we posit that the Poss morpheme a has a [proper] feature which must be shared with its specifier. This is supported by the fact that a is used for three other morphemes in Niuean that bear the feature [proper]. These three are: Absolutive proper case, as shown in (le) and repeated in (12a) (Field Notes.97); Proper article in goal DPs, as shown in (12b) (Sperlich 1997: 41); and Genitive proper case, as shown in (12c) (Seiter 1980: 35)15.

134

Arsalan Kahnemuyipour and Diane Massam

(12)

a .a

Moka ABS.Ρ Moka 'Moka' c. e vaka a ABS.C canoe GEN.Ρ 'Sione's canoe'

b. ki a au to PERS I 'to me' Sione Sione

Thus we see that a has two roles stemming from its two features of [definite] and [proper], which are giving a definite reading to the whole DP and sharing the proper feature with the possessor. Let us now turn to the order (2b). The derivation for this order is shown in (13). Again, (13a) shows the overview of movements, (13b) shows the merge positions, and (13c) (see next page) shows the surface tree. (13) Post-nominal possessor derivation (2b) a. Overview of Movements [DP D [ possp DP Poss PQSS

[oemP .

Dem [AP

b. Merge Positions (hypothetical tree) DP

DPposs

Poss

DemP

Dem

AP

A[,

Patterns of Phrasal Movement

135

(13) c. Surface Tree DP

The first two movements are exactly the same as (11): #P to spec of AP and AP to spec of DemP. The only difference here is that there is no a in Poss. To get the right order, we need the whole DemP to move over PossP to the spec of DP, as shown in (13). The question remains, however, as to why in (13) DemP, rather than PossP, moves to spec of DP. Recall that in (11), it was PossP that moved, which is expected under some version of relativized minimality or shortest move. To explain the phenomenon in (13), we make use of an idea in Rackowski and Travis (2000), where they derive the order of adverbs in Malagasy and Niuean from Cinque's (1999) universal order of adverbs. Let us look at their analysis briefly. Their derivation for the Niuean verb phrase is given in (14). The relevant phrases are Direction Phrase (DirP), Manner Phrase (ManP), Aspectual Adverb Phrase (AspP), Agreement Phrases for subject and object (AgrS and AgrO), and Question Phrase (QP).16 As in (11) and (13) above, (14a) shows the overview of movements, (14b) (see next page) shows the Merge positions, and (14c) (see next page) shows the surface tree. We abstract away from the issue of whether the subject and object DPs are merged in, or move to, their respective Spec/Agr positions. (14) Niuean clausal derivation (adapted from Rackowski & Travis 2000) a. Overview of Movements [QP Q U g r S P _ AgrS [

AG

ROP_

AgrO

[Asp-Advp_

Asp

136

Arsalan Kahnemuyipour and Diane Massam

(14)

b. Merge Positions (hypothetical tree) QP

AgrS

AgrOP

AgrO

AspP

Dir c. Surface Tree QP

VP

Patterns of Phrasal Movement

137

Note the striking parallel between (14) and (13), both of which essentially involve the same series of movements. Thus, in (14), VP moves to spec of DirP, DirP moves to spec of ManP, and so forth. Crucially, when the movement sequence gets to the AgrO and AgrS phrases with arguments in their specifiers, they are skipped and they cannot themselves move. To account for this fact, Rackowski and Travis suggest a restriction on movement given in (15). (15)

Rackowski & Travis (2000: 127) "To avoid this ungrammatical derivation, there must be a restriction in the grammar such that non-contentful phrases like AgrP are invisible to movement and cannot themselves move. In contrast to this, contentful phrases like AdvPs can and, in this case must, move."

The restriction is that non-contentful phrases like AgrP are invisible to movement and cannot themselves move. We suggest that the same restriction is in place for PossP in (13). This seems plausible, given the oftnoted parallels between AgrP and PossP (see discussion in Stowell 1983, Abney 1987, and many subsequent authors), both of which introduce the phrasal subject, and both of which represent an open relation rather than bearing an element with interpretable semantic content.17 In fact, we seem to have come across a striking example to support their proposal here, since we have a head that is contentful in one case and non-contentful in the other. When it is non-contentful as in (13), it is skipped and cannot itself move. In (11), on the other hand, the Poss head is contentful; it contains the interpretable feature [definite] realized by a. In this case, as predicted by Rackowski and Travis, the PossP moves which results in the prenominal possessor order.18

5.

Numerals

As well as variation in word order of possessors, Niuean exhibits variation in word order of numerals. Given what we have proposed for Niuean, we expect an inverse order for numerals compared with adjectives and demonstratives: [N - Adjective - Numeral - Demonstrative], as outlined in (6) from Greenberg (1966) and also discussed in Cinque (1996, 2000, 2005). Let us consider how numerals actually do behave in Niuean.

138

Arsalan Kahnemuyipour and Diane Massam

In fact, in Niuean, numerals are found both preceding and following the noun, as shown in (16). When numerals precede the noun as in (16a), a ligature item e appears between the numeral and the noun, similarly to prenominal possessors. The order of the other elements remains the same. 19 In case of post-nominal numerals, the ligature item does not appear. We have left possessors out of (16) for now, but will return to them below. Note when counting humans, the prefix toko appears on the numeral, which is sometimes written as part of the numeral, or with a hyphen (16a') (de Sousa 2001: 33), and sometimes written as a separate word (16b') (Blanc and Togakilo 1965). (16)

Orders of Elements with Numerals (without Possessors) a. Case+P/C

Num Lig

b. Case+P/C a', toko-lima e PERS-five LIG ' five tall people'

#

Noun

Adjs

#

Noun

Adj

tagata person

loloa tall

Dem Num

Dem

b'.Maori toko ua Maori PERS two ' three Maoris'

Numerals can co-occur with possessors. The most commonly found construction with both a numeral and a possessor is one where the numeral precedes the noun and the possessor follows it, as exemplified in (17) (Blanc and Togakilo 1965). (17)

Ko e toko fa:

e

tama a

PRED PERS four

LIG

boy

GEN

Matakuhifi Matakuhifi

'Matakuhifi's four sons' In deriving these orders, let us consider first the pre-nominal numeral order in (16a) and (17). If we take the base order to be that assumed by Cinque (1996), namely [Dem Numeral Adj N], the order in (16a) and (17) can be derived in a straightforward manner. (The examples given here happen not to include a demonstrative or # marker.) As in the derivations above, (18a) shows the overview of movements, (18b) shows the Merge positions, and (18c) shows the Surface tree.

Patterns of Phrasal Movement a. Overview of Movements [dp D [possP DP p o s s Poss [oemP _ Dem [NumP Num e

[Ap

_ A [#P # [NP N]

b. Merge Positions (hypothetical tree) DP

D

PossP DP Poss

DemP

Dem

NumP Num AP A

#P NP

c. Surface Tree DP DemP D

PossP

NumP Dem

tNumP

DP Poss

Num AP #P A #

NP

t#P

139

tDemP

140

Arsalan Kahnemuyipour and Diane Massam

First, the #P moves to the specifier of AP, just as in all the derivations so far. At this point, it is not possible to move the AP into the specifier of NumP, because this position is filled by the Numeral, while the head of NumP is filled by the particle e. Instead, following Shortest Move, the NumP is moved to the specifier of DemP. This movement exactly parallels the movement in (11) of PossP to DP, in place of movement of DemP to PossP, the latter of which is similarly blocked by a full specifier position. Then, the derivation proceeds as does that in (14), to yield the order in (16a) and (17) where the numeral is pre-nominal. This derivation yields a post-nominal possessor and a pre-nominal numeral as in (17). In Niuean, each of the possessor and the numeral can appear prenominally. (2a') shows a pre-nominal possessor and (16a') shows a prenominal numeral. It is unclear whether both the possessor and the numeral can be pre-nominal in the same phrase, because there is a strong tendency on the part of speakers to avoid having more than one such element on one or the other side of the noun. Similarly, it is unclear whether both the possessor and the numeral can be post-nominal. A pre-nominal possessor and a post-nominal number is also unattested in our data. We leave this question open, pending future fleldwork. Let us now address the post-nominal numerals, as in (16b). Note that there is no ligature item e in these cases. Our analysis of nominal movements laid out in this chapter might predict that in such a situation, NumP will resist movement, as PossP did in (13) to yield a word order [Poss # Ν Α Dem Num], which is not what is found, since numerals precede demonstratives in Niuean, as schematized in (16b). (For an example, see (20).) This incorrect derivation is shown in (19a). A solution lies in the analysis of numerals as heads in such cases (see Shlonsky 2004). If this position is tenable, the correct derivation results, as shown in (19b). Thus, Num is a head unless the head is filled with the ligature item e, in which case Num is merged in specifier position instead of head position. As noted above, we have not come across any examples with both number and possessor on the same side of the noun, thus we assume in (19) that the possessor head is filled with the ligature item a, and that the final order will be [Poss # Ν A Num Dem], though, as noted above, such examples with pre-nominal possessor and post-nominal numerals are also unattested in our data set to date.

Patterns of Phrasal Movement

141

(19)

a. An incorrect prediction [dp D [ possp DP Poss Poss [ e m P _ Dem [ NumP Num null [ AP _ A [ #P # [ NP N] D

[dp

b. A correct derivation D [ possp DP P o s s Poss [ DemP _Dem [NumP

Num [ AP _ A [nP # [ NP N]

This derivation yields the word order [Poss # Ν A Num Dem]. The important question is, does Niuean conform to the Greenberg generalization that when numerals are post-nominal, they follow adjectives? In most examples adjectives and numerals do not co-occur post-nominally, because, as just discussed, there is a strong preference to avoid having strings of multiple nominal "modifiers" (in the loose sense, including numerals and possessors). Thus, when a noun is modified by both a possessor or adjective and a numeral, one or other of them is usually found pre-nominally, while the remaining one is found post-nominally, as in (16a') and (17). This is true of most of the natural examples we have found in texts. However, (20) (Niue: A History of the Island) shows that a numeral and an adjective can both follow the noun. In this example, the numeral follows the adjective post-nominally, in accordance with Greenberg's generalization. 20 (20)

e

tau

ABS.C PL

tagata

matakutakuina

toko-lima

na:

man

awesome

PERS-five

DEM

'These five awesome men' In addition to the uses above, Niuean numerals can also function as sentential predicates as in (21) (Sperlich 1997). (21)

Valu [ e

hui

he

eight

leg

GEN.C

ABS.C

feke ] octopus

'The legs of the octopus are eight.' In this function, numerals often modify noun phrases as predicates within a relative clause modifier, as in (22) (Blanc and Togakilo 1965). This serves as another strategy to avoid pile-up of modifiers in the noun phrase, and is common when there are multiple modifiers, as in (22), and also in (23) from de Sousa (2001: 50).21

142

Arsalan Kahnemuyipour and Diane Massam

(22)

motu

ikiiki mo e tokolalo ne

island small and sandy ' f o u r small a n d sandy islets'

fa:

COMP four

(lit. 'small and sandy islets that are four') (23) Fiamanako au ke sela e tau koloa ikiki oti ia want I COMP sell ABS.C PL store small all DEM haaku ne lima i Niue ne mai e Sione my COMP five LOC Niue COMP give ERG.Ρ Sione Ί want to sell all those five small shops of mine in Niue which I have inherited from Sione.' In this section we have seen that numerals can occur both pre-nominally and post-nominally, and that these orders can be derived in a simple manner which conforms to our movement algorithms, provided we allow for numerals to be either heads or specifiers, in the latter case, with e as head. Some questions arise as to the ways in which numerals and possessors can co-occur. In particular, available data are unclear first as to whether both numerals and possessors can occur pre-nominally or postnominally, since there is a clear preference to avoid having the two expressed together on the same side of the noun, and second as to whether a pre-nominal genitive can co-occur with a post-nominal number.

6.

Conclusion

Following leading ideas of Cinque (1996, 2000, 2005), we have presented a roll-up complement-to-specifier analysis of Niuean DPs that derives the correct word orders and accounts for the variation in position and properties of possessors, and for the variation in position of numerals. Having a filled specifier affects the pattern of movement. In addition, we have claimed, following Rackowski and Travis (2000), that the change in pattern of movement that is seen in some cases in Niuean is tied to the content of functional heads. If the head is filled, movement of the phrase is possible, but if the head is phonologically and semantically empty, the phrase does not move.

Patterns of Phrasal Movement

143

Notes

1.

2.

3.

4. 5.

6.

7.

We would like to thank Ofania Ikiua for her invaluable work as Niuean consultant, Ben Milam for his work as research assistant, and Donna Starks for assistance with conducting field work. We would also like to thank Guglielmo Cinque, two anonymous reviewers, the editors of this volume, and audiences of the University of Toronto syntax research group, AFLA (ZAS, Berlin) and the Canadian Linguistics Association meeting for useful comments and discussion. Funding for this work has been provided by a research grant from Social Sciences and Humanities Research Council of Canada "Issues in Niuean Syntax" (#410-01-1415) and by the SIG program at the University of Toronto. Note that the term N-initial, like V-initial, does not literally mean that the Ν is the first element in the phrase, since it is preceded by functional elements such as case markers and articles (cf. tense and aspect markers in the case of V). Rather, it indicates that the Ν (or V) generally precedes substantive material such as arguments and modifiers. Abbreviations used in the glosses of this chapter are as follows: ABS = Absolutive, c = Common, COMP = Complementizer, DEM = Demonstrative, DU = Dual, ERG =Ergative, GEN = Genitive, LIG = Ligature, Ρ = Proper, PERS = Personal Article, PL = Plural, PRED = Predicative. This marking of the proper / common distinction runs through the entire case system in Niuean (and other Polynesian languages). For example, an ergative common DP begins with the particle he, and an ergative proper DP begins with the particle e, and in (2), the genitive proper form ha is used, whereas in (4), the genitive common form he is used. There are many interesting issues in the Niuean case / article system but we cannot explore them further in this short chapter. The conjunction of number and classifiers is common. For discussion, see Fassi Fehri and Vinet, to appear. This decision has consequences, since number is usually considered to merge low, whereas features such as definiteness are usually considered to merge high. The true nature of these markers is thus worthy of further investigation. Note that, as discussed in Massam and Sperlich (2000), Niuean does not exhibit the well-known alternation between dominant and non-dominant genitive forms, as found in other Polynesian languages. (For discussion, see papers in Fischer 2000). (3) and (4), along with other examples in this chapter appear with a particle ko, which we gloss as "Pred". The function of ko is not uniformly represented in the translations of the sentences. Generally speaking, the particle ko appears before nominals which are not in an argument position (eg. moved FFft-NPs, topics, focused NPs, predicative NPs). When ko precedes a common NP it is

144

8. 9.

10.

11. 12.

13.

14.

15.

Arsalan Kahnemuyipour and Diane Massam followed by the particle e which we have not glossed separately in this chapter. See also Hawkins (1983). While it remains true that many logically possible orders are unattested, the generalization in (6) is clearly too restricted, since in fact some other orders are presented in the literature, as discussed in Cinque (2005). Since Niuean appears, for the most part, to conform to (7) in being a fully inverse language (7d), we need not discuss the exceptions here. Cinque in later work (e.g. Cinque 2005) revises the analysis of the order in (7c) so that it also involves XP movement, rather than head movement, but since (7c) is not the order for Niuean we do not pursue this issue here. Note that we do not assume that adjectives are adjoined, but rather that they form part of the selectional phrasal hierarchy. This type of movement (complement-to-specifier) has been considered illicit by some linguists. We simply note here that if we wanted to avoid such movements, we could posit a null head and specifier between each phrase, in the style of Cinque (2005) and Pearce (2002). We do not do this here (following Rackowski and Travis 2000), in part because the derivations are simpler to display without these intermediate nodes. Another question involves the motivation for such movements. We follow Cinque (2005) in assuming that there is a need for the lexical head to move to the top of the extended projection, although the nature of this requirement remains to be fully explored. It is not fully clear why DP does not move to the specifier of KP. This may, however, be tied to the movement of D to K, realized as a portmanteau morpheme [Case+P/C], as discussed previously. Under this view, the requirement in Κ that is triggering the movement is satisfied by the movement of the head D, rather than by movement of the phrasal DP. An anonymous reviewer considers it unlikely that taha is actually bimorphemic (with ta in specifier and ha in head position), since it is not clear what each morpheme would mean, and since the cognate form in Tongan taha is a numeral "one", whereas ha is an indefinite / nonreferential article, and they can co-occur as in ha taha "whatever, anyone". We recognize the validity of taha as a single vocabulary item (for example it can undergo reduplication in Niuean), but leave open the possibility that in Niuean it consists of two separate feature bundles at Merge (in the sense of Distributed Morphology, cf. Halle and Marantz 1993). Note that the GenP morpheme in (12c) is a, and not ha as given in (2a,b) above. According to Seiter (1980: 84), a is in variation with ha as a genitive proper case marker, with a being used more frequently by older speakers. For further discussion of the syntax of these and other morphemes across Polynesian, see Clark (1976).

Patterns of Phrasal Movement

145

16. The particle system in Niuean is rich and complex. For more information on the particles, see Rackowski and Travis (2000), Seiter (1980), or Massam (2000a). 17. An anonymous reviewer points out that in some cases Poss appears to bear content, rather than being like an Agr node (Barker 1995). For example, in Mayan, there is an intricate system of morphemes indicating the alienable / inalienable status of the possessor (Lehmann 1998). Similarly, closer to home for Niuean, in other Polynesian languages the genitive marker indicates whether the possessor is dominant over the possessed item or not (Fischer 2000), a relational concept which might involve features in the Poss head. In Niuean, however, there is no indication that the possessive head bears semantic features. The issue comes down to whether there are indeed heads whose function is to allow for a more-or-less thematically empty relation of predication to hold between an element in their specifier and complement positions, which are distinct from Focus phrases, Topic phrases etc. The view in this chapter is that there are such heads. 18. See den Dikken (2006) for a different approach to word order and to linking items. 19. In some cases, consultants indicate a preference for there to be no # marker if there is a numeral, but it is clear that there is no outright constraint on the cooccurrence of these elements, since many examples exist where they do cooccur, as in (ia,b) (Niue: A History of the Island), (i) a. ke he tokoua e na: tagata Manu 'a ko Ve 'u mo Ve 'a about two LIG DU person Manu'a PRED Ve'u and Ve'a 'about two Manu'an men, Ve'u and Ve'a.' b. e tau tupua tokolima ABS.C PL tupua five 'the five tupua' (=ancient legendary creature) 20. In elicitation, the order [Numeral A] is also accepted as in (i.a) (Field Notes 2001) and (i.b) (de Sousa.2001: 35). (i) a. e tau manu ua kula fulufuluola e: ABS.C PL bird three red beautiful DEM 'those three beautiful red birds' b. tau ika ua lalahi e: PL fish two big DEM 'these two big fish' The status of this order needs verification, however. Cinque (p.c.) suggests the adjectives here may be predicative. We note that it can be derived by movement of NP to specifier of A, then movement of the same NP to specifier of Num, followed by movement of NumP to specifier of Dem. Similar movement variations are discussed in Cinque (2005), while Pearce (2002, 2003) and Coddington (2003) discuss similar word order variations in Maori.

146

Arsalan Kahnemuyipour and Diane Massam

21. Examples such as (22) and (23) with a relative clause modifier, and examples with a PP argument, as in (23), along with deverbal nominal clauses (described in Seiter 1980 and Massam 2000b) raise further questions about word order in Niuean nominal phrases. Our assumption here is that such arguments are generated high (Kayne 2001), and that movement occurs over them so that they end up on the right of the phrase. For some discussion, see Cinque (2005). A farther question for nominal structure in Niuean, as noted by an anonymous reviewer, is "Raising to Possessor", i.e. the Genitive Relative Construction in Niuean (Seiter 1980, Herd, Massam, and MacDonald 2005). We leave these issues for future research.

References Abney, Steven 1987 The English Noun Phrase in its Sentential Aspect. Ph.D. diss., MIT. Barker, Chris 1995 Possessive Descriptions. Stanford, CA: CSLI Publications. Belletti, Adriana (ed.) 2004 Structures and Beyond - The Cartography of Syntactic Structures. Vol 3. Oxford: OUP. Bittner, Maria and Ken Hale 1996 The Structural Determination of Case and Agreement. Linguistic Inquiry 27: 1-68. Blanc, R. R. V. (and Togakilo - translator) 1965 Ne Toka Hifo e Kuki e Higoa Haana he Tau Aelani. [Captain Cook Leaves his Name in the Islands] Wellington, New Zealand: Islands Education Division, Dept. of Education. Borer, Hagit 1999 Deconstructing the Construct. In Beyond Principles and Parameters, Kyle Johnson, and Ian Roberts (eds.), 43-89. Dordrecht: Kluwer. Carnie, Andrew, and Eithne Guilfoyle, (eds.) 2000 The Syntax of Verb Initial Languages. Oxford: OUP. Cinque, Guglielmo 1996 The Antisymmetric Programme: Theoretical and Typological Implications. Journal of Linguistics 32: 447^464. 1999 Adverbs and Functional Heads: A Cross-linguistic Perspective. Oxford: OUP. 2000 On Greenberg's Universal 20 and the Semitic DP. The University of Venice Working Papers in Linguistics 10 (2): 45-61. 2005 Deriving Greenberg's Universal 20 and its Exceptions. Linguistic Inquiry 36: 315-332.

Patterns of Phrasal Movement

147

Cinque, Guglielmo (ed.) 2002 The Functional Structure of DP and IP - The Cartography of Syntactic Structures. Vol. 1. Oxford: OUP. Clark, Ross 1976 Aspects of Pro to-Polynesian. Auckland: Linguistic Society of New Zealand. Coddington, Anna 2003 DP Internal Movement in Maori. Ms., University of Auckland. den Dikken, Marcel 2003 The Structure of the Noun Phrase in Rotuman. LINCOM Studies in Austronesian Linguistics 05, München, Germany. 2006 Relators and Linkers: The Syntax of Predication, Predicate Inversion and Copulas. Cambridge: MIT Press. Fassi Fehri, Abdelkader, and Marie-Therese Vinet to app. Distribution of Number and Classifier in Arabic and Chinese and Parameterization. Linguistic Research 9.1. IERA Publications. Rabat. Fischer, Steven (ed.) 2000 Possessive Markers in Central Pacific Languages. Thematic volume of: Language Typology and Universals 53.3/4. Ghomeshi, Jila, and Elizabeth Ritter 1996 Binding, Possessives, and the Structure of DP. In Proceedings of NELS 26, K. Kusumoto (ed.), GLSA, University of Massachusetts. Greenberg, Joseph 1966 Language Universals: With Special Reference to Feature Hierarchies. The Hague: Mouton. Halle, Morris, and Alec Marantz 1993 Distributed Morphology and the Pieces of Inflection. In The View from Building 20, Kenneth Hale, and S. Jay Keyser (eds.), 111-176. Cambridge: MIT Press. Hawkins, John A. 1983 Word Order Universals. New York: Academic Press. Herd, Jonathon, Diane Massam, and Catherine MacDonald 2005 Genitive Relative Constructions in Polynesian. In Proceedings of the 2004 Canadian Linguistics Association Annual Conference, MarieOdile Junker, Martha McGinnis, and Yves Roberge (eds.), 12 pages. http://www.carleton.ca/~mojunker/ACL-CLA/ Kayne, Richard 1994 The Antisymmetry of Syntax. Cambridge: MIT Press. 2001 Prepositions as Probes. Ms., New York University.

148

Arsalan Kahnemuyipour and Diane Massam

Laenzlinger, Christopher 2000 French Adjective Ordering: Perspectives on DP-internal Movement Types. Generative Grammar in Geneva 1: 55-104. 2005 French Adjective Ordering: Perspectives on DP-internal Movement Types. Lingua 115: 645-689. Lehmann, Christian 1998. Possession in Yucatec Maya: Structures - functions - typology. Unterschleissheim: Lincom Europa. (LINCOM Studies in Native American Linguistics, 4). Massam. Diane 2000a VSO is VOS: Aspects of Niuean Word Order. In Andrew Carnie, and Eithne Guilfoyle, (eds.), 97-117. 2000b Niuean Nominalization. In Proceedings of AFLA 7: The seventh meeting of the Austronesian Formal Linguistics Association. Marian Klamer (ed.), 121-132. Amsterdam: Vrije Universiteit. Massam, Diane, and Wolfgang Sperlich 2000. Possession in Niuean. In Possessive Markers in Central Pacific Languages. Thematic volume of Language Typology and Universals, Steven Fischer (ed.), 53.3/4: 281-292. Megerdoomian, Karine 2002 Beyond Words and Phrases: A Unified Theory of Predicate Composition. Ph.D. diss., University of Southern California. Nelisi, Lino 1995 Ko e Letio Kula. Wellington: Ministry of Education. Nine: A History of the Island 1982 edited by the Institute of Pacific Studies of the University of the South Pacific. The Government of Niue. Pearce, Elizabeth 2002 DP Structure and DP Movement in Maori. Paper presented at COOL5, University of Canberra, Australia (Ms., Victoria University of Wellington). 2003 Phrasal movement within the Maori DP. Digests of Selected Papers Presented at AFLA X. University of Hawai'i at Manoa Working Papers in Linguistics 34 (2): 41-42. Rackowski, Andrea, and Lisa Travis 2000 V-initial Languages: X or XP Movement and Adverb Placement. In Andrew Carnie, and Eithne Guilfoyle (eds.), 117-142. Ritter, Elizabeth 1988 A Head Movement Approach to Construct-State Noun Phrases. Linguistics 26: 909-929.

Patterns of Phrasal Movement 1991

149

Two Functional Categories in Noun Phrases: Evidence from Modern Hebrew. In Perspectives on Phrase Structure: Heads and Licensing. Susan. D. Rothstein (ed.), 37-62. (Syntax and Semantics 25.) New York: Academic Press. Where's Gender? Linguistic Inquiry 24 (4): 795-803.

1993 Rizzi, Luigi 1997 The Fine Structure of the Left Periphery. In Elements of Grammar, Liliane Haegeman (ed.), 281-337. Dordrecht: Kluwer. Rizzi, Luigi (ed.) 2003 The Structure of CP and IP - The Cartography of Syntactic Structures. Vol. 2. Oxford: OUP. Schoorlemmer, Maaike 1998 Possessors, Articles and Definiteness. In Possessors, Predicates and Movement in the Determiner Phrase, Artemis Alexiadou, and Chris Wilder (eds.), 55-86. Amsterdam: Benjamins. Scott, Gary-John 1998 Stacked Adjectival Modification and the Structure of Nominal Phrases. SOAS Working Papers in Linguistics and Phonetics, Vol. 8, 59-89. Seiter, William 1980 Studies in Niuean Syntax. New York: Garland Press. Shlonsky, Ur 1988 Government and Binding in Hebrew Nominals. Linguistics 26: 951— 976. Shlonsky, Ur 2004 The Form of Semitic Noun Phrases. Lingua 114 (12): 1465-1526. de Sousa, Hilario 2001 Noun Phrase Structure and the Case Marking System in Niuean. Ms., University of Auckland. Sperlich, Wolfgang 1997 Tohi Vagahau Niue / Niue Language Dictionary. Honolulu and Alofi: University of Hawai'i Press and the Government of Niue. Sproat, Richard, and Chilin Shih 1991 The Cross-linguistic Distribution of Adjective Ordering Restrictions. In Interdisciplinary Approaches to Language. Essays in Honor of S. Y. Kuroda, Carol Georgopoulos, and Roberta Ishihara (eds.), 565593. Dordrecht: Kluwer. Stowell, Tim 1983 Subjects Across Categories. The Linguistic Review 2 (3): 285-312. Travis, Lisa 1992 Inner Tense with NP: the Position of Number. CLA Proceedings. University of Toronto Working Papers in Linguistics, 329-346.

Rigidity versus Relativity in Adverbial Syntax: Evidence from Tagalog* Daniel Kaufman

Two diametrically opposed stances have emerged from recent theoretical debates on adverbial syntax. One approach, represented by Alexiadou (1997) and Cinque (1999), espouses a rigid hierarchy of functional projections hosting individual adverbs. The other, represented broadly by Jackendoff (1972), McConnell-Ginet (1982) and most recently Ernst (2002), takes adverb placement to be determined by the semantics of the adverbs themselves as opposed to the functional architecture of the clause. Under the latter view, adverbs may be divided into several categories based on their meaning with each category being licensed in a certain range within the sentence. Here, I undertake a detailed examination of Tagalog adverbs and compare the predictions of the two best articulated recent theories of adverbs, that of Cinque (1999, 2004) and Ernst (2002). The results offer support for some of the basic predictions of the semantically based approach of Ernst. Particularly important are scopal facts which do not obtain a clear explanation under a functional projection-based theory such as Cinque's.

1.

T w o theories of adverbs compared

Because proponents of both theories considered here employ certain adhoc mechanisms for handling exceptions - and are thus able to achieve similar empirical coverage - naturalness must figure prominently as an evaluation metric. 1 Since each theory is suited to naturally handle a well circumscribed set of phenomena the theories under consideration must be evaluated in terms of how well the core predictions are corroborated by the data. For the purposes of this chapter I will refer to the Alexiadou-Cinque theory as the Rigid Approach and Ernst's theory as the Relative Approach}

152

Daniel Kaufman

1.1. Fundamentals and predictions of the Rigid Approach Cinque (1999) bases his proposal on the notion that adverb order is much stricter than can be predicted by scopal properties and semantic domains. He offers evidence to show that ordering relations between adverbs are fixed and transitive. That is to say, if the linear relations between adverbs A, Β and C are such that A>B and B>C then it is possible to conclude that A>C. Two central tenets of Cinque's proposal are: (i) Adverbs are not adjuncts but rather specifiers within an array of strictly and universally ordered functional projections; (ii) Each adverb corresponds to an inflectional head which carries an analogous meaning. Thus, the same order is reflected by verb morphology. Strong evidence for the specifier hypothesis is the relation between single adverbs and the verb in several Romance languages. In these languages, the verb may appear on differing sides of an adverb based on whether the verb is finite, infinitival, participial, etc. This suggests that the verb crosses certain adverbs when raising to an inflectional head. This phenomenon appears to require a purely syntactic explanation since scopally, verbs are not often found to interact with single adverbs in the same way that multiple adverbs interact with each other.3'4 In support of a Spec-Head relationship between adverbs and verbal morphology, Cinque offers data from a wide range of languages to show that the order posited for adverbs is reflected (in reverse) in the domain of the word. Thus, an aspectual adverb is found in the specifier of a functional projection such as PerfectiveP while its corresponding verbal affix constitutes the head of this phrase. Unexpected surface orders (i.e. those not corresponding to the underlying hierarchy of FPs) result from one of the following (Cinque 1999: 3-4): (i) "When an AdvP directly modifies (is the specifier of) another AdvP."; (ii) "When a lower portion of the clause (containing an AdvP) is raised across a higher AdvP (for focus-presupposition requirements)."; (iii) "When one AdvP is wh-moved across another."; (iv) "When one and the same AdvP can be 'base generated' in two different positions in the clause (with one of the two positions to the left, and the other to the right of another AdvP)."; (v) "When a noninherently 'focusing' AdvP (e.g. probably) is used as a 'focusing' adverb (like only and simply)."·, (vi) "When an adverb is used 'parenthetically'." As Cinque notes, (iii) and (vi) are uncontroversial because they are apparent from surface form and tied to

Rigidity versus Relativity in Adverbial Syntax

153

more general phenomena. The other exception-creating circumstances, however, are more contentious in that they are less detectable and more specific to adverbs. The core predictions of the Rigid Approach may be summarized as the following: (i) Ordering of adverbs should be highly restricted both crosslinguistically and within single languages; (ii) Deviant orders should show the hallmarks of XP movement to a higher Spec of a functional projection (esp. FocP, TopP); (iii) Scope may often not correspond to linear order since there need not be a direct link between the scope of an adverb and its position in relation to other elements. To give an example of what prediction (ii) entails, we can observe the following two locality arguments presented by Cinque (2004) (citing Rizzi 2002). First, in Italian movement of adverbs to the pragmatically unmarked ModifierP (Rizzi 2002) appears to be subject to Relativized Minimality effects (1-2).5 (1)

Rapidamente, qualcuno far a sparire i documenti. 'Quickly, someone will make the documents vanish'

(2)

* Rapidamente, qualcuno probabilmente farä sparire i documenti. 'Quickly, someone will probably make the documents vanish'

Second, this displacement appears to be clause-bounded (3) (Cinque 2004: 703 ex. 39-41), unlike topicalization and focalization (4-5) (we return to this data in section 4). (3)

* Rapidamente, credo che qualcuno farä sparire i documenti 'Quickly, I think that someone will make the documents vanish.'

(4)

Rapidamente, credo che nessuno farä sparire i documenti. 'QuicklyT0P, I think that nobody will make the documents vanish.'

(5)

RAPIDAMENTE, credo che qualcuno farä sparire i documenti. 'QuicklyF0C, I think that someone will make the documents vanish.'

In regard to prediction (iii), Cinque suggests that adverbs take scope from their base position and that adverbs which show variable scope have multiple base positions (Cinque 1999: 25-28). Thus, allowing for movement, support for the Rigid Approach should consist of cases where ad-

154

Daniel Kaufman

verbs can obtain several scopal interpretations from a single position in the linear order and conversely, from the absence of syntax-scope pairings that cannot be derived from the underlying order (see section 3.2 below).

1.2. Fundamentals and predictions of the Relative Approach Ernst (2002) develops a theory of adverbs in the former tradition of treating them as adjunctions. Applying a more articulated theory of propositional semantics (based largely on Parsons 1990), Ernst is able to restrict the positions of adverbs according to what type of event type they may modify. These types are referred to as Fact-Event-Objects (FEOs) and are treated as the semantic arguments of adverbs. Based on their semantic class, adverbs are specified to take particular FEO types and are capable of shifting that type as a result of composition. Composition is a step-wise process and layers are added under syntactic sisterhood. The basic FEO types / layers employed are the following (see Ernst 2002, Parsons 1990 for precise definitions): (6)

Speech-Act > Fact > Proposition > Event > Specified Event

The constraints on how these types combine, together with ostensibly universal syntactic principles, are understood to be responsible for the distribution of adverbs within the clause. An adverb may be adjoined freely in the syntax but will be uninterpretable if it cannot access the semantic layer it requires from its position in the sentence. This happens, for instance, when a lower level adverb (e.g. Manner) is composed after a higher level adverb (e.g. Speech Act). A bare verb is associated only with the lowest FEO type which must eventually be built into a Speech Act by the point that the sentence is complete. The correspondence between syntactic phrases and FEO types need only be constrained minimally. Adverb scope is handled via the following Scope Principle·. "An operator A has scope over an operator Β in case A c-commands a member of the chain containing B." (Ernst 2002: 317). Assuming the possibility of right adjunction, adverbs are thus expected to show concentric scope effects around the verb. Total symmetry is mitigated by several factors. One is the special syntactic status of clausal predicational adverbs (e.g. "Clearly, he's not in the building."). According to Ernst, clausal predicational adverbs are subject to the same syntactic

Rigidity versus Relativity in Adverbial Syntax

155

restrictions as predicational heads and thus follow similar ordering principles. This accounts for the fact that these adverbs regularly precede the verb cross-linguistically. Non-predicational adverbs are not bound by such syntactic restrictions and therefore display a wider range of potential positions. Adverb movement falls under the scope of more generally accepted cases of movement such as topicalization, w/j-movement, clefting, etc. Under this view, then, there is no independent motivation for adverb movement and thus it should not be considered a separate "type" of movement on par with those mentioned above. Nor should there exist special phrases to host adverbs such as Rizzi's (2002) ModifierP. 6 For Ernst, the only strict correspondence between FEO types and specific syntactic projections is that only event-internal modification is possible within the "Lsyntax" (i.e., corresponding to VP pace Hale & Keyser 1993). This entails that adverbs adjoined within VP can only receive limited interpretations (e.g. manner, measure). Ernst proposes a semantically motivated taxonomy of adverbs which have their own FEO selection requirements and which may yield different FEO types upon composition. For example, epistemic adverbs such as probably may be represented as [ F A C T A D V [ P ] ] indicating that they take a Proposition as their complement and yield a Fact (a Proposition whose truth is asserted, cf. Parsons 1990). The building up of propositions and the composition of adverbs follows the FEO calculus. The primary principle of this is that any FEO type may be freely converted to a higher FEO type but not lowered (Ernst 2002: 50, 2004: 761). In principle, the FEO type of a sentence fragment may be lowered through composition but this is very rare. In general, adverbs and operators either raise the FEO type or maintain it. Because the FEO type may be freely raised without overt operators, an adverb that selects for a higher FEO type such as Fact or Speech-Act need not occur at a syntactically designated peripheral position but can rather be accommodated in more internal positions through FEO type raising (modulo VPinternal positions, the domain of event-internal modification). The manner / evidential ambiguity with adverbs such as clearly as in (7) results from the fact that manner adverbials are composed via the Manner Rule. This is a rule which obligatorily applies to predicational adverbs within L-syntax and which may optionally apply to constituents of the Event type (although a clausal reading is preferred outside of VP). 7 P R O

156

Daniel Kaufman

(7)

John clearly saw everything written on the chart.

The relevant predictions of the Relative Approach are summarized as the following: (i) An adverb that selects for a lower FEO type should not appear higher than one which selects for a higher FEO type; (ii) The positions in which an adverb is licensed should constitute a contiguous range w.r.t. non-FEO-type changing material; (iii) Adverb scope should be reflected by surface order in a concentric fashion with the predicate as the center due to the symmetric manner in which semantic layers are built in the FEO calculus.

2. Tagalog adverbs Tagalog offers a good testing ground for the predictions of the two theories as the syntactic position of Tagalog adverbs is elucidated by several overt markers which are used to introduce them (cf. Schachter and Otanes 1972: Chap. 6). Adverbs modifying the lower FEO types are marked with nang, the same particle employed to introduce internal arguments.8 Manner and frequency adverbials are typically introduced in this way as seen in (8). The position of the manner adverb in (8) is free within the post-verbal domain. The various positions are interpreted uniformly (although the "focus set" in the sense of Reinhart 1995 may differ). 9 (8)

a. Pag-aral-an (nang maigi) ng piloto (S) TR-study-LV nang well GEN pilot 'The pilot studied the map thoroughly.' b. Pasok (nang madalas) sa opisina (S) enter nang often OBL office 'Ben went to the office often.'

ang mapa (S) SUB map si

Ben (S) P.SUB B.

Adverbs may also be introduced in the same way as topics, that is, in clause-initial position followed by the marker ay. Compare DP topicalization in (9) with (10). (9)

Ang tubero ay nag-trabaho sa kusina. SUB plumber TOP AV.PRF-work OBL kitchen 'The plumber, worked in the kitchen.'

Rigidity versus Relativity in Adverbial Syntax (10)

a. Kadalasan usually 'Usually, he b. Malamang probably

ay hindi siya pa~pasok TOP NEG 3SG.SUB PROG~enter doesn't come to class.' ay nan-day a sila. TOP

AV.PRF-cheat

157

sa klase. OBL class

3PL.SUB

'They probably cheated.' Whereas we see from the above that adverbs may be marked like arguments, it is also possible to introduce adverbs in the same way as adjectives, that is, directly adjacent to the modified head with the linker nal- . Compare (11a) and (1 lb). (11)

a. Madalas na pag-ulan frequent LNK NMLZ-rain 'Frequent rain'

b. Madalas na u~ulan. frequent LNK PROG~rain 'It rains frequently.'

When adverbials are attached with the linker, phrasal arguments may not generally intervene between the adverbial and the verb, as in (12). (12)

*? Pa~pasok PROG~enter

ang

abogadong

SUB

lawyer.LNK o f t e n

madalas

Adverbial notions are also often introduced syntactically as predicate adjectives (13). Strictly speaking, these should not be considered adverbials since such constructions are indistinguishable from non-verbal predications. However, the fact that certain adverbial notions are regularly expressed in this way merits mention.10 (13) Malinaw na clear

mag-sa~salita

sila

sa

COMP AV-IRR~speak 3PL.SUB OBL

miting. meeting

'Clearly, they will speak at the meeting.' (or, 'Its clear that...') Other adverbs may be introduced into the clause without any marker at all. These adverbs are typically mono-morphemic and are often optional second position clitics. The adverb uli 'again', (14), exemplifies this class. (14) Batikos (uli) siya (S) ng mga criticize again 3SG.SUB GEN PL 'He was criticized again by the teachers.'

guro (S) teacher

158

Daniel Kaufman

Finally, there is a class of aspectual, mood, evidential and functional adverbs which are robust clitics restricted to appearing in second position (15). Certain members of this class, such as sana 'OPT' (15c), may also appear in topic position. (15)

a. Balik

na

nga

return already

po

pala

EMPHPOL

'They really already returned!' b. Mag-ta~tago pa rin ba kaya AV-LRR~hide

still

also

TOP

kayo?

Q RHET 2PL.SUB

Ί wonder, will you still hide?' c. (Sana ay) gradweyt OPT

silal

SURP 3PL.SUB

graduate

(sana)

siya

OPT

3SG.SUB n o w

ngayon.

'Hopefully, he'll graduate.' We are now in a position to look at the relation between adverb types and positions in Tagalog. Sentences (16a-d) show the positional possibilities of an ambiguous clausal / manner adverb. (16) Malinaw 'clear(ly)':

(i) Manner reading - [ P C E V E N T ADV [ S P E C E V E N T ] ] (ii) Evidential reading - [STATE ADV [ f a c t ]] a. [Tp Ka~kantai kaj [Vp t-t tj nang malinaw ]] PROG~sing 2SG.SUB nang clear 'You're singing clearly.' {Φ 'Clearly, you're singing.') b. [TPMalinaw kangj [TP ka~kantai [VPI, (/]]] clear

S

2SG.SUB:LNK

E

PROG~sing

'You're singing clearly.' (ψ 'Clearly, you're singing.') c. [χορρMalinaw ay [TP ka~kantai kaj [Vpi; {/]]] clear

TOP

PROG~sing 2SG.SUB

'Clearly, you're singing.' 'You're singing clearly.') d. [TPMalinaw [cp na [Tp ka~kantai / C A [ P / , {/]]]] 7

clear

COMP

V

PROG~sing 2SG.SUB

'Clearly, you're singing.'

'You're singing clearly.')

The adverb in (16a) represents the nang (internal argument-like) attachment showing that only a manner reading is possible from this position; (16b) shows the adverb in the verb-adjacent, adjective-like position11; (16c) exemplifies the adverb in topic position; and finally in (16d) an adverbial meaning is obtained through an adjectival predicate with a clausal

Rigidity versus Relativity in Adverbial Syntax

159

complement. Note that there is a difference in the readings between (16a-b) on the one hand and (16c-d) on the other hand. In the former group the adverb can only obtain a manner interpretation while in the latter group the adverb can only obtain a clausal interpretation. Generally, the argumentlike nang attachment of adverbs necessitates a "lower" reading (e.g. manner, measure, frequency). The positional possibilities for a modal adverb are shown in (17a-d). (17)

Malamang 'probably': modal adverb - [FACT ADV [PROp]] a. * [Tp Ka~kantai siyaj [Vp U tj nang malamang ]]] PROG~sing 3SG.SUB nang probably b. * [FOCPMalamang siyangj [TP ka~kantai [Vi>/, probably

3SG.SUB:LNK

PROG~sing

c. [χορρ Malamang ay [TPka~kantai siyaj [Vp /,· tj]\] probably TOP PROG~sing3SG.SUB 'Probably, he's singing.' d. [tpMalamang [Cp na [TP ka~kanta, siyaj [ V pi, probably COMP PROG~sing 3SG.SUB 'Probably, he's singing.' Table 1 (see next page) shows a summary of the possible positions for the adverb classes discussed here (m = manner reading; c = clausal, epi = epistemic, deo = deontic). What emerges clearly from Table 1 is that the clause appears to be bifurcated at the edge of TP. Differences in interpretation and grammaticality are found to relate to the boundary marked by the topic position. As seen in (a-c), manner readings are only found within TP. The deontic reading of modals is at least favored if not obligatory within TP while a number of adverbs cannot appear within TP at all. The facts in Table 1 support two of Ernst's claims: that manner readings are restricted to (but free within) a lower domain and that clausal predicative adverbs are linearized in relation to their FEO complements in accordance with the general head-complement pattern of head-initial languages. This latter claim is supported by the fact that clausal interpretations can only be obtained in a peripheral position, preceding the entire TP complement. 12

160

Daniel Kaufinan

Table 1. Summary of adverb type-position relations [τορρ ΑΡΥ ay [η· ADV (clitic) [VP nang ADV]]] a. Measure i. lubusan 'completely' b. Pure manner i. mahigpit 'tight(ly)' c. Manner / clausal i. malinaw 'clear(ly)' ii. bigla 'suddenly' d. Exocomparative i. ganito 'like this' e. Subject oriented i. kusang-loob 'willingly' f. Functionalquantificational i. uli 'again' ii. madalas 'often' iii. kadalasan 'most often' g. Modal i. dapat 'should' ii. malamang 'probably' h. Speech act i. sa madaling salita 'briefly'

S

*

*

m

•f m

•f m ·/ m

•/ m •S m

S m

•/ m

•/

•/

V •/ ·/

•S epi S

*

*

S deo

*

*

*

*

*

3. Evidence for Relativity While the facts shown in the previous section suggest a good deal of freedom, they do not provide an iron-clad argument for relativity in and of themselves since there always exists the possibility of movement. In this section I will provide evidence showing that certain ungrammaticalities in Tagalog not predicted by the Rigid view cannot be explained by resort to common constraints on movement.

Rigidity versus Relativity in Adverbial Syntax

161

3.1. Syntactic locality versus the FEO calculus As seen above, a frequency adverb such as madalas 'often' may be introduced in several different positions in the clause without a change in the basic meaning (18).13 (18)

a. u~ulan

nang madalas dito.

PROG~rain nang 'It rains often here.'

b. Madalas ay often

u~ulan

TOP

c. Madalas na often

often

dito.

PROG~rain here

u~ulan

LNK

here

dito.

PROG~rain here

Crucially however, these positions are limited with the presence of another adverb. The modal clitic adverbs sana 'hopefully' and yata 'perhaps' appear in the second position of the clause as in (19a) and (19b). (19)

a. Bisita

sana si

visit

OPT

'Hopefully, Juan visited.' b. Bisita yata si visit

Juan.

P.SUB J.

Juan.

EPST P.SUB J.

'Perhaps Juan visited.' Whereas a frequency adverb such as 'often' may be found in topic position when it occurs alone as in (18b) above, the presence of a modal adverb seems to block the possibility of madalas in topic position as seen in (20). (20)

a. Madalas ay often

TOP

bisita

si

visit

P.SUB J.

'Juan visited often.' b. *Madalas ay bisita often

c. *Madalas often

Juan.

sana si

Juan.

TOP

visit

OPT

P.SUB J.

ay

bisita

yata

si

TOP

visit

EPST P.SUB J.

Juan.

That this is not a categorical constraint on the co-occurrence of these adverbs is seen from (21) where the frequency adverb is attached within TP.

162

Daniel Kaufinan

(21) Bisita visit

(sana /yata) si OPT

Juan nang madalas.

EPST P.SUB J.

nang

frequent

'(Hopefully / perhaps,) Juan visited often.' Unlike the similar examples in (1-5) above cited by Cinque (2004), no elegant Relativized Minimality appears possible here. First, however these adverbs are characterized, it is doubtful that yata can be the same kind of element as madalas since madalas can appear in a spec position, e.g. as a topic in (20a), and may be modified, while yata cannot (masyadong madalas 'too often', but *masyadong yata). Second, as a prosodically placed second-position clitic, yata is ordered at PF and therefore should not interfere with syntactic movement. These facts are expected under Ernst's analysis which does not rely on movement but rather on the FEO calculus to rule out sentences such as (20b) and (20c). Specifically, the presence of modal adverbs (yata, sana) convert the FEO type to Fact while a frequency adverb (madalas) requires a lower type (e.g., Event / Proposition). The prosodic domain within which the clitics are positioned is the TP, excluding the Topic. It is therefore unambiguous that the adverb in topic position must be composed after clitics within TP.

3.2. Concentric scope phenomena Key evidence for Ernst's theory comes from facts about relative scope, in particular, data which shows that outer adverbs tend to scope over inner adverbs on both sides of the predicate and have ambiguous scope relations when flanking the predicate. The predictions of symmetric semantic composition are borne out by the Tagalog data with one caveat which leads us to revise Ernst's original theory. Observe the relative scope of bigla 'suddenly' and lagi 'always' within the TP (22) (see next page). When both adverbs precede the verb, the leftmost adverb must scope over the adverb to its right. This is trivially predicted by most theories of adverbials. What is not handled equally well though is the derivation of the ambiguity in (23) (next page), where two adverbs "sandwich" the predicate. This shows that adverb scope does not necessarily abide by an asymmetric left-to-right hierarchy.

Rigidity versus Relativity in Adverbial Syntax

(22)

a. Biglang

laging

suddenly:LNK

na-lu~lungkot

si

163

Juan

always:LNK STA-PROG~sad P.SUB J.

' Suddenly, Juan is always sad.' suddenly>always; *always suddenly) b. Laging biglang na-lu~lungkot always:LNK suddenly:LNK

si

Juan

STA-PROG~sad P.SUB J.

'Juan is always suddenly sad.' (*suddenly>always; Salways suddenly) (23) Biglang suddenly:LNK

na-lu~lungkot

si

Juan

STA-PROG~sad P.SUB J.

lagi always

'Suddenly, Juan is always sad.' or 'Juan is always suddenly sad.' The similar English sentences in (24) (Andrews 1983) have been widely discussed in relation to this type of ambiguity. (24)

a. John twice intentionally knocked on the door. (twice>intentionally) b. John intentionally twice knocked on the door. (intentionally>twice) c. John knocked on the door intentionally twice. (,twice>intentionally) d. John knocked on the door twice intentionally. (iintentionally>twice) e. John intentionally knocked on the door twice, (ambiguous)

Andrews suggests that the interpretive differences are a result of V' adjunction in combination with the rule, "Apply an adverb to what it is sister o f ' . Cinque (1999: 25) reanalyzes the original adjunction derivation along the lines of his functional projection theory. To capture the scope facts, Cinque posits the existence of two separate base positions for twice, a higher position with an "iterative" meaning, indicating several events, and a lower position with a "repetitive" meaning, indicating several repetitions within a single event. These meanings are retained regardless of movement. The base positions of the adverbs are seen in (25) while the derivations in (26) show how the relative scope between these adverbs is determined. (25) John (twice)ITR [χρ intentionally [YPknocked (twice)REPon the door ]]

164

Daniel Kaufman

(26) a. b. c. d. e. f.

John twice [ intentionally knocked on the door ] (=24a) ?? John intentionally twiceITR knocked on the door (=24b) John [[ knocked on the door ]j intentionally tj\x twiceITR tt (=24c) John [ knocked [ on the door ]\ twiceREP tt ]j intentionally iy(=24d) John [ intentionally [ knocked on the door twicerep ]] (=24e) John [ intentionally knocked on the door ]j [ twiceITR ti ] (=24e)

Free phrasal movement should derive all the possible readings and none of the unattested readings but several problems show this is not so straightforward. First, (26b) is not given a clear explanation. Markedness aside, (26b) strongly tends to be interpreted as intentionally>twice. Cinque, however, is forced to derive this with low-scope twiceITR because movement of twice REP to the left of the verb does not fall into any of his categories of permitted adverb movement (cf. section 1.1). To get the correct order, intentionally is generated as a "framing adverb" in the left periphery, the subject then moving around it. It remains unclear how the correct interpretation is derived and what other adverbs may be subject to a "framing adverb" analysis. Another problem is that (26d) is incorrectly predicted to be scopally ambiguous. If the YP in (25) can raise above twice m (as claimed) then both interpretations are predicted to be available depending on which twice is present. (27) John [ knocked [ on the door ]j (twice)Rßpti ]j (twice)!TR intentionally tj By positing the two projections above for twice, Cinque seems to mistake a simple scope alternation for an orthogonal distinction in frequency. The reason that twice>intentionally may be confused with an iterative meaning is that it implies two separate intentions and thus two separate events. The opposite scope, intentionally>twice, cf. (24d), contains no such implication and thus is ambiguous in terms of a single versus double event of knocking. Crucially though it is not ambiguous in terms of scope. In other words, regardless of whether John's intention was to knock twice in one event or once in two events, it was his intention to knock twice. Furthermore, the time interval that is supposed to distinguish between a single event and double event reading is indepedently indeterminable and has no consequence for truth conditions in the absence of additional scope sensitive material. This approach also requires positing multiple phrases for each of the numerous adverbs that display concentric effects (e.g. almost, again,

Rigidity versus Relativity in Adverbial Syntax

165

always etc.)· Naturally, this leads to an enormous array of projections with essentially redundant content. Finally, we also must explain why the higher twice is marked in its base position even for the iterative meaning ("John twice knocked on the door"). Unmotivated movement of the VP plus PP constituent is necessary to derive the unmarked order for the same reading ("John knocked on the door twice." [2 events]). A free adjunction analysis complemented by a theory of semantic domains or layers avoids the above problems. One difficulty for such an approach, noted by Cinque, is that the symmetry in adverb interpretation is not absolute. Andrews (1983) notes that (28) tends strongly towards the interpretation twice>intentionally. Similarly, the Tagalog (29) tends towards the suddenly>often reading. (28) John twice knocked on the door intentionally. (29) Biglang na-lu~lungkot si Juan nang madalas. suddenly:LNK STA-PROG~sad P.SUB J. nang often 'Juan is suddenly sad often.' (suddenly>often) Ernst (2002: 157) accounts for this tendency as the result of a left-to-right bias in processing but does not offer much discussion. The principle is that when two adverbs sandwich the verb the leftmost tends to take higher scope. Although this suggestion appears plausible at first, we find that precisely the reverse is expected under a processing account. It is important to keep in mind that the type of scope under discussion here is essentially unrelated to that which is used in reference to quantificational items. While scope as used in relation to quantifiers may be closely related to specificity, the scope under discussion here simply refers to the order of semantic composition. Thus, the notation "suddenly>often" as in (29) indicates that often composes with the predicate first and suddenly composes with the constituent including both the predicate and often. Therefore, to say there exists a left-to-right bias in processing a sentence like (29) amounts to positing an opposite (i.e. right-to-left) bias in semantic composition because it is, in fact, the first adverb encountered which tends to be composed last. If processing is responsible for a bias in interpretation, we would expect to find a garden path effect with the first constituent encountered being composed first, as in (30). But this is not the dominant reading.

166

Daniel Kaufinan

(30)

... [ twice knocked on the door ]... %...[[ twice knocked on the door ] intentionally ]

(intent>twice)

Thus, as a strictly linear phenomenon this explanation is unlikely. However, as a hierarchical phenomenon this effect is not at all unexpected if we take certain rightmost adverbs to be VP-internal. If this is correct then we need only state that VP-internal material is composed prior to that which is VP-external; a claim which is already implicit in Ernst's Scope Principle (see section 1.2 above). This incidentally receives strong support from analogues on smaller levels of constituency. As noted by Wechsler (1989), word-internal adverbials are generally composed prior to word-external adverbials as seen with re- in (31).14 (31)

a. b. c. d.

John reswam the English Channel with flippers. (PP>re-) John swam the English Channel with flippers again. (again>PP) He rewrote everything precisely. (precisely>re-) He wrote everything precisely again. (again>precisely)

3.3. Prosody and constituency in concentric phenomena The prosody of English wide-focus sentences also suggests that certain postverbal adverbs are VP-internal while others are not. Assuming that prosodic prominence correlates in some way with embeddedness (Cinque 1993) we find that, in a wide focus context, manner adverbials, for instance, receive phrasal prominence in sentence final position while temporal adverbs, for instance, do not (see also Jackendoff 1972). (32)

a. John sang LOUDLY.

(OK as wide focus)

b. John sang today. c. John sang loudly. d. John sang today.

(OK as wide focus) (not wide focus) (not wide focus)

Furthermore, we see that, in both English and Tagalog, manner adverbs depend on the presence of a verb while temporal adverbs do not. Compare (33a) and (33b).15

Rigidity versus Relativity in Adverbial Syntax

(33)

a. Kape 'Coffee b. *Kape coffee

167

ulit /ngayon /na? again / now / already?' nang mabilis? nang quick (cf. *Coffee quickly?)

It should come as no surprise then that these differences have scopal consequences in addition to the syntactic and prosodic ones noted above; a point which seems to have escaped notice. The following sentences, which are both wide-focus, are unambiguous in terms of adverbial scope. (34) a. Kim intentionally coughed LOUDLY. {/intent>loudly; *loudly>intent) b . Kim intentionally COUGHED j u s t now (*intent>now; ^now>intent) It seems then that the "left-to-right effect" is more a product of the postverbal adverb than the preverbal adverb. Specifically, subject-oriented adverbs and manner-adverbs in post-verbal position tend strongly to compose with the verb before preverbal adverbs of any type, as seen in (35). (35)

a. John almost / always / twice / again knocked intentionally. (S almost /always / twice / again > intent', ?? intent > almost / always / twice / again)

b. John almost / always / twice / again won easily. (/ almost /always / twice / again > easily, ?? easily > almost / always / twice / again)

To make these judgments concrete we can observe the apparent contradiction of (36a) in comparison to (36b). If (36a) were fully ambiguous, we should be able to obtain the reading in (36b).16 (36) I still can't manage to win, but I've gotten to the point where I can... a. %...almost win easily, b. ...easily almost win. (easily>almost) When the postverbal adjunct is not restricted to the VP domain, as with repetitive again, we find the usual ambiguity where the postverbal adjunct has the option of composing after the preverbal one. The contextualized sentences in (37) and (38) show that both readings are possible. The ambiguous (a) sentences require the scope in their unambiguous (b) versions to be felicitous.

168

Daniel Kaufinan

(37)

The first time Razan won was an accident but then... a. ... she intentionally won again, (ambiguous) b. .. .she won again intentionally. (intentionally>again)

(38)

Tina said she would start playing seriously but I think... a. ... she intentionally lost again, (ambiguous) b. .. .she lost intentionally again. (agairi>intentionally)

The role of focus in the relative scope of adverbs is discussed briefly by Phillips (2003) (see also Haider 2000). He contends that apparent rightward scope is an artifact of the sentence final intonation which may induce focus readings on adverbs when they occupy the appropriate position. Ernst (2002: 183-186) shows that this argument is difficult to uphold in the face of conflicting data but does not elaborate on the role of phonology in disambiguation. It is clear, however, that certain prosodic features do correlate with certain scopal readings. I suggest that it is prosodic phrasing and not focus which is the key determinant. First of all, as Ernst notes, it is not clear that focus has a raising effect at all. Work on quantifier and indefinite scope (Diesing 1992: Sec. 2.6) in fact suggests that the opposite is true: focus has a lowering effect on constituents at LF. More importantly, if it is phrasing and not focus which is responsible for disambiguation we have an explanation for why the same effects cannot be duplicated in preverbal positions. If focus induced wide scope on adverbs then we would expect the opposite judgments for (39a) and (39b) where the context requires the scope: almost>intentionally. (39)

a. %John intentionally ALMOST lost, before he realized that wasn't the game he was supposed to throw, b. John almost INTENTIONALLY lost, before he realized that wasn't the game he was supposed to throw.

On the other hand it is intuitively quite clear that the prosodic phrasing indicated by the parentheses in (40) correlates with the order of composition. (40)

a. John (intentionally knocked)(twice). (twice>intentionally) b. John (intentionally)(knocked twice). (intentionally>twice)

Rigidity versus Relativity in Adverbial Syntax

169

Similar effects can be noted in the attachment ambiguity in the mathematical examples (41). When plus two is phrased separately, it is composed after times two composes with the head noun, four. When it is phrased together, the two prepositional phrases are understood as a single constituent which later composes with the head noun. (41)

a. (Four times two)(plus two). b. (Four)(times two plus two).

(=10) (=16)

We may conclude that the above semantic effects of prosody are tightly constrained by constituency. Different prosodies may indicate different bracketings but crucially, prosody cannot subvert the constituent structure, i.e., prosody cannot force Advl to compose before Adv2 in the configuration [Advl Adv2 Verb]. This is expected if the phenomenon in question is prosodic phrasing but unexpected were it focus.

3.4. Inner and outer positions in the Tagalog postverbal domain Certain post-verbal adverbs in Tagalog appear to be VP-internal in being marked with the same marker that introduces VP-internal direct objects (nang), and in being subject to the same extraction constraint on these objects, as shown in (42), cf. (20a). (42)

(* Nang madalas,) nag-basa (nang madalas) nang frequently AV.PRF-read nang frequently 'Jojo read frequently.'

si Jojo. P.SUB J.

As in English, whether or not an adverb is VP-internal determines the interpretive asymmetry discussed earlier. Nang-marked adverbs tend to compose earlier with the verb as seen by the tendency to interpret (43) as shown. (43) Dalawang beses nila akong two:LNK times 3PL.GEN lSG.SUB:LNK singil nang di-sinadya. charge nang accidentally Ί was charged accidentally twice.' (Stwice>accidentally\ laccidentally>twice)

170

Daniel

Kaufman

The other reading is found in (44a) where both adverbs are in clause initial position and in (44b) where both are in clause final position with nang. (44)

a. Di-sinadyang accidentally:LNK

akong lSG.SUB:LNK

dalawang

beses nila

two:LNK

times 3PL.GEN

singil charge

Ί was charged twice accidentally.' (aceidentally>twice) b. Singil nila ako nang dalawang beses charge 3PL.GEN lSG.SUB nang two:LNK times nang di-sinadya. nang accidentally Ί was charged twice accidentally.' (accidentally>twice) The same pattern is exemplified in (45) with the adverbs bigla 'suddenly', lagi 'always' and madalas 'often'. (45)

a. Laging na-lu~lungkot si Juan always:LNK STA-PROG~sad P.SUB J. 'Juan is always suddenly sad.' {/always>suddenly, ?suddenly>always) b. Biglang na-lu~lungkot si suddenly:LNK STA-PROG~sad P.SUB 'Suddenly, Juan is often sad.' (S suddenly> often', lofted suddenly)

nang bigla. nang suddenly

Juan nang madalas. J. nang often

Interestingly, we find that multiple interpretations are more easily available when the final adverb is not attached with nang as is often the case with functional and temporal adverbs. This was seen in (23) above and is due to an attachment ambiguity. However, when two of these adverbs follow the verb the scope is right-to-left as expected. This can be seen in the preferences given the discourse contexts in (46) and (47) (see next page). Another point for Relativity comes from the fact that two readings of often, which Cinque attributes to base generation in different functional projections, are both present on the right edge of the clause in the reverse order which he posits. Because of the nang marker, lower and higher attachment of adverbs is overtly distinguished on the right periphery. In the higher attachment in (48a) (next page), the adverb takes wide scope over negation while in (48b) (next page) it takes narrow scope.

Rigidity versus Relativity in Adverbial Syntax

171

(46) Na-hu~humaling na.naman yata 'yun sa iyo, STA-PROG~obssess again EPST that.SUB OBL 2SG 'He's apparently obssessed with you again, a. ... ta~tawag siya lagi ulit. PROG~call 3SG.SUB always again he's again always calling.' b. %?... ta~tawag siya ulit lagi. PROG~call 3SG.SUB again always he always calls again.' (47) Pagka-baba niya... CMPLT-put.down 3SG.GEN 'After he hangs up, a. %?... ta~tawag siya lagi ulit. PROG~call 3SG.SUB always again he's again always calling.' b. ... ta~tawag siya ulit lagi. PROG~call 3SG.SUB again always he always calls again.' (48)

a. Hindi siya singilQ dalawang beses. NEG 3SG.SUB charge two:LNK times ' He wasn't charged twice.' (iwzce>NEG) b. Hindi siya singil nang dalawang beses. NEG 3SG.SUB charge nang two:LNK times 'He wasn't charged twice.' (NEG>twice)

Also, if topics are considered separately for the purposes of Relativized Minimality, as Rizzi (2002) and Cinque (2004) claim is necessary, there is no principled explanation for why (repetitive) twice could not raise over negation to TopP as in (49) while being interpreted in its base position beneath negation, contrary to fact. (49)

[ Dalawang beses ]i ay hindi siya singil t, two:LNK times TOP NEG 3SG.SUB charge (twice>NEG)

Again, this is predicted by the linear order of the adverbial and negation in regard to the verb. Because negation must be composed with the predicate

172

Daniel Kaufman

before twice, the former must take narrow scope (see section 4 for more on topicalization). Thus, we see that a symmetric approach to adverb scope which requires VP-internal material to be composed with the verb before VP-external material handles all the above facts and may even resolve some of the difficulties noted with interpretive asymmetries in English. If concentric phenomena are mediated through hierarchical relations, scope ambiguity ensues only when attachment is ambiguous, as in the case of two VPexternal adverbs flanking the verb.

3.5. Right-to-Left scope as "dual predication"? Larson (2004) attempts another tack in order to bring concentric scope into line with asymmetric phrase structure. He claims that Andrews' original observation should not be analyzed as a case of scope but rather an instance of predication as in (50) (Larson's [11]). (50)

a. John's intentional knockings on the door were two (in number), b. John's double-knock on the door was intentional.

Adverbs of this type are not viewed as operators but as predicates within a Davidsonian event semantics. Larson adopts Herburger's (2000) account of focus to explain the apparent wide scope of sentence final adverbials. The claim rests on adverbs such as intentionally and twice constituting the nuclear scope of a sentence when appearing in sentence final position as shown in (51) (see next page). When twice, a quantity predicate, appears in the nuclear scope, it is restricted by the material in the restrictive clause thus yielding the apparent wide scope effect. To explain the asymmetric tendency with the sequence twice VERB intentionally Larson proposes that intentionally is ambiguous between a simple event predicate and a scopal operator in preverbal position but that preverbal twice can only be a quantifier. Thus, 'John twice knocked on the door intentionally' tends to be interpreted as in (52) (next page).

Rigidity versus Relativity in Adverbial Syntax

intentionally

3E

Q (52)

173

V

[Ve[Ee->knocking(j,d,e)&intentional(e)]] —^ Restriction

Scope

2e[knocking(j,d,e) & intentional(e)]

Larson's proposal complicates the analysis of adverbs by treating some of them as ambiguous between two very different items. Furthermore, the ambiguity does not appear to be supported by facts other then the scopal tendencies he seeks to explain. The fact that intentionally and twice have predicational equivalents (intentional and two times, respectively) cannot be a diagnostic for adverbs which show concentric effects in the postverbal domain since again, already and before also show these effects (53) but do not possess predicational analogues (*John's knocking was again / already / before). (53)

a. Mickey was drunk already again. (i.e. It's the second time he's been drunk early.) b. Mickey was drunk again already. (i.e. Having just sobered up only 10 minutes ago.)

In support of preverbal twice being quantificational, Larson claims that it is able to take a when-clause as a restriction, unlike clause final twice which cannot. He demonstrates this with (54) (Larson's [55]):

174

Daniel Kaufman

(54) a. Twice when she was in Paris Mary visited the Louvre. Can mean: 'On two separate occasions of being in Paris, Mary visited the Louvre during those occasions.' b. Mary visited the Louvre twice when she was in Paris. Must mean: O n one occasion of being in Paris, Mary visited the Louvre twice during that occasion.' However, he fails to note that the immediate preverbal position of twice in the similar context of (55) strongly favors the second reading above. This is unexpected if twice were truly quantificational in this position. (55)

When she was in Paris, Mary twice visited the Louvre.

Upon further inspection, this data seems to offer additional evidence for the view espoused here. When twice is attached postverbally following a when clause as in (56) there is a tendency to interpret it parallel to (54a). (56) Fido barked when he was hungry twice. This is unexpected from Larson's proposal but is predicted by a theory that allows symmetric adjunction. Because twice is composed with the predicate after the when clause, it tends towards the wide scope reading, referring to two separate events. We can see that in all the above examples the adverb scope is predictable from its surface position in relation to the predicate and other adjuncts. Thus, it seems Larson's proposal cannot handle the interpretation of multiple adverbs as well as a symmetric adjunction approach.17

3.6. Epistemic versus deontic readings of modals Similar to the repetitive / iterative division, Cinque proposes that epistemic and deontic readings of modals are the result of base generation in two different functional projections. He takes the fact that two instances of the same modal can occur simultaneously as evidence for the existence of two separate functional projections. Thus, sentences such as those in (57) could be taken to show that the epistemic phrase is to the left of negation while the deontic (alternatively, "alethic") is to the right.

Rigidity versus Relativity in Adverbial Syntax (57)

175

a. Dapat Hindi sila dapat mag-aral. must NEG 3PL.SUB must AV.INF-study 'It should be that they do not have to study.' b. Maaaring hindi sila maaaring mag-aral. can:LNK NEG 3PL.SUB can:LNK AV.INF-study 'It is possible that they are unable to study.'

As Ernst notes, this necessarily treats the formal similarity between epistemic and deontic modals across languages as accidental homophony. 18 It appears more felicitous to treat modals as yielding an epistemic reading when taking a higher FEO complement (i.e. Proposition) and a deontic reading when taking a lower FEO complement (i.e. Event). Therefore, when modals are stacked on one side of the predicate, the outer modal may only receive an epistemic reading while the inner modal, a deontic one. It must also be noted that many deontic / alethic modals in Tagalog (as in many other languages, cf. de Haan 1997) can appear on either side of negation with transparent scope relations, as seen in (58). (58)

a. Puwede

akong

hindi

can lSG.SUB:LNK ' I ' m able to not sleep.'

b. Hindi ako

puwedeng

ma-tulog.

NEG STA.INF-sleep (deontic-can>NEG)

ma-tulog.

NEG lSG.SUB can:LNK STA.INF-sleep ' I ' m u n a b l e to sleep.' (NEG>deontic-ca«)

Thus, a multiple base generation approach must still rely on movement to capture all the facts while a free adjunction approach accounts for the freedom and scope transparency simultaneously.

3.7. Clitic adverbs Cinque draws a strong connection between affix order and adverb order with the hypothesis that adverbial affixes are the heads of the functional projections containing adverbs of similar semantics as their specifiers. Philippine languages, because of their abundance of clitic adverbs, offer evidence from another domain. In Tagalog, the order of multiple clitics in relation to each other is determined first by prosody; monosyllabic clitics precede disyllabic ones (cf. Schachter and Otanes 1972, Billings 2005).

176

Daniel Kaufman

However, within these phonologically determined domains, there are ordering tendencies that approximate those posited in most general hierarchies. 19 In (59), commas separate items that do not co-occur, the squiggle separates elements that are in free linear variation and the angled bracket separates items that tend (to differing degrees) to enter into a stricter linear relation.20

1σ

(59)

Cna, pa > (na-naman) > man > nga > din • • lang > daw > po > ba D already still

again

even

EMPH

also

L ASP I

ASP II-ITER

only — '

focus / functional

RPRT ^

POL ^

Q ^

EVID POLITE SPEECH-ACT

2σ naman

> yata,

^SWITCH-TOPIC^ apparently

^ focus / func.

V EVID

pala,

sana

>

surprisingly OPTATIVE

^

kaya SPECULATIV

v" SPEAKER-ORIENTED

SPEECH-ACT

The one exception to the strict syllable-based ordering is the compound clitic na naman which appears to be ordered according to its first, monosyllabic, element. Within the otherwise regular clitic domains we find some parallels with the ordering principles of the non-clitic domain. Both Cinque's hierarchy and Ernst's theory predict the position of the reported speech marker daw accurately within the clitic cluster. (60) shows the phrases in the outer periphery of Cinque's hierarchy (Cinque 1999: 106). (60)

[ frankly Moodspeech act [fortunately M o o d e v a i u a t j v e [ allegedly Mood ev identiai

For Ernst, on the other hand, evidentials are necessarily more internal to Speech Act adverbs since evidentials take Propositions and yield Facts (Propositions with annotated truth values) while Speech Act adverbs such as the question marker ba take Facts and yield Speech Acts. Thus, if a

Rigidity versus Relativity in Adverbial Syntax

111

Speech Act adverb occurred internal to an evidential the evidential could not satisfy its selection requirements (given that FEO type lowering is not generally allowed). Aspectual adverbs are the most internal within the monosyllabic domain, also in accordance with general predictions. But whereas, Ernst's theory does not predict a fixed order between iteratives (e.g. again) and aspectual adverbs (e.g. already, stilt), Cinque predicts the opposite order, with the iterative internal to 'already' and 'still'; an order which is incorrect for the Tagalog clitic domain. The two theories appear more or less even in predicting the attested order within single clitic domains. The question now arises as to how these theories can interact with the prosodic component to account for the length-based bifurcation in (59). Within a Cinque-style system, in which adverb order is predetermined by phrase structure, we are confronted with a modular paradox: prosodic conditions must apply before syntactic ordering.21 Similar difficulties have been noted in the literature under the rubric of the "duplication problem". For instance, Chomsky and Halle (1968: 368) observe that there are phonological rules, such as Russian obstruent voicing, which apply both within the word and between words as a sandhi phenomenon. It appears misguided to posit the same rule twice in two parts of the phonology. In the same vein, it seems misguided to posit multiple adverbial hierarchies, one in the phrasal domain, one in the disyllabic clitic domain and one in the monosyllabic clitic domain. On the other hand, this situation is also not unproblematic within the system espoused by Ernst. Although phonological conditions are given a role in adverb ordering it would still be impossible for the FEO calculus to be breached due to overriding prosodic constraints and yet be emergent in cases where prosody was not relevant (i.e., within syllable domains). This pattern is reminiscent of "The Emergence of the Unmarked" effects in Optimality Theory (Prince and Smolensky 1993). When no higher ranking constraints determine the output, lower ranking constraints emerge to eliminate the more marked candidates. We may consider lengthbased requirements to dominate the contiguity required by the FEO calculus. This ranking, shown in Table 2 (see next page), yields the correct facts.

178

Daniel Kaufman

Table 2. The FEO calculus as a violable constraint Input: umalis - AV.PRF:leave, na - already (ASP), ba - question marker (SPEECH ACT), pala - surprise (SPKR-ORIENTED), sila - they a. Umalis {na ba pala} sila?

CLITIC

FEO

1σ>2σ

CONTIGUITY

*

AV.PRF:leave ASP Q SURP 3PL.SUB

b. c. d.

'Did they really leave already?' Umalis {napalaba} sila? Umalis {ba na pala} sila? Umalis {ba pala na} sila?

*! **T *

*!

This accounts for the basic cases but another problem arises. The first element in the sentence hosting adverbial clitics need not be the verb. Adjuncts are found in this position in the adjunct focus construction. If the verb is truly the "center" of the concentric phenomena discussed above, then we would expect that the order in (59) would be reversed when the verb appears to the right of the clitic cluster. However, the order is constant regardless of the position of the verb. Compare (61a) with (61b). (61)

a. Nag-bitaw

na

AV.PRF-quit already

raw

ba ang

REPT Q SUB

komisyoner

kahapon?

commissioner

yesterday

'Did the commissioner reportedly quit yesterday?' b. Kahapon na raw ba nag-bitaw ang komisyoner? Yesterday

already

REPT Q AV.PRF-quit SUB

commissioner

'Was it yesterday that the commissioner reportedly quit?' The verb appears on the right side of the clitic cluster in the presence of a focused adjunct but the internal order of the clitic cluster remains as if the verb is to its left. While I cannot offer a solution to either of these problems here we may note that linear relations in Tagalog become increasingly fixed as one moves from the domain of full words to that of clitics and finally to that of affixes. This is of course not unique to Tagalog, rather, it is probably a feature of natural language in general. As a result we see less semantically transparent alternations between scope sensitive adverbs in the clitic domains. Scope transparency does appear to be a tendency in determining the order of adverbial clitics but it is far from a rule. (62) and (63) from the

Rigidity versus Relativity in Adverbial Syntax

179

internet show scopally transparent orderings of the clitic adverbs lang 'only' and din 'also'. (62) Hindi lang sa NEG

only

OBL

mga

teachers ang

PL

teachers SUB

problema

ο sa

problem

or OBL

estudyante,

hindi lang din

sa

sistema.

student

NEG

OBL

system

only

also

'The problem isn't only with the teachers or with the students, its also not only with the system.' (also>only) (63) Ano

po

what POL

ang

ga~gawa

sa

mga

taong

SUB

PROG~do

OBL

PL

person:LNK

nang-a~away sa kasamahan din lang nila sa opisina? AV-PROG-fight OBL colleague also only 3PL.GEN OBL office 'What is done with the people who fight with their 'just also' colleagues in the office?' (i.e. "with those who are 'only also' like them") (only>also) Further research may determine whether scopal influence on the order of interchangeable clitics is statistically significant. For now, adverbial clitics are seen to pose different, but equally difficult problems for both approaches.

4.

The interpretation of peripheral adverbs and oblique phrases

In this section, we look at the interpretation of Tagalog adverbs and oblique (sa) phrases at the left periphery with the goal of adjudicating between the two theories under discussion. As noted earlier, Ernst allows for adjunct movement only as a subset of more general movement operations such as topicalization, focalization and wA-movement. Others, such as Rizzi (2002) allow for adverb movement without apparent featural / discourse motivations. Taking the sentence in (64) as a base structure we observe the various interpretational possibilities of the oblique phrase sa simbahan in Table 3.22 When the oblique phrase is construed with the matrix clause it is interpreted as "at the church", when it is construed with the subordinate clause it is interpreted as "to the church".23

180

Daniel Kaufman

(64)

utus-an nila akong mag-abuloy ng order-LV 3PL.GEN lSG.SUB:COMP AV.INF-donate GEN pera sa simbahan. money OBL church 'They ordered me to donate money to / at the church.'

Table 3. Interpretive possibilities of oblique phrases MATRIX

a. Inutusan nila akong [mag-abuloy ng pera] sa simbahan] ordered 3 P L lSG.COMP to.donate GEN money OBL church b. Sa simbahan ay inutusan nila akong [mag-abuloy ng pera] c. Sa simbahan nila ako inutusang [mag-abuloy ng pera] d. Inutusan nila akong [sa simbahan ay mag-abuloy ng pera] e. Inutusan nila akong [sa simbahan mag-abuloy ng pera] f. Inutusan nila ako sa simbahang [mag-abuloy ng pera] g. Inutusan nila akong [mag-abuloy sa simbahan ng pera]

SUBORD

always)

The fact that non-topic manner adverbs may not precede a w/z-phrase as in (70b) as opposed to topicalized arguments as in (70c) is taken by Rizzi to indicate the existence of two peripheral positions, one for adverbs and one for DPs. (70)

a. Rapidamente, hanno fatto i compiti. 'Quickly, they did the homework.' b. ?? Rapidamente, che cosa hanno fatto? 'Quickly, what did they do?' c. A Gianni, che cosa gli hanno fatto? 'To Gianni, what did they do to him?'

This too, may be reanalyzed as a violation of the FEO-calculus. The interrogative operator raises the FEO-type to Speech Act after which it is impossible for a manner adverb to modify a lower FEO type. On the other hand, a true topic like the PP in (70c) has a motivation to move from a lower position. The scope would then be calculated from the position of the trace, thereby satisfying the FEO-calculus. With the approach outlined here, the difference between Italian and Tagalog may be reduced to a single parameter: the ability of treating ad-

184

Daniel Kaufman

verbs as discourse topics.26 This is clearly necessary on independent grounds since, unlike Italian, true topicalization of Tagalog adverbs cannot be licensed by discourse, as seen in (71). (71)

A: Paano ko siya puwedeng ma-kita nang madalas? how lSG.GEN3SG.SUB can:LNK PV.INF-see nang frequently 'How can I see her frequently?' B: # Madalas ay walang naka~ka-kita sa kanya. frequently TOP none:LNK AV~PRG-see OBL 3SG 'Frequently, nobody sees her.'

Another argument against treating fronted adverbs in Tagalog as movement comes from the scope facts as seen earlier (20)/(49). Left-peripheral adverbs in Tagalog show no signs of reconstruction, contrary to what might be expected under a movement analysis. Furthermore, under a movement analysis there is no principled way to rule out the topicalization of manner and measure adverbs as in (72). (72)

*MalakaSi loudly

ay

nag-salita

sila

TOP

AV.PRF-speak

3PL.SUB

tt

An approach utilizing the FEO calculus or some version thereof may take advantage of the fact that these are precisely the categories which require access to an internal semantic layer associated with the VP. It would make sense then that access to this layer is not available above sentential operators which require Events. Since the FEO type of a basic clause must at least be a Proposition, Events are unavailable in the peripheral focus and topic positions. Purely structural explanations must claim the existence of a syntactic distinction between these two classes of adverbs as opposed to all others. A more semantically grounded account can make reference to the lexical semantics of measure and manner modification which involves 'core' verbal meaning (Tenny 2000).27

5.

Postscript: Malagasy

If concentric layering of adjuncts around the verb is the rule rather than the exception, then the ordering of adverbs in the postverbal domain of languages like Malagasy (Rackowski, 1998) and Seediq (Holmer, 2004) is

Rigidity versus Relativity in Adverbial Syntax

185

less surprising than previously considered. Rackowski (1998) gives the order in (73) for Malagasy adverbs. She notes that the order in the preverbal domain is predicted by Cinque's hierarchy but that this order is seemingly reversed in the postverbal domain. (73) Na(dia) > Matetika > Tsy > Efa / Mbola > Tsy > VERB > Tsara > 'Even' generally NEG already / still NEG well Tanteraka > Foana > Intsony > Mihitsy > Aza > Ve completely

always

anymore

at.all

though SPEECH ACT (Q)

The order of adverb classes is explained naturally by a theory which treats distance from the verb as indicative of scope. Such a theory also correctly predicts that postverbal adverbs in Malagasy may scope over preverbal ones, as shown in (74). (74) Efa avy mi-sakafo foana Rasoa vao tonga aho. already come AV-eat always Rasoa before arrive lSG. SUB 'Rasoa is always already finished eating before I arrive.' Under an intraposition analysis of (74) where adverbs have syntactically specified base positions, the second adverb foana 'always' is never in a position to c-command efa 'already'. If LF movement is necessary to determine the correct scope then it is unclear how such movement may be constrained to prevent overgeneration of unavailable interpretations. Under a more symmetric analysis, if scope is determined strictly by c-command relations and the subject is in a right branching specifier (as in MacLaughlin 1995, Guilfoyle, Hung and Travis 1992) then foana may be adjoined above efa and below the subject.28 Finally, there is evidence that adverbs within a single class are permutable in Malagasy with corresponding differences in scope (cf. Rackowski 1998: 10). Adverb order should therefore not be taken as a strong argument for intraposition (Pearson 2000, Rackowski and Travis 2000) without evidence that it is possible for a verb-initial language to display the Cinque order of adverbs in the postverbal domain.

186

Daniel Kaufman

6. Conclusion While templatic analyses have been eschewed with good basis in Phonology (McCarthy and Prince 1995), and Morphology (Rice 2000, Stiebels 2003 inter alia) we find that the template has actually gained ground in syntactic theory under the banner of the "Cartographic Approach" (cf. Cinque (ed.) 2002, Rizzi (ed.) 2002, Belletti (ed.) 2002), with Cinque's adverbial hierarchy serving as a prime exemplar. If we take seriously the Minimalist call to reduce the grammar to interface conditions on the articulatory component and the interpretive component (Chomsky 1995), then purely phrase structural solutions should also be eschewed when more integrated solutions are available. Ernst's theory articulates one aspect of the syntax-semantics interface and has been shown here to derive the Tagalog facts well from basic semantic principles. The templatic hierarchy posited by Cinque, on the other hand, makes no reference to semantics at all at the expense of proliferating syntactic categories and functional projections. This is seen to lack in explanatory adequacy in that we have no vista for approaching the obvious question, "Why this order and not another one?". The implicit claim of the Cartographic Approach, as of any templatic analysis, is that the stipulated order is inherently unpredictable. Armed, however, with a basic theory of event ontology (e.g. Parsons 1990) and scope principles we find that adverb order is highly predictable from the perspective of the interpretive component. Syntax should therefore be no exception to the program of eliminating templates whenever possible.

Notes *

1.

Thanks to John Whitman, John Wolff, Draga Zee and the audience at AFLA XI for helpful comments. Thanks also to Ronie Calugay, Rodney Jubilado, Tina Manueli, Raphael Mercado and especially Thess Savella for their judgments. Cinque allows for unmotivated movement through an extensive array of functional projections and multiple loci for base generation of certain adverbs. Ernst on the other hand introduces stipulatory syntactic features such as [+Right] which is satisfied by linearization to the right of a certain head and PF requirements on "heavy" / "light" items which have no basis in the phonological content of the items themselves.

Rigidity versus Relativity in Adverbial Syntax 2.

3.

4.

5.

6.

7.

8.

9.

187

I choose these terms over the more common Tight-fit versus Loose-fit since these terms underemphasize a crucial distinction between the two approaches: the predictions of relative scope between adverbs. This phenomenon cannot be given attention here since verb movement in the languages under discussion here is not "gradated" as in Romance languages, i.e., we do not find overt evidence that the verb is in different positions based on finiteness or other inflectional features. A reviewer points out that there do exist cases such as the German example in (i) where it appears that a verb must be c-commanded by a comparative adjunct (Haider 1997:24, ex.15). (i) Der Wert hat sich weit mehr als bloß verdreifacht, 'the value has itself far more than merely tripled' V2 is blocked in such sentences. But because of the parochial nature of this restriction with comparatives, it would probably be unfounded to unify an account for this with an account of more general verb-adverb ordering in Romance. Relativized Minimality as envisioned in Rizzi 2002a takes the more specific categories quantificational, modificational, topic and argument to be relevant in calculating minimality. Of course, these "types" of movements have all been taken to be instantiations of a single rule since the advent of GB. The empirical prediction underlying the rejection of "adverb movement" as an independent phenomenon is that adverbs which are not interpreted in their surface position are displaced because of positional requirements on topicalized, focused or wA-constituents. Schäfer (2002) takes manner readings to be unmarked and clausal readings to be derived through a process of metaphoric extension. I will have nothing further to say about this issue as nothing hinges on it here. The marker introducing adjuncts is written while that introducing arguments is but these are identically pronounced as /na / and nothing differentiates them besides orthography. The connection is further substantiated by many other Philippine languages which introduce manner adverbs and certain other adjuncts with the same marker used for internal arguments. Abbreviations used: AV = actor voice, COMP = complementizer, CMPLT = completitive, EMPH = emphatic, EPST = epistemic, GEN = genitive, INF = infinitive, IRR = irrealis, LNK = linker, LV = locative voice, NEG = negation, NMLZ = nominalizer, OBL = oblique case, OPT = optative, Ρ = personal/[+human], PL = plural, POL = politeness marker, PROG = progressive, PV = patient voice, Q = question marker, RHET = rhetorical question marker, SG = singular, STA = Stative, SUB = 'subject' case, SURP = surprise, TOP = topic marker, TR = transitive. Some irrelevant morphological categories are conflated in the interest of perspicuity.

188

Daniel Kaufman

10. Second position clitics immediately follow the adverb when it is adjoined directly to the verb as in (1 lb) but must follow the verb when the complement is clausal as in (13). 11. The precise nature of this "linked" position (i.e. with the linker nal- ) requires further investigation. I suppose here that it arises from X° adjunction to the verb, (i) (cf. [16b]) suggests that X° linking is symmetric in the verbal domain (as it certainly is for adjectives in the nominal domain). (i) Kantang malinaw si Jojo. sing:LNK clear P.SUB J. 'Jojo sang clearly.' There is however a rather strong tendency for linked modifiers of both an adjectival and adverbial nature to precede heads. Due to space limitations, I limit the ensuing discussion to the more common types of adverb attachment and thus must postpone investigation of this type to further work. 12. It should be noted that the edge of event-internal modification appears to be higher than VP, where Ernst claims it is. I will not address this here as it does not effect the more far reaching debate between the two theories. 13. A reviewer mentions the possibility that the adverb in (18b) could also quantify over a contextually restricted domain as in, "Often, when it rains...". But this reading is not inherently present and can only result from scoping over a restricting (when) clause or second operator. 14. This is also reflected in the ungrammaticality of (i) which Wechsler (1989) asserts is the result of the PP being unable to enter the scope of the affixal re-. (i) *He reclimbed on the hill (ii) He reclimbed the hill. 15. A reviewer points out the possibility of sentences like the German, Kaffee, schnell! 'Coffee, quickly!'. Although such constructions are often orthographically single sentences I feel that they are better treated as two, the second sentence ('Quickly!') instantiating a common ellipsis. This is supported by the presence of an obligatory prosodic break before the adverb. 16. Under certain intonational patterns (36a) is acceptable here. Crucially, this pattern requires a prosodic break between the verb and the manner adverb. I take this to indicate that the verb and manner adverb are no longer co-inhabiting the VP. Either the manner adverb has been right-dislocated or the functional adverb and the verb have been topicalized. In the first case, the manner adverb follows a prosodic break and is deaccented, in the second case it also follows a prosodic break but receives sentential stress. 17. Two more difficulties should be noted for treating concentric scope as a result of semantic partition. First, it cannot at the same time account for the original phenomena which semantic partition was meant to explain, namely, the interpretation of indefinites. If, as Larson claims, because clauses differ from after clauses in that the former is included in the nuclear scope while the latter is

Rigidity versus Relativity in Adverbial Syntax

18.

19. 20.

21.

22.

23.

189

part of the restrictive clause, we would expect to find additional evidence in the interpretation of quantifiers and indefinites within these phrases. Second, Larson's proposal rests on a correspondence between quantificational and information structure such that the nuclear scope is equated with the informational focus. As shown earlier, in regard to adverb scope, intonation cannot subvert constituent structure and the final postverbal adverb does not require focus. More generally, it is difficult to maintain such a tight connection between information structure and quantification structure. Focus should probably be seen as having certain effects on quantification structure (e.g. triggering lowering into the nuclear scope) but the nuclear scope cannot be so generally considered as equivalent to the focus of a sentence. More recently, Cinque (2004) addresses the homophony problem by suggesting that certain adverbs are 'underspecified' for certain features (e.g. epistemic / deontic) and can thus be compatible with two different positions. However, the fact that these 'underspecified' adverbs are still base-generated separately begs the question of what precisely their connection is. Thanks to Loren Billings for first pointing this out to me. These results are based both upon speaker judgments and statistical data obtained manually from a comprehensive series of Google searches. Most of (59) is in agreement with Schachter and Otanes (1972) but some is not. They claim, for instance, (p.414), that when daw 'reported speech' and ba 'question marker' co-occur they appear as daw ba and ba daw with equal frequency. The internet data showed however that daw ba occurred 2,144 times as opposed to 855 times for ba daw\ a significant difference and one which is expected if the FEO calculus plays a (emergent) role in clitic ordering. Other items in the chart cannot be ordered relative to each other because they only rarely co-occur. Such is the case with yata, pa la and kaya. Another alternative is that the syllable based ordering takes place at PF as expected but that syntactic locality constrains prosodic movement. Under this view, violation of the original "prosody-blind" hierarchical ordering at PF could be treated on par with superiority violations. One of the difficulties inherent in such an approach is deciding on what syntactic conditions apply and don't apply at PF. Clearly, the vast majority of syntactic conditions (e.g. HMC, MLC and versions thereof) cannot be respected at this level. It should be noted that the example employed is a control structure which shows certain restructuring effects (Mercado 2002). Non-restructuring contexts may show slightly different behavior in terms of oblique phrase and adverb movement but a full discussion of these differences will have to await further work. The particle ay follows topicalized elements while the complementizer introducing the subordinate clause surfaces as a velar nasal clitic /- / when preceded by a vowel.

190

Daniel Kaufman

24. Alternatively, to get the matrix construal for (a), a Kaynian analysis could generate the oblique phrase in the matrix clause, raise it to a higher position and then move the remnant higher yet. The problem is that the first movement of the oblique phrase to the left periphery would have to be to an otherwise unattested position which is neither Topic nor Focus (since it is neither morphologically nor pragmatically marked). It would then, idiosyncratically, have to trigger obligatory remnant movement, "coincidentally" covering-up the only evidence for the original, aberrant movement. I feel this appears too suspicious to merit serious consideration. 25. It is not entirely clear that the adverbs in (c) and (e) of Table 4 are in the same position as the oblique phrases in (c) and (e) of Table 3. For one, the linker / complementizer may appear after the adverb but not after the oblique phrase. Also, when an oblique is focused an adverb may follow it directly in the peripheral position shown in (c) and (e) and yet be part of the presupposition. If this is simply adjunction to a (non-pragmatically marked) peripheral projection then we may conclude that adjunction to focus positions is forbidden. Either way, the facts are still consistent with the analysis offered here. 26. A reviewer makes the intriguing suggestion that this in turn may reduce to the presence versus absence of a "bridge accent" in the language. German, which is known to possess such an accent, appears to allow long distance movement of manner adverbs as contrastive topics (i). (i) /HAUfig meint der Hans, dass nur die MaRIA\ Bücher liest. often thinks DET H. COMP only DET M. books reads 'Hans thinks that only Maria reads books frequently.' 27. Ernst actually derives adverbs in topic position through movement in combination with the following Scope Matching Constraint: Scope Matching Constraint on Adjunct Topicalization (Ernst 2002: 420) a. The scope of a topicalized adjunct must match that of its base position. b. Scope matching holds if the lexical material is identical except for "presupposed tense" and the base-position copy of the adjunct. He opts for this approach over a pure adjunction approach based on three facts (Ernst 2002: 424). First, the ungrammaticality of adverbs intervening between a subject and auxiliary verb, as in (i). (i) *He icily had spoken to the lieutenant. Second, the fact that manner adverbs may putatively appear in topic position while measure adverbs cannot. And finally, the fact that speaker-oriented adverbs do not need to be set off by comma intonation while subject oriented and manner adverbs must be. These motivations do not appear to me compelling enough to sacrifice a uniform adjunction analysis of adverbial topics along the lines of Shaer (2003)

Rigidity versus Relativity in Adverbial Syntax

191

and others. My intuition is that the difference between (i) and (ii) is not strong enough to warrant a drastic amendment to the theory, (ii) ?Icily, he had spoken to the lieutenant. Among the languages I am familiar with, manner readings of pragmatically unmarked adverbs in topic position are generally impermissible. Finally, it is very unclear that comma intonation can be taken as a reliable diagnostic for movement as opposed to adjunction. It may be a universal that focus movement and w/j-movement, for instance, are never accompanied by comma intonation. 28. Craig Thiersch (p.c.) has suggested to me an alternative bi-clausal analysis for (64) (Thiersch 2005). Preliminary investigation however shows that the verblike avy is not responsible for the scope facts. Compare always>still in (i) (thanks to Lucien Herimanjaka Rajoarison). (i) Mbola mi-asa foana ζ Rasoa na.dia efa still AV.PROG-work always P.SUB R. even already n-ody aza ny rehetra. AV.PRF-go.home though DET all 'Rasoa is always still working even when everyone has already gone home.'

References Alexiadou, Artemis 1997 Adverb Placement. Amsterdam: John Benjamins. Andrews, Avery 1983 A note on the constituent structure of modifiers. Linguistic Inquiry 14: 695-697. Belletti, Adriana (ed.) 2002 Structures and Beyond. The Cartography of Syntactic Structures. Oxford: OUP. Billings, Loren 2005 Ordering clitics and postverbal R-expressions in Tagalog: a unified analysis? In Verb First: On the Syntax of Verb Initial Languages, Andrew Carnie, Heidi Harley, and Sheila Ann Dooley (eds.), 303339. Amsterdam: John Benjamins. Chomsky, Noam 1995 The Minimalist Program. Cambridge, MA: MIT Press. Chomsky, Noam, and Morris Halle 1968 The Sound Pattern of English. New York: Harper & Row. Cinque, Guglielmo 1993 A null-theory of phrase and compound stress. Linguistic Inquiry 24: 239-298.

192

Daniel Kaufman

1999

Adverbs and Functional Heads: A Cross-linguistic Perspective. Oxford: OUP. 2004 Issues in adverbial syntax. Lingua 114: 683-710. Cinque, Guglielmo (ed.) 2002 The structure of DP and IP. The Cartography of Syntactic Structures. Oxford: OUP. Diesing, Molly 1992 Indefinites. Cambridge, MA: MIT Press. Ernst, Thomas 2002 The Syntax of Adjuncts. Cambridge: CUP. 2004 Principles of adverbial distribution in the lower clause. Lingua 114: 755-777. Guilfoyle, Eithne, Henrietta Hung, and Lisa Travis 1992 Spec of IP and Spec of VP: Two subjects in Austronesian languages. Natural Language and Linguistic Theory 10: 3 7 5 ^ 1 4 . de Haan, Ferdinand 1997 The Interaction of Modality and Negation. New York: Garland. Haider, Hubert 1997 Typological implications of a directionality constraint on projections. In Studies on Universal Grammar and Typological Variation, Artemis Alexiadou, and T. Alan Hall (eds.), 17-33. Amsterdam: John Benjamins. 2000 Adverb placement - convergence of structure and licensing. Theoretical Linguistics 26: 95-134. Hale, Kenneth, and Samuel-Jay Keyser 1993 On argument structure and the lexical expression of syntactic relations. In The View from Building 20: Essays in Linguistics in Honor of Sylvain Bromberger, Ken Hale, and Samuel-Jay Keyser (eds.), 53-108. Cambridge, MA: MIT Press. Herburger, Elena 2000 What Counts. Cambridge, MA: MIT Press. Holmer, Arthur 2004 Intraposition and Formosan adverbial heads. In Proceedings of AFLA XI, ZASPiL 34, Paul Law (ed.), 120-134. Berlin: ZAS. Jackendoff, Ray 1972 Semantic Interpretation in Generative Grammar. Cambridge, MA: MIT Press. Kaufman, Daniel to app. Aspects of pragmatic focus in Tagalog. In The Many Faces of Austronesian Voice Systems, I Wayan Arka, and Malcolm Ross (eds.). Canberra: Pacific Linguistics.

Rigidity versus Relativity in Adverbial Syntax

193

Kimball, John 1973 Seven principles of surface structure parsing in natural language. Cognition2: 15—47. Larson, Richard 2004 Sentence-final adverbs and "scope". In Proceedings of NELS 34. Keir Moulton and Matthew Wolf (eds.), 23-43. Amherst, MA: GLSA. McCarthy, John, and Alan Prince 1995 Faithfulness and reduplicative identity. In University of Massachusetts Occasional Papers in Linguistics 18, Jill Beckman, Laura Walsh Dickey, and Suzanne Urbanczyk (eds.), 249-384. Amherst, MA: GLSA. McConnell-Ginet, Sally 1982 Adverbs and Logical Form: a linguistically realistic theory. Language 58: 144-184. MacLaughlin, Dawn 1995 Wh-movement in Malagasy: An extraction asymmetry. In Theoretical Approaches to African Linguistics I, Akinbiyi Akinlabi (ed.). Trenton, NJ: Africa World Press, Inc. Mercado, Raphael 2002 Verb raising and restructuring in Tagalog. MA thesis, Department of Linguistics, University of Toronto. Parsons, Terence 1990 Events in the Semantics of English. Cambridge, MA: MIT Press. Pearson, Matthew 2000 Two Types of VO Languages. In The Derivation of VO and OV, Peter Svenonius (ed.), 327-363. Amsterdam: John Benjamins. Phillips, Colin 2003 Linear order and constituency. Linguistic Inquiry 34: 37-90. Prince, Alan, and Paul Smolensky 1993 Optimality Theory: Constraint interaction in generative grammar. Ms., Rutgers University and University of Colorado, Boulder. ROA#537-0802. Rackowski, Andrea 1998 Malagasy adverbs. In The Structure of Malagasy, volume II, Ileana Paul (ed.), 11-33. (UCLA Working Papers in Linguistics.) Los Angeles. Rackowski, Andrea, and Lisa Travis 2000 V-initial languages: X or XP Movement and Adverbial Placement. In The Syntax of Verb Initial Languages. Andrew Carnie, and Eithne Guilfoyle (eds.), 117-141. Oxford: Oxford University Press.

194

Daniel Kaufman

Reinhart, Tanya 1995 Interface strategies. Utrecht: OTS working papers in linguistics. Rice, Keren 2000 Morpheme Order and Semantic Scope: Word Formation in the Athapaskan Verb. Cambridge: CUP. Rizzi, Luigi 1990 Relativized minimality. Cambridge, MA: MIT Press. 2002 Locality and Left Periphery. Ms., University of Siena. Rizzi, Luigi (ed.) 2002 The Structure of CP and IP. The Cartography of Syntactic Structures vol.2. Oxford: OUP. Schachter, Paul and Fe Otanes 1972 Tagalog Reference Grammar. Berkeley: Univ. of California Press. Schäfer, Martin 2002 Pure manner adverbs revisited. In Sinn & Bedeutung VI, Proceedings of the Sixth Annual Meeting of the Gesellschaft fur Semantik, Graham Katz, Sabine Reinhard, and Philip Reuter (eds.), 311-323. University of Osnabrück. Shaer, Benjamin 2003 An "orphan" analysis of long and short adjunct movement in English. In Proceedings of WCCFL 22, Gina Garding, and Mimi Tsujimura (eds.), 450-463. Somerville, MA: Cascadilla Press. Stiebels, Barbara 2003 Transparent, restricted and opaque affix orders. In Syntactic Structures and Morphological Information, Uwe Junghanns, and Luka Szucsich (eds.), 283-315. Berlin: Mouton de Gruyter. Tenny, Carol 2000 Core events and adverbial modification. In Events as Grammatical Objects, Carol Tenny, and James Pustejovsky (eds.), 285-334. Stanford: CSLI Publications. Thiersch, Craig 2005 A note on the scope of adverbials in Malagasy. Talk given at the TIN-dag conference, Jan. 29, 2005. Wechsler, Stephen 1989 Accomplishments and the prefix re-. In NELS 19, Juli Carter and Rose-Marie Dechaine (eds.), 4 1 9 ^ 3 4 . Amherst: GLSA.

The Cleft Structure of Malagasy fF/i-Questions* Eric Potsdam

This chapter considers the structural analysis of w/z-questions in the predicate-initial language Malagasy. In addition to a w/z-in-situ strategy, Malagasy allows the w/z-phrase to appear at the front of the clause, as in iza no nihomehy? ('Who laughed?'). I propose that such examples do not involve w/z-movement but are pseudoclefts in which the w/z-phrase is the predicate and the remaining material is a headless relative clause in subject position. Such examples are thus more literally translated as "The one who laughed is who?" Argumentation in favor of the pseudocleft analysis and against a w/z-movement analysis comes from i) predicate properties of the initial w/z-phrase, ii) syntactic and semantic parallels with a focus construction that Paul (2001) independently analyzed as a pseudocleft, and iii) the analysis of certain multiple w/z-questions from Sabel (2003). The chapter concludes that Malagasy does not have any w/z-movement and is a pure whin-situ language.

1. Introduction Malagasy is an Austronesian VOS language spoken on the island of Madagascar, ff/z-questions in Malagasy consist of a clause-initial w/zphrase followed by the invariant particle no and then the remainder of the clause: (1)

iza no nihomehy? who laugh 'Who laughed?'

This chapter considers two structural analyses of this construction. Under the FRONTING ANALYSIS (Sabel 2002, 2003), such questions have an English-like derivation in which the w/z-phrase moves to a clause-initial spec,C:

196

Eric Potsdam

(2)

[CP

izat

[Ct no

[ IP [ V p

who

nihomehy ] tt ]]] laugh

Under the CLEFT ANALYSIS (Dahl 1986, Paul 2000, 2001a), w/z-questions are covert pseudoclefts in which the initial w/z-phrase is an in-situ predicate (pred) and the remaining material is a headless relative (hr) clause in subject position: (3)

tip [pred

iza

]

[DP/hr

no

°Pi

nihomehy

tt ]]

who laugh (lit. 'The one that laughed is who?') Both analyses are compatible with observed word order and have proponents in the literature. The goal of this chapter is to provide evidence for the cleft analysis and it is organized as follows: Section 2 presents some basic facts about Malagasy clause structure and the formation of w/z-questions. Section 3 lays out in more detail the two competing structural analyses of w/zquestions. Section 4 introduces theoretical considerations and empirical evidence in favor of the pseudocleft analysis and against the fronting analysis. The argumentation centers around showing that the initial w/zphrase in such questions behaves like a predicate and not an interrogative argument or adjunct. Section 5 explores some details of the cleft analysis and the final section concludes.

2.

Malagasy syntax and w/z-questions

Malagasy is an Austronesian language spoken on the island of Madagascar. It is well-known for having basic VOS word order in active clauses, as in (4)·1,2 (4)

nividy ny akoho i Bao. PAST.ACT.buy the chicken Bao 'Bao bought the chicken.'

For concreteness I follow Guilfoyle, Hung, and Travis 1992 and adopt the structure for VOS in (5), with the clause-final subject occupying a right specifier of IP. The material preceding the subject is the predicate, con-

The Cleft Structure of Malagasy Wh-Questions

197

stituting a Predicate Phrase, PredP, which is a complement to 1° (Bowers 1993, Chomsky 1995, Kratzer 1996).3 IP

(5)

DP 1°

PredP

i Bao

nividy ny akoho Malagasy also has a well-developed voice system which advances thematically diverse elements to the clause-final subject position. Corresponding to the active sentence in (4), the PASSIVE sentence in (6a) has the direct object as the clause-final subject and the CIRCUMSTANTIAL sentence in (6b) has an oblique element as its subject, in this case a benefactive. The agent in non-active clauses appears immediately following the verb. (6)

a. novidin' i Bao ny akoho. PAST.buy.PASS Bao the chicken 'The chicken was bought by Bao.' b. nividianan' i Bao ny akoho i Soa. PAST.buy.CIRC Bao the chicken Soa 'Soa was bought a chicken by Bao.'

For such non-active clauses, I again follow Guilfoyle, Hung, and Travis 1992 and assign a structure as in (7) in which the surface subject has raised to the right specifier of IP. IP

(7)

DP, 1°

PredP

ny akoho

novidin 'i Bao t, ^ - q u e s t i o n s in Malagasy are formed in one of two ways. For nonsubjects, w/z-in-situ is possible (see Sabel 2002 for description and

198

Eric Po tsdam

analysis). The second method, relevant for this chapter, is to prepose the w/z-phrase and follow it immediately with the particle no (glossed as PRT) and then the remainder of the clause, minus the wA-phrase: (8)

a. iza no nividy ny akoho? who PRT buy.ACT the chicken 'Who bought the chicken?' b. inona no novidin' i Bao? what PRT buy.PASS Bao 'What was bought by Bao?'

It is widely cited that only subjects can be questioned with this strategy (Keenan 1976, 1995, Keenan and Comrie 1977, MacLaughlin 1995, Paul 2000, 2002, Pearson 2001, Sabel 2002, and others). Consequently, it is ungrammatical to w/z-question a non-subject argument, (9). Either wh-msitu can be used or the sentence can first be 'turned around' using voice morphology so that the non-subject corresponds to a subject. (9)

a. * inona no nividy i Bao? what PRT buy.ACT Bao ('What did Bao buy?') b. * iza no novidina ny akoho? who PRT buy.PASS the chicken ('Who was the chicken bought by?')

An exception to this generalization is that some adjuncts, including temporal, locative, and instrumental adverbials, can be questioned without first advancing to subject position (Keenan 1976, Rabenilaina 1998, Paul 2000, 2001a, 2002, Pearson 2001, Sabel 2002, and others). (10) shows that a question with taiza 'where' is compatible with any verbal voice form. Only in the example with circumstantial voice, (10c), does the adverbial plausibly correspond to a subject. (10)

a. taiza no w h e r e PRT

b. taiza no w h e r e PRT

nanafina

ny

lakileko ny

zaza?

hide. ACT

the

key.lSG the

child

nafenin'

ny

zaza

hide.PASS

the

child t h e k e y . l S G

ny lakileko?

The Cleft Structure of Malagasy Wh-Questions c. taiza no nanafenan' ny where PRT hide.CIRC the 'Where did the child hide my key?'

zaza child

ny the

199

lakileko? key.lSG

This restriction, stated in (11) below, will be an important descriptive generalization in the discussions that follow (see MacLaughlin 1995, Sabel 2002, Paul 2002, and Pearson 2005 for possible analyses of the restriction). (11)

3.

Malagasy extraction restriction Only subjects and some adjuncts can be extracted

Two hypotheses

This section introduces two structural analyses of Malagasy w/z-questions. Section 3.1 presents Sabel's (2002, 2003) fronting analysis of w/z-questions and section 3.2 develops the pseudocleft analysis based on Paul (2001a).

3.1. The fronting analysis The fronting analysis of w/z-questions likens their structure to that of whfronting constructions in better studied language families like Germanic, Romance, and Slavic. Sabel (2002, 2003) is the primary articulator of such an approach, although it is assumed in MacLaughlin (1995), Pensalfmi (1995), and Potsdam (2003). fF/z-questions, such as (12a), resemble English w/z-fronting examples and can be assigned a familiar structure, (12b). (12)

a. iza no nihomehy? who PRT laugh.ACT 'Who laughed?'

b.

nihomehy

200

Eric Potsdam

The w/z-phrase moves to spec,C and no is a question complementizer in C°. The motivation for the movement can be as in other languages: a strong [wh] (or EPP) feature on C° no.

3.2. The cleft analysis The cleft analysis, which I will ultimately defend, builds on proposals in Dahl (1986), Pearson (1996), Paul (2001a, 2003b), and others. Under the cleft analysis, w/z-questions in Malagasy are pseudocleft structures in which the w/z-phrase is a non-verbal predicate and the subject is a headless relative clause involving internal movement of a null operator, Op. The closest English translation is with a pseudocleft: Who laughed is who? or The one who laughed is who? The w/z-question repeated in (13a) is assigned the structure in (13b). (13) a. iza who

no

nihomehy?

PRT

laugh.ACT

'Who laughed?'

For now, I take the particle no to be a determiner, following suggestions in Paul (2001a, 2002, 2003a). I return to this issue in section 5.1. I assume that the relative clause is directly selected by the determiner no, as shown.

The Cleft Structure of Malagasy Wh-Questions

201

Given that Malagasy has w/z-in-situ questions, the pseudocleft analysis is simply an extension of this strategy, in which the w/z-phrase is in-situ as a predicate. Malagasy thus has no w/z-movement at all under the cleft analysis. The w/z-phrase is always in-situ, either as an argument, an adjunct, or a predicate. Malagasy does still have A'-movement, of null operators. Several related Austronesian languages are also claimed to employ a pseudocleft structure for questions - for example, Palauan (Georgopoulos 1991), Malay (Cole, Hermon, and Aman to appear), Tsou (Chang 2000), Tagalog (Richards 1998, Aldridge 2002), and Seediq (Aldridge 2002) and in the next section I provide a wide range of arguments showing that the cleft analysis is appropriate for Malagasy also.

4. Argumentation against the fronting analysis This section provides empirical and theory-internal argumentation in favor of the pseudocleft analysis for Malagasy vv/z-questions and against the fronting analysis. Section 4.1 illustrates various ways in which the w/zphrase behaves like a predicate and not a fronted argument / adjunct. Section 4.2 points out parallels with a focus construction that Paul (2001a) analyzes as a pseudocleft. Lastly, section 4.3 argues that the cleft analysis provides a superior treatment of certain multiple fronted-w/z questions.

4.1. Predicational properties of the w/z-phrase One distinction between the pseudocleft and fronting analyses concerns the characterization of the initial w/z-phrase. Under the pseudocleft analysis, it is a predicate. Under the fronting analysis, the w/z-phrase is not a predicate but a preposed constituent - argument or adjunct. We can thus differentiate the two analyses by seeing whether a fronted w/z-phrase has properties characteristic of a predicate or not. To facilitate the discussion, I will use VOS clauses as examples of predicate-initial clauses. There is abundant evidence that the verb plus complement(s) in such clauses constitute a predicate to the exclusion of the subject (Keenan 1976, 1995). Of particular interest is that Malagasy has a variety of particles whose distribution can be described as preceding or following the predicate. What

202

Eric Potsdam

we will see is that such particles treat initial w/z-phrases as predicates, not as fronted constituents.

4.1.1. Post-predicate

particles

Malagasy has a number of particles that immediately follow the predicate, including floating quantifiers like daholo 'all' and avy 'each' (Keenan 1976, 1995), the exclamative particle anie (Keenan 1976, 1995),4 and VP adverbs such as foana 'always' (Pearson 1998, Rackowski 1998). In VOS clauses, these elements only appear immediately after the predicate and before the subject and not elsewhere:5 (14)

a. nihinana vary (daholo) ny vahiny (* daholo) eat.ACT rice all the guest all 'All the guests ate rice' b. manapaka bozaka (anie) Rasoa (* anie) cut. ACT grass EXCL Rasoa EXCL 'Rasoa is really cutting the grass!' c. mihomehy (foana) Rasoa (*foana) laugh.ACT always Rasoa always 'Rasoa is always laughing'

For concreteness, I assume that these particles right adjoin to PredP as shown in (15). This correctly places them between the object and the subject in VOS clauses. (15)

IP

PredP

particle

V Ο The two analyses under consideration make different predictions about where these particles will appear in w/z-questions. As shown in (16a), the

The Cleft Structure of Malagasy Wh-Questions

203

cleft analysis places them immediately after the w/z-phrase. The fronting analysis, in contrast, predicts that they cannot occur immediately after the w/z-phrase but must occur near the end of the clause, (16b). CLEFT ANALYSIS (16)

a.

IP

FRONTING ANALYSIS b.

DP

CP whf

Δ 1°

PredP PredP

no...

C° no

IP

particle

Γ

h

Δ wh

1°

PredP PredP

particle

Δ The prediction of the cleft analysis is correct. Post-predicate particles immediately follow the wh-phrase, (17). The fronting analysis does not permit these grammatical examples. (17)

a. iza daholo no nihinana vary? who all PRT eat.ACT vary ' W h o all ate rice?' b. iza anie no manapaka bozaka? who EXCL PRT cut.ACT grass ' W h o is really cutting the grass?' c. iza foana no mihomehy? who always PRT laugh.ACT ' W h o is always laughing?'

The clause-final position predicted by the fronting analysis is also largely grammatical, (18), although I have no explanation for the ungrammaticality of (18a) with daholo. These data are also accounted for by the cleft analysis, however, because there is a second predicate within the subject

204

Eric Potsdam

headless relative clause of these examples and it permits a PredP adjunct. These data do not differentiate the two hypotheses. (18)

a. *iza no nihinana vary daholo? who PRT eat.ACT vary all ('Who all ate rice?') b. ? oviana no nanapaka bozaka Rasoa when PRT cut. ACT grass Rasoa 'When really did Rasoa cut the grass?' c. zaza iza no mitomany foana? child which PRT crying.ACT always 'Which child is always crying?'

4.1.2. Pre-predicate

anie? EXCL

particles

Malagasy also has a number of particles that immediately precede the predicate in VOS clauses. These pre-predicate particles are largely inflectional in nature and include toa 'seem', tokony 'should' (Paul 2001a), tena 'indeed (affirmative emphasis)', and h(o)- 'future/irrealis': (19)

a. tokony hamangy an-dRabe Rasoa. should visit.ACT ACC-Rabe Rasoa 'Rasoa should visit Rabe.' b. tena hovidin' ny zaza ny fiaramanidina. EMPH buy.PASS the child the airplane 'The child will indeed buy the airplane.' c. ho lasa dokotera Rabe. FUT depart doctor Rabe 'Rabe will be a doctor.'

The cleft analysis of questions correctly predicts that these particles also immediately precede the wh-phrase in a w/z-question because it too is a predicate. See (20) (next page). Under the fronting analysis, such data are unexpected because the whphrase is not a predicate. Instead, the particles should necessarily appear farther to the right, before the verbal predicate. This position is possible, (21), but it is allowed by both analyses because the verb constitutes a predicate under either approach.

The Cleft Structure of Malagasy Wh-Questions (20)

a. tokony iza no hamangy an-dRabe? should who PRT visit.ACT ACC-Rabe 'Who should visit Rabe?' b. ten a inona no hovidin' ny zaza? EMPH what PRT buy.Ρ ASS the child 'What will the child indeed buy?' c. %ho iza no vadinao?6 FUT who PRT marry.PASS.2SG 'Who (among them) will you marry?'

(21)

a. iza no tokony hamangy an-dRabe? who PRT should visit.ACT ACC-Rabe 'Who should visit Rabe?' b. inona no tena hovidin' ny zaza? what PRT EMPH buy.PASS the child 'What will the child indeed buy?' c. iza no ho vadinao? who PRT FUT marry.PASS.2SG 'Who will you marry?'

205

One might defend the fronting analysis by saying that these particles are not pre-predicate particles but clause-initial particles. If that were the correct description, the fronting analysis would account for the data in (20). Then however, the grammatical examples in (21) would remain unexplained since they are not clause-initial. In any case, there is independent evidence that clause-initial is not the correct description of the positioning of the particles. There is a topicalization construction in Malagasy in which a constituent is preposed, followed by the topic particle dia (Keenan 1976): (22)

a. Rasoa dia manoroka an-dRabe. Rasoa TOPIC kiss.ACT ACC-Rabe 'Rasoa, she kisses Rabe.' b. ny fiaramanidina dia hovidin' the airplane TOPIC buy.PASS 'The airplane, the child will buy it.'

ny the

zaza. child

Both Paul (2001a) and Flegg (2003) suggest that the initial constituent in the topicalization construction is not a predicate and, indeed, the pre-

206

Eric Potsdam

predicate particles cannot occur initially in this construction; they must precede the verbal predicate: (23)

a. (* tokony) Rasoa dia (tokony) hanoroka should Rasoa TOPIC should kiss.ACT 'Rasoa, she should kiss Rabe.' b. (* tena) ny fiaramanidina dia (tena) hovidin'

an-dRabe. ACC-Rabe ny

zaza.

EMPH the airplane TOPIC EMPH buy.PASSthe ' T h e airplane, the child will indeed b u y it.'

child

c. (* ho) Rabe dia FUT Rabe TOPIC 'Rabe, he will kiss Rasoa.'

h-anoroka FUT-kiss.ACT

an-dRasoa ACC-Rasoa

If the particles under consideration were clause-initial particles, such data would be unaccounted for since here we see the particles associating exclusively with predicates, not clause-initial position. In summary, the fronting analysis does not allow the grammatical placement of a wide variety of clause-internal particles, while the cleft analysis does. The w/z-phrase in a w/z-question behaves like a predicate with respect to the positioning of these particles, as claimed in the cleft analysis.7

4.2. Parallels with the focus construction Malagasy has a focus construction illustrated in (24) that is formally similar to w/i-questions. Dahl (1986) first proposed that this focus construction is a kind of cleft and Paul (2001a, 2003a) develop this analysis, assigning (24a) the pseudocleft structure in (25). The focused element is the predicate of the clause and the subject is a headless relative clause. (24)

a. Rasoa no nihomehy. Rasoa PRT laugh.ACT 'It was Rasoa who laughed.' b. ny mofo no novidin-dRasoa. thebread PRT buy.PASS-Rasoa 'It was the bread that was bought by Rasoa.'

The Cleft Structure of Malagasy Wh-Questions (25)

[[ pred Rasoa ] [sub|ect/hr no 0Pi nihomehy Rasoa PRT laughed lit. 'The one who laughed was Rasoa.'

207

*,·]]

There are a number of parallels between the focus construction and whquestions which suggest that they should receive the same structural analysis. First, both are formed by preposing a constituent and following it immediately with the invariant particle no. Second, the two constructions place a focus interpretation on the initial XP. fFft-phrases indicate a request for new information in the same way that focused XPs supply new information. Third, the two constructions are subject to the same extraction restriction in (11), that only subjects and adjuncts can be extracted. In (26), an adjunct is focused regardless of the voice of the verb (compare to the w/z-questions in (10)). In (27), we see the ungrammaticality that results from focusing a non-subject argument. (26)

a. ao ambanin' there under nanafina ny hide.ACT the b. ao ambanin' there under nafenin' ny hide.PASS the c. ao ambanin' there under nanafenan' ny hide.CIRC the 'It's under the bed

ny the lakile key ny the zaza child ny the zaza child that the

fandriana no bed PRT ny zaza. the child fandriana no bed PRT ny lakile. the key fandriana no bed PRT ny lakile. the key child hid the key.'

(27)

a. *ny mofo no nividy i Bao. the bread PRT buy.ACT Bao ('It's the bread that Bao bought.') b. *Rabe no novidina ny akoho. Rabe PRT buy.PASS the chicken ('It's Rabe that the chicken was bought by')

Fourth, neither construction allows a resumptive pronoun corresponding to the fronted element:

208

Eric Potsdam

(28)

a. *iza no nihomehy izy? who PRT laugh. ACT 3SG.NOM ('Who laughed?') b. *Rasoa no nihomehy izy. Rasoa PRT laugh. ACT 3SG.NOM ('It was Rasoa who laughed.')

Finally, the focus construction shows the same predicate behavior of the initial constituent with respect to particle placement documented for whquestions in section 4.1 (Paul 2001a). Post-predicate particles immediately follow the initial focused element, (29), and pre-predicate particles immediately precede the initial element, (30). (29)

a. ireo lehilahy ireo daholo no milalao these man these all PRT play .ACT 'It's all these men who are playing ball.' b. Rasoa anie no nanapaka bozaka. Rasoa EXCL PRT cut.ACT grass 'It was really Rasoa who cut the grass!' c. Rasoa foana no mihomehy. Rasoa always PRT laugh.ACT 'It's always Rasoa who laughs.'

baolina. ball

(30)

a. tokony Rasoa no hamangy an-dRabe. should Rasoa PRT visit.ACT ACC-Rabe 'It's Rasoa who should visit Rabe.' b. tena Rabe no mahandro ravintoto. EMPH Rabe PRT cook.ACT pounded.manioc.leaves 'It's indeed Rabe who cooks ravintoto.' 8

Analyzing w/z-questions as clefts immediately accounts for these parallels since the two constructions would have the same syntax. The parallels are perhaps unexpected under the fronting analysis in which the two constructions have rather different derivations and they may require independent explanations in each case.

The Cleft Structure of Malagasy Wh-Questions

209

4.3. Multiple-w/z questions The final argument for the cleft analysis comes from a consideration of multiple-w/z questions introduced in Paul (2000) and Sabel (2003), in which there are two w/z-phrases at the front of the clause.9 (31a,b,c) are from Sabel (2003: 245-246), (31d,e) are from Paul (2002: 202). I will suggest that the cleft analysis provides a better treatment of such data. To see this we need to lay out how each hypothesis would handle such data. I then provide arguments in favor of the cleft-based approach. (31)

a. aiza iza no mividy ny vary? where who PRT buy.ACT the rice 'Who buys rice where?' b. aiza (ny) inona no vidinao? where the what PRT buy.PASS.2SG 'What do you buy where?' c. nahoana iza no mividy ny vary? why who PRT buy.ACT the rice 'Why does who buy the rice?' d. oviana iza no lasa nody? when who PRT left go.home.ACT 'Who went home when?' e. taiza Hay inona no novidin-dRasoa? where that what PRT buy.PASS-Rasoa 'Where did Rasoa buy which of these things?'

4.3.1. Sabel's (2003) fronting

analysis

Sabel (2003) proposes a clever analysis of the multiple-w/z question examples within the fronting approach. It argues that Malagasy is an optional multiple w/z-fronting language like some of the Slavic languages. In such languages, all w/z-phrases move to clause-initial position as seen in the Bulgarian data in (32) (next page) (examples from Pesetsky 2000: 19, see also Rudin 1988, Boskovic 2002, Richards 2002, and references therein). Rudin (1988) first argued that there are two types of multiple w/zfronting languages. In Polish / Serbo-Croatian type languages, fronted w/zphrases move to separate CP specifiers. In Romanian / Bulgarian type

210

Eric Potsdam

languages, the fronted w/z-phrases move to a single spec,C where they form a constituent. Sabel argues that Malagasy is of this latter type: "multiple w/z-elements in Malagasy undergo cluster formation and occupy one sentence-initial spec,C position, similar to languages such as Bulgarian and Romanian" (Sabel 2003: 243). (32)

a. koj kakvo na kogo dade? who what to whom gave 'Who gave what to whom?' b. *koj dade kakvo na kogo? who gave what to whom

Sabel (2001, 2003) lay out the theoretical apparatus that achieves cluster formation. He first assumes that w/z-phrases need to move to spec,C to check a strong [FOCUS] feature of C°. General feature-checking mechanisms given in (33) and (34) allow cluster formation in line with the proposal in (35). (33) Attract F (Chomsky 1995) Κ attracts F if F is the closest feature that can enter into a checking relation with a sublabel of Κ (34)

Closeness (Chomsky 1995) β is closer to Κ than α if β c-commands α

(35)

Cluster Hypothesis (Sabel 2003: 243) A feature F which is attracted by Κ attracts a feature of the same type F

fFTz-cluster formation proceeds as follows: Assume a pre-movement structure of the form [ C° [ ... whl ... [ ... whl ... ]]] in which C° ccommands w/z-phrase 1 which in turn c-commands w/z-phrase2 and all three elements have a [FOCUS] feature that must be checked. By (33), C° can attract whl but it cannot attract wh2 because whl is closer according to (34). If it were to do this however, the strong [FOCUS] feature on whl would go unchecked, causing the derivation to crash. Given the Cluster Hypothesis, an alternative derivation is available: whl attracts whl, forming a cluster. C° then attracts the cluster of whl and whl.

The Cleft Structure of Malagasy Wh-Questions

211

To the multiple-w/z question repeated in (36a), the derivation in (36b) is assigned. Iza 'who' in spec,I attracts aiza 'where', which left adjoins to it. The cluster in spec,I then moves to spec,C. Crucially, the cluster is formed prior to movement to spec,C. (36) a. aiza iza no mividy ny where who PRT buy.ACT the 'Who buys rice where?'

vary? rice

CP

b.

DP, DP*

DP

Δ

Δ

aiza

IP

C° no

iza

Γ

t,

mividy ny vary tk

4.3.2. Paul's (2003a) cleft analysis Paul (2003a) indirectly provides an alternative analysis of the multiple-w/i questions within the cleft approach. The analysis rests on the existence of another construction in Malagasy which Keenan (1976) named the BODYGUARD CONSTRUCTION (see also Paul 2000). Keenan (1976) first observed that when a non-subject is questioned or focused, the subject may optionally appear immediately after the wh-phrase / focus, and before no. The subject serves as a "bodyguard" for the fronted element. (37b) illustrates the bodyguard variant of the w/z-question in (37a) and (38b) illustrates the bodyguard variant of the focus construction in (38 a). The bodyguard construction is only possible when the first XP is an adjunct and the second XP is the subject.

212

Eric Potsdam

(37)

a. aiza no where PRT b. aiza Rasoa no where Rasoa PRT 'Where does Rasoa buy

(38)

nividy no a. omaly PRT buy.ACT yesterday nividy Rabe no b. omaly yesterday Rabe PRT buy.ACT 'It was yesterday that Rabe bought

mividy buy.ACT mividy buy.ACT bread?'

mofo Rasoa? bread Rasoa mofo? bread

vary Rabe. rice Rabe vary. rice rice.'

Paul (2003a) analyzes the bodyguard construction, arguing that the bodyguard phrase is in the specifier position of the DP subject. It is a kind of possessor. This correctly places the bodyguard between the focus and the particle no. To the focus construction with bodyguard in (38a) Paul assigns the structure in (39). (39)

nividy vary tt

The Cleft Structure of Malagasy Wh-Questions

213

Given Paul's analysis, we have a natural account of the multiple-w/z questions, repeated in (40a). They are w/z-questions in which the bodyguard in spec,D is also a wh-phrase, (40b). (40)

a. aiza iza no mividy ny where who PRT buy.ACT the 'Who buys rice where?' b.

vary? rice

IP DP Γ

PredP

DP,

Δ

Δ

aiza

iza

D' D° no

CP Ofj nividy vary t{ t^

4.3.3. Against the fronting analysis of multiple-wh questions In this subsection I present four arguments in support of the cleft treatment of the bodyguard construction and multiple-wA questions, in (40). My discussion owes much to Paul (2000: ch. 4) which investigates the same data and arrives at the same conclusion regarding the untenability of a cluster analysis. The first argument comes from a consideration of the ordering of the initial phrases and superiority. Pesetsky (2000) and Boskovic (2002), following Rudin (1988), show that Bulgarian multiple wA-fronting examples exhibit superiority effects. The wh-phrase that originates structurally higher must appear before the w/z-phrase that originates structurally lower (Pesetsky 2000: 22): (41)

a. koj k de udari Ivan? who where hit Ivan 'Who hit Ivan where?'

b. * k de koj udari Ivan? where who hit Ivan (* 'Where did who hit Ivan?')

Sabel and Paul (2000: 202) show that Malagasy patterns with Bulgarian in restricting the order of fronted w/z-phrases, (42) (from Sabel 2003: 245).

214

Eric Potsdam

(42)

a. *iza aiza no mividy ny who where PRT buy.ACT the (* 'Where does who buy rice?') b. aiza iza no mividy ny where who PRT buy.ACT the 'Who buys rice where?'

vary? rice vary? rice

There is a difference between the two languages, however, in that they show the opposite licit orderings. In Bulgarian we see who where but in Malagasy we see where who. The question that arises under the fronting analysis is why Malagasy shows anti-superiority. Given that the whphrases form a cluster, it must be the case that cluster formation proceeds via left adjunction in Malagasy but right adjunction in Bulgarian. The theory-internal issue is how to capture this parametric variation. It can be stipulated, of course, but this is unsatisfactory and I see no principled explanation given the widely accepted robustness of superiority (at least in the simple cases under consideration here). Under the cleft analysis, by contrast, the correct ordering in (42b) is fixed by the phrase structure. The bodyguard must be a subject and being in spec,D it must follow the initial adjunct w/z-predicate. A second argument for the cleft analysis of multiple-w/z questions comes from examples with multiple fronted w/z-adjuncts. We have already seen that w/z-questions in Malagasy are sharply restricted: only subjects and certain adjuncts may be extracted. When one of each fronts, the subject must be rightmost. Even taking these restrictions into account, if Malagasy is a multiple w/z-fronting language, it should still be acceptable to front other combinations of w/z-phrases, specifically, two w/z-adverbials. (43) shows that this is not possible however, regardless of whether or not a subject bodyguard is present (43d) is from Paul 2000: 203).10 (43)

a. * aiza oviana no mividy vary Rasoa? where when PRT buy.ACT rice Rasoa b. * oviana aiza no mividy vary Rasoa? when where PRT buy.ACT rice Rasoa c. *oviana aiza Rasoa no mividy vary? when where Rasoa PRT buy.ACT rice ('Where does Rasoa buy rice when?')

The Cleft Structure ofMalagasy Wh-Questions 215 d. *nahoana taiza iza no nividy ilay why where who PRT buy.ACT that ('Why did who buy that book where?')

boky? book

By contrast, true multiple w/z-fronting languages like Bulgarian, allow multiple adjunct questions:11 (44)

koga k de jade Ana? when where eat Ana 'When did Ana eat where?'

This lack of parallelism between Malagasy and Bulgarian is surprising if the two are analytically the same kind of multiple w/z-fronting language. If there is a well-formed cluster formation derivation for (44), there should also be one for (43). The cleft analysis in contrast explains (43), as Paul (2000) notes. The sentences cannot be generated since the second wh-adverbial can be neither a second w/z-predicate, since only one predicate is allowed, nor a bodyguard, since it is not a subject. The third problem for the cluster formation / fronting analysis comes from a consideration of the discourse status of the bodyguard when it is not a wh-phrase. Paul (2000, 2003a) showed that the bodyguard is not part of the focus (new information) but, rather, is topical (old information). As such, the bodyguard must be definite, (45), it cannot constitute the answer to an information question, (46), and it is a preferential position for pronouns, (47). The data in (45) to (47) are from Paul (2003a). (45)

tany

an-tokotany

there ACC-yard

* (ny) the

zazavavy

no

nilalao

girl

PRT

play. ACT ball

'It was in the yard that the girls were playing ball.' (46)

a. iza no nanapaka bozaka oviana? who PRT cut.ACT grass when 'Who cut grasss when?' b. # omaly Rasoa no nanapaka bozaka. yesterday Rasoa PRT cut.ACT grass (not acceptable as an answer to (46a)) c. Rasoa no nanapaka bozaka omaly. Rasoa PRT cut.ACT grass yesterday 'It was Rasoa who cut grass yesterday.'

baolina.

216

Eric Potsdam

(47)

a. tany an-tsena izy no nandeha fiara. there market 3SG.NOM PRT go.ACT car 'It was to the market that she went by car.' b. ?tany an-tsena no nandeha fiara izy. there market PRT go.ACT car 3SG.NOM 'It was to the market that she went by car.'

This observation is not compatible with the fronting analysis in which the bodyguard is part of the focus in spec,C. Further, it creates a technical problem in the cluster formation operation. Sabel argued that cluster formation is driven by attractions of [FOCUS] features on C° and the whphrase. However, a bodyguard need not be a wh-phrase as we have already seen, (48), and such non-wh bodyguards do not have characteristics of a focus. (48)

(49)

a. oviana Rabe no nividy ny vary? when Rabe PRT buy. ACT the rice 'When did Rabe buy the rice?' b. omaly Rabe no nividy ny vary. yesterday Rabe PRT buy.ACT the rice 'It was yesterday that Rabe bought rice.' aiza iza no mividy ny vary? where who PRT buy.ACT the rice 'Who buys rice where?'

While a [FOCUS] feature may be driving movement of the w/z-bodyguard in multiple-w/z questions, such as the example repeated above in (49), it is not likely to be driving the movement of a non-w/z-bodyguard. As a result, the fronting analysis would not seem to be able to unify both types of bodyguard constructions. Multiple w/z-fronting would be cluster formation driven by the need to check [FOCUS] features on the w/z-phrases. In nonw/z-bodyguard examples like (48), on the other hand, there is no feature which would license cluster formation that is shared by the focus XP (ioviana 'when' or omaly 'yesterday') and the topic bodyguard (Rabe); in particular, there is no shared [FOCUS] feature. Regardless of how (48) and (49) are ultimately generated, the claim that they are derived by different mechanisms seems theoretically costly. Under the cleft analysis, in contrast, the bodyguard is not part of the predicate focus; it is part of the

The Cleft Structure of Malagasy Wh-Questions

217

subject and has a uniform derivation and discourse status in both examples.12 Fourth, there is constituency evidence against the claim of the fronting analysis that the two initial elements form a constituent. The fronting analysis follows Rudin (1988) in taking the fronted w/i-phrases to form a cluster, as schematized in (50a). Data below will support the pseudocleft analysis, in which the Wz-phrase and the bodyguard do not form a constituent, (50b). (50)

a. [ C P [ WH BODYGUARD] B- [iP [predp

WH

[ c no ...]]

J [DP BODYGUARD no ...]]

FRONTING ANALYSIS CLEFT ANALYSIS

Coordination data in (51) shows that the bodyguard and following material can be coordinated to the exclusion of the w/z-phrase. (51)

oviana Rasoa no nijinja vary ary when Rasoa PRT harvest.ACT rice and Rakoto no nanapaka bozaka? Rakoto PRT cut.ACT grass 'When did Rasoa harvest rice and Rakoto cut grass?'

Such examples are straightforwardly accounted for with the pseudocleft structure since the coordinated material is a constituent, as reflected in (50b). Under the fronting analysis however, there is no constituent being coordinated, see (50a), and the result should be ungrammatical, contrary to fact. Further, the fronting analysis predicts that one should be able to coordinate the wA-phrase+bodyguard, to the exclusion of no and the remaining material. This is completely impossible: (52)

* [ oviana Rabe ] sy/ary [ taiza Rasoa ] when Rabe and where Rasoa no nividy mofo? PRT buy.ACT bread ('When did Rabe buy bread and where did Rasoa buy bread?')

Coordination thus supports the cleft structure over the fronting structure. Post-predicate particles also treat the initial w/z-phrase structurally separate from the bodyguard. When both are present, the particle follows the w/z-phrase, not the bodyguard:

218

Eric Potsdam

(53)

a. taiza (daholo) Rakoto (* daholo) no where all Rakoto all PRT 'Where all did Rakoto go?' b. oviana (anie) Rasoa (* anie) no when

EXCL

Rasoa

EXCL

PRT

'When really did Rasoa cut the grass?' c. aiza (foana) Rasoa (*foana) where always Rasoa always 'Where does Rasoa always buy rice?'

nandeha? go.ACT nanapaka

bozaka?

cut.ACT

grass

no PRT

mividy vary? buy.ACT rice

In conclusion, there is good empirical evidence that Paul is correct in proposing the general structure in (39) for the bodyguard construction and, by extension, the structure for multiple-w/j questions in (40). The bodyguard is part of the subject constituent and not the predicate. No operation treats the two initial phrases as a constituent, contra the fronting analysis but in line with the pseudocleft analysis. Several independent strands of evidence thus point to the cleft analysis as superior for analyzing multiple frontedwh questions.

4.4. Interim summary This chapter has argued thus far that w/j-questions in Malagasy are not derived by wA-movement. They are pseudocleft structures in which the initial w/z-phrase is a base-generated predicate and the remaining material is a headless relative clause in subject position: (54)

IP DP 1°

PredP wh-phrase

D°

CP relative clause

This pseudocleft analysis of w/i-questions was shown to be superior in a number of independent domains. It better predicts the position of predicaterelated particles in w/z-questions, it unites the analysis of w/z-questions and the focus construction, and it provides a superior analysis of multiple-w/z

The Cleft Structure of Malagasy Wh-Questions

219

questions and the bodyguard construction. It is worth pointing out that the above argumentation does not rule out a dual analysis scenario in which w/z-questions have both fronting and cleft structures. We have only ruled out the possibility that w/z-questions have only a fronting analysis. Given that a dual analysis scenario is not the optimal situation, I will not consider it further. Observe that, given (54), the subject of the w/z-phrase predicate need not contain a relative clause. The subject can be a simple DP (D° NP structure). Most generally then, the structure of w/z-questions is as in (55), where the clause-final subject is some independently acceptable Malagasy DP. IP

(55) Γ 1°

DP PredP

wh-phrase This more general structure is instantiated by non-verbal questions as in (56). In these examples, the DP subject is not complex. It is a simple DP, pronoun, or a name. (56)

a. iza ny prezida? who the president 'Who is the president?' b. an' iza ity boky ity? to who this book this 'Whose is this book?'

c. ho

aiza ianao? FUT where 2SG.NOM 'Where will you be?' d. aiza i Soa? where DET Soa 'Where is Soa?'

Strictly speaking, these sentences are not clefts, as they are not biclausal. For simplicity, however, I will continue to refer to the analysis represented by (55) as the cleft analysis regardless of whether or not the subject DP is structurally complex. The picture we are led to is that Malagasy is a w/z-in-situ language. Whphrases may appear in all positions: argument position, adjunct position, or predicate position and there is never w/z-movement. This is a welcome result given Cheng's (1997) claim that there are no optional w/z-movement

220

Eric Potsdam

languages. In the following section, I assume the cleft analysis of whphrase-initial sentences and discuss some analytical challenges that it presents for future work.13

5.

Further issues

In this section I explore some of the details of the cleft analysis. Space considerations prevent me from fully addressing them; however, I introduce them because they are a previously undocumented part of the larger domain of data that any analysis of the Malagasy focus / w/z-question construction must confront.

5.1. The status of no A central analytical detail within the pseudocleft analysis is the treatment of the particle no. Assuming that no is a head, it could be either a determiner or a relative clause complementizer in C°. Both options are shown in the schematic in (57). IP

(57)

DP 1°

PredP

D°

CP Op, c°

IP Γ

h There is another place in the grammar where no appears which suggests that it is a complementizer.14 No is possible in relative clauses just in case the modified noun phrase is of the form na wh-XP na wh-XP 'w/j-ever':

The Cleft Structure of Malagasy Wh-Questions

(58)

221

a. voasazy [ na iza na iza no tsy mamafa lalana ] punish.PASS or who or who PRT NEG sweep road 'Whoever doesn't sweep the road will be punished.' b. * voasazy [ ny mpiasa no tsy mamafa lalana ] punish.PASS the worker PRT NEG sweep road ('The worker who doesn't sweep the road will be punished.')

Such examples are preferred with the DP topicalized, confirming that the bracketed material is indeed a constituent containing a relative clause: (59)

a. [ na or

iza

na iza

no

tsy

mamafa lalana ]

who

or w h o

PRT

NEG

sweep

dia

voasazy.

TOPIC

punish.PASS

road

'Whoever doesn't sweep the road will be punished.' b. [ na boky inona na boky inona no hovidinao ] or book what or book what PRT buy.PASS.2SG dia hovakiko. TOPIC

read.PASS.lSG

'Whatever book you buy, I will read.' c. [ na oviana na oviana no milasy or when or when PRT camp dia avy ny orana. TOPIC come the rain 'Whenever I go camping, it's raining.'

aho ] lSG.NOM

Such strings can also be resumed with a pronoun, again suggesting the indicated constituency: (60)

[ na or dia TOPIC

iza na iza no olo-meloka ] who or who PRT criminal iza no hanasazy azy? who

PRT

punish

3SG.ACC

'Whoever is a criminal, who will punish him?' No in the above examples alternates with the uncontroversial relativizer izay and its null counterpart (see Rajemisa-Raolison 1969: 67 for discussion of Malagasy relativizers):

222 (61)

Eric Potsdam [na

iza

or

who

na iza

0/izay/no

tsy

mamafa lalana ]

or w h o

REL/REL/PRT

NEG

sweep

dia

voasazy.

TOPIC

punish.PASS

road

'Whoever doesn't sweep the road will be punished.' Assuming that there is only one no in the language, these data indicate that it is a relative complementizer. Extending this conclusion to w/z-questions and the focus construction, their subjects must contain a relative clause. The subject is a headless relative, as Paul (2001a) originally asserted and as assumed above. In the next section, I investigate this claim further.

5.2. Headless relatives If no and the following material, what I will call no DPs, are headless relative clauses with an approximate meaning "the one that ...", we expect such strings to have the syntactic characteristics and distribution of other headless relatives in the language. In this section, I explore this general prediction. For comparison, I will look at uncontroversial headless relative clauses in Malagasy formed with izay (Keenan 1976): (62)

a. izay manasa lamba REL wash clothes 'who(ever) washes clothes'

b. izay vakinao REL break.PASS.2SG 'what(ever) you break'

I first point out two syntactic characteristics that no DPs share with headless relatives, supporting their headless relative clause status. I then discuss the distribution of no DPs and suggest that in this domain they do not behave like headless relatives clauses. Their distribution is more restricted. There are two unusual characteristics that no DPs share with izay headless relatives: they cannot be preceded by an overt determiner and the internal predicate may not be a DP. (63) illustrates that izay headless relatives cannot have an overt determiner preceding them. (63)

*ny/ilay izay nihomehy the/that REL laugh ('the/that one who was laughing')

The Cleft Structure of Malagasy Wh-Questions

223

This is also true for no DPs. They also cannot be preceded by a determiner, (64)

a. *iza [ny/ilay no nihomehy ]? who the/that PRT laugh ('Who is the/that one who laughed?') b. *inona [ny no vaovao ]? what the PRT news ('What's the news?')

Given the structure proposed in (57) for a headless relative with no in C°, this is perhaps surprising since the D° head is unoccupied. The fact that they share this property however suggests that no DPs are also headless relatives and the impossibility of an overt determiner is a general property of the structure. If (57) is roughly correct, there must be a null determiner that selects for headless relatives. While this may be stipulative, it appears to be a fact of the language. The second property shared by izay headless relatives in (65) and no DPs in (66) is that the predicate within the (bracketed) relative clause cannot be a DP (see also Paul 2000: 168): (65)

* te

hahalala

[ izay ny

prezida ]

aho.

want know REL the president lSG.NOM ( Ί w a n t to meet the one w h o is the president.')

(66)

a. *iza [no ny prezida]? who PRT the president ('Who is the president?') b. *an' iza [no ity boky ity]? to who PRT this book this ('Whose is this book?') c. *ho aiza [ no ianao ]? d. * aiza [ no i Soa ]? FUT where PRT 2SG.NOM where PRT DET Soa ('Where will you be?') ('Where is Soa?')

The predicates within the subject relative clauses of (66) are all DPs. This is clear when an overt determiner or framing demonstrative is present, (66a,b). 15 It is also reasonably the case with pronouns, (66c), which are analyzed as determiners (Zribi-Hertz and Mbolatianavalona 1999 for Malagasy, Postal 1970, and Abney 1987). Names in Malagasy are also DPs

224

Eric Potsdam

with an obligatory determiner, either i as in (66d) or ra-, which is morphologically incorporated into some other names like Rasoa. The examples in (66) are all fully grammatical if no is removed, (56), in which case the DPs are subjects of the w/z-predicates and there is no headless relative clause. These data show that the predicate in a headless relative cannot be a DP. While I have no satisfactory accounts of the two restrictions, the fact that they are shared by no DPs and izay headless relatives can be taken as evidence that the former are also headless relatives, as asserted by the pseudocleft analysis. At the same time, a more obvious expectation concerning no DPs is that they will have the same external distribution as DPs and izay headless relatives.16 In fact, they do not. No headless relatives only appear in subject position, (67a), and not predicate, direct object, object of preposition or fronted topic positions, (67b-e). (67)

SUBJECT a. iza no mihomehy? who PRT laugh 'Who is laughing?' *PREDICATE b. *no mihomehy Rabe. PRT laugh Rabe ('Rabe is the one who is laughing.') c. *mahalala [no mihomehy ] aho. *DIRECT OBJECT know PRT laugh lSG.NOM ( Ί know the one who is laughing.') d. *nipetraka tao ankaikin' [ no nihomehy ] aho. * P OBJECT sit there beside PRT laugh lSG.NOM ( Ί sat beside the one who was laughing.') e. *[ no nihomehy ] dia iza? *FRONTED TOPIC PRT laugh TOPIC who ('As for the one who laughed, who is he?')

Izay headless relatives are not so restricted in distribution, they can appear in a non-predicate position: (68)

a. iza [ izay mihomehy ] ? who REL laugh 'Who is the one who is laughing?'

SUBJECT

The Cleft Structure ofMalagasy Wh-Questions 225 b. * [ izay mihomehy} Rabe. REL laugh Rabe ('Rabe is the one who is laughing.') c. mahalala [ izay mihomehy ] aho. know REL laugh Ί k n o w the one w h o is laughing.'

d. te

sit

eo

ankaikin'

DIRECT OBJECT

lSG.NOM

hipetraka

want

*PREDICATE

Ρ OBJECT

[ izay

hiresaka amiko ]

aho.

h e r e beside REL talk with.lSG lSG.NOM Ί w a n t to sit beside the one w h o will talk with m e . '

e. izay REL

mihomehy

dia

laugh

TOPIC

faly.

FRONTED TOPIC

happy

'The one who is laughing is happy.' If the pseudocleft analysis is correct, the distribution of no headless relatives is sharply restricted and this will need to be accounted for in future work.

6.

Conclusion

This chapter has argued that w/z-questions in Malagasy are base-generated pseudocleft structures. The initial wh-phrase is a predicate and the remaining material is the clause-final DP subject. (69)

IP DP PredP wh-phrase

Having established the pseudocleft structure, a number of analytical details arose concerning the particle no that appears when the subject DP is not a simple DP. I claimed that it is the complementizer head of a headless relative clause attached to the subject, as shown in (70) (next page). The no

226

Eric Potsdam

DP has some characteristics of other headless relatives in Malagasy but its distribution is more restricted. IP

(70)

DP 1°

PredP wh-phrase

CP

D°

C

Opi C° no

IP

Unlike English and other well-known Indo-European languages then, Malagasy does not employ w/z-movement at all. Rather, it is like Chinese or Japanese in being a w/z-in-situ language. W/z-phrases are either in-situ or they are predicates. Malagasy thus reraises a typological issue - one that has been addressed in many places, in numerous guises - namely, why do some languages make use of overt w/z-movement while others do not?

Notes *

1.

2.

I would like to thank my Malagasy consultants Charlotte Abel-Ratovo, Tina Boltz, Noro Brady, Annie Rasoanaivo, Hasina Randriamihamina, and Voara and Bodo Randrianasolo. All data are from my own notes unless otherwise indicated. Thanks also to an anonymous AFLA abstract reviewer, the reviewers for this volume, the audience at AFLA XI, Paul Law, Ileana Paul, Maria Polinsky, and especially Joachim Sabel for discussions during the writing of this paper. This work is supported by NSF grant BCS-0131993. I use the following abbreviations in glossing: 1/2/3 = person, ACC = accusative, ACT = active voice, CIRC = circumstantial voice, DET = determiner, EMPH = emphatic, EXCL = exclamative, FUT = future, NEG = negative, NOM = nominative, PASS = passive voice, PRT = particle, REL = relativizer, SG = singular. There is considerable debate as to whether this clause-final DP is actually a subject or an A' topic-like element. I continue to refer to it as a subject for

The Cleft Structure of Malagasy Wh-Questions

3.

4.

5.

6.

7.

8.

227

convenience, without taking a stand on the issue. See Pearson (2005) for discussion. Other researchers argue that the subject-final order is derived from subjectinitial order by predicate fronting (Massam and Smallwood 1997, Pearson 1998, 2001, Rackowski 1998, Rackowski and Travis 2000, Massam 2000, Aldridge 2002, Travis 2004, this volume, Chung 2005). For the purposes of this chapter, this innovation is not important. Paul (2001a) suggests that anie may be a second-position clitic. If this is correct, it would not be a post-predicate particle and would not provide evidence in favor of the cleft analysis of w/z-questions. For this reason, I use a variety of particle types above. Pearson (1998) documents the availability of rightward object shift with definite objects in Malagasy. In such cases, an adverb like foana 'always' may occur to the left of the shifted definite object. I ignore such cases here. This position is still not possible for daholo and anie. The possibility of tense marking on w/z-phrases varies with the w/z-phrase but the phenomenon is compatible with the predicate status of wh-phrases. Aiza 'where' is normally marked for tense: ho aiza 'where (fiit.)' and taiza 'where (past)'. Similarly 'when' has two forms oviana 'when (non-future)' and rahoviana 'when (fixture)'. Iza 'who' is more unusual in this context but (20c) shows that tense marking is nonetheless possible in this context as well for some speakers. Malagasy has a number of other particles which, for various reasons, cannot be used in these tests. The post-predicate question particle ve (Paul 2001b) does not occur in w/z-questions, only yes/no questions. The pre-predicate negative particle tsy is semantically incompatible with the main w/z-predicate, as evidenced by the ill-formedness of the English * Who isn't the one who laughed? (Potsdam 2004). The focus construction is not possible with an initial future marker ho and a DP predicate: (i) * ho Rabe no (ho)fidina. FUT R a b e

PRT

elect.PASS

('It will be Rabe who is elected.') I have no explanation for this fact but note that some w/z-questions with initial ho, specifically those with iza 'who' as in (20c), are also dispreferred and judged ungrammatical by some speakers. 9. My consultants did not accept such examples. I am grateful to Joachim Sabel and Elisabeth Ravaoarimalala for help with additional multiple-w/z examples. 10. Such examples are acceptable if the w/z-phrases are coordinated since then they form a single phrase: (i) a. taiza sy oviana no nividy mofo Rasoa? where and when PRT buy.ACT bread Rasoa

228

Eric Potsdam b. oviana when

11. 12.

13.

14. 15. 16.

sy

taiza

no

nividy

mofo

Rasoa?

and

where

PRT

buy.ACT

bread

Rasoa

c. oviana sy taiza Rasoa no nividy mofo? when and where Rasoa PRT buy.ACT bread 'Where and when did Rasoa buy bread?' Thanks to Veronica Gerassimova and Maria Jordan for help with the relevant data in Bulgarian and Romanian. Given Paul's claim that the bodyguard is a topic, a vv/?-phrase bodyguard is perhaps somewhat unexpected since it would not seem to have topic properties. Paul (2000: 197, 203) reports that her consultant who allowed wh-bodyguards required that it be d-linked in the sense of Pesetsky (1987). There must be a contextually salient set of people or things over which the answer to the w/i-phrase ranges. The presence of the determiner on the w/z-bodyguards in (31b,e) supports this proposal. If this general conclusion is correct, it has the consequence that Malagasy will be significant for the analysis of partial wA-movement documented in Sabel (2003), (i). If such examples are not exceptional, they too should not involve any actual w/i-movement, partial or otherwise. (i) heverin-dRabe fa inona no novidin-dRakoto? think.PASS-Rabe that what PRT buy.PASS-Rakoto 'What does Rabe think that Rakoto bought?' Rahajarizafy (1960: 160) documents a number of additional uses of no which I leave for future investigation. See Zribi-Hertz and Mbolatianavalona (1999) and Rajemisa-Raolison (1969) for discussion of Malagasy determiners. Cole, Hermon, and Aman (to appear) argues that some Malay w/i-questions are clefts and that the subject is an unexceptional headless relative. Paul (2000) also notes the problem with Malagasy no and stipulates that no is restricted to equative clauses.

References Abney, Steven 1987 The English noun phrase in its sentential aspect. Ph.D. dissertation, Department of Linguistics and Philosophy, MIT. Aldridge, Edith 2002 Nominalization and wh-movement in Seediq and Tagalog. Language and Linguistics 3: 393—427. Bowers, John 1993 The syntax of predication. Linguistic Inquiry 24: 591-656.

The Cleft Structure of Malagasy Wh-Questions

229

Boskovic, Zeljko 2002 On multiple wh-fronting. Linguistic Inquiry 33: 351-383. Carnie, Andrew, and Eithne Guilfoyle (eds.) 2000 The Syntax of Verb Initial Languages. Oxford: OUP. Chang, Melody Ya-yin 2000 On Tsou w/z-questions: Movement or in situ? Language and Linguistics 1: 1-18. Cheng, Lisa Lai-Shen 1997 On the Typology of Wh-Questions. New York: Garland. Chomsky, Noam 1995 The Minimalist Program. Cambridge, MA: MIT Press. Chung, Sandra 2005 What fronts? On the VP raising account of verb-initial order. In Verb First: Papers on the Syntax of Verb Initial Languages, Andrew Carnie, Sheila Dooley Collberg, and Heidi Harley (eds.), 9-29. Amsterdam: John Benjamins. Cole, Peter, Gabrielle Hermon, and Norhaida Aman to app. Clefted questions in Malay. In Malay/Indonesian Linguistics, David Gil, and James Collins (eds.). London: Curzon Press. Dahl, Otto Christian 1986 Focus in Malagasy and Proto-Austronesian. In FOCAL 1, Paul Geraghty, Lois Carrington, and Stephen A. Wurm (eds.), 2 1 ^ 5 . Canberra: Pacific Linguistics. Flegg, Jill Heather 2003 Topics and clitic left dislocation in Malagasy. Paper presented at the 10th meeting of the Austronesian Formal Linguistics Association, University of Hawaii. Georgopoulos, Carol 1991 Syntactic Variables: Resumptive Pronouns and A' Binding in Palauan. Dordrecht: Kluwer. Guilfoyle, Eithne, Henrietta Hung, and Lisa Travis 1992 Spec of IP and Spec of VP: Two subjects in Austronesian languages. Natural Language and Linguistic Theory 10: 375—414. Keenan, Edward L. 1976 Remarkable subjects in Malagasy. In Subject and Topic, Charles N. Li (ed.), 247-301. New York: Academic Press. 1995 Predicate-argument structure in Malagasy. In Grammatical Relations: Theoretical Approaches to Empirical Questions, Clifford S. Burgess, Katarzyna Dziwirek, and Donna Gerdts (eds.), 171-216. Stanford: CSLI Publications.

230

Eric Potsdam

Keenan, Edward L., and Bernard Comrie 1977 Noun Phrase accessibility and Universal Grammar. Linguistic Inquiry 8: 63-100. Kratzer, Angelika 1996 Severing the external argument from its verb. In Phrase Structure and the Lexicon, Johan Rooryck, and Laurie Zaring (eds.), 109-137. Dordrecht: Kluwer. MacLaughlin, Dawn 1995 Wh-movement in Malagasy: An extraction asymmetry. In Theoretical Approaches to African Linguistics, Akinbiyi Akinlabi (ed.), 117-128. Trenton, NJ: Africa World Press. Massam, Diane 2000 VSO and VOS: Aspects of Niuean word order. In Andrew Carnie, and Eithne Guilfoyle (eds.), 97-116. Massam, Diane, and Caroline Smallwood 1997 Essential features of predication in English and Niuean. In Proceedings of the 27th North East Linguistic Society, Kiyomi Kusumoto (ed.), 263-272. Amherst, MA: GLSA Publications. Paul, Ileana 2000 Malagasy clause structure. Ph.D. dissertation, Department of Linguistics, McGill University. 2001 a Concealed pseudo-clefts. Lingua 111: 707-727. 2001b Ve as a second-position clitic. Oceanic Linguistics 40: 135-142. 2002 On extraction asymmetries. In MIT Working Papers in Linguistics 44: The Proceedings of the Eighth Austronesian Formal Linguistics Association, Andrea Rackowski, and Norvin Richards (eds.), 211— 224. Cambridge, MA: MIT Working Papers in Linguistics. 2003a Multiple topics: Evidence from Malagasy. In Cornell Working Papers in Linguistics 19: Proceedings of the Ninth Annual Meeting of the Austronesian Formal Linguistics Association (AFLA 9), Anastasia Riehl, and Thess Savella (eds.), 137-148. Ithaca, NY: Cornell Linguistic Club Publications. 2003b On the lack of wh-movement in Malagasy. Paper presented at the 10th meeting of the Austronesian Formal Linguistics Association, University Of Hawaii. Pearson, Matthew 1996 Domain phrases and topic arguments in Malagasy existentials. In UCLA Occasional Papers in Linguistics 17: The Structure of Malagasy, Matthew Pearson, and Ileana Paul (eds.), 113-141. Los Angeles: UCLA Department of Linguistics.

The Cleft Structure of Malagasy Wh-Questions 1998

231

Rightward object shift and the syntax of adverbs. In UCLA Occasional Papers in Linguistics 20: The Structure of Malagasy II, Ileana Paul (ed.), 34-49. Los Angeles: UCLA Dept. of Linguistics. 2001 The clause structure of Malagasy: A Minimalist approach. Ph.D. dissertation, Department of Linguistics, UCLA. 2005 The Malagasy subject/topic as an A'-element. Natural Language and Linguistic Theory 23: 381-457. Pensalfini, Robert 1995 Malagasy phrase structure and the LCA. In MIT Working Papers in Linguistics 27: Papers on Minimalist Syntax, Rob Pensalfini, and Hiroyuki Ura (eds.), 209-221. Cambridge, MA: MIT Working Papers in Linguistics. Pesetsky, David 1987 Wh-in-situ: Movement and unselective binding. In The Representation of (injdefiniteness, Eric Reuland, and Alice ter Meulen (eds.), 98-129. Cambridge, MA: MIT Press. 2000 Phrasal Movement and Its Kin. Cambridge, MA: MIT Press. Postal, Paul M. 1970 On so-called pronouns in English. In Readings in English Transformational Grammar, Roderick A. Jacobs, and Peter S. Rosenbaum (eds.), 201-224. Waltham, MA: Ginn and Company. Potsdam, Eric 2003 Evidence for semantic identity under ellipsis from Malagasy sluicing. In NELS 33: Proceedings of the North East Linguistic Society, Makoto Kadowaki, and Shigeto Kawahara (eds.), 285-302. Amherst, MA: GLSA Publications. 2004 Wh-questions in Malagasy. In Ζ AS Working Papers in Linguistics 34: Proceedings ofAFLA 11, Paul Law (ed.), 244-258. Berlin: ZAS. Rabenilaina, Roger-Bruno 1998 Voice and diathesis in Malagasy: An overview. In UCLA Occasional Papers in Linguistics 20: The Structure of Malagasy II, Ileana Paul (ed.), 2-10. Los Angeles: UCLA Department of Linguistics. Rackowski, Andrea 1998 Malagasy adverbs. In UCLA Occasional Papers in Linguistics 20: The Structure of Malagasy II, Ileana Paul (ed.), 21-33. Los Angeles: UCLA Department of Linguistics. Rackowski, Andrea, and Lisa Travis 2000 V-initial languages: X or XP movement and adverbial placement. In Andrew Carnie, and Eithne Guilfoyle (eds.), 117-141. Rahajarizafy, Antoine 1960 Essai sur la grammaire malgache. Antananarivo: Imprimerie Catholique.

232

Eric Potsdam

Rajemisa-Raolison, Regis 1969 Grammaire Malgache. Fianarantsoa: Centre de Formation Pedagogique. Richards, Norvin 1998 Syntax versus semantics in Tagalog wh-extraction. In UCLA Occasional Papers in Linguistics 21, Matthew Pearson (ed.), 259275. Los Angeles: UCLA Department of Linguistics. 2002 Movement in Language. Oxford: Oxford University Press. Rudin, Catherine 1988 On multiple-wh questions and multiple wh-fronting. Natural Language and Linguistic Theory 6: 445-501. Säbel, Joachim 2001 Deriving multiple head and phrasal movement: The cluster hypothesis. Linguistic Inquiry 32: 532-547. 2002 Wh-questions and extraction asymmetries in Malagasy. In MIT Working Papers in Linguistics 44: The Proceedings of the Eighth Austronesian Formal Linguistics Association, Andrea Rackowski, and Norvin Richards (eds.), 309-323. Cambridge, MA: MIT Working Papers in Linguistics. 2003 Malagasy as an optional multiple wh-fronting language. In Multiple Wh-Fronting, Cedric Boeckx, and Kleanthes Grohmann (eds.), 229254. Amsterdam: John Benjamins. Travis, Lisa 2004 VP-, D°-movement languages: Malagasy in Wonderland. Paper presented at the 11th meeting of the Austronesian Formal Linguistics Association, ZAS Berlin. Zribi-Hertz, Anne, and Liliane Mbolatianavalona 1999 Towards a modular theory of linguistic deficiency: Evidence from Malagasy pronouns. Natural Language and Linguistic Theory 17: 161-218.

Three Systems of Remnant Movement II and Extraction from Specifier Position* Craig Thiersch

Kayne's (1994) Antisymmetry Hypothesis and Linear Correspondence Axiom have inspired much fruitful research attempting to adjust syntactic analyses to those permitted by Kayne's restrictive system. Many of these analyses make extensive use of so-called Remnant Movement, raising issues of computational complexity. In this article we compare three such analyses: for Malagasy, for Hungarian, and for German, and examine how they deal with the problems of creating the remnant and avoiding potentially problematic extraction from within specifier positions. We also suggest, tentatively, how the first two cases might be adapted to a more uniform approach, for example as in (Chomsky 2001).

1.

Introduction

In the wake of Kayne's Antisymmetry Hypothesis and Linear Correspondence Axiom (LCA), there has been much fruitful research attempting to adjust syntactic analyses to those permitted by Kayne's restrictive system. In so doing, analyses which at first seem counter-intuitive have been claimed to provide solutions to old problems. Three cases in point are the analyses of Malagasy involving extensive Remnant Movement [henceforth RM] described in (Rackowski and Travis 2000) and (Pearson 2001); the analysis of Hungarian and Dutch verbal clusters in (Koopman and Szabolcsi 2000) [henceforth R&T, Pearson, and K&Sz] 1 ; and the analysis of German in (Müller 2004). The goal of this article is to compare the three systems formally, without making a judgement as to the empirical adequacy of the systems, although we do note briefly some empirical puzzles in Malagasy which are claimed to have been solved by the RM analyses by way of motivation. What we would like to show is that all three systems, although quite different in detail and the nature of their background assumptions, encounter the same "freezing problem": namely, preventing constituents which later need to be moved elsewhere from

234

Craig Thiersch

ending up in (specifier) islands. We examine the details of how these systems have coped with this, and compare them, without however offering more than a tentative assessment as to which is the better approach. Given the perspective of this book, it is striking that three language particular puzzles in Malagasy are claimed to have been solved by analyses involving RM. The solutions are not, however, unproblematical, both technically and empirically, and hence we examine them in somewhat more detail than the other two cases (Hungarian and German). The original motivation (in part) for this author's examining K&Sz and subsequently R&T was that the extensive use of iterated RM could potentially increase computational complexity of the languages generable in Stabler's "Minimalist Grammar" formalism (MG). In addition, requiring specifiers to be islands reduces computational complexity under certain conditions: although MGs with a Shortest Move Condition (SMC) but without restrictions on extraction from specifiers (Specifier Island Conditions, or SPIC) have the complexity of Linear Context-Free Rewriting Systems (LCFRS), it has been noted that adding versions of the SPIC reduces the complexity to a proper-subset, namely those languages generated by LCFRS 12s; cf. (Michaelis 2005) and the references cited therein. 2 Both R&T and K&Sz make extensive use of RM; R&T allow extraction from complex specifiers, while K&Sz do not. Quite aside from the issues of complexity, we nevertheless feel that there is enough intrinsic linguistic interest in RM analyses and in trying to limit extraction possibilities to pursue the comparison of these three systems. Cf. Thiersch (in prep.b) for some additional discussion. This is intuitively plausible, as we know extraction from deeply embedded positions within specifiers is usually ungrammatical 3 : (1)

a. b.

Whoj did Fred say that Susan hoped for Sam to kiss e,· * Whoi did Fred say that for Sam to kiss e,· would create a scandal.

As noted above, extraction from complex specifiers is explicitly forbidden in K&Sz and the prohibition forms a crucial part of their analysis. It appears however to be necessary for R&T to account for the position of nominal objects. A closer look shows that all three systems have a variant of this difficulty, and in this article we compare the three systems in the hope of shedding some light on how three rather different systems involving RM have coped with it. As a terminological matter we distinguish between two types of RM, I and II, following (Müller 2002) and (Thiersch 2002):

Three Systems of Remnant Movement 235 (2)

RM I: a. Erschossen hat er Bin Ladin schon gestern shot has he B. L. already yesterday 'He already shot Bin Ladin yesterday' b. Analysis: [ v p e, erschossen]j hat er [Bin Lading schon gestern ej

(3)

RM II: a. John reads no novels b. Analysis: John [reads e,]j [ Fp [no novels],· ej ]

Although formally identical, in that they involve displacing a constituent a from which some element β has been extracted so that the antecedent, β , no longer C-commands its trace in a , they tend to have different characteristics. For example, RM I consists of two operations which both exist as independent constructions in the language — in (2), "scrambling" and Vorfeldtopicalization 4 — and RM I cannot be iterated indefinitely. 5 RM II, on the other hand, generally can apply repeatedly, neither operation in (3b) occurs independently in English, and its effects are often invisible except for semantic effects: e.g. the two readings of (4a); cf. (Kayne 1998).6 (4)

a. They forced him to marry no one b. Narrow scope reading: i.

They forced him [[to marry e(\j [no one],· ej]

ii. CAUSE (they, Ρ): Ρ =

s.t. MARRY(he, x)

c. Wide scope reading: i. They [forced him to marry ei\j [no one]i ej ii. ->3x CAUSE (they, P(x)): P(x) = MARRY(he, x) In the first reading they forced him to remain unmarried, in the second there is no particular woman he was forced to marry. In the first case the negative DP raises to Spec,NegP in the subordinate clause, and the remnant VP is fronted; in the second, it raises to Spec,NegP in the matrix clause, and raising the matrix VP pied-pipes the subordinate clause. Before we examine a potential difficulty regarding the position of nominal DPs, we briefly sketch some relevant aspects of Malagasy syntax and some sample empirical problems the RM analysis apparently solves. For a more detailed account of Malagasy syntax the reader is referred to (Pearson 2001), (Paul 1999), and the references therein.

236

Craig Thiersch

2. Rackowski & Travis: Malagasy 2.1. The basics: The Malagasy voicing system Roughly, Malagasy is a verb initial language (although there are constructions in which certain "fronted" constituents may precede the verb). It is sometimes cited as being typologically VOS, but this oversimplifies what is really going on. Basically the arguments line up following the verb, and depending on the voice of the verb, one of them is "promoted" to the rightmost position. Schematically, (5)

yvoice2 arg! (arg 2 ) arg3 arg2 ι t

That is, if the verb is in "second" voice, the second argument appears to the right.7 For example [Pearson: Chap. 2, (65)], (6)

a. Mamono akoho amin 'ny antsy ny mpamboly NOMP.kill chicken with'DET knife DET farmer 'The farmer kills chickens with the knife' b. Vonoin'ny mpamboly amin'ny antsy ny akoho ACCP.kill'DET farmer with'DET knife DET chicken 'The chickens are killed by the farmer with the knife' c. Amonoan'ny mpamboly akoho ny antsy CRCP.kill'DET farmer chicken DET knife 'The knife is being used by the farmer to kill chickens'

Indeed, the voicing system has sometimes been compared with Indo-European passive, and the right-most argument called the subject, although the construction has rather different properties, as can be seen in the examples above: the non-promoted semantic subject does not become a chomeur in an optional PP, but remains, presumably in situ* Traditionally, it was often assumed that the Spec,DP was "right-merged". That is, the external argument moved up to a right-hand specifier position, roughly as in (7), on the next page. 9 Pearson and R&T argue that this is not the case, but that the basic structure is consistently Spec-Head-Comp, as in Kayne's antisymmetry proposal, and that the external argument has moved up to the left, and a constituent containing the verb with the remaining arguments (and some adjuncts) has then moved leftward around the externalized argument. Schematically, the sentence in (6b) would have a structure roughly like (8a), on the next page. 10

Three Systems of Remnant Movement

(7)

IP

V (8)

237

DP,

a. [Vonoin ny mpamboly e,· amin'ny antsy]j [[ny akoho\i ej] ACCP.kill'DET farmer with'DET knife DET chicken b. Derivation:

This is grossly over-simplified and ignores the functional categories as well as the considerable difference in hypothesized nodes between, say, Pearson and

238

Craig Thiersch

R&T, but captures the spirit of the analyses under consideration. Pearson in particular motivates a number of functional categories and hence movements, giving a more complete and differentiated account of the facts. We briefly summarize in the following sections how this type of analysis accounts for some empirical observations, suggesting that the RM approach is on the right track, before turning to the issue of correct placement of the DP arguments.

2.2. 2.2.1.

Some empirical problems addressed by the RM analysis Focus "extraction"

Various focus and topic particles can allow a constituent to be "extracted" to the left, but as has been long noted in the literature, this can in general only be the constituent which would have been "externalized" to the rightmost position according to the voice of the verb. 11 E.g., [Pearson: Chap. 2 (36)]: (9)

a.

* Ny

akoho

no

namono

tamin 'ny

antsy ny

DET c h i c k e n FOC PST.NOMP.kill PST.with'DET k n i f e DET

mpamboly farmer 'It's the chicken that the farmer killed with the knife' b.

Ny

akoho

no

novonoinVry

mpamboly

DET chicken FOC PST.ACCP.kill'DET f a r m e r

tamin'ny PST.with'DET

antsy knife c.

* Ny akoho no namonoan 'ny mpamboly ny antsy DET chicken FOC PST.CRCP.kill'DET farmer DET knife

At first glance, this is surprising under an analysis with a right-merged specifier as in (7), as one generally expects objects to be more easily extractable than subjects (but cf. endnote 9). In a structure like (8a), however, the external object is on the main rightward projection line with no barrier-like categories in between, whereas the other arguments are embedded in the fronted TP/VP, i.e. a complex specifier and presumed by R&T to be an island (a point to which we return below). 12 ' 13

Three Systems of Remnant Movement 239 2.2.2.

Placement of discourse markers

Pearson, following (Paul 2001), discusses another phenomenon (noted in passing in R&T), namely that various discourse markers come in peculiar places and don't seem to have any particular pattern. For example, the yes/no question marker ve appears in the penultimate position (left of the externalized argument) in ordinary sentences, but in the second position (between the fronted DP and the particle no) in focus sentences [Pearson: Chap. 2, (14a) & Chap. 4, (102a)]: (10)

a. Vonoin'ny

mpamboly amin'ny

antsy ve ny

akoho?

ACCP.kill'DET farmer with'DET knife QU DET chicken 'The chickens, is the farmer killing (them) with the knife?' b. I Bakoly ve no manapaka bozaka? DET Bakoly QU FOC NOMP.cut grass 'Is it Bakoly who is cutting the grass?' Pearson and Paul point out that we only need to assume that ve is a secondposition clitic which raises the specifier of its complement to its own specifier in order to account for these facts; in (10a) the whole TP/IP is raised, leaving the externalized argument to the right of ve. In (10b), however, only the focused DP is raised (see the structure in endnote 12).

2.2.3.

Adverb order

Finally we come to the phenomenon which forms the centerpiece of the R&T article, namely the order of adverbials. The basic assumption is the universal order of adverbials related to a hypothesized universal hierarchy of functional projections proposed in (Cinque 1999). Language specific deviations were to be accounted for, as usual, by language particular properties (e.g., lexical). As previously discussed by (Rackowski 1998) with a similar analysis, the order of adverbials in Malagasy deviates from the proposed universal order in a surprising way; aside from some minor peculiarities, the pre-verbal adverbs mimic the Cinque order, whereas the post-verbal adverbs are in the mirrorimage order 14 [cf. R&T's (10)-(11)]:

240

Craig Thiersch

(11)

a. Cinque's order: 1 2 3 4 5 6 (speech act) > Generally > Neg > Already > Still > (At.all) > 10 9 7 8 Anymore > Always > Completely > Well b. Malagasy order: ? 2 3 4 5 (3) V Na(dia) > Matetika > Tsy > Efa > Mbola > Tsy > (Verb) > Even > generally > Neg > Already > Still > Neg > Verb > ? 10 9 8 7 6 Tsara > Tanteraka > Foana > Intsony > Mihitsy > Aza > Well > Completely > Always > Anymore > at.all > though > 1 Speech Act

This immediately suggests a "roll-up" operation like the operation proposed in K&Sz to reverse the order of Hungarian verbs. The language particular stipulation R&T need to make is that the upper adverbs (up to and including Neg tsy) are generated in Spec of their functional projections; the lower adverbs (6-10) generated as heads of their projections. 15 Their suggested derivation works as follows: (12)

a. Repeated movement to Spec,Adv^P (roll-up) reverses the order, b. a blocking trigger (e.g., an element in Spec,NegP) stops the roll-up, and c. the external argument (subject) is extracted and a high resulting constituent (e.g., TP) is moved to a projection above it.

The tree in (13), on the facing page, illustrates (12a-b), and the steps (12c) were already illustrated in (8b) for (8a), a sentence without adverbs [cf. R&T (12)—(13), p. 122], Note that without DP arguments, the tree in (13b) does not involve RM; the traces are C-commanded by their antecedents. It is the extraction of an element such as a DP which induces RM. (Similarly in K&Sz's analysis; see below.)

Three Systems of Remnant Movement (13)

a. Tsy manasa tsara tanteraka foana intsony ny lamba mihitsy not wash well completely always anymore the clothes at.all Rakoto Rakoto b. NegP

foanaPj

tanterakaPfc

tsaraP/

VP

tsara

2.2.4.

241

.

.

.

tanteraka

.

foana

intsony

mihitsy

e,·

e,·

e*

e;

e,

V°-fronting vs. "VP-fronting"

We note in passing one other phenomenon (relevant for the discussion of (Müller 2004) below) which the RM analysis of Malagasy addresses, namely why, according to R&T and Pearson, some languages (West Germanic), involve V°-fronting, and others (some Austronesian languages) involve VP (or rather XP) fronting. Pearson's tentative hypothesis is that the crucial distinction is in the morphology — suffixes versus prefixes. If we compare the appropriate stages in the derivation of an Icelandic sentence (14) and a Malagasy sentence (15), on pages 242-243, we see that in the former case, the verb in T° can raise by head-movement, but in Malagasy the only category dominat-

242

Craig Thiersch

ing the T° and the verb is the entire TP [Pearson: Chap.4, (87)-(88)] 16 . While this conclusion is intriguing but tentative, it contradicts the analysis of (Müller 2004), as well as K&Sz, as it allows head-movement. Although one can quibble with many of the details and assumptions, the capturing of these phenomena (other arguments are given in Pearson and R&T) seems sufficient motivation that the RM approach is on the right track, and we turn to the details of DP placement. (14)

a. Bokina

keypti

Jon

book.the bought John 'John bought the book.' 17 TP

b.

Asp e °

ν

Aspe 0

T°

DP

Asp/

eAspe

?

keypDP,

vP

boV eDPt ev

V eDPi

Three Systems of Remnant Movement (15)

243

a. Nohanin'ny gidro ny voankazo PST.AccP.eat'DET lemur DET fruit 'The lemur ate the fruit' PivP

b.

DP,

ny voankazo

Piv' Piv 0

T°

TP

EP

Asp e hanin ny gidro

2.2.5.

.. eAsPeo .. eDPj .. eDPi ..

Summary

In the foregoing, we have cited some empirical puzzles allegedly solved by analyzing Malagasy as a language which derives its surface order by fronting XP (remnant) constituents rather than movement of the "external" argument to the right (and V° movement), by way of motivation for looking at the technical details of the analysis more closely. As noted at the outset, we refrain for the purposes of this article from criticizing the empirical coverage of the approaches, but it should nevertheless be re-emphasized that some of the above claims are controversial and fraught with problems, especially the claim about islandhood induced by the fronted constituent (not limited to R&T, as one reviewer notes, but also made by Rob Pensalfini and Eric Potsdam). As noted in the endnotes, the apparent freedom of extraction of adverbs may not be a counterexample if one can maintain the two classes of adverbs as suggested by Paul (2002). Such a distinction is found in the Northern Salish (YOS) languages, which distinguish sharply in relativization and/or clefting between core obliques (oblique objects and instruments)

244

Craig Thiersch

and, say, locatives. Cf. (Kroeber 1997), sections 4 and 5 for Thompson River Salish. On this note, the same reviewer points out that one might expect other VOS languages to exhibit a similar islandhood of VP internal elements as Malagasy. While examining this in detail is of course beyond the scope of this article, a cursory look at Tzotzil and Salish seems to indicate a greater freedom of extraction. For example, what appears to be "preposition stranding" in Tzotzil [(Haviland 1980), my gloss -ct]: (16)

Jay vo' vinik chabat a-chi?uk? how.many CLF man lMP.2.go 2-with 'How many men are you going with?'

Salish strikingly forms questions and relatives with a cleft-like construction similar to the no construction in Malagasy, but with the righthand element a truly nominalized clause including a determiner [(Kroeber 1997), (23a)] 19 : (17)

Question cleft: swet [NP k who

ART

[s wik-t-xw

]]

see-TRZ-2S-TS

Assumed structure (ct): (COP) who [the you-saw-(him/her)] « 'Who is it that you saw?' = 'Who did you see?'

?

However, as one can see by the preposition-stranding case, comparison is difficult without further analysis [his (45c)]: (18)

st'e?[ndk q'wi-t-'es u-s e smiye what in ART cook-TRZ-3.TS CNJ ART meat 'What did she cook the meat in?'

(ePP20)

]

Since these languages don't have the voicing system of Malagasy, it's not clear the VOS order is derived in the same way, and comparison must wait for a more detailed analysis. Since they have suffixal rather than prefixal verbs, it isn't even clear that they are candidates for a VP-fronting analysis (see comments in section 2.2.4). Palauan, being prefixal, might yield interesting data relevant to this question, but time constraints prevent including it here. This having been said, we turn to the technical details of DP placement.

2.3.

The order of DP objects

The problem is illustrated by two of R&T's examples. Definite direct objects may "optionally appear among or after postverbal adverbials." [p. 125] This

Three Systems of Remnant Movement 245 is illustrated in (19) [R&T: (21a/b)] 21 : (19)

a. Tsy manasa foana ny lamba mihitsy Rakoto. n e g pres.AT.wash always DET clothes at.all R. 'Rakoto does not always wash the clothes at all' b. Tsy manasa intsony mihitsy ny lamba Rakoto. NEG pres.AT.wash no.longer at.all DET clothes R. 'Rakoto does not wash the clothes at all any more'

The adverbs mihitsy and foanalintsony are respectively Adv6 and Adv8 in the Cinque hierarchy, here appearing in 8 > 6 order due to the (remnant) movement of the verb phrase. How might the derivation procede? The first merge would be with the direct object, yielding (20)

[Vp manasa [DP ny lamba]]

This is merged with the lower adverb, foana, and the VP moves to the specifier of foana, yielding (21)

[Adv8P (vp, m a n a s a [dp ny lamba]] foana e,· ]

If we merge this structure again with the higher adverb, mihitsy, and move the Adv8P out, we get (22)

[Adv6P [Adv8P; [vp,· m a n a s a [dp ny lamba]] foanaAdv8 ei ] mihitsym\6 ej ]

This generates neither (19a) or (19b), but is grammatical. 22 Alternatively, suppose we had immediately extracted the DP to the specifier of AgrOP yielding (23)

[AgrOP [dp,· ny lamba] [AgrQ, AgrO 0 [ v p manasa e,· ]] ]]

Merging with Adv6P and raising the remnant VP gives (24a), shown as a tree in (24b) 23 ; and after merging with Adv6, mihitsy, and raising Adv8P, we get the structure for (19a), shown in (25a-b).

246

Craig Thiersch

(24)

a. b.

[AC1V8P [ V P J

manasa

e oana Adv8P i ] if Ad\>&

manasa

e,·

Adv8

[AGROP

(DP;

e

j ]]]

AgrOP

foana DP,· (25)

e,

a. [ A d v 6 P [ A d v 8 P t [ VP . manasa et ] [ foanaAdv8 ej ]]] mihitsy Aäw6 ek ]

b.

[AgrOP [DP. ny lamba]

Adv6P Adv6' Adv8'

manasa

e,·

Adv8

Adv6 ek I mihitsy

AgrOP

foana DP,

&j

What about (19b)? Here there as several possibilities, each with attendant problems. One possibility is to extract the DP immediately, as in the derivation for (19a). But then we would need to extract it again to get (19b). However, we have already "used up" AgrOP and checked the appropriate feature. This means we would have to postulate another functional category. Since there is no limit in principle to the number of adverbs,24 this has the disadvantage of needing to postulate a potentially unlimited number of functional categories (and features) just to effect the extraction. (This is in fact the solution adopted by K&Sz; see below.) The alternative, adopted implicitly by R&T and explicitly by Rackowski (1998), is to take the DP along, as in (22), and wait until the appropriate moment in the derivation, then merge once with AgrOP and extract the DP inside Adv6P to Spec,AgrOP ("·"):

Three Systems of Remnant Movement

(26)

247

AgrOP

Adv8P

AgrO

Adv8'

Adv6P VP

A

Adv6'

V DP Adv6

Adv8

&Adv6P

eVp

Example (26), however, presents us with a potential problem: we have here extraction from a complex specifier, and given a long series of adverbs, the DP could be indefinitely deeply embedded. Suppose we permit unlimited extraction from complex specifiers. We are then in a quandary with respect to the focus extraction: the explanation for the inaccessibility of the DPs for extraction, except for the externalized one, was that they were in a specifier island. In the article R&T in fact do invoke the islandhood of the fronted XP: "The predicate-fronting analysis predicts that, since the subject is the only argument not contained in some kind of island, it should behave differently from the other arguments." [p. 124] How could we differentiate between the two cases? 25 A potential solution comes from the former version of this analysis, namely the version in (Rackowski 1998); here she adopts a Kaynean phrase structure which crucially differentiates between categories and segments. Under this approach the specifier is an adjunct, and not dominated by the category dominating the head of the projection.26 Under these assumptions her structure for (26) was as in (27), on the next page. In (27) the only category which dominates the DP is VP; the rest of the nodes are only segments. Rackowski claims that in terms of a Barriers-like theory, there is nothing to block the properly governed DP from being extracted.

248

Craig Thiersch

(27)

AgrOP

AgrO

Adv8P Adv8P

Adv6P VP

Adv6P Adv8

A

V DP Adv6

&Adv6P

eVp

However, this ignores considerable complications, as Chomsky's (1986) Barriers theory was formulated in terms of a phrase structure without the many embedded functional projections between the VP and IP. In particular, for the VP to be properly L-marked, they would have to be "transparent", and once L-marked, the VP would have to retain its own transparency for extraction under movement, a not entirely obvious desideratum,27 as (28) shows: (28)

a. b. c. d.

Fred said that Sam was beaten by a policeman. Which policeman was Sam beaten by? Fred said that beaten by a policeman Sam certainly was. * Which policeman did Fred say that beaten by Sam certainly was.

Furthermore, the status of TP (IP) is unclear. Chomsky (1986) launches various conceptions of the characteristics of the IP/I' system in the course of Barriers, but seems to conclude that although the IP is a barrier for ordinary government (by inheritance) it is transparent for antecedent government. Hence it is not clear what blocks direct movement from the (moved) TP to Spec,CP, allowed by antecedent government.28 Ignoring these difficulties for the moment, we can ask if this explanation is nevertheless sufficient to block extraction in, say, the focus or topic constructions. This would seem to be the case. Let us look at some steps of a "complete" derivation, much simplified. (Trees on the following two pages.) Note the position of DP in the Spec under VP (29a) « (27) from which it will move. In (29b) it has moved to Spec,AgrOP, and APy has moved to Spec,APz. If TP subsequently moves as in (29c), the DP is again in a specifier position,

Three Systems of Remnant Movement

249

but now dominated by the category AP Z and all categories above AP Z , freezing it in place (na (dia)...aza translates roughly as 'even...though') 2 9 In summary: while the idea that a specifier consisting of a single category otherwise dominated only by segments should be more transparent for extraction than one consisting of multiple embedded categories seems intuitively appealing, it faces considerable obstacles to making it work. Although it appears to be a possible solution to the dilemma of transparency versus islandhood, it depends on the segment/category distinction, which does not seem to be invoked in later work by Travis (this volume), which uses the traditional X-bar theory. It also depends upon reformulating Barriers-style extraction restrictions (as discussed above) 30 , and ignores the potential challenge of adjunct extraction discussed in endnote 13. We also note that this approach is explicitly rejected by Pearson, who makes use of multiple specifiers in his analysis, redefining C-command derivationally along the lines of (Epstein et al. 1998) to allow this. He links the extraction constraint to feature conflict similarly to the exclusion of simultaneous W H and Topic movement in Germanic; cf. discussion at the end of Pearson's 3.4.1, p.133. (29)

AP y

a.

b.

VP

DPs

AP*

VP

V

AkX

DPo

APy

APy

APJC

Ay

eVP

AP Z

e APx

Az

APζ AgrOP

DPo AgrOP

AgrO

e APy

250

Craig Thiersch c.

naP na[dia]

aP

Subj tsy

ZTP

NegP

Neg°

APZ

AP, Az

APZ AgrOP DPo AgrOP AgrO

3.

eApy

The K&Sz system for Hungarian

While it is beyond the scope of this article to discuss the intricate system developed in K&Sz for Hungarian and Dutch verb clusters (cf. Thiersch in prep, a), we would like to look briefly at their solution to the above dilemma involving RM, with an eye to a comparison of the issues involved. The data treated in K&Sz (verbal clusters) lack adverbial constituents, and the roll-up is used to generate the proper order of verbs (and the verbal particle) in the verbal clusters in Hungarian and Dutch. K&Sz are, however, faced with the same difficulty as R&T, namely the DP arguments associated with verbs have to be "gotten out" of the remnant constituents being rolled up. Indeed, as in

Three Systems of Remnant Movement

251

R&T, most of the remnant character of their roll-up comes from repeated evacuation of an argument DP from a VP. First of all, K&Sz deal with extraction from complex specifiers by simply forbidding it by fiat.31 Glossing over many details (cf. Thiersch (in prep.a)) the core of their system is as follows: although with five verbs the Hungarian verb cluster could theoretically have 5!=120 orders, in fact it has only four. For example, with one (finite) true auxiliary, two semi-auxiliaries, a main verb and a VM 3 2 and representing them by the numerals 1-5, repectively, we have (30)

"Inversion" cases:

(31)

"Clitic climbing" case:

a. 1-2-3-5-4

b. 1-2-5-4-3

c. 1-5-4-3-2

5-1-2-3-4

Following Kayne, the verbs are base-generated in the "English" (1-2-3-4-5) order. The first pair (i.e., 4-5: V+VM) must be inverted. Each inverted constituent formed can subsequently be inverted with the next verb to the left, e.g., 3 with 5-4 producing [5-4]-3. If the inversion stops, it cannot restart, e.g. if we have not inverted with 3, we cannot invert anything with 2. Alternatively, the VM may raise (31), preserving the "English" order. 33 Since K&Sz assume (i) only XP movement is involved (no head movement), (ii) obligatory feature checking and (iii) no LF movement or counter-cyclic movement, the surface "English" order (30a) must also be derived. That is, if [5-4-3] has moved above 2 to generate (30c) from (30b), then the same movement must have occurred in (30a), if all feature-checking is obligatory. 34 How do K&Sz prevent this from being circular, and how do they get alternative orders with obligatory feature-checking? The answer is optional piedpiping. K&Sz assume that each V heads its own extended clause (CP). Without going into the details of the derivation, we know pied-piping as a surface phenomenon is optional, so that if we encounter a structure like (32) [K&Sz: (72)=(79)] where the VP4" needs to move to Spec,CP (·) to check a feature, it cannot do so (due to K&Sz's prohibition of extraction from Spec), but we can choose either InfiP + or PredP, pied-piping the rest. 35 This will eventually yield divergent derivations, giving the different orders.

252

Craig Thiersch

(32)

(CP)

(·) PredP

InflP+ VP+

LP [dp]

Δ

Clearly this will run into difficulties if a projection contains a DP, just as in R&T's system, as it will quickly become deeply embedded in a specifier position and trapped inside the verbal complex, contrary to fact. Due to K&Sz's prohibition on extraction from Spec, they do not have the escape-hatch of R&T regardless of how we interpret their X-bar structure. Their solution is twofold: (33)

a. K&Sz forbid "moving two constituents at the same time", [their condition (27), p.43] by which they mean that in a structure like [ x p . . . X° . . . YP . . . ], XP counts as two constituents under "certain circumstances"; inter alia, when XP and YP are different projections, e.g., a VP containing a (phonetically realized!) DP, but not when XP is the extended projection of YP, in the sense of (Koopman 1996), e.g., V P + and VP; see endnote 35. (For brevity, I call this the "1=2" condition.) b. K&Sz allow an arbitrary number of functional "pushing categories" (LPs, or licensing phrases) similar to AgrO, but with vacuous content, which they refer to as stacking positions. Cf. pp.43-44.

Hence after leaving the V P + to check its Case in the first LP(dp) (=AgrOP), the DP must leave subsequent XPs at each step, due to (33a), and can leave due to the availability of pseudo LP(dp)s (33b). (K&Sz note that this is not an entirely satisfactory solution.) While we have not done justice to the complexity and subtlety of their system, and although K&Sz are able to generate the correct orders in Hungarian and parameterize the system for Dutch, there are numerous problems, some of which they themselves note, and some they don't — e.g., many multiple

Three Systems of Remnant Movement

253

derivations for the same string; cf. Thiersch (in prep.a). The question which interests us here is whether K&Sz could relax the prohibition on extraction from specifiers, and use the distinction between category and segment to allow the DP to extract. Using very simplified structures, we demonstrate that this seems unlikely. Suppose we disregard for the moment K&Sz's assumption that each V in the complex heads its own CP, and start with the base structure V Prt DP, e.g., 'take apart the radio-ACC' in (34a)36 and move the PrtP (in their terminology VM-phrase) to the Spec,VP4 as in (34b). We now have two basic options: if we assume (as opposed to K&Sz), that a V can select a VP complement (or the infinitival functional projection thereof), we get a structure like (34c). (34)

a.

VP 4

A A

V 4 PrtP 5

Prt 5 b.

DP VP 4

PrtP 5

VP 4

Prt 5 DP V 4 c.

e prt p 5 VP 3

VP 4 PrtP 5 Prt 5 DP

VP 3 VP 4

V4

V3

eVP4

ePrtP5

Here we see a structure akin to R&T's, and we observe that DP can leave later (it is not too deeply embedded) and the DP must leave at some time (because of feature-checking). (It is not clear what prevents it from leaving too early and becoming embedded in the verb-cluster, which is not permitted

254

Craig Thiersch

in Hungarian, as opposed to West Flemish and Swiss German, for example.) However, we are now faced with a different problem, namely what prevents the inverted cluster from unraveling? In K&Sz's original proposal, where the V selects an infinitival CP complement 37 the inverted cluster becomes too deeply embedded in a specifier and cannot be undone; here, however, after merging (34c) with higher verbs, nothing then prevents the category VP4 from moving up to the specifier of a higher V2 (·), for example, eventually generating the ungrammatical order 1+5+4+2+3 as shown in (35). (35)

VP 2

V2

VP 3 VP 4

PrtP 5 Prt 5 DP

VP3 VP 4

V4

V3

zvr4

ePrtP

Similarly, what stops the roll-up? Roll-up in R&T is stopped at a particular point, i.e., when the next specifier is filled. In K&Sz it was stopped at any point where the verbs were in separate CPs and hence couldn't be moved together due to their "1=2" condition (33a). Suppose we merge before moving VP 4 yielding (36), on next page. But in (36) there is nothing to stop roll-up of VP3, yielding 1+3+2+5+4, also an ungrammatical order. Suppose we do allow the V to select a full CP, as K&Sz do; then we have a structure like (37), next page. But here, if the CP moves up to · (Spec,VP 3 ) to create the inversion, the DP is now embedded too deeply in terms of categories to be extracted. Hence we are faced with a contradiction: we can't freeze verb complexes without freezing the DP. Hence we need the more complex structures hypothesized by K&Sz (as well as their background assumptions) to distinguish between the cases like the inverted verb-complexes, which are frozen in complex specifiers and the cases where we need to extract elements before freezing occurs.

Three Systems of Remnant Movement

(36)

VP2

PrtP5

VP4

Prt5 DP

(37)

V4

tpnP

YP3

PrtP5 Prt5

eDP

VP4 V4

ePr,P5

255

256 4.

Craig Thiersch (Müller 2004): vP fronting

We now turn to the system in (Müller 2004) [henceforth Müller]. His idea is that, contrary to appearances and much of the accepted literature, German does not front the verb by head-movement to second position in main clauses, followed by fronting of an XP constituent, but rather by fronting a remnant containing both to a clause initial position. 38 That is, instead of the traditional structure in (38a), he takes the remnant analysis of (Mahajan 2003) shown in (38b) a step further, moving the XP and the finite verb as a unit to Spec,CP as in (38c) [Müller: (l)-(3)]: (38)

a. [ CP Das Buch.2 [C/ hati+C? [TP Fritz ι [ vP e\ [vp ei gelesen ]