A I A-BAR PARTITION IN DENVATION HROAKl TADA B.A., The University of Tsukuba ( 1982) M.A., The University of Tsukuba ( 1B5) Submitred to the Department of Linguistics and Philosophy in Partial Fulfillment of the Requirements for the Degree of DOCTOR OF PHLOSOPHY at the MASSACHUSETTS INSTITUTE OFTECHNOLQGY The author hereby grants to MIT pen~ission to reproduce and to distribute copies of this thesis docu:nent in whole or in part. Signature of Author. &partrnent of Lingyistics and Philosophy - May 1!993 Certified by , - - Noarn ~hornsF Institute Professor Accepted by , ,-,_- ,,, - .. 1 Wayne O'Neil '~ead, Department of Linguistics and Philosophy MwAcHusElTs INSTW OF TECHNOLOGY A I A-BAR PARTiTlON IN DERIVATIOPI HIROAKI TADA Submitted to the Department of Linguistics and Philosophy on May 23, 1993, in Mia1 Fulfillment of the Requirements for the Degree of DOCTOR OF PHILOSOPHY in Linguistics This thesis investigate some relations of the problem of A/A-bar partition of movement types and the proper formulation of economy principles. In Chapter 2, I will explore two interrelated problems of scrambling. The first one, called the landing site problem, is why and how the apparently single operation scrambling behaves A-movement and A-bar-movement. The second one, called the optionality problem, is whether scrambling is arm optional operation, and if so, why it can v~olate the last resort principle. The answer to the first question leads to the derivational view of AIA- bar parti tion. The answer to the second question lead to a system of effect-based economy principles. In particular, the last resort principle is replaced by the compensation principle which requires the external effect of movement to be compensated for by its internal effect. In Chapter 3, I will discuss the problem of uniformity of chain-steps and suggest that the cnifonnity can be derived from stepwise application of compensation principle, rt ;3 discusses complex cases such as German partial w h-movement and Mayan focus antipassives in light of the compnsation principle and derivational view of A/A-bar- parbtion. In Chapter 4, I will discuss two phenomena related to the problem of multiple spifier in A A-bar systems: extrzction out of multiple specifier configuration and absorption of multiple wh-phrases. They are claimed to supprt Chengls( 1991) view that wh-movement is not driven by feature-checking as W-movement is. Thesis Supervisor: Noam Chomsky Title: Institute Professor It has been a special privilege to have Noarn Chomsky as my thesis supervisor. In particular, he has been generous enough to continue to supervise my thesis this year, even when he was on his sabbatical. Meetings with him were always inspiring and challenging, with full of innovative ideas and penetrating argumentation. I hope the following pages reflect even only a bit of his enormous influence on me and my work. His continuous moral support enabled me to go through this agonizing perid of thesis writing, for which I an1 forever thankful to him. My deep appreciation goes to my other thesis committee members, Jim figginbotham and Alec Marantz: I find myself extremely fortunate to have ken able to work with Jim, who is leaving MIT this year. His insights into formal aspects of language and linguistic representations always stimulated me into pursuing a higher level of formalism &n my thesis. Alec arrived at MIT well in time to raise my interests in ergativity. Discussions with him always reminded me of issues and aspects that I had left untouched. Readers will immdiately notice his influence on many pages. My life at MIT has been all the more richer for the excellent linguists and teachers I have met here: Ken Hale, Moms Halle, Jim Hams, Irene Heim, Michael Kenstowicz, Jay Keyser, Richard Larson, Howard Lasnik and David Pesetsky. They have contributed to my education leading to this thesis in ways I cannot descrik in short passages. Wayne O'Neil, the Departmel,: Chair, and Shigeru Miyagawa deserve a special mention: They were always concerned abut me, linguistically and otherwise, and they were always there when I needed them, for which I am most grateful. I would like to express my gratitude to the secretaries at the Department of Linguistics and Philosophy: Rachel Pearl, Nancy Peters, Marylin Sylva, Wendy Weber and Jamie Young, for their help in many ways: without their efficiency, I would have been at a loss with administrative paperwork, appointments, and so on. Every acknowledgements is filled with the names of fellow students, as well as teachers, the reason for which I dare not mention: I would like to thank those who spent so much time at MIT discussing linguistics and having fun: Harry Ladar, Itziar Laka, Yafei Li, Anoop Mahajan; Eularia Bonet, Lisa Cheng, Hamida Demirdache, Mike Hegarty, Sabine Iatridou, Perter Ihionu, Utpal Lahiri, Paul Law; Tom Green, Doug Jones, Toshifusa Oka, Akira Watanabe; Pilar Barbosa, Jonathan Bobaljik, Tony Bures, Diana Cresti, Hubert Truckenbrodt and Dylan Tsai. Among them, special thanks to Anoop (and Gyanam Mahajan, my TA for Ken Hale's class on I-hndi) for innumerable discussions on agreement and scrambling. I am also indebted to Peter for his Igbo judgments as a native speaker and insights as a linguist when I was working on my phonology generals. Toshi and Akira were my housemates for a year, and were always ready to talk abut a varieiy of topics, and that, in Japanese. 1 am particuiarly thankful to my classmates Phil Branigan, Chris Collins, Jill Gaulding, Bill Idsardi, Miori Kubo, Kumiko Murasugi and Chris Tanciedi, who made my coursework so much more enjoyable and post-generals perid more fruitful. As is the case with anyone working in any field, I have benefited much from a larger linguistic community beyond MIT: I would like to thank the late Osvaldo Jaeggli, Joseph Aoun, Muvet En$, Doug Pullyblank and Tim Stowell for their help and encouragements while I was around the University of Southern California. Special thanks go to Hajime Hoji and Caroline Schemer for the friendship they showed to me. I thank Marnoru Saito for his constant encouragement and interest in my work: It was his efforts that enabled me to visit USC back in 1986 to broaden my lingurstic perspectives. Many, long discussions with him have led to significant improvements in my thesis, too many and too deep to mention every occasion. Naoki Fukui has also been a great help ever since I got initiated to this area and MIT. His enthusiasm with linguistics has always made me aim for the higher. 1 owe many other Japanese linguists including Nobuko Hasegawa, Yoshihisa Kitagawa, Shige-Yuki Kuroda and Koichi Takezawa for so many fruitful discussions, insightful comments and suggestions. Each has been directly or indirectly reflected in what follows. I would like to thank Y asuaki Abe for his generous rditorial help in publishing my paper. I owe Y oshihiro Kubo so much for re-entering the Japanese linguistic academia. MY work on Igbo would have been of even poorer quality but for the revealing comment; and suggestions from Victor Manfredi and Mary Clark, which I deep:y appreciate. Thanks are in order for Jean-Plerre Koenig, my last-minute help for French data and theory: He put up with my incessant e-mail messages asking the most subtle of the gramrnaticali ty judgment. Access to the now-alrnost-extinct Brancieis linguistic community, made possible partly by my marital connection, enriched my life, both linguistically and socially: I would like to thank Claudia Borgonovo, Plroska Csuri, Henrietta Hung, Soowon Kim and Gyanam Mahajan for their linguistic enthusiasm and friendship. Thanks to my Japanese coEleagues I encountered over the course of my linguistic studies: Among them are Jun Ah, Keiko Mumugi, Fusa Katada and Noriko Yoshimura. Thinhng back my roots of linguistics, the one important figure back in Japan is Minoru Nakau, my advisor at the University of Tsukuba: I would like to express my deep appreciation to his persistent attitude towards linguistics and constant help for me and my work. Without his suggestion that I come to the U.S., I would have never continued linguistics. I cannot thank too much my family: My parents Hiroshi and Shigeko Tada have never spared their emotional and financial support throughout my life as a student. Encouragement from my brother Tetsuya Tada have always filled me with energy to pursue my goal. They have always believed in me and what I have been doing, and I will never be able to pay them back. Finally, to Saeko Urushibara, a salute. She contributed to this thesis as much as I did, in a variety of forms from linguistic discussions to secretarial work. Without her patience, encouragement, and above all love and devotion (even during her postpartun? days!), this thesis could not even have been filed. This thesis is dedicated to her and our son Shun-ichi Tada, who entered this world just in time to be included here. TABLE OF CONTENTS Abstract Acknowledgements Table of Contents Chapter 1. Introduction Chapter 2. A/A-bar Partition without Checking: Scrambliilg 2.1. Inthduction 2.2. The Landing Site Problem 2.2.1. Weak Crossover Neutralization 2.2.2. Reciprocal Binding 2.2.3. Strong Crossover 2.2.4. Adjunct Extraction 2.2.5. Summary 2.3. The Optionality Problem 2.3.1. Optionality of L-scrambling 2.3.2. Difference between S-scrambling and Wren A-Movement 2.3.2.1. The Last Resort Principle 2.3.2.2. Optionality of S-scrambling 2.3.3. Optionality of M-scrambling 2.4. A Solution of the Optionality Problem 2.4.1. Geometrical and Dynamic Aspects of Movement 2.4.2. Other Types of Optional Movement 2.4.2.1. Short V-Movement in English 2.4.2.2. Wh-Imperatives in Geman 2.4.3. Reformulation of Some Economy Principles in terms of Morphological J3fects 2.5. A Representational Solution of the Landing Site Problem 2.5.1. L-Relatedness 2.5.2. Problems 2.6. A Derivational Solution of the Landing Site Problem 2.6.1. The Strict Cycle and Adjunction 2.6.2. Reformulation of the Strict cycle 2.6.3. (Anti)reconstruction Effects 2.6.4. AIA-bar Metamorphosis 2.7. Scrambling versus QR 2.8. A Speculation on Checking and Discharging Chapter 3. Uniformity of Chain-steps 3.1. Successive Cyclicity and the Notion of Totentid Stepn 3.2. Non-uniform Chains and the Compensation Principle 3.2.1. A-bar Chain Containing Last Resort Violation 3.2.2. L-tous 3.2.3. Partial NP-movement in there -construction 3.3. Complex Chains 3.3.1. Partial Wh-Movement Constructions 3.3.1.1. McDaniel(l989) 3.3.1.2. Comparison with There -Constructions 3.3.1.3. Corn prison with Scope Reconstruction 3.3.1.4. Uniformity onder hn~position 3.3.2. Fwus Antipassive Constructions in Mayan hnguages 3.3.2.1 .An Overview 3.3.2.2. Problems 3.3.2.3. A Typological Perspective 3.3.2.4. A Solution in terms of Double-ehxtking Appendix 3.3.3. Conclusion Chapter 4. Supports for Counter-Checking Analysis of Wh-Movement 4.1. Introduction 4.2. Multiple Sgecir ier Constructions 4.2.1. Comorovski(l986): Multiple A-bar-Specifiers and Extraction 4.2.2. Nominative Objects in Japanese 4.2.2. I. Basic Facts 4.2.2.2. The Relevance of Stativity 4.2.2.3. The Releirance of Government 4.2.2.4. Nominative Case Assignment as Agreement 4.2.2.5. Conclusion 4.2.3. Multiple A-Specifiers and Extraction 4.2.3.1. Relativized Mnimality EXfects on Object Shift 4.2.3.2. Multiple Specifiers of AGRP 4.2.3.3. Multiple Specifiers of AGRP and Relativized Minimali ty 4.3. Strong Crossover Reconstruction and Absorption 4.3.1. Higginbtham(l983): An Asymmetry between Strong Crossover versus Principle C 4.3.2. A Problem to Higginbotham's Theory 4.3.3. A Lebeaux-Type Andysis 4.3.4. Problems 4.3.5. Absorption and Functional Wh 4.3.6. Deriving Obligatoriness of Absorption CHAPTER ONE INTRODUCTION 1.1. The paths of syntactic derivations appears to be deterministic and lead to at most a unique well-fomd result. To explain this "function-like" (or more precisely, "partial function-like") character of them within the "principla-and-parametem" approach, where the most generous operation Move a (or Affect a) is assumed, there have been proposed a number of conditions and principles which exclude non-well-formed alternatives. Whether these proposais always give a unique and light result is an empirical issue. The minimalist approach program in Chomsky(1992), however, approaches this directly: alternative derivations are directly compared by optimality measures. Partition of syntactic positions and, subsequently, movement operations into A and A-&ir types is one of the most well-supported empirical generalizations. In the minimalist program, this partition of movement directly follows from the partition of types of morphological features of the moving element to be checked at the landing site. Thus NP with an unchecked Case-feature moves to a Spec-IP position with a checking Case-feature, and a wh-phrase moves to a Spec-CP position .with a checking [+wh'j feature. This thesis addresses two problems about partition of movement into AIA-bar: (i) how can scrambling, which appears to be a single operation, show both A and A-bar properties? (ii) uniformity of chain steps: if the driving force of movement comes from the final landing site, why are the intermediate steps are of the same type? These questions lead to a view hat economy principles do not directly apply to movemznt operations per se, but they crucially refers to the effects they would produce in the course of derivation. The abve problems of AIA-bar partition are solved if we focus on the morphological effects the movement operations produce. 1.2. Sumr,~ary of the Following Chapters In Chapter 2,l will explore two interreiated problems of scrambling. The first one, called the landing site problem, is why and how the apparently singie operation scrambling behaves A-movement and A-bar-movement. The second one, &led the optionality problem, is whether scrambling is an optional operation, and if so, why it mi violate the last resort principle. The znswer to the first question leads to the derivational view of AIA- bar partition. The answer to the second question lead to a system of effect-based economy principles. In pcular, the last resort principle is replaced by tlle compensation principle which requires the external effect of movement to be compensated for by its internal effect. In Chapter 3, I will discuss the problem of uniformity of chain-steps and suggest that the uniformity can be derived from stepwise application of compensation principle, and discusses complex cases such as German partid wh-movement and .Mayan focus antipassives in light of the compensation principle and derivational view of AIA-bar- + parh tion. In Chapter 4, I will discuss two phenomena related to the problem of multiple specifier in A and A-bar systems: extraction out of multiple specifier configuration and absorption of multiple wh-phrases. They are claimed to support Cheng's(l991) view that wh-movement is not driven by feature-checking as NP-movement is. 1.3. Definitions The category a dominates @ if every segment of a dominates @. The category a contains fJ if some segment of a dominates $. For a head a, Max(a) is the least full-category maximal projection dominating a. The domain of a ha3 a is the set of nodes contained in M;*X(a) that are distinct from and do not contain a. The complement domain of a is the subset of the domain of a reflexively dominated by the complement of the construction. The residuc of a is the domain of a minus the complement domain of a. For a set S of categories, MIN(S) (minimal S) is the snxllest subset K of's such that for any y E S, some f3 E Kreflexively dominates y. The internal domain of a is the minimal complement domain of a. The checking domain is the minimal residue of a. For a chain CH = (al, ..., a$), the domain of CH is the set of of nodes contained in W(a 1) and 30t containing any ai. L-features of a lexical i tem L are such morphological features as tense, Case, and +-features of L The L-features of an inflectional element I checks those of a lexical item L in the checking domiin of I. A position p is L-related (to a head H) if p is in the minimal domain of H and H has L- features. A position P is narrowly L-related if p is L-related and a nonadjoined position. A position p is broadly L-related if p is L-related and an adjoined position. p is an A-position if p is narrowly L-related. p is an A-bar-positions if p is not L-related. CHAFTER TWO AIA-BAR PARTITION WITHOUT CHECHING: SCRAMBLING 2.1. Introduction Free-word order phenomena in some alleged "non-configurational" languages such as Japanese have been rather successfully analyzed in terns of the movement operaticn called 'scrambling' so that we do not have to assume that the thematic information of lexical items is projected in syntax in different ways between Japanese and configumtional languages such as English except the direction of theta-role assignment. Saito(l985), which is the first extensive study of scrambling in Japanese, argues that scrambling is an adjunction operation which takes place at S-structure, an instance of A-bar-movement. Recent studies on scrambling in Japanese and in other scrambling languages such as German and Hindi, however, have shown that scrambling has not only properties of A-bar movement but also properties of A-movement.1 This duality of scrambling is a challenge to the exclusive partition of XP-positions in to A I A-bar positions in the LGB system developed in ~homsky(1981)2. This challenge is a serious one, since in most (or, perhaps, all) "principles-and parameterhapproaches since Chornsky(l981), the A / A- bar distinction is a cornerstone in determining which principle applies where, and if the definition of A I A-bar positions is not successfully made independent of the pri~clp!~ in question, the explanation becomes circular.3 I call this problem of scrambling the landing site problem. lcf. Webelhuth(l989) for German, Mahajan(l989.1990) for Hindi, and Saito(1992) for Japanese, among others. 2~s Mahajan(1989.1990) points out. the definition of A-position as a potential theta-position in Chomsky(l981) is also in consistent with the VP-internal subject hypotheses (cf. Kuroda, Koopman and Sportiche. Kitagawa, Fukui. among other), where Spec-IP, a typical A-posi tion, my nct be a theta- sition, even potentially. ~rarnno(??) explicitly makes Y s point with respeft to the explanatory power ol Riuifs(1990) Relativizsd Minirnality Principle. Befcre the above-mentioned extensive discussion began, scrambling had been felt to be outqide the core processes of UG and often assumed to be a "stylistic" operation applying at PF. This latter assumption must be abandoned given the effects of scrambling on other syntactic pracesses such as wd crossover and the binding theory, which are used as tests to know whether scrambling behaves as A-movement or A-bar-m0vement.h recent literature on scrambling. One of the reasons for this conception of scrambling as a "stylistic"operation, however, remains intact: Scrambling apparently lacks any syntactic "motivation" such as Case for NP-movement and [+wh] feature for wh-movement, in current terms. This point is sharpened by Saito(1989), who shows ihat A-bar-movement- like scrambling dffers from "standard" A-bar-movement such as wh-movement and topicalization in that it lacks any scope-defining property. This raises another serious problem for one of the cornerstones of the "minimalist program" in Chomsky(1992): the Last Resort Principle, which states that movement of a is pssible only when necessary for licensing (or checking) morphological fmtures of a such as Case or [swh]. I call this problem the o~tionality problem. The point of this chapter is to argue that the landing site problem and the optionaltty problem are correlated and consideration of them leads to a perspective of movement in terms of geometry and dynamics. The summary of the following sections is as follows: In Q 2.2., the landing site problem is discussed using several syntactic tests (weak crossover, reciprwd binding, strong crossover, and ECP). It is argued that there is a correlation between the distance of scrarnbllng and the possibility to have properties of A- movement: S(hort)-scram bling (direct object over indirect object) must show properties of A-movement, M(iddle)-scrambling ((in) direct object over subject) may or may not, and L(ong)-scrambling must not. These results partially argue for Mahajants(1989, 1990) non- uniform analysis, where scrambling is sorted to A-scrambling and A-bar-scram bling, and the distributions of the two types of scrambling are explained by independent principles. "Partially" kame M-scrambling does not pass all the tests for A-movement In 5 2.3, , the optionality problem is discussed based on the three-way distinction of scrambling and it is shown that each type of scrambling lacks the "motivation" it is expected to have. First, it is argued, following Saito(l989), that L-scrambling unlike wh- movement or topicalization is not scopedriven. Second, it is argued that S-scrambling is not Casedriven. Last, it is argued that M-scrambling is neither. The motivation problem, thus, seems to support Webelhuth's(l989) uniform approach. In 4 2.4., it is claimed that the contradiction between the non-uniform approach and the uniform approach is salved by distinguishing two aspects of movement: geometrical and dynamic aspects. The geometrical aspect of an instance of movement is characterized by the geometrical properties of the positions occupied by the members of the chain created. In prbcular, the A 1 A-bar distinction is to be made here. The dynamic aspect of an instance of movement, on the other hand, is characterized by what is "gained" by that instance of movement. The last resort principle is assumed to be defined dynamically. This weakened formulation of the last resort principle allows movement of a without motivation only if nothing isHgained" by that movement. It is argud that scrambling is of this type of "ineffective" movement. A certain type of short V-movement in English and wh-imperatives in German are claimed to be "ineffectiven optional movement. In 5 2.5, a representational solution for the typing of scrambling is proposed which is based on a peculiar property of =-adjoined position, that. is, it is in the minimal domain of both X and the head selecting XP. Although this analysis correctly predicts that VP-adjunction (S-scrambling) always behaves as A-movement, it predicts that IP- adjunction always have properties of both A- and A-bar-movement, contrary to fact. In 5 2.6., a derivational solution is proped which claims that the overlapping of minimal domains are only potential, and when an adjunction takes place in the course of gene;alized tranc,formation determines whether the XP-adjoined phrase actually belongs to the domain of X or that of the head selecting X. This amounts to saying that the strict cycle is to be formulated in terms of monotonic extention of the domafn of the target head, and adjunction as well as substitution obeys it. It is suggested that this version of strict cycle may also extend to singulary adjunction operation and explain the distribution of anti- reconstruction effects in kbeauxts( 1989) sense.. In $ 2.7., the problem of QR is discussed. It is claimed that a dynamic analysis allows to distinguish scrambling and QR, which are geometrically identical: XP-adjunction. 2.1. The Landing Site Problem Two conceptually distinct views of scrambling have been propd to account for the landing site problem. One is in Mahajan(11989, 1W), in which he essentially argues, hed on Hindi data, that 'scrambling' is the cover tern for two distinct movement operations: A- and A-bar-movement, and free word order phenomena can be derived by either of the two operations if no grammatical principle is violated. The other is in Webelhuth(1989), in which he argues, based on Gennan data, that the landing site of scrambling, which he assumes to be an adjoined position, is neither A -nor A-bar-position, but has both of their properties with respect to binding. The two proposals have different predictions about the distribution of A- IA-bar-like properties of scrambling: In Webelhuth's framework, all instances of scrambling should show the me amo~int of A- and A-bar-properties, given the assumption that the landing site of scrambling is uniformly an adjoined position. In Mahajan's framework, whether the landing site of scrambling may show properties of A (A-bar)-position or not depends on whether A (A-bar)- movement to that position is possible or not. The fact that the landing site of long-distance scrambling in Hindi lacks A-like properties is, thus, nicely explained in Mahajm's framework, since in the longdistance case, only A-bar-scrambling is available, given that W-movement out of a tensed clause is prohibited by the binding theory, as Mahajan argues. This fact dm not seem ts be explained straightforwardly in Webelhuth's framework. In this section, I will show, based on Japanese data, that although the landing site of longdistance scrambling exhibits properties of A-bar-psition consistently, the landing site of short-distance scrambling does not always exhibit properties of A-position even if A-movement is required in order for the sentence to be grammatical, and that two types of scrambling are to be distinguished even mong cases of short-distance scrambling, depending on the distance between the landing site and the original position. More precisely, I will show that three types of scrambling exemplified in (la,,b,c) are to be distinguished with respect to the properties of the landing site% (1) a. John-ga kono honq Mary-ni ti ageta. -nom this book-acc dat gave 'John gave this book to Mary.' b. Kono hon-~i John-ga Mary-ni 4 ageta. this book-acc -nom dat gave '(lit) This hk, John gave to Mary.' 'John read this book.' c. kono hon-q Bill-ga [John-ga Mary-ni ti agetag-to omotteiru. this book-acc -nom -nom t gave -camp think '(lit) This book, Bill thinks that John gave to Mary.' 'Bill thinks that John gave this bk to Mary.' 4~s observed by Saito(1985). long-distance scrambling to a position between the subject and the dative in a matrix clause is at best marginal: (i) ?*John-ga kono h0fi-q Mary-nl [Bill-ga ti yondal-to itta.. -nom this book-= dat -nom read-comp said '(lit) John said, this book, to Mary that BiIl read.' I will1 return to Phis problem in 9 2.2.5. In (la), the direct object is scrambled to a periphery position of the VP immediately dominating the original position. In (lb), the direct object is scrambled to the initial position of the clause immediately dominating the original psition. In (lc), the direct object is scrambled out of the clause to the initial pition of the higher clause. I will call (la, b, c) S-, kl-, L-scrambling, respectively. In Section 1, I will examine the properties of the landing sites of the three types of scrambling using several syntactic tests whict ( quire A-movement for the sentence to be grammatical, and show that L- and S- scrambling have typical A-bar-and A-movement properties, respectively, but M-scrambling behaves as A-movement with respect to some tests and as A-bar-movement with respect to the others. In Section 2, I will, however, argue that the three types of scrambling have a common property which neither 'pure' A-bar-movement such as WH-movement nor 'pure' A-movement such as NP-movement in passives has: optionality. In Section 3, I will attempt to account for the 'distancedependency' problem of scrambling, relating it to its optionality. 2.2.1. Weak Crossover Neutralization Weak crossover (WCO) effects take place if neither the trace of an operator (WH or quanbfier) nor the pronoun to be bound by the operator c- commands the other, as shown in (1): (1) a. Whoi 4 loves Risi mother? b. *Whq does hisi mother loves tj? In (la), the pronoun can be bound by the WH-phrase, since it is c-commanded by the trace left behind by WH-movement. In (lb), however, the pronoun, which is not c-commanded by the trace, cannot be bound by the WH-phrase. The same a rnimst shows up in Japanese if we use 'so-' expressions in stead of (so-called) lexical pronoms, as shown in (2): (2) a?Darq-ga scitsui-no sensei-o hihansita no. who-nom the-guy-gen teacher-acc criticized Q. 'Whq criticized the guyj's teacher?' b.*Soitui-no sensei-ga dm-o hihansita no? the-guy-gen teacher-nom wheacc criticid Q '(lit) the guyi's teacher criticized whoi.' 'Who did the guy's teacher criticized?' The contrast in (2a,b) is to be explained by the MrH-movement at, LF, which cream the same configurations as ( la,b). WCQ, however, is neutralized if the W-phrase is preposed to the position where it c-commands the 'so-' phrase by either S-scrambling, as shown in (3), or M-scrambling, as shown in (415: (3) a. *John-ga [[e ~oit~ui-ni aitagatteirui hito]-ni dare-o shookaishita no -nom the guy want-@meet persondat who-am introduced Q '(lit) John introduced (to) [the person who wanted meet the guyi] whoi' 'Who did John introduce to the guy who wanted meet the guyi ?' b.?John-ga darei-o [[e soitsui-ni aitagatteiru] hito]-ni tj shookaisita no -nom who-aa: the guy wit-to-meet person-dat introduced Q '(lit) John introduced jwhq [(to) the person who wanted to meet the guyi ti]] (4) a.*[e soitsui-o hitome mitalhito-ga darei-o sukininatta no -acc one-glance saw person-norn who-acc fell-in-love-with Q '(lit) the person who took a glance at himi fell in love with who;?' 'Who did the person who took a glance at him fall in love wiih?' b. ?dare;-o [e soitsui-o hitome mitalhito-ga tj sukininatta no who-acc the-guy-= one-glance saw person-nom fell -in-love-wi th Q '(lit) Whq, the person who took a glance at the guyi fell in love with ti?' In (3a) and (&), the 'so-'phrase is not c-commanded by the WH phrase in an A- psi tion and, thus, cannot be bound by it In (3b) and (4b), where the WH phrase is preposed to the position c-commanding the 'so-' phrase by S- and M-scrambling, respectively, the WCO effect disappears. This may suggest that S- and M-scrambling are cases of A- movement, since NP movement saves the violation of weak crossover effect, as shown below: (5) Wkoi ti ,seems to hisi mother [ti to be sick]? (6) ?--ga soitui-no sensei-ni ti shmksareta no. who-nom the-guyi-gen teacher-dat was-introduced Q '(lit) Whoi was introduced (t9) hi$ teacher ti.' 'Who was introduced t tc his teacher?' L-scrambling, however, does not seem to neutralize WCO effect, as shown in (716: (7) a. *[e soits&-o hitome mitdhito-ga wary-ga dare-o suhriinaru ] the-guy-acc one-glance saw person-nom -nom who-acc fell-in-love-with- -to omotta no amp thought Q '(lit) The person who took a gZance at the guyi thought that Mary would fall in love with who;' 'Who drd the person who took a glance at the guy thought that Mary would fall in love with?' b. *Dare-o [e soitsui-c hitome mita] hito-ga wary-ga ti who-acc the-guy-ace one-glance saw person-nqm -nom sukininani ]-to omottano fell-in-love-with -amp thought Q '(lit) Whq, the person who took a glance at the guyi thought that Mary would fall in love with ti.' - - - 6~ee Mahajan(l989.1990) for corresponding fiats in Hindi. Saito(1992) claims hit for some Japanese speakers. L-scrambling does marginally neutralize WCO, and that for those speakers, not just AIA-bar distinction but operatorlnon-operator distinction is crucial to WCO. See Lasnik and Stowell(??) for a similar formalization of WCO. In (7b), which is derived from (7a), L-scrambling of WH-phrase does not license the binding of the 'so-' expression in the matrix subject by the WH- phrase, resulting in WCO violation. L-scrambling, however, exhibits WCO neutralization with respect to the 'so-' expression in the embedded subject, as shown klow7: (8) &*Mary-ga [[e soitsui-0 hitome mita] hito]-ga m-9 -nom the-guy-ace one-glance saw person-nom who-ace suhninaru ]-to omottano fell-in-love-with -camp thought Q '(lit) Mary thought that the person who took a glance at the guyi fell in love with whoi' 'Who did Mary think that the person who took a glance at the guy fail in love with?' b. ?Darei-o Mary-ga [[e ~~its~i-~ hitome mitalhito-ga ti wheacc -nom the-guy-acc one-glance saw person-nom sukininm ]-to omofiano fell-in-love-with -comp thought Q '(lit) Whoi, Mary thought that the pemon who took a glance at the guyi fell in love with ti ' We may explain the WCO neutralization in (8b) by assuming, following Mahajan(1989), that (8b) is derived by succesive application of M- and L- scrambling, and that the intermediate trace does neutralize the WCO neutralization. The above examples show that with respect to WCO effects, both S- and M- scrambling behaves as A-movement, whereas L-scrambling behaves as an A-bar- movement. 2.2.2. Reciprocal Binding 'see Mahajm(l989,1990) for corrapcmrling facts in Hindi. 19 Let us turn to the second diaposis, reciprocal binding, to examine the properties of the landing site of the three types of scrambling. The landing site of NP-movement is, of course, a licit position from which to bind a reciprocal, as sho\=~n below: (9) They; seem to each otheri's mother to be stupid. (10) Karerq-ga otagaii-no sensei-ni ti shookaisareta. they-nom e.0.-gen teacher-dat was-introduced '(lit) Theyi were introduced (to) each other's mother tj' Thzy were introduced to each other's mother.' Pure A-bar-movements such as WH-movement or iopicalization, however, do not license reciprocal binding, as shown in (1 1): (1 1) a. *Which students did each other's mother scold? b. *These students, each other's mother scolded. L-scrambli~lg, as is expected, behaves as A-bar-movement, as shown in (12): (12) a *Otagaii-no ryoosin-ga [Bill-ga [John-to Maryli-o sikattal-to ornotteiru each-other-gen parents-nom -nom -and -acc scolded-amp think 'Each other's parents think that Bill scolded John and Mary.' b. *[John-to Mq]i-o otagaii-no rywin-ga pill-ga ti sikattta ]-to omotteiru -and -acc each-other-gen parents-nom -nom scolded -comp think '(lit) [John and Maryli, each otherj's parents think that Bill scolded ti.' The violation of reciprocal binding in (li2a) is not saved by L-scrambling of the intended antecedent of the recipd in (12b). S-scrambling, as is also expected, behaves as A-movement, as shown in (13): (13) a. ?*Bill-ga otagaii-no ryoosin-nl [John-to Mary3i-o shcmkaisita. -nom each-other-gen parents-dat -and -am intraiuced '(lit) Bill introduced (to) each other's ,parents John and ldary.' 'Bill introduced John and Mary to each other's parents.' b. ?Bill-ga [John-to Maryli-o otagaii-no ryoosin-ni tj shookaisita. -nom -and -acc ah-other-gen parents-dat introduced '(lit) Bill introduced [[John and Mary]i [(a) each other's parents ti]] In (13a), the reciprocal is not c-commanded by the antecedent, violating the binding condition A. In (13b), the S-scrambled antdent c-commands, and licenses, the reciprocal. M-scrambling, however, may marginally feed reciprocal binding, as shown in (14): (14) a. ?*Otagaii-no rymin-ga [John-to Maryli-o sikatta. eack-other-gen parents-nom -and -ace scolded. 'Ekck other's parents scolded John and Mary.' b. ??[John-to Maryli-o otagaii-no ryoosin-ga ti sikatta. -and -acc each-other-gen parents-nom scolded '(lit) [John and Maryli, each otheq's parents scolded ti.' For some speakers, the interpretation of (l4b) is not 'standard', in that the antecedent sf the reciprocal is not distributed. Thus, (14b) means not (1%) but (15b): (15) a. John's mother scolded Mary, and Mary's mother scolded John. b. John's mother scolded John and Mary, and Mary's mother scolded John and Mary. This seems to show that the landing site of M-scrambling has an intermediate status between A- and A-'bar-position, with respect to reciprocal bindingg. 2.2.3. Strong Crossover 81 will return to this dual character in 5 ??. The third diagnosis is strong crossover effects. A descripiive chmcterization of the phenomenon is given in (16): (16) A pronoun may not c-command a member of the A-bar-chin containing the quantified NP coindexed with the pronoun at S-structureg. (16), thus, rules out (17): In (13, 'hqQ c-commands &, which is a member of the chain mntaining 'whosei'. As is explicitily stated in (16), only an A-bar-chain is relevant to stlong crossover effects. Thus, (17) does not violate (16): ( 17) [Who~i motherlj t) Seems to himi [tj to be stupid]? In (13, although 'himil,c-commands 'ti', which in the D-structure position of the WH- phrase, 'himi' can be interpreted as a variable bound by 'whosej' since 'ti' is not a member of an A-barchain. A similar example in Japanese is given in ( 18) and (19) 10: (18) [Darei-no senseilj-ga soi tu-ni tj shookaisareta no. who-gen teacher-nom the-guydat was-introduced Q '(lit) phase; teacher] was intraduced (to) the guyi ti.' g~ee Higgisbotbam(l983) and 543. for more details of this condition. losee Nishigauchi(l!3%) for syntactic and semantic properties of wh-constructions with the scope marker mo. Strong crossover, thus, seems to be a good test to exairnine whether the relevant movement is A- ar A-bar-movement. kt us consider S-scrambling cases given in (18) and (19)dn- 12>: (18) &*John-ga soitui-ni darei-nosensei-o shookaisita-no? -nom the gu y-dat whegen teacher-acc introduced-Q '(lit) John introduced (to) the guyi who% teacher' 'Whose teacher did John introduce to the guy?' b. ?John-ga [darei-no senseilj-o soitui-ni tj shokaisi ta-no? -nom who-gen teacher-acc the guy-dat introduced-Q ( 19) a. * John-ga soitui-ni [[e a-ni aitagatteiru] hi to]-mo shookaisita -nom the-guy-dat who-dat want-to-meet person-MO introduced '(lit) John introduced (to) the guyi WO person who wanted to whq] 'For all x,y, x a person, y a person, x wanted to meet y, John introduced x to y.' b. John-ga [[e darei-ni aitagatteiru] hito]-mq soitui-rri tj shookaisita. . -nom whodat want-@meet person-MO the-guy-dat introduced In (1%) and (I%), 'soitu' c-commands the WH-phrase, violating Principle C at S- structure. (18b) and (19b) show that S-scrambling neutralizes a Principle C violation without violating (16). This result again suggesb that S- scrambling has a typical property of A-movement. L-scrambling consistently behaves as A-bar-movement with respect to ( 16), as shown in (20) and (21): (20) a. *Soitui-ga [John-ga dmi-no hahaoya-o naguttaj-to omotteiru-no? the guy-nom -nom who-gen mother-acc hit -comp think-Q '(lit) The guyi thinks that John hit whose mother' 'Whose; mother does the guyi think that John hit?' b. "Darei-no hahaoyalj-o ~oi&y [John-ga t, naguttal-to omotteiru-no? who-gen mother -acc the guy-nom -nom hit -comp think -Q '(lit) Whosq mother, the guyi thinks that John hit ti' a *Soitui-ga [John-ga katteni [[e &rq-ni kita] tegamil-mo the-guy-nom -nom without-permission who-dat came letter -MO yondesimaul-to omotteiru read -comp think '(lit) The guyi tbinks that John reads NO letter that comes to whq] without permission. 'For all x,y, x a person, y a letter, y comes to x, John reads y without permission' b. *[[e darq-ni kita] tegamil-moj soitui-ga [John-ga katteni who-dat came letter -MO the-guy-nom -nom without-permission yondesimau] -to omotteiru read -0omp think '(lit) @4O letter th~i comes to wh~ilj, the guyi thinks that John reads tj' (20b) and (21 b) are derived from (20a) and (2la),respectively, by L- scrambling, and they show strong crossover effects. M-scrambling, however, behaves, not as A-movement, but as A-bar-movement, as shown in (22) and (23): (22) a. *Soi tq-ga darei-no sensei-o nagutta-no? the guy-nom who-gen teacher-acc hi t-Q '(lit) The guyi hit whosei teacher.' 'Whose teacher did the guy hit?' b. * parei-no senseilj-0 SO^ tui-ga tj mgutta-no? whegen teacher-acc the guy-nom hit-Q '(lit) [Whosei teacherlj, the guyi hit ti' (23) a. *Soi tui-ga [[e darei-o sikatta] sensei]-mo hihansita the-guy-nom who-acc scolded sensei -mo criticized '(lit) The guyi criticized [MO teacher who scolded wkoi.' 'For all x,y, x a person, y a teacher, y scolded x, x critic~zed y.' b. * [[e darq-o sikatta] sensei]-mo soitui-ga ti hi hansi ta who-acc scolded teacher-MO the-gu y-nom criticized, '(lit) WO teacher who scolded whq]j, the guyi criticized tj (22bj and (23b), which are derived from (2%) and (22b), respectively, by M- scrambling, exhibit strong crossover effects. One might suppe that their ungammaticali ty is reducible to weak crossover effects, since WH-phrase does not c-command 'soitu' at S- structure. This, however, does not work well, since if 'soitu' does not c-commands the NP containing the WH-phrase at D-structure, A4-scrambling does exhibit weak crossover neutralization, as shown below: (24) a. * Soitui-no hakaoya-ga darei-no sensei* mgutta-no? the guy-nom mother-nom who-gen teacher-ace hit-Q '(lit) The guyli's mother hit whosei teacher.' 'Whose teacher did the guy's mother hit?' b. ?parei-no senseilj-o soittli-no hahaoya-ga tj nagutta-no? who-gen teacher-ace the guy-nom mother-nom hi t-Q '(lit) [Whosei teacherli, the guyli's mother hit ti' (25) a. * Soitupno hahaoya-ga [[e darei-o si katta] sensei]-mo hi kami ta the-guy-gen mother-nom wheacc scolded sensei -mo criticized '(lit) The guyli's mother criticized WO teacher who scolded whoi 1.' 'For x,y,z, x a person, y a teacher, z is x's mother, y scolded x, z criticized y.' b. [[e darei-o sikatta] sensei]-mo soi tui-no hahaoya-ga ti hi hanslta who-acc scolded teacher-MO the-guy-gen mother-no criticized '(lit) [MO teacher who scolded wh~ilj, the guyi's mother criticized tj.' In (24b) and (25b), which are derived from (24a) and (25a), respectively, by M- scrambling, WCO is neutralized even though the WH-phrase contained in the M-scmbled NP does not c-commaqd thc 'so-' expression. 2.2.4. Adjunct Extraction The Condition on Extraction Domains (CED) bars against the extraction out of a domain which is not properly govemedll. Two cases are generally covered by CED: extraction out of subjects and out of ad,uncts, as illustrated in (26) and (27), which are taken from Chomsky( 1986b): (26) *the man who [ [pictures of t] are on the table] (27) *Who did [they leave [before spealung to t]] In Japanese, however, only extraction out of adjuncts is constrained by CED, as shown by Wnik and Saito(1990) and IQlrchi(l987). Consider the following examples from ~tkuchi~ 1983)12: (28) * [[Op, Minna-ga [Paul -ga ti yonda atol-de mpo-ni dekaketa] y orimo] everyone-nom -nom read after-at walking-for went-out than John-wa takusan hon-o yondeita -top many book-ace have-read '(lit) John has read more books than everyone went for a walk after Paul read' (Kikuchi( 1987), P37) (29) ?Paul-wa [[OF+ [[[John-ga ti yonda] kotol-ga akiraka-na]] yorino -top -nom read fact-nom clear is than takusan hon-o yondeita many bmk-ace have-read l~f. Hwg(1982) and Chomsky(l986a). 12The slight marginality is to be related to Complex NP Condition induced by the nominal character of the subject clause, since even if the it appears in an object pi tion, extraction out of ir results in marginality of' the same dep: (i) Waul-wa [[OPi Fizay-ga [[John-ga ti yonda] kotol-o akiraka-ni site]] yorimo -top -ga -nom read fact-ace clear did than takusan hon-o yondeita many book-ace have-read '(lit) John has read more books than Mary made clear the fact that Paul read' '(lit) John has read more books than the fact that Faul read Is clear.' (3Cikuchi(1987), 935)) In (28), the empty operator is extracted out of an adjunct, resulting in the violation of CD. In (29), however, extraction of the empty operator out of the clause in subject position does not induce a CED violation. Moreover, even extraction out of the derived subject in a passive construction dws not induce a CED violation, as illustrated in (30113: (30) ?Paul-wa [[OPi [[[John-ga ti yon&] kotol-ga akiraka-ni sareta] ] -top -nom read fact-nom clear was-done yorimo] takusan hon-o yondeita than many book-acc have-read '(lit) John has read more books than the fact that Paul read was made clear.' NP-movement in English, on the other hand, dm yeld subjects that induce a CED violation, as shown in (30)14: (30) a. Who did John see pictures of t ? b. *Who were pictures of t seen?. In order to explain the non-existence of CED effects with respect to subjects in Japanese, we must assume that every argument position, whether base generated or derived, is properly governed in Japanese. Given this assumption, we may expect CED to be a possible diagnosis for examining whether the relevant movement is A-bar-movement or NP-movement. Let us now consider (3 1): (3 1) a. * John-ga [[e sorei-o sinjiru hazunonai] hito]-ni Mary-ga -nom it -acc believe would-not person-dat -nom (i ttai) dare-o kcrosita] -ttei tutaeta-no? 13see Hoji, Mi yagawa, and Tada( 1989) for several arguments for an h7-movement analysis of direct passives in Japanese. 14see Lasnik and Saito( 1992) for mlevant discussion. 27 the-Re11 whenom killed amp told-Q '(lit) John told the pemn who would not believe i& [that Mary killed who (the hell) li 'WhMthe hell) did John say to the -person who would not believe iti [that Mary killed t ]i?' b. John-ga wary-ga (ittai) dare-o komsita] -tkj [[e s0rei-o -nom war-mom the-hell wheacc killed amp it-aicc sinjiru hazunonai] hito]-nil 3 tubeta no? believe would-not person-dat told Q In (31), 'sore' cannot be coindexed with the CP containing a WH-phrase, which does not @-command 'sore'. If the CP is S-scrambled to a position which c-co~nmands 'sore', the result is grammatical. We may consider this effect b be a special case of WCO neutralization (cf. Sec. 1. l), if we assume that any expression containing a variable bund by a quantifier induces WCO violation. This is confirmed by the behaviors of M- and L- scrambling, as illustrated in (32) and (33), respectively: (32) a * [[e Sore-o sinji temoinai] hi to] -ga wary-ga (ittai) dare-o it-acc did-not-believe person-nom -nom the-hell wheacc korositd-tte itta no killed -amp said Q '(lit) The person who did not believe iti said [that Mary killed wh~]i 'Whq did the person who did not believe iti say [that Mary killed wholi?' b. wary-ga (ittai) dare-o korosital-ttq [[e sorq-o sinji ternoiriai -nom the-hell who-acc killed amp it-= did-not-believe hito] -ga ti i tta no. person-nom said Q '(lit) phat Mary killed who (the hell)], the person who did not believe i& said ti. (33) a *[[e S0rei-o hazime-ni iidasita] hito]-ga [John-ga wary-ga (itti) it-ace originally proposed person-nom -nom nom the- hell dare- korosita]-to hajimeii iidasi td-to ornokiru no who-acc killed -comp originally proposedamp thinks Q The man who originally proposed i& thinks that John originally proposed that Mary died who?' b. *Flary-ga (ittai) dare-o korosital-tq [[e sorej-0 hajimeni -nom the-hell wheacc killed-comp i t-ace originally iidasita] hito]-ga [John-ga hajimeni ti iidasital-to omotteiru no proposed person-nom -nom originally propos~d-comp think Q '(lit) phat Mary killed who]i, the person who originally propad iQ thinks tha.1 John originally proposed ti' M-scrambling of the CP containing a WH-phase makes the coindexing of 'sore' with the CP possible, as shown in (32), but L-scrambling of the CP does not, as shown in (33). This pattern is consistent with the generalization we gave in &tion 1.1. that S- and M- scrambling behaves as A-movement, whereas L-scrambling behaves as A-bar-movement. This pattern, however, breaks down, if the WH-phrase contained in the CP is a 'true' adjunct, as shown in (34-3611% (34) a. *John-ga [[e sore;-o sinjiru hazunonai] hi to]-ni [sensoo-ga -nom it -act believe would-not person-dat war-nom (ittai) naze okotta ] -ttq t tutaeta-no? the- hell why tmk-place-comp told-Q '(lit) John told the person who wodd not believe i& [that the war took place why (the hell)]i 'Why (the hell) did John say to the person who would not believe itj [that the war took place t]i?' b. ?John-ga isnrw-ga (ittai) naze okotta] -ttej [[e sore;-o -nom war-nom the-hell why took-place -comp it-ace sinjim hazunonai] hi to]-nil tj tutaeta no? believe would-not person-&t told Q '(lit) [That the war took place why (the hell)]i, John said that the person who would not believe iti' (35) a *[[e Sorq-o sinjitemoinai] hito] -ga [sensou-ga naze okotta 1-tte itta no it-acc did-not-believe pmn-nom war-nom why took-place -comp said Q 15(34-36) are of course grammatical if sore is not correferential with the CP. 29 "lit) The person who did not believe i@ said [that the war took glace why]i 'Whyi did the person who did not believe iti say [that the war took place Vjji ?' b. *[Sensour-ga araze okotta 1-ttei i[c sorei-0 sinjitemoinai war-norn why took-piace -mmp it-acc did-not-believe hi to]-ga ti itta no. person-nom said Q '(lit) [That the war took place whyli, the person who did knot believe iti said G. (36) a. * [[e sow-o hazime-ni iihi ta] hi to]-ga [John-ga Isensou-ga i t-ace originally proposed person-nom -nom war-nom naze okotta 1-tte ka~imeni iiwi ta]-to omotteiru no why took-place xomp originally pmposed-mmp thinks Q '(1it)The man who originally proposed iti thinks that John originally proposed [that the war took place why]?' 'Why does the person who originally proposed it thinks that John originally proposed [that the war took place t] b. * [Sensou-ga nm okotta 1-ttei [[e sorei-o hajimeni war-nom why took-place amp it-acc originally iidasita] hito]-ga [John-ga bjimeni ti iidasita-to omotteiru no propod person-nom -nom originally prop&-romp think Q '(lit) [That the war took place whyli, the person who originally proposed iti thinks that John originally proposed ti' In (34b), which is derived from (3419 by S-scrambling of the CP, WCO neutralization takes place. In (36b), which is derived from (36a) by L- scrambling of the CP, WCO neutralization does not takes place. These two cases are consistent with ;he generalization given in Section 1.1. M- scrambling of the CP, however, does not induce WCO neutralization, contrary to the generalization, as illustrated in (3%). which is derived from We may assume that this break-down is caused by a CED (or more precisely, the Empty Category Principle (ECP)) effect with respect to extraction of the adjunct Wh-phrase out sf arr adjunct at LF. We may, thus, regard the conbrast between (31b) and (35b) as comparable to that between (37a,b): (37) a. wary-ga nani-o &a]-node John-ga okotta no -mom what-acc did-because -nom got-angry Q '(iiijJohn got angry because May did what?' b.* [Mary-ga YL~~R kita'J-node John-ga okotta no -nom why came -because -rrom got-angry Q '(1it)John got angry because Mary came why?' The parallelism between the standad ECP pheoomena and the "ECPn induced by WCO neutralization is consistent if we replace 'mi' by 'donna riyuu' (= 'for what reason'), a near synonym, as shown in (38) and (39) 16: (38) Flary-ga donna riyuu-de kital-node John-ga okotta no -nom what reason-for came -because -nom got-angry Q '(1it)Joh got angry because Mary came for what reason?' (39) [Semou-ga donna ri yuu-de okotta 1-tq [[e sorei-o sinj iternoinai war-nom what reason-for took-place -camp it-acc did-not-believe 16hglish also exhibits this kind of argument-,adjunct asymmetry with respect to the possibility of comference between a pronoun and a 8 containing a wh-trace: (i) What did you say [Joh bought t]i, although no one would believe iti? (ii) Who did you say [t b~ht your cadi, although no one would believe i~? (iii) %y did you say [John died t]i, although no one would believe itj? The bracketed CP can be coreferential with the pronoun if the trace is an object, as in (i), or a subject, as in [ii). This corderenee, however, is impossible if the trace is a "true" adjunct, as in (iii). This pattern can be muted for if we assume that the CP, which is not a referring expression, must undergo QR in o~der to c- sommand and, hence, bind the pronoun at LF. and that the CP in A-bar position becomes a barrier. It should be noted that the configuration resulting from QR is immune to WCO, since the pronoun is contained in an adjunct clause (cf. Stowell(??)). The cuntmt. between (ii) and (iii) strongly suggests that antecedent government takes place at S-stmctupe for arguments and at LF for adjuncts, Interestingly enough, pied-piped sdjwts patterns with arguments, as in Japanese: (iv) For what reason did you say [Joh died t]i, although no one would believe iti? hito]-ga ti itta no. person-nom said Q '(lit) what the war took place why]i, the person who did not believe iti said ti. In order to explain the pattern illustrated in (34-36) in terms of the ECP, which prevents extraction of 'true' adjunct out the A- bar-psi tion, but mot out of A-psi tion in Japanese, we must conclude that S-wambling behaves as A-movement, whereas M- and L- scraiibling behave as A-bar-movement. 2.2.5. Summary Let us summarize the result given above in (40): (4) S-scrambling M-scmbling L-scrambling WCO-neutralization A A A- bar Reciprocal Binding A A and A-bar? A-bar Strong Crossover A A-bar A- bar Adjunct Extraction A A-bar A- bar We may thus conclude that S- and Lrscrambling has typical properties of A- and A-bar- movement, whereas M-sclramblng has propertiis of both A- and A-bar- movement, but only partially17. A natural question immediately arises: Are the three types of scrambling l7This chacterizati~n 8ccounts for Saito1s(1985) observation that long-distance scrambling to a position between the subject and the dative is at best marginal, as illustrated in (i): (0 ?*John-ga kono hon-oj Mary-ni [Bill-ga ti yondal-to itta. -nom this book-scc dat -nom read-comp said '(lit) John said, this book, to Mary that Bill read! Scrambling to a post-subject position must be always A-movement, as the tats for S-scrambling shows, aud scrambling out of a tensed clause must be always A-bar movement, as those for L-swdmbling stows. These two requirements are in contradiction in sentences such as (i). distinct operations in reality, or are they lfferent manifesbtions of he same single operation? We will see some evidence for the second possibility below. 2.3. The Bptionality Roblem In this section, we will show that all the three types of scrambling, which (ape phenomenally to be distinguished, as we have seen in $2. I., share an essential property : optionality. 2.3.1. Optionality of L-scrambling In ?2.2., we have seen that L-scrambling has all the A'-movement properties, as far as the four syntactic tests can tell. Saito(1989), however, shows that L-scmbling has an important property which is not shared by 'pure' A'-movement such as WH-movement, giving the following examples: (41) a [~pl Mary-ga [m minna-ga [m Jsh-ga dono hon-o toshokan-kara -nom all -nom -nom which book-a@c library-from karidasita] to] omotteiru] ka] siritagatteiru] (koto) checked-out COMB think Q want-to-know fact '(lit) [piMary wants to know [~p2 Q[IP;! everyone thinks [mthat John checked out which book from the library]]]]' 'Mary wants to know which book everyone thinks that John checked out?' b. ??[ [ john-ga don9 hon-o toshokan-kara karidasita] to]: [ Mary-ga [ [mima-ga t. omotteiru] ka] siritagatteiru]] koto '(lit) [~pl I~p3That John checked out which book from the libraryli [rPl Mary wants to know [cP;? Q beveryone thinks tj ]I]] (41b) is derived from (41a) by L-scrambling of the most deeply embedded CP containing a WH-phi=, resulting in a bi.1 marginal status. The point in these examples is that the CP is L-scmbled out of the scope domain of the Q-maker to which the WH-phme must move at LF. If only the WH-phrase moves to Q, the resulting structure violates the proper binding condition, which is schematically illustraeed in (42): (41b) must be as bad as (43a), since (43b), the schematic structure of the LF sf (434, is regresentationally as bad as (42). with =pt to the proper binding condition: (43) a. *bl John-ga dare-ni [C~Z [IP2 Mary-ga kum] b] osieta] (koto) -nom whdat -nom come Q taught fact '(lit) [~pl John told who [@~2 bQ Mary is comingll]' b. [IPI...~~ ...[cP~ IP2 whi-Q] ...I (42b), however, is much better than (43a). In fact, (43) is uninterpretable. Given these facts, Saito concludes that the (L-) scrambled CP is undone at LF, saving the violation of the proper binding condition. Thus, the final LF representation of (41b) is schematically (44) : Saito claims that this kind of 'literal reconstruction' is available to scrambling, which is 'semantically vacuous', but not to 'semantically significant' A'-movement such as topicalization, as shown in (45): (4% *Mary thinks that [the man that laought whatlj, John knows wkoi ti likes tj A similar phenomenon is exhibited in the interaction sf scrambling and QP- scope interpretation. As is claimed by Kur&(l%?,1970) ar,d extensively discussed by Hoji( l985), in the basic word order, QP scope interpretahon is determined by the S- structure configuntion, but scrambling may change the interpretation. Thus in (#a), 'someone' must have scope over 'everyone1, whereas the scope interpretation of (46b) is ambiguous. (46) a. Dareka-ga darerno-o aisiteiru. meone-nom everyone-acc loves 'Someone loves everyone.' b. Daremoi-o, dareke-ga g aisiteiru. everyone-acc someone-nom loves '(lit) Everyone, someone loves.' Let us consider the interaction between Lscrarnbling and QP scope interpretation, as illustrated in (47): (47) a. Dcireka-ga [John-ga daremso aisiteirul- to omotteiru. someone-nom -nom everyone-acc loves amp think 'Someone thinks that John loves everyone.' b. Daremoi-o, dareke-ga [John-ga aisiteiml-to omotteiru. everyone-acc someone-nom -noin loves amp thinks '(lit) Everyone, someone thinks that John loves.' Surprisingly, (4%) as well as (47a) are unambiguous, that is, 'someonef must have scope over 'everyone" This means that Lscm bling of QP cannot determine its scope, coinciding with Saito's characterization of (L-)scrambling as 'semantically vacuous' A'- movement. 2.3.2. Difference between S-Scrambling and 'Pure' A-movement In Section 1, we showed that S-scrambling has all the properties of A- movement detectable in terms of the four syntactic tests. In this Section, we will consider one more typical (or essential) property of A-movement, and examine whether S-scrambling has the property. 2.3.2.1. 'Last Resort' Principle Consider (48), which Is from Lasnik(??): (48) a. It strikes me that Mary is peculiar. b. *Ij strike ti that Mary is peculiar. The subject of 'strike' is in a theta-bar position, as shown in (m), but the NP-movement to that position is impossible, as shown in (48b). Neither theta-criterion nor the binding condition A is violated. (48b) seems to be mled out only by the 'last resort' principle in Chomsky ( l986a): (50) If C = (al, ..., an) is a maximal CHAIN, then a, occupies its unique theta- position and a 1 occupies its unique Case-marked pi tion. (48b), thus, is ruled out, since in the maximal CHAIN (I, t), both the members are in a Case-marked position. This principle essentially rules out unnmssary NP-movement, having a flavor of the 'least effort' principle in Chomsky(1989). We can show that the hi-movement in Japanese also obeys the last resort' principle. Let us consider so-called 'intransitivizing resultatives' (IR) such as (27)(cf. Martin(1975)): (51) Kabin-gai tsukue-ni oi-tearu. vase-nom deskdat put-TEARU 'A vase has been put on the desk.' IR is created by attaching the affix '-tear=' (or more precisely '-te-am') to a verb, resulting in t4e suppression of the external theta-role and optional case conversion (ace -> nom). Hoji, Miyagawa, and Tada(1989) argues that the case conversion is established by NP- movement from the object pition to the subject position, giving the following kind of evidence: (52) a. Kabin-gai tukue-ni ti 3-tu oi-tearu vase-nom desk-dat -cl. put-TEARU Three vases have been put on the desk.' b. Nanika-gai subete-no-tukue-ni ti oi- team something-nom all-gendeskdat put-TEARU 'Something has been put on all the desks.' c.?[[sorei-o huita]zoukin]-gaj dono-tukue-ni-moi oi-team18 it-ace wijxd cloth-nom whichdesk-dat-MO put-TEARU The cloth that was used to wipe, it has been put on every d~ki.' In (524, the numeral quantifier which is not adjacent to its 'ankcexlent' is licensed by the adjacent NP-tme. In (52b), the subject quantifier can be interpreted as within the scope of the quantifier in the indirect object position, since the latter c-commands the trace of the former. In (2&), an empty category (pro) can be construed as a bound variable of the quantifier in the indirect pition, since the quantifier c-commands the trace of the NP containing the ~FG (cf'. Barss(1986)). All the three phenomena show some sort of 'reconstruction effect.' Given the NP-movement analysis of IR, let us consider a case of IR without case conversion: (53) pro tsukue-ni kabin-o 3-tu oi-tearu. desk-dat vase-acc cl put-TEARU '(lit) There have been put three vases on the desk.' Three vases have been put on the desk.' In (53), the object NF is assigned accusative Case by the verb and stays in- situ. This raises a problem, since if the verb with '-tearut can assign Case to the NP object, as in (S), we might expect the object trace in (51) and (52) to be assigned Case, violating the 'last resort' principle. To examine whether this is the case or not, Consider (54)dn->: 18whether NP-movement may show reconskction effects is a mntroversial issue. For at least some speakers, (i) is bctter than standard WCO cases such as (ii): (i) Hisi mother seems to everyonei [t to be sick] (ii) Hisi mother loves eveqonq. (54) a pro John-ni kabin-o 3-hi kaw-ase-tearu. -dat vase-acc -cl. buy-cause-TEARU '(lit) There has been made John buy three vases. 'John has been made to buy three vases. b. *Kabin-gq John-ni ti 3-tu kaw-ase-tearu vase-nom -dat -cl. buy-caus-TEARU '(lit) Three vases have been made John buy t.' If '-team1 is attached to the causative affix which is attached to a transitive verb, the object of the verb can show up as accusative NP, but not nominative W. One might assume that this is the violation of the binding principle A, since the NP trace, an anaphor, is not bound within the complement of the causative, which appears to be the governing category of the NP-trace. This explanation, however, seems to be problematic, since, as Kitagawa(1m) discusses, The goverriingcategory for the anaphor in that position is not the embedded clause, but the matrix, as the grarnmaticality of (31b) against the ungrammaticality of (55a) shows: (55) a. ?*John-gq wary-ga zibudsin-q semeta to] omotteiru -nom -nom self-acc criticized COW think '*John thinks that Mary ciiticized himself.' b. John-gq Mary-ni zibunzisin-oi seme-we-& -nm dat self-acc criticize-caw-past '*John made Mary criticize himself.' Miyagawa(l989) gives an interesting account to similar eases, which contains, not IR, but passives. Consider (56): (56) a John-gaj ti kusuri-o nom-as-are-ta -nom medicine-acc take-caus-pass-past 'John was made to take a medicine.' b. *Kusuri-gai John-ni ti nom-as-are-ta medicine-nom -dat take-caus-pass-pas; '*A medicine was made John take.' If the passive morpheme is attached to the causative morpheme which is attached to a transitive verb, only the dative s~tbject, but not the accusative object, of the verb can be NP- moved. Miyagawa claims that this phenomenon can be explained if we assume that passive morpheme '-rareG must absorb case, and that case absorption obeys an adjacency condition. Thus, (56a) is weli-formed, since '-rareq absorbed the case assigning feature of causative adjacent to it. (Sb), on the other hand, is mled out, since '-rare1 failed to absorb the case assigning feature of the verb which is not adjacent to it. Now, let us retun1 to (53), (52a), md (54) repeated as (57a), (57b) and (58), respective1 y: (57) a pro tsukue-ni kabin-o 3-tu oi-tearu. desk-dat vase-ace put-TEARU Three vases have been put on the desk.' b. Kabin-gaj tsukue-ni ti 3-tu oi-team. vase-nom deskdat -cl. put-TEARU Three vases have been put on the desk.' (58) a pro John-ni kaoinu 3- ti^ kaw-ase-tm. -dat vase-acc -cl. buy-cause-TEARU '(lit) Tnere have been made John buy three vases.' 'John has been made to buy three vases.' b. 'kabin-gai John-ni ti 3-tu kaw-ase-team vase-nom ., ;dat. . .x1, buy-=us-TEARU '(lit) Three vases have been made John buy.' The contrast between (57b) and (58b) shows that NP-movement in IR is possible only .____ _..-r- +- when Case-ab.wrption is possible, according to Miyagawa's theory, which prevents 'long- distance' Case-absorption through a causative morpheme. We may explain this contrast, if we assume that '-team' optionally &sorbs the Case-assigning property of the verb, and that NP- movement takes place only if the Case assigning property is absorbed, obeying the 'last-resort' principle. (58b) is, thus, ruled out, since the 'trace' is assigned the Case which '-tearut failed to absorb. We must, now, conclude that hi-movement in Japanese obeys the 'last-resort' principlelg. One way to show whether S-scrambling obeys the 'last resort' principle is to construct a context where S-scram bling moves an eleme~lt in Case- marked psi tion to another position (either Case-marked or not), and examine whether the landing site exhibits properties of A-movement. This is, however, difficult to pursue, since recently it has been proposed by Chomsky(1W) that structural Case assignment is licensed by some functional category within an articulated IP structure (cf. Pollock(1989)). These abstract $4.2.. I will give an object-shift analysis of nominative objects constructions, and argue that object shift in that sense obeys the last resort principle. functional categories are prly known so far, in parhcular, in Japanese, where no agreement morphology has been detected. So, it is extremely difficult to tell whether the landng site of such a short distance movement as S-scrambling is a Case-marked position or not for independent grounds. There, however, seems to be another way to examine whether S-scrambling obeys the 'last resort' principle or not. Consider .. . the following: (58) a. John believes that Mary will win. b. John believes the claim. (59) a. It is believed that May will win. b. *It is believed the claim. (60) a. That Mary will win is believed. b. The claim is believed. As shown in (58), both CP and NP zan appexu in a Casc pition. But, if Case is not available there, only CP can stay these, as shown in (59). In that context, NP must move to the subject position, whereas CP may move there, as shown irl(60). These examples show that NP must be assigned Case, whereas CP can be, but need not. The optionality of Case assignment, however, is limited only in the complement position, as Lasnik and Uriagereka(l988) points out, giving the following examples: (61) a. I believe [[that John loves Mary] to be surprising]. b. *It is likely [[that John loves Mary] to be surprising]. To explain this mystery, the following data are suggestive**: thank David Pesetsky for infonrliog me that this pattern is discussed in Kitagawa(l986). (62) a. It is likely [t to be believed ithat John loves Mary]]. b. *It is likely [[that John loves Mary] to be believed t]. c. [That John loves Mary] is likely [t to be believed t] (62) shows that CP need not get Case, but if it NP-moves, it must reach a Case manled position, that is, if it moves, it must obey the 'last resort' pnnciple2122 To extend this analysis to (61), dl we have to dc is to assume is the VP-internal subject hypothesis which implies that all IP subjects are derived by NP-movement. If this analysis is on the right track, we should conclude that the 'so-called' small clauses are headed by a functional category, to explain the ungmmmaticality of the following example: (63) *It seems [[that John loves Mary] surprising]. Given the assumption that CP cannot NP-move to a non-Case marked position, we can, now, examine whether S-scrambling obeys the 'last resort' principle. Let us take (34, repeated here as (64): (64) a. *John-ga [[e sorei-o sinjiru hazunonai] hito]-ni [sensotsga -nom it -ace believe would-not person-dat war-nom (ittai) naze okom 1-ttq braeta-no? the-hell why took-place-comp told-Q 0- 21~ven if we assume, following Koster(??), that sentential subjects occupy the topic position, this argument will not be affected. In that case, NiP-movement of the CP to a subject position is (obligatorily) followed by its topidzation. and if nominative Case is not assigned to the variable in the subject sition, it violates the ht Resort, 53.2.4.. I will give an alumative amount within Chomky's(l991.1992) Case Theory, where rhe subject of the embedded clause in ECM constructions is not a Case-marked Position, and moves to Spec- AGR,P to get Case. '(lit) John told the person who would not believe i& [that the war took place why (the hell)]i 'Why (the hell) did John say to the person who would not believe iQi [that the war took place t]i?' b. ?John-ga [sensoo-ga (ittai) naze okotta] -ttei [[e s0rei-O -nom war-nom the-hell why took-place amp it-ace sinjiru hazunomi] hito]-nil ti tutaeta no? believe would-not person-dat told Q '(lit) [That the war took place why (the hell)]i, John said that thc person who would not believe iG' As we have seen in Section 1.4, (64b), which is derived from (64a) by S- scrambling of CP, shows a typical A-li ke property of S-scrambling, that is, it neutralizes weak crossover without creating a Mer for adjunct extraction. We may now examine whether S- scrambling to a non-Case position is possible, if wc passivize (64): (55) a. *[[e Sorei-o sinjiru hazunonai] litd-ni [sensou-ga (ittai) naze itlacc believe would-not persondat war-nom the-hell why okotta 1-tte tutaerareta no took-place amp was-told Q '(lit) It was said to the person who would not believe iti [that the war twk place why (the hell).' 'Whyj (the hell) was i$ said to Mary [that the war took place tjli?' b. ?[Sensou-ga (i ttai) naze okotta 1-tte [[e Sorq-o sinjiru war-nom the- hell why took-place -comp ih-acc believe hazunonai] hito]-ni tutaemta no wodd-not person-dat was-told Q (65b) seems to be as good as (64b). One might suppose that (65) is derived not by S- scrambling, but by NP-movement to the subject position, which does not exhibit CED effects, as we saw in Section 1.4., since the expletive subject in Japanese is not lexical, but empty (cf. Section 2.2.1). There, however, is evidence against the existence of sentential subjects in Japanese. Let us consider the following: (66) a [[Niti-bei-kankei-ga akka shi-teiru] kotol-ga kokurnin-ni tsutae-rare-la Japan-US-relation-mm worse do-prog fact-nom nation-dat announce-pass-past The fact that the Japan-US relation is becoming worse was announced to the Japanese nation. ' b. [[Niti-bei-kankei-ga akka shi-teirul-to kokumin-ni tsutae-rare-ta Japan-US-relation-nom worse do-prog -comp nation-dat announce-pass-past 'It was announced to the Japanese nation that the Japanese-US relation is becoming worse.' or That the Jap-US relation is becoming worse was announced to the Japanese nation.' The minimal difference between (a) and (67a) is that the theme argument of the verb 'tutaeru'(announce) is NP in (67a), but it is CP in (67a). The theme NP is clearly moved to the subject position and assigned nominative Case there. The surface structure of (66b) does not tell whether the theme CP is in the subject position or not, since the CP is not assigned a visible case morpheme. To examine it, let us consider (67a,b) next: (67) a [Witi-bei-kankei-ga akka shi-teid koto]-ga kokumin-ni tsutae-rare-nihi Japan-US-relation-nom worse do-prog fact-nom nation-dat announce-pass-hard The fact that Japanese-US relation is becoming worse is hardly announced to the Japanese Nation.' b.*miti-bei-Wei-ga 3Wra sh-teiru ]-to kokumin-ni tsutae-rare-nikui Japan-US-relation-nom worse do-perf. amp nation-dat announce-pass-hard The adjectival suffix '-nikui' can take as its complement an infinitival version of (664, as shown in (67a), but not an infinitival version of (66b), as shown in (66b). This contrast seems to k related to the following contrast: (68) a. ??Keno-te-no kabiq-ga onnanako-no tukue-ni 4 oi- teari-ni kui . this-kind-gen vase-nom girl-gen desk-on put-TEARU-hard This kind of vase has hardly be put on a girl's desk.' b. *pro kono- te-no kabin-o omanokeno tukue-ni oi-teari-ni kui. this-kind-gen vase-acc girl-gen desk-on put-TEARU-hard (6%) has a nominative subject as is clear from the case morphology. In (68), the matrix subject is expletive pro, since the theme object of IR is in-situ (cf- Section 2.2.1). This contrast suggests that 'nikui' must take a mon-pleonastic subject. We may explain the ungwrnmaticality of (67b) if we assume that subject CPs do not exist in Japanese. Let us return to (65), repeated here as (69): (69) a. *[[e Sorei-o sirnjiru hazunonai] hito]-ni [sensou-ga (ittai) naze it/acc believe would-not person-dat war-norn the-hell why okotta 1-tte tutaerareta no took-place amp was-told Q '(iit) It was said to the person who would not believe iti [that the war took place why (the hell) ]i .' ''Whyj (the hell) was iG said to Mary [that the war took place tj]i?' b. ?[Sensou-ga (ittai) naze okotta I-ttq [[e sorei-o sinjiru war-nom the-hell why took-plm -camp iti-acc believe hazunornai] hito]-ni ti tutaerareta no would-not person-dat was-told Q In (69b), which is derived from (694, the prepsing of the CP containing the WH-adjunct neutralizes WCO without inducing an ECP violation. Given the discussion so far, the preposing may not be NP-movement to the subject psi tion, since subject CPs do not exist in Japanese, but it must be S- scrambling to a VP-periphery position, where no Case is available because of Case-absorption by passive morphology. This conclusion is supported by the following : (70) ?[Sensou-ga (i ttai) naze okotta 1-tte John-ga [[[e sore-o sinjim war-nom the-hell why took-place-mp -nom itpacc believe hazunonai] his-ni 4 turaerareta 1-tte omotteiru no would-not person-dat was-told amp think Q '(lit) phat t5e war took place why]i, John thinks that it was said to the person who would not believe iti ti. 'Why does John think [that it was said $0 [the person who would not believe iti] [that the war took place t]] (70) is created by embedding (59) in a bridge verb complement and L scrambling the CP which contains 'm'. (70) seems to be as good as (69). Given that L-scrambling may not neutralize WCO (cf. Section 2.2. I.), it is neutralized by the same preposing operation as the one in (69), though this is followed by L-scrambling in (70). This prepsing may not be NP-movement to a subject position, since, a Saito(1985) argues, L-scrambling of a subject is prohibited, sks illustrated in (71): (7 1) ?*[[sensou-ga okotta] kotol-gaj John-ga [ti Mary-ni tutaeraretal-to omotkiru. war-nom took-place fact-nom -nom -&it was-told comp thinks '(lit) rhe fact that the war took pla~e]i, John think that ti was said to Mary.' The prepsing of the CP within the embedded clause in (70) must be S-scrambling to a VP- periphery position where no Case is available because of the Case absorption by the . passive morpheme 'rare'. Namely, S-scrambling does not obey the 'last-resort' principle, but it still shows properties of A-movement. Given that M-scrambling has both A- and A'-movement properties, as we saw Section 1, we need to examine whether it is optional or not in both respects, namely, whether it may undergo 'literal reconstruction' or not (optiondity of A'-movement) and whether XP may be M-scrambled to a non-Case position without losing its A-movement property (optionality of A-movement). As for the first pint, let us consider (72-74); (72) John-ga [dono kodomeo sikaril-mo sinakatta. -nom which child-acc scold -MO did-not 'John did not scold any child.' (73) *Dono sensei-@ [John-o sikaril-mo sinakatta which terrcher-norn -act dd -MO did-sot (74) ?[Dono kodomol-oy John-ga [ti sikaril-mo sinakatta. which child -act -nom scold -MO did-not 'John did not scold any child.' In (72), the wh-phrase in the object position is licensed by the scope marker 'mot, which is attached to VP, giving negative polarity 'anyt-like reading (cf. Nishigauchi(l986)). In (73), the wh-phrase in the subject position is not licensed by 'mo', which does not c- command it This is ruled out by the proper binding condition, since the wh-phrase moves to kno' leaving an unbound trace behind at LF. In (74), however, the M-scrambled wh- phrase is licensed by 'mo',outside its c-command domain, though a bit marginally. The relative acceptability of (74). as compared with (73), is explained if we assume that at LF, the M-scmbled wh-phrase is undone before wh- WH-movement takes place. This assumption is also supported by (75): (75) [Senso-ga naze okotta 1-ttei John-ga ti itta no? war-nom why took-place -mmp -nom said Q '(lit) [That the war took place why]j, John said ti.' 'Why did John say [that the war took place t]?' In (75), the CP containing the 'true' adjunct 'naze' is M-scrambled without inducing an ECP violation. As Saito(19$6) argues, this is only explained if we assume that at LF, the CP is literally reconstructed before extraction of 'naze', since the landing site of M- scrambling creates a bamer for the ECP, as we discussed in Section 1.4. Let us turn to the second question: does M-scmmbling show optionality when it has the properties of A-movement? The answer is again 'yes', as shown below: (76) a. * [John-ga sorq-o shuchmitel-irai daremo-ni wary-ga nani-o -nom it-acc asserted -since everyone-by -nom what-ace shital-ttej omow-are.-teim no did -camp think-pass-perf Q '(lit) Since John claimed iti, it has been believed by everyone [that May did whatli.' 'What has it been believed by everyone [that Mq drd t] j, since John claimed iG?' b. ?wary-ga nani-o sib]-ttq [John-ga sorei-o shuchmi te] -ifai -nom what-acc did amp -nom it -ace claimed -since daremo-ni ti omow-areru-yooni natta no everyone-by think-pasamp became Q '(lit) phat NIary did whatli, since John claimed iG, it has been believed ti by everyone. ' In (76a1, the 'so-' expression in the adverbial clause, which I assume is an IP-modifier, fails tg be bound by the CP containing a wh-phrase, a special case of WCO, as we discussed in Section 1.4. In (76), M-scrambling of the CP shows WCO-neutralization, an A-movement-like property. The point here is that Case is not available at the landing site, since the Case-assigning feature of the matrix verb is absorbed by the passive morpheme. The psiblility of MP-movement to the matrix subject where nominative Case is available is also excluded, since as we saw in Section 2.2.2, CP subjects do not exist in Japanese. One might suppose that the scrambling which takes place in (76b) is not M-scrambling, but S-scrambling, whose optionality as A-movement we discussed in Section 2.2.2. This, however, does not seem to be true, as the following shows: (79) * wary-ga naze sinda] -tte [John-ga sorei-o shuchoosite]-irai -nom why died amp -nom it -acc claimed -since daremo-ni ti omow-areru-yooni mtta no everyone-by think-ps-comp became Q '(lit) [That Mary died whyli, since John claimed iG, it has been believed ti by everyone.' 'Whyj has it been believed by everyone [that Mary died tjli, since John claimed iti?' In (77), scrambling of the CP containing the ':rue1 adjunct 'naze' does not show WCO- neutdi~ation, as in (76b) in which the scrambled CP contains an argument wh-phrase. In Section 1.4, we discussed that the lack of WCO- neutralization in these cases is induced by the ECP, and that L- and M- scrambling, but not S-scrambling, shows this kind of ECP effect. We must, thus, conclude that the scrambling involved in (76b) aqd (77) is M- scrambling, and that M-mambling shows it property of A-movement (i.e. WCO- neutralization) even if its landing site is a non-Case position. 2.4. A Solution to the Optionality Problem (outline) The apparent ccntradiction between the non-uniform approach as in Mahajan(1989,lWO) and the uniform approach to scrambling as in Webelhuth(1989) is solved by distinguishing two aspects of movement: geometrical and dynamic aspects. The geometrical aspect of an instance of movement is characterized by the geometrical properties of the positions occupied by the members of the chin created. In particular, the A 1 A-bar distinction is to be made here. For the fonnal characterization sf geometrical aspects of several types of movement, Chomsky1s(1992) minimal domain theory is adopted. The dynamic aspect of an instance of movement, on the other hand, is characterized by the effects caused by that instance of movernene. Following Chomsky(1992), I assume that the effects relevant to economy principles are restricted to morphological ones such as Case and +feature checking. The last resort principle is assumed to be defined dynamically: if movement of a ccntributes to the checking of f3, the movement also contributes to the chechng of a This weakened formulation of the last resort principle allows movement of a without checking motivation only if for any 65, checking of never takes place because of the movement. Thus, assuming that XP- adjoined position is not checked , it follows that scrambling as adjunction to XP is uniform with respect to checking, since no checking takes place. On the other hand, assuming that the AIA-bar distinction is made in the domain-theoretic dimension, it follows that the adjunction site is crucial to determining the type of scrambling. The assumption that an XP-adjoined psi tion is not checked is supported by the nonexistence of a selectional relation between X and a "&-generated "-adjunct, given the generalized transformational view of lexical insertion as singulary operation and move a as binary operation. 2.4.1. Geometrical and Dynamic Aspects of Movement 2.4.2. Other Types of Optional Movement The discussion of scrambling suggests that movement is not so strictly constrained as the last resort principle claims. Rather only the cases where any results toward convergence we gained are so constrained. This pint can be made in other types of movement too. In the following two subsections , wc consider optional short V-movement in English and "semantically vacuous" wh-movement in German imperatives. 2.4.2.1. Short V-movement i~ English: Pesetsky(l989) Though it is often assumed (cf. bonds(??), Pollock(1990),Chomsky(1991)) that nsnauxibiary verbs do not move in English, Besetsky( 1989) shows, convincingly I think, that certain types of verbs do move overtly: (78) a People usually ate apples. b. *People ate usually t apples. (79) a. Fkople usually talk about their countries. b. People talk usually t abut their countries. VP-adverbs may not intervene between V and NP-object, but may intervene between V and PP-object, as shown in (78b) and (79b), respectively. Pesetsky gives nice arguments for an V-movement analysis, which we will not go into here. Pesetsky further shows that this type of V-movement does not pass negation, as shown in (80c): (80) a. People do not usually talk about their countries. b. People do not talk usually t about their countries. c. "People talk not t' usually t about their countries. Though Pesetsky calis the landing site of this type of V-movement p rather than AGR, I assume , following Branigan and Collins(l992), that it is AGRo. Pesetsky accounts for the contrast between (lb) and (2b) in terms of "Case- opacity", a generalization of Pollock's(l990) notion of "theta-opaci ty '! the landing site of this type of V-movement is Case-opaque so that the V loses its Case assigning property if it moves. Thus (lb) is ruled out by Case Filter. (2b), however, has no problem, since it does not have a Case assigning property from the outset. Neat though his account is, it cannot be transferred ts hie framework assumed here, since AGRo does license the case assigning property of the V attached to it. What prevents V-movement in (2b) is the weakness of the features of AGRo, since the procrastinate principle requires that movement take place as late as psible, and weak features ileal not be eliminated by V-movement to have legitimate PF objects. To account for optional V-movement in (2b) with this strict procrastinate principle, we must assume, as BranIgan&Collins(??) do, that verbs with no objects are cptionally strong. ?'his move looks too stipulative. , . Lee us rather assume that the procrastinate principle requires that "effectiven movement take place as late as possible, and that movement is effective iff there is at least one instance of feature-checking toward convergence. Thus, if we assume that intransitive verbs and the associated AGRo lack Case I +-features, then movement of intransitive verb to AGRo dues not result in any fkature-checking lie. it is ineffective). Therefore it does not violate the weaker promutinate principle. It should be also noted that this movement violates the strong last resort but not the weak one, as in the case of scrambling. The point is that what is prevented by economy principles such as the procrastinate and the last rescrt is not just movement as change of position, but rather movement as "action." The possibility of optional short V-movement, thus, depends crucially on the existence of "useless" AGRo, which we should assume to be there only because a verb is downstairs. This point can be made clear by the observation that adjectives with PP objects do not show short &movement unlike verbs. Thus, we have the following minimal pair: (81) a. John made Bill heavily depend on Mary. b. John made Bill depend heavily t on Mary. (82) a. John consider Bill heavily dependent on Mary. b. *John consider Bill dependent heavily ton Mary. Since aijectives never take part in the assignment of structural Case to their object in English, it is natural to assume that there are no AGRo for them23. If the internal structure of adjectival small clauses is (a), as Chomsky(l992) suggests, we should assume that AGRA has weak morphological features which always effectively check the corresponding features of adjectives at LF: 2.4.2.2. Wh-Imperatives in Gennan: Reis and Rosengen(l992) Reis and Rosengen(l992) discusses strange cases of imperatives with wh- movement, which they call wh-imperatives. In German, imperative clauses with indirect questions such as (a) may have another version (85): (84) Sag mir bitte doch mal gleich, wohin Peter gegangen ist Tell me please right away where Peter went (85) Wohin sag mir bitte doch rnd gleich, da.8 Peter gegangen ist Where tell me please ri&t away that Peter went Contrary to appearance, the scope of the wh-phrase is invariant between (84) and (85): They are both imperative clauses with an embedded question. Reis and Rosengen show that wh-imperatives are derived by standard wh-movement to Spec of [+wh] C followed by further wh-movement to Spec of the matrix imperative C. This "semantic vacuousness" of ''1, other languages, however, adjectives may assign structural Case. In fact, in Chapter 4, I will propose that adjectivc~ as well as stative verbs in Japanese take part in Nominative Csse assignment via AGRo. 55 wh-movement seems to be @lel to that of scmbling of wh-phrase out of its scope domain we discussed in 5 2.2. Reis and Rosengen claim that imperatives also allow topicalization of [-wh] phrases to Spec-CP, which behave exactly in the me way as topicalization in declaratives. They, however, point out a crucial difference between declaratives and imperatives: klaratives never allow topicalization of [+wh] to Sp-Cl? (86) a. Sagte er jaldoch, auf wen er er vertrauen kann. He said, i:: whom he wu!d hwe faith. b. *Auf wen sagte er jaldoch, daB er vertrauen kann. In whom he said that he could have faith. To account for the unselectivity of the features of moved phrases in imperatives, Reis and Rmengren suggest that in imperatives, the preverbal position is neither marked [+whj, as it is in wh-interogatives nor [-wh] as it is in declmtives. Reis and Rosengren's suggestion nicely matches with our hypothesis: If there is no feature in C of imperatives, there will be no feature checking, hence, no "action" by movement. This situation allows raiovement of phrases of any type. The standard last resort principle predicts that no movement to the Spec-C in imperative would be possible. 2.4.3. Reformulation of Some Economy Principles in terms of Morphological Effects The abve discusqion of some optional movement suggests that at least some economy principles are to be weakened and formulated with reference to the morphological effects caused by movement operation. We may naturally assume that the checking relations between a Read H and its specifier SP are distinguished with respect to the direction of licensing: checking relations from H to SP and from SP to H. Let us call them checking and counter-checkii relations, respectively. Thus, AGb checks the subject SU with respect to Case and SU counter-checks AGRs with respect to +features in the configuration h~kp SU h~kt AGRs]]. When a movement operation 0 causes as its effects a checking relation El and counterchecking relation E2, let us call El and E2 external and internal effects of 0, respectively. Now the procrastinate principle are reformulated as follows: (87) Minimize internal and external effects before spell-out up to PF-convergence. The self-serving last resort principle is radically reformulated as follows: (88) The external effect of movement operation must be compensated for by its internal effect. I will call (88) the com-pensation principle. 2.5. A Representational Solution of the Landing Site Problem The landing site problem discussed in ? 2. l., seems to support Mahajm's(l990) nonuniform analysis of scrambling, whereaq the optionality problem discussed in $ 2.2. seems to support Webelhuth's (1989) uniform analysis of scrambling. This contradiction, however, seems to be resolved if we distinguish two aspects of the movement analysis: the geometric aspect to be captured by the minimal domain theory, and the dynamic dimension to be captured by the checking theory. Thus, these two problems are claimed to be derived from the two dimensions of the XP-adjoined position, which we assume to be the landing site of scrambling: (i) the XP-adjoined gosibion is representationally both in the minimal domain ~f X and that of the head selecting XP. (ii) the XP-adjoined pi tion is not checked at S-structure. The second problem is solved by (ii) and the assumption that LRP is relative to checking of a feature, namely scrambling vacuously satisfies LRP. The first problem is partially solved by (i) ad (ii), since without the implication of checking, IP- adjoined N13 can be either in the domain of I or of C. 2.5.1. L-Relatedness Chomsky(l992) defines AIA-bar pition in terms of L-relatedness: (89) a. p is an A-position if p is narrowly L-related. b. p is an A-bar-position if p is not L-related. Chomsky leaves open the question whether broadly L-related (adjoined) positions are A- pitions or A-bar positions. The peculiarity of XP-adjoined psi tion becomes clear if we consider it froin a higher node: (90) YP I \ spy Y' 1 \ Y XB I \ a XP I \ specx X' I \ X ZF MIN(MAX(X)) = {ZP, Spec X, a) MIN(MAX(Y)) = {XP, Spec Y, a) a is in the minimal domain of X, since a is contained by XP and no category contained by XP except the projections of X d0minates.a. But a is also in the domain of Y, since a is contained by Y P and no category contained by YP except the projections of Y dominates a. The overl~pping of minimal domain stxms to be useful to characterize types of scrambling. Thus, if we assume that S-scmblling is adjunction to VP and M-scrambling is adjunction to IP, it follows that the S-scmbled phrase is in the domain of V and I, and the M- scrambled phrase is in the domain of I and C. We may, then, slightly revise the definition of AIA-bar-positions in the following way: (92) a. P is L-related to H if P is in the domain of H and H has L-features. b. P-is non-krelated to H if P is in the domain of H and H does not have L- features. According to this definition, the landing site of ?-scrambling is an A-pition, since it is L- related to V and I, whereas the landing site of M-scrambling is simultaneously an A- and A- bar-position, since it is L-related to I and non-1- .related to C. This seems nicely characterize the contrast between S-scrambling and M-scrambling that we observed in $ 2.1.1. 2.5.2. Problems The solution given above, however, faces some problems. First, to account for the fact that L-scrambling behaves only as A-bar-movement, we must consider L-scrambling to be CP-adjunction rather than IP-adjunction, since in the latter case it should behave also as A-movement as M-scrambling does. This works well for the cases where L-scrambling terminates in the matrix clause, since no higher head takes the scrambled phrase in its minimal domain . This analysis predicts that L-scrambling inside a embedded clause behaves as both A- and A-bar-movement, since the scrambled phrase belongs both to the minimal domain of the embedded C and that of the matrix V. This prediction, however, is not borne out: (93) *Mary-ga [hq-o soitsu-no hahaoya-ga [John-ga tj sikattal-to i ttal-to -NOM who-ACC the guy-GEN mo?her-NOM -NOM scolded-that said-that omotteiru. think '(Ii t) Mary thinks who, his mother said that John scolded.' 'Who does Mary think that his mother said that John scolded?' In (93, dare-o (who-ACC), which is L-scrambled within an embedded clause, still fails to neutralize WCO. Second, as observed by Mi yagawa( l990), M-scrambling may behave as either A- or A-h-movement , but not as both simdtaneously: (94) aKarerai-o otagaii-no ryoosin-ga Bill-ni ti shookaisita they-ACC each other-GEN parents-NOM -DAT introduced Them, each other's parents introduced to Bill.' b. Karerq-o Swan-ga [John to Mary];-ni ti shookiisita they-ACC -NOM -DAT introduced Them, Susan introduced to John and Mary c.*Karerq-o ~b@i-n~ ryoosin-ga [John to Maryli-ni ti shmkaisita they-ACC each other-GEN parents-NOM -DAT introduced Them; each other's parents ir~trcduced to John ar,d Mary.' In (944, M-scrambling behaves a9 A-movement, since the plural pronoun karera A-binds the reciprocal atagai. In (94b), M-scrambling behaves as A-bar-Movement, since the binding of John ta Mary by kareru does not result in the violation of Principle C. These two type of binding are not available simultaneous1 y, as the ungrammaticality of (94c) shows. Therefore, attributing both A- and A-bar properties to one position seems to be empirically problematic. 2.6. A Derivational Solution of the Landing Site Problem Although it fails to explain the exclusive disjunction of A/A-bar properties, the previous approach seems to be on the right track in that it predicts the difference beween S- scmbling and M-scrambling while keeping the assumption that scrambling is uniformly XP-adjunction. All we need is to derive in a principled way that overlapping of minimal domains is only potential, and either the higher head Y or the lower head X, but not both, takes the adjoined phrase a in (90) in its minimal domain. In this section, we provide such a solution in terms of the strict cycle. 2.6. I. The Strict Cycle and Adjunction Abandoning postulation of D-structure on conceptual and empirical grounds, Chsmsky(l992) gives a single generalized transformation GT. GT is a substitution operation which targets K, add A, and substitutes K1 for A, forming K*, which must satisfy X-bar Theory. K1 can be either a tree already built by GT, or a phrase contained in K. The former case, which serves the function of phrase stnlcture rules, is a binary operation, and the latter, which corresponds to movement transformation, is a singulary operation. They are illustrated below: (91) a K = [V see] K1- pJP John] K* = [V' see John] b. K = p' was seen John] Kl = [NP John] K* = [IP John was seen t] Chomsky, then, defines the strict cycle for GT: (a) The substitution site A in a GT operation must be external to the target phrase K. This implies that the result of GT K* always includes K as a proper part. As for adjunction operations, either binary or singulary, Chomsky assumes that they do not obey the "extensi~n" version of the strict cycle, since the empirical and conceptual motivation he gives for (92) does not apply to adjunction operations. This assumption seems ts be even necessary if we consider XGmovement as adjunction to XO: When we reach the stage K where V to I is possible, I is already too deeply embedded for V-movement to extend the target phrase. 2.6.2. A Reformulation of the Strict Cycle Now, let us return to the problem of the exclusive partitioning of adjoined phrases. In (%I), an adjoined ph. a representationally belongs to the minimal domain of X and to that of Y: However, we want a to belong to either domain, but not both, in order to explain the distribution of A/A-bar-scraanbling. In the GT framework, this exclusive paQtior, may be obtained derivationally. Suppose that a belongs to the minimal domain of H only if adjunction of a takes place when a projection of H is in process, or, more concretely, immediately after a projection of H is created in the course of (substitutional) GT. Thus, a is in the minimal domain of X if adjunction takes place immediately after Spec X is substituted creating XP, and a belongs to the minimal domain of Y if adjunction takes place immediately after either Y' or YP is created. Hence, overlap of minimal domains becomes impossible, since in GT framework, each operation gives only one projection. But, this conception of domain extension by adjunction seems to have redundancy with the strict cycle, since if we do not want a situation where adjunction of a takes place when a node higher than Y P is in process and a is neither in the minimal domain of X nor that of Y (i.e. scrambling without either A- or A-bar-movement properties), the adjunction operation also must be ordered in the course of GT. So, let us refonnulate the strict cycle in kms of minimal domain exter1sion24: (95) If operation 0 (substitution or abjur.ction) applies to a target K, then only the minimal domain of the head of K is extended. 2.6.3, Anti-reconstruction Effects The "minimal domain extension"version of the strict cycle has interesting consequences with respect to Phe prediction of anti-reconstruction effects in Lebeaux's(l988) sense, illustrated below: ('36) a.*Hei believes the claim that Johni is nice. b.*Hei likes the story that Johni wrote. 24~hns CoUins@.c.) points out to me that (15) is rather closer to the classical strict cyclicity in that more than one operation can be done in one cycle. c.*Whose claim [that Johi is nice3 did hei believe? d. Which story [that 30hi ~vrote] did hq like? (~~ux( 1988: 146)) Adopting Lebeaux's(l988) insight, Chomsky(1992) claims that the difference of rebonstruction effects in (%c,d) is reducible to the difference in the applicability of the strict cjcle to the introduction of the bracketed clauses. Thus, the complement clause must be in'tpoduced cyclically before wh-movement in (96c), but the relative (adjunct) clause may be n?ncyclically introduced after wh-movement. The same contrast is observed in Japanese with scmbling as A-bar-movement: (97) a *Karei- ga wary-ga [[Johni-ni ki- ta] tegmi]-o yondal-to He-NOM -NOM -DAT come-PAST letter-.4CC read-PAST-that it-ta say-PAST 'He said that Mary read a letter that me to John.' b. [[ Johni-fi ki-ta] tegami]j-o kare-ga wary -ga t. yon-da]-to. -DAT come-PAST letter-ACC he-NOM -NOM read-NEG-PAST-that it-la say-PAST The letter that came to John, he said that Mary read.' (98) a*Karq-ga wary-ga [[ Job-ni tegami-ga ki-ta] kotol-o sir-anail-to He-NOM -NOM -DAT letter-NOM come-PAST fact-ACC know-not-that omot-tei-ru think-PROG-PRES 'He thinks that Mary doesn't know the fact that a letter came to John.' b.'!* [[ Johni-ni tegami-ga ki-fa] kotolj-o kq-ga wary-ga tj -DAT letter-NOM come-PAST fact-ACC kare-NOM -NOM sir-anai]-to omot-tei-ru know-not-that think-PROG-PRES The fact that a letter came to John, he thinks that Mary doesn't know.' We may incorporate Lebeaux-Chomsky's account in our fm-ework without abandoning the "minimal domain extension' version of the strict cycle. Even introduction of the relative clause immediately after wh-movement (or scrambling) is cyclic, since it extends the minimal domain of the local C (or the head of adjunction). However, our account predicts, conemy to Lebeaux-Chomsky's, that when a relative clause is more deeply embedded in the prepclsed phrase, it may not be introduced after movement. This prediction seems to be bm out: (W) a *--ga tsuma-ni [[Johni-ni ki-ta] tegamij-o suteru-yooni] rneiji-ta. he-NOM wife-DAT -DAT come-PAST letter-ACC throw away-TO order-PAST 'He told his wife to throw away the letter that came to John.' b. [[Johni -ni h -a Zgamilj-o karq-ga tsuma-ni [ tj suteru-yooni] meiji-ta. -DAT come-PAST letter-ACC he-NOM wife-DAT throw away-TO order-PAST The letter that came to John, he told his wife to throw may.' c.?*[[Johq-ni ki-ta] tegamil-o ~uteru-ywni]j karei-ga tsuma-ni tj meiji-ta. -DAT come-PAST letter-ACC throw away-TO he-NOM wife-DAT order-PAST To throw away the letter that came to John, he told his wife.' In (b), the scrambling of the NP containing the relative clause exhibits anti-reconstruc tion effects. If the complement clause containing the NP is scrambIed, reconstruction effects show up25. 25~oam Chomsky@.c.) points out that the although the contrast between (b) and (c) might be obtained in English to the extent that pposing of infinitival CP is accepted, preposing of an NP whose argument contains a relative clause does not show exwed reconstrudon effects: ti) The award for the book that Johni wrote, hej never received. The same thing happens in Japanese: (ii) [[Johnj -ga kai-ta @on-no hiban]-o Scare,-ga Mary-ni meiji-ta -NOM write-PAST book-GEN aiticism-ACC he-NOM -DAT order-PAST The criticism of the book John wrote,,he ordered Mary (to do).' Interestingly, (ii) minimally contrasts (iii) where the preposed element is not NP but CP: (iii) ?*[[Johni-ga kai-ta ]boa-o hihan-sum]-youni karq-ga Mary-ni meiji-ta -NOM waite-PAST book-ACC criticism-do-to he-NOM -DAT order-PAST Further support to our account is given by the interaction between tense dependency and Principle C under reconstruction. As observed by Ogihara(l989), the present tense of a relative clause can be either dependent on the tense m-commanding it or independent (i.e. denoting the speech time): (100) John-ga [[zitensha-ni not-tei-ru ] otokol-ni monku-o it-&. -NOM bicycle-DAT ride-PROG-PEPES man-DAT complaint-ACC say-PAST 'John complained to the marl who waslis riding the bicycle.' Thus, in (20). the man was riding the bycycle when John complained to him or he is riding now. Consider next the following: (101) [[zitensha-ni not-ki-ru ] otokol-ga [John-ga Mary-ni monku-o bicycle-DAT hide-PROG-PRES man-NOM -NOM -DAT complaint- ACC it-ta ]-to omot-tei-ru. say-PAST-COMP think-PRES The man who is riding the bicycle thinks that John complained to Mary.' (101) implies that the man is riding the bicycle now, but not that he was not riding it when John complained him, since the past tense attached to comphn dm not m-command the present tense attached to ride. This tense dependency also shows up under reconstruction: (102) [[zitensha-ni mot-tei-m ] ot~koli-ni Mary-ga [John-ga ti monku-o bicycle-DAT ride-PROG-PRES man-DAT -NOM -NOM complaint-ACC it-ta ]-to omot-tei-ru. say-PAST-COMP think-PRES To the man who islwas riding the bicycle, Mary thinks that John complained.' To criticize the book John wrote, he ordered Mary .' I have no account for his. (102) may imply that the man was riding tile bicycle when John complained to him.26 Let us now consider the following: ( 103) [[Job-no zitensha-ni not-ki-ru ] ~tok~]j-~~i karei-ga tj monku-o -GEN bicycle-DAT ride-PROG-PRES man-DAT he-NBM cumplaint-ACC i t-ta. say-PA ST To the man who islwas riding John's bicycle, he complained.' In (la), John can take kare as its antecedent showing anti-reconstruction effects, but still the tense of the ~lative clause can be dedepentent on the matrix tense. To accourlt for this, we may assume that the structral condition for tense dependency is "in the domain of' rather than m-command. Thus, the tense of the relative clause is in the domain of the matrix tense even when the relative clause is introduced after M-scrambling. However, tense dependency is in conflict with anti-reconstruction effects when the relevant movement is L-scram bling : ( 104) [[Johni-no zitensha-ni not-tei-ru ] obkolj-ni &-ga wary-ga tj -GEN bicycle-DAT ride-PROG-PRES man-DAT he-NOM -NOM xnonku-0 iw-u]-to omot-tei-ta. complaint-ACC say-PRES-COMP think-PAST 26hterestingly, if the relative clause contains the reflexive zibu. bound from outside, the independent tense reading is impossible whether or not the NP containing the relative dause scrambles: (0 John-ga [[zibuni-no zitensha-ni not- tei-ru ] otokol-ni monku-o -NOM self-GIN bicycle-DAT ride-PROG-PIES man-DAT complint-ACC it-ta say-PAST '(lit.) John complained to the man who is riding self's bicycle.' (ii) [[zibwi-no zi tensha-ni not-tei-ru ] otokolj-ni Mary-ga [Johni-ga 9 self-GEN bicyde-DAT ridePKOG-PRFS man-DAT -NOM -NOM monku-s it-ta]-to omot-tei-ru. complaint-ACC say-PAST-COMP think-PKES To the man who is riding self s bicycle, Mary thinks that 2ohn complained.' To the man who is riding John's bicycle, he thought that Mary will complain.' In (104), the tense of the relative may not be dependent on the matrix pt tense if the naan takes he as its antecedent. To obtain the tense dependency, the relative knse must be in the domain of matrix tense either by reconstruction or in the derived position. The former option is in conflict with the ailti-reconstruction effects on Frinciple C. The latter option is also precluded by the derivational view of (minimal) domain extension, since Gmbling must be non-Lrrelated, that is, in the domain of C, and the relative clause is also in the domain of C if it is adjoined after L-scrambling to obey the strict cycle. Of course, if the pronoun is embedded in a noun phrase and fails to c-command the trace of the L-scmbled phrase, the tense dependency is available by reconstruction, since no Principle C violation would arise, as illustrated in (105): (105) [[Johni-no zitensha-ni not-tei-ru ] ~t~kolj-ni karq-no hahaoya-ga GEN bicycle-DAT ride-PROG-PRES man-DAT he-GEN mother-NOM Mary-ga monku-o iw-u]-to omot-tei-ta. -NOM complaint-ACC say-PRES-COMP think-PAST To the man who islwas riding John's bicycle, his mother thinks that Mary complained.' 2.6.4. AIA-bar Metamorphosis at LF The mechanism given in the previous subsection allows two disjunctive results of adjunction of NP to XP: entering the minimal domain of X or that of the head Y which takes XP in its minimal domain. Either option is equally available with respect to the last resort principle, since both are assumed to be without any external effect and, hence, obeys it vacuously. Whether the landing site of the NP is A or A-bar position depends on whether the host head, whose minimal domain it entered, is a L or non-L category. In VP- adjunction, either host head (i.e. I or V) is a L-category and it shows necessarilly the properties of A-movement. On the other hand, in IP-adjunction, if the host head is I, it behaves as A-movement, and if the host head is C, it behaves as A-bar-movement. Actual dlsributions of A/A-bar scrambling, however, are determined with reference to other principles. hng-distance IP-adjunction (i.e. L-scrambling) must be A -bar- movement, since A-movement out of a tensed clause is prohibited by some locality condition (cf. Mahajan(l990), Saito(l992)). This principle also rules out longdistance VP-adjunction in either option, since boah count as A-iiisvement. What remains to be explained is some pecdiar facts of short-distance IP-adj unction (i.e. M-scram bling) discussed in 8 2.2. The tests for SCO and ECP for adjunct extraction, as opposed to the WCO test, indicate that M-mbling fails to behave as A-movement. The idea 1 would Pike to persue is that this difference reflects that of the levela at which the releva-t principles and conditions apply. Suppose that the SCO condition and the ECP for adjuncts apply at L,F, but that the WCO condition applies at S-structur&7, The impossibility of M- scrambling as A-movement for the former two tests zay follow from FuIl Interpretation, which requires all the elements in syntactic representation be interpreted at LF, and the assumption that XP-adjunction structure is interpretable only as a quantificatlonal structure, namely, A-bar structure. Thus, the landing site of M-scrambling as A-movement at S-structure should become A-bar-position at LF by changing its host head form I to C without actually moving28. This A-bar-to-A rnorphosis may account for the dual character of M- scrambling for reciprocal binding for some speakers if we assume, foilowing Heim, 27~ee Higginbothm(l983) and $ 4.3. fcr discussion that the SCO condition applies at LF, and see Mahajan(l990) and Saito(1992) for the discussion that the WCO condition applies at S-structure. How this distinction is to be treated in Chomsky's(1992) minimalist program, where all the representational conditions apply at the interface levels (i.e. LF and PF), is an interesting question !o be explored in future research. 28~aito(1992) proposes m&anorphosis of the opposite direction (i.e. from A-bar to A) to account for some range of data concerning WCO and reciprocal binding teats which are not totally consistent with ours. See Saito(l992) for details. In this thesis, A-bar-to-A metamorphosis is rather used for the analysis of focus antipassives in myan languages in $3.5.2. Lasni k, and May(??), t!at licensing of reciprocal c~nsists of the binding condition A at S- structure and some interpretational requirements, and that those speakers with non-standard judgements prefer to constant A-binding relation between the antecedent and the reciprocal. 2.7. Scrambling versus QW Any attempt to derive the properties of scrambling only from the representational properties of adjunction, such a3 Webelkuth's(l~9), immediately faces the problem of conflict with rhe theory of QR as IP-adjunction at LF, which has ken generally accepted since May(1977). QR is assumed to be pure At-movement, in that the landing site must be an A'-position and it is not optional but driven by the scopal properties of quantifiers. Therefore, if we maintain our analysis of scrambling as XP-adjunction, we must give up Mayts(1985) analysis of QR as adjunction. It is possible but unrealistic to suppose that there is an abstract non-Lcategory between C and I, and QR moves a. quantifier to its Spec position in order to have some of its features checked, since the number of QPs in a clause is in principle unlimited29. In our derivational approach to adjunction, however, there is a key to a solution. We assumed that M-scrambled (i.e. IP-adjoined) phrase is potentially in the domain of C md I, but derivationally -on& to either, since overlapping of minimal domains is unavailable. This gives the A I A-bar duality of M-scrambling. Thc lack of A-movement- like QR suggests that for some reason, only partition to the minimal domain of C is available in QR It does not seem to be plausible to assume that QR is driven by a weak feature of C to an IP-adjoined position at LF. Rather the driving force seems to come from inside. This komes clear if we adopt the view of qumtifiers as second-order transitive predicates. Thus, in every boy lefr , the universal quantifier every takes the set B of boys -. 29~oam Chomsky@.c.) pints out that movement driven by feature-checking generally exhibits parametric variations among languages with respect to when it applies (i.e. SS or PF), but there is no such parametric variance for QR in that:, that is, no language seems to have oven QR. and the set L of all that left, and return the value true iff I3 is a subset of L. The two "arguments" of a quantifier Q are not realized in ?he same way: the restriction RE of Q is realized as the complement c;' Q , while l.he scope SC is not realized withsn the projections of Q. SC rather dominates Q when it is introduced in the cow of derivation: This configuration itself is not necessarily uninterpretable given an appropriate semantic interpretation system. Rather a reason internal to natural language seems to force QP to move to get SC as its scope. This situation reminds us of another type of movement which is not driven by a standard checking motivation: Larsonian V-movement. Let us consider (2;: If V 1 is a ditmsitive predrcate like put, its theta-role assignment cannot be completed within its maximal projection VPl. Therefore, another "VP-shell" must be necessary for the completion of theta-role assignment If V 1 moves to V2 and V2 does not have any theta-role, V 1 discharges all of its theta-roles (81, 02, 83) within its extended minimal domain (= {a, p, y)). Returning to QR let us suppose that the lexical information of a quantifier Q i,ncludes a grid (ul, d), where a1 is for Q's restriction-role and 02 for Q's scope-role, and that Q may have an internal domain but not external domain for some reason in its maximal projection. Thus, QR is a device to extend the minimal dcmain of Q. A tricky way to do this is to assume that Q liberally projects only up to Q' is ips "D-structure" position, but "regenerates" (cf. van Remsdjik(l9??)) when it adjoins to I?: Though this works for QR without modifying the definitions of Chomsky's(1992) domain theory, it does not for wh-movement, which I will argue is an instance of QR in Chapter 3. So, I would mther like to assume that a2 of Q is discharged strictly external to its maximal projection QP. A possible definition of this relation in terms of domains is as follows: (1 10) The scope of a quantifier must be discharged in the external minimal domain of Q. (1 11) a is in the external minimal domain of a head H iff for some head HI, a and the maximal projection HP of H is the minimal domain of H'. To see how this works, let us consider the following: In (1 12), the head of WhP takes IP in its external minimal domain, since C takes WhY and IP in its minimal domain,and then succeeds in discharging its scope. How about QP? Its head Q may have IP in the minimal externzl domain of Q if and only if it is in the minimal domain of C, that is, if and only if it "A-bar-scrambles" to IP. This is all we needed30. 2.8. A Speculation on Checking and Discharging In the preceding suksection, we assume QR to be similar to hsonian V-movement rather than to NP-movement or V-b1 movement. This characterization is made of course in terns of its motivation or dynamic aspect, but not in terms of the size of the mover, or the geometrical aspect of the landing site, A question arises immediately: what kind of movement is possible in natural language from this dynamic view point? Feature-checlang attd role-discharging are already in the list of possible motivations of movement. However, not all instances of feature-checking are possible motivations for movement. As we will see, checking of' +fkatures or [+wh] alone does not counts as such a motivation for movement, though they may affect when and where movement takes place. The intuition is that +features and [+wh] are inherent for Ws and wh-phrases, respectively, and these pieces of information need not to be aquired in the course of derivation. Thus, whether some feature counl as a motivation for movement or not is relative to the bearer of the feature. Suppose, following Ehizzi(l99?), that the Infl that moved to C in English wh- 3%e definition in (6) also allows us to revive the notion of 1-sequence in May(1985) in a framework where the head of S' is distinguished from the head d S: (9 OP1 and OP2 are in the same Z-sequence iff OP1 and OP2 take the same clausal scope. interrogatives agrees with the wh-phrase in Spec-CP wilh respect to [+wh]. The direction of checking is from the wh-phrase to the Infl (i.e. counter-checking). This means that though [+wh] does not count as a motivation for wh-movement, it counts as a motivation for I to C movement. This is illustmted by the following: (113 *Wa~i glaubst [p du [cpti daB [Ip Hans meiat [~p [mit wem]i Jakob t; gesprmhen hat 11111? WHAT do you believe that Hans thinks with whom Jakob talked? (McDaniel( 1989: (25d))) (1 14) When didi YOU ti leave? The scope maker was in partial wh-movement constructions may not wh-move, since it lacks quantificational force for QR, as we will discuss in 3.5.1.. The periphrastic verb did, on the other hand moves to C for [+wh] checking. Thus, only non-inherent features drive movement. This inherent / non-inherent asymmetry seems to be related to independent / dependent asymmetry in anaphoric reference systems, in the sense that the second part of the pair receives an "interpretation" from the first. The parallelism goes further: many people argue for anaphor-movement (cf. Lebeaux(??), Chomsky(1 Sb), Heim, hni k, and May(??), among others), but nu one argues for antecedent-movement, which would license (3a) with the EF representation (3b): (1 15) a. *Mary showed pictures of himself, to Johni. b. Mary Johni showd pictures of himselfi to ti. If we adopt Willims'(1W) claim that the assignment relation of theta-role to argument is same as the linking relation of anaphor to antecedent in the sense of Higginbotham(l983), we have another asymmetry pair: theia-role 1 argument. Given this, Larsonian V-movement, the motivation of which is theta-role assignment, joins other -es of movement for "interpretationH. As is expected, an NP may not move 'lo get a theta-role: (1 16) *It [vp Johi considered [p ti to be likely that it will rain] The intended derivation of (1 16) is: John is inserted to the embedded Sp-IP, which is neither thefa nor Case psition, to satisfy the extended projection principle, and moves to Spec-VP, to which the external theta-role of consider is assigned at S-structure (and it will move to Spec-AGRo for Case-checking at LF). The four licensing reiations we have assumed as motivations of movement. seems to have the following prspeity: (1 17) The licensee must be uniquely licensed by the licencer, but not necessarily vice versa. In reflexive binding, an antecedent may bind more than one reflexive (John talked to himself stnut himself), but the reflexive may not be bound by split antecedents (*John infmdud Mary to each other's parents). There may be unselective binders (if a man owns a donkey, he always beats it), but there hardly seem to be unselective bindee (example?). Multiple &role assignment is a plausible hypothesis for secondary predicate collstructions (I ate rneat raw), whereas a single 0-role does not seem to license more than one argument. Two quantifiers may share the same IP as their scope in the reading for their independence (Three girls lussed four boys), whereas no quantifier seems to take more than one IP as its scope. As we will discuss in $ 4.2.3., Japanese allows multiple object agreement in Japanese , while Hindi does not. This is because only arguments are to be licensed with respect to the morphoIogica1 realization relation in Japanese, whereas AGRs are also to be licensed by a unique argumentin Hindi. CHARER THREE UNIFORMITY OF CHAIN-STEPS 3.0. Introduction 3.1. Successive Cyclicity and the N~tion of "Potential Stepn Since Ross(1%7), a large number of contributions have been dedicated to the understanding of the nature of locality of syntactic movement. One of the most important concepts to attain descriptive and explanatory adequacy for the thmry of long-distance movement is the succesive cyclicity ,which accounts for apparent non-local relations between the moved element and the tram it left in the original position such as (1) aid (2)in terms of iterative application of local move: ( 1) Who do you think! [tM[that Mary believes [t' [that Bill hates t]]]] The intermediate landing sites used are generally assumed to be relative to the type of movement operation and, more specifically, of the same type as the final landing site. Thus, wh-movement uses Spec-CP positions and NP-movement uses Spec-IP positions. This uniform nature of the landing sites of a chain created by syntactic movement has kn recently given an insightful account by Chomsky and Lasnik(l991) (henceforth, C&L), who attempt to derive Rizzi1s(1990) relativized mininldity from the following simple economy ptinciplel:. 1~zzi's(1990) formulation is as follows: (i) Relativizec! minimality: X a-governs Y only if there is no Z such that a. Z is a typical potential a-governor for Y, (3) Minimize chain links. The crucial assumption for (3) is that snly links to "tentialn landing sites are considered for the calculation of optimal chain formation. Thus formation of a chain CH = (a 1, ..., ai, ai+l, ..., a$ is blocked by formation of another chain CH' = (al, ..., ai, a,, ai+l, ..., an) if aj is in a potential landing site. Uniformity of the landing sites can be derived if only the landing site of the same type are potential one. Thus C6L claim: Conditions quite independent of relativized minimality require that snly heads can move to head positions, and snly elements in A-positions to A-positions. Furthermore, again for independent reasons, XPs can move only to specifier positions, and a ml move only to a pi tion that c-commands it. (Chomsky&Lasnik( 199 1:58)) The above statement gives an impression that whether some position PI is a potential landing site for some category C in some pi tion PO is to be defined strictly locally in the sense that no other category or position is necessary in order to define a potential landing site for C in PO. This impression, however, seems to be false. For example, what prevents movement of NP from A-position depends an the type of the global chain, as illustrated by the following: (4) a. Something is t in the room. b. There is something in the mm. c. What is there t in the room? d. *Something is likely that it is in the room. --- - b. Z cco- Y and does not cad X. (Rizzi(1990:7)) (ii) a. Z is a typical potential govemor for Y, Y in an A-chain = Z is an A specifier c-commanding Y. b. Z is a typical potential governor for Y, Y in an A-bar-chain = Z is an A-bar specifier c-commanding Y. c. Z is a typical potential governor for Y, Y in an xO-chain = Z is a head c-wmmaplding Y. (Rizzi(1990:7)) The NP in the pt-copula position may move to Spec-IP at S-stmcture, as in (4a), or at LF (cf. Chomsky(l986b)), as in (4b). It may, however, also move to Spec-CP at S-structure without being blocked by a "potentialn step to the Spec-IP, as in (4c). On the other hand, a potential step to the intermediate Spec-IP does block the movement of NP to the matrix Spec-IP, as in (46). Thus, the notion of potential step relevant to (3) does not seem to be totally reducible to its local potentiality. "Improper movement" as in (5) seems to be relevant to this problem: (5) *John is possible [t' [t to lave]] A standard explanation of (5) is that the trace t is a variable, since it is locally A- bar bund by t', but it is also A-bound by John in its chain domain (i.e. (John, t' t)), a violation of Principle C. This explanation nicely reduces illegitimacy of a certain type of non-unifonn chain to Principle C, which is independent of the notion of well-kormed chain. This might suggest that binding theory gives, or at least c~ntributes to, the definition of potential landing site. The following example, however, indicates that this does not seem to be the case: (6) "Hei seems to the girl who hates Johni [t to be insane] In (6), the NP-movement of he results in a Principle C violation. Thus, if Principle C contributes to the definition of potential step in that it excludes steps rau1tir.g in its violation from potential steps, the following super-raising example should not violate the shortest step condition (3) but violate just Principle C: (7) *He4 is likely that it seems to the girl who hates John; [t to be insane] Now consider (8): (8) ?*Johni seems to the girl who hates Sohi [t to be insane] The difference between (8) and (6) is that the resulting Principle C violation is milder because of the binder is also the R-expression (cf. Lasnik(??)). If this Principle C violation alsa contributes to the definition of potential step in tire same way, the following should only violate Principle C mildly: (9) *Johi is likely that it seems to the girl who hates Johni [t to be insane] However, (9) is as bad as (8). Therefor Principle C violation may not enter into the definition of potential step. 3.2. Non-uniform Chains and the Compensation Psinciple In this section, I discuss three cases of non-uniform chains, and argue that the uniformity of chain-steps follows from the compensation principle, which I introduced in 5 2.4.3. 4.2.1. A-bar-chain Containing Last Resort Violation Consider the following examples of violation of the last resort principle (cf. Lasnik(??), Chomsky(l992)): (10) a. *John strikes t that it is raining. b. *John seems to t that it is raining. A standard account for ( lOa,b) is that John is already assigned Case in the original psi tion and unnecessary NP-movement is thus to be blwked. The illegitimacy of this step is intact even if it is embedded in well-motivated chain formation: (11) *Whottstrikatthatitisraining? *Who tt seems to t that it is raining? The formation of the A-bar-chain (who, t', t) in (1Oa,b) is motivated by whatever motivates wh-movement, and the links of this chain is even more minimized than the well-formed A- bar chain (who, t) in (12): What makes impossible wh-movement through Spec-IP to Spec-CP, althcugh it does not seem to violate binding principles as the improper movement in (5) dm? In Chapter 2, we introduced the compensation principle, repeated here as (13), in order to account for optional movement such as scrambling: (13) The external effect of movement must be compensated for by its internal effect. We may accounts for (1 1) in terms of the compensation principle if we consider the external effects of the intermediate steps as well as of the final one. Thus, the external effect of the final step in (1 I), which is checking of [+wh] on C, is compensated for by its internal effect, which is standardly assumed to be chechng of [+wh] on the wh-phrase=. 2~ will argue in ???aid Chapter 4 that the internal effect of wh-movement is rather discharge of scope-role of the wh-phrase as a special type of existential qwtifier. I-Iowever, the external effect of the first step, which is checlung of +features on I (or more precisely AGE&), is not compensated for, since no internal effect is produced can who when it is in the subject pition, since it possesses Case. in the original position. Thus, the impxsibility of the first step even embedded in a wh-chain follows from (13). This stepwise application of (13) appears to be problematic to mes with successive cyclic movement such as (1) and (2): (1) Who do you think [tU[that h4ary believes [t' [that Bill hates t]]]] We should assume that the intermediate steps in (1) and (2) have external effects of some sort. Othewise, optional movement to any of them as the final landing site would be possible contrary to fact (14) *You think porn [that Mary hates t]] (15) *It seems [the man to strike t [that it is raining]] Moreover, other languages show direct evidence of the external effects of intermediate steps, For example, McCloskey(l979) argues that each step of successive cyclic wh- movement in irish induces alternation of complementizer forms: Eir siad goN sileann an t-athair Ubposfaidh Sile e say they thinks the father will marry Seila him 'They say that the father thinks that Seila will many him.' b. an fear & Deir siad & sileann an t-athair & bpfaidh Sile - the man say they thinks the father will marry Seila 'the man that they say the father thinks Sheila will marry' (McCloskey(1979:??)) Thus, in (15b), dl the intermediate Cs as well as the find one are aL in stead of goN , which appear when there is no element in its Spec as in (16a). As for successive cyclic NP-movement, Mahajan(l989) argues that each step t~ Spec-AGR induces morphological +feature agreement on a verb, as illustrated in (17): (17) siitaa ayii lagtii thii Sita(f.sg.) come(perf.f.sg) wm(imp.f.sg). be(pst.f.sg.) Sita seemed to have come (Mahajan( 1 W ??) ) The assumption that intermediate landing sites also produce external effects is in conflict with the compensation principle, since the latter requires each external effect must be compensated for by an internal effect, but only the external effect produd by the final step is cornpensated for. The Irish and Hindi data above suggest that not only the intermediate steps produce external effecks, but also they are of the &me type as the final step. Here, we detect uniformity of chain formation with respect to its external effects. We may now solve the problem with the stepwise application of the compensation principle if we assume one internal effect compensates recursively for the external effects of the intermediate steps as well in the following way3: 3~lternatively, one might think that the external effect e of a step s = (ai, aj-1) is to be defined as the effect which is available in ai but not in ai-1 so that all the steps except the first one are externally "ineffective". I will not adopt this "accelation" approach. since it will face a problem in the analysis of German partial wh-movement constraction below. (18) The external effect E of a step s is compensated for by an internal effect EO of a step SO if (i) s = SO, or (ii) the step s' which immediately foilows s is such that the external effect E' of s' is compensated for by EO, and E and E' are of the same type. To illustmte how it works, let us consider the following: (19) John seems [t" to be likely [t' to win]] ., . . .I., (20) members of a chain: John t" steps: s3 s2 external effects: @ 42 internal effects: NOM The external effect @, which overtly realizes on the matrix verb as inflection, is compensated for by NOM, since bth are produced by the step s3(=(John, t")). $2 is also compensated for by NOM, since $3 and +2 are of the same type, is compensated for by NOM, and s3 immediately precedes s2. Given hs extention of the notion of compensation, we may consider a step s 80 be a potential step in a chain CH if adding s to CH does not violate the campensation principle. Successive cyclicity then follows from C&L's shortest step principle (3) and this conception of potential step. In light of the compensation principle, I will now discuss some propflies of chains created by leftward quantifier movement in French as in @la), which is extens~vely studied and called L-mus by Kayne(l973, in comparison with properties of Quantification at a Distance (QAD) as in (21b), which is studied by Obenauer(??): (21) a. I1 a tout consult6s. They consulted everything.' b. I1 a baucoup consultCs. They consulted a lot.' kzzi(1990) claims that brh of the quantifier in (a,b) occupy the Spec-VP position, which he assumes to be an A-bar position, by giving two arguments. First, kzzi observes that these two quantifiers are incompatible: (22) a. *I1 a beaucoup tout consult6s. 'He ate everything a lot.' b. *I1 a tout hucoup tout consult&. Second, &zzi observes that both of them follow a floating quantifier linked to the subject: (23) a. Ils ont tous tout rnang6s. They all everything ate.' b. 11s ont tous hucoup mangCs. They ail ate a lot.' Rzzi assumes, following Sportiche(??), that the floating quantifier occupies the VP-internal subject position, and adopted Manzini(??)'s hypothesis that the VP-internal subject is in VP-adjoined position, which immediakly precedes Spec-VP. This hypothesis is inconsistent with the definition of A/A-bar pition in Chapter 2. But Rizzi's claim that L- tous is A-bar movement is supported by the following WCO tesd: (24) *I1 faut tout, que soni proptietaire leur enleve. 'It is necessary that itsi owner take everythingi away from them.' I leave open the question of how VP-intenma! A-bar position is pclssible, but only exami:~e its properties relevant to the problem of uniformity in chain-steps. Kayne(1975, 1983) show that unlike clitic placement, leftward quantifier movement, which he calls L-taus, can extract elements such as tout (everything) or rieri (nothing) from certain infinitival CP: (25) a. Marie a tout voulu faire. 'Mary has everything wanted to do .' b.*Marie l'a voulu faire Fur-therrnore, L-toru can, for many speakers, extract torit or riert even out of a tensed CP: 4~his test only shows that long-distance L-tous is A-bar movement, and the discussion of the landing site problem of scrambling might suggest that he short-disiance L-tous is possibly A-movement. I have argument for or agziinst it, biit at least it can be said that short-distance L-tous is not Case-driven NP- movement, since as Branigan's(l992) observes, it may not induce past participle agreement, which is induced obligatorily in passives: 0) a. *Yai tous refaits b. *L'Ctudiant a tous faits de ses devoirs. (Branigan(l992: 37)) (ii) Wyne's example] (26) ?Je. veux tout que tu leur enl5ves. 'I want you to take everything (away) from them.' (Kayne(1983, W)) This long distance extractability indicates that L-tow is A-bar-movement. There is, however, a crucial difference between L-tow and WH-movement. With WH-movement, the E?P effect with subject extraction as in (27a) can be nullified by the quelqui rule (cf. Moreau( l971), Kayne( 1976)), as illustrated in (2%): (27) a.*Qui vew-tu que vienne? 'Who do you want that come?' (Kayne(1983,P93)) b, Qui veux-tu qui vienne? 'Whodoyouwmtrhatcome?' (Kayne(1983,P94)) With L-tous, however, the replacement of que by qui in (28b) dm not nullify the ECP effect in (2%): (28) a. *Je veux tout que leur soit enlevC. 'I want everything to be taken away from them.' (Kayne(1983, P93)) b.*Je veux tout qui leur soit enlevd. 'I want everything to be taken away from them.' (Kayne(1983, P93)) Following Kayne(lY83), we assume that the failure of quelqui rule is reducible to the failure of movement through COMP ( or Spec of COMP). The same point is made by Obnauer(??), who observes that L-forrs , unlike wh- movement, does not feed Case assignment to the subject of the embedded clause in ECM constructions: (29) a. * J'avais cru NP Ctre comprehknsi ble. b. Que V NP cru t Ctre mmprehdnsible. c. *Jfavais tous cru t Ctre comprehtnsible. (Obnauer(??: 175) Assuming, following Kayne(??), that the embedded infinitival clause in French is CP and the verb may exceptionally assign accusative Case only to S~K-CP~, it follows that Spec- CP may not be used as an intermediate landing site for L-tous. L-?ous also has an interesting property with respect. to the relativized minimality. As discussed by Obnauer(??), QAD blocks extraction of cornbien : (30) a Combien de problem a-t-elle beaucoup resolu t? 'How may of problems did she solve a lot?' b *Combien a-t-elle beaucoup resolu t problems? 'How many did she solve of problems a Iot?' Rizzi(1990) further fibserves that QAD also blocks extractir., of the manner adverb comment: : (3 1) *Comment a-t-il beaucoup rhlu de problems? 'How did he solve may of the problems t ?' S~lternatively. we may adopt Rizzi's(l990) assumption that only the C in agreement with Spec-CP may exceptionally assign Case to IP-Spec. L-tous , however, does nol seem to show these effect&: (32) a Cumbien de lignes a-t-elle tout laisse traverser t ? 'Mow may of lines did she let everything cross?' b. Combien a-t-elle tout laisse traverser t de lignes1?7 'How many did she let everything cross of lines?' (33) Comment a-t-il tout rdsolu ? 'How did he solve everything?' In C&L1s f'mework, this simply follows from the assumption that movement to C via the landing site of Ltolls is impossible, since only a potential landing site blocks movement when it is skipped. The problem here is that the incompatibility of the step of L-tolu -type i.-d that of wh-movement-type is not derivable from a geon?etrical uniformity requirement on chain- steps, if we assume, follcwing Riz7j1s(1990), that the quantifiers occupy geometrically the same position in QAD and L-lous. The difference between QAD and L-tous may be rather reducible to that of the relations ktwxn the quantifier and the verb. We may assume, following Obnauerf??), that in QAD, the quantifier has some qeantificational relation with the verb, possibly 8-identification of its event slot, in the sense of Higginbotham(??), and indirectly assigns quantificational force to the object, and that in L-tozu , on the other hand, there is no such quantificalional relation between them. Let us further assume that this special binding in QAD is an external effect of the same type as that of abstract A-bar 61 would like to thank Jean-Pierre Koenig for judgements. Amitling to him, () is perfect. and, though (b) is weirder than (a), which is dready weird, the weirdness of (b) does not seem to k another magnitude than (a). agreement between C and its Spec. The incompatibility of L-tous and wh-movement then fol:ows from the compensation principle, since the external effect produced by a step to Spec-CP may n~t be compensated for. since any L-torrs step to Spec-VP does not have an externai effect8. 2 2.3. Partid NP-Movement in there constructions Descriphvely spealang, in existential se!:tences in English such as (I), the expletive there occupies the Case position and its associate occupies a theta-pition: (34) There is [a book on the table]. In cases where the distance between there and its associate is nonlocal as in (2), we may assume that there has moved from the embedded Spec-IP to the matrix one leaving a trace behind? (3.3 There seems [t to be [a bk on the table]]. If there were inserted in the matrix Spec-IP directly, it would violate the Extended Projection Principle (EPP), which applies at S-structure in English, requiring the subject oC infinitival as well as finite clauses: (36) a.It/*O is certain that John will win. b.1 believe [it/*O to be certain that John will win]. This also accounts for Kaynefs(1975) claim that L-tow is optional. g~ng-distance there -movement as in (i) is judged to be marginal at least by some speakers: 0) There sms [t' to be likely [t to be [a book on the table]]] For discussion, see Dresher and Hornstein(??), Aoun(??), and Abe( 1992), among ohers. (37) a. TherePo is a book on the table. b. I believe [therePC to be a book on the table] NP-movement of there 's associate to satisfy EPP in the sentence corresponding to (35), however, is illicit 10: (38) *There seems [a book to be [t on the table]] Lasnik(??) claims that examples like (38) vidate the requirement that the landing site of NP-movement be a Case-position, namely, the Last Resort finciple. A problem for this approach, however, shows up if we consider (39): (39) John believes [Nary to have been criticized t] I0(i) might appear to involve movement of the associate of there to the subject of the small clause in a parallel way to (ii): (i) There is I, someone likely [t to win]] (ii) John considers [p someone likely [t to win] ] a in (i), however, should have different structure from f3 in (ii), since only the AF in (ii) can be extracted: (iii) *How likely to win is there [a someone t]? (iv) How likely to win does John consider [8 someone t]? The structure of a rather seems to be close to that of post-nominal modification such as y in (v): (v) I met [g someone [likely to win]] For the AP may not be extracted out of y either. (vi) *How likely to win did you meet [y someone t] If we adopt Chomsky's(lW1, 1982) Case theory, then the Spec of the embedded IP is not a Case-checking position and Mary must move to the Spec of AGRo at LF. Thus, the Last Resort would also rule out this, contrary to fact. There is, however, a difference between (38) and (39) with respect to their covert processes: at LF, the embedded subject in (38) someone replaces there , whereas the embedded subject in (39) Mary moves on to the Spec-AGRo. Assuming, following Chomsky(l991), that there -replacement is LF affixation, we may consider that its external effect is different from that of movement to vacant Spec-AGRs. Call them r(ep1acement)- type and s(ubstitution)-type. The ungrammaticality of (38) thus follows from the compensation principle: the s-type effect of the S-structure step is never compensated for, since the external effect of the LF step is r-type. On the other hand, (39) is grammatical, since both the external effects are uniformly s-type. (35) is also grammatical, since the external effects of the two covert steps (i.e. succesive replacement of the trace and there ) are uniformly r-typelll2. The uniformity relevant here is not distinguished by the standard A/A-bar distinction, since both replac~ment and substitution steps are A- movemznt. 3.3. Complex Chains In this section, I examine two cases of chains whose steps are to be decomposed into components in some way, and argues that these chains also obey the compensation principle. 3.3.1. Partial Wh-Movement Constructions l~oam Chomsky @.c.) points out to me that tbe step to the trace is unnecessary if we assume that the hace disappears aftcr +feature checking takes place at S-structure, and long-distance exple tive-replacement as adjunction at LF does not violate the Relativid Minimality. In this alternative account. (5) is also trivially satisfied. 12~o munt for the similar pattern with it -expletives we discussed in 8 2.3.2.2.. which is repeated here, we should assume that it is also to be replaced by he CP associate. In this subsection, I discuss some strange properties of partial wh-movement constructions in German mainly based on McDmiel's(l989) data, comparing them with there -bnstructions and scope-reconstruction, and argue that those properties follow from the compensation principle together with the decomposition analysis of wh-movement in Chomsky(1992) and Cheng1s(1991) idea that wh-phrases are not driven by [+wh]. 3.3.1.1. Basic Data from McDaniel( 1989) According to McDiiel( 1989), in some diale~ts of German and Romani, wide- scope WH-construal may be licensed by moving the true WH-operator part~ally to an intermediate Spec of CP and placing scope-markers in all the higher Spec-CP positions upto the one to which the construed scope is associatedl3. Let us consider the following German example: (40) Was glaubst DF du iCP was [IP Hans meint [CP mit wem LIP Jakob t gespmhen hat]]]]] WHAT do you believe WHAT Hans thinks with whom Jakob talked 'With whom do you believe Hans thinks Jakob talked?' (McDani el ( 1989: ??)) In (N), the true WH-phrase mit wem moves to the local CP-Spec and the other higher CP- Specs are occupied by instances of the scope-markers was, which means whut if used as a true WH-phrase. McCanfel gives twc, other versions related to (4r3): (41) Was glaubst PP du [CP mit wem PP Hans meint [CP t' [IP,Jakob t gesprochen hat]]]]] WHAT do you believe with whom Hans thinks that Jakob taked 'With whom do you believe Hans thinks Jakob raked?' (McDani el ( 1989. ??)) 13~c~aniel also discusses pdal movement constmctions with more than one true wh-phrases, which we will not concern here. (42) mit wem daubst [IP du [CP t" [IP Hans meint [CP t' [IP Jakob t gesprden hat]]]]] with whom do you believe that Hans t!!inks that Jakob taked 'Wit! whom do you believe Hans thinks Jakob taked?' (McDaniel( 1% ??)) In (42), the true WH-phrase moves up to the intermediate CP-Spec and a scope-marker occupies the highest CP-Spec. In (42), the true WH-phrase moves to the highest CP- spec, and no scope-marker appears. If, however, the true WE phrase is in situ, the sentence becomes ungrammatical even if all the CP-Spas are occupied by scope markers: (43) * #as glaubst [IP du [CP was PP Hans meint [CP was [IP Jakob mir wern gesprochen hat]]]]] WHAT do you believe WHAT Hans thinks WHAT Jakob taked with whom 'With whom do you believe Hans thinks Jakob taked?' McDaniel extends the notion of chain in a way similar to Chomskyfs(1986) CHAIN, which covers both chams created by NP-movement and ex2letive-argument pairs, so that base generated wh-expletives as well as a moved wh-phrase and its traces count as members of a wh-chain. For McDaniel, the feature [+wh] is distributed to the members of a wh-chain and only the head of the wh-chain is relevant to selectional requirement. Quite interestingly, Thon1ton(l99@) observes cases of partial wh-movement constructions spoken by children in certain stages of acquiring English, as in (44): (45) What do ycu think where this froggy lives? (Kelly 3 ; 1 1) In (43, the true wh-phrase where moves to the Spec of the embedded CP and the scope marker wh~ occupies the Spec of the matrix CP. This clearly shows that Wal wh- movement is a UG available option of wh-chain formation, since those children may not hear adults speak that way. 3.3.1.2. Cornpison of Partial Wh-Construction with There -Constructions Let us kntaiively assume, following McDaniel(l989), that scope markers and the chain created by partial wh-movement constitute a bigger chain, which we may call WH- CHAIN. A WH-CHAIN is characterized in the following way: (46) The whole WH-CHAIN CH = (a,, ..., ai+l, ai, ..., ao) is divided into two parts: (i) MAINCH) = (ai, ..., ao) , where aj is 'Lhe true WH-phrase and ai-1, ..., a0 are ai's traces and i 9. (ii) RE?(CIi) = (an, ..., ai+l) , where a,, ..., a;+l are scope-markers, and n z = i+ 1. Note that in (M), if MAIN(CH) = CH, then the movement is total. Now let us compare the partial WH-movement constructions with there-- constructions, whch involve NP-movement version of partial movement: (47) a. There is [a book on the table]. b. There seems [t to be [a hk on the table]] c. There seems [a' to be likely [t to be [a book on the table]]] We may assume that the "MAIN" for expletive-argument chams are the singleton chain headed. by a man and the "RES" is the part~d chain headed by there.. We may characterize them in the following way: (48) The whole expletive-argument CHAIN CH = (a,, ..., a 1, ad may be divided into two parts: (i) MAIN(CH) = (ao) , where a0 is the argument. (ii) FS?(CH) = (a,, ..., ai+l) , where a,,is die expletive and an+], ..., a1 are the traces of the expletive. The crucial difference between (46) and (48) is that part~al movement applies to MAIN(CI4) in (46) but to RES(CH) in (M), or equivalently, that ncn-movement part is RES(CH) in (46) but MAIN(CH) in (48). Thus, the scope-marker was , unlike there , may not move in parbal wh-movement constructions: (49) * Wmi glaubst [p du [m daB [~p Hans meint [~p [mit wem]i Jakob ti gesprmhen hat 11111? WHAT do you believe that Hans thinks with whom Jakob talked? (McDaniel( 1989:(25d))) On the other hand, MBIN(CH) of there -constructions, unlike parha1 wh-constructions, may not be a non-trivial chain: (50) *There seems [a book to be [t on the table]] In ?3.2.3., I argued that (50) is to be ruled out by the compensation principle, since NP-movement at S-structure and there -replacement at LF are non-uniform with respect to their external effects. The legitimacy of parhal wh-movement corresponding to (33) then implies either that the relation between between scope markers and a true wh-phrase is established interpretively along the line of McDaniel's, or that was -replacement takes place without violatine the compensation principle. I will argue below that the latter is the case. 3.3.1.3. Comparison with Scope-reconstruction Rzzi(1991) points out that partial wh-movement constructions obey Ross's(??) negative island condition, which he redilces to the relativized minimality in Rizzi 1990), aithougk its total movement counter part dm not (51) a. Mit wem glaubst du (nicht), dass Hans t gesprochen hat? With whom do(nlt) you believe that Hans has spoken b. Was glaubst du (*nicht), mit wem Hans t gesprochen hat? WHAT do(n't) you believe with whom Hans has spoken (Rizzi(1991) / McDaniel(1989)) Given Chomsky and Lasnik's(l991) way of deriving the relativized minimality from the economy principle (3) (cf. 8 3.1.), the ungrarnrnaticality of (51b) suggests that rvm - replacement does lake place in LE (3) Mnimine chain-links. Quite similarly, "scope-reconstruction" is also sensitive to wtak islands such as negatiiz and wh-islands (cf. Longobardi(??), Frampton(??)): (52) a. Who do you think that everyone likes? b. Who don't you think that everyone likes? c. ?Who do you won& whether everyone likes? As is well-known, (52a) allows a paired answer (e.g. John, Mary, Bill, Susan ...), but neither (52b) nor (5%) does. Frampton(l991) derives this antecedent government requirement on scope reconstruction from ECP by assuming that only an intermediate trace ean be the target of reconstruction, which he assumes to be an interpretive rule. Thus, all the traces in the embedded clause except the original one in (52b,c) are not antwedent governed and deleted so that scope-rmns truction becon~es impossible. To incorporate Frampton's idea to our analysis of pial wh-movement, we need to assume that wh- expletive replacement takes place at LF. Furthermore, it should tx derived that scope reconstruction is obligatory. Murasugi&Saito(1993) (henceforth, M&S) also derives antecedent government requirement on scope reconstriction from ECP in a way slightly different from Frampton's. based on Chsmskyts(1992) decomposition analysis of wh-movement processes, MLGS claims that wh-movement can be decomposed into two parts: QR part followed by wh-movement part. The QR-part is adjunction of the whole wh-phrase to the local VP, and the wh-part moves the wh-word of the adjoined wh-phrzse to Spec-CP. The head of QR part, which is the wh-phrase minus wh-word, is the offending element in the whole cham if there is an island between it and the final landing site of the wh- movement, since it may not 'be deleted. In light of MM's idea, the antecedent government requirement on he link between wh-expletives and the wh-phrase which partially moved follows if we assume that only the wh-head but not the whole wh-phrase !nay take part in wh-expletive replacement. This assumption is then derivable from the uniformity condition on chain steps if we assume that wh-expletive replacemerlt and adjunction of wh-head to the whole wh-phrase are uniform with respect to their externa! effects14 There are two problems for incarporating M&S to our analysis of partial wh- movement. First, partial wh-movement of an argument XP itself does not show negative i,s!and effectsls: (53) a. Mt wem glaubst du daB Hans nicht meint daD Jakob gesprochen hat? 'With whcm do you believe that Hans does not think that Jakob talked.' b. Was glaubst du mit wern Mans nicht meint daB Jakob gesprcxhen hat? 'WHAT do you believe with whdm Hans does not think that Jakob talked.' c.*Was glaubst du was Hans nicht meint mit wem Jakob gesprochen hat? 'WHAT do you believe WHAT Hans does not think with whom Jakob talked.' Wh-movement over the negation in the intermediate embedded clal~se, whether it is total as in (a) or partial as in (b), is legitimate, but partial wh-movement which does not pass the negation is ruled out. Therefore, only the relation between wh-expletives and the partially moved wh-phrase obeys the negative island condition, or the relativized rninimali ty. In M&S's system, only local VP-adjoined position can be the position where offending QP with the trace of wh-word appear. This problem arises even in scope-reconstruction. Consider the following example given by Frampton(??): (54) HOW many books does Bill need to find out whether Bertrand Russell owned? 14 thank Hubert Truckenbrodt for the judgments. (55) a. Wzt is the number x such that there are x many books that Bill needs to find out whether Bei-trand Russell owned? b. What is the number x such that Bill needs there to be x many books that hc finds out whether Bertrand Russell owned? c. *What is the number x such that Bill needs to finds out whether Bertrand Russcll owned x many books? The wh-phrase may not be scope-reconstructed into island, since its interpretation (c) is *. It inay, however, reconstructed into the intermediate clause. This means M&S1s QR-part should be non-local crossing even a wh-island. The first problem is not a serious one, since M&S1s analysis assumes, but docs not depend on, the Id nature of QR. The second one is more serious: (43), where a11 the Cs are occupied by scope rnarkers and the wh-phrase is in-situ, should be grammatical if M&Sts analysis extends to pama1 wh-movement constructions: (43) *Was glaubst [IP du [CP was [IP Hans meint [CP was [IP Jakob mlt wem gesprochen hat]]]]? WHAT do you believe WHAT Hans thinks WHAT Jakob hked with wi~om 'With whom do you believe Hans thinks Jakob hked?' Adjoin the wh-pfme in-situ to the local VP as the QR-part, and adjoin the wh-word first to the wh-phrase then successive cyclically to the scope markemas the wh-part, the result should be grammatical contrary fact. 3.3.1.3. Cheng(1991) Cheng1s(1991) gives an interesting idea about the motivation of wh-movement, which can be summarized as follows: What forces wh-movement at S-structure in languages such as English is not some strong feature in Co, say [+wh], which "attracts" wh-phrases, as commonly assumed, but, rzther, wh-phrases move to Cg in order to "type" the clause as a wh-question, since these languages lacks wh-particles in Co. On this assumption, wh-movement is the only option to "type" the clause as wh-question via 5,~-Head agreement with Cg, since only wh- phrases are assumed to have the feature [+wh] in these languages. The relevant difference between NP-movement and wh-movement, then, seems to be abstractly expressed in terms of the directionality of licensing: in NP-movement, the recipient of some Case feature moves, whereas in wh-movement, the assigner of [+wh] moves. Or in checking theoretical terminology, we may say that while AGR checks the Case feature of NP in its Spec, the [+wh] of C is counter-checked by the wh-phrase in its Spec. The compensation principle now requires some internal effect to compensate for counter-checking of [+wh]. Discharging the scope-role of the wh-word as existential quantifier seems to be the only internal effect a step in wh-phrase may produce. 'i'his is supported by the impossibility of movement of scope-markers: (49) * Wasi glaubst [~p du [cp daB hp Hans rneint [cp [mit wem], Jakob t; gesprocher, hat 111II? WHAT do you believe that Hans thinks with whom Jakob talked? (McDaniel(1989: (2%))) If a scope-markers is pure realization of of [+wh], then it lacks any quantificational force and it may not move for [+wh] counter-checking bemuse of the compensation principle. 3.3.1.4. Uniformity under Decomposition We may now answer why the landing site of partial wh-movement or QR-part must be Spec-CP rather than IP or VP-adjoined position. Let us consider the following: (56) a. was [wh~i . .. whoj.. .I Q b. WaS ~hO~-[~~~~...tj...] C. whoj-was tj'-Lwhpi ... tj ...I ti Partial movement or QW-part takes place in (a), the wh-word adjoins to the wh-phrase in (b) and the wh-word replaces the scope-marker. Suppose that a wh-phrase is z simple existential quantifier rather than a complex quantifier with two components. This means not only (b,c) but also (a) can be considered as a step of the wh-word. The necessity that the landing of the step (a) be Spec-CP follows from the compensation principle, sine. otherwise the step (b,c) would not k uniform with respect to their external effects. The problem is how to compensate for the external effect of the step (c) by the scope discharge of the step (a), since the compensator must comes later but the step takes place at the beginning. If we adopt Chomsky's(1992) copy-and-deletion theory of movement, then the actual scope-discharge is possible only after the deletion process which follows all the copy processes. We may now consider the deletion as a step, which immediately follows the step (c), produces the discharge of scope-role, and hence can compensate for step (c). 3.3.2. Focus An ti passives in Mayan Languages In this subsection, I analyze focus antipassive constructions in Mayan, and show that the necessary connection between a certain voice morphology and A-bar-m~ement in these constructions is derived from just Case theoretical requirements. The key idea of my analysis is that a certain type of position may have different checking relations with local heads across syntactic levels, and the step which has that position as its landing site changes its external effect accordingly. One important consequence of this analysis is that a non-uniform chain at S-structure becomes a uniform chn at LF, and it obeys the compensation principle. 3.3.2.1. An Overview Basically, Mayan languages are verb initial (either VOS 01. VSO) and have an ergative agreement system (i.e. the objects (patients) of transitives and the subjects of intmitives agree with ABS(0LUTIVE) morphology whereas the subjects (agents) of transitivcs agree with ERG(AT1VE) morphology. As is often found in ergative languages, most Mayan languages have antipassive constructions where a special nlorphology attaches to a transitive verb and "demotes" (or inherently Case-marks) its patient with the result that its agent agrees with ABS. A peculiarity of Mayan languages is that many of them also have focus antipassive constructions which are morphologically and syntactically similar to, but not identical with, antipassives. To avoid confusion, we will call the latter absolutive antipafsives, following Mayanists. Comparison of absolutive and f~vus anti passives is giver, below: AAP FAP (1) a. disappearance of ergative agreement obl. obl. b. extraction of the agent ("fom~er ergative") opt. obl. c. demotion of the patient obl. obl.lopt.limp d. absolutive agreement with the agent obl obl.lcond.limp e. absolutive agreement with the patient imp obl./cond./ imp (abbreviations: opt(ional), obl(i'gatory), imp(ossible), cond(itioned))) The differences between absolutive and focus antipassives are summarized in the following two points: (2) a. Absolutive antipassives are completely intransitive, whereas focus antipassives may retain properties of transitives to some extent (e.g. in Jacaltec, the patient is not demoted and agrees with ABS; in Quiche, the patient is optionally demoted, and if demoted, the agent agrees with ABS, and if not, either the patient or the agent agrees with ABS depending on their +features (see below)). b. Absolutive antipassive morphology "feeds" the extraction of' the agent 3rd pl > 3rd sg This is illustrated below: (12) a. In x-in-il-o-w le achi I ASP- 1A-see-THV-FAP the man 'I saw the man.' b. le achi x-in-il-o-w in the man ASP- 1A-see-THV-FAP me 'The man saw me.' (Davies and Sam-Colop( 1990: 523)) Thus the agent and the patiznt control ths first-person absolutive agreement in (a) aid (b), respectively. A natural explanation of (1 1) in our frrimework , which is based on Davis et. al.'s analysis in spirit, is that ABS abstractly agrees with both the agent and the patient +as a consequence of double checking, but only the higher feature bundle is morphologically spelled-out. Double checking, however, is not a necessary characteristic of the focus antipassives. There are focus antipassives withHdemotion" of the patient in Type 1 Mayan languages. Quiche's focus antipassive with demotion of the patient to dative cleariy shows that double checking by ABS is not involved, since in these constructions, ABS obligatcrily agrees with the agent irrespective of the feature hierarchy. But, extraction of the agent is still required. Adopting Murasugi's(l992) idea that ERG (Tr in her terminology) in ergative languages is [+strong] and licenses the agent to be ergative-marked at S-structure, and ABS (Tense in her system) is [-strong] and licenses the patient only at LF, we may assume that focus antipassive morphology suppresses the Case-checking property of ERG without changing the feature [+strong]. This means that some NP must fill the ERG-spec position, but it cannot stay there, since ERG ibsclf cannot license the NP to be Case-marked there. Therefore, it must move to some Case pition, but ABS is [- strong] and it fails to attract the NP at S-structure. Again, a mixed pasition so!ves the problem: at S-structure, T attracts tie agent via Spec-ERGP and checks its operator feature, whereas at LF, ABS checks the agent w.r.t. its phi-features. Now, let us return to the exceptions to our typological generalization that (i) Type 1 languages have focus antipassives, and (ii) Type 2 languages do not. We have one exaption tc: (i) and two to (ii), illustrated in (13): (13) Exceptions to the typological generalization according to Dayley's (1981) survey: a. Haustec: Type 1 & [-focus antipassive] b. Y ucatec and Ixi1: Type 2 & [+focus antipassive] (13a) is not a problem, since all we have claimed is that if Type 1 languages possess a focus antipassive morphology, they may solve the problem it raises using double-checking. So, this claim vacuously holds with a Type 1 language such as Haustec which iacks the morphology. (13b) is a real problem, since Type 2 languages which lack a mixed position may not solve the problem raised by the focus antipassive morphology. If we look at the alleged focus antipassives in Type 2 languages more carefully, howver, we find a crucial difference between them and those in Type I. In Mayan languages in general, the transitive I intransitive distinction is sometimes made by so-called (in)transitive mode markers. In Type 1 languages, focus antipassives pattern with lexical intransitives, passives, and absolutive antipsives in that they take a intransitive mode marker. Interestingly, even focus antipassives without demotion of the patient in Type 1 languages such as Quiche and Jacaltec take an intransitive mode marker. In our framework, an intransitive mode marker may be taken to indicate that there is only one structural Case licenser (i.e. ABS). In Yucakc and Ixil, however, the "focus antipassives"pttern with transitives and not with intransitives including absolutive anti passives in that they do not take a intnnsi tive mode marker, but a transitive one, even though they do not show ergative agreement morphology, unlike pure trarisitivcs. This is illustrated below: (1 4) Y ucatec focus aiitipassives (from Dayley's (1981) (151)) a. mnu puch-en who,hit-A 1 'who hit me?' b. max puch -e-Q) who hi t-M-A3 'who hit him?' c. max il-ech who see-A2 'who saw you?' According to Dayley(1981), in Yucatec focus antipassives, the verb has no tense, aspect, or mode marking except the phrase final suffix -e (cf. (b)), which is usually used on transitive verbs when no other mode suffix occurs (e.g., in subjunctive). As for Ixil focus antipassives, Ayres(1983) gives three pieces of evidence to show that they retain some properties of transitive constructions: (i) with the aspect marker kat , intransitive verbs require the ernplojment of a phrase-final suffix -i, while monosyllabic transitive verb roots require a suffix -a, and no phrase-final suffix is used with transitive verbs which are not monosyllabic or which bear a derivational suffix, as illustrated bellow: ( 15) a. intransitive verbs: -i kar opoon-i -0 (Ch) ((26)) ASP arrive-M-3A 'iielshelit/they arrived' b. moncsyllabic transitive verb roots: -a kat a- q'os-a (ChkNe) ((27)) ASP 2sE-hi t-M-3sA You hit hirnlherlit/them' c. polysyllabic or derived transitive verbs: no suffix kat w- echb'u-0 (Ch&Neb ((28)) ASP 1sE eat-3A-M 'I ate it' Absolutive antipassives pattern with intransitives, as illustrated in (a), whereas focus antipassives pattern wiih transitives as illustrated (b): (16) a. kat q'os-on-i-0 ASP hi t-AAP-M3A 'Helshelit/they hit' b. in kat q'os-on-0 (Ch&Ne> ((30)) I ASP hit-FAP3A 'I hit itlherlhimlthem' (ii) Similarly, with the aspect marker toq, all intransitive verbs, but no transitive verbs, require a suffix -q: (17) a. intmsitives: -q toq ja7 -oq axh (Ch) ((3 1)) ASP go up-M 2sA 'You are going up' b. transitives: no suffix toq in q'os axh (Ch) ((32)) ASP 1sE hit 2sB 'I am going to hit you' Again absolutive antipassives pattern with intransitives, *vhile focus antipassives pattern with transitives: (18) a. toq q'os on-oq axh (Ch) ((33)) ASP hit AAP-M-2sA You are going to hit' b. in toq q'os-on axh (ch) ((34)) I ASP hi t-FAP 2sA 'I am going to hit you' (iii) Finally, in certain syntactic contexts, and with the aspect marker -nik, only intransitive verbs require the use of ergative agreement morphologies rather than absolutive agreement morphoiog ies: (13) a.nik a- ja7 -e7 ASP 2sE go up 'You are going up' b. nik in-Iun q'os axh (ChlNe) ((36)) ASP 1sE hit 2sA 'I amlwas hitting you' As is expected, the agent of absolutive antipassives controls ergative agreement, as illustrated in (a), while the agent of ergative antipassives controls absolutive agreement: (20) a. nik a- q'os-on s wi7 (ChlNe) ((37)) ASP 2sE hit -AAP on me 'You arelwere hitting me' b. in nik q'os-on axh (ChlNe) ((38)) I ASP hit-FAP 2sA 'I amlwas hitting you' If we keep assuming that an intransitive mode marker is correlated with the lack of ergative Case, then we should conclude that "focus afitipassives" in Y ucatec and Ixil must retain [+Case] ERG and only morphological reali7ation of abstract ergative agreement is blocked. Put it simply, they are "fake" fwus antipassives whose morphology induces only minor change with ERG: something like [+Case, -spellout, +strong]. This is understandable, since [-Case, +strong] ERG necessarily leads to ungrammaticality in these languages, which lack a "mixed" position. These "fake" focus antipassive morphologies still requires extraction of the agcnt, although there is no Case reason. APPENDIX In this chapter, I assume, following Aissen(1992), that Spec-IP (or more precisely, Spec-ABSP) is the landing site of fccus-movement in Type 1 Mayan languages, where the operator counter-checks T with respect to an operaior feature. This, however, contradicts with the analysis of wh-movement as QR in chapter 3, where it is argued that the motivation of wh-movement is h have clausal scope. Therefore, to be consistent, let us suppose that the landing site of wh-movement and focus-movement is Spec-CP in Mayan languages too. To pursue this, we must assume that vacuous I to C takes place at S- structure when an operator moves to Spec-CP, in a way pzallel to subject-aux inversion in Egglish. Thus, the S-structure of operator constructions in type 1 Mayan languages is: (1) CP I \ OF' C' I \ C ABSP I \ 1 \ ABS C ABS' I \ I \ tms TP I \ t~ ERGP 1 \ ERG VP I \ I \ ERG- V tv Thus, 8P (= the agent) in Spec-CP counter-checks T at S-structure and is checked by ABS at LF in focus antipassive constructions in Typ 1 Mayan languages. Furthermore, in a subset of Type 1 Mayan languages such as Quiche, where the patient NF may not be demoted, he patient NP moves to Spec-ABSP at LF. Here, the double-checking function of ABS is localized: the agreement between Spec-CP and ABS, and the agreement between Spec-ABS and tms. Both tvpes of checlung are observed in other languages. In sub;ect- aux inversion constr~ctions, which I assume, following(Chomsky(1986), to be derived by I to C, the subject is checked by the trace of I (or more precisely, AGRs): (2) Where did you t~~b leave? The first type of agreement relation is assumed in kzzi and Rokrts'(1989) analysis of corn;;-:ex inversion in French such as (1): (3) Quel livre Jean a-t-il lu? Which book John has he read The points of R~zzi and Roberts are summarized below: 117 (4) a. The wh-phrase moves to CP-spec. b. V-I moves to C. c. The expletive pronoun, which is base-generated in IP-spec, is incurporated to V- I in C, since I to C movement makes the assignment of Nominative Case to IP-spec impossible. d. The thematic subject, which is base-generated in VP-spec, moves to a pre 1P- spec posi tion (either C'-adjoined pition as a A-bar position, or a second I P-.spec made available after I to C).to which I in C assigns Nominative Case under agreement. The S-structure of (2) is, thus, (2'): (2') [e Quel livre [a Jeani [C' a-t-ilj [[p t'j [vp t; lu I]]]] The agreement relation between Jean and AGRs in the C-I complex corresponds to the agreement between Spec-CP and ABS in Type 1 Mayan languages. What is special in Quich-type Mayan languages is that both the English-type agreement and the French-type agreement are possible. A difference between French and Type 1 Mayans is that in French, there are two positions in agreement with m element of C-I complex, whereas in Type 1 Mayans, Spec-CP does both functicns. This is derived from our theory: in French, both AGRs and [+wh] T are assumed to be strong, and therefore they must be checked at S- structure. This cannot be done using just a single position, since this implies domain- overlapping. In Mayans, where we assume, following Murasugi(1992), ABS is weak, Spec-CP can be in the minimal domain of T at S-structure, and it enters the minimal domain of ABS at LF. Therefore, the whole chain is interpreted as uniform and obeys the compensation principle at LF. CHAPTER FOUR ARGUMENTS FOR [+WH] COUNTER-CHECKING 4.1. Introduction In $3.5. I., we adopted Cheng9s(l992) hypothesis that wh-movement is not driven by [+wh] checking. In this chapter, I will give two supporn for it. In 8 4.2., I will argue that the difference between wh-movement and NP-movement with respect to the extractability in multiple specifier constructions is reducible to the diffelence of their licensers: NP is licensed (or checked) by AGRo, while wh-phm is licensed by (or discharges) some IP as its scope. In $4.3., I will support the claim that C [+wh] is licensed (or counter-checked) by arguing that some strange facts about strong crossover reconstruction are explained by the assumption that wh-absorption takes place obligatorily at LF, and that the assumption itself follows from the assumption that C [cwh] is counter- checking by wh-phrase and the unique licenser requirement (cf. $ 2.8.). 4.2. Multiple Specifier Constructions Though Spec-Head relation is generally assumed to be bijective, constructions exceptional to this assumption are found in a parallel way both in A and A-bar specifier systems: multiple object shift and multiple wh-movement. This parallelism, however, breaks down when we compare their behaviors with respect to Relativized Minimality. This asymmetry is also derivable from the asymmetry in their checking structure: checking vs counter-checking. In $ 4.2.l.,we review Comorovslu's(l986) claim that nonexistence of wh-island effects in Romanian is reducible to its peculiarity of Comp system: Comp (or Spec-C, in our terminology) is multiplicable and can be used either as final or intermediate landing sites. In $4.2.2., as preparation for the following subsection, I give an object shift analysis of nominative object constructions in Japanese, following Mahajan's(1989,W) analysis of object agreement in Hindi. In 9 4.2.3., 1 claim that Spec- AGRo in these corlstructions is multiplicable as final landing sites but never as intermediate. In 8 4.2.4., we exp!ain this asymmetry between multiple wh-movement versus NP-movement by assuming that Relativized Minimality applies from the perspective of the (counter-) checkee's. 4.2.1. Multiple A-bar Specif'iers and Extraction Comorovski(l986) relates two facts about Romanian: (i) it allows m~iltiple WH- movement as illustrated In (I), (ii) it allows rather freely extract violation of wh-island condition, as illustrated (2,3): u ( 1) Cinei cuij % ziceai ca t.j 4-a promis tktj? whoi to-whomj what k you-were saying that ti to-kimj has promised t k tj 'Who did you say promised what to whom?' (Cornorovski( 1986: ( 1))) (2) cu care profesori nu pi cinej tj a promis $ va discuta $? 'With which professor, don't you know whoj tj has promised that he will discuss tj?' (CO~O~-OVS~~( 1986: (4~))) (3) Pentru care cladi vrei sY afli cinej tj nu a decis in& cq va vota tk ti? 'For which paragraphi do you want to learn whq tj has not decided yet whak he will vote ti?' (Comorovski( 1 %: (5~))) Comorovski claims that Romanian allows multiple slots in @ and these slots can be used either as final or intermediate landing sites.(Her arguments to be cited below) 4.2.2. Nominative Objects in Japanese: An Object Shift Analysis 4.2.2.1. Basic Facts As Kuno(1973) observes, the stativity of a verb determines the Case of its object in Japanese. If the verb is [-stati\.e], as in (4a), its object is assigned accusative Case, whereas, if the verb is [+stative], as in (4b), its object is assigned nominative be: (4) a. John-ga hon-o yon-da (koto) -NOM book-ACC read-PAST (fact) '(the fact that) John read a book' b. John-gd-ni nihongo-ga w&x-u (koto)l -NOMI-DAT Japanese-NOM understand-PRES (fact) '(the fact that) John understands Japanese' The object of complex predicates created by the concatenation of cemin types of stative affixes (verbal or adjectival) with nsnstative verbal bases may be assigned either accusative or nominative Case: (5) a. Potential verkal affix: -(rar)e 'can' John-ga nihongo-d-ga hanas-e-n: (koto) -NOM Japanese-ACCI-NOM sm-WT-FSW (fact) '(the fact that) Johq can spk Japanese' vrill not touch issues concerning dative subjects such as in (lb) in this paper. See Kund, 1973), Takerzawa(l!3fF), Ueda(l991), among others. b. Esiderative adjectivd affix: -&I 'want' Boku-ga hon-01-ga yomi-ta-i (koto) I-NOM book-ACCI-NOM read-want-PRES (fact) '(the fact that) I want to read a book' c. Affixes forming 'tought-predicates: -yaw 'easy', -niku 'difficult' Kono teeburu-ga kabin-01-ga oki-niku-i (koto) this table-NOM vase-ACCI-NOM putdifficult-PRES (fact) '(the fact that) this table is difficult to put a vase on' The case amys in (4,s) are schematized in (6): (6) a. SUB-NOM OBJ-ACC V[ -stative ] b. SUB-NOAM OBJ-NOM V [ +stative ] c. SUB-NOM OBJ-ACCI-NOM V[ -slative ] - VIA [ +stative ] The ACC-NOM alternation in (6c) an be accounted for in terms of VeanaIysis", following Kageyama( 1982) and Sugioka( 1984)2: (7) Reanalysis VPIAP VPlAP 1 \ I \ VP \ ---> I VIA [+stative] 2~e may express this "reanalysis" effects in terms of head movement and Government Transparency Corollary in Baker1s(1988) framework 123 I \ \ NP V VIA [-stative] [+stative] Thus, the object is assigned nominative Case, if "reanalysisn applies to the complex verb/ adjective in (6c), and it is assigned accusative Case, otherwise. The structural change induced by (4) is deteckd by the following binding facts: (8) a. Johni-~a Mary-ni Karqlj-o sonkei-sase-rare-ru (kom) -NOM -DAT he-ACC respect-CAUS-POT-PRES (fact) '(the fact that) Johni can make Mary respect himi' b. Johnj-ga Mary-ni kare*ib-ga sonkei-sase-rare-ru (koto) -NOM -DAT he-NOM respect-CAUS-POT-PRES (fact) (cf. Miyagawg 1984)) The difference in grammaticality between (83, 5) is reducible to the difference in the governing category of the, pronoun: the governing category for the accusative-marked pronoun in ($a) is the embedded clause, so it can be bound by the matrix subject. The governing category for the nominative-marked pronoun in (7b), on the other hand, is the matrix clause, since "reanalysis" applies ta the sequence sonkei-sase-rare to license nominative Czse for the object, and this process amounts to monoclawalization. A similar effect is obtained in the scope interpretation of the accusativelnominative object qua~tifier and the stative affix, as illustrated in (9)3: (9) a. John-ga migime-dake-o tsurnur-e-ru, -NOM right-eye-only-ACC close-POT-PRES 'John can close only his right eye.' (i) can > only (John can wink his right eye.) (ii) ?*only > can (It is only his right eye that he can close.) b. John-ga migime-dake-ga tsumur-e-ru. -NOM right-eye-only-NOM close-POT-PRES 'John can close only his right eye.' (i) *can > only (ii) only=.can The accusative-marked quantifier migime-dake-o must be within the scope of the potential affix e, as shown in (pa), whereas the nominative-markpd quantifier miginle-dake-ga must have scope over e. This contrast is also accounted for in terms of "reanalysis" as monoclausalization, since the quantifier must have a clausal scape, but the embedded clause is not available because of the monoclausalization, so it must have as its scope domain the matrix clause, within which the scope of e is contained. 4.2.2.2. The Relevance of Stativity 1 thank Masaki Sano far informing me this sort of scope interaction. See Sane( 1985) for discussion of general scopal properties of duke. We may ask, now, what assigns (or licenses) nominative Case to objeck in stative constructions. Two answers have been proposed in the literature, which are summarized (or restated) in (10): (10) a. Stative predicates assign nominative Case under government. (cf. Kuno( 1973), Kageyarna(1982). Sugf oka( 1984)) b. INFL[+tense] assigns nominative Case to objects under government, when lowered to stative predicates. (cf. Takezawa( 1987)) (10a), an updated restatement of Kunc's(1973) rule-based formulation, straightforwardly accounts for why and when nominative Case is assigned in simple and complex stative predicate constructions we have seen so far, but leaves open the question of why objects of stative predicates and subjects in general share nominative Case. (lob), which is proposed by Takezawg(l987), on the other hand, accounts for the second question straightforwardly, given that INFL [+tense] assigns nominative Case to subjects. it also explains the first question neatly, given the following assumptions: (i)[+stative] predicates may not assign Case. (ii) INFL lowering is possible only when Case is otherwise unavailable to the object NP. Namely, from.(i) and (ii), it follows that only the objects of transitive stative predicates (either simple or reanalyzed complex) can get nominative Case from the lowered INFL[+knse]. Conceptually, therefore, (10c) seems superior to (10a). There are, however, two empirical arguments for (10a). Let us first consider the following: (1 1) a. Kono glnkoo-ga okane-01-ga kai-yasu-i. this bank-NOM money-ACCI-NOM borrow-easy-PRES This bank is easy to loan money from.' b. John-ga okane-01"-ga kan-yasu-i. John-NQLJ money-ACCI-NOM borruw-my-PW '(lit) John is easy t~ loan money.' 'John knows money easily.' (12) a. John-ga hon-01-ga ka-e-ru. John-NOM book-ACCI-NOM buy-POT-PM 'John can buy a book.' b. John-ga hon-o/*-ga kai-u-ru. John-NOM book- ACCI-NOM buy- possi ble-PIES 'It is possible that John will buy a book.' The stative affixes in (1 la) and (1%) license nominative objects while those in (1 lb) and (12b) do not, although the affixes are morphologically closely related in each pair. The minimal difference between them seems to be that the former stative affixes (or the maximal projections headed by them) assign some external theta-role to the subject, whereas the latter stative affixes do not. Thus, we may reduce the impossibility of nominative objects in (1 lb) and (12b) to Burzio's generalization, which permits a verb to assign Case only if the verb assigns external theta-role. This explanation, of course, depends on the assumption that the stative affixes do assign nominative Case. Second, if 1NFL[+tense] assigns nominative Case both to subjects and objects in stative predicate constructions, as Takezawa(1987) argues, then we predict that neither nominative subjects nor nominative objects may appear in [-tense] clauses. This prediction, however, dces not seem to be borne out. To see this, let us first consider (10): (13) a. John-ga [Mary-ol?*-ga kawaiku]omot-ta. -NOM -ACCI-NQM pretty think-PAST '(lit) John thinks Mary pretty.' (cf. Takezawa( 1987)) b.. John-ga Feya-o!*--$a kireini] si-ta. -NOM -ACCI-NOM clean do-PAST 'John made the room clean.' We may attribute the impossibility of nominative subjects in the embedded clauses in (13a, b) to the non-existence of INFL[+tense], following ~akezawa4.. Let us consider (5b) again, repeated here as ( 14): ( 14) Kono teeburu-ga kabin-01-ga oki-niku-i (koto) this table-NOM vase-ACCI-NOM put-difficult-PRES (fact) '(the fact that) this table is difficult to put a vase on' In (14), the object of the complex predicate can be either accusative or nominative because of the affixation of the stative affix niku 'difficult'. If we embed the non-finite version of (14) in an ECM construction such as (13b), we get (15): ( 15) a. John-ga Fono teeburu-o kabin-o oki-nikuku] si-ta -NOM this table -ACC vase-ACC put-difficult do-PAST 'John made this table difficult to put a vase on.' 'kikezawa(l987) gives only think-type ECM constructions such as (13a). I will, however, use m&-type ECM constructions below, since the grammatical contrast in the second type seems to me to be clearer than that in the first type. b. *John-ga mono teeburu- ga kabin-o oki-nikuku] si-ta -NOM this table -NOM vase-ACC put-difficult do-PAST c. ?John-ga mono teeburu-o kabin-ga oki-nikuku] si-ta -NOM this table -ACC vase-NOM put-difficult do-PAST b. *John-ga [kono teeburu-ga kabin-ga oki-nikuku] si-ta -NOM this table -NOM vase-NOM put-difficult dmPAST In (15a), both the subject and the object of the embedded clause are assigned accusative Case by oki 'put' and si 'do', respectively. This case array is independent of the existence of INFL[+tense], so it is no problem. (15b, d) are mled out, irrespective of the Case of the object, since there is not INFL[+tense] which assigns nominative Case to the subject of the embedded clause. The (near) grarnrnatiality of (lsc), however, is a problem to Takezawa's analysis, since the object is assigned nominative Case. (l6a-d) are schematized in (13a-d): (16) a. NP-NOM [NP-ACC [NP-ACC V]-difficult] make b. *NP-NOM [NP-NBM [NP-ACC V]-difficult] make c. ?NP-NONI [NP-ACC [NP-NOM V]-difficult] make d. *NP-NOM [NP-NOM [NP-NOM V]-difficult] make This subject1 object asymmetry of nominative Case assignment in con-finite clauses ((16b) vs. (16~)) seems to show that nominative objects are not licensed by INFL,i+tense].This subject/ object asymmetry of nominative Case assignment in non-finite clauses ((16b) vs. (16~)) seems to show that nominative objects are not licensed by INFL[atense]. 4.2.2.3. The Relevance of Government In the two approaches we just examined, the relevant structural relation for nominative Case assignment to objects is assumed to be that of a head and its complement, that is, government. The inkractians of nominative Case assignment and "possessor ascension", however, seem to indicate that the relevant structural relation is not government. To see this, let us first consider (17) and (18): (17) a. &no hana-ga naga-i. elephant-GEN nose-NOM long-PW 'An elephant's nose is long.' b. bga hana-ga naga-i. elephant-NOM nose-NOM long-PRES (18) a. John-ga zoo-no hana-o tatai-ta. -NOM elephant-GEN nose-ACC hit-PAST 'John hit an elephant's nose.' b. *John-ga 200-0 hana-o tatai-ta. -NOM elephant-GEN nose-ACC hit-PAST 'John hit an elephant's nose.' As Kuno (1973) observes, "possessor ascensionn is possible out of subjects(cf. (17a)), but not out of objects (cf. (18b) in Japanese. We assume, following Shibatani(1977), that the impossibility of "possessor"ascension out of objects is reducible to the condition that a verb, whether simple or complex, may not assign more than one accusative Case. This condition also accounts for the possibility of accusative Case assignment to subjects in causative constructions, as illustrated in (19) and (20): (19) a. John-ga ik-u. -NOM go-PRES 'John goes.' b. Mary-ga [John-d-ni i k]-ase-ta. -NOM -ACCI-DAT go-CAUS-PAST 'May madellet John go.' (20) a. John-ga hon-o yom-u. -NOM book-ACC rad-PRES 'John reads books.' b. Mary-ga [John*-d-ni hon-o yom] -ase- ta. -NOM -ACC/-DAT book-ACC read-CAUS-PAST 'Mary madellet John go.' (cf. Shibatani( 1977)) In (19b), the complex verb consisting sf an intransitive verb and the causative suffix may assign accusative Case to the embedded subject. In (20b), on the other hand, the complex verb consisting of a transitive verb and the causative suffix may not, by virtue of the alleged d condition. In wmplex stative predicate constructions, however, "possessor ascension" takes place in an interesting way: (21) a. Boku-ga sono hon-no syohyoo-01-ga yomi-ta-i. I-NOE4 that book-GEN review-ACCI-NOM read-WANT-PFW 'I want to read that book's review.' b. Boku-ga sono holm-ga syohyoo-01-ga yomi-ta-i. I-NOM that book-NOM review-ACCI-NOM read- WANT-PIES c. XBoku-ga sono hon-o syohyoo-01-ga yomi-ta-i. I-NOM that book- ACC review-ACCl-NOM read-WANT-PRES (22) a, Boku-ga John-no atama-d-ga tatak-e-na-i. I-NOM -GEN head-ACCI-NOM hi t-POT-NEG-PBES 'I cannot hit John's head.' b. Boku-ga John-ga atama-d-ga tatak-e-na-i . I -NOM -NOM head-ACU-NOM hit-POT-MEG-PRES c. *Baku-ga John-o atama-d-ga tatak-e-na-i. I-NOM -ACC head-ACCI-NOM hit-POT-NFS-PRES The "ascending" possessor can be assigned nominative Case, irrespective of the Case of the source possessee NP, as shown in (21b) and (22b). The "ascending" possesp aOr cannot be assigned accusative Case, as shown in (21c) and (22~). The Cae arrays of (21b,c) and (22b,c) are schematized in (23): (23) a. . . . NP-NOM NP-NOM b. . . . NP-NOM NP-ACC C. *. . . NP-ACC NP-ACC d. *. . . NP-ACC NP-NOM The nominative NPs in (21b) and (22b) corresponding to the possessor NPs in (21a) and (22a), respectively, do not seem to be "extra"subjects, since they are not licensed without the stative zffix, as shown in (24b) and (25b): (24) a. Boku-ga sono hon-no syohyoo-o yon-da I-NOM that book-GEN review-ACC read-PAST 'I read the review of that book.' h. *Boku-ga sono hon-ga syohyoo-o yon-da I-NOM that book-NOM review-ACC read-PAST (25) a. Boku-ga John-no atama-o tatai-'a I-NOM -GEN head-ACC hit-PAST 'I hit John's head.' b. *Baku-ga John-ga atama-o tarai-ta. I-NOM -NOM head-ACC hit-PAST Furthermore, the "ascending" possessor may not be the antecedent of the reflexive zibun, which has subject -orientation: (26) Boku-ga Maryi-ga [kanojoi/*zi buni-no oya-no mae-de ] me-o matomo-ni I-NOM -NOM she/*self-GEN parents-GEN front-LOC eyes-ACC directly mi-rare-na-i. s=-POT-NEG-PRES 'I cannot stare at M~i's eyes in front of heri /selfits parents.' (27) a. John-no hahaoya-ga byookida -GEN mother-NOM sick-is "John's mother is sick. " ti. Johni-ga e i/?zi buni-no/?karei-no hahaoya-ga byoolu-da -NOM self-GEN he-GEN mother-NOM sick-is (28) a. Boku-wa John-no hahaoya-ga settoku-deki-nai. -TOP -6EN mother-NOM persuade-can-not "I cannot persuade John's mother. " b. Boh-wa Johni-ga e i/*zibunj-no/?kare-no hahaoya-ga settoh-delu-nai. -TOP -NOM self-GEN he-GEN mother-NOM persuade-can-not We must, therefore, assume that (21b) and (22b) involve "possessor ascension" out of objects and that the "ascending possessor"is assigned nominative Case. What is interesting here is the Case array (23b), where the higher NP is assigned nominative Case and the lower NP is assigned accusative Case. This seems to show that nominative Case assignment tc, objects must be independent of "reanalysisn, which absorbs accusative Case. Thus, if we adopt Yoon1s(1990) theta-theoretically motivated phrase structure for "pssessor ascensionn constructions5, we will have the following schematic representation of (23b): (29) vP2 I VP1 I \ NP2 V'1 NOM I \ NPl V 1 [-stative] ACC (29), however, turns out to be problematic as the representation of (Bb) , if we consider the following: (30) John-ga migimedake-ga mabuta-o tsumur-e-ru. -NOM right-eye-only-NOM lid-ACC close-PO?'-PRE? 'John can close only his iight eyelid.' (i) *can > only 5~oon(19!30) claims that the possessed NP, being a predicate semantically, does not discharge the internal theta-role of the verb, but modifies it by theta-identification in Higginbotharn1s(1985) sense, and that the V' consisting of the verb and the possessed NP assigns the qualified theta-role to the possessor NP. (ii) only > can If (30), which also has the Case array (23b), has the structure (29). then we predict that the "ascending" quantifier (NP2) may be interpreted as being inside the scope of the potential verb (V2), since "reana1ysis"may not take place here in order to assign accusative Case to NPl, and, therefore, the embedded scope must be available to the quantifier by assumption (cf. (9)). (3 1) John-ga migimedake mabuta-o tsumur-e-ru (koto) . -NOM righteye-on1 y lid-ACC close-POT-PEES 'John can close only his right eyelid.' (i) can > only (ii) only > can (32) a. John-ga [[sono-hondake(-o) chyuumonsit~ gakuseij-o sagasiteiru. -NOM that-book-only(-ACC) ordered student-ACC is looking for "John is looking for a student who ordered only that book." b. Sono-hon-dake (?*-o) John-ga chyoomonsita] gakuseil-o sagasiteiru. that-book-only (-ACC) -NOM ordered student-ACC is looking for Cf. Saito( 1985) for base generated vs. derived topics If (29) is theta-theoretically motivated, as we assume, and "reanalysis" does not take place in (29), the quantifier must be forced to move out of VP1 for some reason. The reason is most likely to be a Case-reason. If this is true, then the hypothesis that the stative predicates are nominative Case assigners and the government requirement on nominative Case assignment are in contradiction, since V2 may assign nominative Case to NP2 "exceptionally" if VP1, a non-immediate projection of V1, does not constitute a minimality barrier (cf Chomsky(l%)), and, hence, there is no reason for the quantifier to move. We saw in Section 2 that there are good reasons to assume the former hypothesis. The'-ef~re, we must conclude that the relevant struerural relation is ?lot govemintnt 4.2.2.4. Nominati~e Case Assignment. as Agreement We may now consider that the relevant structural relation is Spec-Head agreement, in stead of government. This, however, contradicts with V RAP internal subject hyptkesis6: the stative predicates which assigns nominative Case have exeernd them-roles, as we saw in Section 2 (cf'.(ll), (12)), so the VP/AP Spec position, to which extenial theta-role is assigned, may not be the nominative-marked psition, to which the ~bject of ike stative predicates moves. We agcin face incompatibility between the hypothesis that srative predicates are nominative Case assigners and the relevat structural relation, Spec- Head agreement, this time. The incompatibiliry, however, readily disappears, if we adopt some version of the Case-theory introduced by Chomsky( 1989) and developed by Mahajan( 1989,1990), among others. The leading idea of the new Case theory is that any structural Case assignment relationship between a Case assigner H and a Case assignee XP must be licensed by some AGR, a spial functional category ,which inherits the Case feature from H and is in Spec-Head agreement relationship with the XP at some level of syntactic representation. Let us assume the following IP-internal phrase structure given in Chomsky(1989) with some minor mdifications: 6~ee Ueda(1991) for some evidence for VPIAP internal subject hypothesis from stative predicate constructions in Japanese. Spec AGRs' I \ TP AGRs / \ T ' / \ AGRoP T I \ Spec AGRo' I \ VP AGRo The licensing condition for nominative Case assignment to object in Japanese can he formulated as follows: (34) [+stative] AGRo licenses nominative Case in its Spec.at S-structure The inheritance of [+stative] from a stative predicate to AGR is established by Head- movement and percolation, as illustrated in (35): (35) a. AGW / \ Spec AGW I \ VPIAP AGR, / \ VP VIA [sstative] verb raising b. AGW I \ Spec AGW I \ VPlAP AGb [+stalive] I\/\ VP tj VilAi AGR, [i-stative] This mechanism expiains the problem raised by (30), repeated here as (36): (36) John-ga migime-dake-ga mabuta-o tsumur-e-ru. -NOM right-eye-only-NOM lid-ACC close-YOT-PRES 'Joh can close only his right eyelid.' (i) *can > only (ii) only > can The relevant part of the S-structure of (3C) is illustrated in (37): NP2 NOM / VPi / \ tNP2 V'1 / \ NP1 V 1 [-stathe] ACC In (37, the quantifier NP2 is already outside the embedded clause to be licensed to get nominative Case, so it will quire only matrix scope at LF, a desirable result. Furthermore, this mechanism solves the conceptual problem we faced in Section 4.2.2.2, that is, there was ro account for why stative predicates assigns the same Case as INFL[+tense] assigns. If we assume that the licensing of' the Case assigned by Tensc is established by AGR, also at S-stmcture, but the Case assigned by [-stative] verbs is to be licensed at LF, we can generalized the nominaiive Case assignmenu licensing in Japanese in the following way: (38) K is nominative iff K is licensed by AGR at S-stnrcmre. 4.2.2.5. Conclusion We have seen that the prcblems raised by nominative Case assignment to objects in Japanese stative constructions is solved neatly by the new Case theory with implementation of the "classical" hypothesis that [+stative] predicates assign nominative Case. This result is particularly interesting, in that the insight of the theory that Case assignment is a subcase of agreement gets support from such a language as Japanese, which is morphologically a fully case-marking language lacking any visible agreement. 4.2.3. Multiple A-Specifiers and Extraction 4.2.3.1. Relativized Minirnality Effects on Objects Shift In 9 4.4.1., we argued that the hypothesis that nominative objects are licensed by stative predicates is supported by (1 1) and (12), repeated below as (39) and (a), respectively (39) a. Kono ginkoo-ga okane-ol-ga kari-yasu-i. this bank-NOM money- ACCI-NOM borrow-eas y -PRES This bank is easy to loan money from.' b. John-ga okane-ol*-ga kan-yasu-i. John-NOM money-ACCI-NOM bornow-easy-PRES '(lit) John is easy to loan money.' 'John borrows money easily.' (40) a. John-ga hon-01-ga ka-e-ru. John-NOM book-ACCI-NOM buy-POT-PRES 'John can buy a book.' b. John-ga hon-olkga kai-u-ru. John-NOM book-ACCI-NOM buy-possible-PRES 'It is possible that John will buy a book.' Our account was that the c~mplex stative predicates in (39b) and (40b) do not assign external theta-role and, hence, fails to assign nominative Case due to Burzio's generalization. Burzio's generalization, however, does not seem to be an axiom of the universal grammar7. To try to derive it from something more fundamental, let us consider the structure of the nominative object versions of (40a,b) schematized in (41a,b), irspectlvdy: In both cases, the object NP moves to the spec of AGW. In (42a), thz moved object receives nominative Case successfully. On the other hand, in (42b), where the subject NP also moves the spec of AGR,P, the object NP fails to receive nominative Case. Therefore, the movement of the subject seems to be crucial to derive this type of Bunio's generalization. I think what is at work here issome version of RePativized Minimality, which is originally proposed by Rizzi1s(1990). The intuition is that the object NP in the spec of AGV blocks the movement the subject NP to the spec of AGkP, since these two spec positions are of the same type8. This approach is supported by the following facts: ' Similar attempts have been made recently by Chomsky(l991. class lectures at MIT) and Marantz(to appear) to derive Burzio's genaalizatiou. 8~ote that Rizzi's(l991) original formalization predicts that PRO in the spec IP position would also block the movement of the object contrary to fact. nom ind abj & passive (42) a. Boku-ga sooyu onna-ni John-o shoohsi-ta-i. -?gd I -nom such woman-DAT -ACC induce-want-PRES -NQM 'I want to introduce John to such a woman' b. Boku-ga sooyuu onna-ni t shookais-are-ta-i. -*ga I -nom such woman-DAT introduce-PASS-want-PFES -NOM 'I want to be introduced to such a woman' (42a) shows that the desiderative adjective -?ui (at least marginally) licenses nominative Case to the indirect object of the verb it attaches to when the verb is active (42b), however, shows that when the verb is psivized, -1ai fails to license to do so. The schematic structures of the nominative indirect object ve~sions of (421,b) are (43a,b), respectively (43) a. [AGRSP Subji ~GR~P Ind-Objj [vp [CRIP PRQ tj Obj V] A] AG&] AGRs] The crucial point is that PRO seems to block the movement of the indirect object only when the PRO is a derived subject as in (43b). Note that in bth (43a,b), -tai assign external theta-role to the matrix subject. Thus, we mnot explain the contrast between them in terms of Burzio's generalization. The same point can be made by the following contrast: (44) a. Boku-ga Mary-ni sooyuu hon-o yom-ase-ta-i -ga -NOM -DAT such book-ACC read-cAUs-w~~~t-PW -NOM 'I want to make Mary read such a book.' b. Bob-ga ~oo>~uu hon-o yom-ase-rare-ta-i -?*ga I-NOM that bmk-ACC read-CAUS-PASS-want-PRESS -NOM 'I want to be made to read such book.' (44a,b) are schematized in (45a,b), respectively. (43 a.h~~s~ Subji ~GR~P Objj [VP [CP PROi [~p [IP Subj tj VJCaus] A] AGRo]AGRs] The movement of the object to the spec of AGRoP is blocked only by the PRO whch is moved to the subjectposition, as in (45b). An interesting point here is that unlike (41 b) and (43b), (45b) is ruled out without violating Path Containment Condition proposed by Pesetsky(1982). Nai~ely the path created by the movement of the object properly contains the path created by the movement of PRO, since PRO asymmetrically c-commands the object at D-structure and the object asymmetrically c-commands PRO at S-structure. Thus, our assumption that (41b), (43b), and (4.%) are ruled out by some version of Relativized mnimality seems to be on the right track. Let us tentatively make the following informal characterization of the relevant Relativized Minimality Efects: (46) Relativized Minimality Effects (RME): A-movement is blocked by an intervening NP in A-specifier position only if the NP is also A-moved. 4.2.3.2. Multiple Specifiers of AGRP When we discussed the interactiens between "possessor ascension" and nominative Case assignment to objects in 6 4.2.2.3., we observed that both the possessor NP and the psessed NP can be assigned nominative Case simultaneously. Let us take the double nominative version of (21 b), repeated here as (47): (47) Boku-ga sono hon-ga syohyoo-ga yomi-ta-i. I-NOM that book-NOM review-NOM read-WANT-PRES 'I want to to read that book's review' (47) now raises the following questions: How is (46) (RME) satisfied in (47)? A plausible hypothesis is that the possessed nominative argument does not count as intervening between the possessor nominative argument and its trace. One way to derive this is to assume that after the whole NP moves to the spec of AGW, "possessor ascensionq' as A- movement moves the possessor NP to the spec of different AGW. In this derivation, the spec of the first AGRoP does not count as intervening, since the spec does not c-commands the trace of the possessor but dominates it. There are, however, two problems to this approach. First, as we will see in the appendix, there is some evidence against movement analysis of "possessor ascension". Second, There are cases where multiple nominative objects are licensed but they are not in possession relation, as illustrated in (48): (48) a. Boku-ga Mary-ni sooyuu hon-o yom-ase- ta-i -NOM -DAT rich book-ACC read-CAUS-want-PIGS 'I want to make Mary read such a book.' b.Boku-ga Mary-ni sooyuuhon-ga yom-ase-ta-i -NOM -DAT such book-NOM read-CAUS-want-PRES c.?Boku-ga Mary-ga suoyuu hon-o yom-ase-ta-i -NOM -NOM such book-NOM read-CAUS-want-PRES d. ?Baku-ga Mary-ga sooyuu hon-ga yom-ase-ta-i -NOM -NOM such book-NOM read-CAUS-want-PRES The subject and object of the complement claw of the causative verb of (48a) can be assigned nominative Case disjunctively , as shown in (48b,c), or conjunctively, as shown in (488). (48d), thus, shows that the two nominative arguments licensed by -tai do not have to be in a possession relation. Another way to explain the nonapplication of RME to (47) and (48d) is to assume that these nominative arguments occupy different spec positions of the same AGR,,P and that thcre is no intervention relation between any pair of the spec positions of the same projection. The second assumption can be formalized as in (49): (49) If a c-commands $, then y intervenes between a and f3 iff (i) a c-commands y, (ii) y c-commands $, and (iii) if a governs y then 0 does not govern a9. (49iii) allows NPl over NP if they are in specifier positions of the same category. The first assumption, of course, violates the standard view that functional categories can project at most one specifier position (cf. Fukui(1986), Fukui & Spzs(l%)), which seems to work well in English and, especially, in Hindi, where at most single NP can be in agreement relation with the verb in both subject and object agreement corstructions (cf. Mahajan(1989,1990)). Ws may not adopt Fukuits(1986) proposal that multiple specs in Japanese are specs of a lexical category rather than a functional category, and lexical categories may project specifiers iteratively but only as daughters of single bar projection. The reason is that, except the multiplicity problem, Japanese nominative object licensing seems to be essentially identical to Hindi object agreement. We may rather relate the difference between Hindi and Japanese in this respect to a more straightforward difference between them, namely, structural Case asigment relation is realized as agreement morphology in Hindi and as case morphology in Japanese. Assuming that each spec-head relation realizes as a single morphology, the difference of morphological realization of double specifiers in Hindi and Japanese can be schematized as in (SUa,b): (50 a. Japanese: NP1-case NP2-case Ii b. Hindi: *NPl NB2 H-agrl-agR We may rule out (50b) by the following condition: 9~f we adopt Rizzi's(1990) conception of Relativized Minirnality as a conditio~l on government in general. (I liii) leads to circularity. We may rather define (1 liii) in terms of spec-head relation which is definable in terms of X-bar theory. 146 (51) Nothing can be the target of morphological realization of abstract agreement more than once. In (%), the morphological realizations of AGR(NP1 ,H) and AGR(NP2,H) are distributed to the NP1 and NP2, respectively. In (Sb), on the other hand, their morphological realizations are accumulated in H, violating (51). (51) predicts that in opposite situations, namely when one NP is in abstract agreement relation with more that one head, only Hindi allows multiple morphologid realizations of them. This seems to be born out, as schematized in (52): (52) a. Japanese: *NPi-casel-case2 [ H 11 H2 b. Hindi: NPi [ ti HI-agr ] H2-agr Mahajan(1989) discusses several cases of (52b), where H1 agrees with NP without assigning Case to it. In Japanese, there seems to be no case where more than one nominative/accusative case are attaciled to one NP, as in (52a). One crucial difference between (50b) and (5%) is that (52.a) becomes grammatical if NPi is assigned on1 y case2, whereas (50b) is ungrammatical even if either agrl or agr2 is dropped. To explain this, we may assume (53): (53 j Agreement for Case must be morphologically realized. Thus, in (52a), asel m be dropped , since AGR(NPi,Hl) is not Case assignment relation, in (50b), on the other hand, neither agrl nor agr2 is dropped, since both AGR(NP1,H) and AGR(NF2,H) are Case assignment relations. Assuming that our discussion of the difference between Hindi and Japanese is on the right track, we may conclude that a functional category may project more than one specifier in Japanese. 4.2.3.3. Multiple Specifiers of AGRP and Relativized Minimality The conclusion in the previous section that a functional category may project more than one specifier position, however, raises a question concerning RME discussed in Section 4.2.3.1. Let us consider the following: (54) NP, AGRl NPj AGR2 ti 9 Suppose that NP, and NPj are assigned Case by AGRl and AGR2, respectively. RME blocks the movement of NP over W, as we have already seen (cf.(41), (49, (43)). Given the assumption that more than one spec position is possible in Japanese, we may have the following alternative derivation: (55) NPi AGRl t'i NPj AGR2 In (53, NP, passes through a specifier position of AGR1. This derivation does not seem to violate RME, since NPj blocks neither the link (t;,tj) nor the link (NPi,Q). The latter link, however, seems to violate the Last Resort Condition (LRC), which prevent any link in NP-movement from a Case-marked position, as we discussed in Chapter 2. Then what about a derivation where NP moves just to a spec of AGR?: (56) does not violate either RME or LRC. (47) and (a) are grammatical instances of this type of derivation: (47) Boku-ga sono hon-ga s yohyoo-ga yomi- ta-i. I-NOM that book-NOM review-NOM read- WANT-PFE? '1 want to to read that book's review' (48d) ?Boku-ga Mary-ga sooyuu hon-ga yorn-ase-ta-i -NOM -NOM such book-NOM rad-CAUS-want-PREC 'I want to make Mary to read such a book.' (39b) and (40b), which we excluded by RME, however, do not seem to.even allow this type of derivation: (39b) * John-ga okane-ga kari-yasu-i. John-NOM money-NOM borrow-easy-PRES '(lit) John is easy to loan money.' 'John borrows money easily.' (40b) * John-ga Iron-ga kai-u-ru. John-NOM took-NOM buy-pssible-PRES 'It is possible that, John will buy a bk.' The schematic S-structure of (39b) and (40b) we assumed was (41 b): The alternative S-structure cf them given by the derivation (56) would be (57): W'nat rules out (57) seems to be the Extended Projection Principle, which requires the subject position of IP (AGRSP) to be filled. Therefore, the ungramrnaticality of (39b) and (ilOb), which we accounted for in terms of Burzio's generalization, is now explained by the conspiracy of RME, LP-C, and EPP. 4.2.4. interpretive Asymmetry and the Relativized Minimality (to be added) 4.3. Strong Crossover Reconstruction In this section, we discuss problems with strong crossover reconstruction, and show that its peculiar property as compared with Principle C reconstruction is naturally accounted for on the assumption that C is counter-checked by a wh-expression in its Spec. 4.3.1. Higginbtham(l983): An Asymmetry between Strong Crossover versus Principle C Chomsky( 1981) claims that strong crossover effects as in (la) are reducible to Principle C Effects as in (Ib):. (58) a.*Whol did he1 see tl? b.*Hel saw Johnl. Thus, Principle C, as stated in (2), applies to the variable t 1 and the name John], both of which are assumed to belong to the same class (Rexpression): (59) R-expressions must be A-free. Mgginbotham(1983), however, claims that strong crossover is not reducible to Principle C, observing that names and variables behave differently in Vmnstruction" contexts: (60) a. mch biography of [which artist] l] do you think he 1 wants to read t? b. mich biography of Picassol] do you think he1 wants to read t? (l-hggin&tl~a~fl( 1%)) According to Higginbotham, (60b) is better than (60a). This contrast Itself might be explained by claiming that (6Oa,b) behave in the same way with respect to Piinciple C reconstruction but (a) additionally violates '.he weak crossover condition. In this account, the contrast is alleged to be reducible to the following one: (61) a. ??[Which biography of [which artistll] do you think his! wife wants to read t? b. [Which biography of Picassol] do you think his1 wife wants to read t? Saito's(l989) following observation, however, excludes this account and supports Higgimbtharn's claim: (62) a. Which book that criticized Jchnl is he1 pissed off at t? b.*Which book that criticized who1 is he1 pissed off at t ? As discussed by many (cf. van Piemsdij k and Williams( l981), Freiden(??), Lebeaux(l~)~, names embedded in relative clauses may fail to exhibit Principle C reconstruction effects, as illustrated in (62a). Strong crossover reconstruction effects, however, show up even in these contexts, as illustrated in (62b). Thus, the contrast in (62a,b) is sharper than that in (61a,b) and does not seem to be reducible to weak crossover: (63) ??Which book that criticized who1 is his1 wife pissed off at t ? figginbotham gives an account for strong crossover in terms of his linking theory, where he adopts, instead of symmetrical coreference relations, asymmetrical linking relations, represented by arrows. Thus, the LF of (64) is represented as follows: ------- --. -.------------ ---------- - -------. ---------- J- I .b. I (64) [which artist] [which biography of t'] do you think he wants to read t] t I To rule out (64), Higginbotham in effect stipulates a principle (65) which already incorporates reconstruction: (65) If a formal variable v is an antecedent of a pronoun P, then P does not c-command (V or) any fomal variable dependent on v. The dependence relation used in (65) is defined as the transitive closure of the containment relation: : (66) X is dependent on Y if (i) Y is contained in an antecedent of X or (ii) for some 2, X is dependent on 2, and Z is dependent on Y. (H's (34)) The antecedence relation used in (66) is, then, defined as the transitive closure of the linking relation: (67) Y is an antecedent of X if X is linked to Y or, for some 2, X is linked to Z and Y is an antecedent. (H's (32)) (64) is, thus, ruled out by (65), since the pronoun he , which takes the formal variable f ' as its antecedent, c-commands the formal variable t which is linked to the phrase conhining t', hence, depends on t'. Pnnciple C effects, on the other hand, are (partially) accounted for by the following condition: (68) If X c-commands Y, then Y is not an antecedent of X (H's (25)). (a), unlike (65), does not incorporate reconstruction effects, and whatever accounts for (weak) reconstruction effects of (68) in (62b), repeated as (69) with linking structure, we may expect the difference between Principle C and strong crossover, since (65) does not apply to (69): - - - - - - - -- - - - -- - - - - - - - - - - - - - - 1 I (69) [Which biography of P~casso] do you think he wants to read t f I 4.3.2. A Problem to Higginbotham's Theory Let CIS consider the following example: (70) Who knows [which book that criticized who11 he1 is pis4 off at t ? I I .- I I a. * I I I b. ?? In the reading of (70a), where who] takes the. embedded scope, who] may not be the antecedent on he], as is expected. However, in the reading of (70b), where it takes the matrix scope, the antecedence relation becomes (marginally) acceptablelo. Their linlung structures are (7 1) and (72), respectively: &---I (71) *who [t knows [[who][which book that criticized t] he is pissed off at t I] 'I It I 'I1 4 ---I 4 ---I (72) ??who who [t knows [[which 'book that criticized t ] he is pissed off at t]] The problem with the condition (65) is that it rules out (72) equally, since it does not take we of the scope of the antecedent of a fonnal variable which is expected to show strong crossover effects. Another type of exception is pointed out by Jim Higginbotharn (p.c.): (73) ??mich book that criticized [each manIl] is her pissed off at t ? lhe marginality seems 1,) be reducible to weak crossover. 154 In (B), each man may take scope over the matrix clause and can be the antedent of he Thus, if the wide scope reading is created by QR, we get a representation in violation of (65) 1 1: ---------------------------*------------------- J. I (74) [each man][which book that criticized t'] is he pissed off' at t t It I Thme exceptions suggest that the reconstmctiof~. effect exhibited in (72) may not be attributed to an inherent property of formal variables, as the condition (65) claims, but mther to a special relation ktween the contained WH-phrase and the WH-phrase containing it. 4.3.3. A Lebeaux-Type Analysis Let us now look at Lebeaiix's(l9t38) account for the following argument I adjunct asymmetry: l~im Higginbotham also points out the following contrast: (i) a. ??[Wtuch book that criticized [which man111 is he1 pissed off at t ? b. *mch book that criticized [who111 is he1 pissed off at t ? This contrast seems to be reducible to D-linked versus non-D-linked wh-phrases in the sense of Pesetsky(1987). If we follow Pesetsky in assuming hat D-linked wh-phrases do not move at LF, as is motivated by the contrast in (ii), (ib) is not a problem to (8): (ii) a.*W did who buy? b. Which book did which man buy? (cf. Pesetsky(l987)) (75) *Who, clainz that John1 is nice did he1 believe? (76) Which sbry that John1 wrote did he1 like? Names contained in complement clauses obey Principle C under reconstruction, as illustrated in (75). Names contained relative clauses, on the other hand, does not, or, following Lebeaux's terminology, they show "anti-reconstruction" effects, as illustrated in (76). Lbeaux claims that relative clauses, unlike coniplement clauses, may be adjoined to the head NPs at S-structure as well as at D-struckre, since they are not required by Projection Principle, and that Principle C applies both at D-structure and S-structure. Thus, there is a good derivation for (76): the relative clause is introduced only after the wh head NP moves satisfying Principle C both at D-structure and S-structure. On the other hand, there is no corresponding derivation for (75), since the complement clause must be introduced at D-structure to obey Projection Principle. A natural way to explain the nonexistence of anti-reconstruction effects in (70a) in Lebeaux's framework is, thus, to suppose that some principle forces the relative clause to be introduced at D-structure when both the contailled and the containing w h-phrases take the same clause as their scope. The relevant relation between the two wh-phrases seems to be an "A-over-AVelation as illustrated in ('77): ( a. Which pictures of who did you buy t ? b.*Who did you buy which pictures of t ? Thus, the containing wh-phrase must move but the contained wh-phrase must be in-situ, in order to obey A-over-A condition, the original formulation of which is given below: (78) A-over- A Condition (Chomsky's(l964)) "... if the phrase X of category A is embedded within a larger phrase ZXW which is also of cakgary A, then no rule applying the category A applies to X (but only to ZXW). To update (78) and avoid Ross's(l%'7) counter-examples, we may consider that the checking features (i.e. [+wh]), rather than the syntactic categories, of X and WW) are relevant to the equivalent class for A-over-A condition. A-over4 condition as stated, however, does not require the smaller phrase % be intraduced at D-structure. We may attempt to derive A-over-A effects from some other principle which have the desired property. For that put-pose, let us look at superiority effects which are formally similar to A-over-A effects in that one of the two wh-phrases "defeatsn the other: (79) a. Who t bought what? b.* What did who buy t? (80) Superiority Condition (Chomsky( 1973)) a. No rule can involve X, Y in the structure ... X...[...Z...WYV...]... where the rule applies ambiguously to Z and Y and Z is superior to Y. b. "... the category A is 'superior' to the category B in the phrase structure if every major category dominating A dominates B as well but not conversely." The A-over-A principle and the superiority are parallel in that they may be suppressed by an additional ~h-~hsel2: (79) a. * W hat did who buy there? b. What did who buy where? (cf. Kayne( 1984)) (80) a. *Who did you buy which pictures of t there? S.??Who did you buy which pictures of t where? A theory of superiority which we may use to account for strong crossover reconstruction effects is hnik & Saitofs(1!392) : (81) a A WH-phrase X in COMP is Odisjoint (operator-disjoint) from a WH-phrase Y if the assignment of the index of X to Y results in the local A'-binding of Y by X. (S-structure) b. If two WH-phrases X and Y are 0-disjoint, then they cannot undergo absorption. A distinctive feature of (81) is that it is formulated as a structural condition for licensing absorption operation at LF. For convenience, we may interpret (81) derivationally as in (82) : (82) a. A wh-phrase X may absorb a wh-phrase Y if X c-commands Y before X moves to COMP. 12~he marginality of (70b) seems to be derived from "specificityn condition. Cf. Chomsky(1973). Fiengo and Higginbotham(l981). See also Diesing(l992) and Mahajan(lW2) for recent accounts for specificity effects. b. Every wh-phrase in-situ must be absorbed. In order to cover A-over-A cases, where the trace of the moved wh-phrase never c- commands the wh-in-si tu, we may collapse the cfistinctio1.i of c-command and domination, as is sometimes suggested for "vertical" binding (cf. Williams(l980)): (83) a. A wh-phrase X may absorb a wh-phrase Y if X c-commands Y or dominates Y before X moves to COMP. b. Every wh-phe in-situ must be absohl. Given (83), we may explain the contrast in (70a,b), repeated as (%,b), in Lebeaux's( 1988) framework: (83) Whw knaws [which book that criticized who11 he1 is pissed off at t ? In (Sa), if adjunction of the relative clause takes place after wh-movement of which book, who1 may not be absorbed by it. If, on the other hand, adjunction of the relative clause takes place before wh-movement, he1 binds wlwl violating hnciple C. Thus, we have no good derivation for (ma). In (83b), the relative clause may be adjoined after wh- movement, since who1 can be absorbed by wh2., which moves after the introduction of the relative clause. First, the requirement that wh-phrases in-situ be either c-commanded or clominakd by a moved wh-phrase is too strong (cf. Watanabe(l991)). Second, pis approach predicis that cases with "vertical" absorption of a wh-in-situ in a relative clause may not show any anti-reconstruction effects. This prediction, however, is not born out: (84) Which book that John1 gave to who2 did hel/*2 like ? In (a), whl is "vertically" absorbed by the moved wh-phrase. So, the relative clause should have been introduced before wh-movement, but still Solut can be the antecedent of he without violating Principle C. This seems to suggest that the strong crossover effect in (84) is not reducible to the standard reconstruction process, since contradictory requirements on (anti-) reconstruction are not satisfied simultaneously, as illustrated in (85,86): (85) John wondered which picture of himself Bill took t (Chomsky(1992:55)) (86) a. Which paper that he1 gave to Bresnan;! did every studentl think that she2 would like? b.*Which pper that he1 gave to Bresnan2 did she2 think that every studentl would like? In (85), the reflexive binding requires the anti-reconstruction of the wh-phrase to obey Principle A, but this make it impossible to get the idiom interpretation for take-picture , which requires reconstruction. In (%a), the reconstruction requirement for the binding of he by everyone and the anti-reconstruction requirement for the obviation of Bresnan from she are simultaneously satisfied, if the reconstruction site is the intermediate SF-CP. In (%a), however, these requirements are contradictory and ungrarnmatidity results. Therefore we should look for another mechanism for non-standard reconstruction effects in A-over-A cases. 4.3.5. Absorption and Functional Wh So far we have not discussed the mechanism of absorption. Let us consider the original version given in Iiigginbotham & May(l981: 49): (87) WH x N' (x)] WH y: N' (y)] -> X, WH Y: N' (x) & N' (Y)] where the variables are those that show and [WH x: N' (x)], [W y: N' (y)] are an adjacent pair of singular or plural wh-phrases. H & M claim that (87) applies optionally, and if it applies, it induces a bijective reading as its distinctive semantic effect when both WHs are singular. A point relevant to our discussion is that H & M claim that (87) does not apply to A-over-A cases like (88): (88) a. Which picture of which person did you see? b. WH! x: x a person] [WH! y: y a picture of x] you saw y In (88b), the structural condition of (87) is violated, since the second wh-phrase contains a variable (x) other than its own (y ). This explains the lack of a bijective reading in (a). We may, however, wonder why absorption is restricted that way, if we look at H Rr M's account for Bach-Peters sentences such as (89) in terms of absorption: (89) Every pilot hit some Mg that chased him. An output of QR and other interpretive rules to (89) is (90): (90) [Every x: x a pilot who shot at it] [Some y: y a Ng that chased him] x hit y Absorption, then, applies to (90) and yields (91): (91) Every x, Some y: x a pilot who shot at it & y a Mig that chased him] x hit y Finally, replacement of pronoums by variables gives: (92) [Every x, Some y: x a pilot who shot at y & y a Mg ihar chased x] x hit y Thus, in this LF derivation, the absorption step ((90) -> (91)) does not violate the structural description in ($7) only because replacement of pronouns by variables follows the application of absorption. A more plausible reason to block application of absorption to (Sb) is that the result of absorption would lose a parallelism between the restriction and the scope: (93) [WH! x, WH! y : x a person & y a picture of x] you saw y In (B), the restriction is a two-place predicate and the scope is a one-place predicate. This may be as odd as vacuous restrictive quantificati~n such as (94): (94) a. 'which mar1 John came. b. WH! x: x a man] John came. In (94b), the LF of (944, the restriction is a one-place predicate and the scope is a zero- place predicate, namely, a saturated claw. We may assume that both (93) and (94b) are ruled out by Full Interpretation in Choms!cy(1986). Now let us return to the strong crossover problem with A-over-A structures. Our tentative conclusion in the previous subsection was that a wh-phrase (or its trace) contained in the moved wh-phrase is "reconstructed" in such a non-standard way that other elements conthined in the moved wh-phrase need not to 5e reconstruck6 (cf.(84)). Curiously, this situation is in contradction with the situation we described to account for the impssi bil i ty of application of absorytion to A-over-A structures: the scope of A-over-A structures lacks the variable of contained wh-phrase, whlch is necessary for absorption to apply. To solve it, one may rather consider, contrary to H & M, that absorption does obligatorily apply to (88b), but the result is not (93) but a representation with a "ranstructed" variable of the contained wh-phe in the scope so as to keep a parallelism between the restriction and the scope13. Thus, the representation must be something like (95): (95) [WH! x, WH! y : x a person & y a picture of x] you saw y-x l3 I do not have an explanation of why (88) lacks a bijective reading. The distribution of bijective readings, however, seems: to be accounted for in terms of more delicate mechanism than H&Mts any way. Consider the following: (0 Which b~yi hdmi~~ which one of hisj sister? (H&M(198 1 :49)) Their explanation of the lack of bijection readings in (i) is that the replacement of the pronoun by a variable makes absorption (87) unapplicable. This explanation, however, contradicts with their treatment of Rach- Peters sentences, where they allow pronoun replacement to apply after absorption, as we saw cbave. A fine grained condition on parallelisms between the restriction and the scope seems to be in order. But, what is the complex object y-x ? A plausible candidate is a functional wh, whch is motivated by the so-called functional readings of wh-questions (cf. Engdahl(l980), Cherchia(l99 1)): (94) a. who does every man love? b. his mother (94b) is a functional answer which would generate a familiar pair-list answer. Thus the representation of (94) is something like (95): (95) f: fa function from men to persons] Every x: x a man] x loves f(x) A functional wh version of (95b) is, then, (%): ('36) [WH! x, WH! f : x a person & fa function from persons to pictures] you saw f(x) Let us now return to (60), repeated as (9'7): (97) Whq knows [which book that criticized who11 he1 is pissed off at t ? I I I I a. * I I I b. ?? The embedded interrogative part of (97a) is: (98) [WH x, WH! f: x a person & f a function from persons to books & f(x) criticized x] he is pissed off at f(x) In (98), the functional wh is necessary in order to have a two-place predicate in the scope. Therefore, he cannot be coindexed with x. (97b), however, can be absorbed without using functional wh, as illustrated in (99): (99) [IW x, WH y: x a person St y a person] y knows w! z: z a book (% z criticizd x] he is pissed aff at z ? The matrix scope of (99) (y lazo\t.s WH! z: z a book & z criticized y] he is pissed of at z ) contains both x and y as free variables without using a functional wh. Hence he earl be coindexed with x without violating strong crossover. A problem for this apprmch is that in (93, the x in the object is bound by the x in the subject in the scope without violating strong crossover. We may r,ct recourse to the difference of the binder, as illustrated by (la)): (100) Who did every man say he loved t? (100) may have the functional reading (101): (101) [WH f: x a function from men to persons] [Every x: x a man] x said he loved x In (101), x is bound by the pronoun, but still strong crossover is not violated. Therefore, we must distinguish the bindees in (95)/(100) versus (96). Let us suppose the following functional determination for the "variablen of functional wh: it is [-anaphor, +pronominal114 when its canonid binder is in A-position, and [-anaphor, -pronominal] l%s is also Cherchia's(l991) assumption. when its canonical binder is in A-bar-position. A canoni~l binder of the "variable" of a functional wh is meant to be what makes the functional reading necessary. Oover To summarize, I have argued that the peculiar process for strong cros, reconstruction effects in A-over-A cases is not reconstructi~n at all, but "construction" of a functional wh structure, which is demanded by the parallelism conchtion on the res!riction and the scope of a binary quantifier, which is the output of obligatory absorption.operation at LF. 4.3.6. Deriving Obligatoriness of Absorption Why, the11, is absorption necessary when more than one wh-phrase have the same interrogative scope except "D-linked" wh-phrases (cf. fn 1.). If we assurne that C checks wh-phrase, then the requirement is that C can check at most one wh-phrase. If, on the other hand, we assume that C is counter-checked by wh-phrase, the requirement is that C can be counter-checked by at most one wh-phrase. The question is: which is more plausible? I think the latter is, since it seems to be generally the case that if there is any uniqueness condtion on either of the arguments of an asymmetrical re1atioLl, X depends on Y, Y tends to be uniquely determined given X, but not vice versa. In reflexive binding, an antecedent may bind more than one reflexive (John talked to himself abut himself), but the reflexive may not k bound by split antecedents (*John intrcduced Mary to each other's parents). There may be unselective binders (if a man owns a donkey, he always beats it), but there hardly seem to be unselective bindee (example?), As we discussed in the previous section, Japanese allows multiple object agreement in Japanese , while Nndi does not. This is because arguments are dependents of the morphological realizatioa relation, whereas AGKs are in Hindi. Thus, if the checking of [+wh] is also an instance of dependency relation, the argument on which 2 uniqueness requirement is imposed seecls to be what the other depends, namely, a unique wh-phrase counter-checks C under agreement configuration. An interesting support comes from Romani, which allows both partial wh-movement and multiple wh-movement , as illustrated in ( a,b), respct.Ively: (102) a.Soi PP niisline [CP [savo filmili [1P o Demiri dikhla ti J]] WHAT do you think which film Demir saw?tj !J]] (McDar,iel( 1989: 569)) b. Kaji hj [IRnisline [CP tj ti so [IP o Demiri dikhla tj ti]]] Where whom do you think that Demir saw? (McDaniel( lm: 600)) Most interestingly, McDaniel observes that two or more wh-phrases with different scopes cannot cooccur, as illustrated in (): (103) *Soi [IP puchlan e chave [CP kaji lcasj [IP ov marja tj ti]]] WHAT did you ask the boy where whom he hit? (McDaniel(l=: 601)) This example clearly shows that multiple wh-phrases in the same Spec-CP must be absorbed and becomes a single unit. References Aissen, J. L. (1992), "Topic and Focus in Mayan," Language, 68. 1: 43 - 80. Aoun, J., N. Hornstein, and D. Sportiche (19 ), "Some Aspects of W~de Scope Quantification," Ayres, G. (1983), "The Antipassive "Voice" in Ixil," IJAL,, 49. 1.20 -45. Branigm, P. (1992), Subjects and Complementizers, Ph.D. Dissertation, MIT. and C. Collins (1992), Verb Movement and the Quotative Construction in English, ms., MIT. Berinstein, A.(1984), Evidence for Multi Attachment in K'ekchi Mayan, Ph.D. dissertation, UCLA. Cheng, L. (1992), On the Typology of Wh-Questions, Ph.D. dissertation, MIT. Chomsky, N. ( 1 986a), Knowledge of language: Its Nature, Origins and Use, Praeger, New Y ork. .(l%b), Barriers, MIT Press, Cambridge. (lWl), Tome Notes on Economy of Derivation and Representation," In R. Frieden (ed), Principles cand Parameikrs in Comparative Grammar, MIT Press Cambridge. .( 1992), A Minimalist Program for Linguistic Theory, MIT Occasional Papers in Linguistics, Distributed by MIT Working Papers in Linguistics. Collins, C. (1992), A Note on Quarltifying into Questions and Relational Readings, ms., MIT. Comorovsh, I.( 1986), "Mu1 tiple Wh-Movement in Romanian," Linguistic Inqzriq, 17.1, i71-177. (1989), Discourse and the Syntax of Multiple Constituent Qiiestions, Ph.D. dissertation, Cornel University. Davies, W.D. and L.E. Sam-Colop (lw), "K'iche' and the Structure of Antipassive," Lunguage, 66.3: 522 - 549. Dayley, J. P. (1981), "Voice and Ergativity in Mayan Languages," Jorrml of Mayan Linguistics, 2. 2: 3 - 81. Ddprez, V. M. (1989), On the Typobgy of Syntactic Positions and the Nature of Chairts: Move a to the specifier of functional projeclions, Ph.D. dissertation, MI'. Frampton, J. (195)1), "Relativized Minimalit, A Review,"The Lirigzristic Review, 8(1), 1- 46 (199 ), The Fine Structure of Wh-Movement and the Proper Fcrmulation of the ECP. Fukui, N.( I%), A Theory of Category Projection and Its Applications, Ph.D. dissertation, MIT Heim, I., H. Lasnik, and R. May( 1991) "Reciprocity and Plurality," Lhtguistic Inquiry, 22, 63-101. I-hggimbtham, J. (1985), "Logical form, Binding, and Nominals," Linguistic Inquiry, 14,395420. Harada, S. (1977), "Nihongo ni Henkei wa Hituyoo da," Gengo 6.10,88-95; 6.11, %- 103. Hoj i, H. (1 BS), Logical Form Cor~traints and Conjigurationul Structures in Japanese, Ph.D. dissertation, University of washington. , f. Miyagawa, and H. Tada (1984), NP-Movement in Japanese, ms., USC, Ohio State University, MIT. Kageyarna, T.(1982), "Word Formation in Japqnese," Lingua ,57,2 15-258. Kayne, R. (1973, French Syntax, MlT Pres, Cambridge. ( 1984), Connectedness and Binurgr Branching, Foris, Dordrec hi. (1989), Facets of Past Participle Agreement in Romance, In Beninca, P. (ed), Dialect Variation in the Theory of Gwrnmar, Foris, Doedrecht. Kawasaki, N.(1991), "Tense, Agreement and Nominative Case in Japanese," In J. D. Bobaljik & T. Bures (eds.), MIT Working Papers irt Linguistics, vol. 14, 13 1 - 146. Etagawa, Y. (1986) Subjects in Japanese and English, Ph.D. dissertation, U Mass. Kroch, A. (1989), "Amount Quantification, Referentiality, and Long Wh-Movement," ms. Universi tjr of Pennsylvania. Kuno, S.(1973), The Structure of the Japmse Language, MIT Press, Cambridge, Mass. Lasnik, H, (1992), "Case and Expletives: Notes toward a Parametric Account," Lirtguistic Inqiry, 23. 3: 381-405. and M. Saito(1992) Move-~Corzditions an its Application and Oictpirt, MIT Press, Cambridge. Li, Y .( 1990), Conditions on Xo-movernent, Ph.D. dissertation, MIT. Mahajan, A. K.(1989), "Agreement and Agreement Phrases,"In I. Laka & A. Mahajan (eds.), MIT Working Papers in Linguistics, vol. 10, 2 17-252. ( 1990), The MA-bar Distinction and Movement Theory, Ph.D. dissertation, MIT. May, R. (1977), The Grarnrnar of Quanti'catiort, Ph. D. dissertation, MIT. (1985), Logical Fom, MIT Press, Cambrige, Mass. McDaniel, D. (1986), Conditions on Wh-Chains, Ph.D. dissertation, CUNY , New Y ork. (1989), "Partial and Multiple Wh-Move~ent," NUT 7, 565-604. Miyagawa, S. (1989), Strxture and Case Marking in Japanese, (Syntax and Semantics; vol22), Academic Press, San Diego. (1991). Case Realization and Scrambling, ms, Ohio State University. Murasugi, K. and M. Saito (1992), Quasi-Adjunct as Sententid Arguments, Talk at Harvard University. Murasugi, K. G. (1992), Crossing anad Nested Paths: I?P Movement in Accusative and Ergative Lungzlages, Ph.D. dissertation, MIT. Oknauer, H-G.(1984-85), "On the Identification of Empty Categories," Tlw Linguistic Review 4: 153 - 202. (19861, "Ddplacer a et A'-Liage Local: Mrivations vs. Representations," In M. Ronat & D. Couquax (eds), h Grmnmire Modulaire, &itions De Minuit, Paris . Pesetsky , D. ( 1988), "Language-Particular Processes and the Earliness Pnnci ple," ms., MIT. Reinhart, T. (I=), Amphora and Semantic Intepretation, University of Chicago Press, (1992), Interpreting Wh-In-Situ, ms. Chicago. Reis, M. and I. Rosengsen (1992), "What Do WH-Imperatives Tell Us About WH- Movemenb?,"NLLT. hemsdijk, H.C. van (1989), "Movement and Regeneration," In Beninca, P. (ed), Dialect Variation in the Theory of Grammar, Foris, Doedrecht. Rizzi, L. ( 1 ma), Relativized Minimality , MIT Press, Cambridge. (1990b), "Speculations on Verb Second," in Grmnmar in Progress, edited by J. Mascar6 & M. Nespor, Foris Publications, Dordrecht. (199 ), Residual Verb Second and Wh-Crierion (199 ), The Definition od A Positions and Improper Movement, Talk presented at MIT 1,inguistics Colloquium. -- and I. Roberts, (1989), "Complex Inversion in French," Probus, 1.1, ppl-30. Rudin, C. (198$), "On Multiple Questions and Multiple Wh Fronting,WUT 6: 445 - 501. Saito, M. (1985) Some Asymmetries in Jnpcrme and their Ilworetical Itnplicatio~is, Ph.D. dissertation, MIT. (1986) "LF Effects of Scrambling," Workshop on Comparative Grammar, Princeton University. (1992) "hng Distance Scrambling in Japanese," Joumul of East Asian Lirrguistics l,69- 1 18. Shlonsky, U. (1991) "Quantifiers as functional heads: A study of Quantifier Float in Hebrew,Wugua, 159 - 180. Tada, H.(1990), "Scrarnbling(s)," Workshop on Japanese Syntax, Ohio State University. Takano, Y. (1992), ProperBinding: A Minimalist Approach, ms., UC Irvine. Thornton, R. (199 ), Ph.D. Dissertation, University of Connecticut. Watanabe, A. (1991), Notes on the Role of AGRo, ms., MIT. Webelhuth, G.(1989), Syntactic Saturation Phenomena mid the Modern Gennartic Lunguagw, Ph.D. dissertation, University of Massachusetts. Yoshimura, N. (1989) "Parasitic Pronouns," Southern California Conference on JapanesefKorean Linguistic, UCLA.