Cedric Boeckx, Norbert Hornstein, Jairo Nunes Control as Movement 2010

This page intentionally left blank C O N T RO L A S M OV E M E N T The movement theory of control (MTC) makes one major claim: that control relations in sentences like ‘John wants to leave’ are grammatically mediated by movement. This goes against the traditional view that such sentences involve not movement, but binding, and analogizes control to raising, albeit with one important distinction: whereas the target of movement in control structures is a theta position, in raising it is a non-theta position; however, the grammatical procedures underlying the two constructions are the same. This book presents the main arguments for MTC and shows it to have many theoretical advantages, the biggest being that it reduces the kinds of grammatical operations that the grammar allows, an important advantage in a minimalist setting. It also addresses the main arguments against MTC, using examples from control shift, adjunct control, and the control structure of “promise,” showing MTC to be conceptually, theoretically, and empirically superior to other approaches. c e d r i c b o e c k x is Research Professor at the Catalan Institute for Advanced Studies (ICREA), and a member of the Center for Theoretical Linguistics at the Universitat Aut`onoma de Barcelona. n o r b e r t h o r n s t e i n is Professor in the Department of Linguistics at the University of Maryland, College Park. ja i ro nun e s is Professor in the Department of Linguistics at the Universidade de S˜ao Paulo, Brazil. In this series 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 r o g e r l a s s : Historical linguistics and language change j o h n m . a n d e r s o n: A notional theory of syntactic categories bernd heine: Possession: cognitive sources, forces and grammaticalization no m i e r t e sc hi k-shi r : The dynamics of focus structure j o hn c o l e ma n: Phonological representations: their names, forms and powers christina y . bethin: Slavic prosody: language change and phonological theory b a r b a r a d a n c y g i e r : Conditionals and prediction cla i re l e f e b v r e : Creole genesis and the acquisition of grammar: the case of Haitian creole heinz g iegerich: Lexical strata in English keren rice: Morpheme order and semantic scope a pril m c m a h o n: Lexical phonology and the history of English m a tth e w y . c he n : Tone Sandhi: patterns across Chinese dialects g r e g o r y t . s t u m p : Inflectional morphology: a theory of paradigm structure j o a n b y b e e: Phonology and language use la uri e b aue r : Morphological productivity t h o m a s e r n s t: The syntax of adjuncts eli z ab e t h c l o ss t r a ug o t t and r i c har d b . d a s h e r : Regularity in semantic change m a y a h i c k m a n n: Children’s discourse: person, space and time across languages d i a n e b l a k e m o r e : Relevance and linguistic meaning: the semantics and pragmatics of discourse markers i a n r ob e r t s and a nna r oussou: Syntactic change: a minimalist approach to grammaticalization d o n k a m i n k o v a : Alliteration and sound change in early English m a r k c . b a k e r : Lexical categories: verbs, nouns and adjectives ca rlot a s. smi t h: Modes of discourse: the local structure of texts rochelle lieber: Morphology and lexical semantics ho lg er di e sse l : The acquisition of complex sentences s h a r o n i n k e l a s and c h e r y l z o l l : Reduplication: doubling in morphology s u s a n e d w a r d s: Fluent aphasia b a r b a r a d a n c y g i e r and eve sweetser : Mental spaces in grammar: conditional constructions hew ba erma n, dunst a n b r own, and g r e v il l e g . c o r b e t t : The syntax–morphology interface: a study of syncretism m a r c u s t o m a l i n : Linguistics and the formal sciences: the origins of generative grammar s a m u e l d . e p s t e i n and t . d a ni e l se e l y : Derivations in minimalism p a u l d e l a c y : Markedness: reduction and preservation in phonology y e h u d a n . f a l k : Subjects and their properties p . h. mat t he ws: Syntactic relations: a critical survey m a r k c . b a k e r : The syntax of agreement and concord g i l l i a n c a t r i o n a r a m c h a n d : Verb meaning and the lexicon: a first phase syntax p i e t e r m u y s k e n: Functional categories j u a n u r i a g e r e k a: Syntactic anchors: on semantic structuring d . r o b e r t l a d d : Intonational phonology, second edition leonard h. babby : The syntax of argument structure b. elan d resher: The contrastive hierarchy in phonology d a v i d a d g e r , d a n i e l h a r b o u r , and laurel j. watkins : Mirrors and microparameters: phrase structure beyond free word order niina ning z hang : Coordination in syntax nei l s mi t h: Acquiring phonology n i n a t o p i n t z i: Onsets: suprasegmental and prosodic behaviour c e d r i c b o e c k x , norbert hornstein, and jairo nunes: Control as movement Earlier issues not listed are also available CAMBRIDGE STUDIES IN LINGUISTICS General editors: p. austin, j. bresnan, b. comrie, s. crain, w. dressler, c. j. ewen, r. lass, d. lightfoot, k. rice, i. roberts, s. romaine, n. v. smith Control as Movement . Brazil .CONTROL AS MOVEMENT CEDRIC BOECKX ICREA/Universitat Aut`onoma de Barcelona N O R B E RT H O R N S T E I N University of Maryland. College Park JA I RO N U N E S Universidade de S˜ao Paulo. Cambridge CB2 8RU. Tokyo Cambridge University Press The Edinburgh Building.cambridge.cambridge. Cape Town. Madrid. no reproduction of any part may take place without the written permission of Cambridge University Press. Delhi. First published in print format 2010 ISBN-13 978-0-511-78955-7 eBook (NetLibrary) ISBN-13 978-0-521-19545-4 Hardback Cambridge University Press has no responsibility for the persistence or accuracy of urls for external or third-party internet websites referred to in this publication. São Paulo. . Norbert Hornstein. Subject to statutory exception and to the provision of relevant collective licensing agreements. accurate or appropriate. New York. and does not guarantee that any content on such websites is. Singapore. UK Published in the United States of America by Cambridge University Press.org/9780521195454 © Cedric Boeckx. Dubai.org Information on this title: www.CAMBRIDGE UNIVERSITY PRESS Cambridge. New York www. and Jairo Nunes 2010 This publication is in copyright. or will remain. Melbourne. 2 2.3 3.1 2.4.2 4.5.2 Introduction Departing from the null hypothesis: historical.4 2.4.1 Configurational properties 3.4 3.3 4.1 3. and empirical reasons Back to the future: elimination of DS and the revival of the null hypothesis Controlled PROs as A-movement traces 3.1 Finite control and hyper-raising 59 59 60 63 70 2.1 The null-case approach 2.1 4. architectural.6 3.4 Introduction Morphological invisibility Interclausal agreement Finite control 4.5 37 43 46 47 49 52 56 vii .3 Phonetic properties and grammatical status Conclusion 36 4 Empirical advantages 59 4.2 Interpretive properties 3.Contents Acknowledgments 1 page x Introduction 1 2 Some historical background 5 2.5.4.5 Introduction What any theory of control should account for Control in the standard-theory framework Control in GB Non-movement approaches to control within minimalism 2.3 2.2 The Agree approach Conclusion 5 5 6 9 16 16 20 35 3 Basic properties of the movement theory of control 36 3.4. 1 6.2.5 Introduction Obligatory vs.4 Phonetic realization of multiple copies and copy control Conclusion 98 102 115 123 5 Empirical challenges and solutions 125 5.2.1 Control with promise-type verbs 5.2 Raising into nominals in Hebrew 5.3 Finite control vs. hyper-raising Nominals and control 5.5.2 Control shift 5.5.3 6. control shift.6 5.3 5.1 Finite control into noun-complement clauses in Brazilian Portuguese 5.5 5.viii Contents 4.5.3.4.2 6.5 4.6.2 Introduction Passives. and intervention effects 4.7 75 79 79 83 142 147 149 152 152 160 .1 Relativizing A-movement 5.2 Finite control.2 Split control Conclusion 125 125 127 132 136 141 169 171 176 181 182 183 190 194 6 On non-obligatory control 195 6.1 Quirky case and the contrast between raising and control in Icelandic 5.4.3 Backward control 4.1 Partial control 5.3 The contrast between raising nominals and control nominals in English Obligatory control and morphological case 5.1 Adjunct control and sideward movement 4.3 Summary Partial and split control 5.6.3.2 Impersonal passives 5.5.3 Summary The movement theory of control under the copy theory of movement 4.1 5. islands. and the logic of minimality 5.2. obligatory control.2 The movement theory of control and morphological restrictions on copies 4. and Visser’s generalization 5.5.6 5.2 Apparent case-marked PROs The minimal-distance principle.4.5.4 6.4 5.3.5.4. non-obligatory control and economy computations Some problems A proposal Conclusion 195 196 202 204 209 4. 2 8.2 Challenges for “simpler syntax” Conclusion 210 210 216 217 226 237 8 The movement theory of control and the minimalist program 238 8.1 Some putative problems for the movement theory of control 7.3 Introduction General problems with selectional approaches to obligatory control “Simpler syntax” 7.4 8.3 Introduction Movement within minimalism and the movement theory of control The movement theory of control and the minimalist architecture of UG Inclusiveness.1 7.Contents ix 7 Some notes on semantic approaches to control 210 7.2 7. and the movement theory of control Conclusion 238 239 References Index 250 261 7. bare phrase structure.4 8.3.1 8.3.5 241 245 248 . 0722648. New York. XVIII Colloquium on Generative Grammar. Marcelo Ferreira. x . third author). Terje Lohndal. NSF (grant NSD. LSA 2005. and at the following meetings: ANPOLL 2003. Tilburg. Harvard. Maryland. GLOW XXX. EVELIN 2004. second author). Edges in Syntax. We would also like to acknowledge the support received from the Generalitat de Catalunya (grant 2009SGR1079. Leiden. Rutgers. We would like to thank these audiences for comˇ ments and suggestions. Special thanks to Zeljko Boˇskovi´c. S˜ao Paulo.Acknowledgments Previous versions of (part of) the material discussed here have been presented at the following universities: Connecticut. Going Romance 2007. Carme Picallo. Hans Broekhuis. first author). Lisbon. Micha¨el Gagnon. Romania Nova II. and V Workshop on Formal Linguistics at USP. and Utrecht. Ways of Structure Building. Stony Brook.BCS. and Johan Rooryck. and CNPq and FAPESP (grants 302262/2008–3 and 2006/00965–2. Lisa Cheng. the movement theory of control (hereafter.g. and Boˇskovi´c 1997]. Hornstein 1999. A-movement. Landau 1999. 1 . Despite virtues that we believe are transparent (see e. Culicover and Jackendoff (2001. and van Craenenbroeck. and (local) scrambling. 2001). 1 See e. Put another way. null case [e.g. be it in terms of the PRO theorem [e. 2005). Chomsky and Lasnik 1993. Under this long-held view.g.. such historical bias should not deter us from a fair evaluation of the conceptual properties and empirical coverage of the MTC.g. Differentiating raising from control in terms of movement has been a fixed point within generative grammar from the earliest accounts within the standard theory to current versions of minimalism (see Davies and Dubinsky 2004).. Landau (2000. 2000. 2004]) significantly hampers our understanding of the phenomenon as it leads to explanations that are roughly as complex as the phenomenon itself. Rooryck. on a par with other phenomena that have been traditionally treated in terms of A-movement such as passive. control cannot.g. which became crystallized in GB with the formulation of the (construction-specific) control module (Chomsky 1981). be evident to all readers.1 We believe that there are several reasons for this.1 Introduction In the following pages we develop an extended argument for a proposal whose conceptual simplicity and empirical success will. raising. as its basic proposal is exactly to analyze control in terms of (A-)movement. we claim that maintaining the constructional specificity of control (in whatever form. We propose that the phenomena that have been used to motivate a special and separate control construction are best explained if control is treated as an A-movement dependency.. It is thus not surprising that the MTC has been welcomed with considerable skepticism. more specifically. or ad hoc “anaphoric” tense-agreement dependencies [e. MTC) has proven to be quite controversial.. Chomsky 1981]. However. The proposal says that (obligatory) control is movement. The first one is historical. Martin 1996. if raising involves movement.. 2003). and van den Wyngaerd (2005) for a useful sample. we trust. Kiss (2005). But before we launch our defense of MTC. backward control) and often empirically succeed by stipulating what should be explained (e.. and Nunes in press) has allowed us to rectify some errors. Over the years. First. This should occasion no surprise. some features of the phenomenon heretofore assumed to be central may not be accommodated at all. we believe that the MTC actually faces few empirical difficulties (and none of principle). One aim of what follows is to make this case in detail. Correspondingly. Moreover. and sharpen the arguments. the distribution of PRO through null case). as it reflects the well-known tension between description and explanation. This stimulating intellectual exercise has led us to better appreciate the consequences of the MTC and has in fact convinced us that it covers even more empirical ground than we at first thought. For all these reasons.” Since the MTC was first proposed. we thought that a detailed defense of MTC required a monograph. it has been regularly objected . In fact. Odd as it may seem.g. failure to cover a data point may be a mark of progress if those that are covered follow in a more principled fashion. if the novel approach is conceptually tighter than the more descriptive accounts that it aims to replace (as we believe to be the case with the MTC). we cannot emphasize enough that MTC does not equate “control” with “raising. particularly if the account is comparatively recent and the full implications of its resources have not yet been fully developed. which we have tried to overcome here. a few notes are in order. Addressing the vigorous critiques of MTC here and in previous work (Hornstein 2003.2 Introduction The second reason behind the controversy is also related to the long interest control has enjoyed within the generative tradition. a tight theory may miss some “facts” and this is not necessarily a vice. 2006a.g. Finally. clarify the proposal. whereas the current alternatives both face very serious empirical hurdles (e. 2004. Boeckx and Hornstein 2003. any new approach will likely fail. as we will argue in the following chapters. The virtues of a proposal can be seriously misevaluated unless one keeps score of both what facts are covered and how facts are explained. it is fair to say that the resistance to MTC is in part due to the inadequacies and limitations of previous versions of the MTC (including our own work). Hornstein.. Nunes 2007. at least initially. We believe that many have been too impressed by these apparent problems without considering how the MTC might be developed to handle them. Boeckx. Consequently. and this is not a virtue. control phenomena have been richly described. to adequately handle some of the relevant data. A weak theory can often be “easily” extended to accommodate yet another data point. It goes back as far as Bowers (1973). it is certainly not the only one possible. passive. Although raising often proves useful in illustrating properties of A-movement that carry over to control. as well as its empirical coverage. Similarly. who already proposed that raising and control should be basically generated in the same way. for movement can only take place once ␪-assignment is taken care of. like the derivation of raising. as the proposal conflicted with core principles of almost every model of UG from Aspects to GB. Thus. However. In the chapters that follow. By contrast. Chomsky’s (1993) proposal that D-structure should be eliminated provided a very natural conceptual niche for the MTC within the generative enterprise as it removed the major theoretical obstacle that prevented movement to ␪-positions. although our specific implementation of MTC is the one that has been extended to the broadest range of data thus far. the MTC is actually not a radically new idea. This scenario drastically changed when the minimalist program came into the picture. for instance. Manzini and Roussou (2000). in a system without D-structure. rather than merely listing the possible controllers. it did not find fertile soil to blossom for a long time. as minimalism aspires to explain why UG properties are the way they are. and control complements coded as features of individual lexical items. Kayne (2002). control predicates. whether or not it is a sound option has to be determined on the basis of the other architectural features of the system. The different properties of constructions involving wh-movement and topicalization. all the MTC is saying is that. movement to ␪-positions arises at least as a logical possibility. and . we are interested in developing a theory of control that deduces the properties of control configurations from more basic postulates. we urge the reader not to dismiss our proposal simply because (unanalyzed) control– raising asymmetries exist. We hope to show that the MTC fits snugly with some leading minimalist conceptions and thus constitutes an interesting argument in its favor. the derivation of obligatorycontrol constructions also involves A-movement. where movement and ␪-assignment intersperse. In a system with D-structure. Third. movement to ␪-positions is a non-issue. controllees. Finally. but raising does not.Introduction 3 that the MTC cannot be right because of features that control has. In other words. it is a ladder that ought to be kicked away as theory advances. control is raising only in the descriptive sense that control is an instance of A-movement. Second. and vice versa. do not argue against analyzing them in terms of A’-movement. or local scrambling constructions. However. but it is not raising qua construction. O’Neil (1995). we in fact argue that control–raising asymmetries generally reduce to independent factors – something we take to be an indication that the MTC is on the right track. Let us close this introductory chapter by providing an overview of the subsequent chapters. Chapter 7 discusses the extent to which the MTC is based on more solid conceptual and empirical grounds than semantic/selectional approaches to obligatory control. Chapter 3 lays out the broad features of our version of the MTC. Chapter 6 presents our take on how non-obligatory control is to be analyzed. in GB.4 Introduction Bowers (2006) share the spirit but not the details of our analysis. Finally. and in non-movement approaches within minimalism. Chapter 5 addresses many of the empirical challenges that have been considered to be fatal to the MTC and proposes solutions compatible with the MTC. Chapter 4 discusses some of the empirical advantages that the MTC has. we will not be able to do proper justice to these works and the reader is invited to evaluate each different implementation in its own right. Chapter 2 offers a brief overview of how control is handled in the standard-theory framework. For reasons of space. Chapter 8 concludes the monograph. . PRO) and construction-specific interpretive systems (e. for instance.1 Introduction Up to very recently. accounting for the positions that the controller and the controllee can occupy. we urge the reader to consult Davies and Dubinsky’s (2004) excellent history of generative treatments of raising and control. in GB (section 2.. the control module). it must specify the kinds of control structures that are made available by UG and explain how and why they differ.. that the controllee can 1 For much more detailed discussion.3). In this chapter.4).g. 2. and in non-movement analyses within the minimalist program (section 2. there had been a more or less uncontroversial view that control phenomena should be analyzed in terms of special grammatical primitives (e. briefly outlining what we take to be the virtues and problems of each approach. for instance. In addition.2 Some historical background 2. Assuming. Assuming. it must correctly describe the configurational properties of control. 5 .5). Second. we examine how this conception of control was instantiated in the standard-theory framework (section 2. First. their differences should be reduced to more basic properties of the system.1 This discussion will provide the general background for us to discuss the core properties of (our version of) the MTC in Chapter 3 and evaluate its adequacy in the face of the general desiderata for grammatical downsizing explored in the minimalist program. that obligatory control (OC) and non-obligatory control (NOC) are different.g. it should provide an account as to why the controller and the controllee are so configured.2 What any theory of control should account for A theoretically sound approach to control – one that goes beyond the mere listing of the properties involved in control – must meet (at least) the following four requirements. 3 Control in the standard-theory framework Within the framework of the standard theory. 1970) for discussion. → John kissed Mary before/after/without [poss John asking if he could] → John kissed Mary before/after/without asking if he could 2 Here we abstract away from issues that are orthogonal to our discussion such as the interaction between END and the rule of complementizer deletion. b. why are controllees so restricted? Third. assuming that controllers must locally bind controllees in OC constructions. John regrets/insisted on/prefers [poss John leaving early] John regrets/insisted on/prefers leaving early → (4) a. control phenomena were coded in the obligatory transformation referred to as equi(valent) NP deletion (END). explaining how the antecedent of the controllee is determined and specifying what kind of anaphoric relation obtains between the controllee and its antecedent (in both OC and NOC constructions) and why these relations obtain and not others.g. the minimal-distance principle is satisfied Irrelevant details aside. why is the control relation so restricted in these cases? Fourth. See Rosenbaum (1967.. ungoverned subjects). and converts them in the corresponding (b)-sentences. for instance. John persuaded/ordered/forced/asked/told Mary [for Mary to leave early] John persuaded/ordered/forced/asked/told Mary to leave early b. (2) a. we briefly review how these concerns have been addressed from the standard-theory model to the minimalist program. it must specify the nature of the controllee: what is its place among the inventory of null expressions provided by UG? Is it a formative special to control constructions or is it something that is independently attested? In the next sections. which has the effect of deleting the term numbered 4 in (1). which for our current purposes can be described as follows:2 (1) X-NP-Y-[S {for/poss}-NP-Z]-W Structural description: 1 2 3 4 5 6 7 → Structural change: 1 2 3 4 Ø 6 7 Conditions: i. (5) a. 2. For instance.6 Some historical background only appear in a subset of possible positions (e. it must account for the interpretation of the controllee. . John tried/wanted/hoped [for John to leave early] John tried/wanted/hoped to leave early → (3) a. b. 2 = 5 ii. END applies to the (a)-structures in (2)–(5). b. that END is a deletion operation. the structure in (ia).3 (6) John persuaded/ordered/forced/asked/told Mary [for John to leave early] Finally. John persuaded a friend of Mary [for Mary to leave] John persuaded a friend of Mary to leave It should be clear how requiring that some sort of command relation hold between the antecedent NP and the deleted one will help screen out cases like (i). Given (1) above. should allow for control by ‘Mary’ in (ib): (i) a. As far as the configurational properties of control are concerned. It is a regular NP in the underlying structure and the fact that the corresponding surface position is phonetically null follows from the kind of transformation END is.e. for example. stipulative) nature remained. the interpretation properties of control are enforced by condition (i). leaving nothing at surface structure. according to the minimal-distance principle. As opposed to what we find in (4a). This general approach was refined within the standard theory as more complex control structures were considered. As for adjunct control in sentences such as (5). sentences such as (4b) must be derived from the structures in (4a) and not from the one in (6) below. It is a deletion transformation that removes the targeted NP.” which was understood in terms of coreference.e. i. which requires that controller and controllee be “identical. the superficial phonetic difference between controller and controllee results not from intrinsic lexical properties of the controllee.2.3 Control in the standard-theory framework 7 According to this approach. the minimal-distance principle is satisfied under the assumption that the embedded clause is adjoined to the matrix clause and. but from properties of the computation itself. as such. This by no means diminishes the value of these earlier approaches to control. it is structurally closer to the subject than it is to the object.. b. . Identifying the different properties of control phenomena with such formal rigor 3 END as stated is not entirely adequate empirically. Thus. To put it differently. but its axiomatic (i. and that the controller must be the closest NP (in compliance with the minimal-distance principle). END explicitly specifies that the controllee (the target of deletion) must occur in the subject position of infinitival clauses (for-clauses) and gerunds (poss-clauses). the antecedent of the controllee in (6) is not the closest NP around. there is nothing of special interest in the nature of the controllee. where the “wrong” NP is chosen. The configurational and interpretive properties of control were analyzed as irreducible features of the END transformation itself.. which would incorrectly allow the understood subject of the embedded clause to be interpreted as being coreferential with the matrix subject. The first one regards an empirical problem that the standard-theory approach faced in relation to the way it handled the interpretive properties of control. b. Data such as (9) are not the sorts of cases Grinder discussed. which were operations governed by command relations. the problem of how to characterize the rules that lead to control structures (are they chopping rules or construal rules?) highlights the tension that we will see constantly recurring: how best to account for both the distribution of the controllee and its interpretation. Rather. including the MTC. with large consequences for theorizing beyond control structures.5 4 If there is an anaphoric relation in control structures. The second point worth mentioning concerns the identification of another type of control.. Seen from a contemporary perspective. two points are worth mentioning which will be relevant to the discussion of these later reanalyses. → (7) a. They appear to require a phonetic gap. This operation also deletes a subject of a non-finite clause but. [John wants [John to win]] John wants to win (8) a. but they fall under the SEND rubric.4 The obvious question then is how to obtain this bound interpretation. what we actually need is a kind of bound anaphor or an expression that can be so interpreted. END was distinguished from a related operation dubbed super-equi (SEND). The problem is that control structures do not appear to leave lexical residues like the other construal operations. as exemplified by the contrast between (7) and (8). coreferential with) its controller in underlying structure. Relatively early on. and it paved the way for subsequent reanalyses in GB and in the minimalist program.8 Some historical background was unquestionably an achievement. As we saw above. then END is unlikely to be a chopping (“gap”-leaving) rule.e. it is more like the rules of reflexivization or pronominalization. it operates across unbounded stretches of sentential material. 5 See Grinder (1970). Problems arise when the controller is not a referential NP.’ This suggests that. [Everyone wants [everyone to win]] → Everyone wants to win Whereas (7a) might be taken to roughly represent the meaning of (7b). as illustrated in (9). the controller and the controllee were taken to be lexically identical and the semantic relation between them was understood as coreference. Before we leave this brief review. (8a) in no way represents the interpretation of (8b). in contrast to END. . b. which should rather be paraphrased as ‘Everyone wants himself to win. instead of an NP identical to (i. b. b. In (9). The following question then arises: what is the relation between END and SEND? Or to put the question somewhat differently: why should UG have two rules that have the same effect (deletion of an identical NP). like the ones in (10). c. b. in contrast to standard END configurations. a base-generated NP containing no lexical material ([NP Ø]). they can also be used to generate structures like (13). for instance. as ‘Mary’ intervenes between the target of deletion (‘John’ in S4 ) and its antecedent (‘John’ in S1 ). 9 [S1 John said [S2 that Mary believes [S3 that [S4 John washing himself] would make a good impression on possible employers]]] → John said that Mary believes that washing himself would make a good impression on possible employers Note that (9) violates the minimal-distance principle. Within GB. and lexical-insertion operations. but apply to different configurations?6 In the next sections we examine some answers to these two issues that were offered within GB and the minimalist program. where the subject of the clause has been generated by the phrase-structure component 6 Grinder (1970) actually collapsed END and SEND.4 Control in GB Building on earlier work in the extended standard theory (EST). as we shall see below. 2. S → NP INFL VP VP → V NP NP → N (11) a. the controllee is not within a clausal complement (or adjunct) of a higher predicate.4 Control in GB (9) a. However. the controllee is a PRO. the controllee is within the sentential subject of S3 . . the GB approach to control is considerably more ambitious and empirically more successful than the standard-theory model. The GB theory of the base includes both phrasestructure rules. like the ones in (11). c.2. Moreover. However. later approaches identified many substantial differences between the constructions underlying END and SEND that are better captured if two kinds of control are recognized. N → John/he/it/Bill V → kiss/see/admire INFL → past/to These two types of rules operate in tandem to generate structures such as (12) below. (10) a. This conception of the controllee as a base-generated non-lexical formative arises as a natural consequence of the GB assumptions regarding the base component. for example) and. In other words. it is hard to explain why PRO must be phonetically null and why it requires an antecedent.10 Some historical background but has not been filled by lexical insertion.. It has no phonetic matrix and its only semantic feature is the requirement that it must be coindexed with a grammatical antecedent. see Hornstein (2001. that the principle of full interpretation does not tolerate contentless structures. grammatical) terms. then we end up with a ‘free variable’ in LF. invoking such features in the construction of lexical items (be it PRO or any other item) is just a way of simulating a grammatical requirement via lexical stipulation. PRO.8 This point is worth emphasizing. is not simply a semantically dependent expression that needs to be interpreted with respect to some salient element in the discourse (e. but this cannot explain why PRO is necessarily anaphoric and null. Moreover.’). If one treats PRO as a lexical element. an improper representation. 9 For a discussion of reflexives and bound pronouns in light of this discussion. .g. on this view. furthermore. like ‘the other’ in ‘John ate one of the bagels.” 8 This point is similar to Chomsky’s (1995) argument against considering Agr as a lexical category. Given that its only features are uninterpretable. this is a very odd lexical feature as it is only definable in configurational (i. PRO is a rather unusual lexical element as it has no positive properties. However. so conceived. Rather. it is possible to stipulate that these two features are inherent properties of a specific lexical item (PRO). In short.7 We wish to stress this point as it is important for some of the discussion that follows. This feature of the GB analysis 7 See Chomsky (1980: 8): “If Coindex does not apply and the embedded clause contains PRO. PRO is specified as needing an antecedent in a particular structural configuration. Harry ate the other. 2007).e. not a sentence but an open sentence. a theory of the base factored into a set of phrase-structure rules and lexical-insertion operations has room for an element like PRO: it is what one gets when one generates an NP structure but does not subject it to lexical insertion. If one assumes that categories without lexical content are uninterpretable unless provided with “content” (by being linked with an antecedent.9 The GB approach offers a sounder alternative as it treats PRO’s properties as the result of interacting grammatical principles.. Of course. all things being equal. is to take these features as belonging to related true lexical categories. a preferable approach. (12) [S [NP John] past [VP see [NP Bill]]] (13) [S [NP Ø] to [VP see [NP Bill]]] This way of understanding PRO has an interesting consequence for the constructions that were captured by END in the standard theory. then the requirement that PRO must have an antecedent follows naturally. Note also that PRO is a non-terminal. an NP without an index. GB strove to derive this fact from the binding theory. known as the PRO theorem.” 12 Saying that PRO was a pronominal anaphor does not imply in the context of GB that it was a lexical formative.e. We return to this point in Chapters 3 and 4. In contrast. that it should be both free and bound in the same domain. but is only coindexed later in the derivation. NP-traces receive their indices as they are created in a movement operation (they must be coindexed with the NP that moves). . as such. they differ only in the way the index is assigned – as a residue of a movement rule in one case. Notice that. x a variable. i.10 What distinguishes them is neither their internal structures nor their interpretation. we are already very close to the MTC. were treated as anaphors (Chomsky 1981) despite their clearly being non-lexical. i. once we take PRO and NP-traces to be indistinguishable at some point(s) in the derivation. The only way for such an expression to meet both requirements is for it to vacuously satisfy them. but how they are introduced in the derivation and how they get their indices. Given that the binding domain for an expression was defined (in one of its formulations) as the smallest clause within which it is governed. they become completely indistinguishable from NP-traces. . . by not meeting the necessary conditions for these requirements to be enforced. and by a rule of control in the other . PRO is in fact quite similar to NP-traces (standard traces of A-movement) in GB. Finally. then PRO does not have a binding domain if it is ungoverned (if it has no governor.” 11 See Chomsky (1977: 82): “trace and PRO are the same element. after PROs get their indices (at S-structure or LF). The GB account of the distribution of PRO is similarly ambitious.. PRO cannot have a binding domain. The proposal.12 Principle A states that an anaphor must be bound in its domain. as a base generated NPx . We consider some possibilities below (see also Chapter 3).e. Being a grammatical. Thus. for instance. namely. principle B that a pronoun must be free in its domain. Traces. if one takes an Infl head to be a governor if 10 See Chomsky (1977: 82): “We may take PRO to be just a base-generated t(x) [trace of x]. for instance). However.2. PRO was taken to be a pronominal anaphor and. they end up imposing contradictory requirements on a pronominal anaphor. non-lexical formative specified as [NP Ø]. Under the assumption that these principles apply within the same domain.11 PRO is inserted at D-structure. Any adequate theory of control should eschew lexically stipulating PRO’s basic properties and specify how grammatical principles interact so that the desired properties of PRO emerge.. Rather than simply stipulate that it appears in the subject position of non-finite clauses. went as follows (see Chomsky 1981).4 Control in GB 11 of control is clearly a desirable one for any theory to have. subject to both principles A and B. PRO cannot have phonetic content. e. as well as examples of its pronominal behavior. Sweat. Fodor (1975). Once again.’ not ‘and Bill expects John to win’) e.g. it cannot be case marked. taking PRO to be a pronominal anaphor also seems to have other welcome consequences as far as its interpretation is concerned. (14) a. b. as illustrated in (15). c. are phonetically null. A side benefit of this reasoning is that it provides an account of why PRO must be phonetically null. will be associated with a structure along the lines of (14b). d. Everyone wants to win [Everyone wants [PRO to win]] At first sight. by taking PRO to be a non-lexical formative. but is rather interpreted as a bound variable. b. and Tears’ speech (16) a. If PRO only appears in ungoverned positions.. ∗ It was expected PRO to shave himself John1 thinks that it was expected PRO1 to shave himself ∗ John1 ’s campaign expects PRO1 to shave himself John1 expects PRO1 to win and Bill2 does too (‘and Bill expects himself to win. as desired. where PRO does not have quantification properties on its own. Williams (1980). One indeed finds examples of its anaphoric behavior. These too occur in caseless positions and. d. case theory requires that nominals with phonetic content bear case and case is taken to be assigned under government. Therefore. see e. the problem posed by quantified expressions in the standard-theory framework dissolves. ∗ It is illegal PRO to park here John1 thinks that Mary said that PRO1 shaving himself is vital John1 ’s friends believe that PRO1 keeping himself under control is vital if he is to succeed John1 thinks that PRO1 getting his resum´e in order is crucial and Bill does too (‘Bill2 thinks that his1/2 getting his resum´e in order is crucial’) [The unfortunate]1 believes that PRO1 getting a medal is unlikely Only Churchill remembers that PRO giving the BST speech was momentous 13 On the properties illustrated in (15) and (16) as well as further data and discussion. for instance.12 Some historical background it is finite but not if it is non-finite (to and ing). as illustrated in (16). this makes PRO very similar to NP-traces. [Only Churchill]1 remembers PRO1 giving the ‘Blood. c.13 (15) a. A sentence like (8b). for otherwise the case filter would be violated. [The unfortunate]1 expects PRO1 to get a medal f. Lebeaux (1985). one then derives the distribution of PRO: it can only appear in the subject position of non-finite clauses. Notice also that. b. not surprisingly. Within GB. and Higginbotham (1992). repeated below in (14a). . f. e. it does not require an antecedent (cf. it makes more sense to assume that PRO is ambiguous between a reflexive and a pronoun. the antecedent need not be local (cf. we are left with no account of the distribution of PRO. [16b]) or c-command it (cf. PRO roughly behaves like a reflexive. Despite appearances. the data in (15) and (16) actually turn out to be quite problematic for the specification of PRO as a pronominal anaphor within GB. In the case of OC. Hence.2. it is tacitly assumed that the control module somehow obliterates the pronominal . By contrast. respectively) within the standard theory. illustrated in (16). (16d) allows both strict and sloppy readings. other (rarely specified but frequently adverted to) principles come into play. Not by coincidence were data such as (15) and (16) handled by two different transformations (END and SEND. In configurational terms. and a bound reading when its antecedent is associated with only (that is.4 Control in GB 13 In (15). [15d]). This accounts for the addition of the control module in the GB framework. [15b]) and ccommand it (cf. than to assume it is a pronominal anaphor. as the theorem crucially assumes the existence of an element that is simultaneously a pronoun and an anaphor. if the PRO theorem falls. Notice that PRO displays either properties of reflexives or properties of pronouns. In addition. [16a]) and. it requires an antecedent (cf. this ambiguity thesis completely undermines the PRO theorem. and (16f) may be falsified by situations in which people other than Churchill recall the import of the BST speech. illustrated in (15). and non-obligatory control (NOC). it is only felicitous if the unfortunate is conscious of who he is and expects himself to get a medal). it only supports a sloppy interpretation under VP ellipsis (cf. Notice that this amounts to saying that. The requirements of the PRO theorem have one further architectural consequence: in order to explain the distributional properties of PRO in terms of the PRO theorem. like in the earlier treatment in terms of END and SEND. some other component of the grammar must be responsible for PRO’s specific interpretation in a given configuration. [15c]). the controller is lexically specified as an argument of the embedding control verb and. Thus. However. The control module recognizes two types of control: obligatory control (OC). On the interpretation side. where there is an antecedent. in the case of non-local control. [15f] can be paraphrased as ‘Only Churchill is such that he remembers himself giving the BST speech’ and not as ‘Only Churchill remembers that Churchill gave the BST speech’). In turn. a de se reading in sentences such as (15e) (i.. [16c]). (16e) permits both de se and non-de se interpretations. [15a]) which must be local (cf. Importantly. OC and NOC are rather distinct types of relations. But in no case does it display properties of both pronouns and reflexives. in (16) PRO behaves like a pronoun in every respect. but that this tacit assumption casts suspicion over the initial specification of PRO as a pronominal anaphor.14 The relevant point for our discussion is that. this apparent success does not survive closer scrutiny either. The first thing to be noted is that the PRO-theorem account of the distributional properties of PRO is intrinsically associated with a specific formulation of binding domains. Why should UG provide PRO with such a specification only to see it blotted out later? After all. (17) ␣ is a binding domain for ␤ iff ␣ is the minimal NP or S containing ␤.15 To put it broadly. in which ␤ is governed and ␣ has a subject accessible to ␤. but a non-finite Infl is not. the account of the distribution of PRO exclusively in terms of government involves an independent axiom. this assumption 14 See Lasnik and Uriagereka (1988) for a good discussion of the notion of accessible subject and the motivations for its inclusion in the definition of binding domain. One could reply that we should learn to live with the PRO module given the nice results we obtain with the PRO theorem concerning the distribution of PRO. this does not happen with standard pronouns and anaphors: they must live with the pronominal or anaphoric specifications stated in their birth certificate. if binding domains are to be formulated along the lines of (17). But. and. Recall that a crucial assumption in the PRO-theorem account is that a finite Infl is a governor. This is not the place to review the various reasons for including the notion of accessible subject in the definition of binding domain within GB. one in which government is essentially the one and only requirement to be satisfied. rather than a theorem. . From an empirical point of view. once accessible subjects become part of the definition of binding domains. conversely. This in turn undermines the account of the distribution of PRO in terms of the PRO theorem. if the correct definition of binding domain ends up including other requirements. its anaphoric specification in NOC constructions.14 Some historical background specification of PRO in OC constructions. Recall that all that matters for PRO to vacuously satisfy both principles A and B is that it does not have a binding domain. for instance. the definition in (17) (see Chomsky 1981). there may be other ways for PRO to lack a domain. 15 See Bouchard (1984) for discussion. The problem is not so much that the details of how this would be achieved are never spelled out. However. PRO may also lack a domain if it does not have an accessible subject. Take. To lack a governor is certainly one way for PRO to be deprived of a binding domain. as government is no longer the only player on the field. But the problem is actually worse than these remarks suggest. If Achains are independently subject to a case-licensing requirement (say. the GB approach has significant empirical and theoretical problems. but a non-finite Infl does not.) for discussion. However. it is not at all obvious how this assumption can be formally encoded in the system. the PROtheorem account would be forced to assume that their finite Infls are optional governors. if a finite Infl can govern its Spec. being a governor cannot be listed as a lexical property for the reasons discussed above.2. Nunes. the account of the distribution of PRO turns out on closer consideration to be 16 See Hornstein. In short. and Grohmann (2005) for a discussion of this point.5.4 Control in GB 15 is challenged by languages like Brazilian Portuguese. . Aoun’s [1979] visibility condition. as it is unnatural to encode structural properties as lexical features or formulate different notions of government for different lexical items. On the down side.2 and 4. Again. Such a lexical specification would be comparable to saying. On the plus side. which allow obligatory control into indicative clauses. PRO’s lack of phonetic content could not be the reason for this exception.2. a finite Infl head m-commands its Spec. What is required is a structural reason for preventing a non-finite Infl from governing its Spec. Even the apparent benefit of the account of PRO’s lack of case has an undesirable consequence. when details are considered. to assume the opposite would be parallel to saying that. as the two structural configurations are identical. Given that government is a structural relation. although the configurational Spechead relation is exactly the same in both cases. as we will see in detail in sections 2. treating PRO as a grammatical formative circumvents the previous problem related to control involving quantified expressions. for instance. which requires that ␪-roles be associated with case in order to be visible at LF). why should A-chains headed by PRO be exempted from such a requirement? Notice in particular that. In order to make room for finite control in such languages. and accounts for why PRO is phonetically null and why (at least in the case of OC) it needs a grammatical antecedent. To sum up: despite its laudable ambitions and its improvement over the standard-theory approach to control.16 This just does not make sense. regardless of the definition of government one assumes. But.4 below. An A-chain in GB must be headed by a case-marked position unless it is headed by PRO.17 This statement is transparently troublesome. 17 See Chomsky (1981: 334ff. given that chains headed by pro and null operators also required case licensing. the distributional properties of PRO do not follow theorematically and it is not even obvious how to convert the PRO theorem into an axiom. so should a nonfinite Infl. that a given lexical item is lexically specified as being unable to c-command. 2. Notice that the contrast in (18) mimics the contrast in the ECM constructions in (19). is ungoverned. It is no wonder that the control module always felt like an appendix to the model and never occupied a bright spot among GB’s theoretical achievements. by assumption. A separate control module must then be added to the theory to specify the interpretive properties of PRO. the construction-specific flavor of this new addition to the model is at odds with the general goal of the principlesand-parameters theory of deducing properties of rules and constructions from the interaction of more basic features. Let us consider how the distributional properties of PRO are handled on this account. one would in principle expect that the abandonment of this assumption should also lead to the abandonment of the control module.5. ∗ John hoped [PRO1 to be elected t1 ] John hoped [PRO1 to appear to t1 [that Bill was innocent]] From the perspective of GB. The ungrammaticality of (18b) is therefore unaccounted for within GB. b. which can be straightforwardly captured if one assumes .16 Some historical background less a theorem than an axiomatic stipulation. gave up on the account of PRO in terms of its alleged pronominal-anaphoric nature. The GB take on control was therefore ripe for a minimalist reanalysis. PRO cannot remain in the object position of the embedded verb in (18a) or the preposition to in (18b).1 The null-case approach Given that the addition of the construction-specific control module in GB was prompted by the problematic assumption that PRO is a pronominal anaphor. Hence. PRO cannot occupy a governed position as it would then meet the requirements for binding theory to apply and would end up violating principle A or principle B. However. However.5 Non-movement approaches to control within minimalism 2. The first minimalist reanalysis of control. Moreover. once it moves to the subject position of the infinitival clause. (18) a. Take the contrast between (18a) and (18b). history and logic are known to frequently go their separate ways. outlined in Chomsky and Lasnik (1993). for instance. but basically left intact the assumption that the interpretation of PRO required a special module in the system. it should circumvent the binding violation in both (18a) and (18b). Moreover. which. the assumption that PRO is a pronominal anaphor leads to empirical problems as the system cannot predict when PRO behaves like an anaphor and when it behaves like a pronoun. Under the assumption that the infinitival to in (18) checks null case. One thing is certain. PRO cannot appear in the subject position of finite clauses or in the object position of a transitive verb. Chomsky and Lasnik (1993) propose a case-based account of the distributional properties of PRO under which A-chains headed by PRO are not exceptional as far as case licensing is concerned. This proposal also extends to the standard cases regarding the distribution of PRO. b. In other words. which cannot. (21) (22) a. which is checked by some non-finite Infl heads.2. under this view. the null-case approach must somehow ensure that the infinitival ‘to’ of control constructions can license PRO.5 Non-movement approaches to control 17 that a given expression cannot move from a case-marked position to another case-marked position. ∗ John hoped (that) PRO could eat a bagel Bill saw PRO Notice that some extra assumption must be made in order to capture the standard contrast between (21) and (22) below. but not the infinitival ‘to’ of ECM or raising constructions. John hopes [PRO to graduate soon] ∗ ∗ I believe [PRO to be nice] It seems [PRO to be nice] 18 The idea of accounting for the distribution of PRO in terms of case finds its origins in Bouchard’s (1984) proposal that PRO cannot appear in a case-marked position.4). One cannot simply say that these heads are lexically ambiguous in terms of their specification for case checking. where it proceeds from a casechecking to another case-checking position. as they were in GB (see section 2. b. Thus. ∗ We expected [John1 to be hired t1 ] We never expected [John1 to appear to t1 [that the job was easy]] Based on the parallelism between pairs like (18) and (19). otherwise.18 The gist of their proposal is that PRO must be licensed by a special kind of case. . b. (19) a. movement of PRO is licit in (18a) as it proceeds from a caseless (passives generally do not check case) to a case-checking position. as respectively illustrated in (20). for instance. dubbed null case. ∗ (20) a. The obvious question is how to independently distinguish the ‘to’ that can check null case in (21) from its siblings in (22). structures corresponding to (22) should be grammatical with the case-checking version of ‘to’. because these are not positions in which null case is checked. but not in (18b). requires an irrealis interpretation (that is. for example. (25) a. the ECM/raising infinitival clauses are taken to be tenseless as they are incompatible with eventive predicates (cf. and licenses VP ellipsis (cf. as well The contrasts above are supposed to show that the control infinitival is tensed as it is compatible with eventive predicates (cf. (23) a. the truth of the complement is left unspecified at the time of the utterance. (26) a. [25b]). and do not license VP ellipsis (cf. b. null case would be very similar to nominative case. 2001) is the most fully worked out version of the null-case approach to the distribution of PRO. ∗ (24) a. At 6. Unfortunately. Building on Stowell’s (1982) proposal that control infinitives are tensed whereas ECM and raising infinitivals are tenseless. [24a]). Pires (2001. Baltin and Barrett (2002). Leo believed Bill to sing in the shower right then Leo decided yesterday to leave tomorrow John believes/believed Mary to be pregnant ∗ ∗ Leo decided [[to leave] [which was/is true]] Leo believes [[John to be smart] [which is true]] Leo doesn’t want John to sing in the shower. [26a]). b.19 Tying null case to tense has the virtue of rendering it more natural and less stipulative. [23a]). as convincingly shown by Wurmbrand (2005). [23b]). triggers a future reading (cf. cf. [24b]). 20 For further discussion and arguments against null case and its ties to tense. b. on the other. [25a]). Martin proposes that only tensed infinitivals check null case. anyway Leo believes John to be honest and she believes Frank to. require a simultaneous interpretation with respect to the embedding clause (cf.20 Take the contrasts between the infinitival complements of the control verb ‘decide’ and the ECM verb ‘believe’ in (23)–(26) (from Wurmbrand 2005). but he decided to. which attempts to couch the distinction between (21) and (22) on more solid grounds. [26b]). and Hornstein (2003). as both would be checked by a tensed Infl.18 Some historical background Martin (1996. . and ECM and raising predicates. Under this perspective. see also Landau (2000). 19 See also Boˇskovi´c (1997) for relevant discussion. Conversely. on the one hand. Leo decided to sing in the shower right then At 6. differing only in terms of their morphological realization. 2006). 21 See Hornstein (1990) for a discussion of this interpretation in the context of sequence-of-tense constructions. b. the proposed independent diagnostics for distinguishing tensed from tenseless infinitivals fail to yield the expected divide between control predicates.21 allow a realis interpretation (cf. the control verb ‘claim’ does not license eventive predicates (cf. c. c. whatever tense properties nonfinite clauses have. which decouple tense properties from control infinitivals.’ an ECM verb. In turn. does not permit a realis interpretation (cf. ∗ At 6. Pires 2001. the clear acceptability of raising examples such as the ones in (29) indicates that the licensing of VP ellipsis fails to cleanly distinguish control from raising. from ‘believe. the infinitival complement of the ECM verb ‘expect’ is compatible with an eventive predicate (cf. as Wurmbrand (2005) shows. b. are seconded by the observation that PRO may also occur in gerundive subject positions. but this varies across verbs and there is no apparent systematic way to distinguish the two classes using the “tense” diagnostics mentioned . despite the fact that gerunds are generally analyzed as not tensed (see Stowell 1982.’ a control verb. 2006).2. (28) a. b. The tower started to fall down and the church began to as well John expects the printer to break down whereas Peter expects the copier to They say that Mary doesn’t know French but she seems to The above arguments. Besides. whereas the control verb ‘manage’ does not trigger a future reading (cf. (29) a. is that the criteria do not generalize to other control and ECM/raising cases. [27c]). There are surely differences between raising and control complements. b. [28c]). For instance. John hated [PRO eating turnips (∗ tomorrow)] John preferred [PRO eating turnips (∗ tomorrow)] The overall conclusion one reaches is that. they do not seem to be useful for distinguishing raising from control configurations. c. (27) a. Leo claimed to sing in the shower right then Leo claimed [[to be a king]. (30) a. and allows a non-simultaneous interpretation (cf. [27b]). b. [27a] below) and allows a realis interpretation for its complement (cf. The problem. where the gerund licenses PRO but not the temporal adverb. which was true] ∗ John managed to bring his toys tomorrow ∗ The bridge is expected to collapse tomorrow The train is expected [[to arrive late tomorrow] [which is true]] The printer is expected to work again tomorrow Wurmbrand (2005) also reviews the VP ellipsis data and observes that the data that purport to demonstrate a distinction between control (where it is allowed) and raising (where it is prohibited) are subject to substantial speaker variation (when the contrast exists at all). [28a]). [28b]). This is illustrated in (30) below.5 Non-movement approaches to control 19 The above paradigm does indeed distinguish ‘decide. What is worth noting here is that simply reducing OC to something like principle A and NOC to something like principle B does not by itself suffice to account for the interpretive properties of OC and NOC configurations. These three stipulations track but do not explain the facts under discussion. With the PRO theorem abandoned. What of PRO’s interpretive properties? Here there is some good news. what remains to be determined is why OC and NOC PROs distribute as they do.2 The Agree approach Let us now consider Landau’s (1999. contrary to fact. that OC PRO requires a local. 2.5. the null-case approach requires three stipulations: (i) PRO has no phonetic content. the attempt to analyze null case as similar to nominative by associating it to a form of tense ends up failing. it is not clear how this is to be implemented grammatically (see Chapters 6 and 7 below). null case finds no independent motivation within the system and follows from nothing but the attested distribution of PRO. of course. Once the distribution of PRO cannot be reduced to a [±tense] feature of T.. stipulate that certain predicates select for OC and so for reflexive-like PROs. 2004) alternative approach to control. unlike proponents of the null-case . (ii) null case must be assigned to PRO. adjunct control seems to pattern like OC and. This is really bad news. c-commanding antecedent follows from its being subject to principle A of the binding theory (or whatever substitutes for principle A). For example. One can. If selection is a head-to-head relation. on the standard assumption that predicates can select complements but not adjuncts. while others do not. And the picture is not very glamorous. It is then possible to reduce the interpretive properties of PRO to the interpretive properties of pronouns and reflexives. Like the null-case approach reviewed in the previous section. Landau takes the existence of PRO for granted but. then adjunct control is expected to be NOC. Second. Thus. then OC is not an obvious case of selection. PRO can be treated as ambiguous. despite its explanatory aspirations. In order to work. First. The fact that NOC PRO does not need an antecedent follows its being pronominal. These are issues that we revisit in later chapters. a null reflexive in some contexts (OC cases) and a null pronoun in others (NOC cases). why reflexive PRO appears in OC contexts and pronominal PRO in NOC contexts. Given such a reduction. In other words. although conceptually appealing. it seems fair to say that the null-case approach amounts to stipulating that PRO appears where it does and that it has the phonetic properties it has.20 Some historical background above. it is not clear how selection of embedded subjects by matrix verbs (so-called control predicates) is to be stated.e. However. 2000. and (iii) only PRO can bear null case. i. but an untensed infinitive such as the complement of “implicative” verbs does not. .4. According to him. Landau argues that a tensed infinitive such as the complement of “desiderative” verbs licenses it. (32) The chair[−Mer] hoped [PRO[+Mer] to gather/meet at 6/to apply together for the grant] It is a great merit of Landau’s work to have shown that partial control is indeed an instance of obligatory control (the controllee requires a local c-commanding antecedent. Let us examine each of these major aspects of Landau’s system.2 below for details). triggers sloppy readings under ellipsis. 2000).5 Non-movement approaches to control 21 approach. ‘committee’ is [+Mer].1 The relevance of partial control Partial control refers to control constructions where an embedded predicate must take a (semantically) plural subject.2. three other aspects of Landau’s approach stand out: (i) the special attention given to “partial-control” constructions. as illustrated by the contrast between (31) and (33) (see section 2.2 below. he takes PRO to bear regular case like any other DP. leaving the discussion of whether or not PRO bears regular case to section 5. Assuming that this interpretive fact shows that controller and controllee are not identical. as illustrated in (32). which he takes to replace his older treatment (Landau 1999. the matrix subject is understood as a member of the set of people denoted by the embedded subject. Landau takes partialcontrol constructions to be a strong argument for a PRO-based account of control.2.5. while ‘chair’ is [−Mer]). (31) The chair hoped [PRO to gather/meet at 6/to apply together for the grant] In (31). see Hornstein (2003) and Landau (2007) for a rejoinder. and enforces de se readings. (ii) the dependence of obligatory control on the postulation of certain features and feature specifications. 22 Here we will primarily focus on Landau (2004)’s analysis of obligatory control. as illustrated in (31). which characterizes group names (for instance. but the antecedent of the controllee is (semantically) singular. 2001) Agree operation. and to have provided a very detailed description of the types of predicates that allow partial control.22 2.5. For discussion of the limitations of his previous treatment. for instance). In addition to this take on case.2. (iii) the interpretation of PRO mediated by (a version of) Chomsky’s (2000. the mismatch in interpretation between PRO and its antecedent results from PRO being independently specified for the semantic feature mereology. [31]). should allow a partial-control reading with a [+Mer] PRO. as opposed to what happens with ‘meet/ gather/apply together’ in (36a).’ Now. . And there are empirical problems. after Landau’s work. ∗ The chair met/gathered/applied together for the grant with Bill The chair sang alike/is mutually supporting with Bill The data in (36)–(37) further show that being compatible with a commitative PP is not sufficient for partial control to be licensed: the commitative must be selected. Hornstein suggests that what seems to distinguish the predicates that support partial control from the ones that do not is that the former can select a commitative PP. b. As observed by Hornstein (2003). whose matrix predicate is of the type that licenses partial control. as well.5. one would expect that a sentence such as (37). So (34b) shows that partial control must in part be determined by properties of the embedded predicate. the amount of ad hoc machinery required to account for partial control in Landau’s system. (34) a. partial control came to be part of the empirical basis that any approach to obligatory control must take into consideration.3. the commitative associated with ‘leave/go out’ is not selected. But this does not happen. (36c) shows that a [+Mer] noun can be the subject of ‘leave/go out. (36) a. ∗ They sang alike/were mutually supporting John hoped/wants [PRO to sing alike/to be mutually supporting] Notice that the matrix predicate of (34b) is of the type that licenses partial control (cf. Example (37) only has an exhaustive control reading. given that in Landau’s system PRO can always be intrinsically specified as [+Mer].2. In fact. b. ends up undermining the initial appeal that a PRO-based theory appears to have. c. partial control: ∗ ) Example (36b) shows that. However.22 Some historical background (33) ∗ The chair managed [PRO to gather/meet at 6/to apply together for the grant] It is fair to say that. as we will see below in section 2. as shown in (35). as shown in (34). The chair met/gathered/applied together for the grant (∗ with Bill) The chair left/went out (with Bill) The committee left/went out (37) The chair preferred [PRO to leave/go out at 6] (exhaustive control: OK. (35) a. b. it is not the case that any predicate that selects a plural subject licenses partial control. In turn. . For purposes of exposition. (38) a.1 below for discussion). below we use I for the tense head T in order to distinguish it from the tense feature [T]. rather than involving a plural subject. in Landau’s system. What is relevant for our current purposes is to point out that. partial control does not intrinsically favor a PRO-based approach and we are back to the original question of what the best account of the null embedded subject of (38a) is (see section 5. the availability of partial control should be quite free once the tense requirements on the infinitive are satisfied. In particular. partial control may in fact involve the licensing of a null commitative argument in a standard (“exhaustive”) obligatory-control construction. Where Landau departs from previous accounts is in the way these features conspire to determine the nature of control. C(ontrol)-subjunctives and F(ree)-subjunctives are distinct in that only the former necessarily require an obligatorycontrol interpretation of their subjects.6.2. as shown in (39) (from Landau 2004: 840). or cannot be PRO.2 [Tense] and [Agr] features and finite control The second major aspect of Landau’s system is the specific typology of control configurations − involving both non-finite and finite clauses – it establishes. whatever accounts for exhaustive control should also cover partial control. Landau assumes that the local environment of the embedded subject must provide all the necessary information to determine whether it must. Following a venerable tradition. a sentence such as (38a) should actually be represented as in (38b) (still keeping PRO for purposes of discussion). 2.5 Non-movement approaches to control 23 The fact that the availability of partial control is contingent on there being a predicate that selects a commitative complement suggests that. And if partial control turns out to be more related to the licensing of null commitative arguments. In other words. It is indeed quite mysterious in his system why partial control should depend on the potential licensing of commitative arguments within the infinitival clause.5. can. The chair preferred to meet at 6 [The chair]i hoped [PROi to[+tense] meet prok at 3] Here is not the place for us to pursue the suggestion encapsulated in (38b) (see section 5.1 below for a suggestion of how partial control can be analyzed under the MTC). where pro is a null commitative argument.23 23 EC and PC in (39) stand for exhaustive and partial control respectively. if something along the lines of (38b) is on the right track. b. Landau takes the relevant local licensing features to be (semantic) [T(ense)] and (morphological) [Agr(eement)]. That is.2.6. the basic intuition underlying the typology in (39) is that obligatory-control configurations do not form a natural class. +Agr] [+T. ∗ Yesterday John hoped to travel tomorrow Yesterday John managed to travel tomorrow As Landau observes. Anything else is possible. selects a CP headed by C[+T] . the generalization is that if I is positively specified for both [T] and [Agr]. +Agr] [+T. which in turn selects an IP headed by I[+T] . −Agr] [−T. 2004. under certain circumstances. Putting aside the case of Hebrew third-person subjunctives for the moment. a verb like ‘hope. non-arbitrary) null subjects in (colloquial) Brazilian Portuguese.’ for instance. this generalization is falsified by “referential” (i. +Agr] Ø Consider the infinitives in (39). it does not trigger obligatory control. as illustrated in (40) below. “the only generalization in this domain that appears to be universal is the incompatibility of indicative clauses with OC. 2009) and Rodrigues (2002. Given that the tense properties of I are predicted by the selecting predicate and that selection is a local relation. but only exhaustive control if specified as [−T]. Thus. −Agr]) will necessarily lead to obligatory control.2. As Landau (2004: 849–850) puts it. b.e. Given this feature distribution. +Agr]) or a negative specification on both ([−T. they are in fact the complement subset of the natural class of non-controlled environments. In sum. −Agr] C0 [−T] [−T] Balkan F-subjunctive indicative [+T. null subjects in Brazilian Portuguese show all the diagnostics of . for instance.24 Some historical background (39) Obligatory control EC-infinitive Balkan Csubjunctive No control Hebrew 3rd-person subjunctive PC-infinitive I0 [−T. +Agr] [+T. non-expletive.1. (40) a. This difference is meant to capture the fact that the infinitival clauses that allow partial control can be temporally independent from the matrix clause. obligatory control is the elsewhere case. +Agr] [+T. the [T] features of I are accordingly replicated on C in (39). On the other hand. Landau has argued that the essential difference between an infinitival that allows partial control and one that disallows it is its tense properties: an infinitival I allows both exhaustive and partial control if specified as [+T]. (+Agr)] [+T. −Agr] or [−T.” However.. +Agr] [+T. 2004). a single negative specification for [T] or [Agr] ([+T.5. As extensively argued by Ferreira (2000. As mentioned in section 2. it follows that indicative complements should not display obligatory control. in all the sentences of (41)–(45). Importantly. [43]). and Rodrigues 2004 for relevant discussion). Take the Brazilian Portuguese sentences in (41)–(45). a null subject in Brazilian Portuguese is interpreted as a bound variable when its antecedent is an only-DP (cf.24 (41) (42) ∗ Comprou um carro novo Bought a car new ‘She/he bought a new car’ [[o Jo˜ao] disse que [o pai d[o Pedro]] acha que vai The Jo˜ao said that the father of-the Pedro thinks that goes ser promovido] be promoted ‘Jo˜aoi said that [Pedroj ’s father]k thinks that hek/∗ i/∗ j/∗ l is going to be promoted’ (43) S´o o Jo˜ao acha que vai ganhar a corrida Only the Jo˜ao thinks that goes win the race ‘Only Jo˜ao is an x such that x thinks that x will win the race’ NOT: ‘Only Jo˜ao is an x such that x thinks that he. As for interpretation matters. the null subject displays the diagnostics of obligatory control despite the fact that it is within a standard indicative clause. 25 Referential null subjects in matrix clauses in Brazilian Portuguese can only be licensed as instances of topic-drop (see Ferreira 2000. 2009) and Rodrigues (2002. too (think that he. for instance. [44]). is going to win the race)’ (45) O infeliz acha que devia receber uma medalha The unfortunate thinks that should receive a medal ‘The unfortunate thinks that he himself should receive a medal’ Example (41) shows that null subjects in Brazilian Portuguese require an antecedent25 and (42). and it only admits a de se reading in sentences such as (45). 2004. it obligatorily triggers sloppy reading under ellipsis (cf. . 2004) for additional tests. Pedro. will win the race’ (44) O Jo˜ao est´a achando que vai ganhar a corrida e o The Jo˜ao is thinking that goes win the race and the Pedro tamb´em est´a Pedro too is ‘Jo˜ao thinks that he’s going to win the race and Pedro does. The existence of finite control into indicative complements in Brazilian Portuguese therefore presents prima facie problems for the typology proposed 24 See Ferreira (2000.5 Non-movement approaches to control 25 obligatory control.2. that the antecedent must be the closest c-commanding DP. Jo˜ao. Modesto 2000. ” However. . the [−R] feature makes PRO a potential goal for agreement.2.26 Below we discuss the implications of this empirical fact within Landau’s Agree-based approach. coindexing their features and licensing PRO’s [−R] feature. [CP C[−T] [IP PRO[−R] I1[−T. . after agreeing with the matrix subject. (i) Greek (Terzi 1997): I Maria1 prospathise Pro1/∗ 2 na divasi the Maria tried. which is given in (48). according to the rule in (46). let us consider the derivation of an exhaustive control construction such as (47) in Landau’s system.3SG ‘Maria tried to read’ .27 26 See Rodrigues (2004) for arguments that Finnish may also allow obligatory control into finite indicative clauses. Finally. Landau (2004: 841) also proposes that DPs must be featurally specified as to whether or not they support independent reference ([R]): lexical DPs and pro are specified as [+R] and PRO as [−R]. (47) John managed to fix the car (48) Agree [DP I2 [ . when occurring on nominal phrases. as represented in (ii). ␤Agr] ∈{I0 . According to Landau (p. 27 The only relevant difference between (48) and a typical obligatory-control subjunctive in Greek such as (i) in Landau’s system is that in the latter.5. I2 agrees with PRO. I1 has overt agreement morphology ([+Agr]). .}: Ø → [+R]/X0 [__] .26 Some historical background by Landau. The feature [R] is also assigned to some functional categories. . . if ␣ = ␤ = ‘+’ Ø → [−R]/elsewhere Given these assumptions.3 Determining the interpretation of obligatorily controlled PRO via Agree In addition to the features [T] and [Agr] to be hosted by C and I. C0 . 841). Agreement between C and I1 then deletes C’s [−T] feature. (46) R-assignment rule (Landau 2004: 842) For X0 [␣T. Agree is a way of achieving that” (p. . −Agr. 843). 2. . −R] [tPRO . rather than “abstract” agreement ([−Agr]). . for “this feature acts as an instruction to coindex the ␾-features of PRO with those of an antecedent. “[b]oth values on [R] are interpretable. tDP . ]]]]] Agree Agree Agree Agreement between I1 and PRO in (48) deletes I1 ’s [−Agr] and [−R] features. .3SG PRT read. a partial-control effect will arise. tDP . Landau assumes (p. . +R] [IP PRO[−R] I1[+T. [CP C[+T. . . he assumes Shlonsky’s (1997) proposal that third-person pros in Hebrew are null Num◦ heads and. . ]]]]] Agree As before. . . tDP .M well ‘Gil promised to behave’ (52) Agree [DP I2[+R] [ . +R] [tPRO . If I2 is specified as [−Mer] and PRO is inherently specified as [+Mer]. . Let us finally consider the last type of obligatory-control configuration listed in (39): Hebrew third-person subjunctives. agreement between I1 and PRO deletes I1 ’s [−Agr] and [−R] features. (49) (50) The chair hoped to meet at 6 Agree [DP I2[+T. which raises the question of how PRO can license its [−R] feature. tDP . . [CP C[+T.eci yitna’heg yafe] Gil promised that will-behave. . +Agr. +Agr. . −R] [tPRO . Agreement between C and I1 now deletes C’s [+T] and [+Agr] features. C then checks its [+R] feature with I2 . +R] [IP PRO[−R] Agree Agree I1[+T. a partial-control construction like (49) is to be derived along the lines of (50). +Agr.5 Non-movement approaches to control 27 In turn. . this feature gets licensed in virtue of I2 agreeing with C. Notice that I2 agrees with C and not with PRO. ]]]]] Agree (ii) Agree Agree Agree [DP I2 [ . . Furthermore.2. they cannot support a third-person feature hosted by a higher D-head. +Agr.28 Given the feature specification for Hebrew subjunctives in (39). because they are null.3SG. −R] [tPRO ]]]]] Agree Agree Agree 28 Landau (2004: 815. . 845). their [Mer] features need not match. −Agr. 846) attributes the lack of a derivation with an uncontrolled third-person pro in Hebrew subjunctives to the non-existence of referential third-person pro in the language. . but not its [+R] feature. the derivation of a sentence such as (51) proceeds as in (52). More specifically. . +Agr. for it mismatches the [−R] feature of I1 . [CP C[−T] [IP PRO[−R] I1[−T. According to Landau (p. . 849) that if I2 and PRO do not agree directly. which in turn is “coindexed” with PRO via I1 . +R] [ . (51) Hebrew (Landau 2004) Gili hivtiax [ˇse. . Even if we put aside the fact . “[t]he fact that C◦ bears [+Agr] does not stop this feature from entering Agree with [−Agr] of I◦ . As the reader can easily check. In other words. the feature specifications and computations proposed above are such that they track. and between I1 and PRO.” That may be so. and between I1 and PRO. but also reinforces the impression that these features are only redescribing the facts to be explained. The second type of relations encompassed by Landau’s version of Agree include “coindexing” relations such as the agreement between I2 and PRO in (48) to license PRO’s [−R] feature (which was assumed to be a [+interpretable] feature. but the resort to features which are motivated neither in LF nor in PF terms not only is completely at odds with core minimalist assumptions. it admits relations between two [−interpretable] features such as the agreement between C and I1 with respect to [Agr] features in (50) or the agreement between I2 and C in (52) with respect to [R] features. According to Landau (p. calling it honest does not make the analysis less stipulative. It is also worth pointing out that. under the label Agree. the remaining unchecked [−R] feature of PRO would be licensed by agreement with I2 . Landau (p. Thus. in addition to the familiar valuation procedure involving a [−interpretable] and a [+interpretable] feature of Chomsky (2001). Unfortunately.” which played the role of case in previous models. If the matrix I2 had a [−R] feature. but not its [+R] feature. However. it is subject to the same criticism made to the null-case approach: the distribution and interpretation of PRO ends up being stipulated under the guise of lexical features. between C and I1 . the distribution and interpretation of PRO.28 Some historical background Agreement between I1 and PRO in (52) checks the [+Agr] feature of I1 . between C and I1 . 842) in fact acknowledges that his Rassignment rule is an “honest stipulation. Landau’s proposal actually groups different kinds of relations. which do not obviously form a natural class. in (52) I2 is specified as [+R]. thus [Agr] on both heads is semantically uninterpretable and phonologically null. Finally. First. if the distribution and interpretation of PRO is to rest on a stipulation. it also includes composite-“coindexing” relations such as the licensing of the [−R] feature of PRO in (50) and (52). as mentioned above). recall that [+Agr] on C◦ represents abstract [Agr] to begin with (in most cases). which involves the conjunction of three basic agreement relations: between I2 and C. 849). The [−R] feature of PRO must then be indirectly licensed in virtue of the agreement relations between I2 and C. Agreement between C and I1 then checks all of the features of C and the [+R] feature of I1 . the Agree operation assumed by Landau encompasses three other types of relations. but do not explain. which mismatches the [−R] feature of PRO. Finally. 843) claims that the reason why PRO cannot be licensed in the indicative configuration in (53) below (Landau’s [40b]) is that “Agree fails due to a feature mismatch in the R value between I◦ and PRO. However. if A is taller than B and B is fatter than C. The reason for C not to be associated with [T] features is that the tense value of I is completely independent from the matrix clause. it cannot be associated with an [R] feature. which licenses A-traces by resorting to a chain coindexing mechanism combining Spec-head agreement with head-to-head government (see Chomsky 1986a. tDP . then one can conclude nothing regarding A’s height or weight as regards C. .5 Non-movement approaches to control 29 that “coindexing” and feature valuation/deletion seem to be of different nature. .” Recall that. . it is worth asking how transitivity arises given that the agreement relations computed do not target the same type of feature. Furthermore. That being so. the proposed composite-“coindexing” relations should be subject to the same skepticism we accord the Barriers approach to A-movement. 840) assumes that the presence of [Agr] on C is parasitic on [+T]. PRO agrees with I1 through some kind of transitivity assumption. +Agr. I2 and C agree with respect to [R] as do I1 and PRO. it cannot have [Agr] either. Landau (p. if indicative C does not have [+T]. . . but C and I1 agree with respect to [Agr]. Note that. As we saw in (39). . according to the R-assignment rule in (46). Thus.2. [CP C [IP I1[+T. . we must assume that this logic is overturned when certain feature sets are involved. but is rather a stipulated feature in Landau’s system. section 11). indicative clauses with independent tense universally do not display OC.” (53) [DP I2 [ . for the account above to work. if it is not specified for both features. which in turn brings the obvious minimalist question: why are these features endowed with their alleged properties? This shows that the proposed transitivity in the account of (51) does not follow as a point of logic. +R] [VP PRO[−R] . it is not at all trivial to explain how the composition of the three agreement relations mentioned should result in “coindexing. In virtue of these two agreement relations. since Landau (p. the feature specification proposed for indicatives involved the features [+T] and [+Agr] for I and no features for C. However. The non-explanatory nature of the proposal is further highlighted when Landau’s account of (51) is examined in light of his take on the impossibility of PRO in indicative clauses. ]]]]] Agree ∗ Agree This specific claim now introduces an additional aspect of compositeagreement relations: feature mismatch is taken to cause a derivational crash under direct agreement. in (50) and (52). like the relation between I1[+R] and PRO[−R] in (53). Thus. 30 Some historical background but not under “composite” agreement. . (55) Agree [DP I2[+R] [ . the [+R] feature of I◦ remains unchecked as no corresponding feature exists on the indicative C◦ . In a configuration like (40b) [= (53) above]. tDP . that I2 is specified as [+R]. Consider for instance the structure in (54). that is. +R] [VP tPRO . [CP C [IP PRO[−R] I1[+T. checking the [+R] feature of the latter. finite control into indicatives becomes freely available. Suppose. which had agreed with C. as represented in (55). . As we saw in the derivation proposed by Landau for Hebrew subjunctive control in (52). before PRO moves. In other words. That is. +R] [VP tPRO . which agrees with I2 . Given that the embedded subject and the embedded I are equidistant (see Chomsky 1995). as opposed to (43b) [= (52) above]. there should be no reason for PRO not to get licensed in (55) via a composite-“agreement” relation. by contrast. mismatch in the values for [R] by itself is not a problem if the features can be licensed later on in the derivation. it is important to point out that this stipulated aspect of composite agreement leads to overgeneration. As such. cannot be the reason for the derivation to crash. . ]]]]] Agree Agree If I2 in (54) is specified as [−R]. like the relation among I2[+R] -C[+R] I1[−R] -PRO[−R] in (50). once the composite-agreement relations proposed by Landau are assumed. +Agr.” Notice. Landau (2004: 846–847) seems to assume that the [+R] feature of I1 cannot be checked by a probe higher than C: “We still account for the fact that indicative complements in Hebrew do not display OC. which depicts the movement of PRO in (53). [CP C [IP PRO[−R] I1[+T. . for instance. it does not seem plausible to exclude the checking of the [+R] feature of I1 in (55) based on the . . . (54) [DP I2 [ . . Putting aside the fact that no motivation was provided for why these two instantiations of “Agree” should yield opposite results. for example. In fairness. that C does not prevent a higher probe from agreeing with the embedded subject in (48). causing the derivation to crash. +Agr. ]]]]] Agree Agree What about the [−R] feature of PRO? Recall from (50) and (52) that PRO can be indirectly licensed by a chain of “agreement” relations. however. but the [+R] feature of I1 will remain unchecked. Notice that the feature mismatch at the derivational step depicted in (53). tDP . . it should be able to agree with I1 . . composite agreement is assumed to be oblivious to feature mismatch. its [−R] feature should be licensed once PRO has agreed with I1 . Crucially. That being so. . which in turn had agreed with I2 . . its [−R] feature is taken to be licensed in virtue of PRO’s having agreed with I1 . it will be able to agree with PRO. In (52). . one may attempt to specify the inner workings of spell-out/transfer in such a way that PRO becomes immune to spell-out/transfer at the relevant derivational step.and R-) features are interpretable (hence. it is plausible to assume that spell-out/transfer must be halted until PRO has its features valued and.2. as discussed in section 2. This in turn indicates that the embedded I must still be available to the computation at the derivational step where PRO is to have its R-feature licensed. however. never erased). Of course.29 Nunes (2007. 2004) in fact analyzes null subjects in Brazilian Portuguese in terms of the MTC. footnote 26) claims that “[a]lthough not at the edge of its phase. such an attempt would require further complications. But that would only add to an already loaded machinery. Thus. A phase-based approach to this conundrum is of no help either. [52] for Hebrew and [ii] in footnote 27 for Balkan languages).and Rfeatures of PRO are only valued after agreement. control into indicative clauses becomes freely allowed. Ferreira (2000. 2008a) reinterprets Ferreira’s proposal in terms of the presence or absence of the feature [person] in T. Landau (2004. in his system a ␾-incomplete T does not value the case feature of the subject of its clause.2. under composite-agreement relations in Landau’s system (cf. He observes that the verbal-agreement paradigm of finite clauses in Brazilian Portuguese is such that the only inflection that overtly encodes both number and 29 Ferreira (2000. if this is so. Although this may be good news for languages such as Brazilian Portuguese.5.5 Non-movement approaches to control 31 intervention of C. which can then undergo A-movement to the matrix clause. it is certainly unwelcome for most languages. that in Landau’s system the ␾.4 Simplifying Landau’s “calculus of control” Let us examine what the relevant property of Brazilian Portuguese indicative clauses is that triggers obligatory control. Still. 2009) (as well as Rodrigues 2002. We will leave a detailed discussion of null subjects in Brazilian Portuguese under the MTC for section 4. To wrap up: if composite relations must be assumed in order to account for Hebrew subjunctive control. 2.4 below. . without actually shedding light on the discussion. 2009) proposes that finite Ts in Brazilian Portuguese are ambiguous in being associated with either a complete or an incomplete set of ␾-features and that obligatory control is licensed in clauses with a ␾-incomplete T. but not I1 .” Note.2. Recall that. PRO is visible to Agree from the outside since its (␾. 2004.2. 2004. Notice also that C in (55) has no features that could block an agreement relation between a higher probe and I1 . we are back to the technical question of why I1 in (55) cannot be checked by the matrix probe if spell-out/transfer is on hold. “which is also coindexed with PRO via I◦ ” (Landau 2004: 845).5. the higher probe agrees with the embedded C. Thus. as illustrated in (56). if T has only a number feature and it is valued as singular in the syntactic component. N:PL (= 3PL) elas ‘they (FEM)’ Nunes proposes that ␾-complete and ␾-incomplete finite Ts in Ferreira’s terms correspond to Ts specified with number and person features or a number feature only. N:default (= 3SG) vocˆe ‘you (SG)’ ele ‘he’ ela ‘she’ a gente ‘we’ vocˆes ‘you (PL)’ eles ‘they (MASC)’ cantam P:default. This can be . N:SG canta P:default. (56) Verbal-agreement paradigm in (colloquial) Brazilian Portuguese cantar ‘to sing’: indicative present eu ‘I’ canto P:1. the corresponding person specification will be added in the morphological component by redundancy rules. as sketched in (57) below. (57) cantar ‘to sing’: indicative present Valuation of T in the syntactic component Addition of [person] in the morphological component Surface form of the verb N:SG P:1. it will later be associated with a default value for person (third) in the morphological component. In case a T with just a number feature is selected. subjunctive control in Hebrew is restricted to the third person. it will later be associated with first person in the morphological component. if the number feature receives any other value in the syntactic component (default or plural). N:PL cantam If finite control in Brazilian Portuguese is related to the possibility that its finite Ts may be specified only for number in the syntactic component. we now find a commonality with Hebrew subjunctive control. That is. N:default canta N:PL P:default. N:SG canto N:default P:default. As argued by Landau (2004). All the other cases involve either number specification with default value for person (third) or default values for both person and number (third singular).32 Some historical background person is the first-person singular inflection. but drastically simplifies the “calculus of control” in Landau’s terms. finding out whether or not a given clause licenses obligatory control does not need to take the features of C into consideration. − ␾ ] untensed uninflected infinitives. − ␾ ] tensed uninflected infinitives. +Agr] and I[+T. third person. . ␾+ ] English indicatives.4 below for further discussion). + ␾ ] Balkan untensed subjunctives.2. this is also a welcome result. where ‘+’ stands for fully specified and ‘−’ for deficient (or null). This in turn paves the way for a considerable simplification in Landau’s typology. No control − [T . for I is sufficient: if either [T] or [␾] is deficient. Balkan tensed subjunctives. etc. etc. (58) also eliminates the suspicious ambiguity of the combination of the specifications C[+T. why should the PF realization of agreement features matter? From the perspective of (58). or both. etc. Landau’s table in (39) can now be revised as in (58). [T+ . which were employed to describe both obligatory control in Hebrew subjunctives and no control in Balkan F-subjunctives (see section 4. If obligatory control is to be ultimately determined in the syntactic component. + [T . i. Under (58). In other words. Hebrew 3rd-person subjunctives. regardless of their later morphological realization. ␾-deficient. the availability of obligatory control is determined by the tense and agreement features that enter in the syntactic component. that is.. +Agr] in Landau’s table in (39). (58) Obligatory control [T − .e. Conceptually. what matters in (58) is not the morphological realization of agreement features.5 Non-movement approaches to control 33 interpreted as indicating that the relevant subjunctive T in Hebrew may also be associated with only a number feature. with the desired effects. As opposed to Landau’s system. obligatory control is possible. Besides simplifying Landau’s system and accounting for finite control into indicatives in Brazilian Portuguese. etc. but rather how specified the set of agreement features associated with I is. in which case it will surface with default-person morphology. heads that are temporally deficient. Brazilian Portuguese indicatives. the environments where one finds obligatory control involve deficient T-heads. The table in (58) shares with Landau the intuition that obligatory control is typologically more diverse. 34 Some historical background The question that now arises is why obligatory control should correlate with deficiency in tense or ␾-feature specification. We return to this correlation between movement/obligatory control and Infldeficiency in section 4. etc. instead of an obligatory-controlled PRO with an independent . the picture embodied in (58) is exactly what one would expect: we can simply replace control by A-movement. it would not be surprising if Landau’s R-assignment rule in (46) could be reformulated in such a way that it should become compatible with the generalizations embodied in (58). Given that the distribution of PRO is handled in such a stipulative manner.g.. We have seen that.2. However.5. as in (59). etc. Landau’s Agreebased approach to control involves a rich array of features that allows him to capture many manifestations of control. Here we have focused on three major pillars of his proposal: (i) the importance ascribed to partial control. which mimicked the case-based approach to PRO in previous models. ␾+ ] English indicatives. notice that tense or ␾-feature deficiency generally characterizes “porous” domains out of which movement can take place (see e. (ii) the typology predicted by his feature system. Hebrew 3rd-person subjunctives. (59) A-movement: √ A-movement: ∗ [T− . Brazilian Portuguese indicatives. 2. 2005). and (iii) the technical details of how the distribution and interpretation of PRO is to be obtained through the operation Agree.4. Boeckx 2003. given their sensitivity to the argument properties of the embedded predicate.5 Summary Combining aspects of syntactic and semantic approaches. Balkan untensed subjunctives. tensed uninflected infinitives. ␾− ] [T+ . ␾+ ] untensed uninflected infinitives. etc. Thus. from the perspective of the MTC. something that we will not pursue here. [T+ . We have seen that Landau’s Agree-based approach was couched on the admittedly stipulated postulation and distribution of [R]-features. ␾− ] [T− . with a very high degree of formal explicitness. partial-control constructions may also be conceived as involving a null commitative argument. Balkan tensed subjunctives. as depicted in (58). etc. 6 Conclusion 35 semantic plural feature. despite their laudable attempts to deduce the distributional and interpretive properties of PRO. the system undergenerates in that it has no room for finite control into indicatives. from the standard theory to minimalism. 2. With this background.2. the existence of partial control is not by itself an argument for PRO-based theories. . we are now ready to examine the major properties of the MTC. All in all. it accounts for the distribution and the interpretation of PRO by ultimately encoding the facts to be explained in the guise of stipulative lexical features. 842) that the theoretical foundations for his approach are on equal footing with the Case-based approach in previous models. but as the output of the computations of the syntactic component. Accordingly. the null-case and Agree approaches take PRO to be a lexical item and. given a minimalist setting. Finally. they end up simply encoding them as lexical features. each of them attempted to account for the distribution and interpretation of PRO in terms of the broad architectural properties of the model of UG then assumed. By contrast. Despite its technical precision and empirical coverage. which is allowed in Brazilian Portuguese. thereby eschewing true explanation.6 Conclusion In this chapter we have discussed different approaches to control within the generative enterprise. It is an interesting fact that the standard theory and the GB approaches took PRO not as a lexical formative. we agree with Landau (p. In other words. As for the typology predicted. the technical apparatus rests on various stipulations regarding the properties of features needed to track the distribution and interpretation of PRO and on composite-agreement relations that are not independently motivated and lead to overgeneration. our fresh minds – unbiased but armed with Occam’s razor – would undoubtedly attempt to capture them in a uniform way. as we shall discuss. Both sentences involve a matrix predicate that embeds a non-finite sentential complement and. unless presented with strong independent reasons for not doing so. we present the basic features of (our version of) the MTC. The seduction of this simple reasoning encapsulates the MTC.1 Introduction If we could start afresh. Section 3. Section 3. Of course. more interestingly. despite claims to the contrary. John seemed to kiss Mary John tried to kiss Mary In face of these structural and interpretive similarities. we would likely be struck by the similarities between sentences like (1a) and (1b) below. But the incorrectness has to be demonstrated and this – in our view – has not been the case with the MTC. leaving a detailed discussion of its empirical advantages and its solutions to problems raised in the literature to chapters 4 and 5. b.4 shows how an analysis 36 . without our historical baggage and the preconceptions that often come with it. and frequently are. In this chapter. The MTC takes it that the null hypothesis for the derivation of raising and control constructions such as (1a) and (1b) should resort to the same grammatical devices. with the same mechanisms. (1b) should prima facie be analyzed as involving A-movement as well.3 shows how the abandonment of D-structure in the minimalist program made it possible and natural to explore the null hypothesis underlying the MTC. (1) a.2 starts with a historical discussion of factors that prevented the MTC from being entertained as the null hypothesis from day one. the unrealized subject of the embedded clause is interpreted as being “the same” as the subject of the matrix clause. Section 3. if (1a) is to be analyzed in terms of A-movement. That is. Thus. ‘John’ is the kisser in both (1a) and (1b).3 Basic properties of the movement theory of control 3. null hypotheses can be. incorrect. Example (1a) is taken to mediate the relation between the two subject positions by moving ‘John’ from the embedded to the matrix position. he is in no sense a seemer. similar distinctions were made in earlier periods using different technology. 3. section 3. Finally. we might say that in (1b) John is described as both a kisser and a trier while in (1a). though he is a kisser still. as represented in (2b). [John1 seemed [t1 to kiss Mary]] [John1 tried [PRO1 to kiss Mary]] It is important to note that enriching the theoretical apparatus by having UG invoke different grammatical resources in order to capture the interpretive difference between (1a) and (1b) is not the only conceivable option. leaving behind a coindexed trace. for they have generally taken the semantic difference revealed by the paraphrases to indicate that these otherwise similar sentences have entirely different generative (derivational) profiles.’ while (1b) has (at best) the very awkward paraphrase ‘John tried for John to kiss Mary’ and no possible paraphrase analogous to the one for (1a): ‘it tried for John to kiss Mary’ is almost incomprehensible. have been more impressed by the last noted difference than the aforementioned similarities. (1b) is taken to relate ‘John’ to the embedded-subject position through some kind of binding relation. one might imagine keeping the theoretical apparatus 1 Unless it is relevant to the discussion. and empirical reasons Consider once again the examples in (1). Infelicitously. By contrast. phonetic.2 Departing from the null hypothesis: historical. Also for presentational purposes. For all their similarities. .3. as illustrated in (2a) below. there is a difference between the two sentences.2 Departing from the null hypothesis 37 of controlled PRO as a trace of A-movement deduces the configurational.1 (2) a.1. as we saw in section 2. This is reflected in the fact that (1a) has a paraphrase like ‘it seemed that John kissed Mary. it is fair to say. As already suggested in section 3. architectural. Generative grammarians. and interpretive properties of obligatory control and how obligatorily controlled PRO can be dispensed with as a grammatical formative. we will employ GB representations in terms of traces and PRO when nothing is at stake.3. In (1b) ‘John’ has an interpretive function in virtue of being the matrix subject that is absent in (1a). we abstract from the VP internal-subject hypothesis in this chapter. showing that there exist no strong reasons for discarding the null hypothesis regarding the derivation of control and raising constructions.5 reviews the architectural features of the MTC and its place within the minimalist program. b. One important legacy of the GB era with the shift from constructions and rules to principles and parameters is that constructions are now viewed as epiphenomena. Given the pervasive role of movement in the grammar – however it is encoded – Occam’s razor should urge us to attempt to make do with movement. as we noted above. resulting from the interaction of more basic operations. the passive rule. given that it is already independently required. the interpretive difference between (1a) and (1b) is to be ascribed to the indisputable fact that there is an additional ␪-role available in the matrix clause of (1b) (the “trier” ␪-role). transformations were complex operations that generated alternative structures typed by construction (e.38 Basic properties of the movement theory of control constant and analyzing (1b) also in terms of movement. as illustrated in (3) below. etc. we now find basic operations such as Move ␣ (or. simply attaining descriptive adequacy was a tremendous challenge as the basic formal tools to handle linguistic data were still in the making.g. it may establish a new thematic relation in (1b). is arguably the null hypothesis for the analysis of (1b). see section 2. raising sentences like (1a) do differ from control sentences like (1b) as far as the thematic powers of the subject ‘John’ are concerned. In this scenario. but not in the matrix clause of (1a). It is therefore unsurprising that these first stages were essentially taxonomic. the wh-question formation rule. Interestingly. One very good diagnostic for differentiating two constructions is their differing effects on meaning and.. this can be attributed to historical reasons regarding the development of the field. establishing the inventory of possible constructions in natural languages and formulating the rules that should yield the catalogued constructions.3) makes good sense. Therefore. Thus. In part. the topicalization rule. In the earliest days of generative grammar. the relativization rule. this has not been a widely explored path in generative grammar2 and it is worth pausing to consider why this is so. differentiating a raising rule from a control rule (the equi(valent) NP deletion rule. . Instead of a (roughly) one-to-one relation between constructions and rules. but not in (1a). more radically still. as ‘John’ moves to the matrix clause. until the early 1980s. (3) [John1 tried [t1 to kiss Mary]] The analysis in (3). From this perspective. ceases to be persuasive with the emergence of the principles-and-parameters approach.). Lasnik and Saito’s [1992] Affect ␣) underlying the derivation of a multitude of different types 2 An early notable exception is Bowers (1973). though reasonable in this background. which essentially embodies the MTC. This motivation. in which constructions are not theoretical primitives. In fact. but in the interaction of their specific properties with other grammatical components. For instance. Once PRO is analyzed as a pronominal anaphor. ␪-theory). But once one goes this far. which must somehow disregard the pronominal specification of PRO in OC constructions and its anaphoric specification in NOC constructions. the control module. the derivation of passive and raising constructions is taken to employ the same grammatical device. The three constructions could potentially be derived by the same grammatical tool (Amovement) and their differences allocated to different components of the grammar.3.4 that the analysis of control in GB could not ultimately be reduced to binding/construal without additional provisos. invites one to eliminate the exceptional theoretical status of control qua construction in the grammar.. Such construction sensitivity looks like a fossil from previous stages of the generative enterprise that one would like to get rid of. there is no obvious conceptual barrier to categorizing control with passive and raising. the principles-and-parameters perspective.g. there remains no logical impediment for entertaining the hypothesis that control structures are derived by movement. all sharing the same generative resources. the account of the interpretive properties of PRO requires a specific grammatical module. rather than resorting to the distinct rules of passive and raising. In section 3. has to do with the general architecture of the grammar standardly assumed prior to the minimalist .2 Departing from the null hypothesis 39 of constructions. these differences do not undermine the claim that their derivations both involve a common (A’-)movement operation. Recall from our discussion in section 2. The source of their differences may reside not in the operations that go into generating them.3 and Chapter 5 below. whatever differences the two constructions enjoy (and there are many). The important point to bear in mind here is that with the abandonment of constructions as grammatical primitives. The second historical reason for why the MTC was not pursued within generative grammar. we will examine in detail potential sources for the differences between control and raising constructions documented in the literature. Nonetheless. Let us make the same point in a slightly different way: nobody assumes that wh-questions are the same as relative clauses. (A-)movement. namely. The MTC asks that this same reasoning be applied to raising and control configurations. it is now widely accepted (among generative grammarians) that. which also accounts for why the construction-specific flavor of the control module was tolerated within GB. The differences between these constructions such as the dethematicization of the external argument in the case of passives are then factored out and analyzed in terms of other independent components of the grammar (e. g. a movement-based derivation along the lines of (5b) for the control construction in (4b) is ruled out at DS. it codes all the relevant argument-structure information a sentence expresses. a passive or unaccusative verb).3 Let us see why. for current purposes the differences are insignificant as are the differences between various models of grammar that included a Dstructure level. More specifically. b. By contrast. phrase-structure rules and lexical-insertion operations) and the input to the transformational component. Hence. if a verb has a logical object but no subject (e. the object position must be lexically filled at DS. However. First. DS has two important properties. but there is no category filling this position in (6b). b. functionally. 4 See Chomsky (1981: 43). In other words. with ‘John’ generated in the embedded-subject position. the DS representation of (5a) and (5b) should be as in (6) below.. whereas the subject position must be empty. The second important aspect of DS is that. if a predicate has a logical subject and object (agent and theme). ‘kiss’ has two arguments and its subject and object positions are correctly filled.. John seemed to kiss Mary John tried to kiss Mary (5) a. in models that include a DS level all lexical-insertion operations precede all movement transformations.g. its subject position is left empty. which indicate that ‘John’ has moved from the embedded to the matrix-subject position.4 Thus. a pure “representation of GF-␪. [John1 seemed [t1 to kiss Mary]] [John1 tried [t1 to kiss Mary]] Given the requirement that lexical insertion must precede movement. ‘try’ does assign a ␪-role to its subject position. ‘Seem’ in (6a) does not assign a ␪-role to its subject position and. in technical lingo.40 Basic properties of the movement theory of control program. accordingly. e. (4) a. as the thematic requirements of ‘try’ are not 3 Deep structure is not quite the same as D-structure. the assumption of D(eep)-structure (DS) as a level of representation left no room for an approach along the lines of the MTC. then both subject and object positions must be lexically filled at DS. It is. for instance. In both (6a) and (6b).” where all and only argument (thematic) positions are filled. it is the output of phrase-building operations (namely. leaving a trace behind. . EST and both early and late GB models. In prior models. Let us now examine how the sentences in (1) repeated below in (4) are analyzed under these assumptions. Take the representations in (5). On the other hand. ‘there’ can then be inserted in the subject position of ‘seem’ after DS without semantic (or grammatical) violence being done. as illustrated in (7a). b. as in (11c). Take the contrasts in (8)–(10). In contrast. If ‘there’ fills this position already at DS.3.’ the thematic properties of ‘try’ will not be satisfied at DS. Notice that a similar DS representation for raising constructions. as in (11b) and later filled with ‘there. we again have an illicit DS representation. is not licit. as illustrated in (7b). (9) a. .5 As the subject of ‘seem’ is not thematic. (10) a. assuming DS in the grammar unavoidably leads to a movement analysis of raising and a construal analysis of control. 5 At least not conventional ones. b. b. with ‘John’ occupying the matrix-subject position and PRO the embedded-subject position. we can find an expletive in the subject of ‘seem’ but not of ‘try. We set aside for the nonce the “quasi”-argument status of ‘it’ in weather constructions like ‘it is raining’ (see Chomsky 1986b for discussion). at DS it must be filled by a category that can bear a ␪-role. If it is left empty at DS. In other words. as the subject position of ‘try’ is thematic. b. b.2 Departing from the null hypothesis 41 satisfied at this level. The derivation of (4b) should therefore have two distinct elements filling the subject positions. for ‘there’ is not a valid ␪-role bearer. it must be empty at DS.’ Why? Because expletives have no semantic content and so cannot bear ␪-roles. DS: [Seemed [John to kiss Mary]] DS: ∗ [Tried [John to kiss Mary]] (7) a. (8) a. as shown in (11a) below. (6) a. ∗ There seems to be someone kissing Mary There tried to be someone kissing Mary The cat seems to be out of the bag (idiomatic interpretation: OK) The cat tried to be out of the bag (idiomatic interpretation: ∗ ) The doctor seemed to examine Mary ∼ Mary seemed to be examined by the doctor The doctor tried to examine Mary = Mary tried to be examined by the doctor In (8). DS: ∗ [John seemed [PRO to kiss Mary]] DS: [John tried [PRO to kiss Mary]] The big empirical virtue of assigning distinct DS representations to raising and control constructions along the lines of (6a) and (7b) is that it derives semantic differences between these constructions in a principled manner. as the subject position of ‘seem’ is filled despite the fact that it is not thematic. for example. both meaning that it seems that the doctor examined Mary. (13) a. though (9b) is well formed and interpretable. (12) a. b. the control counterparts in (10b) have completely different meanings.’ It has nothing to do with secrets revealed or otherwise. that DS is substantively the level at which all and only thematic positions must be filled and functionally the level that feeds movement operations) conspire to eliminate a movement approach to control along the lines of (15) below at DS. In (9a) ‘the cat’ can retain its idiomatic meaning. then an idiom (or part of one) cannot be simultaneously interpreted idiomatically and thematically. b. DS: [The doctor tried [PRO to [examine Mary]]] DS: [Mary tried [to be [examined PRO] by the doctor]] In sum. This is one . as shown in (12b). Thus. This is not possible in (9b). Once ‘the cat’ is in a thematic position. it cannot have an additional idiomatic interpretation. b. As illustrated below. By contrast. with the effort being made by the doctor in the first sentence but by Mary in the second. it means something like ‘the kitty tried to escape the confining sac. c. as represented in (12a) below. They display “voice transparency” in the sense that passivizing the embedded clause has little effect on the interpretation of the whole. as illustrated in (13) below. (9a) can be interpreted as meaning that it seems like the secret has been revealed. DS: [Seems [to be someone kissing Mary]] DS: ∗ [Tried [to be someone kissing Mary]] DS: ∗ [There tried [to be someone kissing Mary]] We can play the same game with the idioms in (9). the two features of DS reviewed above (namely. Idioms are not compositionally interpreted. DS: [Seemed [the doctor to [examine Mary]]] DS: [Seemed [to be [examined Mary] by the doctor]] (14) a.42 Basic properties of the movement theory of control (11) a. at DS ‘Mary’ is the thematic object of ‘examine’ in (14a) but the thematic subject of ‘try’ in (14b). as the subject of ‘seem’ is not thematic and is left empty at DS. DS: [Seems [[the cat] to be out of the bag]] DS: [[The cat] tried [PRO to be out of the bag]] The pairs of sentences in (10) offer a final illustration of the same point. This transparency is captured at DS as ‘Mary’ fills the object position of the embedded verb in the DS representation of both sentences. If we take this to mean that they cannot bear conventional ␪-roles. The sentences in (10a) are rough paraphrases of one another. Lack of voice transparency in (10b) is attributed to the different positions ‘Mary’ occupies at DS in each case. Consequently. b. can be freely interspersed. is to be associated with the (simplified) derivation in (17) (see footnote 1). among others. on the one hand. But questions arise if the architecture of the model changes. without DS there is no obvious architectural reason against reducing raising and control to a common movement source.3 Back to the future: elimination of DS and the revival of the null hypothesis As we discussed in section 3. where movement of ‘what’ in (17e) is sandwiched between different applications of lexical insertion and ␪-assignment. as minimalism argues against the postulation of non-interface levels such as DS. and Hornstein. 3. Specifically. Uriagereka (1998). A sentence such as (16). 6 For relevant discussion. This observation has become especially relevant with the emergence of the minimalist program in the early 1990s. and movement.3 The revival of the null hypothesis 43 reason why the MTC was not considered viable in GB or in previous models that assumed DS. . the interaction of these two features of DS derives subtle interpretive properties of raising and control structures and this was certainly a big achievement. Models that eschew a DS level are therefore free to pursue the MTC. So much so that retaining the clumsy construction sensitivity of the control module in a principles-and-parameters model seemed a reasonable price to pay.3.2. on the other. (15) [John1 tried [t1 to kiss Mary]] Furthermore. which in practice prevents the controller and the controllee from being associated via movement. expressed in (15)? This is the subject of the next section. minimalists have explored the idea that lexical insertion and ␪-assignment. As DS requires that all ␪-roles must be discharged before any movement takes place. assuming DS has drastic consequences for the (simplest) version of the MTC expressed in (15). see Chomsky (1995). the ␪-roles associated with the controller and the controllee must be assigned before any movement operation. Nunes. In other words. for instance. are we not then free to reconsider the null hypothesis regarding control. and Grohmann (2005). what if we give up DS? Can the interpretive differences between raising and control still be captured? If so.6 In particular. for instance. See Chomsky (1995. 2004. j. 2000. (18) John tried to kiss Mary (19) a. does. Once it is independently assumed that ␪-role assignment may follow applications of movement. this merger operation should in principle license ␪-assignment in the same way merger of ‘Mary’ in (19a) or ‘John’ in (19c). Merger of ‘saw’ and ‘what’ + θ-assignment: [saw what] Merger of T: [T [saw what]] Merger of ‘John’ + θ -assignment: [John [T [saw what]]] Merger of ‘that’: [that [John [T [saw what]]]] Movement of ‘what’: [whati [that [John [T [saw ti ]]]]] Merger of ‘say’ + θ -assignment: [say [whati [that [John [T [saw ti ]]]]]] Merger of T: [T [say [whati [that [John [T [saw ti ]]]]]]] Merger of ‘Mary’ + θ -assignment: [Mary [T [say [whati [that [John [T [saw ti ]]]]]]]] Merger of C: [C [Mary [T [say [whati [that [John [T [saw ti ]]]]]]]]] Movement of ‘what’: [what [C [Mary [T [say [whati [that [John [T [saw ti ]]]]]]]]]] b.44 Basic properties of the movement theory of control (16) What did Mary say that John saw (17) a. it is not at all odd to suppose that a control construction such as (18) below could be derived along the lines of (19).7 That is. 2001. if movement of ‘John’ in (19g) involves plugging it into the structure via merger. in press). h. where the “trier” ␪-role is assigned after movement of ‘John’ (cf. . and Hornstein (2001) for relevant discussion. Merger of ‘kiss’ and ‘Mary’ + θ-assignment: [kiss Mary] Merger of T: [T [kiss Mary]] Merger of ‘John’ + θ-assignment: [John [T [kiss Mary]]] b. 7 This holds whether move is identical to merge (is internal merge) or contains it as a subpart (involves copy and merge). c. f. Nunes (1995. e. c. This is even more so if movement is in fact a composite operation that includes merger as one of its basic operations. g. [19g]). d. i. 2001). the null hypothesis. b. f. The doctor seemed to examine Mary ∼ Mary seemed to be examined by the doctor The doctor tried to examine Mary = Mary tried to be examined by the doctor b. The question that we now have before us is empirical. (23) a. As we can verify in the relevant representations of the control structures in (23). d. ∗ [Therei tried [ti to be someone kissing Mary]] [[The cat]i tried [ti to be out of the bag]] [[The doctor]i tried [ti to [examine Mary]]] [Maryi tried [ti to be [examined ti ] by the doctor]] In (23a). Finally. Moreover.3 The revival of the null hypothesis d. g. hence the unacceptability of (20b). once DS is removed. movement of ‘the cat’ to the thematic matrix-subject position in (23b) will only be licit if ‘the cat’ is not interpreted idiomatically in the embedded clause (cf. b. but it is not a licit ␪-role bearer. 45 Merger of C: [C [John [T [kiss Mary]]]] Merger of ‘tried’ + θ -assignment: [tried [C [John [T [kiss Mary]]]]] Merger of T: [T [tried [C [John [T [kiss Mary]]]]]] Movement of ‘John’ + θ-assignment: [Johni [T [tried [C [ti [T [kiss Mary]]]]]]] e. the MTC arises as a theoretical possibility worth considering – in fact. ‘Mary’ is associated only with the “examinee” ␪-role in (23c). ∗ There seems to be someone kissing Mary There tried to be someone kissing Mary (21) a. PRO). ‘there’ moves to a thematic position. The cat seems to be out of the bag (idiomatic interpretation: OK) The cat tried to be out of the bag (idiomatic interpretation: ∗ ) (22) a. it becomes. Under the assumption that a given element cannot be simultaneously interpreted as a thematic argument and an idiom chunk. but with both the “examinee” and “trier” ␪-roles in (23d).3. we believe. c. [21b]). repeated below in (20)–(22)? (20) a. the answer is affirmative. b. are accounted for by invoking the same or comparable . which were used to motivate the derivation of raising and control in terms of distinct grammatical devices (traces vs. In sum. Can the analysis sketched in (19) account for the differences between raising and control illustrated in (8)–(10). semantic contrasts such as the ones in (20)–(22). hence the two sentences in (22b) are not paraphrases of one another. Chomsky (2004). movement) has so far also proven to be empirically adequate. Nunes. OC PRO requires an antecedent: [It was hoped [PRO1 to shave himself]] Its antecedent must c-command it: ∗ [John1 ’s campaign hopes [PRO1 to shave himself]] Its antecedent must be local: ∗ [John1 thinks [that it was hoped [PRO1 to shave himself]]] It cannot appear in case-marked positions: ∗ [John1 said [(that) PRO1 will travel tomorrow]] It gets a sloppy interpretation under ellipsis: [John1 wants [PRO1 to win]] and [Bill does too] (‘and Bill wants himself to win’/∗ ‘and Bill wants John to win’) ∗ b. 8 See Hornstein.3.2) for relevant discussion. In fact. Chapter 8 for discussion). . the merge operation that is not part of movement (“internal” merge). the MTC unifies form and meaning. e. Before we close this section and examine other consequences of the analysis outlined thus far. section 2. In this sense.4 Controlled PROs as A-movement traces By treating OC PROs as traces of A-movement (movement to a thematic position).8 However. it is important to observe that we are not claiming that the MTC follows from the abandonment of DS (although the “simplest” removal of DS does entail this. we examine in the following sections the distinctive characteristics of OC PRO listed in (24) (see section 2. Not only does it account for the lack of phonetic realization of PRO and the distribution of PRO and its controller. This shows that dropping DS does not entail movement into thematic positions. c. In other words. one might be tempted to suggest that any model that rejects DS on principled grounds should welcome the unification of raising and control by seeing both as products of movement. the MTC fits rather comfortably with the broad architectural features of the minimalist program. for instance.4). but it also goes a fair way to accounting for PRO’s interpretive properties. proposes that all thematic information must be discharged via “external” merge. eliminating DS is a necessary condition for the viability of the MTC. as we shall discuss in detail later.46 Basic properties of the movement theory of control assumptions that a PRO-based DS approach must resort to..2. (24) a.e. To illustrate. and Grohmann (2005. that is. 3. d. especially if there is almost no (or rather little) empirical cost in doing so. the use of the minimal technical apparatus independently required (i. cf. ’ (26) a. ∗ [John1 was hoped [PRO1 /t1 to shave himself]] 11 We return to this issue in section 4. where we argue that they are excluded due to a minimality violation. b.4. (i) a.1 below.3. 10 There arises the question of why the sentence in (ia) below cannot be derived along the lines of (ib). fails to license the anaphor. for instance.4 Controlled PROs as A-movement traces 47 f. ∗ John was hoped to shave himself b. where the antecedent does not c-command the A-trace in (26a) and is not local in (26b) due to the intervention of ‘it. PRO requires an antecedent. but if it takes the expletive as its antecedent. therefore. where the trace must take ‘it’ as its antecedent and. Under the standard assumptions that movement (in general) targets a c-commanding position11 and that A-movement in particular is very local. 1970) minimal-distance 9 As discussed in section 2. Koster (1987). the ungrammaticality of (24b) and (24c) reduces to the ungrammaticality of the representations in (26). We postpone the discussion of sentences such as (i) to section 5. where the similarities are expressly emphasized. ∗ ∗ [[John1 ’s sister] was hired t1 ] [John1 seems [that it was likely [t1 to shave himself]]] The analysis of the ungrammaticality of (24c) in terms of minimality is in essence a reincarnation of Rosenbaum’s (1967. where we discuss cases of movement to non-ccommanding positions in adjunct-control constructions. . where the OC PRO/A-trace does have a (local c-commanding) antecedent. It has an obligatory de se interpretation in “unfortunate” contexts: [[The unfortunate]1 expects [PRO1 to get a medal]] (#although he doesn’t expect himself to get a medal) h.9 In (24a).4. the parallels between OC PRO and NP-traces have been often noted.5.’ The same situation is found in (25). It must receive a bound reading when linked to an only-DP: [[Only Churchill]1 remembers [PRO1 giving the BST speech]] (‘Only Churchill is such that he remembers himself giving the BST speech’/ ∗ ‘Nobody else remembers that Churchill gave the BST speech’) ∗ 3.1 Configurational properties The properties illustrated in (24a–d) are standard properties of traces of Amovement (NP-traces in GB terminology).2 below. it cannot license ‘himself. It cannot have split antecedents: [John1 asked Bill2 [PRO1+2 to shave themselves/each other]] g. See.10 (25) ∗ [It1 was expected [t1 to shave himself]] The similarity also extends to the c-command and locality restrictions on the position of the controller. for example. according to which an element is active for purposes of A-movement only if it has not checked/valued its case feature. A-chains are uniformly associated with a case position (see section 5.3).4). Take the object-control construction in (27) below.5.4.5. We will return to the case properties of finite-control constructions in section 4. 12 We return to a discussion of apparent counter-examples such as (i) in section 5.5. notice that here we don’t have two subspecies of A-chains (the canonical ones. which have to be headed by a case position. otherwise the activation condition is violated.48 Basic properties of the movement theory of control principle (see section 2. Under the MTC.2. Brazilian Portuguese allows obligatory control into indicative finite clauses. as ‘Mary’ intervenes. As we saw in sections 2. if OC PRO is an A-trace. . it must occupy a caseless position.4 below for further discussion). the structure of the matrix vP in (27) should be as in (28). as shown in the “hyper-raising” case in (29).2.2 and 2.1 below. it is also a property that applies to A-traces. If it were the trace of ‘John. as was the case in GB (see section 2.’ movement of ‘John’ from the embedded-subject position to the matrix [Spec. for instance. If PRO is a trace. it must be the trace of ‘Mary’ in (28). (i) [Johnk promised Maryi [PROk/∗ i to go]] 13 Notice that this does not entail that OC PRO/trace is always excluded from the subject position of finite clauses. where ‘John’ moves from a position associated with nominative case. vP] would violate minimality.4 below. and the exceptional ones headed by PRO. (29) ∗ [John1 seems [(that) t1 will travel tomorrow]] Under current minimalist technology.12 (27) [Johnk convinced Maryi [PROi/∗ k to leave]] (28) vP v’ Johnk convincedw + v VP Maryi V’ tw [PRO i/∗ k to leave] As for (24d). Again. Under an (updated) Larsonian representation of ditransitive constructions. the fact that one does not find traces in case-marked positions follows from Chomsky’s (2001) activation condition. which are not subject to this requirement).13 Finally. 3.4 Controlled PROs as A-movement traces 49 3.4.2 Interpretive properties The requirement that ellipsis involving OC have a sloppy reading (cf. [24e]) tracks what we find in raising constructions, as exemplified by (30) below, where the second conjunct is understood as ‘Bill also seems to be cooperative.’ Regardless of how ellipsis is to be ultimately analyzed, the similarities of interpretation between (24e) and (30) can receive a straightforward account if they are associated with the same type of dependency relation (i.e., movement), as represented in (31). (30) John seems to be cooperative and Bill does too (31) a. b. [John1 wants [t1 to win]] and [Bill does too] [John1 seems [t1 to be cooperative]] and [Bill does too] The prohibition against split antecedents (cf. [24f]) also finds a straightforward account under the MTC. From the point of view of the MTC, if ␣ is the antecedent of PRO, then ␣ must have moved from the position occupied by PRO, that is, PRO is the trace of the so-called antecedent. That being so, split antecedents would only be possible if two DPs could move from one and the same position. However, a standard assumption within generative grammar is that two expressions cannot occupy the very same position. This restriction is even clearer under the bare phrase-structure system (see Chomsky 1994, 1995), where there is no distinction between positions and elements that undergo merger and merge is assumed to be a binary operation. Under these assumptions there is no way for the computational system to simultaneously merge two unconnected terms ␣ and ␤ to another term ␥ . Thus, the restriction on split antecedents turns out to have more to do with PRO (the source of the movement) than the antecedents themselves (the targets of movement).14 Before addressing how the properties in (24g) and (24h) can be accounted for, it should be noted that PRO differs from standard pronouns, as far as de se readings and bound readings with ‘only’ are concerned. As we saw earlier, the interpretation of a sentence such as (32a) below requires that the unfortunate be conscious of who he is and expect himself to get a medal; hence the addition in parentheses is infelicitous. The use of a pronoun instead of PRO in (32b) does not trigger such a restrictive interpretation. It is compatible with a de se reading, but also admits a non-de se interpretation. For example, it 14 At first sight, the existence of partial control (see section 2.5.2.1) and split-control (see Oded 2006 and Fujii 2006) constructions seems to be problematic for the MTC. However, as will be discussed in sections 5.6.1 and 5.6.2 below, based on Rodrigues (2007) and Fujii (2006), the source of the movement in these apparently problematic cases involves not unconnected expressions, but rather complex “conjunctive” subjects. 50 Basic properties of the movement theory of control admits a reading with the unfortunate expecting that some particular individual should get a medal given what he read about this person without having the knowledge that this individual is actually him, the unfortunate. In this scenario, (32b) may be felicitously followed by the addition in parentheses. Similarly, in (33) the pronoun supports the bound reading required by PRO, as well as a coreferential reading where the sentence may be falsified if anybody remembers that Churchill delivered the BST speech. (32) a. b. [[The unfortunate]1 expects [PRO1 to get a medal]] (#Although he doesn’t expect himself to get a medal) [[The unfortunate]1 expects [that he1 should get a medal]] (Although he doesn’t expect himself to get a medal) (33) a. [[Only Churchill]1 remembers [PRO1 giving the BST speech]] (‘Only Churchill is such that he remembers himself giving the BST speech’/ ∗ ‘Nobody else remembers that Churchill gave the BST speech’) b. [[Only Churchill] remembers [that he gave the BST speech]] (‘Only Churchill is such that he remembers himself giving the BST speech’/ ‘Nobody else remembers that Churchill gave the BST speech’) Having the contrasts in (32) and (33) in mind, let us consider how expressions that have been assigned multiple ␪-roles are to be interpreted. Take the control construction in (34) below, for instance. According to the MTC, it is associated with the (simplified) derivation in (35), where the moved DP in (35d) ends up being marked with two ␪-roles after moving to the thematic-subject position of the matrix clause. The natural interpretation for the thematic relations encoded in (35d) is expressed by the logical form given in (36). (34) John expected to kiss Mary (35) a. Applications of merge: [to kiss Mary] Merger of ‘John’ + assignment of “kisser” θ -role: [Johnkisser to kiss Mary] Applications of merge: [T expected [Johnkisser to kiss Mary]] Movement of ‘John’ + assignment of “expecter” θ-role: [John1 expecter+kisser T expected [t1 to kiss Mary]] b. c. d. (36) John (␭x [x expected x kiss Mary]) Following Reinhart (1983) and Salmon (1986), we can understand (36) as ascribing the property of expecting oneself to kiss Mary to John.15 Importantly, complex monadic predicates such as (36) have an inherently reflexive semantics 15 See the discussion in Grodzinsky and Reinhart (1993: 74), whence this locution is taken. 3.4 Controlled PROs as A-movement traces 51 (note the gloss above: expecting oneself to kiss Mary), thus being semantically very different from structures where two distinct expressions have a dependency relation. In effect, Reinhart and Salmon provide the semantic wherewithal for distinguishing the interpretation of multiple thematic positions within a chain (cf. [32a], [33a], and [34]) from multiple thematic positions in a dependency relation across chains (cf. [32b] and [33b]).16 In other words, the logical forms of (32a) and (33a) are as represented in (37) below and the logical forms of (32b) and (33b), as in (38). Intra-chain “binding” is restricted to de se and bound readings as it involves complex monadic predicates, as opposed to inter-chain binding.17 (37) a. b. [The unfortunate] (␭x [x expected x to win a medal]) [Only Churchill] ␭x([x remembers x giving the BST speech]) (38) a. b. [The unfortunate] (␭x [x expected that he should win a medal]) [Only Churchill] ␭x([x remembers that he gave the BST speech]) In sum, it appears that, by analyzing OC PRO as an A-trace, the MTC can also derive OC PRO’s central interpretive features, whereas this is something tricky to capture in any model that treats OC PRO as a pronoun of sorts, as we have seen in section 2.5. 16 The similarity in interpretation between reflexives and PRO invites us to consider analyses of reflexivization also in terms of A-movement. See Hornstein (2001, 2007), Boeckx, Hornstein, and Nunes (2007, 2008), and references therein for specific proposals and relevant discussion. 17 This is not the way that Reinhart (1983) interprets matters. She understands pronominal binding as another case of ␭-abstraction. Thus, (i) can also have the structure in (ii) in her approach, allowing the interpretation that the property of expecting oneself to kiss Mary is attributed to Alfred. (i) (ii) Alfred1 expected that he1 would kiss Mary Alfred (␭x [x expected that x would kiss Mary]) However, in contrast to (iii) below, (i) need not be interpreted in a de se manner. Moreover, even if we replace ‘Alfred’ in (i) with a quantified DP that binds the pronoun (so blocking the option of a coreferential reading of the coindexed pronoun), as in (iv), the de se reading is still not forced. This suggests that the binding of a pronoun does not result in a complex monadic predicate with an inherently reflexive semantics. To put this another way, there is an important semantic difference between a single expression “binding” two ␪-positions and two expressions each “binding” a ␪-position and themselves in a binding relation. Only the former yields a necessarily reflexive (i.e., de se) reading. For further discussion of this issue, see Hornstein and Pietroski (2009). (iii) (iv) Alfred expected PRO to kiss Mary [Every soldier]1 expected that he1 would kiss Mary 52 Basic properties of the movement theory of control 3.4.3 Phonetic properties and grammatical status Recall from section 2.5 that in non-movement analyses within minimalism, PRO is a primitive lexical formative. Thus, like all of its other properties, its lack of phonetic content is taken to be an irreducible (i.e., non-explainable) lexical property.18 It pays to reexamine what a strange kind of lexical item PRO is, under this view. It has virtually no properties of its own. It has no phonetic properties and its only semantic property is that of a pure variable, in effect, a placeholder for the interpretation of its antecedent. PRO so conceived has even less theoretical appeal than Agr heads, as the latter are (at least) often phonetically visible. Chomsky (1995) has argued that Agr essentially encodes a syntactic relation and, as such, it should be understood as a grammar-internal formative and not as a lexical head. Similar considerations apply to PRO. Treating it as a lexical formative in fact presents more questions than answers. In the case of its phonetic content, the stipulation that it must be phonetically empty brings with it the obvious question of why this should be so (see section 4.5.4 below for further discussion). In contrast to this lexical approach, its predecessors analyzed the lack of phonetic content of the “controllee” as following from properties of the computational system then assumed (see sections 2.3 and 2.4). In the standard theory, the “controllee” was phonetically null as it underwent a deletion transformation (the equivalent NP deletion rule). In turn, PRO in (early) GB was analyzed as a base-generated empty category, [NP Ø], resulting from the phrasestructure rule NP→ N not followed by lexical insertion for N. In addition, if PRO had phonetic content, it should be subject to the case filter and case assignment operated under government. Once PRO could not be governed (the PRO theorem), it then follows that it could not have phonetic content. All in all, the great virtue of the standard theory and GB accounts when compared to non-movement analyses within minimalism is that they attempted to provide a rationale within which PRO’s phonetic emptiness should follow from general considerations. 18 To reiterate: for non-movement accounts there is really no alternative within a minimalist setting given bare phrase structure except to treat PRO as a lexical primitive. This requirement, in turn, prevents such accounts from explaining why control clauses have the properties they have as they must all pack the specific requirements characteristic of control clauses into the lexical specifications of PRO. The reason that PRO needs a local antecedent is that it is lexically specified to require one. The reason that it is phonetically null is that it is lexically required to be so. In effect, given minimalist assumptions, a PRO-based approach to control can at best track/describe the properties of control constructions but it cannot possibly explain them. In this sense, contemporary PRO-based accounts are far less interesting than their GB predecessors. 3.4 Controlled PROs as A-movement traces 53 The MTC also subscribes to the view underlying the standard theory and GB that OC PRO is not a lexical formative, but a product of the grammar, i.e., a trace left by an A-movement operation. Interestingly, the MTC shares specific aspects with both the standard theory and GB. With (early) GB, it shares the view that OC PRO and NP-traces are indistinguishable at LF. Recall from section 2.4 that both OC PRO and NP-traces were taken to be categories of the form [NP Ø], with the only difference between them being the provenance of the index tying them to their antecedents (see Chomsky 1977: 82): for NP-traces, the index is part of the movement operation; for PRO, it arises from a construal rule, the rule of control. However, the conception of PRO and NP-traces as [NP Ø] categories is completely at odds with the bare phrase-structure system adopted in the minimalist program (see Chomsky 1994, 1995). A key feature of bare phrase structure is that it dispenses with the distinction between a lexical element and the position it occupies. Phrases are understood as projections of lexical items and are built through successive applications of merge. Consequently, there are no lexically unfilled positions. In fact, there is no structure other than the structures formed by successively merging (projections of) lexical items. Thus, in a minimalist setting, PRO and NP-traces cannot be associated with the structure [NP Ø], for it is impossible to generate a phrase without a lexical head, given bare phrase structure. Notice that simply taking PRO and NP-traces to be associated with structures such as [NP t] will not do either. One of the core architectural properties of the minimalist program is the inclusiveness condition (see Chomsky 1995), which enforces parsimony in the set of primitives proposed, by requiring that LF objects be built based only on the features of the lexical items that feed the derivation. The inclusiveness condition bans the creation of new objects in the course of syntactic computations, by only allowing restricted manipulation of the (features of the) lexical items that form syntactic structures. Under this view, traces cannot be theoretical primitives as they are not built from items of the lexicon, but are rather created (out of nothing) by the computation itself (i.e., movement operations). Based on conceptual reasons such as the ones discussed here and empirical reasons having to do with reconstruction effects, Chomsky (1993) incorporates the copy theory of movement into the minimalist program.19 According to the copy theory, movement amounts to copying lexical 19 For further conceptual and empirical arguments for assuming the copy theory of movement, see e.g., Chomsky (1995), Hornstein (1995, 2001), Nunes (1995, 2004, in press), Boˇskovi´c and Nunes (2007), the collection of papers in Corver and Nunes (2007), Kandybowicz (2009), and references therein. this minimalist implementation of the MTC not only accounts for the distribution. is analyzed along the lines of (40). in compliance with the inclusiveness condition.54 Basic properties of the movement theory of control items or their projections. In other words. Applications of merge: [T hoped [John to see Mary]] Copying and merger of ‘John’ + θ -assignment: [John1 [T hoped [John1 to see Mary]]] Deletion of the lower copy in the phonological component: [John1 [T hoped [John1 to kiss Mary]]] b. thereby simplifying the general apparatus of the model. which gets deleted in the phonological component. now generalized to . Applications of merge: [was [arrested John]] Copying and merger of ‘John’: [John1 [was [arrested John1 ]]] Deletion of the lower copy in the phonological component: [John1 [was [arrested John1 ]]] b. Thus. It is either a lexical item or a phrase built from lexical items.5 below for further discussion). c. In other words. c. as far as the syntactic computation goes. for instance. (39) John was arrested (40) a. where superscripted indices annotate copies. By taking OC PRO to be a copy resulting from movement to a thematic position. merging the copied material into the structure. interpretation. (41) John hoped to see Mary (42) a. whatever is independently responsible for deletion of traces (lower copies) should also account for PRO’s lack of phonetic content: it is a deleted copy (see section 4. nor a phrase headed by an entity that is not a lexical item. the deletion operation in (42c) can be viewed as a descendant of the equi-NP deletion rule of the standard theory. The control construction in (41) below. PRO is neither a phrase with no lexical head. if OC PRO is an NP-trace. but in fact paves the way to the elimination of PRO as an exotic lexical primitive. for instance. In a sense. That is. is to be derived as in (42). this minimalist implementation of the MTC complies with both the inclusiveness condition and bare phrase structure. and deleting lower copies in the phonological component. and lack of phonetic content of OC PRO. A sentence such as (39). it should also be a copy of the moved element. Copies replicate lexical items or syntactic objects built from lexical items and deletion of copies takes place in the phonological component. as discussed in section 3.’ for example. it differs from its predecessor in being able to correctly account for the meaning difference between (43a) and (43b) below.4. b. Recall from section 2.3 that.2. (43) a. In addition to the required abandonment of DS. two occurrences of the pronoun ‘he. (45) then surfaces as (43a) and is interpreted as a complex monadic predicate. “the syntactic objects formed by distinct applications of Select to LI [lexical item] must be distinguished. we have mentioned earlier that the MTC accords well with the general features of the minimalist program. (45) [Everyone1 T wants [everyone1 to win]] To sum up. When shipped to the phonological component. By contrast. [Everyone1 T wants [everyone1 to win]] [Everyone2 T wants [everyone1 to win]] We have two copies of ‘everyone’ in (44a). incorrectly predicting that the two sentences should have the same meaning.4 Controlled PROs as A-movement traces 55 constructions other than control. we have seen in this section that the implementation of the MTC in terms of the copy theory complies with the inclusiveness condition and 20 As Chomsky (1995: 227) points out. (44) a.20 Under this view. Everyone wants to win Everyone wants everyone to win By contrast.’ which is merged in the embedded clause and later copied to be merged in the matrix clause.” . In other words. (44b) surfaces as (43b) and each quantifier ranges over a different variable. the structures in (44) will then receive different treatment despite their superficial similarity. the equi-NP deletion rule should derive (43a) from the structure corresponding to (43b). b. by relying on phonetic identity. but not the second independent occurrence of ‘everyone’ in (44b).’ each of which is merged in a different clause. Importantly. the numeration underlying (44a) has only one instance of ‘everyone. the numeration underlying (44b) has two instances of ‘everyone. By contrast. may have entirely different properties at LF. but two distinct occurrences of ‘everyone’ in (44b). deletion will target the lower copy of ‘everyone’ in (44a).3. as shown in (45) below. That is. the sentences in (43) are respectively associated with the structures in (44). the deletion operation seen in (40c) and (42c) deals with copies and not simply with elements that happen to have the same phonological shape. explaining why the controlled element appears where it does. More importantly. interpretation. However. it also adopts the minimal-distance principle. but ended up eliminating PRO as a primitive of the grammar.56 Basic properties of the movement theory of control bare phrase structure. the phonological component may realize a lower copy instead of the head of the chain or even multiple copies. This substantial result will be strengthened further as we examine its empirical consequences in the next chapters. If this is correct. Like EST and GB. Third.4 below. it must account for the interpretation of the controlled element.5. as a residue of movement).5 below for more detailed discussion). This is in many 21 For instance. paying exclusive attention to OC (we return to NOC in Chapter 6 below).5 Conclusion As mentioned in section 2. namely move. the MTC insists that these are both formed via the same grammatical operation. it must determine the nature of the controlled element. it treats the controlled element in OC contexts as a non-lexical formative (more precisely. in contrast to (virtually all) earlier theories that strictly distinguished raising from control. we outlined the broad features of (our version of) the MTC. and lack of phonetic content of OC PRO. The version of the MTC explored here borrows liberally from the insights of its predecessors (see Chapter 2). 2004) has argued that in certain well-defined circumstances. the version of the MTC outlined here not only accounted for the distribution.2. Second. it must enumerate the kinds of control structures.21 3. specifying if and why obligatory control (OC) and non-obligatory control (NOC) are different. First.3 and 4. it must account for the distributional properties of control. Like EST.5. an adequate theory of control must meet (at least) the following four requirements. we should in principle expect to find OC constructions (under the relevant circumstances) with the “OC PRO” copy phonetically realized. And fourth. by sticking to these core minimalist precepts. Nunes (1999. This is made possible by the renunciation of one assumption that has been held constant since the earliest days of generative grammar: we give up the assumption that movement into ␪-positions is impossible. in fact deriving it in terms of relativized minimality (see section 5. specifying its place among the inventory of null expressions provided by universal grammar. We return to specific cases in sections 4. showing how PRO’s antecedent is determined and what kind of anaphoric relation obtains between PRO and its antecedent in OC and NOC structures. In this chapter. . The only difference between the A-movement found in control configurations and that found in more familiar cases of raising and passive is that the one that occurs in control moves the DP through multiple ␪-positions. it is the head of the (multiply ␪-marked) A-chain of which it is a link. PRO can occur as the subject of these kinds of clauses. The possible antecedents for OC PRO are then defined in terms of the positions to which the “controllee” can move. the MTC allows a considerable 22 For instance. its antecedent is simply the element that has moved from that position. 2004. In short. OC amounts to the formation of a multiply ␪-marked A-chain and is interpreted as standard A-chains are. OC PRO is a copy left by a movement operation.1 below for relevant discussion. OC is an instance of A-movement. which is later deleted in the phonological component. given that OC PRO is a residue of A-movement. Thus. The specific answers provided by (our version of) the MTC to the questions above are the following. under the copy theory of movement. as the controlled element is simply a residue of A-movement. First.5 Conclusion 57 ways the key innovative feature of MTC.5. It is well known that subjects can A-move from non-finite clauses (not only infinitives and gerunds. we expect OC to be subject to these locality restrictions as well. 2001. See section 4. Taking a page out of the GB playbook. sideward movement to thematic positions is argued to be involved in the derivation of parasitic gap and ATB constructions (see Nunes 1995. the controlled element in OC contexts should be interpreted essentially like a locally bound anaphor. it can only appear in positions from which A-movement is possible. we assume with Chomsky (1981) that A-traces and reflexives are both anaphors.5. the other species of control is non-obligatory control (NOC).2. As we will discuss in detail in Chapter 6 below. occurring when movement cannot take place. Second. In all other respects. So OC PRO does not surface with phonetic content for the same reasons that “traces”/lower copies do not. which amounts to the elsewhere case.22 and we will see in the next chapters that this innovation has some nice empirical consequences. but also ␾-defective finite clauses as discussed in section 2. Thus. the core case of control is obligatory control (OC).3. Indeed. We will see ample evidence for this in the next chapters. . given that A-movement is subject to strict locality conditions. and Hornstein and Nunes 2002). although it is not specific to it.4). Moreover. Finally. The controlled element is the residue of A-movement. Third. it is the same operation in both instances and the “traces”/copies left behind are expected to be indistinguishable from one another. Thus. Hornstein 2001. The most appealing feature of the MTC for us is the fact that the theory not only provides a principled account for PRO’s distribution. but we assure the reader that the following chapters will show that MTC is indeed even more successful than we have made it seem in this chapter.58 Basic properties of the movement theory of control simplification in the grammar by eliminating OC PRO as a theoretical primitive. see Chapter 6 below). This makes the MTC rather unique given that other current approaches to control treat these two facets of PRO as quite unrelated. This all seems too good to be true. but the account it provides for the syntactic distribution of OC PRO immediately extends to cover most (if not all) of OC PRO’s interpretive properties. as well as parts of the control module responsible for its interpretation (on NOC. . In particular. and being allowed in the subject position of finite clauses when the finite T is not an obligatory case assigner/checker (section 4.5.3 on backward control and section 4.2 Morphological invisibility It has long been observed that PRO differs from A’-traces in not blocking sandhi phenomena. We start by discussing the welcome results one obtains by assuming that OC PRO is a trace of A-movement.. regardless of the view on traces one takes.2).5. and cases where OC PROs are traces of sideward movement.5.4 on copy control).g. Finally.3). Boeckx (2000). we discuss empirical consequences of the MTC when the copy theory of movement is taken into consideration. as illustrated in (1) and (2). section 4. the most well-known example of such being wanna-contraction in English.1 (1) [Who1 do you want PRO to banish t1 from the room] → Who do you wanna banish from the room? 1 For relevant discussion. we will discuss cases where an OC PRO behaves like an overt element in being subject to morphological restrictions (section 4.1 Introduction In this chapter we explore some empirical consequences of (our version of) the MTC.4 Empirical advantages 4.. Jaeggli (1980). Postal and Pullum (1978).6 presents our conclusion that the data covered by the MTC discussed in this chapter prove fatal for any PRO-based account of OC. i.5. being transparent for interclausal agreement (section 4.e. and references therein. 4. cases where OC PROs are phonetically realized (see section 4.2). movement from one tree to another independent tree (see section 4. Next. see e.1 on adjunct control). Lightfoot (1976). 59 .4). We will see that OC PRO and standard A-traces pattern alike in being invisible to some morphological computations (section 4. 60 Empirical advantages (2) ∗ [Who1 do you want t1 to vanish from the room] → Who do you wanna vanish from the room? Interestingly, as has been known since Lightfoot (1976), A-traces also allow contraction: (3) a. b. c. [John1 is going t1 to kiss Mary] → John is gonna kiss Mary [John1 used t1 to kiss Mary] → John usta kiss Mary [John1 has t1 to kiss Mary] → John hasta kiss Mary The parallel behavior of PRO and A-traces in licensing contraction is exactly what is predicted by the MTC, regardless of what the correct analysis for this contrast between A’- and A-traces is. Take, for instance, the influential proposal that the contrast has to do with case (see e.g., Jaeggli 1980), that is, case-marked elements block contraction, but caseless elements do not. If OC PRO is an Atrace, as advocated by the MTC, it must sit in a caseless position and, therefore, should not prevent contraction under this view. Moreover, as observed by Boeckx (2000), if the case-based account of the contrast between (1)/(3) and (2) turns out to be correct, it poses serious questions for any approach to control that takes OC PRO to be case marked, be it in terms of null case (see e.g., Martin 2001) or in terms of regular case (see e.g., Landau 2004). The MTC also accounts for superficially similar constructions such as (4) (from Postal and Pullum 1978), where contraction cannot take place. (4) [I don’t want [[PRO to undress in public] to become standard practice]] → ∗ I don’t wanna undress in public to become standard practice Notice that the infinitival clause in (4) is in the subject position of the embedded clause. In other words, it is a subject island and should prevent movement from the position occupied by PRO. From the perspective of the MTC, that amounts to saying that the empty category occupying the subject position of the infinitival clause in (4) cannot be an OC PRO/A-trace (see Chapter 6 below). Given that only A-traces are invisible for purposes of contraction, the contrast between (1) and (4) now follows straightforwardly. 4.3 Interclausal agreement OC PRO and A-traces also pattern alike in triggering agreement in their local domain, matching the features of their antecedent.2 Take case concord, for 2 Although this is the general pattern, there are well-known cases (also in Latin) where an OC PRO seems to mismatch the features of its antecedent. We postpone the discussion of these potentially problematic cases to section 5.4.2 below, where we show that the problems are either apparent or are due to independent factors. 4.3 Interclausal agreement 61 instance. As discussed by Cecchetto and Oniga (2004), in languages with rich case morphology such as Latin, an adjectival predicate agrees with the subject in case (as well as in number and gender), as illustrated in (5a) below. Under the standard assumption that copular constructions involve raising from the embedded predicate, (5a) is to be represented as in (5b). (5) a. b. Latin (Cecchetto and Oniga 2004): Ego sum bonus I.NOM am good.NOM ‘I am good’ [TP egoi sum [SC ti bonus]] What is relevant for our discussion is that similar case agreement is also found in OC PRO clauses, as shown in (6a) with subject control and (6b) with object control. (6) a. b. Latin (Cecchetto and Oniga 2004): [Ego volo [PRO esse bonus]] I.NOM want to-be good.NOM ‘I want to be good’ [Ego iubeo te [PRO esse bonum]] I.NOM order you.ACC to-be good.ACC ‘I command you to be good’ The embedded adjectival predicate exhibits nominative case in (6a) and accusative case in (6b). This follows if PRO is not case marked within the embedded clause. If it were, we would have to make the awkward assumption that the infinitival T-head in Latin assigns either nominative or accusative depending on the kind of control involved (subject or object control). As pointed out by Cecchetto and Oniga, case matching of the type illustrated in (6) is thus problematic for any approach that takes PRO to be a bearer of structural case. By contrast, the agreement pattern exhibited in (6) is exactly what is expected under the MTC. If the OC PROs in (6) are actually A-traces, as respectively represented in (7) below, they should pattern like the A-trace of (5b). In other words, the embedded predicates in (7) must agree with the antecedent of the trace in the subject position of the infinitival clause. (7) a. b. [Egoi volo [ti esse bonus]] [Ego iubeo tei [ti esse bonum]] It should be clear that our point here is not to debate on how to technically capture the agreement relation between the embedded predicate and the antecedent of the trace in (5b) and (7). For concreteness, we may assume that the embedded predicate surfaces the way it does because it is in agreement relation with a 62 Empirical advantages link of the chain headed by ‘ego’ in (5b) and (7a) and ‘te’ in (7b) or that the trace in a local agreement relation with the embedded predicate is a copy of the moved element. What should be borne in mind is that, from the perspective of the MTC, whatever the technical analysis is that enforces case matching in (5b), it must also apply to (7). The requirement of interclausal agreement in raising and OC configurations also holds of ␾-features, as discussed by Rodrigues (2004, 2007) with respect to gender. Rodrigues examines the agreement pattern triggered by nouns such as the Romance counterpart of ‘victim,’ which is invariably [+feminine], regardless of whether it refers to males or females. In the raising constructions in (8), for instance, the adjectival predicates take the feminine form even in the context where a man has been hurt. (8) a. b. Italian (Rodrigues 2004): La vittima sembra essere ferita/∗ ferito The victim seems be injured.FEM/injured.MASC Brazilian Portuguese (Rodrigues 2004): A v´ıtima parece estar ferida/∗ ?ferido The victim seems be injured.FEM/injured.MASC ‘The victim seems to be injured’ The agreement seen in (8) is replicated in OC constructions such as (9), but not in non-OC constructions such as (10), again in a context where the victim is a male. (9) a. Italian (Rodrigues 2004): La vittima ha cercato di essere trasferita/??trasferito The victim had tried of be transferred.FEM/transferred.MASC alla stazione di polizia di College Park to-the station of police of College Park b. Brazilian Portuguese (Rodrigues 2004): A v´ıtima tentou ser transferida/??transferido para a The victim tried be transferred.FEM/transferred.MASC to the delegacia de pol´ıcia de College Park station of police of College Park ‘The victim tried to be transferred to the police station at College Park’ (10) a. Italian (Rodrigues 2004): La vittima ha detto che essere ∗ portata/portato alla The victim has said that be brought.FEM/brought.MASC to-the stazione di polizia non era una buona idea station of police not was a good idea ‘The victim said that being brought to the police station was not a good idea’ 4.4 Finite control b. 63 Brazilian Portuguese (Rodrigues 2004): A v´ıtima disse que ser ??transferida/transferido para The victim said that be transferred.FEM/transferred.MASC to outra cidade n˜ao e´ uma boa id´eia other city not is a good idea ‘The victim said that being transferred to another city is not a good idea’ As Rodrigues notes, the agreement contrast between OC and NOC in (9) and (10) requires non-trivial provisos under a PRO-based analysis of control. From a purely formal point of view, PRO should have a uniform behavior regarding agreement. In other words, in both types of constructions PRO should either be the element triggering agreement with the embedded predicate (independently from its antecedent) or be transparent to interclausal agreement. Thus, the fact that the null subject of the infinitival clause of the sentences of (9) and (10) does not have a uniform behavior suggests that we are not dealing with the same type of empty category in these constructions. Rodrigues further points out that the MTC, on the other hand, provides a straightforward account of the contrast between (9) and (10). In (9) we have OC (more specifically, subject control). Thus, the null subject inside the infinitival is an A-trace and, as such, it should pattern with the A-trace in the embedded subject position of the raising constructions in (8). Hence, we have a transparent domain for interclausal agreement in (9) and the agreement morphology on the embedded predicate must match the gender feature of the antecedent of the embedded subject. Again, this should be so independently of the specific analysis one assumes for interclausal agreement in standard raising constructions. The sentences in (10), on the other hand, cannot be analyzed as involving A-traces in the subject of the infinitival clause, as the infinitival is a subject island. Once (10) cannot be analyzed in terms of A-traces, interclausal agreement is blocked and the embedded predicate takes an (arguably default) masculine form. 4.4 Finite control As we saw in detail in section 2.5.2, Landau’s (2004) Agree-based approach to control faces several problems. It resorts to various stipulations regarding the features employed to track the distribution of PRO and the system of composite agreement relations proposed to account for the interpretation of PRO, in addition to not being independently motivated, leads to overgeneration. In this section, we will pay closer attention to a salient undergeneration problem of Landau’s system, which we believe provides decisive evidence in favor of the MTC, namely, the existence of control into indicative clauses. 64 Empirical advantages Recall that the typology of control structures predicted by Landau’s system given in (11) below (from Landau 2004: 840; cf. [39] in Chapter 2) explicitly blocks OC into indicative clauses. In his own words (pp. 849–850), “the only generalization in this domain that appears to be universal is the incompatibility of indicative clauses with OC.” (11) Obligatory control Hebrew 3rd-person subjunctive No control EC-infinitive Balkan Csubjunctive PC-infinitive I0 [−T, −Agr] [−T, +Agr] [+T, +Agr] [+T, −Agr] C0 [−T] [−T] Balkan indicative F-subjunctive [+T, +Agr] [+T, +Agr] [+T, (+Agr)] [+T, +Agr] [+T, +Agr] Ø However, we have seen that null subjects in finite indicative clauses in Brazilian Portuguese display all the diagnostics for OC (see section 2.5.2.2). In a sentence such as (12) below, for instance, the embedded subject can only be interpreted as controlled by the most local c-commanding DP, namely, s´o o irm˜ao do Jo˜ao ‘only Jo˜ao’s brother.’ Furthermore, the null subject has only a bound interpretation, only licenses sloppy readings under ellipsis, and must receive a de se reading in the appropriate contexts.3 (12) Brazilian Portuguese: [[O Pedro]i disse [que [s´o o irm˜ao d[o Jo˜aok ]]m estava achando The Pedro said that only the brother of-the Jo˜ao was thinking [que Øm/∗ i/∗ k/∗ w deveria ganhar uma medalha]]] that should receive a medal ‘Pedro said that [only Jo˜ao’s brother]m was thinking that hem should get a medal’ Once finite control into indicative clauses is empirically attested in Brazilian Portuguese, one has to determine which special property allows it and why it is considerably rare from a crosslinguistic point of view. As discussed in section 2.5.2.4, Ferreira (2000, 2004, 2009) has proposed that indicative Ts in Brazilian Portuguese are ambiguous in that they may be associated with a complete or an incomplete ␾-set. Nunes (2007, 2008a) has reinterpreted this ambiguity in terms of how the person and number features of T are combined in the course of the computation. More specifically, Nunes proposes that finite Ts in Brazilian Portuguese may enter the numeration specified for number and person or for number only. When T is only specified for number, wellformedness conditions in the morphological component trigger the addition 3 See Ferreira (2000, 2004, 2009) and Rodrigues (2002, 2004) for a discussion of these properties of null subjects in Brazilian Portuguese in the context of the MTC. 4 Finite control 65 of the feature person in accordance to the redundancy rule sketched in (13) below. Verbal-agreement paradigm in (colloquial) Brazilian Portuguese cantar ‘to sing’: indicative present eu ‘I’ canto P:1.4. N:default canta N:PL P:default. add [P:default]. the other two inflections involve a default value (third) for the person feature. [57] in Chapter 2). [56] in Chapter 2) is such that the only form that distinctively encodes person and number is the syncretic inflection for first-person singular. Crucially. (15) cantar ‘to sing’: indicative present Valuation of T in the syntactic component Surface form of the verb P:1. as shown in (16) (cf. N:default (= 3SG) vocˆes ‘you (PL)’ eles ‘they (MASC)’ elas ‘they (FEM)’ cantam P:default. N:PL cantam (16) cantar ‘to sing’: indicative present Valuation of T in the syntactic component Addition of [person] in the morphological component Surface form of the verb N:SG P:1. N:SG canto N:default P:default. the paradigm of verbal-agreement morphology in (colloquial) Brazilian Portuguese given in (14) (cf. if N is valued as SG. N:default canta P:default. (13) (14) When T is only specified for number (N): (i) Add [P:1]. N:PL cantam . N:SG vocˆe ‘you (SG)’ ele ‘he’ ela ‘she’ a gente ‘we’ canta P:default. the three different verbal inflections available in (14) can be obtained in two ways: (i) T is specified for both person and number throughout the derivation. N:SG canto P:default. N:PL (= 3PL) Thus. (ii) otherwise. as in (15). or (ii) T is only specified for number and the feature person is associated with T in the morphological component in accordance with (13). a new car]] [vP Jo˜ao[case:u] said [CP that [TP t T[N:default]/EPP [vP t buy. . The next finite T to enter the derivation comes from the numeration with a complete ␾-set (person and number). etc. It may then raise to the matrix [Spec.’ valuing its case feature and having its own features valued. Brazilian Portuguese indicatives. . [58] in Chapter 2). Hebrew 3rd-person subjunctives. yielding (18c). proceeds along the lines of (18) (with English words for convenience). + ␾ ] Balkan untensed subjunctives. the derivation of a sentence such as (17). c. as shown in (19) (cf. − ␾ ] untensed uninflected infinitives. It then agrees with ‘Jo˜ao. as illustrated in (18e). N:default]/EPP [vP t said [CP that . First. as shown in (18d).” as Landau calls it. not only does it allow us to incorporate Brazilian Portuguese into the picture. (17) Brazilian Portuguese: O Jo˜ao disse que comprou um carro novo The Jo˜ao said that bought a car new ‘Jo˜ao said that he bought a new car’ (18) a. e. [T+ . Whether or not a given clausal structure allows for OC may be determined solely based on the tense and ␾-feature properties of T. . . etc. where ‘+’ stands for fully specified and ‘−’ for deficient or null.’ as shown in (18b). but it also makes it possible to considerably simplify the “calculus of control. − ␾ ] tensed uninflected infinitives. b. ␾+ ] English indicatives. [17]). ]]]] [TP Jo˜ao[case:NOM] T[P:default. ]]] The past indicative T in (18a) comes from the numeration with just a number feature. etc. (19) Obligatory control [T − . where it receives an additional ␪-role. [TP T[N:u]/EPP [vP Jo˜ao[case:u] buy. [T + . vP].a new car]]]] [TP T[P:u. Balkan tensed subjunctives.a new car]] [TP Jo˜ao[case:u] T[N:default]/EPP [vP t buy. Given that only a ␾-complete T is able to check/value the case feature on a DP (Chomsky 2000. a default person feature is added to T and the embedded verb surfaces with the “third-person singular” form ‘comprou’ (cf. No control [T − . for instance. 2001). which gets valued (as default) after agreeing with ‘Jo˜ao. etc. In the morphological component.66 Empirical advantages Under this view. N:u]/EPP [vP Jo˜ao[case:u] said [CP that [TP t T[N:default]/EPP . d. ‘Jo˜ao’ remains active after agreeing with the ␾-incomplete T in (18b). . The advantages of this analysis of OC into indicative clauses in Brazilian Portuguese in terms of ␾-incompleteness are twofold. eci yitna’heg yafe] Gil promised that will-behave. the embedded and the matrix subjects are coreferential. the realization of the embedded subjects in (21b) and (21c).4. Vocˆes/eles falaram que (vocˆes/eles) comeram o bolo You. +Agr] – I[+T. Hebrew subjunctive Ts may enter the derivation fully specified for person and number or specified for just number and receive a default (third) person in the morphological component.PL/they ate. the simplification of (11) as in (19) also eliminates the conceptually suspicious ambiguity of the composite C[+T. a natural analysis of Hebrew constructions such as (20) below. [21a]). Vocˆe/ele/a gente falou que (vocˆe/ele/a gente) comeu o bolo you.SG/he/we ate.3SG the cake ‘You(SG)/he/we said that you(SG)/he/we ate the cake’ c. which is used to describe both OC in Hebrew subjunctives and no control in Balkan F(ree)-subjunctives.PL/they spoke. as argued by Landau. for instance.4.1SG the cake ‘I said that I ate the cake’ b.5. like Brazilian Portuguese indicative Ts. By contrast. is truly optional. a good number of speakers prefer an overt pronoun when the embedded subject triggers first-person agreement (cf.3SG. In other words. 2000) shows that the percentage of null subjects with third person is significantly much higher than with first person in both spoken and written corpora of Brazilian Portuguese. For speakers who do not 4 Duarte (1995. For all speakers of BP. the parallel between Hebrew subjunctives and Brazilian Portuguese indicatives is even clearer if we consider the idiolectal variation in Brazilian Portuguese illustrated in (21): (21) a. which trigger default (third-person) agreement. is to take the T-head of their subjunctive clauses to be also ambiguous with respect to how person and number features are associated. Brazilian Portuguese (Nunes 2008a): Eu falei que %(eu) comi o bolo I spoke.3PL that you.3SG that you.3PL the cake ‘You(PL)/they said that you(PL)/they ate the cake’ In (21a–c). (20) Hebrew (Landau 2004): Gili hivtiax [ˇse.1SG that I ate.4 Nunes (2008a) attributes this idiolectal variation to the presence or absence of the specification of the redundancy rule in (13i) across speakers’ grammars.2. Given that OC in Hebrew subjunctives is restricted to third person. .M well ‘Gil promised to behave’ In fact. +Agr] in Landau’s system.4 Finite control 67 As mentioned in section 2.SG/he/we spoke. Although patterns with opposite values for T and ␾ are not garden-variety species across languages. Boeckx (2003.g. . Martins (2001). ␾− ]). Uchibori (2000) for Japanese and Uriagereka (2006) for Romance. finite control into indicatives is like what we find in Hebrew subjunctives: it is only possible with subjects that trigger (default) third-person agreement. is not absolute. where the embedded subject can leave the subjunctive clause and undergo A-movement to the matrix-subject position.. 2008) for relevant discussion.5 and “porous” subjunctives such as (23) in Greek. Witness.3PL ‘The children do not seem to work’ Bearing this in mind. they do exist. A-movement from 5 See e. as seen in (19). However.68 Empirical advantages have (13i) in their grammars.3PL the game was really a surprise ‘Their winning the game was a real surprise’ (23) Greek (Alexiadou and Anagnostopoulou 1998): Ta pedhia dhen fenonte na doulevoun The children not seem. 6 See e. which in the case of A-chains is defined in terms of [T+ .. Alternatively. see e.3PL SUBJ work. Since the incorporation of the case theory into GB. Roussou (2001).g. The second advantage of the analysis of OC into indicatives in Brazilian Portuguese in terms of ␾-incompleteness is that it makes it possible to understand why this subtype of OC is rare from a crosslinguistic point of view. it has been standardly assumed that there is a strong correlation between finiteness and the presence of a full ␾-set.g. As is well known.. as happens in finite control in Brazilian Portuguese (cf. On other cases of raising out of subjunctive clauses. (22) Brazilian Portuguese: Eles ganharem o jogo foi realmente uma surpresa They win. ␾+ ] (see Richards 2001 and Rizzi 2006 for related ideas). Raposo (1987. let us now consider why finite control provides convincing evidence in favor of the MTC. in the case of indicative OC. 1989). inflected infinitivals such as (22) below in Portuguese. Terzi (1997).INF. and Pires (2006) for relevant discussion. Alexiadou and Anagnostopoulou (1998). the correlation. [18]). albeit strong.6 Thus. we have the mismatching specification [T+ . where the subject is licensed with nominative case within the infinitival. Varlokosta (1993). 2008) argues that the reason movement can take place out of defective finite domains is that a chain can be extended up to its point of maximal checking. The unmarked situation is for finite Ts to be ␾-complete ([T+ . the fact that OC into indicative clauses is rather uncommon is related to the marked character of mismatches between T and ␾ with respect to full specification. ␾+ ]) and for non-finite Ts to be ␾-incomplete ([T− . A-movement in (23) is consistent with last resort if the subject has not been case-licensed in the embedded clause. ␾− ]. for instance. and Boeckx (2003. (25) A-movement: √ A-movement: ∗ [T− .T. Balkan tensed subjunctives. Boeckx (2003. etc. ␾− ] [T− . b. Raising constructions such as (ii). Kinande control structures behave as usual. ␾+ ] English indicatives.EXT ‘We seem to have left’ Ebitsungu bi-li-nga mo-by-a-huk-ir-w-e potatoes. [23]). it should allow both control and raising constructions. have matching person/number/case agreement on both the matrix and the embedded verbs. Brazilian Portuguese indicatives. Kinande (Mark Baker.PASS.be-if AFF. As Mark Baker pointed out to us (personal communication). 2008) argues that the relevant property that renders a given domain porous to Amovement is its deficiency with respect to ␾-features (cf. etc.T. as we have just seen in (23). on the other hand.EXT INF. [24a]) or with respect to tense (cf. personal communication): Tu-li-nga mo-tw-a-na-gend-ire 1PS. etc. with person/number/class agreement on the matrix verb.4.cook. ∗ John is likely [t to be home] John is likely [t is home] However.FV wood ‘We refused to chop the wood’ (ii) a. tensed uninflected infinitives.8. uninflected clauses. That being so. as illustrated by the contrast in (24).8 8. (i) Kinande (Mark Baker. ␾+ ] untensed uninflected infinitives.1PS. . [59] in Chapter 2). b.refuse.be-if AFF. personal communication): Mo-tw-a-gan-ire eri-seny-a olukwi AFF. (24) a.chop.1PS. the picture in (19) can in fact be subsumed under the more general table in (25) (cf.go. In other words.7 In 7 It is interesting to note in this context that control is sometimes much more well behaved than raising.4 Finite control 69 subject positions typically takes place from non-finite. there are indeed cases where a finite (inflected) subjunctive clause does not block A-movement.INDEED. the MTC predicts that modulo idiosyncrasies of selection by a matrix predicate. as illustrated in (i) below. and infinitival morphology on the embedded verb. if the relevant Infl head of a given domain is negatively specified for T or ␾. etc. [T+ .EXT ‘The potatoes seem to have been cooked’ So Kinande raising seems to flout the generalization that ␾-complete agreement freezes the relevant DP.T. Hebrew 3rd-person subjunctives. ␾− ] [T+ . Balkan untensed subjunctives. the paradigm in (26) and (27) is exactly what the MTC leads us to expect. tense marking cannot go on ‘seem’ in (ii). it is fair to concede that one could reasonably argue that the paradigm in (26)–(27) is only consistent with the MTC and not a knockout argument in its favor. as argued by Landau (2004). as in (26).3PL ‘The children do not seem to work’ Romanian (Dobrovie-Sorin 1994): Copiii t˘ai par s˘a fie foarte obosit¸i Children your seem. This suggests that raising in (ii) may actually involve an inflected modal adjunct.1 Finite control and hyper-raising It is a well-known fact that languages such as Greek and Romanian resort to subjunctive clauses for both control and raising constructions. TP].g. and Alboiu (2007) on Romanian.4. 4. for the pattern in (26)–(27) is associated with a morphological gap in these languages.70 Empirical advantages the sections that follow we show that this prediction is indeed borne out and explore some of its consequences. b.3PL SUBJ work. Dobrovie-Sorin (1994). Varlokosta (1993). b. Greek (Terzi 1997): I Maria prospathise na divasi The Maria tried. 2008) that Tdeficient or ␾-deficient domains are transparent for purposes of A-movement. If these subjunctives are specified as T− . as in (27). . Terzi (1997). or directly to the matrix [Spec. but a full range of tense morphemes is possible on the embedded verb.3SG SUBJ leave. see Henderson (2006) and Boeckx (2008). namely.3SG ‘Maria tried to read’ Romanian (Dobrovie-Sorin 1994): Ion vrea s˘a plece devreme mˆıine Ion want.8 Assuming with Boeckx (2003. For alternatives to reconciling raising in Bantu and the strong-agreement freezing effect. (27) a. (26) a. the lack of true infinitival However.3SG SUBJ read. and Roussou (2001) on Greek and Grosu and Horvath (1984).. as Baker notes. they should not block A-movement. 8 See e. Alexiadou and Anagnostopoulou (1998). as respectively illustrated in (26) and (27) below.3SG early tomorrow ‘Ion wants to leave early tomorrow’ Greek (Alexiadou and Anagnostopoulou 1998): Ta pedhia dhen fenonte na doulevoun The children not seem.3PL very tired ‘Your children seem to be very tired’ However.3PL SUBJ be. Whether we obtain a control or a hyper-raising construction (in the sense of Ura 1994) then depends on whether the embedded subject moves first to a ␪-position. they differ with 9 See Ferreira (2000. [28b]). So one could say that at an abstract level these are infinitival constructions in the syntactic component and we are back to square one as to what the best theory to account for infinitival OC is. as seen in (28a).9 (28) a. allowing all unvalued features to be valued. 10 It is worth pointing out that. Duarte (2004). nothing need be stipulated to ensure this result. T is associated only with number and the subject does not have its case valued after agreeing with T. ]]] If the embedded T in (29a) were fully specified with respect to ␾-features. the subject would have been case licensed in the embedded clause.3PL/finished. Nunes 2008a): Parece/acabou que os estudantes viajaram mais cedo Seem. as illustrated in (28b). yielding an impersonal construction. as represented in (29d).3SG that the students traveled. as discussed earlier. After further computations. although both the matrix and the embedded verb surface in the third-person plural form (cf.3SG/finished. it has both the uninflected and the inflected varieties). Martins and Nunes (2005.10 Notice that. [TP T[N:u]/EPP [vP [the students][case:u] traveled earlier]] [TP [the students][case:u] T[N:PL]/EPP [vP t traveled earlier]] TP T[P:u. and enters into an agreement relation with the embedded subject.3PL that traveled. as it does not lack infinitives (in fact.3PL more early ‘The students seem to have traveled earlier’/‘The students ended up traveling earlier’ The hyper-raising constructions in (28) are derived along the lines of (29) (again with English words. N:u]/EPP [vP seem/turned out [CP that [TP [the students][case:u] T[N:PL]/EPP . Although both ␾-complete and ␾-incomplete finite Ts are legitimate options for any . b. as shown in (29c). In this regard. Brazilian Portuguese provides us with the relevant test case. N:PL]/EPP [vP t seem/turned out [CP that . . and Nunes (2007. (29) a. b. 2009 for discussion). 2004. as shown in (29b). a fully inflected T is selected. ]]] [TP [the students][case:NOM] T[P:default. although finite control and hyper-raising constructions necessarily involve a ␾-incomplete finite T in the embedded clause and ␾-complete finite T in the matrix clause.4 Finite control 71 morphology (in these contexts). d. c. . this is not what happens in (29a). Brazilian Portuguese (Ferreira 2000. However. but allows OC into indicatives. in press). for convenience).4. The asymmetry between matrix and embedded clauses is trivially derived from UG principles (see Ferreira 2000. . . it also allows hyper-raising out of indicative clauses. 2009). 2008a) for relevant discussion. 2004.3PL more early ‘It seems/turned out that the students traveled earlier’ Os estudantes parecem/acabaram que viajaram mais cedo The students seem. As the MTC predicts. 2 below. being no different from other types of ␾-incomplete Ts. The matrix vP in (29) does not count as a (“strong”) phase as its head is not a “transitive” light verb (see Chomsky 2000. only has a number feature as it enters in the derivation. [18]) and the derivation of hyper-raising constructions sketched in (29) is that. In other words. For purposes of the current discussion. there is no source of case assignment for the matrix subject and the derivation simply crashes. discussed in Nunes (2008a). where we discuss data bearing on Visser’s generalization which add an interesting twist to this discussion. which show that idiom chunks. Rodrigues 2004. the embedded CP in (28b)/(29) does not count as a phase. can also appear in hyper-raising constructions. The matrix T verb enters the numeration with person and number features. If the matrix clause is associated with a ␾-incomplete finite T. Brazilian Portuguese (Martins and Nunes 2005. consider the Brazilian Portuguese data in (31)–(32). . The embedded T. Martins and Nunes (2005. on the other hand.11 (30) a. given numeration in Brazilian Portuguese. which are generally resistant to A’-movement. and Martins and Nunes in press). a ␾-incomplete finite T will only yield a convergent derivation if it sits within an embedded clause. in press). the MTC predicts that the two types of constructions should go hand in hand. After being valued in the syntactic component. In this regard. We return to this issue in section 5. 2001).72 Empirical advantages respect to how this specification is carried out. as the head of its complement bears only a number feature (cf. 11 See Ferreira (2000). 12 Technical questions arise regarding phase-based computations. hence. the matrix light verb does not have another ␪-role to assign and the embedded subject moves directly to the matrix [Spec. TP]. 2008a) for additional independent evidence. it suffices to assume with Ferreira (2000) that a C-head that selects for a ␾-incomplete TP does not count as a strong phase head. the number feature is then combined with a default person specification in the morphological component in accordance with (13ii). in press): [a vaca]i acabou que ti foi pro brejo The cow finished that went to-the swamp Idiomatic reading: ‘It turned out that things went bad’ [o pau]i parece que ti comeu feio The stick seems that ate ugly ‘It seems that there was a big discussion/fight’ Notice that the only relevant difference between the derivation involving OC into finite clauses discussed earlier (cf. [29a]). UG principles determine whether or not the choice and the structural locus of a ␾-incomplete finite T give rise to a convergent derivation. But what about the embedded CP? Should it not count as a phase and block movement of the embedded subject? Several different answers have been proposed to address this potential problem (see Ferreira 2000. b. and Nunes (2007. in the latter. Independent evidence for the derivation sketched in (29) is provided by the sentences in (30). which then get trivially valued in the syntactic component through Agree.12 But putting aside this independent difference. 2008a. such as the infinitival T-head of standard raising and OC constructions. Nunes 2007. [32b] violates the ␪-criterion. (33) a.4 Finite control (31) Brazilian Portuguese (Nunes 2008a): [Ningu´em mexeu um dedo para me ajudar] Nobody moved a finger to me help ‘Nobody lifted a finger to help me’ a. b. in Brazilian Portuguese the embedded null subject is a trace of the matrix subject. b. we see that once the embedded clause is porous due to the availability of an indicative T-head specified as ␾− . as argued above. The embedded null subject in sentences such as (32) in European Portuguese. Thus. (32) a. [32a]) or a hyper-raising construction (cf. [TP nobodyi [vP ti said [CP that [TP ti would [vP ti lift a finger to help me]]]]] [TP nobodyi [vP seemed [CP that [TP ti would [vP ti lift a finger to help me]]]]] Again. as illustrated in (33) below (with English words). Hence. if we have a null rather than an overt embedded subject in sentences analogous to (31b). the derivation will yield a control (cf.4. Given that Brazilian Portuguese allows finite control into indicative clauses. [32b]) depending on whether or not . the minimizer can be licensed by the clause-mate trace of the negative quantifier (or it can be licensed before the quantifier leaves the embedded clause). Even more interesting is the fact that it is not the case that any type of null subject will do. 73 ∗ [Ningu´em disse [que a Maria mexeu um dedo para me ajudar]] Nobody said that the Maria moved a finger to me help ‘Nobody said that Maria didn’t lift a finger to help me’ Brazilian Portuguese (Nunes 2008a): [Ningu´em disse [que ia mexer um dedo para me ajudar]] Nobody said that went move a finger to me help ‘Nobody said that he wasn’t going to lift a finger to help me’ [Ningu´em parecia [que ia mexer um dedo para me ajudar]] Nobody seemed that went move a finger to me help ‘It seemed that nobody was going to lift a finger to help me’ The contrast in (31) illustrates the well-known fact that a negative polarity item such as the minimizer um dedo ‘a finger’ and its licenser (in this case. sentences analogous to (32) are unacceptable in this dialect. the contrast between the two dialects with respect to (32) receives a straightforward account from the perspective of the MTC. Curiously. ningu´em ‘nobody’) must be in the same clause. as shown in (32a). Although contrasts such as (31) also hold in European Portuguese. which is a prototypical pro-drop language. these sentences are ruled out in European Portuguese because the minimizer and its licenser are not in the same clause (in addition. b. the minimizer can now be licensed by the matrix subject. as there is no ␪-role available for the matrix subject). By contrast. is pro. Take the verb cantar ‘to sing. Although the paradigm is considerably meager in (colloquial) Brazilian Portuguese. and compare its uninflected form (‘cantar’) with the paradigm of inflected forms in (i) below. Thus.3PL someone ‘These teachers rarely praise someone’ .’ for example. the existence of inflected infinitives.INF. Nunes (2008a) suggests that the relevant trigger for this reanalysis in Brazilian Portuguese was the existence of inflected infinitives in the language. namely. For relevant discussion and alternative approaches. but hyper-raising is not allowed. Nunes (2008a) suggests that the weakening of finite verbal morphology in Brazilian Portuguese led learners to generalize this pattern and uniformize the whole paradigm. for instance. as illustrated in (ii) (see section 5.2. (i) Inflected infinitives in European Portuguese: cantar ‘to sing’ (ii) Inflected infinitives in (colloquial) Brazilian Portuguese: cantar ‘to sing’ 1SG (eu) cantar 1SG (eu) cantar 2SG (tu) cantares 2SG (vocˆe) cantar 2SG (vocˆe) cantar 3SG (ele) cantar 3SG (ele) cantar 1PL (a gente) cantar 1PL (n´os) cantarmos 2PL (vocˆes) cantarem 1SG (a gente) cantar 3PL (eles) cantarem 2PL (v´os) cantardes 2PL (vocˆes) cantarem 3PL (eles) cantarem Brazilian Portuguese (Nunes 2008a): a. both dialects have a considerable number of forms that are ambiguous between being inflected or uninflected. successful acquisition of infinitives in both dialects requires that learners must postulate that (certain) infinitival forms are ambiguous between being ␾-complete (the inflected ones) and ␾-incomplete (the uninflected ones). see also Ferreira (2000). hyper-raising should be a very common phenomenon. the ambiguous morphological paradigm cannot be the whole story. More specifically. it should arguably be a marked property. the inflected realization of some forms is the same as the uninflected form. Thus. for all Portuguese verbs. verbal morphology is considerably weak.3PL someone b. TP] (cf. for in English. Interestingly. [33]). That being the case. Nunes proposes that. hyperraising also became possible with inflected infinitives in Brazilian Portuguese. otherwise. Furthermore. E´ dif´ıcil desses professores elogiarem algu´em Is difficult of-these teachers praise. while finite verbal morphology started becoming weakened. Brazilian Portuguese learners still had to acquire a marked property of Portuguese.74 Empirical advantages the moving subject is assigned an additional ␪-role on its way to the matrix [Spec.3 below for further discussion). and Martins and Nunes (2009). whatever the relevant property turns out to be. taking both infinitival and indicative Ts to be systematically ambiguous. After all.INF.13 13 A learnability question that arises is what exactly led indicative Ts to be analyzed by children as ambiguous between ␾-complete and ␾-incomplete in Brazilian Portuguese. Rodrigues (2004). Esses professores s˜ao dif´ıceis de elogiarem algu´em These teachers are difficult of praise. islands. e. [ie]). [id]). c. Rodrigues (2002. 15 That constructions such as (36a) are indeed cases of OC is shown by the fact that the embedded subject: (a) cannot have an arbitrary interpretation (cf. we should expect it to exhibit island and intervention effects. 2004). (c) can only have a bound interpretation when controlled by only-DPs (cf. [ib]). whose simplified structures under the MTC are given in (35b) and (36b) respectively. d. and Nunes (2009a) for further discussion.2 Finite control. See Landau (2003) and Barrie (2007) for relevant discussion. (d) only licenses sloppy reading under ellipsis (cf. let us examine the sentences in (34). finite control and hyper-raising should also pattern alike in this regard. ∗ Brazilian Portuguese (Nunes 2008a): [O Jo˜ao disse [que [o bolo [que comeu]] n˜ao estava bom]] The Jo˜ao said that the cake that ate not was good ‘Jo˜ao said that the cake that he ate was not good’ [O Jo˜ao parece [que [o bolo [que comeu]] n˜ao estava bom]] The Jo˜ao seems that the cake that ate not was good ‘It seems that the cake that Jo˜ao ate was not good’ Under the proposal that both finite control and hyper-raising involve Amovement. [ia]). this movement crosses two islands (the relative clause itself and the embedded subject containing it).15 14 See Ferreira (2000.4.4. given that hyper-raising constructions are also derived by A-movement. Moreover. In other words. (i) a. island effects such as the ones documented in (34) provide additional evidence for the MTC. explaining why the resulting sentences in (34) are unacceptable. and (e) requires de se readings in the appropriate contexts (cf. (34) a. and Mary doesn’t either (‘and Mary also doesn’t know what she/∗ he should eat’) The unfortunate wondered how to get along with people after the war (‘[The unfortunate]i wondered how [he himself]i was going to get along with people after the war’) . However. (b) must be interpreted as the most local c-commanding DP (cf. ∗ b. b.4 Finite control 75 4. the sentences in (34) should be derived by moving ‘o Jo˜ao’ from the subject position of the relative clause to the matrix subject position. 2009). 2004. and intervention effects If finite control is also derived via A-movement. ∗ John doesn’t know what PROarb to eat Peteri said that [Johnk ’s mother]w doesn’t know what PROw/∗ i/∗ k to read A: Only John wondered what to do B: No! I also wondered what I/#he should do John doesn’t know what to eat. With this in mind.14 Let us now consider the OC constructions in (35a) and (36a) below. as advocated by the MTC. [ic]). b. as all of these different constructions involve the same derivational device: A-movement. CP] is a licit operation in the case of standard non-finite control.76 Empirical advantages What is relevant for our discussion is that the derivations of these sentences involve a step where the embedded subject moves to the matrix [Spec. crossing a filled [Spec. Brazilian Portuguese (Nunes 2009a): O que o Jo˜ao disse que comeu? What the Jo˜ao said that ate ‘What did Jo˜ao say that he ate?’ [CP whati [TP Jo˜aok [vP tk said [CP ti that [TP tk ate ti ]]]]] ↑ Brazilian Portuguese (Nunes 2009a): O que o Jo˜ao parece que comeu? What the Jo˜ao seems that ate ‘What does Jo˜ao seem to have eaten?’ [CP whati [TP Jo˜aok [vP seems [CP ti that [TP tk ate ti ]]]]] ↑ . b. First. CP] does not count as a proper intervener for A-movement of the embedded subject. vP]. The Brazilian Portuguese data in (38a) and (39a). CP]. this observation makes two predictions with respect to finite control and hyperraising constructions. [vP Johnk try [CP whati C [TP tk to do ti ]]] ↑ b. What did John try to do? [CP whati did [TP Johnk [vP tk try [CP ti C [TP tk to do ti ]]]]] (36) a. as illustrated in (37). (35) a. b. as seen in (35) and (36). b. show that this prediction is indeed fulfilled. it should also be legitimate in cases of finite control into indicatives and hyper-raising. whose simplified derivations are provided in (38b) and (39b) (with English words). (39) a. John wondered what to do [TP Johnk [vP tk wondered [CP whati [TP tk to do ti ]]]] (37) a. Leaving aside matters of technical implementation. if A-movement of an embedded subject across a filled [Spec. [vP Johnk wondered [CP whati C [TP tk to do ti ]]] ↑ The acceptability of the sentences in (35a) and (36a) indicates that we must assume some version of Rizzi’s (1990) relativized minimality under which an element in [Spec. (38) a. b.4.3PL that in any moment SUBJ explode ‘The bombs can go off any minute’ (Grosu and Horvath 1987) a. As discussed in Nunes (2008a). as illustrated in (40)–(43).her help only Petre ‘Ion wants Mary to be helped only by Petre’ (Dobrovie-Sorin 1994: 124) a.her help ‘Ion is beginning to help Maria’ (Dobrovie-Sorin 1994: 124) ca ˆın orice moment s˘a explodeze b. both finite control and hyper-raising should be affected. (41) Romanian Ion vrea ca pe Maria s-o ajute numai Petre Ion wants that PE Maria SUBJ. ∗ Bombele pot The-bombs can. (42) Brazilian Portuguese (Nunes 2009a): [O Jo˜ao]i disse que o bolo.3SG that the-bombs SUBJ explode in any moment ‘It is possible that the bombs will go off any minute’ (Grosu and Horvath 1984: 352) b.4 Finite control 77 The second prediction made by the MTC is that if A-movement of the embedded subject is blocked by elements in other left-periphery positions. ∗ Example (40) shows that Brazilian Portuguese allows a left-dislocated element in the embedded clause.PRES. ti comeu The Jo˜ao seems that the cake ate ‘Jo˜ao seems to have eaten the cake’ a. ti comeu The Jo˜ao said that the cake ate ‘Jo˜ao said that he ate the cake’ ∗ b. o Jo˜ao comeu Seems that the cake the Jo˜ao ate ‘It seems that Jo˜ao ate the cake’ a. [O Jo˜ao]i parece que o bolo.PRES. (43) ∗ Romanian: Ion ˆıncepe ca pe Maria s-o ajute Ion starts that PE Maria SUBJ. o Jo˜ao comeu Someone said that the cake the Jo˜ao ate ‘Someone said that Jo˜ao ate the cake’ Parece que o bolo. Se poate ca bombele s˘a explodeze ˆın orice moment REFL can. (40) Brazilian Portuguese (Nunes 2009a): Algu´em disse que o bolo. whereas (41) illustrates the fact that the subjunctive complement introduced by the complementizer ‘ca’ in Romanian must have a . this prediction is borne out for both types of constructions regardless of whether they involve indicative or subjunctive embedded clauses. but not the adverbial elements quando ‘when’ or por que ‘why’ in (ib). Nunes suggests that the argumental wh-element in (ia) lands in a left-dislocated position on the way to the embedded [Spec. thus indicates that left-dislocated elements do count as proper interveners for A-movement out of the embedded clause. hence the argumental DP que livro ‘which book’ in (ia) counts as an intervener for the embedded subject. hyperraising of subjects involves movement through the embedded [Spec. and (38a) should also be unacceptable. This raises the independent question of why elements in [Spec. Ferreira’s proposal incorrectly predicts that obligatorycontrol sentences such as (35a). Ferreira (2000. Under this view. where the minimality effect is much more salient. 17 Based on the interesting contrast in (i) below. If the embedded . The idea is that once the embedded subjects of (i) are moving to receive the matrix external ␪-role. 2009) proposes that [Spec. ∗ O Jo˜ao n˜ao sabe que livro leu na The Jo˜ao not knows which book read in-the week past ‘Jo˜ao doesn’t know which book he read last week’ tk b. as sketched in (iii) below (with English words). which are not potential ␪-role bearers. (ii) Brazilian Portuguese (Ferreira 2000): Jo˜ao disse que esses livros. 2001. the contrast between the unambiguous left-dislocation structure in (ii) and (ia) can be accounted for if (ia) marginally allows the argumental DP to skip the embedded left-dislocation position. on the one hand. Kato 1999. and Negr˜ao 1999). namely. Grosu and Horvath (1984) and Dobrovie-Sorin (1994) for relevant discussion. In turn. as such. leu na semana passada The Jo˜ao said that these books read in-the week past ‘Jo˜ao said that he read these books last week’ ∗O Given the pervasive use of topic/left-dislocation structures in Brazilian Portuguese (see e. CP]. In other words. depending on the nature of its occupant. Britto 1997. CP] and left-dislocated elements contrast in this fashion. O Jo˜ao n˜ao sabe quando/por que leu esse livro The Jo˜ao not knows when/why read this book ‘Jo˜ao doesn’t know when/why he read this book’ Nunes (2009a) argues that the key to this puzzle is to be found in another contrast noted by Ferreira. Pontes 1987. (i) Brazilian Portuguese (Ferreira 2000): semana passada a.g.78 Empirical advantages left-dislocated constituent. Although able to correctly distinguish the acceptability patterns of (ia) from (ib). 2004.16 The contrast between (40)/(41).m 18 But see Nunes (2009a) for the suggestion that the relevant distinction is that C is a phase head and. minimality considerations should prevent them from crossing elements that are potential ␪-role bearers. (36a).g. CP] is a potential ␪-role bearer. CP]. (iii) [TP Jo˜aoi doesn’t [vP ti know [CP [which book]k [LD tk [TP ti read tk last week]]]]] <-∗ --.. Galves 1998. it allows multiple Specs for successive cyclic movement. given that the DP in the embedded [Spec. as in (ia). CP] may count as a proper intervener for the embedded subject. 2000. and (42)/(43). the marginality of (ia) is not due to a wh-phrase in [Spec. CP].18 it is worth stressing that whatever 16 See e. but to its trace in the left-dislocated position. as in (ii). on the other. and an analogous left-dislocation structure.17 Although at the moment we do not have anything conclusive to say on this issue. CP]. the contrast between a filled [Spec.. we saw that finite control and hyperraising constructions pattern alike with respect to reconstruction effects regarding the licensing of minimizers (section 4. and movement actually intersperse. if a given ␪-role fails to be assigned at DS. A-movement. indeed as a natural expectation. convergence cannot be enforced by having the relevant ␪-relation established by a later movement operation because the derivation is already ruled out at DS.1). and minimality (section 4. As they involve full-fledged clausal structures. The overall conclusion that their parallel behavior leads to is that they are derived through the same derivational resources.4. there is no room for movement to thematic positions.2). this picture completely changes.5 MTC under the copy theory of movement 79 the relevant minimality notion is that makes the correct distinctions. But it is worth emphasizing that. finite control provides one of the strongest empirical arguments for the MTC. in models having D-structure (DS) as one of their components.4. once DS is gone.4. it must group finite control and hyper-raising as a natural class and this is exactly what the MTC does. Thus. 4. Both are products of A-movement. no minimality issue arises as the two Specs are equidistant (see Chomsky 1995). lexical insertion and ␪-role assignment precede all movement operations. One can of course retain the old properties of DS by assuming that ␪relations can be established by merge but not by move (see Chomsky 1995). namely. ␪-role assignment. as is the case within minimalism. movement to thematic positions arises as a logical possibility. they permit a much more varied testing. In our opinion. once it is assumed that lexical insertion. Recall that under DS-based models. More specifically. the game is different and the CP already has a filled Spec. It is hard to see how alternative approaches on the market can account for the parallel behavior between finite control and hyper-raising without adding stipulations isolating finite control from standard non-finite control or replicating the A-movement operation involved under the guise of different technologies. . island effects.5 The movement theory of control under the copy theory of movement As we discussed in Chapter 3.4. 4. if the notion of DS is abandoned. as opposed to the leaner structures generally found in non-finite control.3 Summary Finite-control constructions are especially illuminating in the debate on how to analyze OC. As we also pointed out. In the sections above. and SS-based analyses by LF-based analyses in the case of the binding theory (Chomksy 1993). we have defended the claim that OC PRO is an empty category left by A-movement. In Chapter 3 and in the sections above. to name just the most prominent ones (see Chapter 2 for other additional assumptions required in different models).g. and if traces are copies under the copy theory of movement. everything that was discussed so far could be implemented in terms of traces or copies. Hornstein (2001). in fact. Nunes (1995. it permitted the replacement of DS. as it accounts for the distribution and interpretation of OC in terms of A-movement without enriching the theoretical apparatus. 2007). Hornstein and Nunes (2002). without pausing to discuss the ontology of this empty category. under the general architecture of the minimalist program. Hornstein.. which bans creationism in syntax in the sense that all syntactic structures must be built based on lexical material present in the numeration. in the domain of parasitic gaps and across-the-board movement. 2008) in the domain of anaphora. As such. Zwart (2002). it is the ban on movement to ␪-positions that requires proper justification. Thus far. copies are in compliance with this condition as they are replicas of lexical material present in the numeration or structures built from this lexical material. although traces violate the inclusiveness condition as there are no such entities in the lexicon.4. 19 For further evidence for movement to ␪-positions. . First. see e.3 we in fact mentioned that if OC PRO is a trace. Recall that the original motivation for the incorporation of the copy theory within the minimalist program was twofold. 2004). Thus. the copy theory is plausibly one of the solid architectural pillars of the minimalist program. Nunes and Uriagereka (2000). In other words. it is the ban on movement to ␪-positions that represents an enrichment of the theory if we keep the same assumptions constant and such an enrichment brings with it a full bag of additional provisos such as PRO as a special type of empty category and the control module. If movement to ␪-positions is a theoretical possibility allowed by the model. and Nunes (2007. we have argued that. and Lidz and Idsardi (1997). and Boeckx. Second. it made it possible that the output of movement operations was compatible with the inclusiveness condition (Chomsky 1995).19 Again. Kayne (2002). Hornstein (2001. OC PRO should also be a copy under the copy theory. However. 2001. the burden of proof is on proposals that exclude it. In section 3. the null hypothesis in a model that eschews DS receives substantial support in the domain of OC. as advocated by the MTC. as it is an independent proviso in a model lacking DS. Grohmann (2003).80 Empirical advantages evaluation of the assumptions of the new model also changes. We will shift gears now and examine the implications of the copy theory for the MTC in more detail. was1 . and the independently required merge operation. The relevant steps for our discussion are the ones represented in (45h– i). given in (45).4. promoted0 } VP = [promoted John␪ ] T = was Merger of ‘was’ and VP: TP = [was promoted John␪ ] Copying of ‘John’: TP = [was promoted John␪ ] N = John␪ Merger of ‘John’ and TP: TP = [John␪ was promoted John␪ ] Deletion in the phonological component: TP = [John␪ was promoted John␪ ] c. d. h. was1 . Consider the derivation of (44). was0 . there is actually no operation of movement employed in the derivation in (45).5 MTC under the copy theory of movement 81 It is worth stressing that this reanalysis of movement operations is not a notational variant of the trace theory just involving the substitution of one type of category without phonetic realization for another. As we can see. [45c–d]) or complex syntactic objects assembled in previous derivational . b. which replicates the targeted constituent (cf. promoted1 } Selection of ‘promoted’: Num = {John1 . j. [45h]). e. was1 . for instance. g. (44) John was promoted (45) a. promoted0 } V = promoted N = John Merger of ‘promoted’ and ‘John’: Num = {John0 . promoted0 } V = promoted Selection of ‘John’: Num = {John0 . promoted0 } VP = [promoted John␪ ] Selection of ‘was’: Num = {John0 . which assembles larger syntactic objects by combining lexical items (cf. was1 . f. promoted0 } VP = [promoted John] θ-marking of ‘John’: Num = {John0 . was1 . i. Rather. it has wholesale conceptual and empirical implications within minimalism. Num = {John1 . Displacement in natural languages is reanalyzed under the copy theory as the output of two basic operations: copy. [48d]) preserves the relevant configurations associated with each thematic relation. the matrix light verb should be able to assign its external ␪-role to the copy of ‘John’ with which it has merged. it is an unavoidable property of the system that merge can license ␪-assignment (cf. We will put this possibility aside for the sake of brevity. 20 Merge itself may be a composite of the more basic operations concatenate and label. Any departure from this expectation should be welcomed with skepticism and accompanied by solid evidence. as assumed in the minimalist program. d. as sketched in (48) below. c. the natural expectation is that in the derivation of an OC construction such as (46) below. merge. and copy: TP = [John␪1. . or a configurational view. if movement is not a primitive operation of the computational system. and copy: vP2 = [hopes [John␪1 to [vP1 John␪1 win]]] Copying of ‘John’: vP2 = [hopes [John␪1 to [vP John␪1 win]]] N = John␪1 Merger of ‘John’ and vP2 : vP2 = [John␪1 hopes [John␪1 to [vP John␪1 win]]] θ-marking of ‘John’ by the matrix v: vP2 = [John␪1. vP] and another copy merged into the matrix [Spec.␪2 hopes [John␪1 [vP John␪1 win]]]] Deletion in the phonological component: TP = [John␪1. establishing how to state this prohibition in a natural way is no trivial matter. θ-assignment. (46) John hopes to win (47) a. e. as sketched in (47c–d). Notice also that the point remains valid regardless of whether one assumes a featural view on ␪-roles. where ‘John’ can be interpreted as establishing the thematic relations ␪ 1 and ␪ 2 in virtue of having one of its copies merged into the embedded [Spec.␪2 T [John␪1. as in DSbased models. f. but through various applications of merge. as argued by Hornstein and Nunes (2008) and Hornstein (2009). keeping the DS-based ban on movement to ␪-position by assuming that merge but not move licenses ␪-assignment requires further elaboration.␪2 hopes [John␪1 [vP John␪1 win]]]] b. merge. [45d–e]). Applications of select.␪2 T [John␪1. However.2).␪2 hopes [John␪1 [vP John␪1 win]]] Further applications of select. as illustrated in (47). Crucially. both (47) and (48) can be appropriately interpreted as a complex monadic predicate such as (49) in the semantic component (see section 3. In other words. Notice that if syntactic structures are not built in a single step. for instance. That being so.4.82 Empirical advantages steps (cf. vP]. the structure that reaches LF (cf. [45h–i]).20 Thus. which treats movement as just one more case of merge (see Chomsky 2001. 4. If the copy of a given syntactic object can participate in syntactic relations that are licensed by merge such as case-. the copy theory unifies the kinds of thematic discharge found in control and non-control structures. [45d–e]). or EPP-checking. 2004). Once move reduces to copy plus merge.21 In the next sections we focus on specific empirical consequences of the copy theory for the MTC. N = John Merger of ‘John’ and vP2 : vP2 = [John hopes [John to [vP1 John win]]] ␪2 ␪1 d. even those arising from “movement. as given in (51). all instances of ␪-assignment. Take the derivation of (50). ␾-feature.1 Adjunct control and sideward movement Let us now examine another theoretical possibility that is made available once DS is eliminated and syntactic objects are built step by step through the operation merge. merge. and copy: TP = [John T [John hopes [John to [vP1 John win]]]] e. In addition.” result from merging into a thematic position. and copy: vP2 = [hopes [John to [vP1 John win]]] ␪1 b. for instance (cf. Further applications of select. should find the idea that ␪-roles could be discharged under remerge just as congenial. 83 Applications of select. the copy theory highlights the conceptual awkwardness of keeping the DS-based ban on movement to thematic positions within a system where traces are independently conceived of as targets of a copying operation.5 MTC under the copy theory of movement (48) a. [45h–i]). for instance. it should also be able to participate in ␪-relations as these are also licensed by merge (cf.5. Copying of ‘John’: vP2 = [hopes [John to [vP1 John win]]] ␪1 c. . if being merged into a ␪-position suffices for a given DP to get a ␪-role. (50) The man saw Jane 21 Those partial to the merge/remerge variety of the copy theory. merge. Deletion in the phonological component: TP = [John T [John hopes [John [vP John win]]]] ␪2 (49) ␪1 John (␭x [x hopes x to win]) In sum. being remerged should as well. In other words.4. Jane1 } V = saw Selection of ‘Jane’: Num = {the1 . T1 . two items must be selected from the numeration (cf.84 Empirical advantages (51) a. T1 . Jane0 } VP = [saw Jane] D = the Selection of ‘man’: Num = {the0 . e. man0 . [51c]) and under bare phrase structure each of these items constitutes a (root) minimal-maximal projection (Chomsky 1995). saw0 . k. saw0 . T1 . Jane0 } VP = [[the man] [saw Jane]] T Merger of T and VP: TP = [T [[the man] [saw Jane]]] Copying of [the man]: TP = [T [[the man] [saw Jane]]] DP = [the man] Merger of DP and TP: [[the man] T [[the man] [saw Jane]]] The exaggeratedly detailed presentation of the steps in (51) is meant to make it transparent that it is ordinarily common for a given derivational step to involve more than one root syntactic object (more than one “tree”) at a time. This is trivially true in the first steps of any derivation. Jane0 } VP = [saw Jane] Selection of ‘the’: Num = {the0 . man0 . saw0 . d. Jane0 } VP = [saw Jane] D = the N = man Merger of ‘the’ and ‘man’: Num = {the0 . man0 . g. Jane0 } VP = [[the man] [saw Jane]] Selection of T: Num = {the0 . T1 . T0 . But there are in addition two other general cases . f. l. T1 . man1 . j. Num = {the1 . Jane0 } VP = [saw Jane] DP = [the man] Merger of DP and VP: Num = {the0 . saw0 . b. man1 . c. Jane1 } Selection of ‘saw’: Num = {the1 . man0 . T1 . saw1 . saw0 . man1 . Jane0 } V = saw N = Jane Merger of ‘saw’ and ‘Jane’: Num = {the1 . i. man1 . In order for merge to start operating. saw0 . T1 . man1 . saw0 . h. T1 . saw0 . Applications of select.] M=␣ Merger of ␣ and L: K = [.. the syntactic object in (51h) can only be assembled if ‘the’ and ‘man’ in (51f) merge first and the resulting object [the man] then merges with [saw Jane] (cf. abstractly sketched in (52)..␣.. . a copy of ␣ from within K is made (cf... We are abstracting from this difference here. its derivation must involve a step analogous to (51f). b. [52b]) and merged with L. The interesting point for our discussion is that the derivational steps exemplified in (51) make room for an additional possibility. 23 Steps like the ones in (52) are referred to as an “interarboreal operation” by Bobaljik (1995) and Bobaljik and Brown (1997). c.] N = [␣ [L . [51g–h]).4. and as “paracyclic movement” by Uriagereka (1998.. whenever a given structure has a complex specifier. Nunes (1995) calls the steps sketched in (52) ‘sideward movement.. As exemplified in (51k). chapter 4).␣.. The first one is illustrated in (51e–h). for instance. yielding the syntactic object N in (52c).. before the copy gets merged we again have more than one root tree in the derivational working space.. note that there is no intrinsic difference between the “upward” movement seen in (51j–l). . with (at least) three root syntactic objects in the derivational working space.␣.] L = [.24 22 The same holds for complex adjuncts. Thus. we have trivial applications of movement..] L = [. ]] After the computational system builds the root syntactic objects K and L in (52a).’23 Terminological metaphors aside. Under the assumption that merge can only target root syntactic objects (Chomsky’s [1995] extension condition). If ‘man’ first merges with [saw Jane]..22 Finally. In both cases. viewed as copy plus merge..5 MTC under the copy theory of movement 85 where the computational system must handle more than one root syntactic object at a time.. See Uriagereka (1999) and Nunes and Uriagereka (2000) for relevant discussion. 2004) system. then this is not quite right. (52) a.. the last general case of more than one root syntactic object in a given derivational step arises when the computational system copies a substructure of a root syntactic object. . as in Chomsky’s (2000.] Copying of ␣: K = [. 2001. merge. and copy: K = [. and the “sideward” movement sketched in (52) in what regards the computational steps involved.. 24 If movement is parasitic on agree and agree requires c-command. the next merger yields the unwanted structure ∗ [the [man [saw Jane]]]. . Under the standard architectural features of minimalism. It is just a description of a specific interaction between copy and merge. . ] T0 If T and VP had merged in (53a). 2008). 2004). V . . However. as it has been consistently misunderstood. V . ] K = [T0 V [T0 T]] Merger of K and VP: TP T0 V [VP .25 (53) a. multiple root syntactic objects in a single derivational step not only are allowed in principle. and copy: VP = [ . . for instance. in order for the extension condition to be satisfied (cf. . The fact that ␣ in (52b–c). . . The derivation of V-to-T movement under the sideward-movement analysis sketched in (53) illustrates this point. may follow from two independent (therefore irrelevant) reasons: (i) that ␣ in (52b) has two root syntactic objects around to merge with. c. ] T V Merger of T and V (by adjunction): VP = [ . [51d–h]). b. Uriagereka (1998). V . Within DS-based models such as GB. does not merge with the structure that contains the “source” of the copy. Once DS is abandoned and structure building is carried out by merge. yielding [TP T VP]. . the picture is completely different within minimalism. . but merger with K may not. and Hornstein and Nunes (2002. for instance. as T would no longer be a root 25 For relevant discussion. . and (ii) that merger of ␣ with L in (52b) may be licensed by last resort. merge. Applications of select. but must indeed be employed in derivations involving complex specifiers. sideward movement is not a theoretical possibility.86 Empirical advantages This point is worth emphasizing. see Bobaljik (1995). ] T Copying of V: VP = [ . . Nunes (1995. V . . Bobaljik and Brown (1997). whereas [the man] in (51j) only has one. . d. sideward movement is therefore not a novel operation or a new species of movement. . for all syntactic computations operate with a single root tree – the one made available by DS. . . as opposed to [the man] in (51j–l). 2001. the extension condition should then prevent the verb from adjoining to T. . 27 For reasons that will become clear in section 4.g. However. 28 These cases were pointed out to us by John Nissenbaum (personal communication). adjunct-control PRO obligatorily requires a de se interpretation:28 The unfortunate wrote a petition (in order) PRO to get a medal ‘[The unfortunate]i wrote a petition so that [he himself]i would get a medal’ b. but not the latter.1. This is not the place to defend the virtues of sideward movement. As illustrated in (55). it is excluding sideward movement as a grammatical possibility that requires additional theoretical devices. V-to-T adjunction can comply with the extension condition if it proceeds as in (53b–d). Adjunct-control PRO requires a local c-commanding antecedent: Johni said [that [Maryk ’s brother]m left [after PROm/∗ i/∗ k/∗ w eating a bagel]]27 Adjunct-control PRO only licenses sloppy readings under ellipsis: John left before PRO singing and Bill did too ‘and Billi left before hei /∗ John sang’ Adjunct-control PRO can only have a bound interpretation when controlled by only-DPs: Only Churchill left after PRO giving the speech ‘[Nobody else]i left after hei /∗ Churchill gave the speech’ In the appropriate type of adjuncts (e. d. the copied V must merge with T rather than VP as it arguably has features to check with the former. adjunct control has virtually the same properties as OC into complements. given the standard minimalist assumptions reviewed above.2 below. Thus.5 MTC under the copy theory of movement 87 syntactic object. and Drummond (2009) for an extensive discussion of the conceptual and empirical virtues of sideward movement. Hornstein (2001). sideward movement naturally arises within the system without enriching the grammatical apparatus. apparent problems. we have argued that the properties illustrated in (55a–d) are signature properties of A-movements/chains. c. Crucially. Let us then examine how sideward movement allows OC into complements and 26 See Nunes (1995. purposives). sideward movement comes for free as one more instantiation of copy plus merge. once the derivational step in (53b) is reached. and solutions to prevent many cases of overgeneration. To put it in general terms. 2004). (54) Johni saw Mary after/before/while PROi eating a bagel (55) a. unless explicitly prohibited..4.5. not those that allow it. The theories that exclude sideward movement by invidiously distinguishing the merger in (51k–l) from the one in (52b–c) are the methodologically profligate ones. 2001. . Consequently. PRO in (55a) can be interpreted as ‘John’ if the adjunct clause is interpreted as modifying the matrix verb.26 Rather we wish to illustrate how it allows a smooth analysis of adjunct-control constructions such as (54) below. In the preceding sections and chapters. T␾+ 1 . d. h. lunch0 } [VP [VP John saw Mary] [PP after John T␾− eating lunch]] T␾+ Merger of T ␾+ and VP: TP = [T␾+ [VP [VP John saw Mary] [PP after John T␾− eating lunch]]] Copying of ‘John’: TP = [T␾+ [VP [VP John saw Mary] [PP after John T␾− eating lunch]]] N = John Merger of ‘John’ and TP: TP = [John [T␾+ [VP [VP John saw Mary] [PP after John T␾− eating lunch]]]] Deletion in the phonological component: TP = [John [T␾+ [VP [VP John saw Mary] [PP after John T␾− eating lunch]]]] b. TP] (cf. Note that both the sideward movement in (57b–c) and the upward movement in 29 See Hornstein (1999. ‘saw’ still has its external ␪-role to assign. lunch0 } PP = [after John T␾− eating lunch] VP = [saw Mary] Copying of ‘John’: PP = [after John T␾− eating lunch] VP = [saw Mary] N = John Merger of John and VP: PP = [after John T␾− eating lunch] VP = [John saw Mary] Merger of PP and VP: [VP [VP John saw Mary] [PP after John T␾− eating lunch]] Selection of T ␾+ : Num = {John0 . but there is no remaining element in the numeration to receive it. [57c]). T␾− 0 . for instance. gets copied and merged into) [Spec. . [57d]).. 2001. Notice. allowing the external ␪-role to be discharged (cf. Take the (simplified) derivation of (56). merge.e. that ‘John’ has not checked case within the gerund clause (its T-head does not have a complete ␾-set) and is therefore still active for purposes of A-movement.29 (56) John1 saw Mary after PRO1 eating lunch (57) a.88 Empirical advantages adjunct control to be both subsumed under a movement analysis. [57b]) and merges it with the VP (an instance of sideward movement). after0 . eating0 . [57g–h]). however. Mary0 . The computation then creates a copy of ‘John’ (cf. saw0 . Mary0 . i. T␾+ 0 . and copy: Num = {John0 . the matrix TP is built (cf. [57f]). given in (57). Applications of select. c. After VP and PP have been assembled in (57a). T␾− 0 . saw0 . The PP then adjoins to VP (cf. after0 . g. 2003) for discussion. f. e. and the matrix subject moves to (i. eating0 . The derivation sketched in (57) shows that A-movement of ‘John’ from within the adjunct to a thematic position (the external argument of ‘saw’) yields a control structure.MASC to o hospital the hospital ‘The victim died after being brought to the hospital’ In the next sections we discuss three additional welcome consequences of the sideward-movement analysis of adjunct control sketched in (57). this pattern of agreement also obtains in adjunct-control configurations such as (58). Were ‘John’ not to move.4. Recall from section 4. As observed by Rodrigues (2004). just like the trace left by upward movement in instances of raising and subject control (cf. Italian (Rodrigues 2004): La vittima mori’ dopo essere stata trasportata /??stato The victim died after be been. if PRO in (56) is a residue of (sideward) movement to a thematic position. the derivation would crash as it could not get its case checked within the adjunct clause. The adjunct-control analysis in terms of sideward movement outlined in (57) also accounts for the fact that adjunct control does not differ from subject or object control in terms of interclausal agreement. .3 that epicene nouns such as the counterpart of ‘victim’ in Romance trigger [+feminine] agreement even in contexts when the victim is a male.MASC to-the hospital b. Notice also that sideward movement provides an escape hatch for ‘John’ to have its case checked. Thus.FEM brought.MASC trasportato all’ ospedale brought. As the data in (55) indicate. The MTC when interpreted under the copy theory predicts that any instance of obligatory control should involve applications of copy and merge and obligatory adjunct control should be no different. Brazilian Portuguese (Rodrigues 2004): A v´ıtima morreu depois de ser trazida /??trazido para The victim died after of be brought. It is worth pointing out that no special proviso was added to the system in order to achieve this result. which indicates that the trace left by sideward movement of ‘the victim’ triggers [+feminine] agreement on the embedded participial morphology. our expectations are not disappointed. (58) a.5 MTC under the copy theory of movement 89 (57g–h) extend the targeted tree. we expect it to function like an (A-)trace and we expect the relation between ‘John’ and PRO in (56) to manifest all the properties of A-chain dependencies.FEM brought.FEM been. [8] and [9] above). merge. the minimal-distance principle blocks object control into adjuncts on the assumption that ‘John’ but not ‘Mary’ c-commands the adjunct. 31 See Nunes (1995. eating0 . T␾+ 1 . contrary to fact (cf.’ which is still active for purposes of A-movement in virtue of having its case unchecked. where move is understood as copy plus merge. as Orson Welles taught us: (60) John will drink [no wine]i before iti is ready for drinking If we assume that in order to be interpreted as a bound variable a pronoun must be c-commanded by its antecedent. Roughly speaking. 2001.1 below. If ‘Mary’ is selected and merged with ‘saw.5.’ which is still in the numeration. T␾− 0 . That is. after0 . saw0 .’ 30 But see section 7.3. But then objects should be able to control into adjuncts. .31 With this in mind. this proposal can be interpreted as saying that the operation copy is costly and should be employed only under convergence pressure. ‘saw’ must assign its internal ␪-role and there are two potential candidates to receive it: ‘Mary.2.1. this implies that ‘it’ in (60) must be c-commanded by ‘no wine’ at least at LF. For example. merge trumps move when both are available and both lead to convergent derivations. (61) Applications of select. as illustrated in (59). [59]).90 Empirical advantages 4. this assumption is not obviously correct. lunch0 } PP = [after John T␾− eating lunch] VP = saw In (61). The sideward-movement account outlined in (57) allows us to explain this subject/object asymmetry if we assume with Chomsky (1995) that movement is subject to economy. More specifically. where we discuss cases of adjunct control in Portuguese where we find subject or object control depending on whether we have wh-in situ or wh-movement. and copy: Num = {John0 . Unfortunately. Mary1 . consider the derivational step in (61). it is possible for objects to bind pronouns found within adjuncts. 2004) and Hornstein and Nunes (2002) for relevant discussion. In the context of the copy theory.1 Subject–object asymmetry in adjunct control A very distinctive property of adjunct control is that PRO must be controlled by the subject and not the object of the next higher clause. and ‘John. as discussed earlier. Chomsky has proposed that move is less economical than merge.30 (59) Johni saw Maryk after PROi/∗ k eating lunch Rosenbaum (1970) attempts to account for this subject–object asymmetry by extending the minimal-distance principle to cases of adjunct control. Notice that once ‘Mary’ becomes the object of ‘saw’ (cf. if ‘John’ is copied and merged with ‘saw. 2004). VP] (cf. the derivation should in principle converge as well. [57a]). [62a]) at a derivational step where selection and merger of ‘Mary’ would suffice to yield a convergent result (cf. . as there is no alternative option that leads to convergence. There are various ways of ensuring preference of merger over movement in these contexts. e. [57b–e]). g.5 MTC under the copy theory of movement 91 as seen in (57a). if economy independently restricts movement/copying and sideward movement is just an instance of copy plus merge. however.32 32 The result is actually a bit more robust than this. Copying of ‘John’: PP = [after John T␾− eating lunch] V = saw N = John Merger of ‘John’ and V: PP = [after John T␾− eating lunch] VP = [saw John] Selection and merger of ‘Mary’: PP = [after John T␾− eating lunch] VP = [Mary saw John] Merger of PP and VP: [VP [Mary saw John] [PP after John T␾− eating lunch]] Selection and merger of T␾+ : TP = [T␾+ [VP [Mary saw John] [PP after John T␾− eating lunch]]] Copying and merger of ‘Mary’: [TP Mary [T␾+ [VP [Mary saw John] [PP after John T␾− eating lunch]]]] Deletion in the phonological component: [TP Mary [T␾+ [VP [Mary saw John] [PP after John T␾− eating lunch]]]] Note. In this scenario. that the derivation in (62) violates economy as copying of ‘John’ is employed (cf.’ as it becomes inactive for purposes of A-relations. it receives accusative case and therefore cannot check the external ␪-role of ‘saw.4. In sum. In turn. d. this time yielding an object-control structure. copying of and merger of ‘John’ are indeed legitimate (cf. This possibility is explored in Nunes and Uriagereka (2000) and Nunes (2001. then the restriction to subject control into adjuncts is what we expect (and find). one may piggyback on Chomsky`s (2000) proposal that a numeration is organized in subarrays. and require that material that has been selected and integrated into the structure can be accessed again only after all relevant elements of the active subarray have been used. the derivation converges as a subject-control structure. f. after ‘John’ undergoes sideward movement to [Spec.’ as shown in (62). [57]). (62) a. For example. b. c. [57f–g]). each is an independent syntactic object. TP] depicted in (67). it should obviously apply to adjuncts. incorrectly bleeding CED effects? The challenge for the MTC is thus to provide an account that at once explains why movement leads to CED effects in (63) but not in (64). the question that arises is why sideward movement in the adjunct-control construction in (64) below (cf. Copying from within the PP should in fact be no different from the copying that takes place in standard instances of movement such as movement of the subject from [Spec. showing that such movement does not lead to acceptable results. to put matters slightly differently. and copy: [Did [John [vP [vP talk to Mary] [PP [which book] after he read [which book]]]]] Whatever the ultimate analysis of adjunct islands turns out to be.92 Empirical advantages 4.1. displays a typical CED effect (see Huang 1982). for instance. This is exactly the case in (65). but a relational notion. what prevents sideward movement from applying in (63). merge. (64) [Johni [vP [vP ti saw Mary] [PP after ti eating lunch]]] Theoretically accounting for the difference is less difficult than it may appear to be. But notice that the notion of being an adjunct is not an absolute. (65) Applications of select. The PP is not an adjunct of the vP at this derivational step. A sentence such as (63). Or. In other words. (63) ∗ [[Which book]i did [John [vP [vP talk to Mary] [PP after he read ti ]]]] Given the unacceptability of (63). but an independent tree. Let us more closely examine the derivational steps prior to movement of ‘John’ in (64) and ‘which book’ in (63). as respectively represented in (65) and (66). Before that. . and copy: PP = [after John T␾− eating lunch] vP = [saw Mary] (66) Applications of select. [57]) does not violate the CED. the proposal that adjunct control is also a product of (A-)movement seems to face problems when familiar instances of movement out of adjuncts are taken into consideration.2 Adjunct control and CED effects At first sight. a given constituent X is an adjunct of Y only after X and Y are merged. vP] to [Spec.5. merge. the PP has already adjoined to the vP (cf. To derive (63) without incurring a CED violation requires moving ‘which book’ from within the PP before the PP is adjoined to the vP. [66]). if PP is adjoined to the vP “at the right time” (cf. CP]. Thus. it becomes an adjunct island for later movements from within it. However. c. b. if it is to remain porous to movement. there is no convergent continuation for (68c). extraction out of it should indeed yield a CED effect. [68]). Therefore. merge. merger between PP and vP violates the extension condition. at the derivational step where the wh-phrase should move out of the PP. [66]). and copy: PP = [[which book] after he read [which book]] CP = [did [John [vP talk to Mary]]] Copying of ‘which book’: PP = [[which book] after he read [which book]] CP = [did [John [vP talk to Mary]]] DP = [which book] Merger of ‘which book’ and CP: PP = [[which book] after he read [which book]] CP = [[which book] [did [John [vP talk to Mary]]]] Although sideward movement of ‘which book’ in (68b–c) allows the relevant wh-features to be licensed. Applications of select. c. as illustrated in (i) below. However. where the PP remains an independent syntactic object throughout the computation.33 33 One wonders what blocks a derivation in which the wh-element of (63) undergoes sideward movement to an outer [Spec. the extension condition prevents PP from merging with vP. once vP has been integrated into a larger structure. In either case. b. To sum up. If adjunction of PP is delayed to allow feature checking later on in the derivation (cf. the adjunct cannot be merged until C has been added to the derivation. 93 Applications of select and merge: TP = [T [vP he saw her]] Copying of ‘he’: TP = [T [vP he saw her]] D = he Merger of ‘he’ and TP: [TP he [T [vP he saw her]]] As for the derivation of (63). a potential derivation of (63) employing sideward movement should involve the derivational steps in (68). vP] before PP adjoins to vP. this movement targets [Spec.5 MTC under the copy theory of movement (67) a. the derivation of (63) is correctly blocked. which .4. Notice that the PP in (68c) must adjoin to vP for interpretive reasons. (68) a. In other words. See Nunes (2008c) for further discussion. although the copy of ‘which book’ inside PP is trapped within the adjunct. however. the copy in the outer [Spec. as shown in (70) below. CP]. (i) a. the wh-phrase will then remain in the edge of PP until PP adjoins to vP. but only the subject of the next clause up. as it would be crossing an adjunct island. where the gerund is adjoined to the embedded clause. Notice. and copy: PP = [[which book] after he read [which book]] vP = [vP John talk to Mary] b. PRO cannot take the matrix subject as its antecedent. as we are assuming here. vP]) is introduced in the derivation it is already too late. when the relevant target for the movement of ‘John’ (the matrix [Spec. Similar to what happens to the wh-phrase in (66). . Copying of [which book]: PP = [[which book] after he read [which book]] vP = [vP John talk to Mary] DP = [which book] c. Crucially. which basically forces elements that are in the complement domain of a phase head to move to its edge so that later relations can take place (see Boˇskovi´c 2007 for relevant discussion). Movement of ‘John’ to the matrix [Spec. If computational decisions are to be made locally without look-ahead. (69) [Johni left the room [after Maryk answered the questions without PROk/∗ i understanding them]] In order for the matrix control in (69) to be obtained under the MTC. vP] is free to undergo further movement to [Spec. should then yield the sentence in (63) in compliance with the extension condition. Merger of PP and vP (by adjunction): [vP [vP [which book] [v’ John talk to Mary]] [PP [which book] after he read [which book]]] We would like to suggest that sideward movement in (ib–c) is prevented by last resort and the ban on global computations with look-ahead. as ‘John’ is trapped within an adjunct island. Applications of select. for instance. vP] should then yield a CED effect. A standard assumption within the phase-based model (Chomsky 2000.94 Empirical advantages Notice that this analysis also derives locality effects in adjunct-control constructions. In a sentence such as (69). Thus. 2004) is that movement to the edge of a phase is not driven by regular feature checking. 2001. merge. but by the phase impenetrability condition. ‘John’ should be generated in the adjunct clause. copying of the wh-phrase at this derivational step is not licensed by last resort. that the wh-phrase in (ia) is already at the edge and is not in the complement domain of any phase head. Merger of DP and vP: PP = [[which book] after he read [which book]] vP = [[which book] [v’ John talk to Mary]] d. before that. Hornstein 2001. every application of merge (including adjunction) is subject to the extension condition. and Hornstein and Nunes (2002). Deletion in the phonological component: [[everyone having to get up early] upset everyone] 36 Starting with Chomksy (1993). For a reanalysis of the principal empirical arguments for assuming that adjunction is not subject to extension. 2004). the operation of copy is subject to last resort. one cannot idly create a copy and 34 For a full elaboration see Hornstein (2001).34 First. Merger of L and M: K = [everyone having to get up early] N = [upset everyone] d.4. sideward movement in a sense bleeds CED by applying before CED becomes relevant. Applications of select. and copy: K = [everyone having to get up early] L = upset b. once syntactic structures are built step by step. Nunes (2001. analyze Higginbotham’s (1980) PRO-gate phenomena illustrated in (i) in terms of sideward movement of a given DP from within an XP before this XP becomes a subject island. see Nunes (1995. and Nunes 2001. it is a porous domain like any other root syntactic object. 2004).2 below for further discussion). for relevant discussion). computed in a local fashion (without look-ahead). relational notions such as adjunct of are only instantiated if the relevant objects have been merged. 2004. (i) [[PROi having to get up early] upset everyonei ] (ii) a. for instance.35 A given XP will become an adjunct island only after it adjoins to its target. . Copying of ‘everyone’: K = [everyone having to get up early] L = upset M = everyone c. as illustrated in (ii) (see section 6. merge.36 Third. it has been often assumed that the extension condition does not apply to adjunction. 35 The same considerations of course apply to the notion specifier/subject of (see Nunes and Uriagereka 2000. Clearly an approach that treats all instances of merge as subject to extension holds the methodological high ground. Thus. Hornstein and Kiguchi (2003) and Kiguchi (2004). Merger of K and N: [[everyone having to get up early] upset everyone] e. 2001.5 MTC under the copy theory of movement (70) 95 PP = [after Mary [vP [vP answered the questions] [PP without John understanding them]]] vP = [left the room] It is worth stressing that the analysis of the contrast between (63) and (64) advocated above explores very natural assumptions within minimalism. Second. Thus. 96 Empirical advantages leave it dangling because it will be useful later on in the derivation.1. d. 38 For data and relevant discussion.37 In the next section we will see that this analysis also receives independent support from finite control.38 (71) Brazilian Portuguese: O Jo˜ao viu a Maria depois que entrou na sala The Jo˜ao saw the Maria after that entered in-the room ‘Johni saw Maryk when hei /∗ shek entered the room’ The (simplified) derivation of (71) proceeds along the lines of (72) (with English words for convenience). b. which displays all the diagnostics of OC. This analysis also extends to finite adjunct clauses. e. When associated with T␾− . see Ferreira (2000. yielding finite-control configurations. which then remains active for purposes of A-movement. we saw that Brazilian Portuguese displays finite control into indicative clauses. Nunes (2001. and copy: K = [CP after that Jo˜ao T␾− entered the room] L = [vP saw Maria] Copying of ‘Jo˜ao’: K = [CP after that Jo˜ao T␾− entered the room] L = [vP saw Maria] M = Jo˜ao Merger of L and M: K = [CP after that Jo˜ao T␾− entered the room] N = [vP Jo˜ao saw Maria] Merger of K and N (by adjunction): [vP [vP Jo˜ao saw Maria] [CP after that Jo˜ao T␾− entered the room]] Selection and merger of T␾+ : [TP T␾+ [vP [vP Jo˜ao saw Maria] [CP after that Jo˜ao T␾− entered the room]]] Copying and merger of ‘Jo˜ao’: [TP Jo˜ao [T’ T␾+ [vP [vP Jo˜ao saw Maria] [CP after that Jo˜ao T␾− entered the room]]]] 37 For additional arguments and relevant discussion. merge. (72) a. which was attributed to the possibility that its indicative Theads be associated with person and number (T␾+ ) or number only (T␾− ). Copies are not made until prompted by their targets. f. 2004). 4.4.5. . 2009) and Rodrigues (2004). as illustrated in (71). Applications of select. 2004.3 Finite adjunct control In section 4. c. indicative finite clauses do not value the case feature of their subject. see Hornstein (2001). and Hornstein and Nunes (2002). 5. ‘Jo˜ao’ would have to be generated in the adjunct clause. [72b–c]). when sideward movement occurs (cf.1.2 predicts that a CED effect should be detected. as argued earlier. [76] with English words).5. Crucially. the null subject inside the adjunct can only be interpreted as ‘Maria. That this prediction is correct is shown by the sentence in (34a). the null embedded subject of (71) cannot be controlled by the object. and section 7. vP] in (74) (with English words).1 below for further discussion). as in the case of non-finite control. CP is an independent tree and no adjunct violation arises. ‘Jo˜ao’ can then be copied and merged with vP and receive the external ␪-role of the matrix clause. Therefore.1.2. .39 By contrast. [69]).’ If it were to be interpreted as the matrix subject. but movement to the matrix [Spec.1. 97 Deletion in the phonological component: [TP Jo˜ao [T’ T␾+ [vP [vP Jo˜ao saw Maria] [CP after that Jo˜ao T␾− entered the room]]]] The T-head of the adjunct clause in (72a) is not ␾-complete and therefore cannot freeze ‘Jo˜ao’ for purposes of A-movement.5 MTC under the copy theory of movement g. (73) ∗ (74) Brazilian Portuguese: [O Jo˜ao disse [que [o bolo [que comeu]] n˜ao estava bom]] The Jo˜ao said that the cake that ate not was good ‘Jo˜ao said that the cake that he ate was not good’ [vP said that [the cake [that Jo˜ao T␾ ate]] was not good] Another correct prediction that the analysis presented in section 4. vP] should then yield a CED effect (cf. as illustrated in (75) below. (75) Brazilian Portuguese: [O Jo˜ao saiu da sala [depois que a Maria gritou The Jo˜ao left of-the room after that the Maria screamed porque estava com medo]] because was with fear ‘Jo˜ao left the room [after Maria screamed because she/∗ he was scared]’ 39 Notice that. if the relevant clause has already been adjoined.5.2 (see also Modesto 2000. Rodrigues 2004. Nunes 2008c. This subject–object asymmetry can therefore also be accounted for in terms of the “merge-over-move” economy preference discussed in section 4.2 is that finite adjunct control should also display sensitivity to locality (cf. the discussion in section 4.3.4. repeated below in (73). adjunct finite control in Brazilian Portuguese provides additional evidence that adjunct control should be analyzed in terms of sideward movement. which should involve the illegitimate movement of ‘Jo˜ao’ to the matrix [Spec. When the adjunct modifies the intermediate clause. The subject orientation observed in adjunct control can be accounted for in terms of the “merge-over-move” economy preference. Moreover. The reason is that adjunct control cannot be reduced to the thematic requirements of an embedding control predicate (as is standard in cases of control into complement clauses).2 and 7. In effect. based as they are (at least in part) on selection properties of the embedding predicate. we have been discussing interesting consequences revolving around the reinterpretation of the operation move as copy plus merge. Nonetheless. We find it interesting that cases of adjunct control can be analyzed in terms of movement.2 The movement theory of control and morphological restrictions on copies So far. given some natural minimalist assumptions. whatever control one finds here cannot be a function of the properties of the matrix predicate. Moreover. Most other approaches to control. and the fact that the sideward movement involved in adjunct control is not subject to the CED is explained in terms of the derivational timing when such movement takes place. at least in adjunct control. The MTC’s ability to unify complement and adjunct control thus constitutes another argument in its favor.40 4. This suggests that. Thus.3. as there is no thematic relation between the matrix verb and the adjunct. We now turn to 40 See sections 7. we believe. it is possible to assimilate adjunct control to the MTC quite straightforwardly. . have had little to say about these cases despite their having all the signature properties of OC found in the case of complements.4 Summary The above discussion shows that. the controller is structurally specified.2 below for a detailed discussion of this point. we have seen that the account proposed also captures finite adjunct control. adjunct control displays all the diagnostic properties of OC found in control into complements.1. as is usually assumed in cases of control into complement clauses.5.98 (76) Empirical advantages [vP [left the room] [PP after Maria [vP [vP screamed] [because Jo˜ao T␾− was scared]]]] 4. if the properties of OC PROs within adjuncts are structurally determined.5. then it would be very odd to treat cases of control into complements (where the very same properties appear) as derived by entirely different operations and in entirely different ways. the MTC offers a complete analysis of adjunct control within a framework of minimalist assumptions. Martins and Nunes propose that (77a) and (77b) are to be respectively analyzed along the lines of (79) and (80) below (with English words). By contrast. Custa-nos (∗ a) haver pessoas com fome Costs-us to have. they argue that the infinitival subject of (77a) has its case licensed clause-internally. In (77a)/(79).5 MTC under the copy theory of movement 99 other empirical implications of the implementation of the MTC under the copy theory by inspecting the copies themselves.INF the house ‘It was hard for me/∗ him/∗ her to succeed in selling the house’ Martins and Nunes (2008) show that these sentences contrast not only in their meaning. and expletive subjects.3PL the house ‘It saddened me that they had to sell the house’ ∗ ‘It was hard for me to get them to greet the boss’ b. the null subject inside the infinitival clause is a pro. Custa-me (∗ a) tˆe-lo despedido Costs-me to have.4. (78) European Portuguese (Martins and Nunes 2008): Custou-me (∗ a) venderem a casa Cost-me to sell. b. the embedded null subject in (77b)/(80) is a trace of the experiencer. they show that only structures such as (77a) − without the preposition − allow inflected infinitives. which may be coreferential with the experiencer complement of the matrix verb.INF people with hunger ‘It saddens us that there are hungry people’ ∗ ‘It is hard for us to succeed in causing people to be hungry’ a. European Portuguese (Martins and Nunes 2008): Custou-me vender a casa Cost-me sell. as respectively illustrated in (78). Taking the presence or absence of the preposition a ‘to’ as a diagnostic for which structure is at stake.INF-him fired ‘It saddens me that I had to fire him’ ∗ ‘It is hard for me that I succeeded in firing him’ c. Based on these differences.INF the house ‘It saddened me that I/he/she had to sell the house’ Custou-me a vender a casa Cost-me to sell. independent tense. but also in their structural properties.INF. More specifically. as opposed to . in other words. Consider the superficially similar constructions in (77) from European Portuguese. (77) a.41 41 Martins and Nunes (2008) show that constructions such as (77b) indeed display all the diagnostics of obligatory control. contrasting with constructions like (77a). For instance. but not the infinitival subject of (77b). (77b) is an OC construction. as in (81b).SE early ‘Jo˜ao got up early’ a. In turn.SE early in-this country ‘One can get up early in this country’ Example (81) shows that in Portuguese. Custou-lhe depor as armas Cost-him lay-down the arms ‘It saddened him that the documentary character had to lay down his arms’ b. (81) European Portuguese (Martins and Nunes 2008): O Jo˜ao levantou-se cedo The Jo˜ao raised.100 Empirical advantages (79) [TP proexpl cost mei [proi sell the house]] (80) [TP proexpl cost mei to [ti sell the house]] What is relevant for our present discussion is an additional contrast discussed by Martins and Nunes. custou-lhe tamb´em Cost-me to drink that and to him cost-him too ‘It was hard for me to succeed in drinking that and it was also hard for him to succeed in drinking that/∗ in having me drink that’ (ii) European Portuguese (Martins and Nunes 2008): [Context: an amnesic soldier watches a documentary in which he is the protagonist.SE well in-this city ‘One lives well in this city’ (82) a. #Custou-lhe a depor as armas Cost-him to lay-down the arms #‘It was hard for him to succeed in laying down his arms’ . Custou-me a beber aquilo e a ele. or an indefinite. b. (82) shows constructions such as (77a). as in (81a). Vive-se bem nesta cidade Lives.SE raise. as illustrated in (i) and (ii) (see Martins and Nunes [2008] for the other diagnostics and further discussion). the third-person clitic se is ambiguous between a reflexive. Custou-me beber aquilo e a ele. ∗ European Portuguese (Martins and Nunes 2008): Levanta-se-se cedo neste pa´ıs Raises. constructions such as (77b) require sloppy reading under ellipsis and de se interpretation in the relevant contexts.SE-SE early in-this country ‘One gets up early in this country’ Pode-se levantar-se cedo neste pa´ıs Can. Let us first consider the background data in (81)–(82). but he does not remember that the person being shown is him himself] a. custou-lhe tamb´em Cost-me drink that and to him cost-him too ‘It was hard for me that I had to drink that and it was also hard for him that I/he had to drink that’ b. (i) European Portuguese (Martins and Nunes 2008): a. ∗ b. In (82b) the clitics are each associated with a different verb and an ungrammatical result still obtains as the clitics are co-occurring within the same clause (the finite verb is a modal auxiliary). (83) European Portuguese (Martins and Nunes 2008): Custou-me levantar-me cedo Cost-me raise-me early ‘Getting up early is hard for me’ a. ∗ Custou-me a levantar-me cedo Cost-me to raise-me early ‘It was hard for me to succeed in getting up early’ These sentences indicate that the first-person reflexive clitic me can co-occur with a homophonous experiencer in the matrix clause when control is not involved (cf. it is not at all obvious how contrasts such as the one in (83) can be captured under a PRO-based account. As Martins and Nunes argue. From this perspective.4. b. and the question is why this should be so. . [83b]). With this restriction in mind. (84) [TP proexpl cost mei [proi [raise me early]]] (85) [TP proexpl cost mei to [mei [raise me early]]] In (83a)/(84) the embedded reflexive has as its clause-mate a pro coreferential with the matrix experiencer. we find two instances of ‘me’ in the embedded clausal domain in (85). the contrast in (83) can receive a natural explanation if one adopts the MTC as implemented under the copy theory. In the OC construction in (83b)/(85). Example (82b) further shows that the problem with (82a) is not that we have two clitics associated with a single verb. As Martins and Nunes also point out. Thus. the trace in the OC structure in (80) is a copy of the matrix experiencer. given that the reflexive and the experiencer should be in different clauses in each of the sentences in (83). the OC construction replicates the restriction seen in (82a). Thus. the embedded subject is a copy of the experiencer. where no control is involved. consider now the interesting contrast in (83). [83a]). the relevant structures associated with (83a) and (83b) are the ones given in (84) and (85) (with English words and superscripted indices annotating copies). but not when OC is at stake (cf. This indicates that the ungrammaticality of (83b) in effect reduces to the independent restriction in the morphological component banning two identical clitics in the same clause seen in (82). as represented in (86) and (87) (with English words). on the other hand. but not in (84).5 MTC under the copy theory of movement 101 that the two uses cannot coexist within a single clause. In other words. Backward control.. . and Alexiadou. is also attested in several languages. ]] [DP V 1 [DP1 .5. by assuming the MTC and the copy theory. that OC is derived by copying/movement. [1 V [DP1 . . ]] [DP V DP1 [1 . In this section we discuss the implications of the existence of backward-control constructions for the debate on how to better account for OC. the controller appears in a lower position than the controllee.3 Backward control So far we have examined customary cases of forward control. (88) a. We then proceed to present actual examples of backward control in two different languages: Tsez and Korean. Alboiu (2007) on Romanian. Potsdam (2006) on Malagasy.1 Backward control and PRO-based approaches to control PRO-based approaches assume that control is an inter-chain relation in which the phonetically unexpressed ␪-role carrier is PRO and that this inter-chain 42 For further illustrations of backward control. b. .42 4. . More precisely. 4. Monahan (2003) on Korean. we expect OC PRO to be subject to whatever computations and restrictions apply to its antecedent in the phonological component. Haddad (2007) on Telugu and Assamese. b. ]] Backward-control constructions have the following two defining properties: (i) they exhibit a control interpretation in that a single DP is interpreted as being associated with two or more ␪-roles. . and Marchis (2008) on Greek. We start by indicating the problems that backward control poses to PRO-based theories and how they can receive a natural explanation under the MTC.g. ]] (89) a. Anagnostopoulou. Iordachioaia. as sketched in (89). we expect OC PRO to behave like a regular copy of its antecedent. . [DP1 V [1 . and (ii) in overt syntax. .102 Empirical advantages (86) [TP proexpl cost mei [proi [raise me early]]] (87) [TP proexpl cost mei [PROi [raise me early]]] To conclude. where ‘’ is meant to be neutral regarding the grammatical nature of the controllee. This is exactly what we find in OC involving experiencers in European Portuguese.5. see e. therefore. as represented in (88) below. in which the controllee is c-commanded by the controller. . The fact that the morphological restriction ruling out two identical clitics applies to OC PRO but not to a coreferential pro strongly indicates that OC PRO is a copy of its antecedent and. . Polinsky and Potsdam (2002) on Tsez.3. . especially given . PRO is allowed to occupy both the subject and the object position. The conservative strategy. [PRO1 V [DP1 .e.. Under this view. (91) a. Consider the distributional properties of PRO. if binding generally involves c-command (i. this should induce a disjoint reference or strong crossover effect.g. regardless of the tense properties of the clause containing it. However. when PRO occupies the subject position.. PRO-based accounts of control violate the basic principles of binding and so one or the other must be abandoned. More precisely. or the tense properties of the heads C and T of the clause containing PRO (e. In standard cases. Martin 2001). Anaphors must be bound by their antecedent. if ␣ binds ␤ then ␣ c-commands ␤) and something like principle C obtains (i. We think that it is reasonable to conclude that any theory of control that must abandon either or both is very unlikely to be correct.g. Chomsky 1981). . In (90). . PRO c-commands DP1 and so should induce a principleC effect and thus be unable to bear the same semantic value as PRO. the abstract scheme in (89) takes the form in (90). b. Thus. then PRO-based accounts of backward control must violate one or both of these assumptions. . null Case (e. The standard assumption is that controller and controllee stand in binding/anaphoric relation. . the structures in (90) manifest the same control relations we find in (89). (90) a. Clearly. ]] Backward-control configurations as in (90) pose deadly problems for any account of OC in terms of PRO. ]] [DP V PRO1 [DP1 . In effect.. PRO-based analyses of backward control are inconsistent with the two central principles of binding informally described in (91).g. there is no specific tense restriction on the type of Infl the clause may have. Landau 2004). Moreover. for instance. b.. the problems become even more damaging. for PRO in (90) can appear in governed positions marked with regular case.4. Every theory of binding in the generative tradition has adhered to the principles in (91). When the interpretive properties of control are taken into account. Anaphors cannot bind their antecedent. DP1 fails to bind PRO and hence should not be a potential antecedent for PRO. if ␣ c-commands ␤ then ␤ cannot bind ␣). Worse still. this is incompatible with the fact that in the relevant languages and constructions. PRObased accounts are forced into this uncomfortable situation. (90) is problematic for the standard accounts that tie the distributional properties of PRO to government (e. In (90).5 MTC under the copy theory of movement 103 relation is a species of binding (see Chapter 2). given the existence of backward control.e. the existence of backward control creates a serious theoretical conundrum for PRO-based accounts of control.. ␣ . but backward control is in fact a phenomenon to be expected when we examine the MTC under the copy theory of movement. . In the general case. should be the abandonment of PRO-based theories.. as sketched in (94a).␣.] (93) a. 1999.2 Backward control and lower-copy pronunciation The MTC finds itself in a much more comfortable situation as regards backward control. as illustrated in (92)–(93) below.. . Indeed. Copying of ␣: K = [.. 44 See Boˇskovi´c and Nunes (2007) and references therein for an extensive review of cases of lower-copy pronunciation. for instance. the object wh-phrase does not appear to move 43 See Nunes (1995. 2004) and the collection of papers in Corver and Nunes (2007) for relevant discussion.. ]] b. . movement under the copy theory reduces to applications of copy and merge. . the existence of backward control constitutes a strong counterexample to PRO-based theories of control. discussed by Boˇskovi´c (2002). In this sense. .44 (92) K = [. ␣ .. Example (95) shows that Romanian is multiple wh-fronting (MWF). ␣ . as we will see below. (94) a. not one that is readily massaged away.43 However. b. However. ]] L = [␣ [ . when a given structure containing copies is spelled out.5.␣. . standard PRO-based accounts appear to forbid backward control as they would lead to binding violations. . Deletion in the phonological component: L = [␣ [ . the highest copy is kept and the lower ones are deleted.. as represented in (94b). . 4.104 Empirical advantages the straightforward account provided by the MTC. . .. The reasoning goes as follows. . ]] Consider the Romanian data in (95) and (96) below. an increasing body of literature has been showing that it is also possible to find cases where it is the highest copy that gets deleted. . In sum. As we have discussed above. Not only can it provide an account of backward control that is consistent with standard assumptions regarding binding.3.] ␣ Merger of α and K: L = [␣ [ . . . repeated in (98) and (99) for convenience. as illustrated in (96). [DP1 V [1 . as . the higher copy of the object wh-phrase is deleted and the lower one is pronounced instead. [1 V [DP1 . . . [DP1 V [DP1 . As for the exceptional pattern in (96a). Their difference is a matter of which copy is phonologically expressed. . where the two instances of DP1 are copies. . ∗ Ce ce precede? What what precedes? ‘What precedes what?’ (97) Deletion in the phonological component: [ce cei precede cei ] Given this general scenario. the lower copy is deleted. . .and object-control configurations in the syntactic component: a.4. . Ce precede ce? What precedes what? b. In the more common case. in order to comply with the PF constraint on adjacent homophonous elements. ∗ Cine precede ce? Who precedes what? ‘Who precedes what?’ (96) Romanian (Boˇskovi´c 2002): a. Boˇskovi´c proposes that Romanian has a low-level PF constraint against adjacent homophonous wh-phrases. The abstract syntactic configurations for forward and backward control in (88) and (89). (95) Romanian (Boˇskovi´c 2002): a. ]] b. . (98) a. . [DP V DP1 [1 . the control relation is a chain-internal relation in both forward and backward control. ]] In other words. . [DP V 1 [DP1 . as illustrated in the simplified structure in (97). it is easy to observe that the MTC is able to offer a very simple account of backward control. [DP V DP1 [DP1 . ]] (99) a.5 MTC under the copy theory of movement 105 if it is homophonous with the fronted-subject wh-phrase. ]] b. ]] (100) Subject. reduce to the configurations in (100) under the copy-theory implementation of the MTC. which rules out (96b). . ]] b. Boˇskovi´c argues that it also involves multiple wh-fronting in the syntactic component but. Cine ce precede? Who what precedes? b. . but backward control is even to be expected in a minimalist context where the copy theory obtains.ERG cow. (101) Deletion in the phonological component (forward control): a. ]] Interestingly and importantly. ]] b. The example in (103). . We turn next to illustrations of backward control and the arguments that support their existence. as shown in (102).3.3 Empirical illustrations of backward control Polinsky and Potsdam (2002) document a case of subject backward control in Tsez. As nothing theoretically prevents the grammatical option of pronouncing lower copies. not only do the problems discussed in section 4. Furthermore.5. they require that their subjects be [+animate] and [+volitional]. When the upper copy is deleted instead. 4. [DP1 V [DP1 . the matrix verbs in constructions such as (103) are thematic in that they impose selectional restrictions on their subjects.ABS feed. this possibility constitutes the null hypothesis. thus excluding sentences such 45 Both the verbs -oqa ‘begin’ and -iˇca ‘continue’ in Tsez can occur in configurations such as (103). First. yielding forward-control constructions. . [DP1 V [DP1 . illustrates a case of backward subject control. .1 evaporate in the context of the MTC in conjunction with the copy theory. .3.INF began ‘The girl began to feed the cow’ The evidence for the proposed structure in (103) − with the null subject in the matrix clause and the overt subject in the embedded one − is the following.5. . ]] (102) Deletion in the phonological component (backward control): a. [DP V DP1 [DP1 .106 Empirical advantages shown in (101). .45 (103) Tsez (Polinsky and Potsdam 2002): biˇsra] yoqsi] [1/∗ 2 [kidb¯a1 ziya girl. a language of the Caucasus. we have cases of backward control. for instance. it is tempting to conclude (and we will not resist this temptation) that the null hypothesis under the copy-theory implementation of the MTC is that backward control should be an option of UG. . [DP V DP1 [DP1 . ]] b. For instance. given that there is reasonable empirical evidence in favor of this option being empirically realized (see footnote 44). . embodying the pattern in (89a)/(99a) above. it is also clause bounded. the overt subject cannot scramble with matrix-clause elements.INF begin.INF begin.ERG cow feed began b.ABS eat. t’ont’o¯h-¯a buq bac’-xo darkness.PRES ‘The sun has been eclipsed’ (lit. #kw art’-¯a cˇ ’ikay yexur-a roq-si hammer.PRES ‘The eclipse of the sun has begun’ Second.DAT father. from (107a) we can obtain (108a) but not (108b). However.EVID ‘The girl began to suffer’ (105) Tsez (Polinsky and Potsdam 2002): a. and this is what we expect if the overt subject ‘kidb¯a’ is part of the complement clause.EVID ‘The girl began to hear the father’s story’ Third. Moreover. when one does. In -oqa constructions. Thus.PAST. the verb teqa ‘hear’ takes a dative subject regardless of whether it is embedded under -oqa: (106) Tsez (Polinsky and Potsdam 2002): kid-ber babiw-s xabar teq-a y-oq-si girl.INF begin. (107) Tsez (Polinsky and Potsdam 2002): a.ABS see.III. as illustrated in (105).DAT suffering.ERG sun.II. the case marking on the overt subject is always that which is found on subjects in the embedded clause.ABS eat.ABS begin.ERG yesterday cow feed began ‘Yesterday the girl began to feed the cow’ .INF hear.4.GEN story. the subject cannot be left behind but must scramble with the rest of the clause. ¯huł [kidb¯a ziya biˇsra] yoqsi yesterday girl.5 MTC under the copy theory of movement 107 as (104) below. For example. In particular. Furthermore.PAST. they cannot host idiomatic expressions. as illustrated in (107) below. it is possible to scramble the whole embedded clause and.ERG sun. in Tsez. With this in mind.ERG glass. ∗ t’ont’o¯h-¯a darkness.ABS break. scrambling out of an infinitive is not permitted.EVID ‘The hammer began to break the glass’ b. scrambling is rather free both to the left and the right. #kid-ber hazab bukad-a yoq-si girl. (104) Tsez (Polinsky and Potsdam 2002): a.PAST. ∗ kidb¯a ¯huł [ziya biˇsra] yoqsi girl. like standard control verbs. ‘Darkness eats the sun’) buq bac’-a boq-xo b. scrambling can be used as a diagnostic of sentence structure. 108 (108) Empirical advantages Tsez (Polinsky and Potsdam 2002): a. ¯huł yoqsi [kidb¯a ziya biˇsra] yesterday began girl.ERG cow feed kidb¯a yoqsi [ziya biˇsra] b. ∗¯huł yesterday girl.ERG began cow feed ‘Yesterday the girl began to feed the cow’ Finally, let us examine the event quantification in (109) and (110) below. (109a) is ambiguous with uyrax a¯ tiru ‘four times’ modifying the embedded verb (four feedings), as shown in (109b), or the matrix verb (four beginnings), as shown in (109c). In contrast, (110) only has the reading in which the embedded clause is modified (four feedings), which is what we would expect if the overt subject were in the embedded clause. (109) (110) Tsez (Polinsky and Potsdam 2002): a. uyrax a¯ tiru kidb¯a ziya biˇsra yoqsi fourth time girl.ERG cow feed began ‘The girl began to feed the cow for the fourth time’/‘The girl began for the fourth time to feed the cow’ b. [uyrax a¯ tiru kidb¯a ziya biˇsra] yoqsi c. uyrax a¯ tiru [kidb¯a ziya biˇsra´] yoqsi Tsez (Polinsky and Potsdam 2002): kidb¯a uyrax a¯ tiru ziya biˇsra yoqsi girl.ERG fourth time cow feed began ‘The girl began to feed the cow for the fourth time’/∗ ‘The girl began for the fourth time to feed the cow’ Polinsky and Potsdam present other types of evidence all pointing to the same conclusions, namely, that in Tsez backward construction the subject position of the matrix is obligatorily null, thematic, and obligatorily bound by the embedded overt subject. The copy-theory implementation of the MTC can account for these facts as follows. The surface form in (111a), for instance, is derived from the syntactic structure in (111b), where there are two copies of ‘kidb¯a,’ followed by deletion of the higher copy in the phonological component, as shown in (111c).46 46 Polinsky and Potsdam (2002) account for these facts in terms of the MTC by proposing that backward-control constructions involve covert movement of the embedded subject to the matrix ␪-position at LF. This also yields the desired LF structure in (111b) for the overt surface form (111a). However, if one assumes that derivations involve a single cycle, a current minimalist assumption (see Chomsky 2000), this is not a viable analysis as single-cycle theories reject LF-movement operations. Overt movement combined with lower-copy interpretation has the effect of covert LF-movement but without requiring that covert movement exist (see Polinsky and Postdam [2006] for a discussion of this possibility and comparison with backward subject raising). 4.5 MTC under the copy theory of movement (111) 109 Tsez (Polinsky and Potsdam 2002): a. [[kidb¯a ziya biˇsra] yoqsi] girl.ERG cow feed began ‘The girl began to feed the cow’ b. [kidb¯a [kidb¯a ziya biˇsra] yoqsi] c. [kidb¯a [kidb¯a ziya biˇsra] yoqsi] Polinsky and Potsdam provide independent evidence for the proposed analysis (see footnote 46), based on data such as (112) below. Example (112a) shows that Tsez reflexives are clause bound. However, a reflexive in the matrix clause of a backward-control construction can be bound by a DP in a lower clause, as shown in (112b). This makes perfect sense if the lower DP raises to the matrix clause and from there binds the reflexive. At the C–I interface, there is a copy of ‘irbahin-¯a’ in the matrix subject position licensing the matrix reflexive in (112b) and this contrasts with (112a), where there is no copy of uˇza¯ ‘boy’ in the matrix clause. (112) Tsez (Polinsky and Potsdam 2002): a. babirk nes¯a nesirk/∗ i [uˇza¯ i ␥utku roda] retin father REFL.I.DAT boy house build wanted ‘The father wanted for himself that the boy should build the house’ b. nes¯a nesiri [irbahin-¯ai halma␥or ␥utku roda] oqsi REFL.I.DAT Ibrahim.I.DAT friend house make began ‘Ibrahim began, for himself, to build a house for his friend’ Polinsky and Potsdam offer further elaborations of the proposal sketched here and discuss various technical issues related to its implementation. However, their main point is twofold. First, that it is quite unclear how the standard theories of control that involve PRO and binding would account for backward-control constructions. In fact, as noted in section 4.5.3.1, on PRObased approaches to control, backward control should be simply impossible. And second, it is easy to explain the properties of backward-control phenomena if one adopts a movement approach to control (even more so under the copy theory). Let us now consider a case of object backward control from Korean, as discussed by Monahan (2003).47 In Korean, predicates like seltukha ‘persuade,’ sikhita and kangyohata ‘force,’ chungkohata ‘advise’ and jeanhata ‘suggest’ allow for two kinds of control configurations, which are superficially distinguished by the case of the controller, as illustrated in (113). 47 For further data and discussion, see Monahan (2003) and Polinsky, Monahan, and Kwon (2007). 110 (113) Empirical advantages Korean (Monahan 2003): Chelswu-nun Yenghi-lul/ka kakey-ey ka-tolok seltukha-ess-ta Chelswu.TOP Yenghi.ACC/NOM store.LOC go.COMP persuade.PAST.DECL ‘Chelswu persuaded Yenghi to go to the store’ There are two important facts to note concerning (113). First, these sentences are close paraphrases of one another regardless of whether ‘Yenghi’ bears nominative or accusative case. Second, the case ‘Yenghi’ carries correlates with whether it resides in the matrix or the embedded clause in overt syntax. When marked nominative, it patterns like an embedded subject. When marked accusative, it patterns like a matrix object. Taken together, these pairs of facts imply that these sentences have the structure in (114), with ‘’ indicating a phonetically null position. (114) Korean (Monahan 2003): a. Chelswu-nun Yenghi-luli [i kakey-ey ka-tolok] seltukha-ess-ta Chelswu.TOP Yenghi.ACC store.LOC go.COMP persuade.PAST.DECL b. Chelswu-nun i [Yenghi-kai kakey-ey ka-tolok] seltukha-ess-ta Chelswu.TOP Yenghi.NOM store.LOC go.COMP persuade.PAST.DECL ‘Chelswu persuaded Yenghi to go to the store’ The case of interest here is (114b). Sentences like these have the interpretations of object-control constructions but the overt syntax of expect-type predicates. Thus, like standard object-control constructions and unlike ECM constructions, their matrix verbs impose animacy restrictions on their complements, as shown in (115), and do not display voice transparency, as illustrated in (116) (as indicated by the translations, [116a] and [116b] do not paraphrase one another). (115) (116) Korean (Monahan 2003): #Chelswu-nun tol-i tteleci-tolok seltukha-ess-ta Chelswu.TOP rock.NOM fall.COMP persuade.PAST.DECL #‘Chelswu persuaded the rocks to fall’ Korean (Monahan 2003): a. Chelswu-nun Yenghi-ka Swuyeng-ul intophyu ha-tolok Chelswu.TOP Yenghi.NOM Swuyeng.ACC interview do.COMP seltukha-ess-ta persuade.PAST.DECL ‘Chelswu persuaded Yenghi to interview Swuyeng’ b. Chelswu-nun Swuyeng-i Yenghi-eykey intephyu pat-tolok Chelswu.TOP Swuyeng.NOM Yenghi.DAT interview pass.COMP seltukha-ess-ta persuade.PAST.DECL ‘Chelswu persuaded Swuyeng to be interviewed by Yenghi’ 4.5 MTC under the copy theory of movement 111 On the other hand, there is convincing evidence that the nominative controller in constructions such as (114b) is located within the embedded clause. First, the matrix verbs of these constructions only permit accusative marking on their complement DP, as illustrated in (117) below, which does not have a sentential complement. Thus, the case of the controller in (114b) must be licensed within the embedded clause. (117) Korean (Monahan 2003): Chelswu-nun Yenghi-lul/∗ ka seltukha-ess-ta Chelswu.TOP Yenghi.ACC/∗ NOM persuade.PAST.DECL ‘Chelswu persuaded Yenghi’ Second, the interpretation of temporal adverbs correlates with the case marking on the control DP, as illustrated in (118). (118) Korean (Monahan 2003): nayil kakey-ey mayil Chelswu-nun Yenghi-lul/∗ ka Chelswu.TOP Yenghi.ACC/∗ NOM tomorrow store.LOC every-day ka-tolok seltukha-l ke-ya go.COMP persuade.PAST.DECL ‘Tomorrow Chelswu will persuade Yenghi to go to the store every day’ The contrast between the accusative and nominative marking on ‘Yenghi’ in (118) follows straightforwardly if the nominative DP resides in the embedded clause and the accusative DP, in the matrix clause. Under the standard assumption that adverbs can only modify the clauses they are part of, the adverb nayil ‘tomorrow’ is in the embedded clause when following the nominative DP and, therefore, it will clash with the interpretation of mayil ‘every day,’ leading to unacceptability. On the other hand, when the controller has accusative case, ‘nayil’ can reside in the matrix clause and the sentence is fine. Third, clausal scrambling must pied-pipe a nominative controller, but not its accusative counterpart, as shown in (119). Again this can be accounted for if only the nominative controller is inside the complement clause. (119) Korean (Monahan 2003): Chelswu-nun [kakey-ey ka-tolok]i Yenghi-ka/∗ -lul ti Chelswu.TOP store.LOC go.COMP Yenghi.ACC/∗ NOM seltukha-ess-ta persuade.PAST.DECL ‘Chelswu persuaded Yenghi to go to the store’ The upshot of these various considerations is the following. The thematic properties of the controller are the same irrespective of the case it carries. However, if marked accusative, it resides in the matrix clause and, if marked 112 Empirical advantages nominative, it is the subject of the embedded complement. Thus, control clauses with a nominative controller in Korean are backward-control configurations and, as argued by Monahan (2003), can be easily accounted for if OC is movement and if lower copies can be pronounced. In other words, the copy version of the MTC assigns a single syntactic structure to the two sentences of (114), for instance, as illustrated in (120). (120) [Chelswu-nun Yenghi [Yenghi kakey-ey ka-tolok] seltukha-ess-ta] Chelswu.TOP Yenghi Yenghi store.LOC go.COMP persuade.PAST.DECL If the matrix copy is phonetically realized, as illustrated in (121a) below, it appears with accusative case (cf. [114a]) and we have a standard instance of forward object control. If the embedded copy is pronounced instead, as illustrated in (121b), it appears with nominative case (cf. [114b]) and a backward object-control construction is yielded.48 48 Cormack and Smith (2004) challenge this analysis and propose an alternative based on the fact that Korean allows clausal scrambling and null objects quite freely. According to them, the derivation of a sentence like (114b), for instance, involves scrambling of the clausal complement to the left of a null object pronoun, as represented in (i). (i) a. [Chelswu-nun [Yenghi-ka1 kakey-ey ka-tolok]2 pro1 t2 seltukha-ess-ta] b. [DP [TP DP1 . . . ]2 pro1 t2 V] Although reasonable, there is good evidence to suggest that this alternative does not work for Korean (we would like to thank Sungshim Hong and Sun-Woong Kim for patient and invaluable critical discussions concerning these constructions and the data reprised below). Take the contrast between (ii) and (iiia) below, for instance. Example (ii) does not involve object control and the embedded nominative quantificational subject cannot bind the null pronoun inside the matrix adjunct, as the former cannot scope over the latter when not in the same clause. By contrast, backward-control constructions in (iiia) allow pronominal binding, which indicates that its matrix object position is occupied not by a null pro, as proposed by Cormack and Smith, but by a deleted copy of the quantified embedded subject, as represented in (iiib). (ii) ∗ pro 1 ttayloin hwuey, Chelswu-nun Yenghi-ka [motun salam1 -i hit.MOD after Chelswu.TOP Yenghi.ACC everyone.NOM ttena-tolok] seltukhassta leave.C persuaded ∗ ‘After hitting him , Chelswu persuaded Yenghi that everyone left’ 1 1 (iii) a. pro1 ttaylin hwuey, Chelswu-nun [motun salam1 -i ttena-tolok] seltukhassta hit.MOD after Chelswu.TOP everyone.NOM leave.C persuaded ‘After hitting him1 John persuaded everyone1 to leave’ b. pro1 ttaylin hwuey, Chelswu-nun motun salam1 [motun salam1 -i ttena-tolok] seltukhassta In fact, were (ib) the correct structure for Korean backward-control configurations, the relation between the DP1 and pro would be an instance of coreference and not binding, for the embedded As the MTC is the only current account of control that makes it subject does not c-command pro. They1 go there once in a while’ ∗ John-i [motun salam-i1 ttena-tolok] ku-tul1 seltukhassta John. coupled with pronunciation of the lower copy. in backward-control constructions.TOP Yenghi Yenghi.4. as they would require that. Note. . the existence of backward control constitutes. Like the Tsez data examined earlier. this is not possible with constructions analogous to (ib). It should be stressed that this is a genuine option within UG once one allows movement to ␪-positions and adopts the copy theory of movement.ACC Yenghi store.NOM store. (iv) (v) Motun saram-tul-i ku siktang pica-lul coahanta.ACC persuaded ‘John persuaded everyonei that theyi should leave’ Finally.GEN pizza. for Malagasy allows backward control but not null objects (see Potsdam 2006 for data and discussion). the movement analysis of control can treat the nominative version of the Korean constructions above analogously to the Tsez cases via a process of overt movement of the embedded subject to the internal ␪-position of the matrix verb. something that is strongly disallowed in all other cases of binding. as shown in (v).3. [Chelswu-nun Yenghi-lul [Yenghi kakey-ey ka-tolok] Chelswu.1).COMP seltukha-ess-ta] persuade. Potsdam (2006) shows that Cormack and Smith’s (2004) proposal does not constitute a general crosslinguistic alternative either.C they.NOM everyone.LOC go. however. that.PRS. On the assumption that the null pronoun should function like the overt one in this case. if the embedded subject in (ib) were quantificational.TOP Yenghi. eschewing traces in favor of copies. PRO) binds the DP that licenses it.5.NOM the restaurant.PL.5 MTC under the copy theory of movement 113 (121) a.NOM leave. then pro would be an E-type pronoun.e. in our view.. [Chelswu-nun Yenghi [Yenghi-ka kakey-ey ka-tolok] Chelswu. a very powerful reason for rejecting PRO-based accounts of control (see section 4.LOC go. as illustrated in (iv) below.DECL In sum.DECL Ku-tul-un kakkum keki-e ka-n-ta they once in while there go ‘All the people1 like the pizza restaurant. the unacceptability of (v) again indicates that a null pronoun is not actually available in backward-control cases such as (114b).PAST.COMP seltukha-ess-ta] persuade.DECL b. although we can have cross-sentential relations between an overt E-type pronoun and a preceding quantificational expression. Thus. All person. the Korean data discussed above pose a very serious problem for PRO-based accounts of control. Thus.ACC like. an anaphor (i.PAST. but this is already required to handle cases of multiple reflexivization in the envisaged framework. being at odds with standard binding principles. is compatible with backward-control data.5. wherein the related DPs are generated together and then one moves away to check another ␪-role. only a PRO-like element. This is of great significance. but is in fact ubiquitous (see the references in footnote 42 and Polinsky and Potsdam 2002 for a brief review). and regular control results when the doubled DP moves to check the controller ␪-role. Furthermore. However. The single chain is derived from a process wherein the PRO cliticizes to the controller (or a region very near it) resulting in the unification of the two disparate chains (i. This proposal resembles the MTC in allowing for a single chain at LF comprising two ␪-roles. as proposed by Hornstein (2001). but is instead expected under a copy-theory version of the MTC. the putative PRO appears in the matrix object position. the putative PRO under a PRO-based account c-commands the controller. . Similar problems afflict approaches to control like that proposed in Manzini and Roussou (2000) on the assumption that feature movement cannot be in a downward direction. We believe that the moral of these constructions is clear: if backward control exists.49 4. it is no longer descriptively accurate to maintain that PRO only occurs in the subject position of untensed clauses for..4 Wrapping up Backward control is not restricted to the languages surveyed. As discussed in section 4.3.3. Backward control would result when the “PRO” moves to check the controller ␪-role. in backward objectcontrol constructions. it appears to us that PRO-based accounts are doomed to inadequacy. 49 It is worth pointing out that backward control also argues against “mixed” theories of control such as the one in Martin (1996).e. Interestingly.114 Empirical advantages possible to dispense with OC PRO. One last point: this suggests that local reflexivization and OC should be treated via the same mechanisms. in both backward subject-control constructions such as the ones found in Tsez and in backward object-control constructions such as the ones found in Korean. as backward control poses fatal problems for PRO-based accounts of control.1. This approach requires allowing rather complex “doubling” structures as there is no upper bound on the number of possible “PROs” in a control configuration.5. then PRO-based analyses of control are incorrect. it differs from the MTC in having a PRO-like element that merges into the controlled ␪-position. a doubling account of control like the analysis of reflexivization in Zwart (2002). the existence of backward control points ineluctably to the conclusion that some version of the MTC is correct. the two chains “collapse” into one). As we believe that there is considerable evidence that backward control obtains. it cannot rely on the copy theory to provide the right phonological options as there is no copy of the controlled expression down below. This proposal has problems with backward-control cases for it predicts that the phonetic gap will always appear in the “lower” position. Unlike the MTC. This said. made evident in the case of backward control. we tacitly assumed that. the properties of backward control are easy to explain given the copy-theory version of MTC on the assumption that control complements are clausal. 4. However. be they high or low. This is interesting for there is a long tradition within generative grammar suggesting that control complements are not actually clausal. and Nunes (2007. Chierchia (1984).50 This is plausible when the (null) controllee is in the embedded clause as there is nothing phonetically evident in such cases. that case may be the responsible factor.5. . any version of the 50 See e.2 for further discussion). a full account of backward control must also specify what licenses the phonetic interpretation of copies. Monahan (2003). This is consistent with the idea that variation is limited to the PF/A–P side of the grammar with the LF/C–I side being uniform across languages. backward control is both consistent with the MTC and expected when the MTC is coupled with the copy theory of movement. Hornstein. then we can see that control complements are clausal. in languages that allow backward control. holds for control quite generally (see section 7. and Boeckx. the details of a full case proposal are still to be worked out to round out a full account of backward constructions in the context of the MTC.4). It is extremely unlikely (at least it would require a lot of evidence) that backward control involves clausal complements while forward control involves VPs or predicates of some kind. 2008). [120]). for their properties are identical. copy-control constructions.. Theoretically. and the discussion in section 7. However. given a series of copies left by movement.4 Phonetic realization of multiple copies and copy control Let us now consider another type of control construction which is predicted to exist by the copy version of the MTC. Descriptively speaking.5 MTC under the copy theory of movement 115 It is important to observe that backward control also raises challenges for the MTC. Bresnan (1982).g. In the discussion above. As we have noted. it appears that. its existence implies that control complements are clausal. as far as some details of technical implementation go. We have tacitly assumed. The best conclusion then is that control complements are indeed clausal and that this fact. However.4. only one gets pronounced at PF. as we saw in the case of Korean. namely. following Polinsky and Potsdam (2002). Backward control has another useful consequence. for example.2 below. as appears to be the case. After all. copying a DP may leave its case feature stranded (cf. Like finite control (see section 4. if backward control has all of the properties of standard control. we can “see” the controller sitting in the embedded clause. As for the choice of the link to survive deletion. Thus. but a chain is a discontinuous object and cannot be mapped onto a single position at PF. 1999. given that these copies are non-distinct. [Johni [was [arrested Johni ]]] b. with two copies of p. Nunes (1999. derivations could not converge as they could not be linearized and so would not receive PF interpretations. Nunes (1995. for instance. by Kayne’s [1994] LCA). there do exist constructions where a lower copy is pronounced instead. Again. 2004) argues that. K can be phonetically realized as well. [Johni [T hoped [Johni to see Mary]]] To put this somewhat differently. under certain conditions. the phonological component demands that syntactic structure be converted to linear precedence (say.’ (122) a.3 that. they should in principle be able to surface overtly. this actually happens. once the syntactic structure in (123a) below. The null hypothesis is that all copies should in principle be on an equal footing with respect to phonetic realization. The same applies to (122b): ‘John’ must precede and be preceded by ‘hoped. Suppose for instance that. yielding instances of backward control if the relevant movement is movement to a thematic position. the two copies of ‘John’ (annotated by the superscripted indices) induce contradictory linearization requirements. is spelled out. The gist of his proposal is that copies count as “the same” for purposes of linearization in the phonological component because they are non-distinct elements (technically. the morphological component fuses (in the sense . they relate to the same occurrences of lexical items of the numeration) and this creates problems. 2004) proposes that the ban on phonetic realization of multiple copies has to do with linearization considerations. in order for a structure containing a chain to be linearized. the reason that all copies but one are phonetically null is that. if they were not deleted.’ which must in turn precede the second copy. In each of the structures in (122a) and (122b) below. this fits within the null hypothesis regarding the copy theory: if a constituent K is a replica of K’ and K’ can be phonetically realized.’ which is a contradictory requirement. Here is the reasoning.5. Thus.116 Empirical advantages copy theory must account for why realization of multiple copies is forbidden. all of its links but one must be deleted. although the common situation is for the highest copy to be pronounced. However. This line of thinking predicts that if copies do not interfere with linearization. we have already seen in section 4. this amounts to saying that ‘John’ must precede and be preceded by ‘was. The first copy of ‘John’ in (122a) must precede ‘was. Spelled-out structure: M pi L r k pi m b. Koopman (1984) shows that the two verbal occurrences in (124) cannot be separated by islands. it can be indirectly linearized in (123b) in virtue of being an integral part of #mp#: given that the upper copy of p asymmetrically c-commands r and that r asymmetrically c-commands #mp#. as illustrated in (124) below. 2004. with no internal structure accessible to further morphological or syntactic computations. and references therein. Fusion in the morphological component: M pi L r k #mpi # The content of #mp# in (123b) cannot be directly linearized with respect to r or the upper copy of p because it is an inaccessible part of #mp#. which indicates that they should 51 For further illustration. once the lower copy of p in (123b) becomes invisible for standard linearization computations. we should obtain the linear order p>r>#mp#. yielding the atomic blended terminal #mp# (or #pm#. surface with two copies of p at PF. In other words. but must. for that matter). Thus.’ But. the structure in (123b) not only can. From an LCA perspective. crucially.5 MTC under the copy theory of movement 117 of Halle and Marantz 1993) the terminals m and p. the blended material within #mp# is not accessible to c-command computations. the material inside #mp# gets linearized in a way analogous to how the phoneme /l/ is indirectly linearized in ‘John loves Mary’ due to its being part of the lexical item ‘loves. .4. see Nunes (1999. as sketched in (123b). in press). However. the collection of papers in Corver and Nunes (2007). Boˇskovi´c and Nunes (2007).51 Consider verb clefting in Vata. (123) a. An example should make this idea clear. the linearization problems caused by the presence of multiple copies discussed as regards (123) cease to exist. for instance. This makes sense if these particles render the verb . that the fronted verb in these focus constructions must be morphologically unencumbered. then it should not be possible to linearize the structure containing them in accordance with the LCA. Since the pronoun a` ‘we. asymmetrically c-commands and is asymmetrically c-commanded by (a copy of) the verb li ‘eat. is that. . they certainly should not be able to undergo the morphological fusion with Foc◦ depicted in (125a) and should not be allowed in predicate clefting constructions. . . he analyzes verb clefting in Vata as involving verb movement to a focus head. Deletion of copies: [FocP #[Foc0 Vi [Foc0 Foc◦ ]]# [TP . [125b]) and the structure is linearized as in (124). Vi . if these occurrences are to be treated as copies under the copy theory.118 Empirical advantages be related by movement. as illustrated in (126) below. ]]] b.’ (124) Vata (Koopman 1984): li a` li-da zu´e sak´a eat we eat. . More precisely. in particular. Fusion: [FocP #[Foc0 Vi [Foc0 Foc◦ ]]# [TP .’ for example. The second piece of evidence is provided by the fact. . . If these verbs cannot participate in any morphological process. . The problem. . the LCA only “sees” the lower two after the highest copy gets fused with Foc◦ . .PAST yesterday rice ‘We ATE rice yesterday’ Nunes (2004) proposes that this possibility does not in fact arise because the highest copy of the clefted verb gets morphologically fused and thereby evades the purview of the LCA. Vi . Of the three verbal copies in (125a). followed by fusion in the morphological component between the moved verb and the focus head. [T0 Vi [T0 T◦ ]] [VP . . as discussed above with respect to (123). The first one relates to Koopman’s (1984: 158) observation that the restricted set of verbs that cannot undergo clefting in Vata has in common the property that they cannot serve as input for morphological processes that apply to other verbs. with two copies of the verb phonetically realized. however. as represented in (125a) below. [T0 Vi [T0 T◦ ]] [VP . (125) a. . none of the tense or negative particles that occur with the verb in Infl may appear with the fronted verb. ]]] Nunes (2004) presents two pieces of evidence in favor of this account of verb clefting in Vata. also observed by Koopman. The lowest copy is then deleted (cf. .’ the LCA should induce the contradictory requirement that ‘`a’ precede and be preceded by ‘li. it is phrasal).5 MTC under the copy theory of movement 119 morphologically too complex. Now comes the punch line. If multiple copies may be phonetically realized when fusion allows the linearization problem to be circumvented and if control is movement.eat. the less likely it is that it will undergo fusion and become part of a terminal. (126) Vata (Koopman 1984): a. (∗ na`-) le wa n´a`-le-ka NEG eat they NEG. the duplication of focused material only affects verbs and many languages only allow multiple copies of wh-elements. It also predicts that such constructions should display hallmarks of fusion such as sensitivity to morphological information (fusion may or may not take place depending on the morphological properties of the copies involved) and morphological complexity (the more morphologically complex a given copy is. but only the ones that satisfy the morphological requirements of one another.e.4. As a rule.TP they eat. thereby preventing the verb from undergoing fusion with the focus head. Hornstein. This seems to be what is going on in (126). In Vata. After all. Boeckx. the copy version of the MTC predicts that control constructions with more than one copy phonetically realized should exist. the more morphologically complex a given element is. li(∗ -wa) w`a li-wa zu´e eat. 2008) show that these predictions are indeed fulfilled. given that multiple copies are only allowed when some copies get morphologically reanalyzed as being part of a fused terminal. Of course..FT ‘They will not EAT’ b. This may be interpreted as a reflex of the morphological (categorial) restrictions a given head may impose on the copy with which it may fuse. if a given copy is syntactically complex (i. The first kind of relevant information regards the feature composition of the elements that are to be fused. it is also morphologically complex and not a good candidate to undergo morphological fusion. with the addition of Infl particles to the fronted verb. The second kind of information concerns morphological complexity. the addition of specific morphemes (which may vary from language to language) may make the resulting element morphologically “too heavy” to become reanalyzed as part of a word. Thus.TP yesterday ‘They ATE yesterday’ What is relevant for our purposes here is that these restrictions indicate that the realization of multiple copies should indeed be very sensitive to morphological information. and Nunes (2007. the more unlikely it is to undergo fusion). for instance. not any two terminals can get fused. . as shown in (129). b. R-c`aa` a’z Gye’eihlly g-ahcn`ee Gye’eihlly Lia Paamm HAB.want Mike IRR. Hmong (Quinn 2004) Pov xav kom Pov noj mov Pov want/think so-that Pov eat rice ‘Pov wants to eat’ Maiv ntxias Pov kom Pov rov qab noj Maiv persuaded Pov so-that Pov return back eat ‘Maiv persuaded Pov to eat’ Pov pais tom qab Pov hais goodbye rau tus Pov left after back Pov said good-bye to CLF ‘Pov left after saying good-bye to the woman’ mov rice pojniam woman . B-`ıi’lly-ga’ Gye’eihlly zi’cyg`aa’ nih cay-uhny Gye’eihlly z`ee` iny PERF.persuade priest Mike IRR. B-qu`ıi’lly bxuuhahz Gye’eihlly ch-iia Gye’eihlly scweel PERF.want Mike IRR.go Mike school ‘The priest persuaded Mike to go to school’ c. They also trigger a sloppy reading under ellipsis.do Mike work ‘Mike sang while he worked’ Each of the sentences in (127) shows a bound copy in the embedded-subject position. and the bound copy displays complementarity with a coreferential pronoun.120 Empirical advantages Consider the data in (127) below from San Lucas Quiavin´ı Zapotec. discussed by Lee (2003). as illustrated in (i) below (see Boeckx. c.sing.do Mike work PERF. (128) San Lucas Quiavin´ı Zapotec (Lee 2003): a. Hornstein. as shown in (128).52 (127) San Lucas Quiavin´ı Zapotec (Lee 2003): a. and Nunes 2007. the similarities of these constructions with standard control constructions go beyond translation.eat Mike grasshopper ‘Mike wants to eat grasshopper’ b.help Mike FEM Pam z¨e’cy cahgza’ Li’eb likewise Felipe ‘Mike wants to help Pam. 2008 for relevant discussion). Zi’cyg`aa’ nih cay-uhny Gye’eihlly z`ee` iny b-`ıi’lly-ga’ Gye’eihlly While that PROG.also Mike z¨e’cy cahgza’ Li’eb likewise Felipe ‘While Mikei was working. and so does Felipe (want to help Pam/∗ want Mike to help Pam)’ b.sing-also Mike while that PROG. hei sang. Interestingly. and so did Felipek (sing while hek worked)’ 52 Hmong also seems to allow structures analogous to (131). (i) a. R-c`aa` a’z Gye’eihlly g-auh Gye’eihlly bxaady HAB. personal communication): a.. they propose that these constructions involve morphological fusion of the controllee copy with the null “self” morpheme available in this language.PROX FEM Pam ‘Mikei wants himk/∗ i to help Pam’ b. 2008) for details and further discussion.5.POSS bx:`aady grasshopper ‘Mike told his brother to eat grasshoppers’ Let us reexamine the adjunct copy-control case in (127c). ∗ R-e’ihpy Gye’eihlly behts-ni’ HAB. Hmong (Mortensen 2003): Pov yeej qhuas Pov Pao always praise Pao ‘Pao always praises himself’ 54 See Hornstein.shave Mike Mike ‘Mike shaved himself’ b. movement to thematic positions.5 MTC under the copy theory of movement (129) 121 San Lucas Quiavin´ı Zapotec (Felicia Lee. and Nunes (2007. ∗ Yra’ta’ zhy`aa’p r-c`aa` a’z g-ahcn`ee’ yra’ta’ zhy`aa’p Lia Paamm Every girl HAB. whose links contain an anaphoric possessor. 2008) argue that the data in (127)–(128) are indeed cases of control.e. fusion will be blocked and phonetic realization of more than one copy leads to an ungrammatical result. (i) a. Boeckx. Mikek sang’ Hornstein.3SG. .sing. and Nunes (2007.REFL. i. 2008) argue that fusion with this null “self” morpheme is also what underlies the existence of copy-reflexive constructions in San Luca Quiavin´ı Zapotec and Hmong (see footnote 52) such as the ones illustrated in (i). That this prediction is correct is illustrated by the copy-control constructions in (130a).REFL. As discussed in section 4. if a control chain involves morphologically encumbered copies.want Mike IRR.also Mike ‘While hei/∗ k worked. and Nunes (2007.eat brother. Boeckx.54 (130) San Lucas Quiavin´ı Zapotec (Lee 2003): a. The fact that 53 Hornstein.help.3SG.help every girl FEM Pam ‘Every girl wants to help Pam’ g-a’uh behts-ni’ b.POSS IRR.do. Boeckx. R-caaa’z Gye’eihlly g-ahcn`ee-¨eng Lia Paamm HAB. and in (130b).PROX work PERF. San Luca Quiavin´ı Zapotec (Lee 2003): B-gwa Gye’eihlly Gye’eihlly PERF. More specifically.1. which involves a quantifier phrase. adjunct control involves sideward movement.53 As we should expect given the discussion above. with both the controller and the controllee copies phonetically realized.tell Mike brother. Zi’cyg`aa’ nih cay-uhny-¨eng z`ee` iny b-`ıi’lly-ga’ Gye’eihlly While that PROG.want IRR.4. GEN anger raise.122 Empirical advantages sideward movement may also lead to phonetic realization of multiple copies thus further stresses the point that sideward movement is nothing more than one of the instantiations of copy plus merge. we should in principle expect such movement to also yield constructions with multiple copies. 2009) shows that adjunct copy-control constructions such as (131) and (132) below (CNP stands for conjunctive participle particle) display all the traditional diagnostics of obligatory control and argues that they should also be analyzed in terms of sideward movement and phonetic realization of multiple copies.CNP Kumar.NOM movie watch. the subject “does not exceed one or two words.NOM popcorn ate. if movement into ␪-positions is possible. and against PRO-based accounts of control.NOM movie watch.NOM my house destroyed. it comes as no surprise that multiple copies are only possible if. which indicates that the relevant head that triggers fusion in these languages is within the adjunct clause. provided that we have evidence that one of the copies is morphologically reanalyzed. Interestingly.3. powerful) argument for the copy version of the MTC. (131) Telugu (Haddad 2007): [[Kumar sinima cuus-tuu] [Kumar popkorn tinnaa-Du]] Kumar. they ate popcorn’ To summarize. in Haddad’s (2007: 87) words. (133) ∗ Telugu (Haddad 2007): [[Kumar maryu Sarita sinim cuu-tuu] [Kumar maryu Kumar. Haddad (2007.SG. Kumar ate popcorn’ (132) Assamese (Haddad 2007): [[Ram-Or khong uth-i] [Ram-e mor ghorto bhangil-e]] Ram. The existence of copy-control constructions therefore provides another (in our view.NOM and Sarita.CNP Ram.” as illustrated by the ungrammaticality of (133).M ‘While watching a movie.NOM popcorn ate ‘While Kumar and Sarita were watching a movie. After all.NOM and Sarita popkorn tinna-ru]] Sarita. there are languages which only allow adjunct copy control. what can be more convincing for the copy version of the MTC than the phonetic realization of both controller and controllee as copies? . Ram destroyed my house’ Given the role of morphological fusion in making the phonetic realization of multiple copies possible. In his detailed study on control structures in Telugu and Assamese.CNP Kumar.3 ‘Having got angry. Also common are cases that appear to contradict the theory by having a status at odds with what it predicts. .. Or. it leads to the conclusion that no PRO-based approach to control can be right. the theory can be saved by judicious twiddling. given the copy theory of movement. backward and copy-control constructions prove fatal for accounts of control that are PRO-based. accounts that assume a primitive expression like PRO which at once bears a ␪-role and has the controller as antecedent. the larger background assumptions are saved and the more specific proposals sacrificed.6 123 Conclusion Counter-examples come in various flavors.56 In such cases defusing the counter-example requires reneging on well-established background principles.55 Counter-examples are more serious when more Janus-faced. at least in the linguistics literature.e. Why this brief divagation into the epistemology of counter-examples? Because we believe that part of the data discussed in the previous sections constitute arguments of the strongest type for the debate on obligatory control. backward and copy control are not only fully consistent with the MTC. In each case. though it seems. These are surprisingly rare. the “exceptions prove the rule” in the originally intended sense that the counter-examples for one proposal “prove” the second by at once disconfirming the former and confirming the latter. Another plausible example is the island and subjacency arguments against Chomsky’s (1982) approach to parasitic gaps in terms of a functional interpretation of empty categories (see Kayne 1984). In such circumstances. but are in fact expected. that is. Methodological kudos does (and should) accrue to theories that are able to clean up the ad hoc messes of their competitors. when they are perched between two competing theories and smile towards one but frown towards the other. More significant yet are counter-examples that seem to cut to the heart of a theory. 56 Examples include Chomsky’s (1955. more precisely. intuitively. Though often tinged with adhockery. 1957) famous arguments against finite-state grammars and simple phrase-structure grammar as models for linguistic competence. Such problematic data are deeply telling for they suggest either that the favored theory is clearly wrong or that the larger theory in which it is embedded is in need of extensive revision. that it should as the data are of a piece with those that it does.6 Conclusion 4. this kind of maneuver is both common and well understood. In particular. Moreover.4. This in turn leads directly to the conclusion that some version of 55 Nor is it always inappropriate. If a theory is basically correct then an ad hoc addition can be interpreted as claiming that the putative counter-example is not actually a real one. appending a codicil that adds or excludes the relevant problematic data point. Typically. A common variety consists of cases that the theory does not cover. i. given that the only current alternative to PRO-based accounts of control is the MTC. As discussed in Chapter 3. However. Finally. Before closing this chapter. By exploring the logical consequences of these pillars of the minimalist program.57 whatever problems of technical implementation the MTC currently faces cannot be grounds for arguing in favor of PRO-based accounts. 57 Another possible approach would revive some version of the earliest equi-accounts. we believe that such an approach only appears to be different from the MTC. but also substantially broadens the empirical coverage of previous analyses. an adequate equi-account will have to incorporate the MTC if it is to overcome the original difficulty for equi-approaches posed by the contrast between (ia) and (ib) below (see section 2.3). If this is correct. the MTC not only provides a conceptually well-grounded analysis of control. we have shown that the copy theory of movement makes room for the existence of cases of lower-copy and multiplecopy pronunciation and that the copy implementation of the MTC correctly predicts that control structures may also exercise these options. In sum. then equi-accounts are just subspecies of the MTC. an additional argument in its favor. Everyone wants to leave Everyone wants everyone to leave . In fact. a feature of the MTC that we believe to be of more than passing interest. the MTC follows naturally under central features of the minimalist program: the elimination of DS and the copy theory of movement. Without DS. It is in this sense that the MTC is a quintessentially minimalist account of control. the nice fit between the MTC and these minimalist ideas constitutes. Baltin and Barrett (2002) make such a proposal. More particularly. b. we would also like to call attention to the fact that the MTC and the minimalist program snugly fit together conceptually. we believe. the MTC fits well with the elimination of DS. Given that (ia) and (ib) do not mean the same. (i) a. the requirement that all ␪-role assignment be prior to all movement operations is set aside. But this is just to endorse the MTC. distinguishing between the putative deleted instance of ‘everyone’ in the embedded clause of (ia) and the non-deleted instance in (ib) requires assuming that only (ib) results from two selections of ‘everyone’ in the derivation.124 Empirical advantages the MTC must be correct. and this opens the possibility that ␪-roles may also be assigned under movement. In this chapter. Hornstein. as the controllee appears to be distinct from the controller. However. under very reasonable assumptions (most of which are standard in GB and/or minimalism). 2004. we will show that. Boeckx and Hornstein (2003. but control is.2 1 See Hornstein (2001.2. As the reader will see. 2009a. 5. 125 . which at first sight seem to require the postulation of PRO. 2006b.4 discusses cases in which the different morphological patterns associated with raising and control have been interpreted as showing that control constructions involve a case-marked PRO. Sections 5.3 examine cases where raising is not allowed.6 we discuss partial and split control. 2007). the apparent problems have more to do with a proper characterization of the relevant constructions than with the MTC itself. in sections 5. obligatory control. and Visser’s generalization Landau (2003) has claimed that the MTC overgenerates by incorrectly ruling in passives of subject-control predicates (see also Kiss 2005 and section 5.2 and 5. For the sake of brevity. which appear to be at odds with the minimality assumptions underlying the MTC. 2010). Finally.1 Introduction In this chapter we discuss a fair sample of the empirical challenges that have been perceived by many as providing lethal counter-arguments to the MTC.5 Empirical challenges and solutions 5. 2003). 2006a.5. and Boeckx. and Nunes (in press) for discussion of possible solutions to other empirical objections that have been raised.1 We will start with apparent problems that stem from contrasts between raising and control constructions. the apparently problematic data can receive a streamlined analysis under the MTC and in many cases turn out to actually provide compelling evidence in its favor. we examine control constructions involving promise-type verbs and control-shift phenomena. In sections 5. Nunes (2007. Sections 5. It is beyond the tenets of this chapter to offer an independent full-fledged account of each of them.2 Passives. here we will not review details of our earlier attempts to address the issues to be discussed in this chapter. 2 2 See Nunes (2007. if the matrix verb of a finite-control structure such as (6) is passivized. . contrary to fact. it contends that raising and obligatory-control constructions are derived by the same operation – A-movement. let us enlarge our data set.4 that Brazilian Portuguese allows both finite control and hyper-raising. control and raising should pattern alike when this ␪-position is eliminated. this is what happens if the subject-control predicate is passivized. b. ∗ ∗ John was tried to kiss Mary [Johni was tried [ti to kiss Mary]] At the risk of beating a dead horse. in order for Landau’s argument to be valid. But before examining these derivations in detail and providing an account of (3a). as illustrated in (4) and (5). b. 2010) for discussion. b. John was persuaded to kiss Mary [Johni was persuaded [ti to kiss Mary]] (5) a. So. we would like to reiterate a point made in the previous chapters: the MTC is not a raising theory of control. Thus. The argument runs as follows. hyper-raising is blocked. it should be first demonstrated that the licensing conditions for the relevant A-movement to apply in (1b). We show in the next section that. Recall from section 4. Rather. b. Relevant to the present discussion is the fact that. the observation that subject-control verbs do not behave like raising predicates when passivized goes beyond structures involving non-finite complements. movement of the embedded subject first targets a ␪-position. given that object control and ECM verbs do not shy away from passivization. The unacceptability of (3a) is generally taken to fall under Visser’s generalization (see Bresnan 1982). sentences such as (3a) should be licit under the derivation sketched in (3b). John seems to love Mary [Johni seems [ti to love Mary]] (3) a. as illustrated in (6) and (7) below. b. (1) a. (4) a.126 Empirical challenges and solutions below). and (3b) are the same. Arguably. in the former. 2008a. (2b). If the only difference between the derivation of subject control and raising constructions is that. John was expected to kiss Mary [Johni was expected [ti to kiss Mary]] Interestingly. as shown in (8). John tried to kiss Mary [Johni tried [ti to kiss Mary]] (2) a. as illustrated in (1) and (2) below. when the three derivations are carefully inspected. this is actually not the case. 3. the contrast between raising and subject control seen in (2) and (3) also arises when A-movement out of finite clauses is involved.2 Passives. (8) Brazilian Portuguese: Os meninos foram ditos que n˜ao fizeram a tarefa The boys were said.1 Relativizing A-movement Leaving the discussion of finite control and hyper-raising construction to section 5. (9) a. John tried to kiss Mary [vP v␪ [VP tried [CP C [TP John to kiss Mary]]]] (10) a.2. b. (7) Brazilian Portuguese: Os meninos parecem que n˜ao fizeram a tarefa The boys seem. b.5. as sketched in (9)–(13) (irrelevant details omitted). OC. b.PL the homework ‘The boys said that they didn’t do their homework’ [[Os meninos]i disseram [que ti n˜ao fizeram a tarefa]] a. ∗ [[Os meninos]i foram ditos [que ti n˜ao fizeram a tarefa]] a. ∗ Let us then consider how the MTC can provide an account of the data in (1)–(8) by paying close attention to the licensing conditions involved in each case where the embedded subject undergoes A-movement to the matrix clause. 5.PL that not did. b. let us start by considering the derivational step preceding the movement of the embedded subject to the matrix clause in the derivations of (1)–(5).MASC.2. and Visser’s generalization 127 In other words. So. b. whatever the ultimate analysis for the unacceptability of (3a) is.PL the homework ‘It was said that the boys didn’t do their homework’ b. John seems to love Mary [TP T␾ [VP seems [TP John to love Mary]]] ∗ John was tried to kiss Mary [PpleP -en␾ [VP tried [CP C [TP John to kiss Mary]]]] (12) a.PL the homework ‘It seems that the boys didn’t do their homework’ [[Os meninos]i parecem [que ti n˜ao fizeram a tarefa]] a. it should in principle apply to (8a) as well. b.PL that not did. b. (11) a. John was persuaded to kiss Mary [VP persuaded␪ [CP C [TP John to kiss Mary]]] (13) a. (6) Brazilian Portuguese: Os meninos disseram que n˜ao fizeram a tarefa The boys said. John was expected to kiss Mary [PpleP -en␾ [VP expected [TP John to kiss Mary]]] .PL that not did. (10b) and (13b) involve no CP layer in the embedded clause. if movement of ‘John’ is to be anchored on ␾-agreement. and (11b).3 Let us first examine the instances of A-movement triggered by ␾-agreement. 2010) argues that this difference is what underlies the contrast between (10b) and (13b). as exemplified in (i). as respectively shown in (15) and (16). Movement in (10b). [10b] and [13b]) and the one where it is not (cf. it cannot have its case feature checked/valued (the embedded C/T is not a case checker/assigner) and the derivation crashes. as depicted in (14) below. Assuming with Chomsky (2008) that clausal ␾-features are actually hosted by C (they are associated with T only by inheritance from C). Once ‘John’ is prevented from undergoing A-movement. (14) [PpleP -en␾ [VP tried [CP C␾ [TP John␾ to kiss Mary]]]] ∗ By contrast. 2001) that case checking is contingent on ␾-agreement and that passive participial heads are associated with (defective) ␾-features (and case) regardless of whether or not these features are overtly realized. One salient difference between the structures where such movement is allowed (cf. Nunes (2007. movement of ‘John’ couched on ␾-agreement is licit. [11b]) has to do with the categorial nature of the embedded clause. (i) To kiss Mary was tried (by John) repeatedly throughout the evening . the intervening ␾-features of C induce a minimality violation. 4 Notice that passivization of the whole clause may yield an acceptable result.4 More specifically. (15) [TP Johni T␾ [VP seems [TP ti to love Mary]]] ↑ OK (16) [PpleP Johni -en␾ [VP expected [TP ti to kiss Mary]]] ↑ OK 3 We assume with Chomsky (2000. whereas control verbs select for CP. on the other. Nunes proposes that the agreement relation between ‘-en’ and ‘John’ in (11b) is blocked due to the intervention of C. for there is no intervening ␾-feature bearer. and (13b). on the other hand. Under standard assumptions. the trigger for the movement of ‘John’ is ␪-related: the matrix light verb in (9b) and the matrix main verb in (12b) need to assign their remaining ␪-role. on the one hand. (11b). raising and ECM verbs select for TP.128 Empirical challenges and solutions In (9b) and (12b). is motivated by agreement in ␾-features with a finite T or a passive participial head. Thus. ␾-agreement with the passive participial head can license A-movement of the embedded object in (19). in long passives. .NOM SELF to-imagine tried was ‘since they tried to recall the image of the fish’ Assuming that Wurmbrand’s analysis of long passives is correct (see section 5. movement for ␪-reasons in (20a) provides an escape hatch for ‘John’ to enter into ␾-agreement relations later in the derivation. given this relativized minimality approach to Amovement.2. there is no appropriate antecedent to license the anaphor. PpleP] without any problems. the matrix control verb is a restructuring verb that takes VP for a complement. Wurmbrand (2001) uses contrasts such as the one in (18) between an impersonal and a long passive to argue that.ACC to-imagine ‘They tried to imagine what the fish would look like with stripes’ a. not ␾-agreement. Once these complements involve just the lower shell of the verbal skeleton. as there are no intervening elements bearing ␾-features. when the passive participial head is introduced. as shown in (20b). OC. Hence.2 below for a discussion of impersonal passives). (17) German (Wurmbrand 2001): dass die Traktoren zu reparieren versucht wurden that the tractors. the ␾-features of C do not block the movement of ‘John. hence the unacceptability of (18b).5.NOM to repair tried were ‘that they tried to repair the tractors’ (18) German (Wurmbrand 2001): Es wurde versucht [PROi sichi den Fisch mit Streifen vorzustellen] It was tried SELF the fish with stripes. for there is no intervening element that hosts ␾-features. ∗ weil {sich} der Fisch {sich} vorzustellen versucht wurde since SELF the fish. (19) [PpleP -en␾ [VP tried [VP repair [the tractor]]]] OK As Nunes points out. the relevant derivational step underlying (17) is as sketched in (19) (with English words for convenience).2 Passives. b. movement of the embedded subject is motivated by ␪-considerations. ‘John’ has already moved out of the CP and can therefore enter into an agreement with -en and move to the [Spec. [12b]).’ as represented in (20a) below. the acceptability of (12a) becomes very illuminating. and Visser’s generalization 129 Similar considerations apply to the derivation of “long passives” in German such as (17) below. Later on. Although there is a CP layer in the complement clause (cf. In other words. Under the first one. Alternatively. one wonders why it does not act as a proper intervener for the movement of ‘John.is possible. but not in raising structures. 2010) analysis makes it clear that the lack of passivization of subject-control verbs is not at all a problem for the MTC. C does not qualify as a proper intervener because CP is not an appropriate element to carry the external ␪-role assigned by the matrix light verb in (9b) or the experiencer ␪-role assigned by ‘persuaded’ in (12b).’ After all. If so. Nunes’s (2007. for in general only animate DPs can be controllers. C gets ␪-marked as it is the head of the complement of the matrix lexical verb. Example (21a) involves an animate subject and the infinitival complementizer me. whereas (21b) has an inanimate subject and me. that movement cannot be too local. C does not count as a proper intervener for movement of the embedded subject. b.’ which is ambiguous between a control and a raising verb. a given element cannot resort to movement to check two ␪-roles inside the same thematic domain for reasons of anti-locality.130 Empirical challenges and solutions (20) a. (21) a. Landau (2003: 488) claims that the MTC does not seem able to account for the crosslinguistic generalization that infinitival complementizers are found in control structures. once CP is not an eligible candidate to receive the unassigned ␪-role of (9b) and (12b). Given the blocking role played by C with respect to ␾-agreement. following Abels (2003) and Grohmann (2003). [VP Johni persuaded␪ [CP C␾ [TP ti to kiss Mary]]] ↑ OK [PpleP –en␾ [VP Johni persuaded␪ [CP C [TP ti to kiss Mary]]]] OK Let us now examine the instances of A-movement for ␪-purposes in (9b) and (12b) more closely. given the connection between infinitival complementizers and control. Quite the opposite! This account of Visser’s generalization in fact provides an answer for another of Landau’s criticisms. In sum. cease. This empirical generalization is illustrated by the contrast in Hebrew in (21) below involving the verb xadal ‘stop. Suffice it to say that either of them correctly allows movement of an embedded subject to a ␪-position in control configurations. propositions are simply incompatible with these ␪-roles. we may assume. There are two possible approaches to this issue. Hebrew (Landau 2003): Rina xadla (me-)le’acben et Gil Rina stopped (from-)to-irritate ACC Gil ‘Rina stopped irritating Gil’ . As Landau points out. the contrast in (21) is to be expected.is blocked. Thus. We will leave the choice between these two approaches for another occasion. as respectively shown in (23) below. Nunes’s (2007. standard raising constructions are incompatible with (␾-feature-bearing) complementizers (but see section 5. it is easy to see that movement of ‘John’ is legitimate in (23a) (the ␾-features of C do not induce an intervention effect for ␪-related movements) and in (23c) (there is no intervener bearing ␾-features). but not in (23b).2 Passives. and van den Wyngaerd (2005).2. Rooryck. Their reasoning is the following. the fact that the three sentences above can be derived through Amovement does not entail that the relevant movements have the same motivations (and restrictions) or that the structural configurations are kept constant. and Visser’s generalization 131 Ha-muzika ha-ro’eˇset xadla (∗ me-)le’acben et Gil The-music the-noisy stopped (from-)to-irritate ACC Gil ‘The loud music stopped irritating Gil’ b. If ‘John’ can move in the control structure in (22a) because it does not have its case feature checked. OC. ∗ [Johni tried [ti to win]] [Johni is important [ti to win]] [Johni is likely [ti to win]] Again. its subject will not be able to undergo A-movement for agreement/case purposes due to the intervention of C (a ␾-feature carrier). 2010) proposal reviewed above provides a straightforward account of this generalization. (23) a. [TP is-T␾ likely [TP John␾ to win]] ↑ OK .3 for further discussion). By contrast. c. Hence. why can it not move in (22b) on a par with the raising derivation in (22c)? (22) a. Once these points are disentangled. but by ␾-agreement reasons in (22b) and (22c). The proposal reviewed above also provides an answer to a related challenge posed by van Craenenbroeck. hence control structures may involve an overt complementizer.5. the control and the impersonal constructions in (22a) and (22b) involve CP infinitivals. [TP is-T␾ important [CP C␾ [TP John␾ to win]]] ↑ ∗ c. Moreover. whereas the raising construction in (22c) involves a TP infinitival. b. under standard assumptions. if the movement is ␪-related. [vP v␪ [VP tried [CP C␾ [TP John to win]]]] ↑ OK b. C does not count as an intervener. movement of ‘John’ is sanctioned by ␪-reasons in (22a). for the ␾-features of C induce a minimality effect. In the case of (22). If an infinitival clause has a CP layer. despite their intrinsic interest. (24b) is taken to show that.AUX.3SG wished to dance The second potential counter-argument has a familiar format.INF ‘Somebody wished to dance’ Es scheint getanzt zu werden It seems danced to PASS.1 also accounts for contrasts such as (24) and (25). German (Kiss 2005): Der Mann wurde zu tanzen gew¨unscht The man PASS. as seen in (25a). as shown in (25b).2. b. Johnson.2 Impersonal passives Kiss (2005) claims that German impersonal passives pose two types of problems for the MTC.AUX. In other words. which have also been claimed to offer deadly counter-evidence to the MTC. the contrasts in (24) and (25) are more relevant to a proper analysis of (impersonal) passives. a raising verb can.3SG wished danced to PASS.132 Empirical challenges and solutions Let us now consider how this proposal can be extended to impersonal passives.3SG to dance wished Es wurde gew¨unscht zu tanzen It PASS. than to control per se. In turn. passivization of an embedded subject in a subject-control structure is disallowed in German. Interestingly. (25) a.2. just as we saw in English. the interpretation of (24b) is that the implicit argument of the matrix verb controls the external argument of the embedded verb. one must first establish the relevant property that languages like German have that allows a simple impersonal passive such as (26) below. (24a) shows that.AUX. For concreteness. 5. if no movement takes place.AUX. ∗ German (Kiss 2005): Es wurde gew¨unscht getanzt zu werden It PASS. the matrix subject-control verb can appear in the passive voice. ∗ b. and Roberts’s (1989) proposal and show that the analysis of Visser’s generalization reviewed in section 5. however. before analyzing (24) and (25). It is beyond the scope of this volume to provide such an analysis. The first one involves contrasts such as the one in (24) below. (24) a. yielding an impersonal construction.AUX. we will assume the gist of Baker. Although a passivized subject-control verb cannot take an impersonal passive for a complement. .INF ‘It seems that someone is dancing’ It seems to us that. Johnson. we take the passive morpheme to be a type of light verb with one distinctive property: its external argument may be null (IMP) or realized as a by-phrase. the contrast between English and German with respect to impersonal passives is due to the different case-licensing possibilities the passive morpheme has in each language: accusative in English and accusative or nominative in German. vP] to receive the external ␪-role assigned by -en in a derivation 5 From this perspective. d. Johnson.2 Passives. vP].5 As for the difference between English and German. (i) John was arrested (ii) a. and Visser’s generalization (26) 133 German (Jaeggli 1986): Es wurde getanzt It was danced ‘There was dancing’ Baker. Under this view. where -en assigns the external ␪-role and inherent case to IMP. we assume that in English IMP is licensed by structural (accusative) case.5.1. (28) [vP IMP [v’ -en␾ [VP wished [CP C␾ [TP [the man]␾ to dance]]]]] ∗ One could ask what prevents the embedded subject from moving to the matrix [Spec. and Roberts’s proposal and updating it in current parlance. Adapting Baker. [vP IMP [v’ -en␾ [VP arrested John]]] [vP Johni [v’ IMP [v’ -en␾ [VP arrested ti ]]]] [TP was-T␾ [vP Johni [v’ IMP [v’ -en␾ [VP arrested ti ]]]]] [TP Johni was-T␾ [vP ti [v’ IMP [v’ -en␾ [VP arrested ti ]]]]] . the derivation of a standard passive such as (i) below proceeds along the lines of (ii). as the two Specs are equidistant. OC.2. TP] first stops in the outer Spec of vP (cf. given the configuration in (28) (with English words). Notice that movement of the object to [Spec. c. movement of the embedded subject to the outer Spec of -en for purposes of ␾-agreement (see footnote 5) is blocked by the ␾-features of C. b. (27) [it was [vP IMP [v’ -en [VP danced]]]] Assuming an analysis along the lines of (27). whereas in German IMP can be licensed by either structural or inherent case. let us now return to (24). Example (24a) is basically subject to the same analysis applied to English in section 5. [iib]). That is. and Roberts propose that the passive morpheme is a clitic of sorts which is assigned the external ␪-role of the predicate and is doubled either by a “by-phrase” or by an empty category (IMP). This intermediate step circumvents a potential minimality violation induced by IMP in the inner [Spec. the impersonal passive in (26) is to be represented as in (27) (with English words). According to them. McGinnis 1998.134 Empirical challenges and solutions without IMP. as shown in (32a) below (with English words). zu dem Treffen It PASS. [vP -en␪ [VP wished [CP C␾ [TP IMP to dance]]]] [vP IMPi [v’ -en [VP wished [CP C␾ [TP ti to dance]]]]] ↑ OK Let us now examine the contrast in (25). as sketched in (30) (with English words). which is not the case in (24a). IMP is generated in the embedded clause. (29) German (Kiss 2005): Es wurde von dem Mann gew¨unscht. where it receives the external ␪role of the embedded predicate. b. and Rezac 2004). However. The answer should be: “Nothing!” for the ␾-features of C are oblivious to ␪-related movements. In other words. its derivation proceeds as in (31) below (with English words). Assuming that inherently casemarked elements are inert for purposes of A-movement (see e. The derivation starts by building an impersonal passive. starting with the grammatical raising construction in (25b). this hypothesized derivation is what actually underlies impersonal sentences such as (29). IMP becomes inert after it receives inherent case and cannot move to check the EPP in (32b). as shown in (31b). where it receives another ␪-role and inherent case from -en. vP]. Further computations then assemble the infinitival TP in (32b).AUX. movement of IMP in (31b) is triggered by ␪-considerations and the ␾-features of C do not render it a proper intervener for such a movement.3SG by the man wished to the meeting zu kommen to come ‘The man wished to join the meeting’ (30) [it was [vP [the man]i [v’ -en␪ [VP wished [CP C␾ [TP ti to come to the meeting]]]]]] ↑ OK As for (24b). if -en assigns its external ␪-role to an overt DP. Crucially. which must have its EPP-feature checked. Notice that (31a) is the typical configuration for subject control to obtain: the matrix light verb (the passive -en) still has to assign its external ␪-role and the embedded subject is still active for purposes of A-movement as it has not checked its case yet. and further computations yield the structure in (31a). The embedded subject then moves to [Spec.. it gets morphologically realized as a by-phrase.g. Hornstein and Nunes 2002. (31) a. where IMP receives inherent case from -en. Expletive . . d. The embedded expletive still has its case unchecked but it cannot enter into an agreement relation with the matrix T due to the intervention of the ␾-features of C. d. the puzzling contrasts in (24) and (25) 6 See Baker. OC. and Visser’s generalization 135 insertion solves this problem. c. IMP in (33b) is a potential ␪-role bearer but it has already received inherent case from the lower -en and has become inert for purposes of A-movement. In turn. after the matrix T is introduced in the derivation. and Roberts’s (1989) empty category IMP. (33) a. Johnson. yielding the convergent result in (32e). [32c]). postulated to represent the external argument in passives. The important point to bear in mind is that. and Roberts (1989: 228–229) for some remarks on the similarities and differences between IMP and arbitrary PRO. as there is no CP layer in a standard raising configuration. upon close inspection. [TP it to be [vP IMP [v’ -en [VP danced]]]] [vP -en [VP wish [CP C␾ [TP it to be [vP IMP [v’ -en [VP dance]]]]]]] [vP IMP [v’ -en [VP wish [CP C␾ [TP it to be [vP IMP [v’ -en [VP dance]]]]]]]] [TP T␾ [vP IMP [v’ -en [VP wish [CP C␾ [TP it to be [vP IMP [v’ -en [VP dance]]]]]]]]] ∗ There are a couple of details to spell out in the approach outlined above. An unsolvable problem then shows up in (33d). as shown in (32c). let us see what goes wrong with (25a). Johnson. The matrix -en in (33b) has to assign its external ␪-role. So the matrix -en can only assign its external ␪-role if another IMP is merged in its Spec. The embedded subject cannot move to receive this ␪-role because it is an expletive. b. e. such as the nature of Baker. After (32d) is built. as represented in (33c). c. Further applications of merge then yield (33b). In addition. vP]. hence the unacceptability of (25a).2 Passives. b.6 But we would like to emphasize that such details have to do with the ultimate analysis of passives and not directly with obligatory control. Its derivation proceeds in an identical fashion to the derivation of (25b) until the point when (33a) below is formed (cf. ‘it’ intervenes between the matrix and the embedded [Spec. No minimality issue arises. [vP IMP [v’ -en [VP danced]]] [TP toEPP be [vP IMP [v’ -en [VP danced]]]] [TP it to be [vP IMP [v’ -en [VP danced]]]] [TP T␾ [VP seems [TP it to be [vP IMP [v’ -en [VP danced]]]]]] [TP iti T␾ [VP seems [TP ti to be [vP IMP [v’ -en [VP danced]]]]]] Finally. (32) a. the expletive then moves to check the EPP and the ␾-features of the matrix T.5. As shown in (37)–(40). as represented in (36). Brazilian Portuguese (Nunes 2008a): Parece [que os meninos fizeram a tarefa] Seems that the boys did the homework ‘It seems that the boys did their homework’ . b.3 Finite control vs. ∗ Brazilian Portuguese: Os meninos foram ditos que n˜ao fizeram a tarefa The boys were said.1. Given the configuration in (35) below (with English words). Rather than being fatal counter-evidence to the MTC. the unacceptability of (34a) is not surprising. (37) a. (35) [vP [the boys] [v’ pro [v’ -en␾ [VP said [CP that␾ [TP t didn’t do the ↑ ∗ homework]]]]]] [TP [the boys] T␾ [VP seem [CP that␾ [TP t didn’t do the homework]]]] ↑ (36) Nunes (2007. The question then is why the movement depicted in (36) does not yield a minimality effect.1.MASC.136 Empirical challenges and solutions prove to be amenable to a streamlined MTC approach. (34) a.2.2. which also involves A-movement for ␾-agreement purposes across a ␾-feature bearing C. movement of the embedded subject for purposes of ␾-agreement is blocked by the intervening ␾-features of C. The unexpected case is the hyper-raising construction in (34b). as illustrated in (34). 5. 2010) proposes that the contrast between (34a) and (34b) is related to an interesting correlation between movement of the embedded subject and movement of the embedded clause.PL that not did.PL that not did. 2008a. the data involving impersonal passives brought up by Kiss (2005) actually turn out to lend support to the MTC and the approach to Visser’s generalization in terms of minimality discussed in section 5.2. hyper-raising Let us now return to the contrast between passivization involving finite control and hyper-raising in Brazilian Portuguese. movement of the embedded subject for purposes of ␾-agreement is possible just in case the embedded CP cannot move.PL the homework ‘It was said that the boys didn’t do their homework’ Os meninos parecem que n˜ao fizeram a tarefa The boys seem.PL the homework ‘It seems that the boys didn’t do their homework’ From the perspective of the approach to Visser’s generalization reviewed in section 5. Brazilian Portuguese (Nunes 2008a): Acabou [que os estudantes viajaram mais cedo] Finished that the students traveled more early ‘It turned out that the students traveled earlier’ a. ruling out (40c) (cf.5. b.2 Passives. If the embedded C counts as an intervener for ␾-related movement. [35]). (40) Brazilian Portuguese (Nunes 2008a): N˜ao foi dito/mencionado [que os meninos fizeram a tarefa] Not was said/mentioned that the boys did the homework ‘It was not said/mentioned that the boys did their homework’ a. b. [[Que os meninos fizeram a tarefa]i parece ti ] That the boys did the homework seems ‘It seems that the boys did the homework’ [[Os meninos]i parecem que ti fizeram a tarefa] The boys seem that did the homework ‘The boys seem to have done the homework’ (38) b. the contrast between (40b) and (40c) follows straightforwardly. ∗ c. 137 [[Que os meninos fizeram a tarefa]i n˜ao foi dito/mencionado ti ] That the boys did the homework not was said/mentioned ‘That the boys did their homework was not said/mentioned’ ∗ [[Os meninos]i n˜ao foram ditos/mencionados que ti fizeram a tarefa] The boys not were said/mentioned that did the homework ‘It was not said/mentioned that the boys did their homework’ Again. it is not surprising that its projection can indeed undergo movement . ∗ (39) Brazilian Portuguese (Nunes 2008a): Periga [que aqueles funcion´arios v˜ao ser demitidos] Is-in-danger that those employees go be fired a. OC. and Visser’s generalization b. [[Que os estudantes viajaram mais cedo]i acabou ti ] That the students traveled more early finished ‘It turned out that the students traveled earlier’ [[Os estudantes]i acabaram que ti viajaram mais cedo] The students finished that traveled more early ‘The students ended up traveling earlier’ c. ∗ [[Que aqueles funcion´arios v˜ao ser demitidos]i periga ti ] That those employees go be fired is-in-danger ‘Those employees are in danger of being fired’ [[Aqueles funcion´arios]] perigam que ti v˜ao ser demitidos Those employees are-in-danger that go be fired ‘Those employees are in danger of being fired’ c. c. b. 8 As observed by Nunes (2008a). thereby freezing movement of the embedded CP and freeing movement of the embedded subject (cf. In turn. (39c) is not as acceptable as (37c) for some speakers. For instance. ∗ Maryi seems to him [ti to be nice] It seems to himi that Johni is nice (42) a. Recall also that finite Ts in Brazilian Portuguese may be ␾-incomplete (see section 4. allowing for hyper-raising. Once C is assigned inherent case. Nunes (2007. despite the fact that they arguably c-command into the raising domain. it becomes inert for purposes of ␾-agreement. this proposal is able to accommodate some micro-variation among speakers. The idea is that if the embedded CP and the embedded subject can both undergo A-movement to participate in a given ␾-agreement relation.’ acabar ‘turn out. Given that inherent case is a lexical property that is to some extent idiosyncratic. 2008a. as sketched in (43). variation across speakers with respect to the lexical idiosyncrasies of specific impersonal predicates is therefore unsurprising. inducing principle-C effects (cf. [41b]/[42b]). Nunes proposes that verbs like parecer ‘seem. In other words. [41a]/[42a] below).8 7 Alternatively. movement of CP blocks movement of the embedded subject as it defines a shorter path towards the targeted specifier. they become immobile for A-purposes and do not count as proper intervener for the movement of ‘Mary’ in (41a) and (42b). ∗ Maryi struck him [ti as a fool] It struck himi that Johni was a fool Returning to (37b)/(38b)/(39b). as in (40b). We will leave a comparison between the A-over-A and the c-command approaches to another occasion.7 The challenge is to determine what renders the embedded C in (37b)/(38b)/(39b) inert for purposes of ␾-agreement. it does not block movement of the embedded subject. it should behave like the experiencers of (41) and (42). [37c]/[38c]/[39c]). .4). b.’ and perigar ‘be on the verge of’ in Brazilian Portuguese assign inherent case to the head of their CP complements. Only when it receives inherent case does it become transparent (the same considerations apply to the impersonal constructions in [46] and [47] below). (41) a. Nunes (2007) has proposed an account of the paradigm in (37)–(40) based on Hornstein’s (2009) reinterpretation of Chomsky’s (1964) A-over-A condition in terms of paths. 2010) argues that this issue is related to the well-known fact that English experiencers in raising constructions do not block movement (cf. which accounts for the immobility of the CP. Under the assumption that the experiencers in (41) and (42) are assigned inherent case by the raising verb. if C is inert for ␾-agreement purposes.138 Empirical challenges and solutions for ␾-agreement purposes. The ungrammaticality of the passive construction in (40c) thus indicates that C counts as an intervener for ␾-purposes regardless of whether it is ␾-complete or ␾-incomplete (see Nunes 2008a for discussion). OC. as sketched in (45). [ib]). and Visser’s generalization (43) 139 [TP DPi T␾ [VP parece/acabou/periga [CP queinherent case [TP ti . condition for hyper-raising to be permitted. but optionally in Brazilian Portuguese (see Ferreira 2000. 2007. [44c]) due to the intervention of C.4). Que eles viajaram parece o´ bvio That they traveled seems obvious ‘That they traveled seems obvious’ b.5.’ Thus. even though inherent-case assignment to the embedded C in (ib) makes it transparent for purposes of A-movement. ]]]] seems/turned out/is on the verge of that ↑ OK Nunes presents two pieces of evidence for this proposal. By contrast. b. The first one involves the contrast between (37) and (44). [44b]) and hyper-raising is blocked (cf. 2009. where ‘parecer’ takes a small clause as its complement. . However. CP can move (cf. . and Nunes 2008a). hyper-raising is not allowed in English (cf. it is arguably the case that ‘seem’ in English also assigns inherent case to its complement CP. ∗ Eles parecem o´ bvios que viajaram They seem obvious that traveled ‘It seems obvious that they traveled’ In (44) CP is not an argument of parecer ‘seem’ but of o´ bvio ‘obvious. . It is also worth pointing out that assigning inherent case to C is a necessary. in Brazilian Portuguese the embedded subject may be active if the embedded finite T is associated with an incomplete set of ␾-features (see section 4. The relevant difference between English and Brazilian Portuguese is that finite Ts assign case to their subjects obligatorily in English. Accordingly. ∗ [[that John left]i seems ti ] seems [that ti left]] ∗ [John i . ]]]]] seems obvious that ↑ ∗ The second piece of evidence regards the paradigm in (46)–(47).2 Passives. Thus. Rodrigues 2004. c. (45) [TP DPi T␾ [VP parece [SC o´ bvio [CP que␾ [TP ti . ‘parecer’ cannot assign inherent case to CP and the embedded C is active for purposes of ␾-agreement relations. (i) a. . 2004. but not sufficient. as is well known. the embedded subject has already checked/valued its case and is inactive for A-movement purposes. Given that the embedded clause of (ia) below is immobile. (44) Brazilian Portuguese (Nunes 2008a): Parece o´ bvio que eles viajaram Seems obvious that they traveled ‘It seems obvious that they traveled’ a. . Nunes takes ‘de’ to be a realization of inherent case. movement of the infinitival is possible just in case ‘de’ is not present.3PL praised o diretor the director a.3PL someone is well easy/difficult ‘These teachers easily/rarely praise someone’ . .140 Empirical challenges and solutions (46) Brazilian Portuguese (Nunes 2008a): E´ f´acil/dif´ıcil (d)esses professores elogiarem os alunos Is easy/difficult of-these teachers praise. ]]] ↑ OK (50) Brazilian Portuguese (Nunes 2008a): (∗ D)esses professores elogiarem algu´em e´ (bem) f´acil/dif´ıcil Of-these teachers praise. which is (optionally) assigned by some impersonal predicates to their CP complements. If ‘de’ is not present or is not licensed.3PL the students ‘These teachers often/rarely praise the students’ b. By contrast. movement of the embedded subject is blocked by C. As we should expect given the present analysis. . b. . as illustrated in (50). as sketched in (49). and does not block movement of the embedded subject. ∗ Os professores s˜ao bem prov´aveis/lament´aveis de terem elogiado The teachers are very probable/regrettable of have. as shown in (48) below. C is assigned inherent case. Importantly.3PL praised o diretor the director ‘It is very likely/regrettable that the teachers praised the director’ Examples (46a) and (47a) show that impersonal predicates such as ‘to be easy/ hard’ in Brazilian Portuguese allow the dummy preposition de ‘of’ to precede their infinitival complements. [46b]). In turn. (48) [TP DPi T␾ is easy/difficult/probable/regrettable [CP C␾ [TP ti . (47) Brazilian Portuguese (Nunes 2008a): E´ bem prov´avel/lament´avel (∗ d)os professores terem elogiado Is very probable/regrettable of-the teachers have. if ‘de’ is present. (46b) and (47b) show that only the predicates that license the dummy preposition admit hyper-raising.3PL the students ‘It’s easy/hard for these teachers to praise the students’ a. Esses professores s˜ao f´aceis/dif´ıceis∗ (de) elogiarem os alunos These teachers are easy/difficult of praise. whereas predicates such as ‘to be probable/regrettable’ do not. hyper-raising can only take place in the presence of the dummy preposition (cf. ]]] ↑ ∗ (49) [TP DPi T␾ is easy/difficult de [CP Cinherent case [TP ti . thereby becoming inert for A-movement. contrasts such as the one in (51) should be universal under the semanticsbased approach. the semantics-based approach will find itself in a very uncomfortable position. finitecontrol and hyper-raising structures in Brazilian Portuguese behave exactly as the MTC predicts. we discuss an argument that Culicover and Jackendoff (2001) presented against the MTC. (51) a. as illustrated in (51). the contrast between passivization of finite-control structures and hyper-raising in Brazilian Portuguese (cf. Under the plausible assumption that the argument structure and the conceptual structure associated with ‘attempt’ and ‘appearance’ are kept constant across languages.3 Nominals and control 141 To wrap up. on the other hand. is not committed to the universality of contrasts such as (51). the two constructions in (51) must be derived by A-movement of the embedded subject motivated by ␪-reasons in (51a) and ␾-agreement/case reasons in (51b). these contrasts “raise no particular problem for theories of control based on argument structure or conceptual structure. 5. In these theories an implicit argument is precisely a semantic/functional argument that has no NP corresponding to it in phrase structure. Thus. ∗ John’s attempt to leave John’s appearance to leave By now. namely. The MTC.” It is worth noticing the different empirical predictions this semantics-based approach and the MTC make with respect to crosslinguistic variation. based on another instantiation of the contrast between control and raising: control from within nominals is allowed in English. From the perspective of the MTC. yielding contrasts . Once this independent fact is taken into account.5. However. if it turns out that there are languages where raising into nominals coexists with control from within nominals. 501–502). [34b]) is due to an independent property. why doesn’t it pattern with raising?” But Culicover and Jackendoff’s criticism purports to go beyond the specific analysis of (51). According to them (pp.3 Nominals and control In this section. but raising into nominals is not. as it is incompatible with such variation. the fact that an element marked with inherent case is inert for A-relations. [34a] vs. it should not be surprising if the syntactic configurations involved in control nominals and raising nominals vary across languages. b. the reader can already anticipate the refrain that accompanies this line of objection: “If control involves A-movement. as such contrasts are taken to favor a semantics-based analysis. b. movement makes it possible for the embedded subject to receive inherent case from ‘afirmac¸a˜ o’ in (52a). the contrast in (52) follows straightforwardly. but not in (52b) as ‘probabilidade’ does not have an additional ␪-role to assign. the differences between the argument or conceptual structures of ‘afirmac¸a˜ o’ and ‘probabilidade’ should suffice to account for both (52) and (53). where the embedded null subject is licensed when its clause is embedded under afirmac¸a˜ o ‘statement.SUBJ done the trabalho] e´ alta job is high ∗ ‘Jo˜ao’s probability that he did the job is high’ From the perspective of the MTC. After all.’ but not under probabilidade ‘probability. the contrast in (52) is replicated in (53) below. After showing that these structures can be adequately handled by the MTC but are problematic for the semantics-based approach outlined by Culicover and Jackendoff. 5. for instance. Under a semantics-based approach. Below we discuss two different cases that bear on the issue of crosslinguistic variation: finite control into indicative noun-complement clauses in Brazilian Portuguese and raising into nominals in Hebrew. where the embedded subject is not null.1 Finite control into noun-complement clauses in Brazilian Portuguese Consider the contrast in (52) in Brazilian Portuguese. Notice the contrast in (52) alone is not enough to make a case against a semantic approach. indicates that its syntactic configuration in English must be such that it prevents movement of the embedded subject. but whether it can account for it.142 Empirical challenges and solutions such as (51) in some languages but not in others. ∗ A probabilidade d[o Jo˜ao]i de [que Øi tenha feito o The probability of-the Jo˜ao of that has. The ungrammaticality of (51b).3. we will then suggest an analysis for the English contrast in (51). Given that nominals in Brazilian Portuguese only assign inherent case. . The question then is not whether the MTC is compatible with crosslinguistic variation with respect to contrasts such as (51) (it is!).’ (52) Brazilian Portuguese (Nunes 2009b): A afirmac¸a˜ o d[o Jo˜ao]i de [que Øi fez o trabalho] e´ falsa The affirmation of-the Jo˜ao of that did the job is false ‘Jo˜ao’s statement that he did the job is false’ a. but from this one cannot conclude that every language will display the same syntactic configuration in this domain. However. 143 ∗ A probabilidade do Jo˜ao de [que a Maria tenha feito o The probability of-the Jo˜ao of that the Maria has. if it is referential. For instance. In this regard.3 Nominals and control (53) Brazilian Portuguese (Nunes in press): A afirmac¸a˜ o do Jo˜ao de [que a Maria fez o trabalho] e´ falsa The affirmation of-the Jo˜ao of that the Maria did the job is false ‘Jo˜ao’s claim that Maria did the job was false’ a.10 9 See Nunes (2009b) for further evidence that referential null subjects within noun-complement clauses in Brazilian Portuguese also behave like A-traces. ∗ Brazilian Portuguese (Nunes 2009b): A hip´otese de [que Ø vai ser eleito] e´ de rir The hypothesis of that goes be elected is of laugh ‘The hypothesis that he’s going to be elected is laughable’ A afirmac¸a˜ o de [que Ø fez o trabalho] e´ falsa The affirmation of that did the job is false ‘The statement that he did the job is false’ Second. for instance.9 (54) a. for the absence of the preposition yields gibberish (see Nunes 2009b for further discussion). In general. as shown in (54).’ as shown in (55) below. 10 In (55) and (56). [57]). the sentence becomes unacceptable. ‘de’ remains optional (cf. Recall from section 4. crucially hinges on independent syntactic properties of Brazilian Portuguese. there is an interesting correlation in Brazilian Portuguese between the presence of a dummy preposition preceding the noun-complement clause and the licensing of the embedded null subject.5. b. ∗ b. if there is no antecedent for a null subject inside a noun complement clause. . [56]). there is no stylistic difference when a referential null subject is involved (cf.SUBJ done the trabalho] e´ alta job is high ∗ ‘Jo˜ao’s probability that Maria did the job is high’ However. on the other hand. there are several aspects that show that the derivation of (52a). 2004. noun-complement clauses may be optionally preceded by the dummy preposition ‘de. an intriguing contrast arises when the noun-complement clause involves a null subject: if the null subject is an expletive. the alternatives with the preposition are generally associated with formal style and written language. 2009 and Nunes 2008a). However. [57]). as Nunes (2009b) observes.4 that referential null subjects in (colloquial) Brazilian Portuguese behave like A-traces rather than null pronominals and this was attributed to its finite Ts being able to host an incomplete ␾-set (see Ferreira 2000. it is worth pointing out that the null subject of (52a) displays the same behavior as the other instances of referential null subjects in Brazilian Portuguese. ‘de’ becomes obligatory (cf. Brazilian Portuguese (Nunes 2009b): A hip´otese (de) [que a Terra e´ chata] n˜ao foi esquecida The hypothesis of that the Earth is flat not was forgotten ‘The hypothesis that the Earth is flat was not forgotten’ Ele comentou a afirmac¸a˜ o do Jo˜ao (de) [que a Ana He commented the affirmation of-the Jo˜ao of that the Ana era inocente] was innocent ‘He commented on Jo˜ao’s statement that Ana was innocent’ Brazilian Portuguese (Nunes 2009b): A hip´otese do Jo˜ao (de) [que Øexpl n˜ao existe The hypothesis of-the Jo˜ao of that not exists movimento-wh nessa l´ıngua] parece estar errada wh-movement in-this language seems be wrong ‘Jo˜ao’s hypothesis that there doesn’t exist wh-movement in this language seems to be wrong’ b. (56) a. (57) a. Building on Stowell (1981). A-movement under the MTC). If ‘de’ encodes a noun-complement configuration in virtue of realizing inherent case. A afirmac¸a˜ o (de) [que Øexpl nunca chove aqui e´ exagerada] The affirmation of that never rains here is exaggerated ‘The claim that it never rains here is an exaggeration’ Brazilian Portuguese (Nunes 2009b): A hip´otese d[o Jo˜ao]i (∗ de) [que Øi vai ser eleito] e´ de rir The hypothesis of-the Jo˜ao of that goes be elected is of laugh ‘Jo˜ao’s hypothesis that he’s going to be elected is laughable’ A afirmac¸a˜ o d[o Jo˜ao]i (∗ de) [que Øi fez o trabalho e´ falsa] The affirmation of-the Jo˜ao of that did the job is false ‘Jo˜ao’s statement that he did the job is false’ Assuming that (57) is also a case of finite control (therefore. Nunes takes ‘de’ in these constructions in BP to be the realization of the inherent case assigned by the subcategorizing noun to the embedded clause. Nunes (2009b) uses contrasts such as (58) below to argue that. the presence or absence of ‘de’ in these constructions respectively signals whether we are dealing with a true complement structure or an appositive of sorts. one wonders why the referential null subjects/A-traces require the presence of the preposition. its presence in (58) yields unacceptable results as these sentences involve a predication configuration. in Brazilian Portuguese.144 Empirical challenges and solutions (55) a. (58) a. b. A hip´otese e´ (∗ de) que o Jo˜ao tenha feito isso The hypothesis is of that the Jo˜ao has done this ‘The hypothesis is that Jo˜ao did this’ . More specifically. b. receives the external ␪-role associated with the subcategorizing noun and is also marked with inherent case (cf. the noun and CP must then undergo pair-merge (in the sense of Chomsky 2000). both inherent cases are realized as ‘de’ in the morphological component (cf. Movement of the embedded subject + ␪-role assignment: [Jo˜aoinherent case affirmation [that Jo˜ao T[N] did this]inherent case ] ↑ d. which remains active for purposes of A-relations. Applications of merge and move: CP = [that Jo˜ao T[N] did this] N = affirmation b. after ‘o Jo˜ao’ moves to the relevant ␪-position associated with ‘afirmac¸a˜ o. the embedded subject cannot move out of it as it would induce a CED 11 The linear order of (57b) indicates that. By contrast. [59f]). After the noun and the CP undergo set-merge (in the sense of Chomsky 2000). However.3 Nominals and control b. the version of (57b) without ‘de’ has no convergent derivation at its disposal. Merger between N and CP + inherent-case assignment: [affirmation [that Jo˜ao T[N] did this]inherent case ] c. . Deletion of copies in the phonological component: [affirmation [Jo˜aoinherent case affirmation [that Jo˜ao T[N] did this]inherent case ]] f. the CP is assigned inherent case in virtue of the ␪-role it receives from the noun (cf. Finally.’ the latter moves to a higher position. [59b]). the embedded subject moves. (59) a. Next. Realization of inherent case: [affirmation [de Jo˜ao] [de that did this]] Like in the other instances of finite control in Brazilian Portuguese. Movement of the head noun:11 [affirmation [Jo˜aoinherent case affirmation [that Jo˜ao T[N] did this]inherent case ]] ↑ e. the derivation of the version of (57b) with the preposition involves the steps sketched in (59) (with English words).5. if the CP becomes an adjunct. The nature of such positions is orthogonal to the current discussion. 145 A alegac¸a˜ o e´ (∗ de) que a Maria viaja muito The allegation is of that the Maria travels much ‘The allegation is that Maria travels too much’ With this overall picture in mind. Given that the absence of ‘de’ signals that the embedded CP is an adjunct rather than a complement. the embedded T is associated with an incomplete ␾-set in (59a) (see Ferreira 2000 and Nunes 2008a) and is unable to check the case of the embedded subject. [59c]). and circumvent potential linearization problems (cf. (60) a. the only relevant possibility to be considered is the one in which the embedded subject undergoes sideward movement before CP becomes an adjunct (see sections 4.5. the upper copy of ‘Jo˜ao’ c-commands and forms a chain with the lower copy. By contrast.’ as well as precede itself (see Nunes 1999. Applications of merge and move: CP = [that Jo˜ao T[N] did this] N = affirmation b.146 Empirical challenges and solutions violation. . we account for why they require an antecedent and why the clauses containing them must be true complements (preceded by ‘de’) and not adjuncts (lacking ‘de’). Movement of the head noun (see footnote 11): [affirmation [NP [NP Jo˜aoinherent case t] [CP that Jo˜ao T[N] did this]]] Nunes (2009b) argues that.3). Assuming that deletion of copies can only operate with chains. the MTC is able to account for all the data concerning referential null subjects within noun-complement clauses in Brazilian Portuguese. 2004 for discussion). i.1. as illustrated in (60) (with English words). Failure to delete one of the copies of ‘Jo˜ao’ in (60d) in turn causes linearization problems as the system gets contradictory instructions: ‘Jo˜ao’ should precede and be preceded by ‘that.5. 12 It is immaterial for the purposes of our discussion if CP adjoins to a projection higher than NP. Under the assumption that such subjects are A-traces. Going back to the comparison between the MTC and approaches based on argument or conceptual structure. [59e]).2 and 4. Short of ad hoc provisos. Sideward movement (copy +merge)+ θ-role assignment: CP = [that Jo˜ao T[N] did this] NP = [Jo˜aoinherent case affirmation] c. Thus.. when we have a true noun-complement structure as in (59d). Adjunction of CP to NP:12 [NP [NP Jo˜aoinherent case affirmation] [CP that Jo˜ao T[N] did this]] d. Note that when ‘de’ is present instead.e. the final output cannot be linearized. chain reduction (see Nunes 2004) cannot be employed in (60d).1. although the derivational steps in (60) are licit. Notice that the copies of ‘Jo˜ao’ in (60d) are not in a chain configuration as they do not stand in a c-command relation. the empirical coverage of the semantics-based alternative is limited to (52). allowing chain reduction to apply in the phonological component. there seems to be no coherent way to explain the behavior of referential null subjects within noun-complement clauses in Brazilian Portuguese and their apparent requirement of a dummy preposition. delete the lower copy. 5. it is fair to say that the semantic account of the contrast in (52) turns out to be spurious.FEM. Therefore.13 (62) a. when all pertinent data are taken into consideration.3 Nominals and control 147 based solely on the argument or conceptual structure of the relevant noun. ∗ b. In sum. and idiom chunks respectively. along with standard control from within nominals. Hebrew also allows constructions that involve raising into nominals.2 Raising into nominals in Hebrew The problem posed by Hebrew to semantics-based approaches to control is even stronger than the one presented by finite control into noun-complement clauses discussed above. In turn. b. the b-sentences involve a raising noun and the element case-marked by it has raised from the embedded clause. (61) Hebrew (Sichel 2007): ha-nisayon Sel rina [le-hagi’a ba-zman] the-attempt of Rina to-arrive on-time ‘Rina’s attempt to arrive on time’ a. therefore being incompatible with inanimate elements. (63) a. As convincingly argued by Sichel (2007). ha-sikuyim Sel rina [le-hagi’a ba-zman] the-chances of Rina to-arrive on-time ‘Rina’s chances to arrive on time’ Evidence that (61b) does involve raising is provided by the pairs in (62)–(64) below. Sichel concludes.’ which imposes no such restrictions. as illustrated in (61). 5.SG slim quite ‘The chances of the theory being correct are pretty slim’ ∗ Hebrew (Sichel 2007): [ha-nisayon Se ze likrot [Se-bibi yibaxer]] hifti’a otanu the-attempt of it to-happen that-Bibi will-be-elected surprised us [ha-sikuyim Se ze likrot [Se-bibi yibaxer]] tovim the-chances of it to-happen that-Bibi will-be-elected good ‘The chances of it happening that Bibi will be elected are good’ 13 See Sichel (2007) for further evidence and discussion. The a-sentences of (62)–(64) show that the control noun corresponding to ‘attempt’ imposes selectional restrictions on the DP associated with it. expletives. the b-sentences show that the opposite holds of the noun corresponding to ‘chances. Hebrew (Sichel 2007): [ha-nisayon Sel ha-te’oria lihiyot nexonot] hirgiz otanu the-attempt of the-theory to-be correct annoyed us [ha-sikuyim Sel ha-te’oria lihiyot nexona] kluSim le-maday the-chances of the-theory to-be correct. b.3. . the movement is triggered by ␪-related reasons. Importantly. as illustrated in (66). the licensing of the embedded subject of (66a) is similar to what we see in the ECM constructions in (67).1). Although both structures in (61). in nominal constructions a negative embedded subject must be licensed by the matrix and not by the embedded verb. In (61a). ∗ Hebrew (Sichel 2007): [ha-nisayon Sel ha-kerax le-hiSaver be-macav ka-ze] hu tipSi the-attempt of the-ice to-break in-situations like-this is silly [ha-sikuyim Sel ha-kerax le-hiSaver be-macav ka-ze] kluSim the-chances of the-ice to-break in-situations like-this slim ‘The chances of the ice breaking in this kind of situation are slim’ (idiomatic reading) The fact that raising nominals are attested in Hebrew provides strong evidence against an account of the impossibility of such nominals in English in terms of argument or conceptual structure. In particular. hence. or idiom chunks (cf.148 Empirical challenges and solutions (64) a. [64a]). Leaving English to the next section. As Sichel points out. [62a]). the motivation is different in each case. as shown in (65) below. the two of them should not be different with respect to raising. negative DPs must be licensed by clause-mate negation. (65) Hebrew (Sichel 2007): af takmid (∗ lo) nice’ax no student NEG won ‘No student won’ . for instance. In Hebrew. involve movement of the DP case-marked by the dummy preposition. b. the incompatibility with elements that cannot bear this ␪-role such as inanimate elements (cf. Hence. the relevant A-movement involved must be triggered by ␾-agreement/case considerations. let us consider under what conditions raising nominal constructions should be allowed in Hebrew from the perspective of the MTC. If so. [63a]). which under standard assumptions should not involve a CP layer. the MTC is much better equipped to handle these cases as it relies on the different syntactic configurations that may underlie raising nominal constructions in Hebrew and English. namely.2. there can be no intervening element bearing ␾-features. As for the raising structure in (61b). It is reasonable to suppose that the argument and conceptual structures of these nominals are the same in the two languages. as C hosts ␾-features (see section 5. By contrast. to allow the external ␪-role associated with ‘attempt’ to be assigned. expletives (cf. there should be no CP projection intervening between the raising nominal and the embedded subject. Evidence that this conjecture is correct is provided by Sichel’s (2007) discussion of negative concord. which looks very similar to what happens with hyper-raising out of inflected infinitivals in Brazilian Portuguese (see section 5. “there do exist examples that appear to be parallels of raising in nominals. 149 ∗ (67) he’emanti [ba-sikuyim/netiya Sel af talmid lo le-hitkonen] believed-I-in the-chances/tendency of no student NEG to-prepare ‘I didn’t believe in the chances/tendency of any student preparing’ Hebrew (Sichel 2007): lo zaxarti [af talmid mitkonen] NEG remembered no student preparing ‘I didn’t remember any student preparing’ The existence of raising into nominals is therefore not surprising from the perspective of the MTC. (69) John’s likelihood/probability of winning A comparison between (68a) and (70) below. 5.3 The contrast between raising nominals and control nominals in English Now we have seen that raising into nominals should not be excluded as a matter of principle. All depends on the specific syntactic structures involved. footnote 10) acknowledge. repeated here in (68).3.3 Nominals and control (66) Hebrew (Sichel 2007): lo he’emanti [ba-sikuyim/netiya Sel af talmid le-hitkonen] NEG believed-I-in the-chances/tendency of no student to-prepare ‘I didn’t believe in the chances/tendency of any student preparing’ a. b. the existence of data such as (69) is quite problematic for approaches to control based on argument or conceptual structure. is very suggestive. mentioning the data in (69) below. and (69). (68) a.3). the contrast between control and raising in the nominal domain may shed more light on the structure of nominal expressions than the nature of control. on the other. (70) ∗ John’s likelihood/probability to win The acceptable pattern is possible when the dummy preposition ‘of’ is present. As Culicover and Jackendoff (2001.2. In this sense. let us return to the contrast in (51). ∗ John’s attempt to leave John’s appearance to leave The first thing to point out is that it is not exactly correct that English never allows raising into nominals.5. on the one hand. b.” As in the case of Hebrew. Following a suggestion . In other words. as illustrated in (73) (cf. [iia]). Suggestive evidence for such an approach is the contrast in (i).’ Once CP receives inherent case. as illustrated in (72). (i) a. as opposed to to-infinitivals (cf. of-gerunds behave like standard complements (cf. The derivation of (70) involves movement of the embedded subject for ␾-agreement/case reasons skipping C. [ib]/[iib]).150 Empirical challenges and solutions by Lisa Cheng (personal communication). the subcategorizing nominal assigns inherent case to its complement. b. [60]). The contrast between (69) and (70) now follows the interaction between inherent case and ␾-intervention discussed earlier with respect to hyper-raising in Brazilian Portuguese. the presence of the dummy preposition ‘of’ signals a complement from which movement is allowed. proposes that ‘of’ in constructions like (69) is in fact the realization of the inherent case assigned to the embedded non-finite clause. which is morphologically realized as ‘of. (73) a. ∗ The likelihood/probability was of winning The desire/attempt was to win ∗ The driver is of my car The book is about Chomsky . However. b. which yields a minimality violation. in predicative environments. English does not allow raising of expletives or idiom chunks in this context. conversely. (71) [DP ’s␾ [NP likelihood/probability [CP C␾ [TP John␾ to win]]]] ↑ ∗ As for (69). Nunes (2010). and explains why a dummy preposition should be resorted to. this proposal accounts for the fact that only some nominals admit raising (inherent case is to some extent a lexical idiosyncrasy). [63b] and [64b]). we can extend the analysis of finite control into nominals in Brazilian Portuguese (see section 5. the absence of the dummy preposition may indicate an adjunct to NP and then there is no way to derive the structure licitly via sideward movement (cf. as sketched in (71).1) to English. b.3.14 (72) [DP ’s␾ [NP likelihood/probability [CP Cinherent case [TP John␾ winning]]]] ↑ OK Although admittedly sketchy. it is no longer active for A-purposes and its head is not computed for purposes of A-minimality. (ii) a. ∗ ∗ its likelihood of raining/annoying me that Jane is late the shit’s likelihood of hitting the fan in these situations (Sichel 2007) 14 Alternatively. it does not explain why raising into the nominal domain in English is much more restricted than what we saw in Hebrew. which shows that. In particular. Movement of the embedded subject then proceeds without problems. That is. the moved subject seems to occupy a position lower than the subcategorizing noun (cf. ∗ What headway do you wonder [how PRO to make t on this project] ?What project do you wonder [how PRO to make headway on t] (Rizzi 1990) In the case of raising. but allowed by the verbal -ing in (76b). N in English should induce a minimality effect for non-referential elements similar to what we find in the A’domain.3 Nominals and control 151 Following Nunes (2010). the contrasts in (77) below can receive the same minimality account if we assume with Rosenbaum (1967) and Hornstein and Witkos . on the other hand. . (76) a. but verbal in (76b) (see e.5. b. Perhaps this is what makes it transparent for the movement of true arguments. we suggest that the difference between English and Hebrew has to do with the span of the relevant A-movement in each language. b. blocking non-referential expressions from raising. as illustrated in (75). .. ]]] ↑ In English. but also the subcategorizing N (cf. . movement of the idiom chunk is blocked by the nominal -ing in (76a). as sketched in (74). . [74b]).. That being so. ]]]] ↑ Hebrew: [NP N [ .g. English: [DP ’s [NP N [CP Cinherent case [TP DP . If so. contrasts such as the ones illustrated in (76) below follow from the fact that the functional head associated with -ing is nominal in (76a). the moved subject crosses not only C. The cat’s being out of the bag was a big problem for the government (idiomatic reading: ∗ ) The cat being out of the bag was a big problem for the government (idiomatic reading: OK) As for expletives. within NP. where referential and non-referential wh-phrases sharply contrast with respect to movement across a weak island (see e. because it is ultimately a ␾-feature bearer. As for why referential expressions are not subject to such intervention. . it is worth noting that. (74) a. (75) a. if we are to judge by the surface order of (61)–(64). [74a]). the subcategorizing noun functions as a predicate and not as an argument. In Hebrew. . b.g. Chomsky 1970 and Reuland 1983). Rizzi 1990). [TP DP . it is plausible to think that the raising nominal induces (weak) intervention effects for ␾-related movements. San Martin 2004. .152 Empirical challenges and solutions (2003).’ We believe there are good reasons to extend the insight to ‘there’ (see Hornstein and Witkos 2003). By contrast.g. Sigurðsson (2008). the contrast with constructions involving control nominals. b.1 Quirky case and the contrast between raising and control in Icelandic Landau (2003. But it is worth emphasizing that this fine-grained range of (im)possible instances of raising within a single language raises the same kind of problem as crosslinguistic variation regarding raising into nominals poses to semantic-based approaches of the type envisioned by Culicover and Jackendoff (2001). 5. Landau 2003. is of a different nature – ␪-related rather than ␾-related. is unsurprising. It has also been claimed that these correlations constitute knock-out evidence against the MTC (see e. 2006. the matrix subject of raising constructions surfaces with the quirky dative case specified by the embedded 15 Rosenbaum (1967) made this proposal only for ‘it.15 (77) a. which are much less diversified. that ‘it’ and ‘there’ are generated together with their “associates” (the CP complement and ‘someone’ respectively) before moving to the subject position of the gerund. for the relevant movement involved in control. these fine-grained distinctions are very congenial to the MTC as their source arguably stems from minimality issues governing movement for ␾-agreement/case purposes. 5. We focus on these two languages as their overt morphology presents interesting case/agreement correlations that have been taken to shed light on the nature of control. and Bobaljik and Landau 2009). Sigurðsson 2008. As illustrated in (78) and (79) below. it seems to us. 2007. more on the ultimate structure of nominals per se than on the MTC.4 Obligatory control and morphological case Let us now examine the case and agreement patterns found in control contexts in Icelandic and Basque. although being of the A-type. 2006. Let us then see whether the MTC is up to the challenge. Furthermore. and Bobaljik and Landau (2009) have claimed that the fact that raising and control constructions differ with respect to the realization of quirky case constitutes a fatal problem for the MTC.. among others. It/∗ its seeming that we would get a raise motivated everyone to work harder There/∗ there’s being someone here was surprising Whether this general account will prove fruitful will depend.4. 2007). case realization should be the same in both types of constructions. b..DAT have been helped. when the relevant properties involved in quirky-case assignment and agreement are sorted out. quirky case in Icelandic displays properties of both inherent and structural case (see e.DFLT ‘The boys are believed to have been rescued’ (Andrews 1990.DFLT of the-mountain ‘He hopes to be rescued from the mountain’ (Andrews 1990.DAT/∗ NOM ‘The men seem to have both been helped’ (Sigurðsson 2008) Hann/∗ Honum vonast til að verða bjargað af fjallinu He. we show below that. [78b]/[79b]).MASC. [78a]/[79a]).DFLT/∗ PL.g. Zaenen.DFLT Men-the. reproduced in Bobaljik and Landau 2009) Icelandic: Str´akunum er talið (hafa The-boys. As is well known.SG believed. reproduced in Bobaljik and Landau 2009) Str´akarnir vonast til að verða hj´alpað/∗ hj´alpaðir/∗ hj´alpuðum The-boys. as opposed to the matrix subject of control predicates. the MTC in fact makes the right cut with respect to contrasts such as (78) and (79). If both raising and control involve A-movement. b. (79) a. elements marked with quirky case can undergo standard A-movement in passives and ECM constructions. so the argument goes.NOM/∗ PL. As is the case in other languages. Hornstein.DAT ‘The boys hope to be helped’ (Sigurðsson 1991. is. Thus.5. whose case realization is determined solely by the properties of the matrix domain (cf. (78) a. Icelandic has a morphologically rich case-agreement system in which structural and quirky case are associated with different agreement paradigms (for comprehensive overviews. see Sigurðsson 1991 and Thr´ainsson 2008). Maling.DFLT to-have verið) bjargað been rescued.PL.4 Obligatory control and morphological case 153 verb (cf. DAT. On the other hand. reproduced in Bobaljik and Landau 2009) Following Boeckx. for instance.NOM/∗ DAT hopes for to be rescued. it is unlike inherent case in that it does not render its recipient frozen for purposes of A-movement. as respectively .NOM hope for to be helped. it is associated with a ␪-role and is lexically determined. and Nunes (in press). it rather behaves like structural case in requiring an agreement relation with a ␾-complete head in order to be deactivated for A-purposes. Icelandic: M¨onnunum/∗ Mennirnir virðist b´aðum hafa verið hj´alpað seems both. and Thr´ainsson 1985). Like inherent case and unlike structural case. 17 (81) a.PL was rescued. c. the verb merges with DP and assigns quirky case to it. before jumping to hasty conclusions. as shown in (81b). b. But. e. we should first examine in more detail how quirky-case assignment and checking obtain in a simple sentence. . φ:d :df lt] OK [VP rescued t]]]] [TP [the boys][Case:DAT] [T’ T[φ [φ:d :df lt] be [vP t [v’ IMP [v’ -en[Case:dflt. In other words. ␾:?] [VP rescued [the boys][case:DAT] ]]] OK [vP [the boys][case:DAT] ] [v’ IMP [v’ -en[Case:dflt. d. [80a]). agreement on the finite T is determined by the nominative object but it cannot be first or second person (see e. Assuming the structure of passives outlined in section 5. f.DFLT ‘I believe the boys to have been rescued’ Given (80). Sigurðsson 1996). the quirky case marked DP is still active for purposes of A-relations and its case feature must 16 In addition. for instance. φ:d :df lt] [VP rescued t]]]]]] Given the derivational step in (81a). elements bearing quirky case are indeed quirky mainly from a morphological point of view. [80b]) and systematically fail to trigger agreement on the finite verb (cf.2.PL to-have been rescued. let us consider the derivation of the quirky passive sentence in (80a). in this section we will assume that T has already inherited the ␾-features of C (see Chomsky 2008). b. but not under control. reproduced in Bobaljik and Landau 2009): Str´akunum var bjargað The-boys.. Icelandic (Andrews 1990.DAT. [80a]) or ECM (cf.2. 17 For ease of exposition. Such an assignment only means that the case feature of the DP has been valued as dative. if a given clause involves a quirky subject and a nominative object. Crucially.MASC. as they do not lose their morphological case under passivization (cf.DAT.g. it indeed appears to be surprising from the perspective of the MTC that quirky case is preserved under raising.16 (80) a.154 Empirical challenges and solutions illustrated in (80) below. V = rescued DP = [the boys][case:?] [VP rescued [the boys][case:DAT] ] [vP IMP [v’ -en[case:?.MASC.DFLT ‘The boys were rescued’ ´ tel Eg str´akunum (hafa verið) bjargað I believe the-boys. as sketched in (81) (with English words). φ:d :df lt] [VP rescued t]]] ↑ [TP T[␾:?] be [vP [the boys][case:DAT] ] [v’ IMP [v’ -en[Case:dflt. 4 Obligatory control and morphological case 155 be checked against a ␾-complete probe in order to be deactivated. It is. however.2) (82) a. as illustrated in (82) (see section 3.g. when a given probe enters into an agreeing relation with an element marked with quirky case. [88]/[89] below). yielding the structure in (81f). the DP [the cat] in (82a) acquires the idiosyncratic meaning of the idiom when it merges with [out of the bag]. but not under control. an ungrammatical result obtains. As mentioned above. which leaves the idiosyncratic meaning specification of [the cat] unaltered. movement in (82b) is ␪-related and. they can enter into an agreement relation and the object can move to [Spec. neuter. which has case and an incomplete set of ␾-features (gender and number). Either [the cat] cannot move because it is not a potential ␪-role bearer once it has become 18 This is also the pattern found with nominal and adjectival predicates (cf. and singular. On the pattern displayed by secondary predicates and floating quantifiers.18 Notice that. By contrast. Agreement between the ␾-complete T and the quirky DP then sets the value of the ␾-features of T as default (third-person singular) and all the uninterpretable features – including the case feature of the subject – get deleted (for LF purposes). in this case. b.5. which are deleted (for LF purposes) once valued. .4. the features of -en in (81d) get valued as nominative. the case feature of the moved object remains active as it has not entered into an agreement relation with a ␾-complete probe yet. the features of the probe get default values if there is no nominative goal around (see footnote 16). the remarks above constitute no innovative treatment of quirky case. Given that both -en and the object DP are active. They just spell out in Agree parlance the old intuition that quirky case has both inherent and structural characteristics (see e. hence. The next step of the derivation in (81c) introduces the passive -en. as opposed to the features of -en.. as represented by the outlined characters. Freidin and Sprouse 1991 and Chomsky 2000). as sketched in (84). movement of [the cat] to the matrix subject position in (82a) and (82b) has different implications. Obviously. the logic to be exploited here is no different from the one we used to account for why idiom interpretation is preserved under raising. which surfaces as (79a). as shown in (81d) (see footnote 5).2 below. see section 5. sufficient for us to tackle the contrast between raising and control illustrated in (78) and (79). movement is triggered for ␾-agreement reasons. as represented in (83) below. This only happens when the finite T enters the derivation in (81e). That being so. The cat seems to be out of the bag The cat tried to be out of the bag (idiomatic interpretation: OK) (idiomatic interpretation: ∗ ) Arguably. In the case of (82a). In fact. -en]. φ:d :df lt] . Agreement between T and the quirky DP + movement: [TP [the boys][Case:DAT] [T’ T[φ [φ:d :df lt] be [vP t [v’ IMP ↑ [v’ -en[Case:dflt. if it moves.156 Empirical challenges and solutions an idiom chunk. . Agreement between the passive participle and the quirky DP + movement: [vP [the boys][case:DAT] ] [v’ IMP [v’ -en[Case:dflt. ]]]]] e. ]]]]] ↑ f. ]]]] g. Assignment of quirky case: [rescued [the boys][case:DAT] ] b. Applications of merge and move: [vP IMP [v’ -en[case:?.PL is. . The derivation of the raising construction in (79a).DFLT to-have been rescued. φ:d :df lt] [VP believed . b. .DFLT] . [vP v␪ [tried [CP C␾ [TP [the cat]idiom chunk to be [t out of the bag]]]]] ↑ ∗ b. . [vP [the cat]idiom-chunk v [tried [CP C␾ [TP t to be [t out of the bag]]]]] ↑ Similar reasoning accounts for the raising-control contrasts in (78) and (79).DFLT believed. Merger of a φ-complete probe: [TP T[␾:?] be [vP [the boys][case:DAT] ] [v’ IMP [v’ -en[Case:dflt. . proceeds along the lines of (86) (with English words). [the-boys. . φ:d :df lt] [VP believed [TP t . ]]]]] h. . as represented in (85a).DAT. φ:d :df lt] [VP rescued t]]] ↑ d. for instance. (83) a. assignment of the external ␪-role to it obliterates its idiomatic specification and it can no longer be interpreted as an idiom chunk. as shown in (85b). or. Merger of a φ-incomplete probe: [vP IMP [v’ -en[case:?. ␾:?] [VP rescued [the boys][case:DAT] ]]] c. ␾:?] [VP believed [TP [the boys][case:DAT] to have been ↑ [vP t . DP = [the cat] PP = [out of the bag] [[the cat]idiom chunk [out of the bag]] (84) [TP [the cat]idiom chunk T␾ seems [TP t to be [t out of the bag]]] ↑ OK (85) a. . (86) a.MASC. Agreement between the passive participle and the quirky DP + movement: [vP [the boys][case:DAT] ] [v’ IMP [v’ -en[Case:dflt. . Movement and θ -assignment: [vP [the boys][Case:?] [v’ v [VP hope [CP C [TP t to be [vP t . proceeds as sketched in (87) (with English words). the embedded object triggers default agreement on its way to the matrix [Spec. case valuation. . Recall that a given quirky case is intrinsically tied to a specific ␪-role. ␾:?] [VP helped [the boys][case:DAT] ]]] c. Agreement between the passive participle and the quirky DP + movement: [vP [the boys][case:DAT] [v’ IMP [v’ -en[Case:dflt.5. and movement: [TP [the boys][Case:NOM] [T’ T[φ [φ:3PL] :3PL] [vP t [v’ v [VP hope [CP C [TP t to be ↑ [vP t . ]]]]]]] g. it is natural to assume that assigning an additional ␪-role to an element bearing quirky . for instance. . . Merger of a φ-complete probe: [TP T[␾:?] [vP [the boys][Case:?] [v’ v [VP hope [CP C [TP t to be [vP t . . Thus. . ]]]]] ↑ e. [86h]). Merger of a φ-incomplete probe: [vP IMP [v’ -en[case:?.DFLT] The derivation of the embedded clause proceeds like the derivation of raising constructions until we hit the derivational steps in (87d–e).NOM hope. ]]]]]]]] h. ]]]]]] ↑ f. By contrast. [81]). As opposed to the derivational steps in (86c) or (86e). Applications of merge and move: [vP v␪ [VP hope [CP C [TP [the boys][Case:DAT] to be [vP t . [86c] and [86e]). φ:d :df lt] [VP helped t]]]] ↑ d. as shown in (86g). (87) a. The quirky DP will only become inactive after agreeing with the finite T (a ␾-complete probe). [the-boys. Assignment of quirky case: [helped [the boys][case:DAT] ] b.3PL to be helped. Agreement between T and DP. .4 Obligatory control and morphological case 157 After having its case feature valued in (86a). . movement of the quirky DP in (87e) is triggered by the ␪-properties of the matrix light verb and this makes a very big difference. setting their features to default values (cf. TP] and surfaces with the quirky case it received in the most embedded clause (cf. Like in the derivation of a simple passive (cf. which involve movement driven by ␾agreement. but remains active for purposes of A-movement as passive morphemes are not associated with a complete ␾-set (they do not have the feature person). the derivation of the control structure in (79b). the quirky DP enters into an agreement with the two passive morphemes. . Given that in (87) the probe is a finite T.FEM[case:DAT] ]] OK c.ACC longs not for to be cold.FEM[case:DAT] ] b. obliteration of the quirky-case value seen in (87e) is similar to what we saw in (85b). Notice that the controller surfaces with structural case in (87h) because the ␪role it received in the matrix [Spec. ]]]]]] ↑ 19 Example (i) shows that the embedded predicate of (88) (under the intended meaning) assigns quirky dative to its subject.19 (88) Icelandic (Sigurðsson 2008): Hana langar ekki til að vera kalt Her.3SG.3SG. with the DP having its case valued through agreement with a ␾-complete probe (cf. [87h]). (i) Icelandic (Sigurðsson 2008): M´er er kalt Me.DFLT ‘I am (feeling) cold’ .FEM[case:DAT] to be [IP t . In other words. Agreement between Infl and the quirky DP + movement: [IP pron.3SG cold. Applications of merge and move: [vP v␪ [VP want [CP C [TP pron. Movement and θ-assignment: [vP pron. the derivation then proceeds in a standard fashion. This is how we propose control structures with quirky case assigners in both the matrix and the embedded clause are to be derived. . The derivation of (88).3SG. . vP] was not tied to any specific morphology.FEM[case:ACC] [v’ v [VP want [CP C [TP t to be [IP t .3SG. where ␪-assignment to an idiom chunk deletes the idiom specification. though. ]]]]] ↑ e. . Once the quirky-case value in (87e) is eliminated.DAT is. Merger of a φ-incomplete probe: [IP Infl[␾:?] [AP cold pron.158 Empirical challenges and solutions case may obliterate the quirky-case value previously specified. the controller will then surface with the last quirky case it received and will trigger default agreement of the finite T.DFLT ‘She doesn’t want to be (feeling) cold’ (89) a. [87g]).3SG.FEM[case:DAT] [I’ Infl[φ [φ:d :df lt] [AP cold t]]] ↑ d. proceeds as sketched in (89) (with English words). This is not the only possibility. If the ␪-role of the matrix predicate is associated with quirky case. Assignment of quirky case: [AP cold pron. the moved DP surfaces as nominative (cf. for example. . . We would like to stress that assignment of a ␪-role to an idiom chunk and assignment of a ␪-role to an element marked with quirky case are similar but not identical. the previous quirky-case value is eliminated and the one associated with the ␪-assignment of the matrix predicate (accusative) is specified. .DFLT not to be cold.ACC wants. which surfaces as (89h) (cf. Agreement between T and DP + movement: [TP pron.FEM[Case:ACC] [v’ v [VP want [CP C [TP t to be [IP t . the preposition preceding the controller is determined by the embedded rather than the matrix verb. [her. Merger of a φ-complete probe: [TP T[␾:?] [vP pron.5.3SG. [88]). as seen in (89g). [87e]). [89c]). The matrix light verb needs to assign its external ␪-role and the embedded subject is still active for the computation as it has not entered into an agreement relation with a ␾-complete probe. The interesting step for the current discussion is the one after (89d) is assembled. That different 20 Nothing will substantially change if it turns out that the quirky case is assigned by the matrix main verb instead of the light verb. Further checking with finite T finally deactivates the case feature of the moved pronoun and values the ␾-features of T as default. As we saw earlier.3SG. .FEM[Case:ACC] [T’ T[φ [φ:d :df lt] [vP t [v’ v [VP want [CP C ↑ 159 [TP t to be . Interestingly. assignment of a ␪-role to an element marked with quirky case obliterates the quirky specification previously established. it enters into an agreement relation with the pronoun and its ␾-features receive default values (cf. The latter scenario can be illustrated by dialects of Spanish that allow embedded quirky morphology on the controller of some obligatory control verbs. as seen in (89e). [89b]). the matrix light verb is also a quirky-case assigner. deletion of the idiom specification in (85b) is arguably triggered by interpretability at the C–I interface. In particular. . after the Infl probe associated with adjectival predicates is merged (cf. ]]]]]]] g. ]]]]]]] h.20 Thus. in instances of ␪role assignment to an element previously marked with quirky case (cf. for instance.4 Obligatory control and morphological case f.DFLT] After the pronoun merges with the adjective in (89a). it receives quirky case and. the issue is a morphological one: does the morphology of the grammar in question allow preservation of quirky-case specification when a new ␪-role is assigned? It is not inconceivable that different grammars may have opposite answers and even different answers depending on specific quirky values or ␪-roles. as discussed by Boˇskovi´c (1994) based on work by Gonz´alez (1988. 1990). By contrast. In (90) below. this at face value appears to show that the controller of the agreement cannot be a trace/copy of . the MTC can handle the contrast between raising and control perfectly well. (90) Spanish (Gonz´alez 1988. Icelandic (Boeckx and Hornstein 2006a): J´on vonast til að leiðast ekki einum Jon. [91a]) and (quirky) accusative case (cf.1 Icelandic The second type of challenge posed to the MTC coming from Icelandic involves control configurations in which embedded floating quantifiers and secondary predicates display case-agreement morphology that at face value seems to be independent from the controller in the matrix clause (see Landau 2003. [91b]).NOM hopes to to be-bored not alone. Amovement in control is motivated by ␪-considerations and. by making use of fairly standard assumptions regarding quirky case. A-movement in raising is related to ␾-agreement and is therefore oblivious to ␪-relations that the relevant DP may have participated in. (91) a. as illustrated in (92) below. it may in principle be sensitive to ␪-related issues. the matrix subject bears (structural) nominative case (cf. which is the quirky case assigned by the embedded verb.160 Empirical challenges and solutions possibilities may be accommodated by morphology undoubtedly underlies part of the variation found in speakers’ judgments regarding quirky case.4. If the secondary predicates in (91) mismatch the case of the matrix subject but exhibit the quirky agreement licensed by the embedded verb. Sigurðsson 2008.2 Apparent case-marked PROs 5.4.ACC wanted not to to be-bored alone. and Bobaljik and Landau 2009). 5. In the sentences in (91) below.DAT ‘Jon hopes not to be bored alone’ Bjarna langaði ekki til að leiðast einum Bjarni. therefore.DAT ‘Bjarni wanted not to be bored alone’ The reason why (91) seems intriguing from the perspective of the MTC is that floating quantifiers and secondary predicates agree in case and ␾-features with the nominal expression they are associated with. for instance. 1990): A Juan le quiere gustar Marta To Juan CL.DAT wants like Marta ‘Juan wants to like Marta’ In sum. the difference between raising and control with respect to the preservation of quirky-case morphology lies exactly where the MTC would lead us to look. By contrast. b. In fact. but the secondary predicate in the embedded clause shows up with dative case.2. 5. Rather. The derivation in (91a). φ:SG. Agreement between T and DP + movement: [TP J.DFLT a´ fundinn to meeting-the ‘The brothers were both invited to the meeting’ Again.MASC] ]] e. φ:SG.[case:?] alone[case:?. is as sketched in (93) (with English words): (93) a. Movement and θ-assignment: [vP J.PL invited.MASC] :SG.PL elected ´ı atj´ornina to board-the ‘The brothers were not both elected to the board’ b. appearances are misleading and a close look at the relevant derivations promptly reveals the source of the case dissimilarity between the matrix subject and the embedded secondary predicate.MASC.MASC] ]]]]] h. ␾:?] ] b.MASC] ]]] . Merger between DP and the secondary predicate: [J.[case:DAT] alone[case:?. Merger of the verb + assignment of quirky case: [VP be-bored [J.MASC] :SG.MASC] ]]]] f. (92) a.PL was both. Icelandic (Sigurðsson 2008): Braeðurnir voru ekki b´aðir kosnir Brothers-the. φ:SG. Concord: [VP be-bored [J. 2008).[case:?] alone[case:?.MASC.[case:?] [v’ v [VP hope [CP t not to be-bored t alone[Case:DAT. Braeðrunum var b´aðum boðið Brothers-the.PL were not both.MASC] :SG.[case:?] [v’ v [VP hope [CP t not to be-bored t alone[Case:DAT. DAT. φ:SG.4 Obligatory control and morphological case 161 the matrix subject. Merger of a φ-complete probe: [TP T[␾:?] [vP J.MASC] :SG.NOM. Applications of merge and move: [vP v␪ [VP hope [CP C [TP J.[case:DAT] alone[Case:DAT.MASC.MASC] ]]]] ↑ g. c. the agreement on the secondary predicate should be determined by a case-marked PRO (see Sigurðsson 1991. ] ]] d.MASC] :SG. Concord: [J.NOM.[case:DAT] not to be-bored t ↑ ]] alone[Case:DAT. φ:SG. for instance.DAT.MASC.[Case:NOM] [T’ T[φ [φ:3SG] :3SG] [vP t hope not to be-bored t ↑ alone[Case:DAT. MASC] :SG.4. the surface form of a given DP depends on the last caseassigning/valuing head it interacts with. The crucial step is the next one. Upon merger. the previous quirky-case value is overwritten by the new one.21 Notice that the derivation of (91b) is essentially identical to the derivation of (91a).1. which provides evidence for the view that case must be assigned/valued rather than checked. assignment of a ␪role to an element marked with quirky case obliterates the previous quirky-case value. φ:SG. ]] Assignment of quirky case: [VP be-bored [B. φ:SG. merger between the secondary predicate and the DP in (93a) allows the ␾-features of the secondary predicate to be valued (cf.MASC] ]]] Let us pause for a moment to reconsider the valuation of the uninterpretable features of the secondary predicate in the derivations outlined in (93) and (94). ] ]] Concord: [VP be-bored [B. b. Movement of ‘Jon’ to the embedded [Spec.MASC] :SG.4. when it assigns its external ␪-role to the moved subject. .MASC] :SG. c. the secondary predicate get its ␾-features valued by the 21 Just as we saw in section 5. [93b]).[case:ACC] [vP v [VP wanted [CP t not to be-bored t ↑ alone[Case:DAT. That being so.[Case:ACC] [T’ T[φ [φ:d :df lt] [vP t hope not to be-bored t ↑ alone[Case:DAT. [93c]). Agreement with a φ-complete T + movement: [TP B. it values the case feature of ‘Jon’ as dative (cf.[case:?] alone[case:?. (94) a. The only relevant difference is that the matrix light verb is a quirkycase assigner (see footnote 20) and.162 Empirical challenges and solutions Assume that concord between floating quantifiers/secondary predicates and nominal expressions takes place under mutual c-command. φ:SG. as sketched in (94) (with English words).[case:DAT] alone[Case:DAT.MASC] ]] Movement and θ -assignment + quirky valuation: [vP B. Once a quirky-case assigner is introduced in (93c). d. This is what happens in (93f) after ‘Jon’ moves to receive the external ␪-role of the matrix light verb. Merger between DP and the secondary predicate + concord: [B. TP] in (93e) strands the secondary predicate. [93h]). which in turn allows the secondary predicate to have its case feature valued via concord (cf.MASC] ]]]] e.[case:DAT] alone[case:?. valuing (and deleting for LF purposes) the uninterpretable features of the floating quantifiers/ secondary predicates. As discussed in section 5. Finally.1. the moved subject agrees with a finite T and surfaces as nominative (cf. [93d]). [case:?] to go [t alone[case:?. We propose that.PL elected ‘The brothers disliked not being both elected’ b.NOM. Its case feature. Bearing this in mind. regardless of the case specification of the controller.4 Obligatory control and morphological case 163 corresponding features of the nominal expression it associates with. in such circumstances.[case:?] alone[case:?. ↑ ] ]]]]] Movement and θ -assignment + quirky valuation: [vP O. the case of the secondary predicate or floating . b.DAT. Interestingly.ACC longed to go alone. ]] Applications of merge and move: [vP v␪ [VP longed [CP C [TP O.5. vP].[case:ACC] [v’ v [VP longed [CP C [TP t to go [t alone[case:?. The ␾-features of the relevant nominal expression are interpretable and hence valued at every derivational step. on the other hand. ↑ ] ]]]] In (96) ‘Olaf’ gets its case value after it moves to the matrix [Spec. ´ ´ı veisluna Olaf langaði að fara einn Olaf. is uninterpretable and is unvalued upon merger. leaving the stranded secondary predicate with its case feature unvalued.[Case:ACC] [T’ T[φ [φ:d :df lt] [vP t longed t to go [t alone[case:?. [93c–d]) and [94b–c]). c. ↑ ] ]]]]]] Agreement with a φ-complete T + movement: [TP O. the floating quantifier/secondary predicate surfaces as nominative. Only after the case feature of the nominal expression is valued can it value the case feature of the secondary predicate.PL liked ill to be not b´aðir kosnir both. case valuation of the controller takes place after it leaves the floating quantifier/secondary predicate stranded. as exemplified in (95). Merger between DP and the floating quantifier + concord: [O.MASC. but not its case feature.NOM to party-the ‘Olaf wished to go alone to the party’ In absence of quirky-case assignment in the embedded clauses of (95). d. (96) a. this happens after the embedded predicate assigns quirky case to the nominal expression (cf.MASC. This has a straightforward explanation. for instance. let us now consider instances where there is no quirky-case assignment in the embedded clause. Consider the simplified derivation of (95b) given in (96) below. (95) a. In the derivations discussed. Icelandic (Sigurðsson 2008): Braeðrunum l´ıkaði illa að vera ekki Brothers-the. NOM to be hired. the default value for case in Icelandic is nominative. It is worth noting that. as illustrated in (97). it is reasonable to analyze nominative as the unmarked case in Icelandic.NOM hopes he/Eric. the embedded subject does not have its case valued and the derivation crashes. reproduced in Bobaljik and Landau 2009) ´ bað Mar´ıu ´ ´ ∗ Asta) Eg [að (∗ hun/ fara ein andað] I asked Maria.MASC.” . in order to be licensed. the MTC has a very simple explanation for why control infinitives in Icelandic cannot license an overt subject. However. and Nunes in press). why are the sentences in (97) out? When all is weighed. adapted in Bobaljik and Landau 2009) ´ Eg vonast til [að (∗ m´er/∗ J´oni) verða hj´alpað] I. the element bearing quirky case needs to agree with a ␾-complete probe and there is no such element within the embedded clause.NOM. in (97c) the embedded subject is valued with quirky case.SG there ‘I asked Maria (for her/Asta) to go there alone’ (Thr´ainsson 1979. b. If this is so. (97) a.SG ‘John hopes for him(self)/Eric to be hired’ (J´onsson 1996. Boeckx.4. as in many other languages. Sentences such as (97) are in fact quite problematic for approaches that take the morphological facts reviewed above to indicate that control infinitives in Icelandic license a PRO marked with structural or quirky case. By contrast.164 Empirical challenges and solutions quantifier is assigned a default value in the morphological component (see Boeckx and Hornstein 2006a. adapted in Bobaljik and Landau 2009) Since there is no local ␾-complete probe in the embedded clause of (97a) and (97b). PRO is specified 22 This is in consonance with Sigurðsson’s (2008: 419) observation that “Icelandic is unusual in overtly marking many of its nominatives in morphology. Hornstein. ∗ Icelandic: J´on vonast til [hann/Eir´ıkur að verða r´aðinn] Jon. The reasoning above is quite simple: no case licensing.1). As we have already seen when discussing default specification for passive morphemes (see section 5.NOM hope for to me/Jon.22 hence the nominative specification on the floating quantifier and secondary predicate of (95) irrespective of the case value of the controller.NOM go alone. Maling. the derivation of (97c) also crashes. At the end of the day. hence.ACC to she/Asta. Even so.FEM. and Thr´ainsson 1985. no convergence.NOM. c. contrary to Sigurðsson (2008) and what Bobaljk and Landau (2009) claim. it seems that approaches that take control infinitives to license a PRO marked with “regular” case are very similar to the null-case approach.DAT be helped ‘I hoped (for myself/Jon) to be helped’ (Zaenen. ∗ Við b´aðum hana að verða boðna We. as illustrated in (98e). Recall that. let us consider a final set of data.23 He also notes that. .GEN to be calm. etc.DAT found pleasurable to be the-first-one. where the numbers in parentheses indicate how many out of 15 informants in his survey judged the relevant agreement pattern as OK. where the embedded predicate assigns quirky dative.ACC to go just alone.5. Importantly.NOM asked her.SG eina alone. when quirky case assignment is available in the embedded 23 See Sigurðsson (2008) for relevant figures regarding judgments classified as ? or ∗ .ACC to party-the ‘She asked Olaf to just go alone to the party’ ´ b.ACC.GEN ‘We shouted to Olaf to be calm’ a. (98) Icelandic (Sigurðsson 2008): ´ ´ı veisluna (12/15) H´un bað Olaf að fara bara einan She. the value of the quirky case.). [98d]). for example. Sigurðsson (2008) points out that. as respectively exemplified with structural and quirky accusative in (98a) and (98b) below. Olafi fannst gaman að vera fyrstum (3/15) Olaf. although it is sensitive to a variety of as yet poorly understood factors (the type of predicate. Olaf langaði að vera fyrstan (2/15) Olaf. Við k¨olluðum til Olafs að vera r´olegs (0/15) We shouted on Olaf. case matching with the controller is marginally possible.FEM. that it is easier for controllers bearing structural accusative case to transmit their cases to a downstairs predicate than it is for controllers bearing quirky case.4 Obligatory control and morphological case 165 as being licensed by a peculiar kind of case which can license no other nominal expression. For the sake of completeness. [98c]).ACC ‘Olaf wanted to be the first one’ ´ c.ACC. He notes.ACC to be invited.NOM asked Olaf.DAT ‘Olaf was pleased to be the first one’ ∗ ´ d. transmitting quirky genitive is not an option (cf. idiolectal variation.SG ‘We asked her to get invited alone’ The sharp contrast between (98a) and (98e) is exactly what our approach predicts. e. although transmitting quirky dative is marginally possible (cf. with nominative specification for embedded agreeing elements. case transmission never applies if the embedded predicate has a quirky-case property. in addition to the “central facts” discussed above.ACC longed to be the-first-one.FEM. In a nutshell.24 In sum. [93d] and [94c]).’ as illustrated in (i). It is then plausible to assume that this leaves the door open for competing morphological strategies (default nominative assignment or longdistance case copying). [98a]) – vs. Case transmission also appears to be sensitive to the presence of an intervening DP. Hence. including sensitivity to the type of case (structural – cf. Landau (2007) also reports a great deal of variability in situations of case transmission in Russian. assignment of a new ␪-role to an element marked with quirky case obliterates the quirky-case value previously specified. Hornstein. San Martin’s argument 24 See Boeckx. reporting data from Andrews 1982): Þeir telja hana hafa lofað honum að vera They believe her. quirky – cf.SG ‘They believe her to have promised him to be good’ .DAT to be g´oð/∗ g´oða good. the dissimilarity between the features of the controller and an embedded floating quantifier or secondary predicate is not an insurmountable problem for the MTC. the nominal expression bearing quirky case is able to enter into agreement/concord relations and value case and ␾-features of the agreeing elements around (cf. its case realization will be determined from that point on in the derivation. and Nunes (in press) for further discussion.FEM. 5. which San Martin (2004) has claimed runs against the expectations of MTC in that it also appears to involve a casemarked PRO (see also Landau 2006). once a quirky-case-marked element moves to a ␪-position. with no connection with the quirky agreement it had previously triggered. It in fact follows very naturally as a by-product of the dynamics of the derivation.2 Basque Let us finally turn to control in Basque.2. [98b]) and to the specific case value. If case transmission is indeed a morphological process. Interestingly.166 Empirical challenges and solutions clause. Once quirky cases are tied to ␪-roles. Thus.FEM.NOM. as the contrast between dative and genitive in (98c) and (98d) illustrates. the corresponding elements in the embedded clauses of (98a–d) do not have their features valued in the syntactic component.SG/ACC.ACC have promised him. case transmission is blocked in sentences such as (98e) because the relevant elements have already had their features valued. as it is blocked under ‘promise. it would not be surprising that it would be subject to a variety of morphological factors. aside from the murky status of case transmission (a status it has under any approach to control). Landau notes that the presence of an intervening argument also blocks case transmission. (i) Icelandic (Ussery 2008. By contrast. as is the case in (98e). contrary to what is frequently claimed.4. 3ABS. Assuming that the absolutive case on the controller in the matrix clause disqualifies it as a potential candidate to be the element bearing ergative case. Thus. among others.ERG bread. This in turn indicates that ergative case must have been assigned in the embedded clause. Jon etxera joan da Jon. another with ergative case. a zero morpheme). we believe that it just shows how control structures such as (100) can be handled in a PRO-based approach. consider the control structure in (100) below.ABS Mary. (99) a. finally. if a given clause has a single argument DP. Marantz (1991).DAT present. given that one argument is realized as dative. it is marked with absolutive case (morphologically. the case patterns that arise in Basque are not related to the nature of the predicate involved. Jonek ogia erosi du Jon. (100) Joni [Øi Mariari ogia ematen] Jon.3ABS ‘John has tried to give bread to Mary’ San Martin’s argument is indeed very ingenious.DET.3ABS ‘John has gone home’ b.DAT bread.ABS house.5. thereby accounting for the presence of dative case. As she observes. not that the MTC cannot account for such constructions.4 Obligatory control and morphological case 167 is the following.ERG Mary. .NOMIN. one is marked with absolutive case.ALL go AUX.ABS buy AUX.1ERG ‘I have given the present to Mary’ With this in mind. and Jackendoff (1987). but rather to the number of argument DPs. San Martin concludes that the embedded clause of (100) must involve a PRO marked with ergative case. Nik Mariari oparia eman diot I.3ERG ‘John has bought bread’ c. The embedded clause of (100) displays the case pattern associated with three arguments.ABS give AUX. if it has three argument DPs. However.DET. Maling. and Harley (1995).INN saiatu da try AUX.25 This is respectively illustrated in (99).DET. one is marked with absolutive case and the other with ergative case.ABS give. Basque has the canonical morphological profile of an ergative language in the sense 25 This is reminiscent of the morphological case-assignment mechanisms proposed in Yip.3ABS. if it has two argument DPs. and the remaining element with dative case.3DAT. [vP John[case:?] [v’ v [tried [t give bread[case:ABS] Mary[case:DAT] ]]]] ↑ d. More importantly. the presence of three arguments activates the three case specifications and the case of ‘John’ is specified as ergative. Further case computations then specify ‘John’ as absolutive as it is the sole argument DP in the matrix clause. German.1ABS ‘I have tried to buy the present’ b.2. Basque arguments behave syntactically on a par with what is found in nominative–accusative languages. Nii [Øi etxera joaten] saiatu naiz I. Basque is much closer to Icelandic than. And as far as its inherent case system goes (which. vP]. it receives another ␪-role and its previous case specification is obliterated. we have witnessed case mismatch between the controller and the controllee (its trace/copy) which arises in the course of the derivation. say. following Laka [2006]. But morphology aside.1 to account for apparent case-marked PROs in Icelandic extends straightforwardly to Basque. after ‘John’ moves to the matrix [Spec. proceeds along the lines of (102) (with English words). in the domain of control. we take to include ergative and dative). hence.ALL go.INN try AUX.1ABS ‘I have tried to go home’ Thus. However. (101) a. [vP v␪ [tried [John[case:ERG] give bread[case:ABS] Mary[case:DAT] ]]] c.ABS house. as illustrated by the agreement on the auxiliary in [99]–[101]). [TP John[case:ABS] has[3ABS] [vP t [v’ v [tried [t give bread[case:ABS] ↑ Mary[case:DAT] ]]]]] In (102a).INN try AUX. as it allows inherently case-marked elements to function as regular subjects (and even enter into ␾-agreement with a full ␾-probe.ABS present. Nii [Øi oparia erosten] saiatu naiz I. For example.ABS buy. ‘John’ is generated in the embedded clause.NOMIN. The derivation of (100).DET. [John[case:ERG] give bread[case:ABS] Mary[case:DAT] ] b. . as illustrated in (101) below with subjects of transitive and unaccusative predicates respectively. the mismatch arises exactly where predicted by the MTC: when an inherently case-marked element receives an additional ␪-role.4. it is subjects that are invariably controlled.168 Empirical challenges and solutions that transitive objects and intransitive subjects are equally marked absolutive. the analysis we proposed in section 5. (102) a. Again. for instance.NOMIN. [ persuaded+v [Mary tpersuaded [John to . control shift. The argument form is impeccable. the conclusion reached by some is that it too must be false. control shift. .3). (103) a. John said that Mary tried [PRO to wash herself/∗ himself] b. . . The first involves control into the complement of verbs like ‘promise. John persuaded Mary [PRO to wash herself/∗ himself] The MTC captures the empirical virtues of the minimal-distance principle via minimality (see section 3. Culicover and Jackendoff 2001 and Landau 2003). ]] ↑ ∗ b. and the second involves cases of “control shift. [John said that [Mary tried [t to . [105]). As the MTC assumes the minimaldistance principle in the guise of minimality. [John persuaded+v [Mary tpersuaded [t to . ]]] ↑ ∗ (105) a. many (though equally impressed with the intimate connection between the MTC and the minimal-distance principle) have concluded that the MTC is fatally tainted empirically precisely because of this tight relation. Conversely.5 The MDP. [ said that Mary tried [John to . Whether the facts are fatal is less clear. ]]] ↑ OK b. .g.” as illustrated in . . However.’ as illustrated in (106) below. and the logic of minimality 5. as we will argue. and the logic of minimality Rosenbaum (1970) proposed the minimal-distance principle to account for why. There are well-known counter-examples to Rosenbaum’s minimal-distance principle that (some claim) demonstrate that the minimal-distance principle is incorrect.. ‘Mary’ can be the controller in (103). typically. As this violates (relativized) minimality.’ as shown in (104) below. as there is no intervening DP (cf. . ‘John’ cannot be an antecedent of PRO. ]]] ↑ OK We believe it to be a virtue of the MTC that it so elegantly derives Rosenbaum’s minimal-distance principle. . .5 169 The minimal-distance principle. the antecedent of an obligatorily controlled PRO is the most proximate nominal expression.4. (104) a.1): if obligatorily controlled PRO is a residue of A-movement. The putative problems for the minimal-distance principle come in two varieties (see e.5. then moving ‘John’ from the embedded-subject position in (103) to the matrix-subject position traverses the c-commanding intervening nominal ‘Mary. as illustrated in (103) (see section 2. this learnability feature of verbs like ‘promise’ has received less attention than the acceptability of (106) (for some speakers). a standard assumption at least since Aspects. It appears that there are speakers for whom these cases are never considered acceptable. they are far too systematic to be dismissed as ‘highly marked’ exceptions. control by the matrix subject should be unavailable. dismisses the relevance of the acquisition puzzle. Rosenbaum (1967) had already observed (citing C. We take it that a complete account of these cases has three parts: (i) that the cases in (103) are the central examples of complement control (as Rosenbaum originally proposed). . John asked Mary [PRO to be allowed to shave himself/∗ herself] There has been an animated discussion regarding how compelling these facts are.’ As this seems to involve a minimality violation under the MTC. for example.” Boeckx and Hornstein (2004) observe that the reasoning behind this line of argument is flawed if one accepts that theories of UG aim to provide an answer to the logical problem of language acquisition (Plato’s Problem). In both (106) and (107b) ‘John’ appears to control PRO. John asked Mary [PRO to shave herself/∗ himself] b. claiming that “although cases of type [106] are not as common as those of type [103b]. (106) John promised Mary [PRO to wash himself] (107) a. after all. as emphasized by Boeckx and Hornstein (2004). which appear to be acquired quite straightforwardly. Why.27 As Rosenbaum correctly notes. (ii) that some speakers find (106) acceptable. describe this type of sentence as “only marginally grammatical. and Partee (1973: 536). Schachter. this is a problem for those who consider these cases as identical to the standard cases in (103). the promise-examples are not uniformly deemed acceptable. and (iii) that some speakers either never come to admit cases like (106) or take a 26 Stockwell. for example. Chomsky 1969) that control cases like (106) are mastered rather late in the acquisition process. despite the fact that ‘Mary’ surfaces between PRO and ‘John.26 Importantly. if mastered at all. For example. do the acquisition profiles of sentences like (106) differ so significantly from those in (103) if they are unexceptional cases of control? Curiously.” 27 See Courtenay (1998) for discussion. The interesting acquisition profiles that verbs like ‘promise’ have constitute prima facie evidence that all is not quite standard with these cases from a grammatical point of view and that whatever evidence they provide against the minimal-distance principle requires some further massaging. Landau (2003: 480).170 Empirical challenges and solutions (107b). (108) [DP1 [persuade [DP2 [PRO∗ 1/2 to go home]]]] The movement from the position occupied by PRO (understood as an Atrace) to the position occupied by DP1 is prohibited if we assume that DP2 triggers a minimality violation. 5. Rather. Some may take this to indicate that controller selection is not a lexical property of the embedding predicate but a compositional fact about the embedding predicate coupled with the composed semantic contribution of the embedded sentence. In fact. if selection is a local head-to-head relation (the standard assumption). these works do not explain why these readings arise or why these factors are relevant. This holds if DP2 “intervenes” between DP1 28 This seems to be the position of Culicover and Jackendoff (2001) and van Craenenbroeck. There is currently no good account for why control shift obtains.28 However. In section 5.5. However. For concreteness. consider a persuade-structure like (108). Rooryck. it seems to us that it is not clear what the control reversal facts in (107) tell us about the adequacy of the MTC. control shift is incompatible with standard analyses of control in which the controller choice is catalogued as a (diacritical) fact about the selection properties of the embedding verb.5. we still await a (non-ad hoc) explanation of how these different readings arise.1 below. for instance. they basically reiterate the observed facts and factors by noting that the isolated factors are dispositive and that they contribute in some way to the observed interpretations. and van den Wyngaerd (2005).5 The MDP. 29 There is a considerable literature discussing various factors involved in control-shift interpretations that build on Farkas (1988). we outline an approach that covers all three data points. For instance.1 Control with promise-type verbs Let us review exactly how the MTC derives the MDP as a special case. We attempt to do so in section 5. it transcends the reaches of selection. The putative counter-example in (107b) is equally problematic.5.5.2 below. An adequate theory of these cases should tell us why control shift occurs in just this narrow range of cases.29 Again. to allow the verb ‘ask’ to select its object as the controller in (107a) but its subject as the controller in (107b) requires allowing it to “see” the verb of the embedded clause: if it is something like ‘allowed’ then the subject controls. otherwise the object (see Farkas 1988). However. . Whatever goes on in control shift cases. and the logic of minimality 171 long time to acquire them. What is clearer is that the reversal only applies in certain stylized situations in which the object of ‘ask’ authorizes the embedded event in some way. control shift. the embedded verb is just too remote from ‘ask’ to be visible and so cannot be exploited in this way to determine the “correct” controller. though seldom noted. repeated below in (109a). the preposition ‘to’ may surface in other syntactic frames associated with ‘promise. the apparent direct object of ‘promise’ in (109a) should then be mapped into the complement of a preposition. [John promised [XP X Mary] [t to wash himself]] ↑ OK There is indeed a considerable amount of evidence that points to the conclusion that ‘Mary’ in (109) is not the complement of ‘promise. (109) a. 2003). If any of these clauses fails to obtain. should not count as an intervener if it fails to ccommand the embedded subject. and this leads us to the second set of 30 See e. Given this logic. and DP1 c-commands DP2 .. Landau (1999. Given (a non-relativized version of) Baker’s (1988. . The question is whether it makes sense to suppose that there is an extra layer of structure in promise-cases preventing the object from counting as an intervener.’ but the complement of a null preposition.30 (110) [John vowed/committed [PP to Mary] [t to wash himself]] ↑ OK ‘Vow’ in particular is of interest as it is semantically very close to ‘promise’ and so it is not unreasonable to suppose that the thematic role of the nominal argument of the two verbs is identical. as sketched in (109b). DP2 intervenes if it c-commands PRO. John promised Mary [PRO to wash himself] b.g. My promising to/∗ of Mary to leave In fact. ‘promise’ is often classed with other subjectcontrol verbs whose nominal complement is preceded by the preposition ‘to.’ as shown in (112) below. [109b]). despite the fact that ‘of’ is what normally precedes structural objects within nominalizations: (111) a.’ as illustrated in (110) below. where ‘Mary’ is the complement of a null head. the object of control cases involving ‘promise’ like (106). John’s promise to/∗ of Mary to leave b. thus obviating any minimality problems (cf. then DP2 is not an intervener and movement of DP1 from the PRO position does not violate minimality. First. DP1 c-commands PRO. 1997) uniformity of theta assignment hypothesis (UTAH).172 Empirical challenges and solutions and PRO. This conclusion is bolstered by the oft-noted observation that the nominalization of ‘promise’ requires ‘to’ rather than ‘of’ before the nominal complement. as shown in (117) below. ∗ Whoi did you promise ti to leave the party? c. as shown in (ii). Baker (1997) argues that the goal of double-object constructions such as (116c) patterns like indirect objects because it is actually a prepositional complement. despite the lack of an overt preposition. as opposed to the object of standard object-control verbs. (112) I didn’t promise this to Mary (113) a. again paralleling double-object constructions (cf. In 31 Notice that sentences like (ia) below with an overt preposition are only weakly unacceptable. wh-movement improves quite a bit. ?John promised to Mary to leave the party early b. [ib]). ∗ You gave ti a book [every man that you met]i (116) a. (114) a. and secondary predicates (cf. when the preposition is overt. This is exactly what we are saying regarding the object of ‘promise’ in constructions like (109a). a fronted thematic goal surfaces without its preposition. the nominal object of ‘promise’ in constructions like (109a) patterns like the goal of double-object constructions and not like the object of control constructions involving verbs like ‘persuade. ?John gave to Mary a book (ii) (?)To whom did John promise to leave the party early? .5 The MDP. John persuaded Mary1 to go to the party undressed1 b.5. the object of ‘promise’ and the goal of a double-object construction disallows wh-movement (cf.’ Thus. ∗ John gave Mary1 a book undressed1 These secondary-predication structures are particularly interesting for it is well known that PPs cannot be subjects of such predicates. (i) a. [114]). Interestingly. John gave Mary a present Importantly. that is. ∗ John promised Mary1 to go to the party undressed1 c. John gave a present to Mary b. ∗ Whoi did you give ti a book? (115) a.31 heavy NP shift (cf.’ The two syntactic frames of ‘promise’ seen in (109a) and (112) resemble what we find with standard double-object constructions in English such as (113). ∗ You promised ti to leave [the party every man that you met]i c. [115]). Whoi did you persuade ti to leave the party? b. and the logic of minimality 173 facts that indicate that ‘Mary’ in (109a) is not the theme/patient complement of ‘promise. [116]). control shift. You persuaded ti to leave [the party every man that you met]i b. Note that the problem posed by (i) is not different from what we find in (ii). ∗ John promised heri to visit Maryi b.32 (117) ∗ John gave a book to Mary1 undressed1 (118) ∗ John vowed to Mary1 to leave the party early undressed1 In sum. assume Kitahara’s (1997) and Boeckx’s (1999) proposal regarding principle-C effects. Alternatively.3. Thus. if something along these lines is correct. according to which ‘to’ in (ii) is a marker of inherent case and inherently case-marked elements do not induce A-intervention. the pronoun undergoes covert movement to a position from where it c-commands the clausal complement. [ promised [P Mary] [John to donate money to the library fund]] ↑ OK Moreover. Again. the only difference being the overt preposition in the latter case. if the object of ‘promise’ in sentences like (119a) below is actually tucked inside a PP phrase (see footnote 32). despite the fact that the nominal within PP induces a principle-C effect with respect to material inside the clausal complement. accordingly. with an overt preposition.33 (119) a. it should be noted that. the object of ‘promise’ behaves just like the object of ‘vow’ (cf. as illustrated in (ia) below. As the reader can see. at the point when raising takes place in (iia). the present analysis does not require the postulation of null prepositions in English.. John promised Mary to donate money to the library fund b. both approaches make the same empirical predictions regarding the material discussed here. with a null preposition.174 Empirical challenges and solutions effect.g. However. The matrix object does not c-command the embedded subject and so does not intervene for purposes of minimality. this null-preposition hypothesis supplies the ingredients for an answer as to why some speakers never seem to allow sentences like (119a) or 32 Strictly speaking. generating the subject-control reading is expected to be possible. . this suggestion carries over to (i) straightforwardly. there arises the question of why the matrix object induces principle-C effects with respect to material inside the embedded clause. 33 If movement of ‘John’ in (119b) is allowed because the matrix object does not c-command it. ∗ John vowed to heri to visit Maryi (ii) a. (i) a. we will keep using representations like (119b) in the discussion that follows and. For purposes of presentation. as Baker [1997] proposes). later on in the derivation.’ just like ‘give’ in double-object constructions. renders an overt preposition. the pronoun is within PP and does not induce intervention effects. we can exploit the account of (ii) suggested in section 5. e. it applies to (i) in a straightforward fashion. It suffices for what we need here that ‘promise. where raising over an experiencer PP is allowed in English.’ null at some point of the derivation (perhaps via incorporation. [118]). as far as we can see. movement of embedded subject over the matrix object is a licit operation. thereby inducing principle-C effects (see also Boeckx 1999). ‘to. and (ib). [Johni seems to her [ti to be nice]] b. as sketched in (119b).2. ∗ [It seems to heri [that Maryi is nice]] Kitahara (1997) has proposed that. It should not be surprising to find that these methods occasionally do not apply uniformly across speakers and that what some speakers catalogue as theme/patient others consider oblique. the data used above to motivate its presence is too exotic to be considered part of the primary linguistic data. putting in an overt preposition will signal that a DP is not a direct object of the verb. One way to alleviate the obscurity is by syntactic means. if such overt evidence is absent. What is true for ‘promise’ also holds for ‘threaten. if one takes a coarse-grained view of ␪-roles and adopts a Dowty-style protorole analysis to underwrite the projection of proto-roles to syntactic positions. .’ another putative subjectcontrol predicate. Given the meaning of ‘promise. Were this to happen. Those that do not have enough of such properties but have goal/path/location properties will be treated as oblique and mapped to the object of some preposition. then there will be cases where the relevant proto-role may be obscure (and/or ambiguous). e.5.34 The question that then arises is whether ‘Mary’ in cases like (119a) has more oblique.or more theme/patient-like consequences. It should be no surprise to discover that null prepositions may be difficult to pin down. For these speakers. It might not be too great a stretch to assume that some speakers never receive the relevant input in sufficient amounts. 36 See Landau (2003). Thus. then arguments will be projected to grammatical-function positions on the basis of their thematic proto-role consequences (see Dowty 1991). the DP associated with the promisee would be treated as a direct object and minimality would block movement across it. then ambiguity will be rife and more subtle calculations of semantic consequences will be needed to settle matters. 35 It is worth noting that the authors are acquainted with some speakers that cannot understand sentences like ‘John promised Mary to leave’ as involving subject control. Clearly. As Baker (1997) emphasizes.36 (120) John threatened Mary to kiss Sue 34 See Baker (1997) for one reasonable elaboration of these mapping principles.5 The MDP. and the logic of minimality 175 that they are late in getting to them. these sentences considerably improve with the subject-control reading if ‘to’ is inserted before ‘Mary’ (‘John promised to Mary to leave’). DP arguments that are internal to VP and that have sufficient theme/patient properties will be assigned to object positions. control shift..g. If one adopts a non-relativized version of UTAH (see Baker 1997). It appears that some speakers accept sentences like (120). There is a second confound here as well.35 However. the main evidence for the grammaticality of sentences like (119a) are actual instances of such sentences used in the appropriate context.’ it would not be surprising if a child concluded that the promisee was affected (in some suitable sense) by the proffered promise. the more proximate object blocking the more remote subject. The MTC requires that post-verbal DPs in sentences like (119a) not intervene in order for the subject control reading to be generated. In these latter cases. on the other hand. One way of being affected is to have one’s psychological state changed. Do promises alter one? Do threats? It is subtle questions like these that determine the thematic syntax of these constructions (on a Baker–Dowty approach) and that also determine whether a given DP will act as an intervener in an MTC account. and how they contrast with control constructions like ‘persuade. The evidence that such DPs act like objects of prepositions and not like direct objects is just what one would expect were the MTC correct. for instance. find (120) very unacceptable. In (121b). (121) a. John1 asked/begged/petitioned Mary2 [PRO2/∗ 1 to leave the party early] b. It seems reasonable to conclude from (120) that Mary’s state of mind might be affected by John’s action. the matrix 37 As in Landau (2000) and Culicover and Jackendoff (2005). for example. In fact. why they are not allowed by all speakers. Put tendentiously. we could go further. .176 Empirical challenges and solutions Some. The question on a Baker–Dowty account of ␪-role projection is what one takes the proto-role of ‘Mary’ to be in (120). while other verbs are not clear cut in their semantic consequences. If so.’ Our proposal above offers a sketch that addresses all three concerns and is compatible with the MTC.5. Consider (121). Note that the indicated licit indexation observes the minimal-distance principle. Of course. some may conclude that Mary is merely a recipient of a threat and so not primarily an affected object and so might map ‘Mary’ in (120) to an oblique position.’ ‘force’) are unproblematic. we should expect variation and we appear to find it. thereby permitting A-movement across it. John1 asked/begged/petitioned Mary [PRO1 to be allowed/permitted to leave the party early] The cases in (121a) are typical cases of object control. We noted that an account of subject control sentences involving verbs like ‘promise’ needs to explain how they can be generated. ‘Mary’ will be a proto-theme/patient in (120) and so be mapped to the syntax as a direct object.37 5.2 Control shift Let us now examine control shift. this is what the MTC predicts. including one of the present authors. The same cannot be said for more conventional approaches which stipulate the antecedents of PRO via ad hoc diacritics that annotate the argument structure of embedding verbs. It should not be surprising that the core cases (‘persuade. To conclude. in violation of both the minimal-distance principle and minimality (if PRO is understood to be a residue of movement).5. the DP is mapped to the complement position of a preposition. However. as illustrated in (122). is (124a) below. passive may matter and that the passive makes the source reading more readily available on the matrix object.5 The MDP. Control shift can be licensed without an overt ‘allow’/‘permit’ verb in the embedded clause. . Thus. control shift. Here we would like to propose that control-shift cases are in fact amenable to essentially the same kind of analysis proposed in section 5. (122) a. In the latter case.5. thereby allowing movement across it without any 38 In addition.38 (124) a. as illustrated in (124b). an optimal paraphrase of (122a). when ‘John’ controls. it has never been made clear why control shift is sensitive to such a tight interpretive constraint.1. John asked the guard to be allowed to smoke one more cigarette b. (i) contrasts with (123). Chomsky (1980) has noted that control shift is rather sensitive to the right embedded content. the subject-control reading of the sentences in (122) becomes infelicitous if the guard/manager has no authority to grant the request. one speculation is that the distinction active vs. and the logic of minimality 177 subject is a licit antecedent for PRO and it appears that the control relation is established over the matrix object. John asked the manager [PRO to pitch in the last game] These cases are ambiguous with either subject or object being potential controllers. ∗ John asked the guard to be allowed by the general to smoke one more cigarette Landau (2003: 480) observes that control shift “is sensitive to pragmatic factors” like “authority relations” and that languages differ in how sensitive they are to such factors. John asked the guardi to be allowed by himi to smoke one more cigarette b. the sentence becomes unacceptable. (i) ∗ John 1 asked Mary [PRO1 to have/get permission to shave himself] Why these cases contrast with (123) is unclear. with an overt instance of ‘allow’/‘permit. What happens in such cases is that a matrix object may be treated as a thematic object or as a thematic oblique (a “source” to be specific).’ (123) a. John asked the guard [PRO to smoke one more cigarette] b. for instance. However. In effect. When the object cannot be interpreted as the source of permission/authority. John asked the manager to be permitted to pitch in the last game Moreover. the only acceptable reading is as paraphrased in (123). However. then it should syntactically project as the object of a preposition.40 (126) [ asked [PP P Mary] [John to be allowed to leave]] ↑ OK In fact. it is understood as the source of the authority/permission to carry out the request whose content is specified by the embedded clause). and paths into one proto-role. that is. 39 Baker (1997: 108) proposes collapsing sources. locations. the interaction between control shift and control constructions involving ‘promise’ has a very curious result. where the control shifts from the object to the subject. goals. one reasonable possibility is that the thematic function of the matrix object is different in (121a) and (121b) and.” as seen in (124). So far. if the matrix object in (122). Consider the data below. as illustrated in (ia) below. cases where the control shifts from the subject to the object. if so. [119b]). (125) [John asked [PP P Mary] [PRO to be allowed to leave]] The structure in (125) should look very familiar. 40 As was seen with promise-cases.. as shown in (126) (see footnote 33).39 This source interpretation underlies the implicit understanding of the matrix object as the agent of the “allowing. for instance. ∗ John ∗ John asked her1 PRO to be allowed to visit Mary1 asked of/from her1 permission to visit Mary1 . and again this holds regardless of whether the preposition is null or overt.e. However. we have discussed standard cases of control shift. (i) a. as illustrated in (125) below. it gets a source interpretation (i. just as in the case of ‘promise. they should be projected into different syntactic positions under (a non-relativized version of) UTAH (see Baker 1997). b. however. promise-cases yield “reverse” control-shift effects. is a theme/patient then it will be a complement of ‘ask. However. as shown in (ib). The contrast between (121a) and (121b) with respect to the subject-control reading is in fact unexpected only if the matrix object of each sentence is associated with the same syntactic configuration.178 Empirical challenges and solutions violation of minimality. for example. the nominal inside the PP complement in control-shift structures induces principle-C effects with respect to material inside the clausal complement. See footnote 32 for two alternative accounts that are compatible with the MTC. ‘Mary’ in (125) does not c-command PRO.’ If.’ Let us consider the details. or does not c-command it at that point in the derivation where ‘John’ moves over it. It is what we proposed for matrix object of subject-control constructions with ‘promise’ (cf. As in promise-cases. Specifically. [127b]). the empty counterpart of a by-phrase (see section 5. TP]. however. as illustrated in (129). on the other hand. as sketched in (130).2.5 The MDP. the derivation of (128) parallels the derivation of passives involving object-control verbs such as ‘persuade’ (cf. ]]]] c. Our proposal has a straightforward account for the contrast between (127b) and (128). [20] in section 5. the derivation crashes. this entails that the matrix object must be buried in a PP layer in order for minimality to be observed. In other words. . it is then reasonable to assume that it cannot be passivized (cf. But if ‘Mary’ in (129) is the complement of a preposition. control shift.1): ‘Mary’ is generated in the embedded clause and moves to the specifier of ‘promise’ before landing in the matrix [Spec.’ where the object does not block movement of the embedded subject. if ‘Mary’ is generated as a standard verbal complement. (130) a. In turn. ‘Mary’ is the controller (and not the source of authority). passivization becomes licit. this amounts to saying that ‘John’/IMP must be generated in the embedded clause and move to the Spec of the passive morpheme.5. . [vP Maryi [v’ [IMP/by John] promised+-en␾ [VP ti tpromised ↑ OK [ti to be allowed . . In (127b). Either way. John promised Mary to leave b. Given the MTC. Conversely. and the logic of minimality 179 (127) a. [TP Maryi T␾ [vP ti [v’ [IMP/by John] promised+-en␾ [VP ti tpromised ↑ OK [ti to be . is the intended controller of the embedded subject.2. “Authority” restrictions translate naturally into a . In turn. being a DP complement. [VP Maryi promised␪ [ti to be allowed ti to leave]] ↑ OK b. ∗ Mary was promised (by John) to leave (128) Mary was promised (by John) to be allowed to leave Example (127a) is a typical example of subject control with ‘promise. (129) [ -en␪ [promised [PP P Mary] [IMP/John to leave]]] ↑ OK In (128). (127b) shows that the matrix object of (127a) cannot be passivized. . ‘John’ or IMP. ]]]]] Note that this account ties together control shift and the interpretive restrictions that characterize it.2). as seen in (128). it should be able to undergo passivization. it blocks movement of the embedded subject. Interestingly. if the embedded clause contains a predicate that makes a control-shift reading salient. the MTC explains why control-shifted contexts are semantically restricted to cases where the matrix object is thematically oblique. the preposition is required when the source is at the end of the clause but can be deleted if it is in medial position.5. In short. wh-movement is possible under standard object control (cf. [132c]) and double-object constructions (cf. John asked/begged (of/from) Mary permission to leave b. [132d]). which is more on a par with promise.1). (131) a. as illustrated in (131). In effect. These must be interpreted with Mary authorizing the leaving. the diagnostics operative in double-object constructions and promise-cases (see section 5. First. Second. Examples (131a) and (131b) seem to display something analogous to the dative alternation found in double-object constructions and promise-constructions (see section 5. if we are correct that in control-shifted clauses the matrix object is actually an oblique inside a PP. [132a]). Given this mapping. There is additional independent empirical evidence for the PP structure in (126).(cf.41 Given that these verbs already have structures with oblique source arguments.5.1) also extend to these configurations. it is reasonable that such roles can arise in the control-shifted readings as well. John asked/begged permission to leave of/from Mary c. just as in the control-shifted readings above. the matrix object does not function as an intervener for another DP that moves across it. Interestingly. John petitioned from/??of Mary a permit to leave Here the preposition is overtly manifest and the source reading on ‘Mary’ is forced. licensing subject control.180 Empirical challenges and solutions source role for the nominal object of the matrix clause. Source roles being oblique are syntactically mapped into PP configurations. [132b]). the main difference residing in the deletion of the preposition in the shifted cases. but strained under the control shift (cf. The fact that this occurs thus constitutes a further argument in favor of the MTC. then the purported empirical problem that control shift posed for the MTC is actually an argument in its favor and an argument against more standard accounts that decouple control selection from the minimal-distance principle or minimality. there are related forms that overtly show prepositions. 41 The deleted version may be slightly preferred. . For example. To put this another way: the MTC (coupled with UTAH) requires that the matrix nominal object receive an oblique thematic interpretation in order for subject control to be possible in such syntactic configurations. under the subject-control reading. the strongly preferred reading is one in which ‘who’ is the controller.5. though acceptable in the object-control case. unsure of herself. (134) a. John1 asked/begged [the guard]2 [PRO2/1 to smoke a cigarette] b. c. 181 I wonder [whoi John asked ti [PROi to leave early]] ??I wonder [whoi Johnk asked ti [PROk to be allowed to leave early]] ??I wonder [whoi Johnk promised ti [PROk to leave early]] ??I wonder [whoi John gave ti a book] This effect is also evident in those cases where either a theme/patient or a source reading can be attributed to the matrix nominal object. d. ??Who2 did John1 ask/beg t2 [PRO1 to be allowed to smoke a cigarette] In turn. John1 asked/begged Mary2 .5.and in control-shifted cases constitutes strong evidence against the empirical adequacy of Rosenbaum’s minimal-distance principle and. is ambiguous. However. when the matrix object undergoes wh-movement. (133b) patterns like the transparent control-shift construction in (133c). b. and the logic of minimality (132) a. [PRO1 to be allowed to sing at the gala] In sum. We have noted that the arguments depend on the premise that . which has the minimal-distance principle as a consequence. [PRO2 to sing at the gala] b. heavy NP shift is allowed under object control. This then provides independent support for the structure proposed in (125)/(126). the MTC. That is. Example (133a) below. [John1 asked/begged t2 [PRO2/??1 to stop working] [every employee that he met]2 ] b. the diagnostics used to argue that the matrix object of control constructions with ‘promise’ and the goal of double-object constructions are actually contained in an underlying PP also extend to the control-shift cases. Several have claimed that subject control over “objects” in promise.3 Summary Let us recap and conclude. unsure of herself. [∗ John1 asked/begged t2 [PRO1 to be allowed to stop working] [every employee that he met]2 ] Finally.5 The MDP. ∗ John1 asked/begged Mary2 . as illustrated in (134). but unavailable under control shift. secondary predicates modifying the matrix object are disallowed in the shifted reading. as shown in (135). (135) a. pari passu. control shift. for instance. Who2 did John1 ask/beg t2 [PRO2/??1 to smoke a cigarette] c. (133) a. 5. 4. where the embedded predicate requires a semantically plural subject but the controller is singular and must be interpreted as a member of the set of referents denoted by the embedded subject. it is far from clear that “a theory that does not derive the [minimal-distance principle] is. but in partial control the syntactic number of the controllee is determined by the controller. the MTC requires that this apparent direct object actually be analyzed as an indirect object at that point in the derivation where the movement applies and this requires that it get an oblique thematic interpretation given UTAH. In split control the controllee is also syntactically plural. Darcy at the end of Pride and Prejudice: “[our] feelings are quite different. The first case involves partial-control constructions like (137) below.” 5. (136) a. (137) [[The chair]i decided [PROi+k to meet at 6]] (138) [Johni proposed to Maryk [PROi+k to meet each other at 3]] Although semantically plural.182 Empirical challenges and solutions the matrix object DP in constructions like (136) is a direct complement of the matrix verb. In effect. just the opposite in fact. We have reviewed evidence pointing to precisely this conclusion and so these cases shift from being problems for the MTC to being evidence in its favor.2. In (137). In this section we discuss two cases where controller and controllee seem to be semantically distinct. this DP is actually embedded within a PP. the controllees of these constructions differ with respect to syntactic number (see Landau 2000). John1 asked Mary [PRO1 (to be allowed) to smoke a cigarette in her apartment] If. then it would not serve as an intervener and so minimality would not block movement across it. however. for instance. Pace Landau (2003: 481).6 Partial and split control In section 5. as illustrated in (138).” Our view is well expressed by Elizabeth Bennet’s reply to Mr. The second case involves constructions where the controlled subject has split antecedents. as the moving DP (the controller) receives a ␪-role that is incompatible with the quirky-case value previously assigned. John1 promised Mary [PRO1 to leave the party early] b. the controllee . we discussed cases where controller and controllee differ with respect to the agreement patterns they trigger and argued that the dissimilarity arises in the course of the computation. ceteris paribus. better off than one that does. like its controller.1 Partial control The fact that controller and controllee need not match in semantic number in partial-control constructions has been taken as a strong argument for a PRO-based analysis. If the controllee is a trace/copy of the moved element. Similarly. Given this overall picture. ‘each other’ can be licensed in the split-control structure in (138). lexical idiosyncrasies. In the next sections we discuss how partial and split control can be handled as special cases under the MTC. the controller should completely determine the referential properties of the controllee. how can it be interpreted as semantically plural in (137). exhaustive control receives a straightforward analysis under the MTC. whatever turns out to be the ultimate analysis of these constructions. If obligatorily controlled PRO is simply a residue of A-movement (i. but not in the partial-control structure in (139). or have distinct binders in (138)? It is worth observing that there is in fact a clear-cut divide between (standard) exhaustive control. on the other. This can be seen by the contrast between (138) and (139). Given that a plural anaphor must be licensed by a syntactically plural antecedent. for instance. (139) ∗ [[The chair]i decided [PROi+k to meet each other at 6]] Examples (137) and (138) clearly contrast with (standard) exhaustive-control constructions such as the ones in (140). 5.e. where the controller must be unique and have the same (semantic and syntactic) number specification as the controllee.1).6. and non-uniform judgments among speakers. and split and partial control. . on the one hand. The latter is much more amenable to crosslinguistic variation. (140) a.4. for example.and split-control constructions look especially challenging for the version of the MTC explored here. a copy of the antecedent interpreted as a bound variable).5. ∗ [[The chair]i managed [PROi+k to meet at 6]] b. there can be no split antecedents for the controllee because two distinct elements cannot move from the exact same position (see section 3. it should treat them as somewhat special when compared to exhaustive-control constructions. So. partial. where the feature [+Mer] characterizes group names. Landau (2004)..6 Partial and split control 183 is syntactically singular. analyzes the partial-control construction in (141a) along the lines of (141b). ∗ [Maryi expects Johnk to try [PROi+k to leave]] As discussed throughout this volume. 184 Empirical challenges and solutions (141) a.2. which license partial control. (ii) the matrix T and the embedded C. (142) a. which do not.4). [141a]). ∗ [[The chair][−Mer] managed [PRO[+Mer] to meet at 6/apply together for the grant]] Landau obtains the wanted results by requiring that PRO have its ␾-features licensed and ascribing a specific set of features to the C.5. something that is allowed in Brazilian Portuguese.2 and 4. In addition.5. the presence of an embedded predicate that requires a plural subject is not sufficient for partial control to be licensed. All it says is that PRO is semantically plural. crucial parts of the postulated feature system (the R-assignment rule. the availability of partial control seems to be related to properties of the embedded predicate as well. The chair hoped to meet at 6/apply together for the grant b. it does not distinguish tensed infinitives such as (141a). As observed by Hornstein (2003). [[The chair][−Mer] hoped [PRO[+Mer] to meet at 6/apply together for the grant]] Note that the representation in (141b) by itself is not able to ensure that PRO is interpreted as being a set which includes the referent of the matrix subject as a member. Thus.5.and T-heads in tensed infinitivals which then allows PRO to get its ␾-features valued through a chain of agreement relations involving: (i) the matrix T and the matrix subject. . One such incorrect prediction was already mentioned in section 2. as shown in (143). ∗ The chair managed to meet at 6/apply together for the grant b. for instance. Here we will put these problems aside and focus on some incorrect empirical predictions that a PRO-based analysis a` la Landau makes with respect to partial control. for instance) are not independently motivated (something that Landau [2004: 852] himself acknowledges) and the typology predicted is empirically incorrect in that it has no room for finite control into indicatives. As discussed in detail in section 2. from untensed infinitives such as (142). and (iv) the embedded infinitival T and PRO (see sections 2. (143) ∗ John hoped [PRO to sing alike/to be mutually supporting] Notice that in (143) the matrix predicate is of the type that licenses partial control (cf. In addition. the use of a specific chain of agreement relations to account for partial control overgenerates and incorrectly allows finite control in languages like English. in contrast with the MTC.2. (iii) the embedded C and the (tensed) infinitival T. ∗ The chair sang alike/was mutually supporting with Bill b. (144) a. we may go one step further and show that the MTC analysis encapsulated in (147) is empirically superior to PRO-based approaches. The chair left/went out (with Bill) c. as shown by (144c)/(145c). [145a]). In fact. the commitative must be selected. The chair met/applied together for the grant (∗ with Bill) (145) a. Compare the data in (144) and (145). [[The chair]i hoped [ti to meet procommitative at 6]] b. The chair hoped to meet/apply together for the grant Given that predicates like ‘sing alike’ and ‘be mutually supporting’ require a plural subject.5. [The chair hoped [PRO to meet procommitative at 6]] b. as shown in (147) below. (147) a. The chair hoped to leave/go out ∗ ‘The chair hoped that he and other people would leave/go out’ c. . (144a) shows that the plural meaning cannot be obtained via a commitative. [The chair hoped [PRO to apply together procommitative for the grant]] Note that. Rather. As seen in (147). the embedded subjects of the partial-control constructions in (146) should in fact be garden-variety NP-traces. Accordingly. the derivation of partialcontrol structures in fact seems to involve the licensing of a null commitative. Under the MTC. ∗ The chair hoped to sing alike/be mutually supporting b. the appeal of the PRO-based analysis gets bleached. as represented in (141b). (146) a. the MTC is also compatible with partial-control phenomena. for example. [[The chair]i hoped [ti to apply together procommitative for the grant]] Data such as (144) and (145) thus show that it is not the case that partial control necessarily favors PRO-based analyses or that it is fatal to the MTC. Rather than involving a PRO with different semantic features. Examples (144b)/(145b) in turn show that being compatible with an (adjunct) commitative is not sufficient to license partial control ([145b] only admits an exhaustive interpretation for the embedded subject). as illustrated in (146). one need not commit hostages to postulating PRO or assigning special properties to it. When the facts in (144) and (145) are taken into consideration. these predicates do not license partial control (cf.6 Partial and split control 185 Hornstein (2003) in fact suggests that only predicates that select a commitative PP can support partial control. if partial control just involves the licensing of null commitatives. ∗ The chair met at 6 b. Bearing this in mind. under PRO-based analyses. (150) [[The chair] hoped [PRO to meet with the president]] Let us now consider the tense restrictions. whereas under the MTC partial control is related to the licensing of null commitative complements. (148) is to be represented as in (150) below. this reading is impossible in (148). (151a) and (152a) are excluded because the subject is semantically singular and there is no position available for a semantically . (151) a. (148) is to be represented as in (149) below. partial control is something related to the embedded null subject of tensed infinitivals. However. Take the sentence in (148). The chair cannot apply together for the grant Under a PRO-based approach. At first sight. (148) The chair hoped to meet with the president Under the MTC approach outlined in (147). PRO-based approaches admit unavailable readings with verbs that select commitatives. However. If so. there is no gain in such translation. for instance. where the embedded subject is a trace of the matrix subject. Suppose that PRO in (150) is semantically plural (it is marked [+Mer] in Landau’s [2004] terms). ∗ The chair applied together for the grant b. Notice that the structure in (149) can only support an exhaustive control reading: the chair is the only person to meet with the president. (149) [[The chair]i hoped [ti to meet with the president]] By contrast. based on Rodrigues (2007). Under a PRO-based approach.186 Empirical challenges and solutions First. (150) should allow an interpretation under which the chair hoped that a group of people including him would meet with the president. but tenseless infinitivals do not. consider the contrasts in (151) and (152). Why an overt commitative eliminates an otherwise available partial-control reading is therefore quite surprising under PRO-based analyses. Suppose we adapt Landau’s proposal that tensed and untensed infinitivals contrast in licensing partial control and assume that tensed infinitivals license null commitative complements. we should in principle be able to find interpretive effects associated with partial control even when no infinitival clauses are involved. a possibility available given that the infinitival is tensed. The chair can only meet tomorrow (152) a. Thus. the two approaches do make distinct predictions. when we look closer. Given that secondary predication is clause-bound. As Rodrigues (2007) points out. contrary to fact. this is no surprise. John hates to meet angry b. Note that the null commitative is an indirect object of sorts (it corresponds to an overt PP) and. as sketched in (iib). the DP of (i) moves to the matrix clause. (153) a. the secondary predicate modifies ‘John. PRO-based accounts of partial control seem unable to capture the interpretation of the secondary predicate in sentences such as the ones in (153). one cannot say that this modification is indirectly allowed in virtue of PRO being plural and ‘John’ being included in the denotation of PRO. predicting that these sentences should be equally unacceptable. Following Wurmbrand’s (2006) proposal that “tensed” infinitivals actually involve an abstract modal operator (‘woll’). [[The chair]i decided [ti to [[pro ti ] meet at 6]]] .’ Under the null commitative approach suggested above coupled with the MTC.5. [Johni hates [ti to meet procommitative angry]] b. this reasoning should carry over to sentences like (151b) and (152b). we have reinterpreted her original proposal of a complex DP subject in terms of a null commitative complement.5. as we saw in section 5. for instance. (i) [pro DP] (ii) a. leaving pro stranded. Thus. Were that possible. Rodrigues’s proposal does not account for the commitative restrictions on the embedded predicate illustrated by (145) and (148). The chair decided to meet at 6 b. Although compatible with the MTC. Moreover. secondary predicates cannot modify indirect objects. In the derivation of a partial-control sentence such as (iia). the secondary predicate of the structures in (155) below cannot modify the matrix subject. (154) a.42 Another argument against PRO-based approaches to partial control and in favor of the MTC involves secondary predicates (see Rodrigues 2007). for example. As illustrated in (154) below. the secondary predicate modifies the trace of the moved subject. Consider the sentences in (153). John wants to meet ready for all contingencies In each of the sentences in (153).6 Partial and split control 187 plural PRO. [Johni wants [ti to meet procommitative ready for all contingencies]] By contrast. 42 Rodrigues (2007) actually proposes that partial-control constructions involve a complex DP subject with null pronoun along the lines of (i) below. Rodrigues argues that standard partial constructions involving infinitivals and data such as (151b) and (152b) can receive a uniform analysis if the null pronoun that acts as the trigger for the semantically plural reading is licensed by the modal. as illustrated in (158) (with English words). for instance.FEM casually]] Rodrigues (2007) provides one final very ingenious argument in favor of the MTC based on Torrego’s (1996) work on control constructions in Spanish such as (159a). where the secondary predicate agrees with the trace of the matrix subject. this can be clearly seen when the controller involves nouns such as v´ıtima ‘victim’ in Brazilian Portuguese. for the floating DP in the embedded clause is understood as a subset of the set denoted by the matrix subject. (155) a. As she observes. Given the subject-doubling .FEM victim. which is feminine regardless of whether its referent is male or female. agreeing in gender and number with the matrix subject. Fuimos los ling¨uistas went. the secondary predicate must surface in the feminine singular form. Again. No sabemos si firmar los ling¨uistas la carta Not know. with the secondary predicates modifying individuals denoted by the semantically plural subject. which are arguably related to the possibility of subject doubling in (159b). (159) Spanish (Torrego 1996): a. (157) Brazilian Portuguese: A v´ıtima decidiu [se reunir vestida informalmente] The. Rodrigues (2007) also shows that embedded secondary predicates syntactically agree with the matrix subject in partial-control constructions.INF the linguists the letter ‘We don’t know whether the linguists among us should sign the letter’ b.188 Empirical challenges and solutions we should find a similar reading in a sentence such as (156). for example. In a sentence like (157) below.FEM victim.FEM]i decided [ti to gather dressed.FEM decided REFL gather dressed.1PL the linguists ‘The linguists among us went’ Rodrigues refers to cases like (159a) as inverse partial control. [Johni hates [PROi+j to meet angry]] b.FEM casually ‘The victim decided to gather dressed casually’ (158) [[The. However. these agreement facts are exactly what one expects under the commitative approach coupled with the MTC. the secondary predicates of (156) must hold of the whole committee and cannot be restricted to its leaders.1PL whether sign. as indicated by the translation. [Johni wants [PROi+j to meet ready for all contingencies]] (156) [Its leaders said [that the committee met drunk/angry/ready for anything]] In addition. . it is plausible that the inverse partial-control construction in (159a) also involves subject doubling.INF the linguists ‘We don’t know whether the linguists among us should go’ (Rodrigues 2007) b. under the PRO-based structure in (161). if the matrix subject is overt. In (160). that is. a copy. The relevant question is whether (159a) is to be represented as in (160). The existence of partial-control interpretations is often taken to be incompatible with movement theories of control.1PL whether go. the floating DP and the controllee should form a doubling structure. This section has reviewed approaches to partial-control phenomena that are fully compatible with the 43 See Rodrigues (2007) for additional arguments. as the MTC would require. (162) Spanish: ir los ling¨uistas a.43 To conclude.5. or with the structure in (161).1PL the linguists ‘The linguists among us went’ (Torrego 1996) The contrast between (159) and (162) shows that a null first-person plural pronoun can be doubled. but with PRO. the empirical coverage provided by the MTC with respect to inverse partial-control constructions is superior to PRO-based approaches. (160) [proi not know. By contrast. So.1PL [whether [TP ti sign [vP [ti the linguists] tsign the letter]]]] (161) [proi not know. if this association is licit in (159a). but not by the structure in (161). the floating DP is not directly associated with the matrix subject. Again.e. i. of the matrix subject. but an overt one cannot. it should also be allowed in (162a). ∗ Nosotros no sabemos si We not know. Thus. which mimics the contrast between (159b) and (162b) observed by Torrego. the floating DP doubles a trace. where the floating DP forms a doubling structure with the trace of the matrix subject. That being so. contrary to fact.6 Partial and split control 189 structure in (159b). ∗ Nosotros fuimos los ling¨uistas We went. where the floating DP doubles PRO. it should not allow doubling for the same reasons (162b) does not license it. Rodrigues concludes that the ungrammaticality of (162a) can only be accounted for by the MTC structure in (160).1PL [whether [TP PROi sign [vP [ti the linguists] tsign the letter]]]] Rodrigues builds an answer based on the contrast between (159a) and (162a). . although most verbs do not allow split control.TOP Hiroshi. each marker is associated with a different type of obligatory control: the intentive marker is associated with subject control. as illustrated in (163) below. Turkish (see Oded 2006). 5. some do. But even in this regard the MTC proves to be no worse than PRO-based alternatives. Hebrew.6.2 Split control Let us now turn to split control. There remain several open questions. the imperative marker with object control. (163) Johni proposed to Maryk [PROi+k to help each other] As Fujii shows. which do not have an explanation for these facts either.45 Interestingly. .C ordered ‘Yoko ordered Hiroshi to eat my bagel’ 44 Similar facts have been observed in German.ACC eat.ACC eat. it can shed more light on what licenses split control and how it is derived. Here we will focus on exhortatives. why the extensive speaker and language variation regarding the accessibility of such readings? All speakers easily get the exhaustive readings and for many the partial readings are difficult to get if attainable at all.44 In this section we present an MTC approach to split control based on Fujii’s (2006) work on control in Japanese. the imperative marker -e/-ro. (164) Japanese (Fujii 2006): a. and the exhortative marker with split control. Japanese has three mood particles that trigger obligatory control: the “intentive” marker -(y)oo. Landau (2000) points out that. Taroi -wa [PROi boku-no beeguru-o tabe-yoo-to] keikakusita Taro. See Fujii (2006) for discussion.C planned ‘Taro planned to eat my bagel’ b. as respectively illustrated in (164) below. All things considered. 45 Japanese exhortative constructions are interpreted in a way similar to English let’s-constructions. it is fair to say that the problems that partial-control phenomena allegedly pose to the MTC are more related to our lack of understanding of these phenomena than to architectural features of the MTC. As Japanese ties the possibility of split control to a specific particle.INTENT. Yokoi -wa Hiroshik -ni [PROk boku-no beeguru-o tabe-ro-to] meireisita Yoko.TOP my bagel. this should not be taken to imply that we fully understand partial control. and Japanese (see Fujii 2006).190 Empirical challenges and solutions MTC and has also discussed constructions that present serious problems for PRO-based approaches. However.IMP. and the “exhortative” marker -(y)oo.DAT my bagel. For example. C said. .EXHORT.C thinks. [164b]).ACC respect. [164c]).RECIP. .6 Partial and split control 191 c. on the other hand. the flipside of what exists in the imperative mood. ‘kare. and split control with exhortative mood (cf.RECIP.ACC told ‘Taroi told his brother that hei said to Hiroshik to respect each otheri+k ’ b.TOP brother.TOP Hiroshi.’ (165) Japanese (Fujii 2006): a. . Fujii (2006) points out that this leaves out one possibility: subject control in a two-DP argument structure.ACC teiansita help.DAT each other. [164a]). Note that in (165a) there are two local controllers for PRO in the intermediate clause. MOOD . ] V] Both the ungrammaticality of (165b) and the unavailability of mood markers associated with the control configuration in (166) have the flavor of a minimality/minimal-distance principle effect. Taroi -wa Hiroshik -ni [PROi+k otagai-o tasuke-a-oo-to] Taro. i. ‘kare’ and ‘Hiroshi.e. . Relevant to our current discussion is the fact that the antecedents of exhortative constructions must be local to one another. how 46 See Landau (2000) for similar observations regarding split control in English.ACC sonkeisi-a-oo-to] omotteiru-koto]-o tugeta respect.DAT otagai-o sonkeisi-a-oo-to] itta-koto]-o tugeta each other. as sketched in (166). object control with imperative mood (cf. ∗ Taroi -wa Hiroshik -ni [karei -ga [PROi+k otagai-o Taro.C.C.EXHORT. exhortative constructions such as (164c) test positive for all the standard diagnostics of obligatory control. there is only one local antecedent available.DAT he.’ In (165b). aside for the ban on split antecedents.46 as illustrated by the contrast in (165) below.NOM Hiroshi. that is.RECIP.C proposed ‘Taro proposed to Hiroshi to help each other’ Fujji shows that.5.DAT he.TOP Hiroshi. Taroi -wa otooto-ni [karei -ga Hiroshik -ni [PROi+k Taro. they cannot be in different clauses. As seen above. .ACC told ‘Taroi told Hiroshik that hei thought that theyi+k should respect each other’ Also relevant to the present discussion is a particular gap in the mood paradigm in Japanese. Japanese allows subject control with “intentive” mood (cf..EXHORT. But if minimality obtains.NOM each other. (166) ∗ [DP1 DP2 [CP PRO1 . but reinterpreted it under a commitative structure. ␤ does not count as an intervener and ␣ can move to [Spec. [MoodP [␣ [+ ␤]] [Mood’ -(y)oo TP]]]]]] ↑ Going back to English. [ia]). In order to block (iib). [vP John v [VP washed [John + Bill]]] . Fujii proposes that they license some sort of coordinate structure in its Spec. John washed Bill b. for example. (ii) a. we have kept the gist of Fujii’s original proposal that split control involves some sort of coordination. vP] without any minimality problems.192 Empirical challenges and solutions are split-control configurations like (165a) derived? Fujii argues that the mood particles highlighted in (164) head mood phrases (MoodPs) and that there is no case available for the subject of MoodPs. The sentence in (163). In (168a). movement of ␣ in (168b) is parallel to what we saw with subjectcontrol cases involving ‘promise’ (see section 5. In the particular case of exhortative -(y)oo. repeated in (169) below. can be derived as sketched 47 The derivation that Fujii (2006) actually proposes for licit instances of split control involves the steps illustrated in (i) below. [vP ␣ [v’ v [VP [+ ␤] [V’ V . . as in (168). [MoodP [␣ + ␤] [Mood’ -(y)oo TP]]]]]] ↑ Although compatible with the MTC. b. (i) a. it is plausible that its split-control constructions also involve a derivation with an exhortative MoodP. the derivation above seems to make the incorrect prediction that a sentence such as (iia). Under the assumption that [+ ␤] is complex with + akin to a null commitative preposition. [MoodP [␣ + ␤] [Mood’ -(y)oo TP]]]] ↑ [vP ␣ [v’ v [VP [␣ + ␤] [V’ V . ␣ then moves to the matrix [Spec. [MoodP [␣ [+ ␤]] [Mood’ -(y)oo TP]]]] ↑ b. In the relevant respects. [+ ␤] moves to the internal ␪-position of the subcategorizing V. If so. .5. yielding (ib). .’ given the structure in (iib).47 (168) a. as represented in (167). . .1). ␤ moves to receive the ␪-role of the matrix V. could mean ‘John washed Bill and himself. vP]. . [VP [+ ␤] [V’ V . [MoodP [␣ + ␤] [Mood’ -(y)oo TP]] (167) Suppose that the complex specifier in (167) is some sort of commitative expression. pied-piping ␣ (cf. . split-control structures such as (164c) or (165a) can be derived along the lines of (168) below. [VP [␣ + ␤] [V’ V . for instance. . but it is harder to understand why imperatives must resist them. For instance. According to Landau (2000: 55). . Why interactions like these should hold is far from clear. [MoodP [John ↑ [+Mary]] . why this is so and what exactly characterizes exhortations so that they essentially differ from imperatives and other kinds of moods remains. we have witnessed above another interaction between commitatives and modals (see section 5. why is split control limited to exhortatives in Japanese? Why do imperatives not support such readings in Japanese? The problem becomes more pressing when one sees that the semantic apparatus that handles imperatives is easily extended to accommodate exhortatives. For example.5. were there a semantic . nor are recommendations to do so any odder semantically. . However. [MoodP [John [+Mary]] to help each other]] ↑ b. Landau (2000) and Fujii (2006) propose that this follows from the semantics. This said. we believe. ]]] The account reviewed above leaves several questions open. Why can one not order someone or recommend to someone to engage in a collaborative activity. . As for the question of why certain verbs resist split control.6 Partial and split control 193 in (170). .1). This does not seem semantically untoward.g. The considerable amount of overlapping between the verbs that allow exhortative -(y)oo in Japanese and the verbs that allow split control in English suggests that something along these lines may indeed be on the right track. [vP John proposed-v [VP [to [+Mary]] tproposed . . recommend and order do not allow split control – for obvious reasons. given that in order to engage in some action. [VP proposed .6. one does not recommend to/order other people to do it.” We do not find this at all obvious. (169) John proposed to Mary to help each other (170) a. . quite unclear. e.. It is easy to understand why exhortatives require plural (commitative) subjects. “[u]nlike propose and ask. The fact is that split control seems to be very restricted. washing each other? Why can ‘John ordered Mary to wash each other’ not mean that John ordered Mary to engage with him in the activity wherein each of John and Mary washes the other. perhaps applying only to verbs that support exhortative interpretations. It appears that they only differ in that imperatives encumber the addressee’s to-do list while exhortatives fill in the speaker’s to-do list as well. for instance. In the previous sections we have examined a variety of empirical phenomena that have been claimed to pose insuperable problems for the MTC. We think that this difference is not significant enough (at least not in a framework like minimalism. under close scrutiny.7 Conclusion As discussed throughout this volume. and (local) scrambling. all the allegedly deadly counter-examples can receive plausible analyses under the MTC. The only relevant difference between obligatory control. and raising. rather than presenting fatal counter-examples to the MTC as often claimed. . is that the relevant A-movement in the case of obligatory control is triggered by ␪-reasons. it could be easily combined with the MTC (as Fujii observes). most of the phenomena reviewed in this chapter end up lending strong conceptual and empirical support to the MTC. on the other. on the one hand. 5. All in all. raising that it reduce to the extra thematic relation established under control. which does not recognize any substantive notion of D-structure) to warrant a special control operation/construction/rule. We have shown that. as a research strategy. all the answers given stemmed from two simple ideas: (i) that. of course. What we have done is make the relevant configurations explicit so that minimality can be properly computed and ␪-marking properly characterized. say. if control involves movement. we would like it to be true of any difference between obligatory control and. However. (relativized) minimality must be obeyed. it seems to us that. and (ii) that quirky-casemarked DPs must be stripped of their quirky case if they are to be further ␪marked. Even more importantly. the MTC takes obligatory control to be established via A-movement. Indeed. we do think this thematic difference is important when it comes to explaining some divergences between obligatory-control constructions and the other types of constructions that rely on A-movement. passivization. It is. up to the reader to decide if we are right and to what degree the purported difficulties for the MTC reviewed and reanalyzed here vitiate the project of reducing control to movement.194 Empirical challenges and solutions restriction of the kind Landau and Fujii suggest. 6 On non-obligatory control 6. 2003. with the focus of inquiry resting on obligatory control (OC). NOC does not resort to movement. the two are to some extent independent from one another. Thus. However.4 outlines an approach according to which the interpretation of NOC PRO is a dual function of the grammar and the parser. a disclaimer is in order. That said. Section 6. non-obligatory control (NOC) has been pushed to the side. As NOC is the elsewhere case (when movement is not involved). 2007) proposal that NOC PRO is a null pronoun and that the complementary distribution between OC and NOC is couched on an economy competition between movement and pronominalization. this does not affect the essence of the preceding chapters. the MTC effectively has something to say about control relations that exhibit movement diagnostics but not much about construal relations that are not derived by movement. as opposed to OC. Nonetheless. Section 6.2. 1 The discussion to be presented below is primarily based on Boeckx and Hornstein (2007). 195 .1 Introduction Within the MTC. As mentioned above. In section 6. there may be different types of NOC which in turn may be subject to different licensing conditions. The chapter is organized as follows.1 But before we proceed.5 concludes the chapter. we discuss configurations where OC and NOC are in complementary distribution and present Hornstein’s (1999. The most obvious reason for this is that.3 examines (apparent) counter-examples to this complementarity and section 6. 2001. should it turn out to be partially or totally incorrect. The discussion below presents the beginnings of an account of NOC that we believe is quite reasonable. although we think that the proposal to be discussed below fits snugly with the version of MTC advocated in this volume. the MTC is incomplete without an account of the distribution of NOC and in this chapter we would like to present our thoughts on this issue. let us move to the discussion proper. Hornstein (2001. b. b. [5b]). if it has an antecedent. non-obligatory control and economy computations The pairs of examples in (1)–(6) below illustrate the systematic contrast between OC and NOC (see Chapter 2). Example (1a) shows that OC PRO requires an antecedent. (5a) that OC PRO must be interpreted as a bound variable when associated with an only-DP. (5) a. ∗ Johni ’s campaign expects PROi to shave himself Johni ’s friends believe that PROi keeping himself under control is vital if he is to succeed (4) a. Johni expects PROi to win and Billk does too (‘and Billk expects himself to win. ∗ It was expected PRO to shave himself It is illegal PRO to park here (2) a.196 On non-obligatory control 6. 2007) accounts for this correlation by reinterpreting in minimalist terms the old idea that (resumptive) pronouns are employed as a last resort saving strategy . 2b]) or in a c-commanding position (cf. Sweat. the antecedent need not be local [(cf. and it allows both strict and sloppy readings under ellipsis (cf. the corresponding bexamples show that exactly the opposite holds of NOC PRO: it does not require an antecedent (cf. (1) a. [4b]). the complementary distribution between OC and NOC generally correlates with environments where movement can or cannot take place. and de se and non-de se readings (cf. b. (3a) that the antecedent must be in a c-commanding position.’ not ‘and Billk expects himi to win’) Johni thinks that PROi getting his resum´e in order is crucial and Bill does too (‘Billk thinks that hisi/k getting his resum´e in order is crucial’) [Only Churchill]i remembers PROi giving the ‘Blood. (2a) that the antecedent must be local. b. [3b]). [6b]). In fact. and Tears’ speech Only Churchill remembers that PRO giving the BST speech was momentous [The unfortunate]i expects PROi to get a medal [The unfortunate]i believes that PROi getting a medal is unlikely Note that the b-examples in (2)–(6) also illustrate a typical environment where an NOC PRO can be found: an island configuration.2 Obligatory vs. On the other hand. 2003. bound and coreferential readings when associated with only-DPs (cf. and (6a) that OC PRO only admits de se reading in “unfortunate” contexts. b. ∗ Johni thinks that it was expected PROi to shave himself Johni thinks that Mary said that PROi shaving himself is vital (3) a. (4a) that OC yields sloppy readings under ellipsis. (6) a. b. [1b]). it admits both strict and sloppy readings under ellipsis (cf. Crucially. b. Rodrigues 2 But see section 6. Thus. and general discussion on the competition between movement and pronominalization with respect to derivational economy.3 below for a discussion of cases where null and overt pronouns do not go hand in hand. non-obligatory control 197 when movement fails. Hornstein argues that movement is more economical than pronominalization. b. e. 2003. More concretely. for instance. Floripi (2003). [9c]) antecedent.2 That is. [9d]).6. (7) a.2 Obligatory vs. an overt pronoun does not need a linguistic antecedent (cf. (8) a. [9f]). as illustrated in (9) below. bound and coreferential readings in sentences like (9e). (9) a. 2007) for further arguments. it may be associated with a non-local (cf. technical implementation. . [[The unfortunate]i expects [ti to get a medal]] [[The unfortunate]i believes that [[proi getting a medal] is unlikely]] ∗ ∗ [[The unfortunate]i expects [proi to get a medal]] [[The unfortunate]i believes that [[ti getting a medal] is unlikely]] The distribution and interpretation of pro in the b-examples of (1)–(6) to a great extent mimic the distribution and interpretation of overt pronouns. NOC PRO can be analyzed as a null pronoun (pro) which is resorted to when movement is not possible. as argued in the previous chapters. they cannot be represented as in (8): in (8b) movement of the embedded subject should induce an island violation and in (8a) there is an economy violation as the less economical option of pronominalization was employed instead of movement (cf. d. c. 3 See Hornstein (2001. if OC PRO is a residue of movement under the MTC. [9a]). and de se and non-de se reading in “unfortunate” contexts (cf. [9b]) or nonc-commanding (cf. It is illegal for him to park here Johni thinks that Mary said that hisi shaving himself is vital Johni ’s friends believe that hisi keeping himself under control is vital if he is to succeed Johni thinks that hisi getting his resum´e in order is crucial and Bill does too (‘Billk thinks that hisi/k getting his resum´e in order is crucial’) Only Churchill remembers that his giving the BST speech was momentous [The unfortunate]i believes that hisi getting a medal is unlikely This general competition between movement and pronominalization extends beyond standard instances of control.3 For instance. b. are to be represented along the lines of (7) below. [7a]) in a configuration where both options would lead to convergent results. Under this view. the structures in (6). f. with a trace in (7a) and pro in (7b). we find a different behavior: the antecedent for the null possessor need not be within its clause (cf. [12c]). [10a]) and only admits sloppy readings under ellipsis (cf. Assuming Hornstein’s (1999. (11) [[S´o o Jo˜ao] ligou para a m˜ae ec] Only the Jo˜ao called to the mother ‘Only Jo˜ao called his mother → Nobody else called his own mother’ NOT ‘Nobody else called Jo˜ao’s mother’ [Non-de se context: Jo˜ao doesn’t remember who he is or that the person under discussion is his brother] #[Jo˜ao passou a admirar o irm˜ao ec] Jo˜ao passed to admire the brother ‘Jo˜ao came to admire his brother’ (de se reading only. [12a]) and the null possessor is compatible with strict and sloppy readings (cf. and de se readings (cf. bound readings when associated with only-DPs (cf. if the null possessor sits within a subject. Brazilian Portuguese (Floripi and Nunes 2009): [O Pedrom acha que [o amigo [d[o Jo˜ao]i ]]k telefonou The Pedro thinks that the friend of-the Jo˜ao called para a m˜ae eck/∗ i/∗ m ] to the mother ‘Pedrom thinks that [Jo˜aoi ’s friend]k called hisk/∗ i/∗ m mother’ b. (10) a. 2001) theory of control. these authors analyze the empty category in (10) as a trace left by movement of the possessor to the specifier of the closest vP. and de se and non-de se readings . the empty category in (10) below must be interpreted as the closest c-commanding antecedent (cf. Thus. and Floripi and Nunes (2009) show that. bound and coreferential readings (cf. too’ (sloppy reading only) c. d. as observed by Floripi (2003) and Floripi and Nunes (2009). [10c]). infelicitous in this context) Brazilian Portuguese: [TP [o Jo˜ao] [vP ti ligou [PP para [DP a m˜ae ti ]]]] The Jo˜ao called to the mother ‘Johni called hisi mother’ Interestingly. [[O Jo˜ao]i vai telefonar para a m˜ae eci ] e [a The Jo˜ao goes call to the mother and the Maria tamb´em vai] Maria also goes ‘Jo˜ao will call his mother and Mary will call her mother. in Brazilian Portuguese. as illustrated in (11).198 On non-obligatory control (2004). [10d]). [10b]). a null possessor sitting within an object displays all the diagnostics of OC. [12b]). b. the null possessors in (12) pattern like their overt counterparts in (13). Brazilian Portuguese (Floripi and Nunes 2009): [[O Jo˜ao]i disse que [[o amigo delei ] vai viajar]] The Jo˜ao said that the friend of-him goes travel ‘Jo˜ao said that his friend is going to travel’ [[A Maria]i vai recomendar a pessoa [que [um amigo The Maria goes recommend the person that a friend [o Jo˜ao]k tamb´em vai] delai ] entrevistou] e of-her interviewed and the Jo˜ao also goes ‘Maria is going to recommend the person that a friend of hers interviewed and Jo˜ao is also going to recommend a person that a friend of his/hers interviewed’ (sloppy and strict readings available) c. Brazilian Portuguese (Floripi and Nunes 2009): [[O Jo˜ao]i disse que [[o amigo eci ] vai viajar]] The Jo˜ao said that the friend goes travel ‘Jo˜aoi said that hisi friend is going to travel’ [[A Maria]i vai recomendar a pessoa [que [um amigo eci ] The Maria goes recommend the person that a friend entrevistou] e [o Jo˜ao]k tamb´em vai] interviewed and the Jo˜ao also goes ‘Maria is going to recommend the person that a friend of hers interviewed and Jo˜ao is also going to recommend a person that a friend of his/hers interviewed’ (sloppy and strict readings available) c. In other words. [[S´o o Jo˜ao] leu o livro [que [a m˜ae dele] indicou]] Only the Jo˜ao read the book that the mother of-him recommended ‘Only Jo˜ao read the book that his mother recommended’ → ‘Nobody else read the book that his own mother recommended’ or ‘Nobody else read the book that Jo˜ao’s mother recommended’ . [Non-de se context: Jo˜ao doesn’t remember who he is or that the person under discussion is his brother] Jo˜aoi se surpreendeu [quando [o irm˜ao eci ] fez um discurso] Jo˜ao REFL surprised when the brother made a speech ‘Jo˜ao got surprised when his brother made a speech’ (non-de se reading available) (13) a.6. [12d]).2 Obligatory vs. (12) a. b. [[S´o o Jo˜ao] leu o livro [que [a m˜ae ec] indicou]] Only the Jo˜ao read the book that the mother recommended ‘Only Jo˜ao read the book that his mother recommended’ → ‘Nobody else read the book that his own mother recommended’ or ‘Nobody else read the book that Jo˜ao’s mother recommended’ d. non-obligatory control 199 (cf. After the two independent syntactic objects in (16a) are built.4 4 It is also consistent with what follows if we assume a derivation reminiscent of Belletti and Rizzi’s (1988) analysis of psych-verb constructions. which surfaces as (15) after the copy of ‘Mary’ within the clausal subject is deleted in the phonological component (cf. if pronominalization were generally available. as illustrated in (14). consider the PRO-gate sentence (see Higginbotham 1980) in (15) below. Once movement is possible. is taken to proceed along the lines of (16).’ as shown in (16b). it preempts pronominalization. the computational system copies ‘Mary’ and merges it with ‘delighted. as shown in (17). [Non-de se context: Jo˜ao doesn’t remember who he is or that the person under discussion is his brother] surpreendeu [quando [o irm˜ao delei ] fez um discurso] Jo˜aoi se Jo˜ao REFL surprised when the brother of-him made a speech ‘Jo˜ao got surprised when his brother made a speech’ (non-de se reading available) Adopting Hornstein’s economy approach to the contrast between OC and NOC. A crucial feature of the derivation sketched in (16) is that. and that pronominalization is sanctioned in these cases as movement out of the subject position is not allowed. First. and movement of . for instance. with the surface subject in (15) being basegenerated in a position lower than the position occupied by the surface object. Further computations yield the simplified structure in (16c). the sentences in (12) should exhibit no interpretive restrictions thanks to the alternative derivation with pro. if movement is possible. [16d]).5. pronominalization is blocked.2).200 On non-obligatory control d. Hornstein and Kiguchi (2003) and Kiguchi (2004) show that PRO-gate structures like (15) display all the diagnostics of OC and propose that they involve sideward movement from within the infinitival subject to the object position. (14) Brazilian Portuguese: [[O Jo˜ao]i disse que [[o amigo proi ] vai viajar]] The Jo˜ao said that the friend goes travel ‘Jo˜aoi said that hisi friend is going to travel’ Let us examine a slightly more complex case which sheds additional light on the correlation between the impossibility of movement and the availability of pro (and hence NOC). Crucially. Floripi (2003) and Floripi and Nunes (2009) argue that the null possessor in (12) is a null pronoun. The derivation of (15).1. as opposed to what we saw in the b-examples in (2)–(6) and in the null-possessor constructions in (12). Again. movement of ‘Mary’ in (16b) is licit as it takes place before the non-finite clause becomes a subject island (see section 4. once movement of ‘John’ from within the gerund to the matrix clause in (18b) is out due to the subject island. c. b. John said that [[PROi /∗ heri washing herself] delighted Maryi ] Johnk said that [[prok /himk washing himself] delighted Mary] Example (18a) is unsurprising as it replicates what we saw in (15)/(17): if movement of ‘Mary’ to the object of ‘delighted’ is allowed (cf. ‘Mary’ proceeding in an upward fashion. it should not. (i) a. pronominalization is possible. d. (18) a. it is relations that are OC or NOC. as illustrated in (18). The contrast between (18a) and (18b) shows that structures should not be classified as OC or NOC.e. i. Applications of select and merge: [Mary washing herself] delighted Applications of copy and merge (sideward movement): [Maryi washing herself] [delighted Maryi ] Application of merge: [[Maryi washing herself] delighted Maryi ] Deletion in the phonological component: [[Maryi washing herself] delighted Maryi ] b. b. identify sentences as OC/NOC clauses.6. OC and NOC describe relations between nominal expressions. not selection/ subcategorization relations between predicates and types of clausal complements. OC and NOC are descriptive predicates that are more analogous to bound and free than to interrogative and declarative. for a given structure may allow OC or NOC. the important point to bear in mind is that movement of ‘Mary’ is licit and therefore preempts pronominalization. (17) ∗ 201 [[heri washing herself] delighted Maryi ] Now consider what happens when structures such as (15) are embedded. d.2 Obligatory vs. non-obligatory control (15) [[PROi washing herself] delighted Maryi ] (16) a. As grammatical theory does not distinguish clauses as reflexive or pronominal depending on whether they contain anaphors or pronouns. That the MTC treats OC and NOC as relations is an important point that is worth emphasizing. pronominalization is not. [16] and (i) in footnote 4). c.. Rather. by parity of reasoning. Regardless of whether (15) is to be derived along the lines of (16) or (i). Conversely. as sketched in (i) below. [VP delighted [Mary washing herself]] [VP Maryi [delighted [ti washing herself]]] [vP delightedk [VP Maryi [tk [ti washing herself]]]] [TP [ti washing herself]m [vP delightedk [VP Maryi [tk tm ]]]] . 1). Assuming that pronominalization and movement compete and that movement is derivationally more economical than pronominalization (see section 6. this account does not explain why the illicit movement of ‘John’ in the derivation of (19) does not make room for pronominalization. In other words. In contrast. movement of ‘Mary’ from the embedded clause to the Spec of ‘persuade’ does not violate minimality. Lexical items cannot be so coded on a strict reading of inclusiveness.. for it codes grammatical restrictions in lexical selection. let us now consider some potential problems.e. one which treats control as a selection relation between a predicate and its embedded sentential complement. For more discussion see section 7. PRO in (20) is a copy/trace of A-movement and this explains why ‘Mary’ is the antecedent and ‘John’ cannot be (see section 3. vP].3 Some problems Consider the representation of the sentence in (19) given in (20).2 below. 6. this makes as much sense as claiming that a predicate selects for an embedded reflexive structure.4.202 On non-obligatory control Consequently. why can (19) not be associated with the structure in (21b). If what we say in footnote 5 is correct. (19) John persuaded Mary to leave (20) John1 persuaded Mary2 [PRO2/∗1 to leave] Under the MTC. we account for why (19) cannot be associated with the structure in (21a) below. However. Successful movement of ‘Mary’ to the Spec of ‘persuade’ blocks pronominalization. For ‘John’ to be the antecedent requires that it move over ‘Mary’ on its way to [Spec. it cannot be the antecedent. 6 One possible answer to this question is not open to us: that ‘persuade’ selects for an OC complement.. with no explanatory ambitions.5 With this general picture in mind. On one reading of the inclusiveness condition this kind of selection would be ruled out.e. a copy/trace of A-movement) from binding a pro (i. a null pronoun) in the same position?6 5 This claim has the following consequence: verbs cannot be classified as taking OC complements and so a standard approach to OC.2). where pronominalization “rescues” a failed movement connection between the two subject positions? What prevents a DP that cannot licitly antecede PRO (i. as this violates minimality. is conceptually misguided. . to say that a given predicate selects/subcategorizes for an OC or NOC structure can only be viewed as a descriptive statement. as shown in (26).5. (23) John1 kissed Mary2 without PRO1/∗2 getting embarrassed If sideward movement of ‘John’ from the position of PRO in (23) is licit. given the preference for merge over move. we have assumed that a coupling between an antecedent and a pronoun is licit just in case movement cannot establish the same relation.1. which is indeed the case. However.6.e. we expect a coreferential pronoun to be ruled out in this position. (19) cannot be interpreted with ‘John’ as the leaver. the question is why pronominalization cannot save the derivation. If one can move from a position to another. why can Mary in (22) not be the one who gets embarrassed. the overt counterpart of (25) yields an acceptable result. if sideward movement of ‘Mary’ in (23) is not allowed. adjunct control involves subject control. we also see that it is impossible.3 Some problems (21) a. movement from the null-possessor . this also implies that if movement is not possible between positions A and B then binding should be. as shown in (24) below. As opposed to (21b). ∗ John1 persuaded Mary [him1 to leave] John kissed Mary1 without her1 getting embarrassed A similar pattern is found in null-possessor constructions in Brazilian Portuguese. However. b. That is. Consider another problematic case: (22) John kissed Mary without getting embarrassed Example (22) is a case of adjunct control.. b. ∗ 203 John1 persuaded Mary2 [pro2 to leave] John1 persuaded Mary2 [pro1 to leave] To phrase the problem differently. By the same reasoning. In a sentence such as (27) below. the position of the trace. given the availability of the structure in (25)? (24) ∗ John1 kissed Mary without him1 getting embarrassed John1 kissed Mary2 without pro2 getting embarrassed (25) Note that the unacceptability of (22) with the structure in (25) is even more troublesome than the unacceptability of (19) under the representation in (21b). As discussed in detail in section 4. a DP in the “target” cannot bind a pronoun in the “launch” site. What we see in (21b) is a concrete example of this option. Furthermore. i. (26) a. as represented in (23). adjunct control involves sideward movement from the subject position of the embedded clause before it becomes an adjunct. rather than object control. allow the relevant readings suggests that more than grammatical requirements are at issue. repeated below in (31). the question is why the null possessor in (27) cannot be interpreted as the matrix subject under the structure in (29b) with a null pronoun. vP] without violating minimality. What else could be at stake? Following Boeckx and Hornstein (2007). (30) ∗ [O Jo˜ao]k acha que [o Pedro] vai ligar para a m˜ae ti [O Jo˜ao]k acha que [o Pedro] vai ligar para a m˜ae prok Brazilian Portuguese: delek [O Jo˜ao]k acha que [o Pedro]i vai ligar para a m˜ae The Jo˜ao thinks that the Pedro goes call to the mother of-him ‘Jo˜aok thinks that Pedro is going to call hisk mother’ We outline a possible answer in the next section. the fact that the overt pronouns in (26b) and (30). (25). get a unified approach (not an obvious requirement. as sketched in (28a). we would like to suggest a parsing-based approach.7 7 If we also assume that producers and parsers meet similar constraints. once movement of ‘o Jo˜ao’ in (29a) below is excluded due to the intervention of the embedded subject. repeated in (32). but not a bad one either). More particularly. then this would not be produced either.4 A proposal If we insist that the problems in (21b). the null possessor in (27) must be interpreted as the lower subject and economy considerations regarding the derivational cost of movement and pronominalization exclude the representation in (28b). as illustrated in (30). and (29b). Such an assumption is natural in any kind of analysis-by-synthesis model. we propose that the structures in (31) are not blocked by the grammar. . Thus. b Brazilian Portuguese: [O Jo˜ao]k acha que [o Pedro]i vai ligar para a m˜ae eci/∗ k The Jo˜ao thinks that the Pedro goes call to the mother ‘Jo˜ao thinks that Pedroi is going to call hisi mother’ ∗ [O Jo˜ao] acha que [o Pedro]i vai ligar para a m˜ae ti [O Jo˜ao] acha que [o Pedro]i vai ligar para a m˜ae proi By the same token. So why is a null pronoun with the same reading unacceptable? (29) a. despite the fact that an overt pronoun allows this interpretation. but neither would ever be accepted by a well-behaved parser.204 On non-obligatory control position can only go as far as the lower [Spec. b. (27) (28) a. 6. (34) gets the parse in (35). see Berwick and Weinberg (1984). and we get (35) below. John kissed Mary1 without her1 getting embarrassed Brazilian Portuguese: delek [O Jo˜ao]k acha que [o Pedro] vai ligar para a m˜ae The Jo˜ao thinks that the Pedro goes call to the mother of-him ‘Jo˜aok thinks that Pedro is going to call hisk mother’ 205 Let us make the following (as far as we can tell. Finally. Parsers move from left to right and project structure rapidly and deterministically on the basis of local information Parsers are transparent with respect to grammars.6. John1 persuaded Mary [pro1 to leave] John kissed Mary2 without pro2 getting embarrassed Brazilian Portuguese: [O Jo˜ao]k acha que [o Pedro] vai ligar para a m˜ae prok The Jo˜ao thinks that the Pedro goes call to the mother ‘Jo˜aok thinks that Pedro is going to call hisk mother’ (32) a. Our assumption only implies that the parser respects the design features of grammars. b.8 Given the assumption in (33b). if grammars encode a condition. we arrive at ‘to’ and the parser realizes that it must assign a subject to the embedded clause. it will prefer to treat a potential gap as a copy/trace (rather than a pro) if it can. Thus. . So. repeated in (34). c. either a pro or PRO (= trace/copy). (34) John persuaded Mary to leave As the sentence is parsed. we expect parsers to be sensitive to earlier information. fairly standard) assumptions: (33) a. As it can. For discussion of the transparency relation between grammars and parsers. it would require that at ‘to’ the parser drop a pro in the subject position. parsers respect it. Moreover. which requires that ‘Mary’ be the antecedent of “PRO. In addition.” (35) John persuaded Mary [t to leave] As for (31a). the trace/copy in (35) must have ‘Mary’ as its antecedent due to minimality. As the parser incorporates the principles of the grammar and grammars “prefer” movement to pronominalization. it does. b. for the only empty category that could take ‘John’ as antecedent is a 8 This does not imply that grammars are identical to parsers (Phillips 1996) – a position which we think is untenable (Phillips [2004] appears to agree on this). the parser “sees” that the subject is a null category. for instance. Take the sentence in (19). As a parser builds structure left to right. the parser “prefers” to drop a trace here if it can. consequently. we expect parsers to prefer traces to pronouns (if grammars prefer movement to pronominalization) and.4 A proposal (31) a. b. that parsers will treat gaps as copies/traces in preference to analyzing them as null pronominal pros. if transparency holds). it drops a trace and not a null pronoun.. As it prefers dropping traces if it can. One point is worth emphasizing here. In other words. . The problem is not one where extra resources would help. Thus. then ‘John’ must be the antecedent in (31b) due to mergeover-move computations and ‘o Pedro’ must be the antecedent in (31c) due to minimality. center-embedding structures can be parsed given more memory “space. once the parser analyzes the null subject of the gerund and the null possessor as traces. on encountering a phonetic gap. the account survives even if in place of parsers all we have is parsing. See below for a case where a pro can be posited in a place where a “PRO” is licit in order to advance another parsing desideratum. however. to drop a pro requires ignoring the parser’s (built-in) preference for a trace/copy over a pronoun. which prefers movement over pronominalization. However. a trace is preferred to pro. for example. When the parser gets to the subject position within the gerund in (31b) or the possessor position in (31c) and needs to drop an empty category.’ 9 Our discussion has reified parsing by adverting to properties of a well-designed “parser. Thus.e. cannot be so easily ameliorated. It is often assumed that parsing strictures can be overridden given greater resources. Thus. however.” The suggestion above. the indicated readings in (31b) and (31c) become unavailable. then a parser cannot circumvent these principles by using additional memory or attention resources. The “preference” the parser displays arises as a design feature of a parser that conforms to transparency (a very good condition − perhaps an optimal one − for regulating the relation between grammars and parsers). the parser must “decide” what sort of empty category to drop in the relevant position. Observe that this account turns on there being an empty category to be parsed. If it is. This makes (31a) computationally unavailable and this accounts for the lack of the indicated interpretation. However. it must be. it prefers a “PRO” to a pro.10 The parse is simply not available. it must drop a trace/copy if it can. if there is an overt pronoun occupying the same position. As a “PRO” can be licitly dropped here. all things being equal − a preference the parser has in virtue of being structurally transparent to the grammar. 10 This does not mean that a pro can never be placed where a “PRO” can be. parsing adopts those principles prized by the grammar.” However.206 On non-obligatory control null pronoun. If parsing principles must respect grammatical ones (i. when using the grammar to divine the nature of an empty category. So. By “seeing” nothing there. we can derive sentences like (32a) and (32b) with an overt pronoun anteceded by ‘Mary’ and ‘o Pedro. The transparency assumption then would amount to saying that.9 The same account extends to (31b) and (31c). the parser is not faced with any choice as to what it must do as overt pronouns are grammatically licit here. Observe that if the empty category were analyzed as a copy/trace. As a “PRO” cannot provide the support for this relation.4 A proposal 207 Consider now the last set of cases. for discussion. are fine with the indicated interpretation. This kind of structural and lexical information is standardly assumed to be available on-line to the parser. for example. at the point where the gerund is encountered. . 13 See. pronouns quite generally greedily appropriate suitable interpretive antecedents (referential anchors) very rapidly on-line. Note. However. Nicol and Swinney (1989). (36) Johnk said that [[prok washing himself] delighted Mary] Here the parser gets to the subject gerund and “encounters” an empty category. this does not end matters.12 So it seems that the parser can drop a trace here. however. 12 Either sideward movement (cf. Thanks to Nina Kazanina for very helpful discussion and references. However. then in cases such as (36) and (37). the relevant information that the gerund is a subject (and hence an island) is known. the connection with ‘John’ would be illicit as it would require movement from an island. In contexts like (36) and (37) these desiderata pull in opposite directions: if the empty category is understood as a pro it can be related to ‘John’ and so can rapidly be provided with an interpretation at this point. it must treat it as a pro. On the other hand. (37) John said that [[PROi washing herself] delighted Maryi ] The PRO here is a residue of movement. If we add the assumption that parsers are interpretively greedy to our previous two assumptions in (33). the parser has competing preferences: it would like to assign an interpretation to the empty category (at this point in the parse) and it would prefer to treat the empty category as a trace rather than a pronoun. [16]) or movement as in a psych-verb construction (cf. Why then does this not prevent dropping a pro in (36)? The answer is that the parser here must weigh a competing parsing demand.13 Thus. We noted that examples like (18b). that the empty category is inside an island and if the parser wants to link ‘John’ to this element. Osterhout and Mobley (1995). [i] in footnote 4). and Badecker and Straub (2002). a pro is licensed by the grammar. It is known that parsers like to assign interpretations to empty categories (and dependent elements in general) very quickly. We have seen that (37) is also acceptable. this will also require overriding its preference for traces over pronouns.11 It can treat it as a copy/trace or a pro. Thus. if it drops a trace 11 The parser “knows” that this is a subject because it follows ‘that’ and because it knows that ‘believe’ does not take gerundive complements. repeated here in (36).6. As such. it should be able to have ‘Bill’ as antecedent in (38).14 It is instructive to compare (36) and (37) with the structures in (31). in (36) and (37) the potential antecedents for the empty categories are on opposite sides. 14 It is possible that different speakers weigh these options differently. there is no parsing advantage to interpreting the empty category as a null pronoun (there is nothing to link the empty category to so that it can be quickly interpreted). Consider a sentence like (38). “PRO” over pro). creating a situation in which transparency and quick assignment of interpretation to the empty category pull in opposite directions. giving us a structure like (39). as both options have their virtues. Thus. 15 See Hornstein and Kiguchi (2003) and Kiguchi (2004) for details and evidence that the empty category in (38) is an OC PRO and not a pro.’15 Note that. (39) [PRO having to wash behind the ears] . [37]).. In (31) the potential antecedents are both to the left of the empty category. [33a]). were there a pro here. the parser not only complies with transparency (i. Recall that this is a case where the antecedent will only become visible downstream and our parser operates from left to right (cf. However.. we would expect the parser to drop a “PRO” here.208 On non-obligatory control here. (38) Having to wash behind the ears made Mary angry at Bill Here. This suggests that these speakers value transparency more than reference resolution. It goes without saying that the proposal above is not a fully worked-out account and that much more detailed work needs to be done to flesh it out. For some interesting work extending this line of reasoning. it can adhere to its preference for traces over pronouns (i. this reading seems unavailable in (38). . we have NOC (cf. If this analysis is roughly on the right track. . By contrast. In short. . If we substitute ‘his’ for ‘the’ in (38). if the empty category is analyzed as a trace. Nina Kazanina (personal communication) has found many speakers for whom sentences like (37) with ‘John’ as antecedent become very odd when ‘Mary’ is encountered. we suggest that both parses are available. Thus.e. but also satisfies its desire to interpret the empty category quickly as it may rely on the local information already parsed. then some predictions follow. [36]). the parser will analyze the empty category as a residue of A-movement. see Snarska (2009). if quick interpretation is enforced. dropping a trace). In (38) the controller of PRO will then necessarily be ‘Mary. we get OC (cf. yielding two distinct outputs: if transparency is enforced. for instance. but it cannot resolve the interpretation of the empty category at this point (as there is no antecedent yet available for the copy/trace).e. uttered discourse initially. We have relied on the assumption. i. they are transparent to grammars.. Hornstein has argued that its distribution is to a large extent determined by economy considerations: assuming that movement is more economical than pronominalization. We have also assumed that parsers like to resolve the interpretation of empty categories very quickly. because of parsing preferences. what cannot occur here. that an overt pronoun can have ‘Bill’ as antecedent: (40) Himi having to wash behind the ears made Mary angry at Billi The reason for acceptability of (40) under the intended reading is that the pronoun is grammatically permitted in this position and the parser does nothing more than put what it hears where it hears it. then being unparsable may provide a reason for the unexpected unacceptability of instances of NOC where OC/movement is not an option. Following Boeckx and Hornstein (2007). however. 2003. 2001. which is analyzed as an A-trace. 2007) has proposed that.. the resort to pro (i. NOC PRO is essentially a null pronoun (pro).e.. In combination. namely.e.5 Conclusion 209 Mary is doing the washing behind Bill’s ears! Note. OC) fails. If acceptability reflects both generability and parsability. we have seen that the assumption that movement is more economical than pronominalization is. In this chapter. pro. we have argued here that the distribution of pro is a dual function of the grammar and the parser. 6.6. as opposed to OC PRO. Thus. . however. is a null pronoun. The interpretation of an NOC PRO then reduces to figuring out the distribution of pro. insufficient to block some unacceptable instances of NOC whose movement counterpart is not licit. that parsers use the same sorts of principles and target the same sorts of entities as grammars do. NOC) will be sanctioned only when movement (i. common in the parsing literature. these two assumptions provide the beginnings of an account of NOC by predicting where and when pro is available.e.5 Conclusion Hornstein (1999. Section 7. in particular the identification of its antecedent. Typically. or invoke a rich inventory of semantic types or conceptual structures that bypass syntax. One initially plausible way of capturing this locality effect focuses on the observation that the interpretation of PRO.2 discusses some general problems with selectional approaches to control. 7. the antecedent is an argument of the embedding predicate (the subject for subject control.3 focuses on Culicover and Jackendoff’s “simpler-syntax” approach. We have provided many conceptual and empirical arguments in favor of a movement-based approach and against the major alternative syntactic treatments reviewed in Chapter 2. is determined in 210 .4. Many of the arguments we will produce against such approaches are independent of our favorite syntactic analysis and can be formulated in various frameworks. we would like to consider the major non-syntactic approaches to control that can be found in the literature and argue that they are inadequate on various grounds. the object for object control). One of the most obvious is that it requires a local antecedent. In this chapter. The approaches we will focus on seek to reduce control to selection. However. The locality requirement is one of the key features of control configurations and must be accommodated.1 Introduction Much of this book has been devoted to a defense of a specific syntactic approach to (obligatory) control.7 Some notes on semantic approaches to control 7.2 General problems with selectional approaches to obligatory control There are several critical features of the interpretation of OC PRO. Section 7. we think that it is when we contrast non-syntactic approaches to control with a movement-based account that the former more clearly reveal their explanatory weaknesses. The chapter is organized as follows. Finally. a brief conclusion is presented in section 7. 2007). However. Boeckx and Hornstein 2007). as Wurmbrand (2001) shows in detail for German. Chierchia (1984). One linguistically reasonable way of implementing this observation is to treat antecedent identification as a species of selection restrictions that the matrix predicate imposes on its embedded complement clause.2). Rooryck (2001. whereas the one between ec and ‘John’ is one of NOC (see e. those that can be analyzed as involving just a VP layer.g. VPs. which brings back the question of how to identify the content of their OC PRO subjects. Landau (2000).g. it is not structures but relations that are OC or NOC (see section 6. John said that [[eci washing herself] delighted Maryi ] Johnk said that [[eck washing himself] delighted Mary] One selection approach that does not appear to face this problem involves treating control infinitives as properties.2 General problems with selectional approaches 211 some way by the embedding predicate.. Other types of obligatory control infinitivals in German show evidence of a full-blown clausal structure. Kiguchi 2004.g. this proposal can capture at best a subset of control structures. ..7. The virtue of so proceeding lies in the possibility of tracing the locality of antecedent selection in control clauses to the very tight locality that characterizes selection more generally (as everyone knows. rather than clausal complements (see e. Under this view. Hornstein 2001. As discussed in section 6. If control complements are properties. Culicover and Jackendoff (2005).2 1 See e. as illustrated in (1) below. for the same structural configuration may allow both OC and NOC. they can be directly selected by the control predicate and need not involve an empty category such as PRO.e.2.1 Rather than discussing specific technical implementations of this idea.. Hornstein and Kiguchi 2003).. i. the semantic identification of the “concealed” embedded subject is determined by entailment relations triggered by the meaning of the control predicate. here we would like to highlight a couple of problems that seem to be common to all selectionbased approaches. the relation between ec and ‘Mary’ is one of OC (see e. for instance. Chierchia 1984). (1) a. this is a general intuition that underlies both “purely” semantic and mixed semantic–syntactic approaches to control. In other words.g.. 2 See Wurmbrand (2001) for relevant data and arguments for the two types of OC infinitivals in German. b. Wurmbrand (2001). To a greater or lesser extent. In the case of (1). one cannot simply say that a given predicate selects for an OC structure. selection in natural languages is confined to very local domains). The first one regards what is being selected. and copycontrol constructions wear their clausal structure on their phonological sleeves.2).212 Some notes on semantic approaches to control In previous chapters we in fact discussed three kinds of data that argue that control configurations are indeed syntactically clausal/propositional and do have embedded subjects.eat Mike grasshopper ‘Mike wants to eat grasshopper’ If establishing an OC relation cannot be obtained through direct selection. On the assumption that control involves the same grammatical relation across languages. In effect. The first case involves finite control into indicatives in Brazilian Portuguese. as illustrated in (3) (see section 4. We claim that it is exactly .2. 3 See. (2) Brazilian Portuguese: [[O Jo˜ao] disse que [o pai d[o Pedro]] acha que ec vai The Jo˜ao said that the father of-the Pedro thinks that goes ser promovido] be promoted ‘Jo˜aoi said that [Pedroj ’s father]k thinks that hek/∗ i/∗ j/∗ l is going to be promoted’ (3) Tsez (Polinsky and Potsdam 2002): biˇsra] yoqsi] [1/∗2 [kidb¯a1 ziya girl.want Mike IRR.4). backward. where the embedded clause displays the same clausal material that is present in its non-control counterpart with an overt subject (see section 2. since the antecedent of the infinitival subject is already pre-specified as part of the meaning of an obligatory control predicate.3.3). a syntactic subject and the application of (syntactic) control mechanisms is in a sense vacuous in obligatory control constructions.INF began ‘The girl began to feed the cow’ (4) San Lucas Quiavin´ı Zapotec (Lee 2003): R-c`aa` a’z Gye’eihlly g-auh Gye’eihlly bxaady HAB. something that should be impossible if control complements were VPs.5. the third case involves copy-control sentences such as (4). for instance. the other option is to attempt to enforce the desired relation by some sort of indirect selection. where the controller is pronounced in the embedded clause. Wurmbrand (2001: 250) on obligatory control involving infinitivals larger than VP in German: “Assuming that obligatory control is determined lexically/semantically.5. The second case involves backwardcontrol configurations.3 Consider for instance an object control construction such as (5). where the embedded subject is realized as a copy of the controller (see section 4.ERG cow. Finally. the existence of backward-control and copy-reflexive structures indicates that the controller in OC structures has moved from an embedded clausal complement.ABS feed. as illustrated in (2) below.5. Things are actually much worse. for instance). Though not implausible. It must be due to some feature of OC PRO. this kind of transitivity of selection must be used sparingly for. selection is no longer local and thus is not suited to explain the locality of antecedent selection in control clauses. just as C selects T. T does not impose selection restrictions on arguments of the predicate. but no other head further down. this redundancy that licenses (but does not necessitate) the omission of a syntactic subject in obligatory control constructions. comports badly with stative predicates: ∗ ‘John is knowing/understanding/recognizing the problem. given that C selects T. For example. it voids the attractive locality feature of a selection-based account. if not restricted. However. For example. But. it is not clear how to prevent it from working in unwanted cases. requiring that the subject be animate or that the object have some specified thematic role or be some kind of DP (a reflexive. But on a non-movement account. ‘persuade’ can select C. this requirement must be an intrinsic property of OC PRO. at the very least. more problems surface. TP].2 General problems with selectional approaches (5) 213 [John [VP persuaded Mary [CP C [TP PRO to leave]]]] Given that head-to-head selection is a very local process. how they do so and what information can get imparted. One of the core properties of OC PRO is that it needs an antecedent. So. what prevents a matrix verb from imposing restrictions on the argument of a verb embedded two clauses down if transitivity of selection is assumed? Generalized in this way. What intrinsic property? One could say it is [+anaphoric] and this is what requires that it have an antecedent.. e.” . If these chains of indirect-selection relations could be licensed. it seems plausible that ‘persuade’ indirectly imposes restrictions on the embedded T in virtue of selecting C. T selects V.’ Nonetheless. One can then attempt to stretch this reasoning and assume that ‘persuade’ could also impose restrictions on the element occupying the embedded [Spec. for instance. What feature? On the MTC the fact that OC PRO requires an antecedent follows from OC PRO being a residue of movement and one cannot find a residue of movement without something having moved.7. Progressive-present tense. Assuming that a technical way out of the transitivity problem may be found. as soon as we resort to transitivity of selection for control. there should in principle be room for accommodating the fact that in (5) the subject of the embedded clause must be controlled by the matrix object. a selection-based account of control needs to specify exactly which heads interact.g. We do not see how this can be achieved in a non-stipulative fashion. so far as we know. Why? This cannot be a matter of selection. Attempts to eliminate the diacritic smell of selectional approaches by couching the interpretation of OC PRO on other semantic grounds have not been successful either. Identification of [unrealized] (C-)T with the [unrealized] subevent entails identification of (C-)T anaphoric phifeatures with the phi-features of all and only those argument(s) contained in the [unrealized] subevent. 287). To explain what we find requires showing why the analogue of OC PRO without the [+anaphoric] feature does not exist in “OC environments. Observe that this does not imply that such an approach to control is incorrect.” Why is there no analogue of OC PRO without this feature? Tracing the requirement that OC PRO have an antecedent to a [+anaphoric] specification only tracks the observed fact. Kim forced Sue to leave Subevent structure: [Tr e1 ∗ act (Kim. leave)) {en+1 } leave (Sue)]Tr Syntactic structure: [V force] ( . Rooryck’s (2007) approach. Take. resulting in a subevent at an undetermined moment after e1 ∗ . which do the job of PRO.” Anything short of this does little more than redescribe the facts. a cover term for accomplishments and achievements): (6) a. Perhaps.214 Some notes on semantic approaches to control However. we should be dragged kicking and screaming to this rather nihilistic conclusion. for instance. One must explain why it must be [+anaphoric] in “OC environments. such an approach cannot be extended to finite-control sentences such as (2) in Brazilian Portuguese. d. under which the sentence in (6a) is analyzed as in (6b) and (6c) (Rooryck 2007: 287. The tense of the indicative complement . ) [V leave]]  [CP (C-)T      {en+1 } {Tunrealized } Plain English: At the event time e1 ∗ . Kim undertakes action with respect to Sue’s leaving. However.” Putting aside the conceptual issue of why temporal identification should provide the basis for argument identification. it is clear that this is not so much a theory of control as the claim that control has no theory. But it implicitly denies that there is anything interesting to be said about the properties of control configurations beyond specifying that what we find is due to the idiosyncratic lexical properties of OC PRO and the special selectional/ semantic features (very broadly understood) of the embedding predicate. Control thus rides piggyback on the temporal identification of the infinitive by the matrix verb. at which Sue leaves. Methodologically. . this is not enough. Tr stands for transition. In Rooryck’s own words (p. c. (Sue. as the feature on OC PRO that forces the presence of an antecedent boils down to a diacritic. . “infinitival [unrealized] (C-)T has anaphoric phi-features. b. 4). Wurmbrand (2005) shows that not all control complements show the same kinds of temporal dependencies.5. Adjunct-control PRO requires a local c-commanding antecedent: Johni said [that [Maryk ’s brother]m left [after PROm/∗ i/∗ k/∗ w eating a bagel]] Adjunct-control PRO only licenses sloppy readings under ellipsis: John left before PRO singing and Bill did too ‘and Billi left before hei /∗ John sang’ Adjunct-control PRO can only have a bound interpretation when controlled by only-DPs: Only Churchill left after PRO giving the speech ‘[Nobody else]i left after hei /∗ Churchill gave the speech’ 4 As we saw in section 2. Ontem [o Jo˜ao]i disse que eci/∗ k vai viajar amanh˜a Yesterday the Jo˜ao said that goes travel tomorrow ‘Yesterday John said that he’s going to travel tomorrow’ c Ontem [o Jo˜ao]i disse que eci/∗ k tinha viajado na Yesterday the Jo˜ao said that had traveled in-the semana passada week past ‘Yesterday John said that he had traveled last week’ Let us finally turn to – we think – the most lethal problem for selection-based accounts.2 General problems with selectional approaches 215 clause is completely independent from the matrix clause.4 (7) a. Thus. as illustrated by (7). adjunct control shows all the characteristics of complement control. One prominent property of selection is that it holds for complements. Brazilian Portuguese: Ontem [o Jo˜ao]i disse que eci/∗ k est´a estudando muito nos Yesterday the Jo˜ao said that is studying much in-the u´ ltimos dias last days ‘Yesterday John said that he has been studying a lot lately’ b. c.1. as illustrated in (8).5. This. The distribution of the latter is quite free. one expectation of a selection-based account of control interpretation is that control into adjuncts should have properties quite different from control into complements. however.2 and 4. at least for a large class of adjuncts.1. (8) a. b.7.5. not adjuncts. As we have discussed in section 4. so it is not clear that we could generalize the above approach. we have proceeded on the premise that it is prima facie reasonable to codify antecedents in control configurations in terms of selection. . is incorrect. and yet OC obtains as in cases of non-finite complements (see sections 2. So far.2. First. “the control properties of heads . And third. under a movement approach of the sort we have advocated. though they note (p. Some notes on semantic approaches to control In the appropriate type of adjuncts (e. 445. the central problem in control consists in “identifying the factors that determine possible controllers in any given circumstance” (2005: 416) (i.g. 416). controller selection is largely a function of the “lexical semantics of the predicate that selects the infinitival or gerundive complement” (p. which assumes a very rich. footnote 19).6 This lexical meaningbased perspective relies on Jackendoff’s (1983. footnote 1) that their arguments are “largely neutral” concerning PRO. . couched in terms of conceptual structure” (p. follow insofar as possible from their meanings.7 Second. The parallel between adjunct and complement control suggests that such accounts are on the wrong track..5 Let us now consider Culicover and Jackendoff’s “simpler-syntax” approach. In other words. generative apparatus outside narrow syntax. 2005) propose an approach to control that aims to explain the interpretive properties of control structures directly in terms of the lexical meanings of their constituent parts. determining the antecedent of “PRO”). 7. .2.. . Crucially.3 “Simpler syntax” Culicover and Jackendoff (2001. As we find the exposition is streamlined if we include PRO. C&J make three principal claims regarding control. purposives). we do so in what follows. adjunct-control PRO obligatorily requires a de se interpretation: The unfortunate wrote a petition (in order) PRO to get a medal ‘[The unfortunate]i wrote a petition so that [he himself]i would get a medal’ That adjunct control shows all the diagnostic properties of controlled PRO is somewhat unexpected if these properties are the result of selection restrictions imposed by a higher predicate on its complement. 416.e. The following quote well illustrates 5 See section 7.216 d.1 below for additional problems to semantic approaches to control brought up by the interaction of adjunct control and wh-movement. the control relation is not simply a diacritic feature of a given verb or class of verbs but is “an organic part of their meaning” (p. 6 As Culicover and Jackendoff’s approach to control is more fully developed in their 2005 book. 7 C&J avoid using the PRO-notation in discussing control.5. where the requisite aspects of meaning are explicitly represented in the predicate’s conceptual structure.1). 420). 1990) framework of conceptual semantics. we will base our critique on that work (henceforth C&J).3. control into adjuncts falls out from the same mechanism as control into complements (see section 4. there could not be a verb that meant the same thing as persuade but had the control behavior of promise” (p. b. . it might plausibly be able to control a PRO. Let us consider them in turn. it cannot be the antecedent of PRO on the assumption that adjectives cannot antecede PRO. They simply cite the facts and conclude that they speak for themselves. Every American1 military attempt at pacification in Iraq was a failure. we argue that C&J’s approach fails in its explanatory ambitions and reduces to little more than a list of cases. we will begin our discussion of C&J’s treatment of control by defusing the critical points they make against the MTC. . 7. .3 “Simpler syntax” 217 C&J’s core conception of what control is: “Our own intuition . For this reason. which is taken to show that OC PRO is licensed despite the absence of an appropriate antecedent in the syntactic structure. they offer several arguments against the MTC. the MTC should analyze it along the lines of (11). As such. 7.’ However. as illustrated in (10) below. The reasoning is supposed to go as follows. Ultimately.3. As C&J develop this lexical-semantic perspective to control.1. Then we turn a critical eye towards C&J’s own proposals. It1 was finally forced to approach the UN Every American presentation of itself in international venues is appalling Assuming that (9) indeed involves OC. (10) a. 420).8 PRO in (9) is interpreted as ‘America’ though it does not appear in the structure. where ‘American’ gets the external ␪-role of ‘dominate’ and 8 C&J do not actually spell out the problem that (9) poses for syntactic theories of control. (9) An American attempt [PRO to dominate the Middle East] But why assume that ‘American’ in (9) cannot be a controller? It appears to be able to “corefer” with a pronoun and even bind a reflexive. is that the control behavior of persuade and promise is an essential part of their meanings.7.1 Control by adjectival expressions One of the objections raised by C&J involves sentences such as (9) below.3.1 Some putative problems for the movement theory of control C&J offer three major kinds of objections to syntactic approaches to control. The most similar potential antecedent is ‘American. We hope that we have thrust the right words into their mouths. if what ‘American’ is doing in (9) is not typical of adjectives. In sum. as it stands. but this seems not to hold in these cases. [10b]). Such binding normally requires categorical identity. i. adjectives like ‘American’ seem to be able to bind nominal anaphors (cf.218 Some notes on semantic approaches to control then moves up to the thematic domain of ‘attempt’ and receives its external ␪-role. but the sentences are understood as the attempter being the leaver and the person ordered leaving. (11) [DP an [NP American attempt [CP C [TP American PRO to [vP American dominate the Middle East]]]]] Is the movement from a nominal to an adjectival position in (11) illicit? It is not actually very clear what is wrong with this sort of derivation. Any such attempt (by Fred) [PRO to leave] will be punished Yesterday’s orders (to Fred) [PRO to leave] have been cancelled C&J provide a possible way out for the syntactically inclined. alternatively one could stipulate a null by-phrase in [12a] and a null to-phrase in [12b]” .3.e. Moreover.. then the interrelation between conceptual structure and syntax will have to reflect this exceptional behavior by expressly permitting it. If not. They note (without much enthusiasm) that “[o]ne could stipulate a phantom position in the specifier that can never be realized by anything but a null NP. The alleged problem with (12) is that there is no controller at all in the overt syntax (ignoring the parenthetical material). Such an account would have to allow the adjective ‘American’ to saturate the attempter ␪-role in the mapping between conceptual structure and syntax despite the fact that adjectives do not typically do this. In other words.1.2 (Apparent) lack of antecedent for PRO Another challenge that C&J pose to the MTC involves sentences such as (12) below. the observation that sentences like (9) involve control does not in itself argue against syntactic accounts or for semantic ones. it is likely that a parallel assumption would have to be made were one to provide an account of (9) in semantic terms. it is not clear why cases like (9) are more problematic for syntactic theories than for semantics-based accounts. there is no obvious reason why the same categorical laxity should not extend to movement. b. For whatever reason. Unless it can be demonstrated that the quirky behavior of adjectives like ‘American’ is actually smooth and just what one should expect when viewed from C&J’s perspective. as far as we can tell. (12) a. 7. it is ‘Fred’ when the PP is overt. the route motivated on theory-internal grounds is methodologically preferred. In fact.9 C&J note (2005: 418) two further cases where there is no conceivable source for the controller: (13) a. otherwise. Nunes (2008b).10 the sentences in (13) are not cases of OC but of NOC. inherent case is not overtly realized if its bearer is a phonologically empty category. This is incorrect. and Nunes (2008). it receives the relevant ␪-role and can be licensed by the inherent case associated with nominals (see Chomsky 1986b). Note. these are cases of what C&J call free control and we call non-obligatory control. How about PRO taking a swim together? PRO undressing myself/ourselves/yourself in public may annoy Bill Here C&J suggest that only at pain of adopting Ross’s (1970) theory of performative deletion could these facts be accounted for in a syntactic theory of control like the MTC. If the theory that provides the requisite “internal grounds” is desirable independently (which we think the MTC is). among others. there is nothing untoward about having a null pronoun move from within the embedded clause of (12a) and (12b) to the thematic domain of ‘attempt’ and ‘orders.’ Deverbal nouns arguably have thematic requirements analogous to those of their verbal counterparts and project into the syntax in a well-behaved UTAHlike manner (see Baker 1997). that. Thus. Martins. but are instances of pro. then argument is needed not to follow the theory’s promptings. Independent of Ross’s proposal. Can you think of any way they can clean up? How about washing each other in the outdoor shower?’ Consequently. However. given the discussion of pro in Chapter 6. b. after a pro moves to the thematic domain of ‘attempt’ and ‘orders’ in (12). as we do not share their aesthetic. they are not the residues of A-movement. 10 Ross’s earlier analysis has been revived recently in Etxepare (1998) and Nissenbaum (2000). We can see why C&J’s global theoretical predilections for a very spare syntax would find these motivations weak. pace C&J. this is not 9 As argued by Hornstein. it is not clear why we should not adopt their apposite proposal. C&J claim that this move would be problematic as it would be motivated on theory-internal grounds and hence with “no independent syntactic motivation. and Kato and Nunes (2009). . This places the burden of proof on C&J to demonstrate that such an extension fails empirically for.3 “Simpler syntax” 219 (2005: 418). As such. incidentally. these PROs can be anaphoric to anything: ‘Bill and Martha are filthy from all the work in the garden. Depending on the conversational context.” Why theory-internal grounds in general are not motivation enough is unclear to us.7. a false dichotomy has been drawn here between lexical meaning and syntactic structure. 419) asseverate that only “the dogma that control is syntactic” could motivate a syntactic distinction in these cases on which to rest an explanation for the witnessed differences. C&J (p.” 7.1 that.1. in some cases “the same syntactic configuration can be associated with different controllers” and in other cases “the controller can appear in different syntactic configurations. b.3. 419).220 Some notes on semantic approaches to control a particular problem for the MTC or any other syntactic account unafraid of null elements and not wedded to the specific theoretical assumptions in C&J’s “simpler syntax. (14) and (15) illustrate cases of same structure but different controllers and (16).5. As such. A predicate’s thematic-argument structure and how this structure is syntactically realized are closely connected.3 Apparent syntax–semantics mismatches C&J advance a third kind of argument against syntactic accounts of control. Since the earliest days of generative grammar it has been recognized that these two notions are closely related. c. b. while the latter does not. ‘Sarah’ in (14) is a direct object when . it is reasonable to suppose that if predicates have different argument structures they also have different syntactic structures. cases of different structures but same controller. (14) a. but we see precious little beyond confident assertion here. ‘Fred’ is the antecedent for PRO despite the “blatant” syntactic differences among the examples. b. contrary to surface appearances. d. This may be correct. (15a) contrasts with (15b) in that the former allows an impersonal (gen) reading with PRO. For C&J. Johni talked about [PROi/gen dancing with Jeff] Johni refrained from [PROi/∗ gen dancing with Jeff] (16) a. while preserving meaning.” According to them. Using such an assumption based on Baker’s (1997) UTAH. First of all. Johni persuaded Sarahj [PROj/∗ i to dance] Johni promised Sarahj [PROi/∗ j to dance] (15) a. we argued in section 5. Bill ordered Fredi [PROi to leave immediately] Fredi ’s order from Bill [PROi to leave immediately] The order from Bill to Fredi [PROi to leave immediately] Fredi received Bill’s order [PROi to leave immediately] In (14) PRO is controlled by ‘Sarah’ with ‘persuade’ but by ‘John’ with ‘promise.’ In turn. “intuition suggests that the differences are a consequence of what the verbs mean” (p. As regards (16). They claim (2005: 419) that “the choice of controller can be doubly dissociated from syntactic configuration.” that is. C&J do not. We believe that the ‘persuade’/‘promise’ contrast is a good illustration of this. . a lexical-semantics account seems reasonable. in the latter case. to our knowledge. allows PRO to be interpreted impersonally.’ This difference accounts for why.3 “Simpler syntax” 221 the complement of ‘persuade. Example (15a). as attested by the examples in (17).7. John talked about one’s/her dancing with Fred John1 refrained from ∗ one’s/∗ her/his1 /PRO1 dancing with Fred Example (17b) shows that ‘refrain from’ requires that the subject of its gerundive complement be anteceded by the subject of the predicate regardless of how the gerundive subject is expressed.’ but the object of a null preposition (as in dativeshifted structures) when associated with ‘promise. Consider now the contrast in (15). Generative grammarians have regularly noted that surface form can be misleading. Of course. even when it is unclear how to develop the account in detail. Why so? Chomsky (1986b) proposed the following rule of thumb: if an overt impersonal is disallowed in a certain structure. one might attribute the oddities in (17b) to a fact about the meaning of ‘refrain from. In the case of ‘talk about’ vs. However.5. it implies one can only refrain from one’s own activities. then it suggests that considerable care must be taken when diagnosing underlying syntactic structure from apparent surface form. (17) a. if it is not. It is just a bald statement saying that whatever ‘refrain from’ means. If this is correct. in the former. Note that. then it is no surprise that a covert one with PRO is barred as well. it is not. We have no problem with this sort of answer in this case precisely because the constraints ‘refrain from’ imposes on interpretation are insensitive to the particular syntax used to express it. their double-dissociation argument weakens.. b. even if this account 11 To say this is not yet to explain in virtue of what this makes no sense. ‘refrain from’ it is very plausible that part of the relevant difference reflects the meanings of these predicates. it is plausible that part of the contrast revolves around the semantics of the matrix predicates in these constructions and not specifically with the interpretation of PRO. When this obtains. Given this. as argued in section 5.’ e. but not (15b). pace C&J’s assumption that the syntax of ‘persuade’ and ‘promise’ in (14) is the same.1. semantic coherence obtains and the sentences are acceptable. ‘John’ is a possible controller while.11 When identity of reference between subjects is established (be it by binding or control). explain why this contrast exists besides attributing it to the lexical meanings of the predicates involved.g. it makes no sense for one to refrain from anyone else’s doing something. and Chapter 6 above). Thus. if this is correct. In other words. but A-movement out of genitiveassigning gerunds is not. at least if we test the relevant cases using alternative syntactic realizations. it is useful to contrast ‘refrain from’ with ‘talk about. ‘persuade’: (19) a. To get a better bearing on when a lexical-semantics account is apposite. Consider one more illustration of the problem. Note that one cannot simply attribute this to the meaning of ‘talk about’ for.g. as ‘him’ cannot be interpreted as anteceded by ‘John’ in this configuration. However. it is far from clear that the meaning of the predicate precludes the control relations claimed to be impossible. (18) Mary1 said that John talked about her1 /∗ PRO1 washing herself C&J tend. whatever its meaning is. b. movement. b. If pronouns can only be used where movement is forbidden (see Hornstein 2001. the fact that different pronouns are semantically coherent in (17a) raises the question of why the antecedent of PRO in (15a) remains ‘John’ even in cases like (18) below. . then some account of this fact that does not rely exclusively on the lexical semantics of ‘talk about’ is required. to treat all cases of control as if they were like the ‘refrain from’ example above. it does not imply that the PRO in (15b) is not a trace of movement. as the MTC would propose). Johni promised Billk PROi/∗ k to leave on time Johni persuaded Billk PROk/∗ i to leave on time 12 Note that replacing ‘his’ with ‘him’ in (17b) renders the sentence unacceptable. in our view. 2007.. C&J’s favorite contrast involving ‘promise’ vs. as the acceptability of ‘one’s’ and ‘her’ in (17a) demonstrates. where an overt pronoun can have ‘Mary’ as antecedent.12 There is no contradiction (or even infelicity or dogma) in assuming that the meaning of ‘refrain from’ demands certain antecedence relations and at the same time assuming that these relations are established by certain grammatical processes in cases like (15b) (e.’ The latter’s meaning contrasts with the former’s in tolerating a DP in the subject of its gerundive complement that is not controlled by its subject. this suggests that A-movement out of accusative-assigning gerunds is allowed. it is clearly compatible with having ‘Mary’ as antecedent (as seen if we replace PRO with a pronoun). But. in most of the cases of interest. it cannot be that the required interpretation of (15a) with ‘John’ being antecedent of PRO follows from the meaning of ‘talk about’ in a way parallel to ‘refrain from’ in (15b). Johni promised Billk that hei/k would leave on time Johni persuaded Billk that hei/k should leave on time (20) a.222 Some notes on semantic approaches to control of the semantics of ‘refrain from’ is correct. it appears to us that.’ That being so. Thus. hence defeasible given sufficient data. C&J adopt a version of the MDP which they interpret in markedness terms and which can be overridden by “predicate and complement semantics.. which is indeed correct. Rosenbaum argued that ‘promise’ was not a counter-example to the MDP despite its being a subject-control predicate. In fact. . In the body of their control chapter. why does it fail to work its magic in (19)? It would appear that more than the meaning of the embedding predicate is relevant. 435.” This sounds very much like Rosenbaum’s original proposal concerning the MDP. The cases that C&J mention are cases of NOC and. but not in cases of NOC. Nonetheless. John talked to Sarah about defending himself). i. According to C&J (p. why is it that in (20a) the only possible antecedent is ‘John’ while in (20b) it is ‘Bill’? Note that the embedded predicates in all the cases of (19) and (20) are “actional” (using C&J’s categorization).7.” .” Note. the MDP is subsumed under minimality (a condition on movement). 432). despite invidious remarks to the contrary. have aimed to bridge the gap between what lexical semantics plausibly supplies and what is actually observed.4. which explains why ‘John’ is not a potential controller in (20b) (see section 3. Chomsky’s work on the acquisition of ‘promise’ and noted that its late mastery is just what we would expect if the MDP were a markedness condition on determining controllers. as such. they should not show MDP/minimality effects. Chomsky [1969]) to get subject control readings.g. including the MTC. as shown by C. a full account of the control properties involved here is underdetermined by the lexical semantics of the embedding predicates. the control properties of ‘promise’ and ‘persuade’ are nothing like those of ‘refrain from. If the “meaning” of the matrix verb operates to fix antecedence in (20).4. in (19) the antecedent of ‘he’ is not fixed while in (20) the antecedent of PRO is. As such. C&J’s attitude towards MDP is somewhat difficult to discern. 440). and this may be a default constructional meaning that makes it hard for some speakers (especially young ones. Syntactic theories of control. Rosenbaum’s (1967) minimal-distance principle (MDP) should be abandoned because it “fails to account for long distance control in subject complements and for free and nearly free control in object complements (e.13 In sum. precisely because of its odd acquisition profile. However. footnote 10) they concede that “[i]t is true . that this poses no problem at all for the MTC. Minimality does not block similar movement in (20a) as ‘Bill’ is within a PP (see section 5. however. We have argued that PRO is a residue of A-movement and movement to the matrix-subject position in persuadeconstructions violates minimality. they repeatedly claim that the MDP is empirically inadequate (see pp. that there is a strong bias toward interpreting NV V NP to VP as object control. within the MTC. which was understood as a markedness condition. . As the movement is licit. in a curious footnote (p. 434.3 “Simpler syntax” 223 Example (19) shows that the embedded pronoun can be coherently anteceded by either ‘John’ or ‘Bill.’ 13 One additional point. At the very least. Recall from section 3. ‘John’ can control PRO in (20a). the MDP/minimality is relevant only when movement obtains – in cases of OC.1).5. Rosenbaum cited C. 432. the syntactic structure involved also contributes to the attested control properties.e.1 that.1). Thus. Bill’s order to Fred [PRO to ready ourselves for departure] c. Mary saw the order from Bill to Fred [PRO to ready ourselves/oneself/ herself for departure] b. it does not seem to be the case that in (21) we necessarily have a single interpretation associated with different syntactic forms. depending on the context. The different syntactic configurations in (22). . Thus. the import of C&J’s point regarding the examples in (21) relies on two assumptions: (i) that the same control relations are realized in all these configurations. they would be no more problematic than the sentences in (22). With this understanding of the to-PP. Do verbs and their nominalized counterparts have the same interpretive properties? Not completely. Bill ordered Fredi [PROi to leave immediately] Fredi ’s order from Bill [PROi to leave immediately] The order from Bill to Fredi [PROi to leave immediately] Fredi received Bill’s order [PROi to leave immediately] (22) a. This. is not particularly surprising given the assumption that different syntactic configurations can arise from the same basic underlying form. c. b. Were the sentences in (21) syntactically related. different syntactic structures result in identical control configurations. (21) a. repeated below in (21). ∗ Bill ordered Fred [PRO to ready ourselves for departure] In (23a) and (23b) ‘Fred’ can be interpreted as the person registering the order though not necessarily ordered himself. Such a reading is not available in (23c). as implied by C&J. are not a problem for any syntactic theory of control (including the MTC) given the standard assumption that actives and passives are transformationally related. for example. d. one can interpret PRO rather freely. the to-PP inside a nominal is not an infallible guide to the thematic role of its complement or. Though the sentences in (21) all have interpretations in which PRO is controlled by the individual who has been ordered. (23) a. and (ii) whether it is reasonable to suppose that they share an underlying form. John ordered Bill PRO to leave Bill was ordered (by John) PRO to leave Thus. Let us consider these questions in turn. to put it in different words.224 Some notes on semantic approaches to control Consider now the observation that in (16). in itself. and (ii) that these different configurations are not syntactically related. So the relevant questions are: (i) whether the interpretive properties of verbs and their deverbal nominalizations are indeed identical. (23a) and (23b) are acceptable while (23c) is not. b. for example. this need not be identical to the recipient of an order. to our knowledge most generative grammarians assume that there is a syntactic relation between verbs and their deverbal counterparts. Here categorical identity is a rather surfacy property. 16 In fact. that there are compelling arguments against verbs and their nominalizations sharing a common underlying syntactic form. If this is correct. the main argument C&J have against syntactic accounts of control in general and the MTC in particular is that the requisite syntactic structure 14 This was the assumption in Lees (1960). given the state of flux concerning the details of the relation. via UTAH).. at least to us. If this is so. . this assumes that the semantic properties of nominals are determined directly from the structure of the nominal and not via some relation to its corresponding sentence. This is certainly not an exotic assumption and it has been fairly common since the earliest days of generative grammar. there is a reading of (21b–d) in which ‘Fred’ is understood as having been ordered and. Chomsky (1970) argued against Lees’s transformational analysis but not against the view that verbs and nouns could have parallel underlying representations.16 In sum. We also accept their implicit argument that the MTC is committed to the claim that verbs and their deverbal nominalizations are syntactically related. it controls PRO in the nominal cases. although they do assume that they share an underlying conceptual structure. This is certainly implicit in ‘Remarks’ where it is assumed that the object of ‘destruction’ and ‘destroy’ are semantically and syntactically analogous. The same basic assumption carries over to recent distributed-morphology analyses.3 “Simpler syntax” 225 But putting such cases aside. just as in the verbal counterpart in (21a). their conclusion that this necessarily requires computing control non-syntactically at the level of conceptual structure does not follow. Suffice it to say that we know of no reason why movement could not operate within either domain with the same interpretive effects if we assume that the way thematic information projects within the verbal and nominal domains is guided in the same way (e. We leave this possibility aside for now.3).g. then nothing prevents treating nouns and verbs as essentially derived from a common structure that is categorically neutral.7. under this reading. it is not clear. although C&J are right that one should strive for a unified account of control with nominal and verbal predicates. 15 Actually.14 C&J present no arguments against it and so we see no reason as yet to reject it. However.15 If control is due to movement in the former then it is also due to movement in the latter (see section 5. However. we have nothing concrete to propose about the structure of verbs and nouns. This conclusion requires arguing that verbs and their deverbal nominals are not syntactically related and we see no reason to accept this assumption at this point. This brings us to our second question: do verbs and their deverbal nominalizations share a common underlying syntactic form? C&J presume that they do not. which indicates that control within nominals strongly parallels what we find in their verbal counterparts. 2.1 Adjunct control As discussed in section 7. it is very unclear how C&J could integrate adjunct control into their basic story. In fact.5. .1 and reviewed in section 7. when there is an adjunct in need of a controller. So why should this be if obligatory control is primarily a lexical-semantic fact about embedding predicates? Given that adjunct control exhibits all the diagnostic properties of complement control but without the specific lexical predicates whose semantics is supposed to ground these properties. in adjunct-control configurations.2. though not one that is particularly exotic given current assumptions. we believe that C&J fail to fully appreciate how severe a challenge adjunct control is to their entire enterprise. 7. it will be a backward step when compared to earlier accounts that pursued a conceptually more appealing route subsuming adjunct-control cases to the MDP. As such. C&J concede that “[o]ne point in control theory where some syntactic constraint seems unavoidable is in the control of adjuncts” (2005: 425).3. 416).226 Some notes on semantic approaches to control is unavailable. But. the local syntactic subject must function as the unique controller will not have any explanatory heft. yet the very same properties associated with complement control arise. their arguments against syntactic approaches like the MTC rest on premises that syntactic-centric theories have no problem rejecting and. consequently. As discussed in section 4.” control is fundamentally a fact about “the lexical semantics of the predicate that selects the infinitival or gerundive complement” (p. For the MTC to capture these parallels in terms of movement requires tolerating a more abstract syntax than C&J seem comfortable with. However. adjunct control shows all of the typical marks of complement control.2. In fact.2 Challenges for “simpler syntax” Let us now consider some problems for the approach based on conceptual structure advocated by C&J. adjunct control is quite problematic for semantic approaches to control and the “simpler-syntax” approach is not exceptional in this regard. 7. most of their objections simply beg most of the relevant questions. The strongest argument revolves around the parallel properties of nouns and verbs as regards control. there is no special property in the lexical semantics of the embedding predicate that could determine control.3. Stipulating a very specific rule to the effect that. In an approach like “simpler syntax. By contrast.7. depending on whether or not the matrix object undergoes wh-movement. Curiously. Thus. 18 For original discussion of the finite counterparts of (24) in Brazilian Portuguese.5.’ It is unclear why not. the fact that they show identical properties is no surprise. (ii) a. if indeed it is the implicit agent that is doing the controlling. (24c) has whmovement and now both subject and object control are possible. he shows that in Brazilian and European Portuguese the subject of infinitival adjunct clauses may be controlled by the matrix subject or the matrix object. However. However.’ Example (iib) is more unusual as it seems that the implicit agents of the sinking are the collectors. as in (24a). with no wh-element involved.18 Example (24b) has a wh-in situ in the matrix clause and the result is subject control. More specifically. For example.1. even here there must be something else going on. this interpretation is not available in other cases of adjunct control: ∗ ‘The ship sinks before/after furthering the plot. as illustrated in (24) below. The ship sinks (in order) PRO to further the plot b. the MTC accommodates adjunct control via sideward movement. This allows the MTC to give a unified account to both complement and adjunct control. and Nunes (2008c).3 “Simpler syntax” 227 As noted in section 4. Example (iia) can be seen as a typical case of event control with the paraphrase ‘The ship sinks in order for its sinking to further the plot. 147) also discuss cases such as (ii) below. as we have shown. The distinctive properties of these constructions can be largely derived given recent minimalist assumptions. 17 C&J claim that even in the case of adjunct control “some nonsyntactic influence is necessary” (p. things are even worse than this for C&J’s analysis. these cases of control do not license anaphors. even with the by-phrase present: ∗ ‘The ship was sunk (by John) to make himself famous. Rodrigues (2004). (i) Such a brutal interrogation of the suspect without PRO considering the legal repercussions could lead to disaster C&J (p. as it would be on a mixed theory of the kind mooted in “simpler syntax. Recent work by Nunes (2008c) on adjunct control in Brazilian and European Portuguese shows that independent syntactic properties may yield subject or object control in adjunct-control configurations. which illustrate the so-called “event” control originally discussed by Williams (1985). . see Modesto (2000).”17 In fact. The unacceptability follows if even here we have some kind of event control. They note that there can be control of adjuncts by implicit arguments within nominals in examples such as (i) below and conclude that this tells against syntactic approaches. a claim that we have argued is contentious.’ PRO is not controlled by an implicit agent but by the event. 426). The ship was sunk (in order) PRO to collect the insurance C&J analyze these as involving control by an implicit agent. whether this is so is quite unclear. a grammatical option made available once D-structure is dispensed with. However. this position again assumes that the syntax cannot have phonetically null arguments. for merger of ‘a Maria’ in this position is more economical. ∗ O Jo˜ao [[cumprimentou [quemuF ]i ] [depois de ti entrar The Jo˜ao greeted who after of enter. let us now consider the contrast between (24b) and (24c) regarding object control.5.1. In the case of (24a). if ‘o Jo˜ao’ is in the subject position of the adjunct clause. with optional whmovement. O Jo˜aoi cumprimentou quemk depois de PROi/∗ k entrar na sala? The Jo˜ao greeted who after of enter in-the room ‘Who did Jo˜ao greet after entering the room?’ c. So. Nunes argues that they involve the derivations sketched in (25a) and (25b) respectively. Moreover. (25) a. Bearing this in mind. for instance. Quemk e´ que o Jo˜aoi cumprimentou ti depois de PROi/k entrar Who is that the Jo˜ao greeted after of enter na sala? in-the room ‘Whoi did Jo˜aok greet after hei/k entered the room?’ Assuming with Boˇskovi´c’s (2007) that the strong feature that triggers successive cyclic movement is hosted by the moving element. it cannot undergo sideward movement to the complement of the matrix verb. after ‘a Maria’ is merged. na sala in-the room ‘Jo˜ao greeted Maria after entering the room’ b.INF na sala]] in-the room ‘Whoi did Jo˜ao greet after hei entered the room?’ b. Recall that subject control over object control is enforced in adjunct control due to merge being more economical than move (see Hornstein 2001 and section 4. yielding subject control. the presence of this feature in the derivation has consequences for economy computations regarding merge-over-move. [quem√F ]i e´ que o Jo˜ao [[cumprimentou ti ] [depois de ti Who is that the Jo˜ao greeted after of entrar na sala]] enterINF in-the room ‘Whoi did Jo˜ao greet after hei entered the room?’ .1 above). vP]. this feature is lexically optional on wh-elements. ‘o Jo˜ao’ can only move to the matrix [Spec.228 Some notes on semantic approaches to control (24) European and Brazilian Portuguese (Nunes 2008c): O Jo˜aoi cumprimentou a Mariak depois de PROi/∗ k entrar The Jo˜ao greeted the Maria after of enter a. Nunes (2008c) proposes that in languages like Brazilian and European Portuguese. 19 In conclusion. C&J’s treatment does not address any of the other factors often taken to be central to control phenomena. curiously.2. 7. it is not at all obvious how C&J can incorporate adjunct control in their system as it attempts to derive obligatory control from semantic relations between embedding predicates and their complements.INF na sala]] in-the room ‘Whoi did Jo˜ao greet after hei entered the room?’ . If the wh-element sits in the subject of the adjunct clause and sideward movement to the matrix-object position is possible. Now. but can also account for cases where adjunct control may be of both the subject and the object type. (25a) is unacceptable not because movement of the wh-element from the adjunct clause to the matrix-object position violates merge-over-move. Hence.3. but because the strong feature of the wh-phrase remained unchecked. C&J’s overall approach to syntax (“simpler syntax”) is against using abstract expressions like PRO (though. vP]. they have no qualms about abstract 19 The subject-control reading of (24c) is obtained from a derivation in which the wh-phrase gets merged in the matrix-object position and ‘o Jo˜ao’ moves from the adjunct clause to the matrix [Spec.2). if merge-over-move is circumvented in the presence of a strong feature. To put this more starkly: the fact that one gets all the properties of obligatory control in adjunct-control configurations in the absence of the relevant properties that C&J identify as the main causal agent suggests that they have nabbed the wrong suspect. Most conspicuously absent is any account of the distribution of PRO. the derivation converges. In addition to the problems this sort of approach suffers from (see section 7. such movement must take place. the MTC not only provides a unified movement account of both complement and adjunct control. This requirement of the strong feature now overrules merge-overmove.2 The problem of the distribution of PRO C&J characterize the empirical problem of control as essentially that of controller selection. When it is checked by moving to [Spec.7. as in (25b). CP]. By contrast. which in turn requires that the wh-phrase must move if possible. yielding an object-control reading. as illustrated in (i) (see Nunes 2008c). (i) [Quem√F ]i e´ que o Jo˜aok [tk [cumprimentou ti ] [depois de tk entrar Who is that the Jo˜ao greeted after of enter.3 “Simpler syntax” 229 The wh-element of both derivations in (25) entered the numeration specified with a strong feature uF. this strong feature must be checked. for things are not equal anymore. these instructions are represented as in (26). . Note first that there is no obvious semantic reason for why only syntactic subjects are controlled. Even if one takes control to be a relation to an actional property.3. The question of the distribution of PRO amounts to the question of why it is that this controllee within the ACT complement is always a syntactic subject. if one restricts oneself to conceptual-structure information alone. (26) X␣ INTEND [␣ ACT] The ␣s in (26) indicate that X. However. 419).e. Anything that can be a syntactic subject in a clause (either base generated or derived) is a potential controllee under the appropriate predicate. it suffers from the sorts of problems often noted to be endemic to such approaches. but this would be unsatisfactory for obvious reasons.2. But. for example. a lexical predicate’s control obligations are embedded in its conceptual structure as an instruction that one of its arguments antecede some argument of the embedded complement. 7. one can always stipulate that the second ␣ in (26) is the subject of the embedded ACT complement. Consider a concrete analysis of control by C&J. semantics).230 Some notes on semantic approaches to control entities in either semantics or phonology). that are controllees. Note. whether one adopts PRO or eschews it. it is syntactic subjects. as noted. it does not follow that the relevant open variable must be in subject position. The problem of the distribution of PRO amounts to explaining why the control relation (most particularly OC) is generally restricted to controllee subjects of non-finite clauses. According to them. controls the ␣ in the embedded action complement. as such. that is “an appropriate level for stating control relations” (p. As C&J restrict their attention largely to actional complements. we say syntactic subjects for we take it as given that there is no thematic restriction on controllees.3 Problems with C&J’s decompositional approach The final problem we would like to point out with respect to the approach to control defended in “simpler syntax” is that it is essentially based on a decompositional approach to natural-language predicates and. The problem C&J’s approach faces is clear: C&J stress that it is conceptual structure (i. not conceptual subjects. the logical subject of INTEND. one is still left with the task of explaining why controllees must be syntactic subjects and why they so typically reside in non-finite clauses.. and not overt syntax. namely. Thus. that it is syntactic subjects that are controlled. it is unclear how to track one of the most basic facts about control. Of course. C&J 20 Henceforth.’ It is arguable that it is part of the basic semantics of this predicate that one cannot refrain from someone else’s doings. For example. the conceptual structure of ‘persuade’ also involves it as it means ‘cause to come to intend.7. consider once again the discussion above of ‘refrain from. (28) John persuaded Maryi PROi to leave early Other primitives include OBLIGATE.20 (27) Johni intended PROi to leave early However. C&J take ‘intend’ to be a primitive controller with the conceptual structure in (26).’ which are not simply ungrammatical. For example. which requires as part of its inherent meaning that the persuadee INTEND to do what s/he is persuaded to do. INTEND can also be part of more complex conceptual structures. The second step is to propose that all cases of obligatory control stem from including one of these primitive-control predicates in their meaning as represented in their conceptual structure. incoherent. . the effects of the meaning of the predicate are exercised across disparate syntactic forms and enforce the same condition: one can only refrain from acts that one commits oneself. As an illustration of what this means. and CS. three conceptual-structure predicates.3 “Simpler syntax” 231 namely. the binding that one sees in (27) results from the fact that ‘intend’ (the word) is associated with the conceptual structure of INTEND in (26).’ Thus. but. natural-language predicates (words) will be in single quotes and conceptualstructure predicates will be capitalized. First. 21 Actually. 447–448). they are either clearly wrong or (where they are not vacuous) largely stipulative. Evidence that this is a basic feature of the semantics of ‘refrain from’ (and not merely a feature of the control configuration) comes from noting the semantic acceptability of ‘John refrained from his leaving early’ when ‘John’ is understood as anteceding ‘his. The basic idea behind C&J’s analysis of control consists of two steps. more likely it involves CAUSE to COME to INTEND. Thus. This is manifested in the status of sentences like ‘John refrained from Mary’s leaving early. SHOULDroot . one might contend.’21 Thus the reason that one finds control in (28) is in virtue of the conceptual structure of ‘persuade’ including INTEND. a complex stand-in for verbs of the force-dynamic class of predicates (pp. The binding between X and the ACT is taken to be “inherent” within (26). they take control to be a semantically primitive relation inherent in the meaning of certain basic predicates. Why the resort to upper case? One reason is that these denote primitives in the conceptual system. which covers those cases where the predicate takes an ACT complement. This raises the following puzzle. Here is the source of our skepticism. this being a semantic class that cuts across the finite/non-finite category (cf. For example.’ which we took to be a plausible model for what might be meant by “controlling in virtue of its meaning. OBLIGATE. as far as one can see. not natural-language predicates. . C&J propose that these cases all involve primitive predicates like those noted above which.232 Some notes on semantic approaches to control restrict their study to the class of voluntary actions.’ nor is CS a predicate of English at all. ‘agree. for example.” For example. ∗ ‘John agreed with Bill to help each other’ is not. and (ii) showing how complex predicates include these and manifest control in virtue of including these in their conceptual structures. etc. 429–431). see e. First. as part of their inherent semantics. OBLIGATE. Fodor 1975): how are we to understand predicates like INTEND. and so on? If. their discussion on pp. ‘intend’ imposes no requirement on the subject of its complement. we take INTEND to mean ‘intend’ and we take control to mean what it conventionally means (namely. consider the primitive-semanticcontrol predicates. it should be clear that the success of this approach to control relies on two details: (i) finding a class of primitive predicates for which it is plausible that their control properties are inherent in their meaning. as C&J note in their discussion of coercion (p.’ If ␣ agrees with ␤ then ␣ and ␤ agree. for example. Nonetheless.22 From the above. There are verbs that one could imagine would semantically fit with more than one controller. providing the following kinds of examples: 22 It is unclear why conceptual structures require unique control. unlike INTEND. uniquely specifying the value of the syntactic-subject argument of the ACT complement) then it is hard to see that it requires control as part of its meaning.g. The reason is that ‘intend’ behaves quite unlike ‘refrain from. which we dub Fodor’s problem (as Jerry Fodor is someone who has worried about this issue consistently since the heyday of generative semantics. select a unique antecedent from the arguments of the embedding predicate to control the “actor” for the ACTion that is complement to that predicate. We believe that C&J succeed in neither task and as a result fail to successfully motivate this approach to control. As such. The astute reader will have noticed the flurry of capital letters in defining the primitives: INTEND. 452). Other options come easily to mind. a stipulation. Why conceptual structures specify exactly one controller is. though ‘John and Bill agreed to help each other’ is perfectly acceptable. INTEND is not the same as ‘intend. However. to allow INTEND to mean ‘intend’ and to show that ‘intend’ always involves control. (31) Hilary intended for Bill to come to the party though. it is not clear to us that (31) below is a contradiction. In other words.3 “Simpler syntax” (29) a.7. they do assume that INTEND means ‘intend’ and that all uses of ‘intend’ actually involve control. then? Here is one place we do not wish to go: it means just what ‘intend’ means but it requires that the embedded subject be controlled as with ‘refrain from. So. Thus.’ (30) Hilaryi intends PROi to bring it about that Bill come to the party So. INTEND cannot mean ‘intend. though not necessarily in the sentence at hand but in some more abstract way. now we are faced with the second problem for decompositional approaches like this one: paraphrases are not meanings. C&J are aware of this problem and they seem to take a different tack.’ The reason is that we have no reason to think that such a predicate exists and even if it did. but what they seem to say is the following (the relevant discussion is in C&J 2005: 452). C&J expand the concept of control to include cases of control of an implicit subject of an implicit ACT complement as well as explicit ones. 233 Hilary intends for Ben to come to the party Hilary intends that Ben come to the party Note this is precisely what ‘refrain from’ disallowed (∗ ‘Hilary refrained from Bob coming to the party’) and this suggests that it does not follow from the very meaning of ‘intend’ that the subject of the embedded ACT complement must be controlled by the matrix subject. for example. and (30) involves control of the PRO of the interpolated predicate ‘bring it about. What these sentences mean is paraphrased “approximately” as (30). Thus. They analyze ‘plan’ as also involving INTEND.’ But what does it mean. C&J are not very precise here so we may have misunderstood them. From what we can gather. which it should be if (29) meant (30). Though we agree that the sentences in (29) can be paraphrased as (30) for many circumstances. it is not clear in what sense the primitivecontrol property follows from the meaning of INTEND except by stipulation. b. she intended to do nothing whatsoever to bring this about We can make a stronger case to the same effect with another of C&J’s examples. one problem with capitalization is that it threatens to bleach all explanatory value from the exercise as it stipulates what it claims to explain. In cases like (29) control exists despite all appearances to the contrary. (32) should be interpreted as (33): . being lazy and complacent. this does not imply that (29) means what (30) does. Moreover. Note that under C&J’s approach. but in the end the problem illustrated here is. Rather. Recall what the aim of the game is. her plans can be based on the assumption that Ben will come without her doing or having to do anything about it. non-finite clauses. Until we know what INTEND means exactly. or the meanings are left unspecified (albeit with a nod and a wink towards the lower-case analogue).’ One could always fish around for another primitive controller to ameliorate matters. if it does not. Fodor has repeatedly pointed out that the capitalized predicates exploited in semanticdecomposition proposals rarely mean what their lower-case brothers do. there are no PROs. In other words. a quite general instance of Fodor’s problem. We can assume that ‘plan’ does not involve INTEND as a subpart. C&J in effect propose that there are cases of semantic control even in the absence of syntactic control. we are left with no explanation of control at all. (34) Hilary is planning that Ben will come without planning to bring it about that Ben will come There is a way of repairing this problem. However. as all control ultimately rests on the meaning of these primitive predicates. delimited. we believe. . There is yet another problem with C&J’s strategy. The project is to show how what grammarians think of as control actually piggybacks on a more basic notion of semantic control that arises from a small class of primitive predicates that enforce control as an inherent part of their meanings. or apparent binding. it is an unpromising strategy to aim to explain how their putative semantic powers arise from their meanings. we cannot have any confidence that its meaning underlies its control powers. it leaves unexplained why ‘plan’ heads a control structure when coupled with a non-finite complement as in ‘Johni planned PROi to leave. In short. As such. either what is delivered is not an explanation but a stipulation. what C&J claim is that there exists a type of implicit control in sentences like (29) and (32) that appears not to be an instance of grammatical control at all. But this leads to a prediction: licit control interpretations are semantically. here it is quite clear that Hilary could be planning what we have in (32) without planning what we have in (33). (34) is clearly not contradictory. it is never quite clear what they do mean. not syntactically. This general problem carries over to C&J’s control account. as what the primitivecontrol predicates enforce is not syntactic control but semantic control. However. More often than not. This seems clearly incorrect.234 Some notes on semantic approaches to control (32) Hilary is planning that Ben will come to the party (33) Hilary is planning to bring it about that Ben will come to the party However. But if this is so. Perhaps because the complement when finite need not be an ACTion. it is. i. then it cannot be that ‘persuade’ (somewhat) means ‘cause to come to intend’ but. at most. It should be fine with the meaning paraphrased in (36). why can (37a) with indexation in (37b) not mean what (35) means? (35) Ben’s mother said that Hilary is planning that Ben will come to the party (36) Ben’s mother said that Hilary is planning to bring it about that Ben come to the party (37) a. Thus. that ‘persuade’ with an infinitival complement means this. This is what is done in (37b)..’ But what is it about this combination that results in this meaning? C&J do not say. But so construing control should allow a syntactic PRO to be bound by an antecedent other than its grammatical controller so long as semantic control can obtain. C&J suggest that the object-control properties of ‘persuade’ come from its meaning “cause to come to intend” (p. one can cause someone to come to intend to leave by threatening or humiliating them rather than persuading them). 446). this move effectively divorces syntactic and semantic control as control configurations obtain even in the absence of a syntactically bound PRO. a part of its meaning. the “fix” C&J propose to allow INTEND to be a basic controller in virtue of its meaning actually ends up leaving unexplained standard cases of syntactic control.g.7. But anyone can see that this is not what ‘persuade’ means (e.’ As noted.e. ‘Hilary’ controls the subject of ‘to bring it about’ and so the control imposed by the “meaning” of INTEND has been satisfied. (37a) does not have this meaning. ∗ Ben’s mother said that Hilary is planning PRO to come to the party Ben’s1 mother said that Hilary is planning PRO1 to come to the party What gets us (37b) is the mechanism C&J exploit to “regularize” (29) and (32) in service of getting INTEND to mean ‘intend. However. if (36) is a fine reading of (35).3 “Simpler syntax” 235 (35) below is a control configuration with the rough interpretation in (36). such as it is. Thus. Thus (38) is not well paraphrased as (39) despite ‘persuade’ being the matrix predicate. But note here it is an ACTion (‘what Frank did was . However. the meaning.. Another illustration of Fodor’s problem as pertains to control is appropriate. is a product of the composition of the non-finite complement and ‘persuade. b. at most. (38) John persuaded Mary that Frank had come (39) John caused Mary to come to intend that Frank had come If so. But even this is only clear for those cases where the complement is infinitival. The problem is that compositional theories never really manage to do this. we have not been offered an account of how the meaning of ‘persuade’ results in control. given the meaning of ‘persuade’ noted above? Why does it become semantic control with a non-finite complement and why grammatical control when the non-finite complement has a PRO subject? C&J do not say. Rather. Why not. As such. Moreover. then we end up with the problem analogous to that in (37a): why can (42) below not mean what (43) means? One possible answer is that control semantics only arises when the infinitival complement has a PRO subject. Though (40) is degraded. the point is that C&J owe us a theory of how to compose complex conceptual structures from primitive ones so that we see why control occurs with non-finite complements but not with finite ones. However. (42) John1 persuaded Mary PRO1 to come (43) John1 persuaded Mary to bring it about that he1 come There are several moves one might make to finesse this line of reasoning. We could go on and on. it is perfectly meaningful. In place of capitalization. the point is not whether one could do this in this particular case. even if the complement is infinitival unless it has a PRO subject. However. C&J need to provide an account of the meanings of the primitive predicates and principles of composition for forming complex conceptual structures from basic ones.e.236 Some notes on semantic approaches to control come/Frank came voluntarily’). This would get the result but it would clearly not be very satisfactory for we would want to know what it is that PRO contributes and this is precisely what the proposed theory of control is supposed to explain.. we hope that the main objection with C&J’s semantic project is clear. We understand it to mean something roughly like (41). unless C&J say why control with ‘persuade’ relies on these conditions (i. Theirs is an essentially decompositional approach to control. When ‘persuade’ has a finite clausal complement there is no control. Either the meanings of the conceptual structures . it suffers the well-known vicissitudes of all such theories. How does the fact that it need not have been an action (although it is one) affect the compositional properties of the predicate? C&J again do not say. unless some rules of composition are offered). However. we do not see syntactic control. as illustrated in (40) below. (40) ??John persuaded Mary for Bill to come (41) John persuaded Mary to bring it about that Bill come Note that if we assume that (40) is actually a case of semantic control (like in [29] and [32]). a better understanding of the lexical semantics of embedding predicates in complement-control structures will certainly supplement syntactic accounts in explaining why a given binding relation must obtain regardless of whether the embedded subject is overt or not. for instance). In effect. without sacrificing empirical coverage. we are provided with a list.and overgenerate. . [29]. In the second. As shown in the sections above. and unified account of all the problems faced by semantic approaches to control. We hope that the reader appreciates how much more the MTC offers. That semantic approaches to obligatory control fail to achieve an adequate level of theoretical explanation becomes even more transparent when a detailed comparison with the MTC is made. On the other hand.3. Once they are essentially based on the semantic relations between control predicates and their complements. conceptually motivated.4 Conclusion 237 are identified with that of the words that embody them (IDENTIFY/‘identify’) or their meaning is left either unspecified or stipulated. the overall conclusion of the discussion in the previous sections is that. decompositional accounts generally resolve into elaborate discussions of individual cases.’ for instance. the MTC is able to provide a non-stipulative.7. As such. they undergenerate in being unable to account for adjunct control and for the syntactic conditioning one may find in the specification of subject or object control in adjunct configurations (see section 7. However.4 Conclusion This may be a good moment to note that accepting the MTC does not entail rejecting semantic contributions to determining control.2. 7. In the first instance it is easy to argue that the proposed meanings cannot be correct. left to their own devices. no real account is offered at all. semantic approaches to obligatory control do not constitute a viable alternative to syntactic approaches as they are bound to both under. As noted above with respect to ‘refrain from. in place of a general account. [32].1). they overgenerate in incorrectly predicting obligatory control where the subject of the embedded complement to a control predicate is overt (see the discussion of [19]. First. a prejudice embodied in the inclusiveness condition. In this concluding chapter. despite the latter’s superficial notational similarity to the GB view. we will remind readers of the fact that (obligatory) control relations exemplify canonical properties of movement. more accurately. much to our surprise. and in one sense much more consistent with classical generative analyses than are PRO-based accounts. our main interest in the MTC has much more to do with minimalism than with control. we return once again to how the minimalist program and the MTC conceptually intertwine. The take-home message is that if you like the 238 . This chapter is a defense of this claim. Truth be told. the latter presupposing the truth of some of the central tenets of the former − in particular.8 The movement theory of control and the minimalist program 8. especially as regards its relation to the broader concerns of the minimalist program.1 Introduction The movement theory of control (MTC) has been. which we interpret to forbid (among other things) coding of formal properties as lexical properties. we could be persuaded to go so far as to claim that a minimalistically respectable account of control will necessarily have some version of the MTC at its core. Or. we believe that control is currently interesting because of its near perfect fit with certain central tenets of the minimalist program as realized in the MTC. Second. In the process of making these claims we note that the MTC is actually a conservative extension in a minimalist setting of the classical approaches to control. a controversial proposal. The main reason for this is that PRO-centered accounts of control must run afoul of the anti-construction bias characteristic of the minimalist program. we would like to return to the MTC’s conceptual foundations. Indeed. This book has aimed to outline (and trumpet) its virtues and to consider (and parry) purported vices. The defense will proceed through various layers of abstraction. minimalistically construed. Third. the elimination of D-structure. we argue that the MTC alone is consistent with the explanatory ambitions of the minimalist program. subject control with ‘promise’ and control shift. Note that this same account explains why it is that OC PRO must be a syntactic subject (see section 7. . for instance. the three major ones being locality. apparent exceptions to this generalization.1 1 Still compatible with minimality are potential cases with a non-subject OC PRO where PRO sits in a caseless position. (1) John1 persuaded Mary2 PRO∗1/2 to leave Moreover. For example. proposes that “reflexive” predicates like ‘wash. and explanatory aspirations of the minimalist program. With respect to locality. e. ‘John’ is not a possible antecedent of PRO.3.1). .2.g. Consider the structure in (2). which represents A-movement from a non-subject position: (2) [ . is the source of the minimaldistance principle’s restrictions on possible antecedents of PRO. architecture.’ and ‘dress’ in English optionally assign case and may therefore license an OC PRO/A-trace in the object position. We have argued that the dependency exhibited in obligatory control (OC) acts as if governed by the same restrictions. Thus. in cases like (1).2). Let us consider them in turn.2 Movement within minimalism and the MTC 239 minimalist program. . . starting with minimality. an OC chain will always bottom out in a syntactic subject. [DP2 V DP1 . OC is a dependency generated by movement. as they fit poorly with the principles. . Hornstein (2001). following a suggestion by Alan Munn.’ ‘shave. we have argued.5).8. Let us illustrate. as represented in (i). ]] The dependency illustrated in (2) between the two instances of DP1 violates minimality as they span an intervening DP2 . DP1 . and copying. 8.2 Movement within minimalism and the movement theory of control Within the minimalist program there are several signature properties of movement. there are two cases: minimality effects and freezing effects. given standard minimalist marks of movement. arise precisely because the apparent intervening DPs do not actually c-command the PRO position and so do not function as interveners (see section 5. This. . Intervening DPs in A-positions block the establishment of further A-dependencies by prohibiting A-movement across them. as moving across ‘Mary’ is prohibited by minimality (see section 3. . for all other dependencies will necessarily violate minimality. economy. In other words. then you should love the MTC and eschew PRO-based accounts of control.. Given that OC requires an antecedent.4. and not construal/Agree operations. because ‘Bill’ intervenes. the MTC is unique among current approaches to control in providing a principled explanation of the contrast in (3) (see section 2. If control involves movement. Freezing explains why control. finite indicative clauses can be ␾defective and. . where ‘Mary’ cannot be the antecedent of PRO.1. like raising.’ can control PRO due to merge-over-move (see section 4. e. and indeed they are. exceptions to freezing are expected to be possible control configurations.1).1). as illustrated in (3) below.. If merge-over-move is at the root of this restriction. In (4) below. Given that in English finite clauses are ␾-complete domains. Second. infinitives and gerunds. freezing prohibits A-movement from the subject position of the embedded clause of (3b). Let us now examine another characteristic feature of movement within MP. economy in the sense of Chomsky 1995. for example. is possible into non-finite clauses but is generally prohibited into finite ones.3.5). First.4). as merge-over-move regulates movement.g. All other analyses are reduced to stipulating the distribution facts of OC PRO − that it appears exclusively in the subject position of ␾-defective clauses. For example. This once again shows that it is the possibilities for movement to take place that lead to subject or object control (see Nunes 2008c).240 The movement theory of control and the minimalist program Let us now consider the effects of freezing. ‘John. (3) a. then there should be copies. it is instrumental in explaining the restriction to subject antecedents in adjunct-control configurations. This expectation is clearly borne out The effects of minimality in these constructions can be observed in (ii).2 (4) John1 saw Sue2 [before PRO1/∗2 leaving the party] A fourth diagnostic property of movement within the minimalist program is that it produces copies (hence the copy theory of movement). (i) (ii) Bill1 washed/shaved/dressed PRO1 Mary1 wants Bill2 to wash/shave/dress PRO2/∗1 2 Recall that Portuguese allows subject or object control in adjunct configurations depending on whether or not the matrix wh-object is in situ or has undergone movement (see section 7. in Brazilian Portuguese.2. they do support OC dependencies (see section 4. accordingly. b. ∗ John hoped [PRO to win] John hoped [PRO won] Two points are worth emphasizing here. It too governs OC dependencies. In the guise of merge-over-move.’ but not ‘Sue.5. which in the case of A-relations prohibits A-movement from a “␾-complete” domain. then OC dependencies must be generated by movement. In other words. . Iordachioaia.. in our view. and Marchis (2008). Anagnostopoulou. The MTC is clearly incompatible with (i) and thus its theoretical viability requires 3 To date we know of no PRO-centric analysis of backward control. the MTC is incompatible with D-structure. the copy theory of movement. at least from a minimalist perspective. may be taken to follow from the “simplest” definition of merge (see Chomsky 2004). Together.4 control must be a movement relation on minimalist grounds given that it exhibits what the minimalist program takes to be the core properties of movement. recall. For recent further novel examples of backward control. it is a duck). 8. Of the group. then PRO-based analyses are incorrect. it follows that OC is a dependency generated by movement.4 and 4. these two properties (i) prohibit movement into ␪-positions and (ii) require that all argument DPs begin their derivational lives in ␪-positions.g. D-structure.5. which replaces the less methodologically valued trace theory.8. This suggests that there is a consensus: should backward control exist. Fujii (2006) and Alexiadou. as we have argued.3 The movement theory of control and the minimalist architecture of UG The MTC rests on the assumption that movement into ␪-positions is grammatically viable. minimality and freezing have been central features of generative grammar since the late 1980s and are excellent candidates of “least effort” principles. walks. Importantly. the above principles and their concomitant technology are central features of minimalist grammars.5. and flies like a duck. a particularly clear illustration of the tight conceptual fit between central tenets of the MTC and the minimalist program. it seems to us that only the MTC can account for backward control for such control configurations in PRO-based theories necessarily violate principle C and so should be impossible. though also the most deeply embedded in current phase-based approaches.3).g. 4 For example. Indeed. movement). Note that these four features of OC configurations reflect the operation of very general principles of movement in the minimalist program.3 The MTC and the minimalist architecture of UG 241 in copy-control languages and languages that allow backward control (see sections 4. merge-over-move is the most conceptually suspect. Following the duck principle (if it quacks.3 That the MTC combines so neatly with the copy theory of movement to provide a straightforward account of these phenomena is. To the extent that they reflect more fundamental minimalist conceptions (and most do). is the syntactic level where all and only ␪-relations are coded. which are the hallmark of the minimalist program. It is also the input to all transformation processes (e.. In turn. from the fact that OC dependencies look as if they respect these principles. see e. let us take a brief digression. This sentence is derived if ‘John’ is merged into the embedded [Spec. The MTC requires that movement into ␪-positions be possible. The purported grammatical violation (the lack of a ␪-role in ‘John’) will only become evident downstream. However. Chomsky (1995) notes that (5b) respects merge-over-move. there exists a tight conceptual connection between the minimalist program and the MTC. Belletti 1988 and Lasnik 1995). and not at the point of the derivation where ‘John’ is introduced. vP] to get the external ␪-role and up to the matrix [Spec. it is not clear that the grammar will locally block the derivation. However. Though incompatible with (i). This is correct. TP] to get case checked. for the only empirical argument against movement into ␪positions that we know of. As disposing of Dstructure (a methodologically unwelcome grammar-internal level) is a central architectural feature of the minimalist program. if this prohibition is not primitive but actually a subpart of the more general prohibition against movement into ␪-positions. Moreover. Before considering this in more detail.’ ‘John’ then moves up to the matrix [Spec. a central architectural feature of the minimalist program − the elimination of D-structure − is both necessary and sufficient for the MTC. as it were. Thus. b.. this approach is not in consonance with his general attempt to minimize “look-ahead. If we assume that ‘someone’ in (5b) can check (partitive) case with ‘be’ (see e. So the (primitive) prohibition against first merging into a non-thematic position is not only sufficient to rule . for example. TP] instead of ‘John’ being merged into that position).” He observes that prohibiting first merge of an argument into a non-thematic position will allow the illicit nature of (5b) to become immediately evident. (5) a. the MTC is not incompatible with (ii). We mention this. TP] in preference to raising ‘someone. this derivation should block the derivation of (6). as illustrated in (5b). (6) ∗ John expects to be someone kissing Sam [TP John [vP John expects [TP John to be [vP someone kissing Sam]]]] John expects someone to be kissing Sam Chomsky (1995) uses the ban on movement into ␪-positions to rule out the derivation in (5b). then the derivation should converge. the absence of Dstructure is sufficient to license the MTC. whereas the derivation underlying (6) would not (‘someone’ is raised to [Spec. but it is agnostic as to whether argument DPs must begin their derivational lives in ␪-positions.g. which was offered in Chomsky 1995. can rely on (ii) only. Consider the sentence (5a). Moreover.242 The movement theory of control and the minimalist program the elimination of D-structure as a grammatical level. How tight? The MTC clearly implies the absence of D-structure. the MTC is quintessentially minimalist. that a necessary condition for the theoretical viability of the MTC (that D-structure not exist) is one of the central tenets of the minimalist program. However. so long as subsequent movement into ␪-positions is countenanced. There is thus a very close conceptual connection between the minimalist program and the MTC. to the extent that the elimination of D-structure is a central feature of the minimalist program. it is a sufficient one as well: (7) MTC ↔ no D-structure In other words. Furthermore. the ungrammaticality of (5b) does not tell against the MTC because the MTC is consistent with the requirement that argument DPs enter derivations through a thematic door. not only is eliminating D-structure a necessary condition for the MTC. then the MTC is a grammatical option. Drury. This is good news for the MTC. for current purposes. why has this not been observed previously? The main reason is that eliminating D-structure does not imply . if so. their theoretical apparatus (though not the technology used to express control dependencies) is largely independent of it. However. and Grohmann 1999 and Epstein and Seely 2006). So interpreted. it is not clear why the proposed restriction against merging into non-thematic positions should exist if D-structure does not (not a small matter in the context of minimalist theorizing). the reader should not conclude that the proposed prohibition against first merge into non-thematic positions is correct. Specifically: once D-structure is eliminated as a grammatical level. nothing prohibits movement into ␪-positions. though important. If this is correct. are incidental to the MTC proper and − what is important here − that the data in (5) and (6) are not problems for it.3 The MTC and the minimalist architecture of UG 243 out the derivation in (5b). the reader may be asking. Not only does the MTC imply the absence of D-structure. The conceptual connections between the minimalist program and the MTC are stronger still. 2004) or by assuming that non-finite clauses do not have TP specifiers (see Castillo. This contrasts with most other theories of control currently being considered. though they might be compatible with the minimalist program. This said. They do not rely on any distinctive minimalist assumptions and thus. Thus.8. it is worth noting that these issues. There are accounts that derive the facts in (5) and (6) by assuming that ‘be’ is not a case checker and that ‘someone’ cannot check accusative case against the matrix light verb due to the intervention of the lowest copy of ‘John’ (Nunes 1995. To recap: one of the central architectural features of the minimalist program has been the elimination of D-structure. but also more congenial to local computations. but the absence of D-structure is sufficient for the MTC given standard ancillary assumptions. to wit. D-structure within GB. or restricting a DP’s ␪-role assignment to its first merge. Chomsky (1993) actually retains the thematic restrictions coded at D-structure. More recent avatars of the minimalist program take merge and move to be different instances of the very same operation. but in another form. . meaning that all Dstructure operations precede all transformational operations. If D-structure has certain properties. the latter to satisfy all the other grammatical dependencies) should be given up as well. for example. on both methodological and theory-internal grounds. the MTC requires that movement into ␪-positions be an option. Here is some Whig history: Chomsky’s (1993) argument against D-structure was actually quite narrowly focused.244 The movement theory of control and the minimalist program removing D-structure conditions from the grammar. however. It proposes banning movement into ␪-positions. correspondingly we believe. but the segregation of functions between lexical insertion and movement (the first being designated to satisfy ␪-relations. it commits no hostages to whether argument DPs must first merge into ␪-positions. and it represents “pure GF-␪. The 1993 vintage of the minimalist program distinguishes two different operations: move and merge.5 How reasonable is this? On methodological grounds. In sum. the prior differentiation between move and merge is conceptually difficult to retain and. The former contrasts with the latter in being greedy and being driven by feature-checking requirements. 5 More accurately: as discussed above. in Chomsky (1993). The MTC requires that it be complete: not only must Satisfy be rejected. the wholesale elimination of D-structure and its restrictions is clearly the preferred option. Moreover. then either both should be subject to feature-checking requirements or neither should be. is a level with many distinctive properties: it is input to the transformational component. Whichever tack one takes.” Chomsky (1993) argues that the first noted feature of D-structure (Satisfy in the parlance of Chomsky [1993]) must be dispensed with and grammars must adopt generalized transformations that allow derivations to interleave operations akin to lexical insertion with operations akin to movement. we believe that current theoretical assumptions internal to the minimalist program lead to the same conclusion. If so. the prohibition against movement into ␪-positions becomes theoretically awkward to enforce. we believe that there is every reason to retain the methodologically superior option (the complete elimination of D-structure and its properties) that underwrites the MTC. Thus. The upshot then is that the MTC requires a more radical elimination of D-structure than considered in Chomsky (1993). the elimination of D-structure is only partial. then eliminating D-structure entails removing the restrictions coded in D-structure from UG. we think it a virtue of the MTC that it follows quite seamlessly from the elimination of D-structure. We have argued extensively that the facts do not require this. This makes the GB conception of PRO above a non-starter. namely [DP e]. However. why not just build the structure required by the head that is inserted − in other words. bare phrase structure.4 Inclusiveness. consists of two kinds of operations: phrase-structure rules and lexical-insertion operations. X’-conventions regulate phrase-structure rules: XP→ .6 Of course. . regardless of how this plays out empirically. this derivational option no longer exists. In this context. D-structure. PRO is a perfectly sensible grammatical element. So. 1995). In a minimalist approach that adopts bare phrase structure (see Chomsky 1994. X . In other words. It is what arises via the application of a DP phrase-structure rule sans lexical insertion. The facts may force us to back away from the minimalist optimum. . The second option is the one explored 6 See for example Chomsky’s (2004) discussions on treating move as just internal merge. Removing D-structure and its attendant properties from UG is both necessary and sufficient for the MTC to be grammatically viable. one cannot have phrases that are lexically headless. how is PRO to be described? There are two options: as a primitive lexical item or as a grammar-internal formative. . the original motivation behind X’-theory was to remove the redundancy in systems that generated phrase structure and lexical insertion that pruned it. no phrase structure. recall. 7 One can adopt the endocentricity assumption behind X’-theory without assuming that phrases are projections of lexical heads. but not doing it is methodologically suspect as it comes for free unless specifically blocked. especially in the context of the minimalist program. It is consistent with this that there be lexical-insertion rules that replace X with lexical items of category X. However. the only way of preventing it is to encumber UG with principles that are not otherwise methodologically or theoretically required. On such an interpretation. The minimalist ethos frowns on this.4 Inclusiveness. one need not do this. bare phrase structure. . build to specifications contained in the head? This assumes not merely endocentricity but the stronger notion that phrases are projections of heads. Bare phrase structure strongly embodies the X’-theoretic conception that phrases are projections of lexical elements. and the MTC 245 In sum: the MTC is closely tied to a central feature of the Minimalist Program – the elimination of D-structure. 8. As he observes.8. once D-structure is removed. If this is so. this does not mean that the MTC is correct. and the movement theory of control The elimination of D-structure has a second important consequence for the theory of control.7 No lexeme. On this latter view. . The pruning seemed to imply that lexical items coded structural consequences. 246 The movement theory of control and the minimalist program by the MTC. Within GB. traces and PROs are formally identical at LF. the formal similarity embedded within GB carries over to the minimalist program if PROs are copies as well.’ etc. the only formal difference being the source of their indices. PROs are what we call A-traces that have wandered into ␪-positions. Generative grammar has generally analyzed control properties as grammatical by-products. let us quickly see how the MTC is actually a conservative extension in a minimalist setting of the classical GB conception. In this respect. PROs are cousins to traces. Of course. 9 One argument against traces is that unlike copies they violate the inclusiveness condition. just like any other lexical item or phrase. movement supplying the requisite index for traces and construal doing the same for PROs. as noted. On the assumption that derivations cannot “add” new elements to the derivation.’ ‘dog. as control relations − like A-trace dependencies − are grammatical products formed by move/(re)merge. So.’ ‘bring. a grammatical formative − an expression whose existence and properties derive from the organization of the grammar. the properties of control structures are expected to derive from general principles of grammar. then classical GB PROs are suspect elements as well. if PROs and traces are identical. However. it is worth considering for a moment how radical a departure this is from the classical conceptions of control. There are no problems with bare phrase structure on this conception because PRO functions like any other (nominal) expression drawn from the lexicon.8 The MTC models this similarity by identifying PROs and A-traces. Nevertheless. being purely theory-internal formatives. What is critical to note here is that copies are perfectly welldefined elements within minimalism and occurrences/copies of an expression are licit entities consistent with bare phrase structure. PRO is like ‘the. the main difference being their grammatical provenance. Before considering the first option. which like PROs are bereft of case. PRO is a phonetic gap that results from deletion under equi. as in GB.9 Moreover. the MTC embodies the assumption that the properties of control configurations derive from (and so directly reflect) the underlying operations and principles of UG. On this conception. Taking “PRO” to be the product of this operation aims to explain its 8 PROs are identical to A-traces in particular. In the standard theory. This is the crux of the MTC. both elements are structurally similar (both have the shape [DP e]).’ ‘this. Let us now consider the option of treating PRO as a lexical item and not a grammatical formative. traces are methodologically suspect elements. As copies replace traces in the minimalist program. In GB. PRO is. traces being by-products of movement. It lives in the lexicon and it can merge and move. . in contrast to A’-traces. the analysis of “PRO” reflects the view that control facts directly follow from basic operations and organizing principles of grammars (see sections 2. in effect. bare phrase structure. the “control construction” directly reflects the idiosyncratic properties of a distinctive lexical item. in a generative context.4). What does this mean? It is the claim that the fundamental principles of grammar operate independently of the lexical items that they manipulate. their structures. Indeed. The inclusiveness condition can be interpreted as a prohibition against confusing lexical and structural information.4 Inclusiveness. and their basic operations and principles? They cannot be. but because they all involve A’-movement. In effect. grammatical operations and restrictions have the properties they do not because of the functional features of the “constructions” in which they apply. Topicalization. and the MTC 247 semantic and phonetic properties. The analysis of PRO as [DP e] is meant to account for its distribution and its semantic and phonological interpretation. generative grammarians have assumed that constructions do not exist. In other words. focus. PRO is [DP e]. relative clauses are not islands because they involve particular lexical heads or contain particular lexical items. where it is coded in the inclusiveness condition. and relativization do not obey islands because they involve topic.or ␾-)defective domains. assuming that PRO is a lexical item (rather than a grammatical formative) is. to treat its special licensing requirements as lexical quirks and this seems to us quite wrongheaded. But this is what reducing the grammatical properties of control structures to the lexical requirements of a lexical item PRO does. In turn. one can go further. rather than the basic operations and organization of the grammar. So treating PRO is to endorse a form of “constructionism.” Here is what we mean.8. The problem with treating PRO as a lexical item is that it amounts to analyzing control configurations as constructions: control properties follow from the unique properties of the lexical item PRO.3 and 2. or relative heads. in GB. It is in this sense that constructions do not exist. focus. The GB antipathy to construction accounts carries over to the minimalist program. In both cases. syntactic antecedent and can only be licensed within (tense. For example. PRO needs a local. PRO’s requirements are grammatical licensing requirements. In contrast. c-commanding. Postulating PRO makes no sense except in a grammatical . Indeed. Since the early 1980s. but because they instantiate particular structures. which defines the construction. many (if not all) the properties of this “lexical” item cannot even be identified independently of the grammar. they are not the fundamental units of syntactic analysis. How are these requirements to be stated in purely “lexical” terms? How can they be expressed except by adverting to grammars. but because of the formal properties that these constructions instantiate. It discourages coding structural information onto lexical formatives. then the evidentiary bar relevant to rejecting the MTC should be quite high. the kind of explanation we should provide incorporates the MTC. explanations no). then. The reason is that only the MTC evades constructionism and tries to derive the properties of control structures from general principles of grammar rather than from the special licensing conditions of a peculiar lexical item. However. if the fit between the minimalist program and the MTC is as tight as we have suggested. We take this to be a positive feature of the analysis. this implies that only the MTC is compatible with the minimalist program. if we are right. the MTC is correct. If this is correct. Even describing what they are requires reference to principles and operations of the grammar. There is thus a very tight conceptual fit between the MTC and the minimalist program. as this is an inherent feature of minimalism that ineluctably follows the elimination of D-structure. Moreover. Consequently.5 Conclusion In the earlier chapters we have tried to elaborate a movement theory of control. If we are correct. The reader will have noticed that this reinforces our earlier conclusion reached above regarding the MTC and the elimination of D-structure. it does suggest that those with minimalist aspirations should smile on the MTC and that the burden of proof must be with those that reject it. We have tried to address the empirical difficulties attributed to the MTC as well as outline what we take to be its empirical strengths. because the MTC fits well with the minimalist program. given minimalist standards. But as the only non-PRO-centric theory of control is the MTC. treating PRO as a lexical element violates the spirit of the inclusiveness condition and so renders PRO a suspect element.248 The movement theory of control and the minimalist program context. This is the (very high) cost of violating the inclusiveness condition. Dispensing with D-structure implicates the MTC. In the end. one should not conclude that. postulating lexical elements like PRO to account for the attested properties of control cannot possibly yield explanations of these properties (descriptions yes. only the MTC has the capacity to move beyond description to explanation. among the alternatives on offer at present. Bare phrase structure implicates it as well. Once again. Its requirements are entirely grammar-internal. In this last chapter we have briefly recapped why we think that the MTC is a particularly interesting . for a lexical item like PRO codes as part of its content the very properties that are supposed to be explained. Furthermore. 8. then the upshot is that PRO-based accounts of control within minimalism are not compatible with the explanatory ideals of the minimalist program. 8. We have argued that. about the scope and prospects of the minimalist project. it is an almost perfect fit with the minimalist program.. but we do not believe that it will be wrong in a trivial way. Given how deeply interweaved it is with general minimalist precepts. showing that it is inadequate should tell us a lot about how control in particular and anaphoric dependencies in general are grammatically coded and. and theoretically. . backward control and copy control) and points to novel kinds of derivations (sideward movement) that have led to an appreciation of the subtle possibilities inherent in the modern minimalist approach to grammar. methodologically.5 Conclusion 249 theory in the context of the minimalist program. principles. and operations. The MTC might be wrong (though we doubt this). importantly.g. its considerable empirical coverage is a delightful bonus. the MTC is essentially the null hypothesis about control. Conceptually. in a minimalist context. The MTC not only covers virtually all of the classical facts in a principled manner – it even leads to the discovery of new kinds of data (e. Given these virtues. The processing role of structural constraints on the interpretation of pronouns and anaphors. The representation of case in Modern Icelandic. and Kathleen Straub. Modern Icelandic Syntax. Alboiu. 424–503. Natural Language and Linguistic Theory 16(3): 491–539.). D. Linguistic Inquiry 19: 1–34. Cornell University. 73–138. 2002. and Leslie Barrett. Dordrecht: Kluwer.). Successive cyclicity. Linguistic Inquiry 20: 219–252. Dordrecht: Springer. Alexiadou. In W. Gianina Iordachioaia. Bresnan (ed. Passive arguments raised. 1990. verb-movement and EPP-checking. In L. 1997. On government. Thematic Roles and Syntactic Structure. Baker. MIT. 187– 234. A stronger argument for backward control. San Diego. Kyle Johnson.). 1979. New Horizons in the Analysis of Control and Raising. Barrie. anti-locality.References Abels. In W. 263–279. 250 . MA: MIT Press. The null content of null case. Control and wh-infinitivals. Baker. New York University. Artemis. University of Chicago Press. 2007. Josef. Avery.). 1989. New Horizons in the Analysis of Control and Raising. Case structures and control in Modern Icelandic. 2007. and Elena Anagnostopoulou. Aoun. The Mental Representation of Grammatical Relations. Davies and S. Belletti. Dubinsky (eds. and adposition stranding. and Mihaela Marchis. Moving forward with Romanian backward control and raising. 2008. Mark. Unpublished manuscript. Parametrizing Agr: word order. Memory and Cognition 28: 748–769. The case of unaccusatives. University of Connecticut. 187–211. 1982. Paper presented at NELS. Incorporation: A Theory of Grammatical Function Changing. 2003. Elena Anagnostopoulou. Gabriela. A. 2002. Michael. Baltin. Dubinsky (eds. Badecker. Alexiadou. Andrews. Journal of Experimental Psychology: Learning. Klaus. and clitic placement. Cambridge. Davies and S. In J. 1988. Mark. Dordrecht: Springer. 1998. D. and Ian Roberts. Unpublished manuscript. Mark. Elements of Grammar.). In Joan Maling and Annie Zaenen (eds. Doctoral dissertation. 1988. CA: Academic Press. case marking. Artemis. William. Haegeman (ed. Overt copies in reflexive and control structures: a movement analysis. Zeljko. Copy-reflexive and copycontrol constructions: a movement analysis. Dubinsky (eds. The Syntax of Nonfinite Complementation: An Economy Approach. On the Content of Empty Categories. Cedric. D. and Norbert Hornstein. 1994. Cedric. and Norbert Hornstein. Dennis. Bobaljik. and Samuel Brown. Amsterdam: John Benjamins. Boeckx. 2003. Jonathan. Boeckx. Jonathan..umd. Davies and S. Linguistic Analysis 24: 247–286. and Amy Weinberg. Cedric. Dordrecht: Foris. Cedric.). 2007. Boeckx. 2007. Boeckx. D-structure. Amsterdam: John Benjamins. 2004. Cambridge. Conflicting c-command requirements. Icelandic control is not A-movement: the case from case. 13–74. www. Bouchard. ˇ Boˇskovi´c. Berwick. 2006a. Cedric. 2010. Studia Ling¨u´ıstica 53: 227–250. 1999. Bobaljik. Oxford University Press. Boeckx. New Horizons in the Analysis of Control and Raising. and Jairo Nunes. A. and Norbert Hornstein. 2008. On the locality and motivation of move and Agree: an even more minimal theory. Islands and Chains. Bobaljik. Control in Icelandic and theories of control. Psych verbs and theta theory. MA: MIT Press.). Linguistic Inquiry 40: 113–132. 2008. 2002.References 251 Belletti. 1997. Cedric. 2009. The virtues of control as movement. Cedric. ˇ Boˇskovi´c. and Norbert Hornstein. 2007. On multiple wh-fronting. . On (non-)obligatory control. In W. 2007. and Idan Landau. Boeckx. Movement under control. theta-criterion. Linguistic Inquiry 41(1): 111–130. and movement into theta-positions. Norbert Hornstein. and Jairo Nunes. Natural Language and Linguistic Theory 6: 291–352. A note on contraction. Linguistic Inquiry 31: 357–366. 1988. Cedric. In terms of merge: copy and head movement. The Copy Theory of Movement. Norbert Hornstein. Linguistic Variation Yearbook 8: 61– 100. Dordrecht: Springer. The Grammatical Basis of Linguistic Performance.ling. Reply to “Control is not movement. Rizzi. 1997. MIT Working Papers in Linguistics 27: 41–64. 2006b. Cambridge. Norbert Hornstein. In Norbert Corver and Jairo Nunes (eds. 2003. 251–262. Zeljko. Syntax 9: 118–130. and Norbert Hornstein. 1984. Cedric. The copy theory of movement: a view from PF. and Jairo Nunes. 1984.” Linguistic Inquiry 34: 269–280. Robert. Icelandic control really is A-movement: reply to Bobaljik and Landau. MA: MIT Press. 1995. University of Maryland Working Papers in Linguistics 15: 1–45. Inter-arboreal operations: headmovement and the extension requirement. Linguistic Inquiry 38: 589–644. Linguistic Inquiry 33: 351–383. and Jairo Nunes. Boeckx. Boeckx. Linguistic Inquiry 37: 591–606. Linguistic Inquiry 35: 431–452. Jonathan. 2000. Linguistic Inquiry 28: 345–356. and L. Bare Syntax.edu/publications/volume15 Boeckx. 252 References Bowers, John. 1973. Grammatical relations. Doctoral dissertation. MIT. 2006. On reducing control to movement. Syntax 11: 125–143. Bresnan, Joan. 1982. Control and complementation. In Joan Bresnan (ed.). The Mental Representation of Grammatical Relations. Cambridge, MA: MIT Press, 282–390. Britto, Helena. 1997. Deslocados a` esquerda, resumptivo-sujeito, ordem SV: a express˜ao do juizo categ´orico e t´etico no portuguˆes do Brasil. Doctoral dissertation. Universidade Estadual de Campinas. Castillo, Juan Carlos, John Drury, and Kleanthes Grohmann. 1999. Merge over move and the extended projection principle. University of Maryland Working Papers in Linguistics 8: 63–103. Cecchetto, Carlo, and Renato Oniga. 2004. A challenge to null case theory. Linguistic Inquiry 35: 141–149. Chierchia, Gennaro. 1984. Topics in the syntax and semantics of infinitives and gerunds. Doctoral dissertation. University of Massachussetts. Chomsky, Carol. 1969. The Acquisition of Syntax in Children from 5 to 10. Cambridge, MA: MIT Press. Chomsky, Noam. 1955. The Logical Structure of Linguistic Theory. New York: Plenum Press. 1957. Syntactic Structures. The Hague: Mouton. 1964. Current Issues in Linguistic Theory. The Hague: Mouton. 1970. Remarks on nominalisation. In R. Jacobs and P. Rosenbaum (eds.). Readings in English Transformational Grammar. Waltham, MA: Ginn, 184–221. 1977. On wh-movement. In P. W. Culicover, T. Wasow, and A. Akmajian (eds.). Formal Syntax. New York, NY: Academic Press, 71–132. 1980. On binding. Linguistic Inquiry 11: 1–46. 1981. Lectures on Government and Binding. Dordrecht: Foris. 1982. Some Concepts and Consequences of the Theory of Government and Binding. Cambridge, MA: MIT Press. 1986a. Barriers. Cambridge, MA: MIT Press. 1986b. Knowledge of Language: Its Nature, Origin and Use. New York: Praeger. 1993. A minimalist program for linguistic theory. In Kenneth Hale and Samuel Jay Keyser (eds.). The View from Building 20: Essays in Linguistics in Honor of Sylvain Bromberger. Cambridge, MA: MIT Press, 1–52. 1994. Bare phrase-structure. MIT Occasional Papers in Linguistics 5. 1995. Categories and Transformations. In The Minimalist Program. Cambridge, MA: MIT Press, 219–394. 2000. Minimalist inquiries: the framework. In Roger Martin, David Michaels, and Juan Uriagereka (eds.). Step by Step: Essays on Minimalist Syntax in Honor of Howard Lasnik. Cambridge, MA: MIT Press, 89–155. 2001. Derivation by phase. In Michael Kenstowicz (ed.). Ken Hale: A Life in Language. Cambridge, MA: MIT Press, 1–52. 2004. Beyond explanatory adequacy. In A. Belletti (ed.). Structures and Beyond. Oxford University Press, 104–131. 2008. On phases. In R. Freidin, C. P. Otero, and M. L. Zubizarreta (eds.). Foundational Issues in Linguistic Theory: Essays in Honor of Jean-Roger Vergnaud. Cambridge, MA: MIT Press, 133–166. References 253 Chomsky, Noam, and Howard Lasnik. 1993. The theory of principles and parameters. In Joachim Jacobs, Arnim von Stechow, Wolfgang Sternefeld, and Theo Vennemann (eds.). Syntax: An International Handbook of Contemporary Research. Berlin: Walter de Gruyter, 506–569. Cormack, Annabel, and Neil Smith. 2004. Backward control in Korean and Japanese. University College of London Working Papers in Linguistics 16: 57–83. Corver, Norbert, and Jairo Nunes. 2007. The Copy Theory of Movement. Amsterdam: John Benjamins. Courtenay, Karen. 1998. Summary: subject-control verb PROMISE in English. http://linguistlist.org/issues/9/9-651.html Culicover, P., and R. Jackendoff. 2001. Control is not movement. Linguistic Inquiry 30: 483–512. Culicover, P., and R. Jackendoff. 2005. Simpler Syntax. Oxford University Press. Davies, W. D., and S. Dubinsky. 2004. The Grammar of Raising and Control: A Course in Syntactic Argumentation. Oxford: Blackwell. Dobrovie-Sorin, Carmen. 1994. The Syntax of Romanian: Comparative Studies in Romance. Berlin: Mouton de Gruyter. Dowty, David. 1991. Thematic proto-roles and argument selection. Language 67: 547– 619. Drummond, Alex. 2009. On the surprisingly constrained nature of sideward movement. Unpublished manuscript. University of Maryland. Duarte, Maria Eugˆenia. 1995. A perda do princ´ıpio “evite pronome” no portuguˆes brasileiro. Doctoral dissertation. Universidade Estadual de Campinas. 2000. The loss of the “avoid pronoun” principle in Brazilian Portuguese. In M. Kato and E. Negr˜ao (eds.). Brazilian Portuguese and the Null Subject Parameter. Madrid and Frankfurt am Main: Iberoamericana and Vervuert, 17–36. 2004. On the embedding of a syntactic change. Language Variation in Europe, Papers from ICLaVE 2: 145–155. Epstein, Samuel D., and T. Daniel Seely. 2006. Derivations in Minimalism. Cambridge University Press. Etxepare, Ricardo. 1998. The syntax of illocutionary force. Doctoral dissertation. University of Maryland. Farkas, Donka. 1988. On obligatory control. Linguistics and Philosophy 11: 27– 58. Ferreira, Marcelo. 2000. Argumentos nulos em Portuguˆes Brasileiro. MA thesis. Universidade Estadual de Campinas. 2004. Hyperraising and null subjects in Brazilian Portuguese. MIT Working Papers in Linguistics 47: Collected Papers on Romance Syntax, 57–85. 2009. Null subjects and finite control in Brazilian Portuguese. In J. Nunes (ed.). Minimalist Essays on Brazilian Portuguese Syntax. Amsterdam: John Benjamins, 17–49. Floripi, Simone. 2003. Argumentos nulos dentro de DPs em Portuguˆes Brasileiro. MA thesis. Universidade Estadual de Campinas. Floripi, Simone, and Jairo Nunes. 2009. Movement and resumption in null possessor constructions in Brazilian Portuguese. In J. Nunes (ed.). Minimalist Essays on Brazilian Portuguese Syntax. Amsterdam: John Benjamins, 51–68. 254 References Fodor, Jerry. 1975. The Language of Thought. New York: Thomas Y. Crowell. Freidin, Robert, and Rex Sprouse. 1991. Lexical case phenomena. In R. Freidin (ed.). Principles and Parameters in Comparative Syntax. Cambridge, MA: MIT Press, 392–416. Fujii, Tomohiro. 2006. Some theoretical issues in Japanese control. Doctoral dissertation. University of Maryland. Galves, Charlotte. 1998. T´opicos, sujeitos, pronomes e concordˆancia no Portuguˆes Brasileiro. Cadernos de Estudos Ling¨u´ısticos 34: 7–21. 2001. Ensaios sobre as Gram´aticas do Portuguˆes. Campinas: Editora da UNICAMP. Gonz´alez, Nora. 1988. Object and Raising in Spanish. New York: Garland. 1990. Unusual inversion in Chilean Spanish. In Paul M. Postal and Brian D. Joseph (eds.). Studies in Relational Grammar 3. University of Chicago Press, 87–103. Grinder, J. 1970. Super equi-NP deletion. Chicago Linguistic Society (CLS) 6: 297– 317. Grodzinsky, Yosef, and Tanya Reinhart. 1993. The innateness of binding and of coreference. Linguistic Inquiry 24: 69–101. Grohmann, Kleanthes K. 2003. Prolific Peripheries. Amsterdam: John Benjamins. Grosu, Alexander, and Julia Horvath. 1984. The GB theory and raising in Romanian. Linguistic Inquiry 15: 348–353. Grosu, Alexander, and Julia Horvath. 1987. On nonfiniteness in extraction constructions. Natural Language and Linguistic Theory 5: 181–196. Haddad, Youssef. 2007. Adjunct control in Telugu and Assamese. Doctoral dissertation. University of Florida. 2009. Copy control in Telugu. Journal of Linguistics 45: 69–109. Halle, Morris, and Alec Marantz. 1993. Distributed morphology and the pieces of inflection. In Kenneth Hale and Samuel Keyser (eds.). The View from Building 20: Essays in Honor of Sylvain Bromberger. Cambridge, MA: MIT Press, 111–176. Harley, Heidi. 1995. Subjects, events, and licensing. Doctoral dissertation. MIT. Henderson, Brent. 2006. Multiple agreement and inversion in Bantu. Syntax 9: 275–289. Higginbotham, J. 1980. Pronouns and bound variables. Linguistic Inquiry 11: 679–708. 1992. Reference and control. In R. Larson, S. Iatridou, U. Lahiri, and J. Higginbotham (eds.). Control and Grammar. Dordrecht: Kluwer, 79–108. Hornstein, Norbert. 1990. As Time Goes By. Cambridge, MA: MIT Press. 1995. Logical Form: From GB to Minimalism. Oxford: Blackwell. 1999. Movement and control. Linguistic Inquiry 30: 69–96. 2001. Move! A Minimalist Theory of Construal. Oxford: Blackwell. 2003. On control. In Randall Hendrick (ed.). Minimalist Syntax. Oxford: Blackwell, 6–81. 2007. Pronouns in minimal setting. In Norbert Corver and Jairo Nunes (eds.). The Copy Theory of Movement. Amsterdam: John Benjamins, 351–385. 2009. A Theory of Syntax: Minimal Operations and Universal Grammar. Cambridge University Press. Hornstein, Norbert, Ana Maria Martins, and Jairo Nunes. 2008. Perception and causative structures in English and European Portuguese: ␸-feature agreement and the distribution of bare and prepositional infinitives. Syntax 11(2): 198–222. References 255 Hornstein, Norbert, and Hirohisa Kiguchi. 2003. PRO gate and movement. Penn Working Papers in Linguistics 8: 101–114. Hornstein, Norbert, and Jairo Nunes. 2002. On asymmetries between parasitic gap and across-the-board constructions. Syntax 5(1): 26–54. Hornstein, Norbert, and Jairo Nunes. 2008. Adjunction, labeling, and bare phrase structure. Biolinguistics 2: 57–86. Hornstein, Norbert, Jairo Nunes, and Kleanthes Grohmann. 2005. Understanding Minimalism. Cambridge University Press. Hornstein, Norbert, and Paul Pietroski. 2009. Obligatory control and local reflexives: copies as vehicles for de se readings. Unpublished manuscript. University of Maryland. Hornstein, Norbert, and Jacek Witkos. 2003. Yet another approach to existential constructions. In Lars-Olof Delsing, C. Falk, G. Josefsson, H. Sigurðsson (eds.). Grammar in Focus: Festchrift for Anders Platzak. Lund: Lund University, 167– 184. Huang, C.-T. James. 1982. Logical relations in Chinese and the theory of grammar. Doctoral dissertation. MIT. Jackendoff, Ray. 1983. Semantics and Cognition. Cambridge: MIT Press. 1990. Semantic Structures. Cambridge: MIT Press. Jaeggli, Osvaldo. 1980. Remarks on to-contraction. Linguistic Inquiry 11: 239–245. 1986. Passive. Linguistic Inquiry 17: 587–622. J´onsson, J´ohannes G´ısli. 1996. Clausal architecture and case in Icelandic. Doctoral dissertation. University of Massachusetts. Kandybowicz, Jason. 2009. The Grammar of Repetition: Nupe Grammar at the Syntax– Phonology Interface. Amsterdam: John Benjamins. Kato, Mary A. 1999. Strong and weak pronominals and the null subject parameter. Probus 11(1): 1–38. 2000. The partial pro-drop nature and the restricted VS order in Brazilian Portuguese. In M. Kato and E. Negr˜ao (eds.). Brazilian Portuguese and the Null Subject Parameter. Madrid and Frankfurt am Main: Iberoamericana and Vervuert, 223–258. Kato, Mary A., and Jairo Nunes. 2009. A uniform raising analysis for standard and nonstandard relative clauses in Brazilian Portuguese. In J. Nunes (ed.). Minimalist Essays on Brazilian Portuguese Syntax. Amsterdam: John Benjamins, 93–120. Kayne, Richard. 1984. Connectedness and Binary Branching. Dordrecht: Foris. 1994. The Antisymmetry of Syntax. Cambridge, MA: MIT Press. 2002. Pronouns and their antecedents. In S. D. Epstein and T. D. Seely (eds.). Derivation and Explanation in the Minimalist Program. Oxford: Blackwell, 133–166. Kiguchi, Hirohisa. 2004. Syntax unchained. Doctoral dissertation. University of Maryland. Kiss, Tibor. 2005. On the empirical viability of the movement theory of control. Unpublished manuscript. Ruhr-Universit¨at Bochum. Kitahara, Hisatsugu. 1997. Elementary Operations and Optimal Derivations. Cambridge: MIT Press. Koopman, Hilda. 1984. The Syntax of Verbs. Dordrecht: Foris. Koster, Jan. 1987. Domains and dynasties. Dordrecht: Foris. 256 References Laka, Itziar. 2006. On the nature of case in Basque: structural or inherent? In H. Broekhuis, N. Corver, J. Koster, R. Huybregts, and U. Kleinhenz (eds.). Organizing Grammar: Linguistic Studies in Honor of Henk van Riemsdijk. Berlin: Mouton de Gruyter, 374–382. Landau, Idan. 1999. Elements of control. Doctoral dissertation. MIT. 2000. Elements of Control: Structure and Meaning in Infinitival Constructions. Dordrecht: Kluwer. 2003. Movement out of control. Linguistic Inquiry 34: 471–498. 2004. The scale of finiteness and the calculus of control. Natural Language and Linguistic Theory 22: 811–877. 2006. Severing the distribution of PRO from case. Syntax 9: 153–170. 2007. Movement-resistant aspects of control. In W. D. Davies and S. Dubinsky (eds.). New Horizons in the Analysis of Control and Raising. Dordrecht: Springer, 293– 325. Lasnik, Howard. 1995. Case and expletives revisited. Linguistic Inquiry 26: 615– 633. Lasnik, Howard, and Mamoru Saito. 1992. Move A. Cambridge, MA: MIT Press. Lasnik, Howard, and Juan Uriagereka. 1988. A Course in GB Syntax: Lectures on Binding and Empty Categories. Cambridge, MA: MIT Press. Lebeaux, David. 1985. Locality and anaphoric binding. The Linguistic Review 4: 343– 363. Lee, Felicia. 2003. Anaphoric R-expressions as bound variables. Syntax 6: 84– 114. Lees, Robert B. 1960. The grammar of English nominalizations. The Hague: Mouton de Gruyter. Lidz, Jeff, and William Idsardi. 1997. Chains and phono-logical form. UPenn Working Papers in Linguistics 8: 109–125. Lightfoot, David. 1976. Trace theory and twice-moved NPs. Linguistic Inquiry 7: 559– 582. Manzini, Maria Rita, and Anna Roussou. 2000. A minimalist theory of A-movement and control. Lingua 110: 409–447. Marantz, Alec. 1991. Case and licensing. In Germ´an F. Westphal, Benjamin Ao, and Hee-Rahk Chae (eds.). Proceedings of the Eighth Eastern States Conference on Linguistics. Columbus, OH: Ohio State University, 234–253. Martin, Roger. 1996. A minimalist theory of PRO and control. Doctoral dissertation. University of Connecticut. 2001. Null case and the distribution of PRO. Linguistic Inquiry 32: 141–166. Martins, Ana Maria. 2001. On the origin of the Portuguese inflected infinitive: a new perspective on an enduring debate. In L. J. Brinton (ed.). Historical Linguistics 1999: Selected Papers from the 14th International Conference on Historical Linguistics. Amsterdam: John Benjamins, 207–222. Martins, Ana Maria, and Jairo Nunes. 2005. Raising issues in Brazilian and European Portuguese. Journal of Portuguese Linguistics 4: 53–77. Martins, Ana Maria, and Jairo Nunes. 2008. Personal and impersonal infinitives in European Portuguese and obligatory control. Unpublished manuscript. Universidade de Lisboa and Universidade de S˜ao Paulo. Linearization of chains and phonetic realization of chain links. Aboh. In E. UC Berkeley. On the identification of null arguments. Unpublished manuscript. McGinnis. 2008c. Philip J.). Sideward movement. Somerville. Backward object control in Korean. Cambridge. 2004. 1999. A-over-A. A note on wh-islands and finite control in Brazilian Portuguese. 2001. Martha. Crisma and G. Doctoral dissertation. Romance Languages and Linguistic Theory: Selected Papers from “Going Romance” 2007. Relativizing Minimality for A-movement: ␾. Nicol. In E. Swinney. J. J. Estudos da Lingua(gem). Nunes. GLOW Newsletter 58. The role of structure in coreference assignment during sentence processing. P.References 257 Martins. Longobardi (eds. Doctoral dissertation. Jon. Cambridge. MIT. E. Boeckx (ed. and output. 2000. Investigations of covert phrase movement. Linguistics in Potsdam 28: Optimality Theory and Minimalism: Interface Theories. Sideward movement: triggers. In G. The copy theory of movement and linearization of chains in the minimalist program.). Doctoral dissertation. 2000. Linearization of Chains and Sideward Movement. Preposition insertion in the mapping from spell-out to PF. In P. 133–156. 2009. Quer. and Jairo Nunes. The Complementiser Phase: Subjects and Wh-Dependencies. O Portuguˆes Brasileiro: uma l´ıngua voltada para o discurso. WCCFL 22 Proceedings. Monahan. In press. Dummy prepositions and the licensing of null subjects in Brazilian Portuguese. inherent case. Ana Maria. Marcello. Tsujimura (eds. 217–249.). Nissenbaum.). Negr˜ao. Doctoral dissertation. Apparent hyper-raising in Brazilian Portuguese: agreement with topics across a finite CP.. Syntactic change as chain reaction: the emergence of hyper-raising in Brazilian Portuguese. Esmeralda. Garding and M. The Oxford Handbook of Linguistic Minimalism. van der Linden. MA: Cascadilla Press. Amsterdam: John Benjamins. Journal of Psycholinguistic Research 18: 5–20.). 2003. 2007. In Samuel David Epstein and Norbert Hornstein (eds.). 2009a. MA: MIT Press. David. Oxford University Press. Panagiotidis (ed. Probus 22: 1–25. Jairo. 1999. Doctoral dissertation. . and P. Paper presented at Ways of Structure Building. The copy theory. 1995. Sleeman (eds. 2008b. Ana Maria. University of Southern California. 356–369. Modesto. timing. 243–265. Oxford University Press. Linguistic Inquiry 31(2): 303–344. and relativized probing. Mortensen. and D. MA: MIT Press. Universidade de S˜ao Paulo. 2010. Working Minimalism. 2009b. Two kinds of variable elements in Hmong anaphora. Inherent case as a licensing condition for A-movement: the case of hyperraising constructions in Brazilian Portuguese. Locality in A-movement. Journal of Portuguese Linguistics 7: 83–108. 2003. In press. and Jairo Nunes. 2008a. Historical Syntax and Linguistic Theory. MIT. In C. 1998. University of Maryland. 1989. Oxford University Press. 144–157. University of the Basque Country. Martins.and ␪-relations. O’Neil. Unpublished manuscript. Reinhart. Amsterdam: John Benjamins. Movement in language. Colin. Luigi. O T´opico no Portuguˆes do Brasil. Lightfoot (ed. Polinsky. 277– 305. 2004. Field notes on white Hmong reflexives. L. Out of control. Linguistic Inquiry 33: 245–282. Dordrecht: Foris.). Syntax 9: 171–192. David Montero. Cambridge. Cyclicity and extraction domains. 1995. and Eric Potsdam. 1990. Eric. Rezac. 97–133. 1987. 2006. Linguistic Inquiry 14: 101–136. 2006. MA: Cascadilla Press. The real-time status of island constraints. Pontes. Raposo. and Eric Potsdam. Object control in Korean: how many constructions? Language Research 43: 1–33. 1996. Tanya. Reuland. 1983. Event-related brain potentials elicited by failure to agree. Linguistic Inquiry 9: 1–29. C. WH-Movement: Moving On. Relativized minimality.258 References Nunes.). Jairo. Order and structure. In Donald Baumer. 2001. MA: MIT Press.). Pires. 2007. Syntactic Effects of Morphological Change. A. Governing -ing. Nu˜nez Cede˜no (eds. . Traces and the description of English complementizer contraction. J. 2001. Phillips. and Nayoung Kwon. Prepositional infinitival constructions in European Portuguese. and Michael Scanlon (eds. 1989.). 2004. Backward control. Cambridge. 1983. Quinn. Unpublished manuscript. 328–336. On the form of chains: criterial positions and ECP effects. 160–178. The Null Subject Parameter. Studies in Romance Languages. MA: MIT Press. Eunice. 2006. Polinsky. Jaeggli and K. Neidle and R. Bogazici University. In O. Expanding the scope of control and raising. University of Maryland. Eduardo. 1978. In Lisa Cheng and Norbert Corver (eds. Rizzi. NELS 25: 361–371. Paul. In C. Doctoral dissertation. 2002. Campinas: Pontes. Romance infinitival clauses and case theory. and Geoffrey Pullum. 1995. Morphology and null subjects in Brazilian Portuguese. Backward object control in Malagasy: evidence against an empty category analysis. 237–249. Osterhout. In D. Proceedings of the 25th West Coast Conference on Formal Linguistics. 2006. and L. Cilene. 2002. Safir (eds. MA thesis. Control in Turkish. Acrisio. Somerville. Philip Monahan.. Postal. Elements of cyclic syntax: Agree and merge. and Juan Uriagereka. A. Eric. Syntax 3: 20–43. The syntax of gerunds and infinitives: subjects. London: Croom Helm. Doctoral dissertation. 2000. University of Toronto. Harvard University. case and control. Dordrecht: Kluwer. The Minimalist Syntax of Defective Domains: Gerunds and Infinitives. Oxford University Press. Oxford University Press. Norvin. Maria. Richards. Milan. Maria. MIT. Mobley. 2006. Polinsky. Potsdam. Ilknur. Anaphora and Semantic Interpretation. 2004.). Rodrigues. 1987. Oded. Journal of Memory and Language 34: 739–773. University of Maryland. Doctoral dissertation. Maria. and Barbara Hall Partee. In Mar´ıa Luisa Rivero and Angela Ralli (eds.. London: Routledge. Tim. MIT. Cambridge. Waltham. Torrego. Notre Dame Journal of Formal Logic 27: 401–429. 1979. Reflexivity.). The Grammar of English Predicate Complement Constructions. Jacobs and P. Doctoral dissertation. Davies and S. Halld´or Armann. Icelandic finite verb agreement. Salmon. D.). Dubinsky (eds. P. University of Maryland. Shlonsky. 1997. D. split control and super-equi. On certain issues of control in English and Polish: partial control. 1996.). Oxford University Press. Control and raising in and out of subjunctive complements. 2008. New Horizons in the Analysis of Control and Raising. 2007. Control via selection. Johan. 1981. New York: Holt. Natural Language and Linguistic Theory 26: 403–450. Dubinsky (eds. Dordrecht: Springer. Doctoral dissertation. 20–29. Davies and S. MA: MIT Press. 1973. Roussou. Sichel. On declarative sentences. 1970. Rosenbaum. 1970. N. Linguistic Inquiry 27: 111–126.). In R. Paul Schachter. 2001. Anna. Itziar. In W. MA: Ginn. Doctoral dissertation. Configurations of Sentential Complementation: Perspectives from Romance Languages. H¨oskuldur. MA: Ginn. Rinehart and Winston. . 2007. Poznan. The tense of infinitives. Readings in English Transformational Grammar. New Horizons in the Analysis of Control and Raising. Harvard University. 1982. Jacobs and P. A principle governing deletion in English sentential complementation. 15–34. Rooryck. Adam Mickiewicza University. Waltham. In W. Oxford University Press. On subordination and the distribution of PRO. Ross. Readings in English Transformational Grammar.). Natural Language and Linguistic Theory 9: 327–363. 213–229. 1991. PRO and null case in finite clauses. The Syntax of Icelandic. Stockwell. 2008. Working Papers in Scandinavian Syntax 57: 1–46. 2001. 74–104. Ivy. Robert P. Clause Structure and Word Order in Hebrew and Arabic: An Essay in Comparative Semitic Syntax. On complementation in Icelandic. Esther. Anna. Arontho. Agreement and flotation in partial and inverse partial control configurations. The Linguistic Review 14: 335–360. In R. 2007. Doctoral dissertation. Dordrecht: Springer. University of Maryland. Comparative Syntax of the Balkan Languages. 222– 272. Davies and S. 1986. The Major Syntactic Structures of English. Doctoral dissertation. Impoverished morphology and A-movement out of case domains. Rosenbaum (eds. Stowell. John R. Dubinsky (eds. 1997. D. 281–292. Dordrecht: Springer. 2004. 1996. Ur. San Martin. On quantifier float in control clauses. Terzi. ´ Sigurðsson. 2009. Cambridge University Press. Linguistic Inquiry 13: 561–570. S. Icelandic case marked PRO and the licensing of lexical arguments. New Horizons in the Analysis of Control and Raising.). The origins of phrase structure. In W. Thr´ainsson.References 259 2004. Snarska. Rosenbaum (eds. Raising in DP revisited. The case of PRO. 1967. Doctoral dissertation. 267–298. Varieties of raising and the feature-based bare phrase structure theory. MA: MIT Press.). What it means to agree: the behavior of case and phi features in Icelandic control. 2008. Case in tiers. Tense in infinitives. Zaenen. 2000. 269–304. Language 63: 217– 250.). Yip. Haynie (eds.” Harvard University. 1985. Unpublished manuscript. Rhyme and Reason: An Introduction to Minimalist Syntax. Wurmbrand. Epstein and T. MA: MIT Press. it fails to copy. Proceedings of WCCFL 26. 2006. J. Asako. Paper presented at the LSA Linguistic Institute Workshop “New Horizons in the Grammar of Raising and Control.). Varlokosta. 2005. Doctoral dissertation.” Harvard University. Johan Rooryck. Edwin. Williams. Working Minimalism. Juan. Infinitives: Restructuring and Clause Structure. 1980. 8–10. Joan Maling. 1985. University of Connecticut. WollP: where syntax and semantics meet. If control raises.-W. Spyridoula. Ura. Control in Modern Greek. University of Maryland. 1994. 1987. Uriagereka. D. July. Cambridge. Predication. D.260 References Uchibori. MA: Cascadilla Press. July. van Craenenbroeck. Seely (eds. Oxford: Blackwell. Moira. reconstruct. Natural Language and Linguistic Theory 3: 277– 295. Multiple Spell-Out. Cherlon. C. and linearize. Paper presented at the LSA Linguistic Institute Workshop “New Horizons in the Grammar of Raising and Control. MIT Working Papers in Linguistics 23: 297–316. In Samuel David Epstein and Norbert Hornstein (eds. . 2005. In C. 1998. 251–282. Agreement systems. and Guido van den Wyngaerd.). B. 2006. J. 1993. In S. Linguistic Inquiry 11: 203–238. Derivation and Explanation in the Minimalist Program. Susanne. Cambridge. Amsterdam: John Benjamins. Chang and H. Annie. The syntax of subjunctive complements: evidence from Japanese. Issues relating to a derivational theory of binding. Somerville. PRO and the subject of NP. Berlin: Mouton de Gruyter. Case and grammatical functions. Jeroen. 8–10. University of Connecticut. 2001. In Cedric Boeckx (ed. Complete and partial Infl. Joan Maling. 2002. Hiroyuki. 1999. Ussery. and H¨oskuldur Thr´ainsson. Natural Language and Linguistic Theory 3: 441–483. Zwart. 480–488. and Ray Jackendoff. 75. 46. 200. 196. 100–101. 21. 72. 126. 38. Tibor. 136. 8 exceptional case-marking verb. 185–193 control adjunct. 48. 53. 79. 239–240 Fujii. 247 inherent case. 59. 151. 138. 141–152. 68. 18–25. 66. 242. 67. 21. 70 heavy NP shift. 129. 196–198 Kinande. 90. 77. 49–51. 143. 153–154. 128. 140. 133 commitative PP.Index acquisition. 9–16. 240 Culicover. 53. 151. 57. 208. 216 Japanese. 52. 57. 234 freezing. 122 equivalent NP deletion. 194. 216 German. 41. 182 copy theory of movement. 39–43 partial. 238. 80. 132–134 government and binding theory. 101. 190–193 economy. 53. 48. 228. Noam. 195–209. 165–166 c-command. 99. 16. 62. 125. 219. 238. Ray. 133–153. 90. 92. 173–181 D-structure. 221. 84. 80 ˇ Boˇskovi´c. 14. 90. 45. 132–136 261 . 80–90. 46–48. 203. 244. 72. 108. 239 Assamese. 228 Ferreira. 11. 47. Zeljko. 13. 88. 155–159 inclusiveness condition. 196–199 double-object construction. 13. 49. 159. 130. 3. 42. 170. 170. 128. 57. 158. 5. 239–240 ellipsis. 68 case transmission. 149 Icelandic. 90. 223 activation condition. 86. 126–128. 34. 245 Basque. 25. 49. 107. 127. 244 clitic. 247 Italian. 78. 160. 70. 104. 79. 147. 93 bare phrase structure. 12. 99. 3. 148. 160–166 Fodor’s problem. 148. 7. 168. 36. Tomohiro. 151. 16. 64. 11–14 binding theory. 226. 147–149 Hmong. 120. 52. 47. 64. 190–193 case theory. 36. 92. 121. 150. 29. 16–19. 31. 70. 53–55. 28. 194. 27. 86. 241–248 Jackendoff. 22–34. 63. 134. 102 Chomsky. 141–152. 181 Hebrew. 120. 1. 145. 143 floating quantifier. 216. 13. 69 Kiss. 48. 68. 87. 63 A-movement. 117. 6. 226 module. 246 Greek. 24. 43. 26. 190–193 expletive. 219 island. 80. 104. Marcelo. Peter. 16. 35. 1. 150. 89 de se interpretation. 1. 153 exhortative construction. 207. 160–168 idiom. 166–168 binding domain. 75. 121 hyper-raising. 240 complement. 60. 215. 91. 48 Agree. 57. 173 extension condition. 54 super-equi. 20–24. 80. Cilene. 206. 195. Peter S. 194 reconstruction effect. 85. 172. 116–119. 227 Wurmbrand. 36. 71–74. David. 53. 96–97. 197. Susi. 203. 90. 56. 15 Visser’s generalization. 179. 170. 214 Rosenbaum. 62–63. 52 strict reading. 99–101. Johan.262 Index Koopman. 63. 227 ´ Sigurðsson. 147–150 sideward movement. 44. Tanya. 125. 152. 187. 49. 43. 79 redundancy rule. 103. 3. 110–112. 39. 124. 125. 89. 90. 188–189 split antecedent. 215.. 24. 175 quirky case. 191. 200. 21. 129 phase. 39. 111 selection restriction. 242 Latin. 239 minimalist program. 147. 92. 224 transparency. 219 parser. 197 Romanian. 117–119 Korean. 182 standard theory. 31–33. 46. 201. 64. 50 San Lucas Quiavin´ı Zapotec. 129. 182. 153. Halld´or Armann. 126. 77. 177. 17–20. 213. 133. 16. 228. 53. 199 Telugu. 238 nominalization. 105 Rooryck. 69. 223 Salmon. 196. 203. 219–225 Vata. 132. 176. 204–209. 118–119 visibility condition. 60 linear correspondence axiom. 130. 113 Landau. 136 voice transparency. 76. 89. 38 PRO theorem. 152. Hilda. 16. 38. 237 Lightfoot. 227 principle C. 64–66. 47. 110 wanna-contraction. 17–20. 224 null case. 172. 6. 169. 127. 6. 32. 122 thematic role. 211–216. 200. 244 scrambling. 136–140. 79. 224 long passive. Howard. 197–198 Spanish. 193 Lasnik. 24. 146. 198–200. 145 minimal-distance principle. 66. 183–186. 212 uniformity of theta assignment hypothesis (UTAH). 83. 170. 7. 18. 74. 244 pair-merge. 169. 190. 211 . 67–68. 77. 20–35. 13 pronominalization. 120–121. 49. 138. 107. 196–198. Nathan. 160. 227 null pronoun. 212 Satisfy. 57. 49. 65 Reinhart. 216–226. 180–181. 61 lexical semantics. 204–209 passive. 169. 194 Rodrigues. 130. 164–165 sloppy reading. 229 Sichel. 47. 121. 59 wh-movement. 151. 142–146. 145 set-merge. 53. 62–63. 50 relativized minimality. 212. 116–118 linearization. 129. 188. 197–209 proto-role. 31 Portuguese Brazilian. 228 European. 90. 86. 152. 106–109. 227. 146 merge. 171. 160. 11–12. 172–182. 187. 186–189. Idan. Ivy. 241 principles-and-parameters theory. 70. 42. 204. 78. 120. 106. 206–208 Tsez.