University of Southern California Dissertations and Theses

Copy theory of movement and PF conditions on spell-out

(USC Thesis Other)

Copy theory of movement and PF conditions on spell-out

PDF

Download a page range

Download transcript

Copy asset link

Request this asset

Transcript (if available)

Content COPY THEORY OF MOVEMENT AND PF CONDITIONS ON SPELL-OUT Chorong Kang Submitted to the USC Graduate School in partial fulfillment of the requirements for the degree of Doctor of Philosophy in Linguistics at the University of Southern California August, 2017 Approved by: ________________________________ Andrew Simpson, chair ________________________________ Maria Luisa Zubizarreta, member ________________________________ Yen-hui Audrey Li, member ________________________________ Mario Saltarelli, member ii Acknowledgements Your word is a lamp for my feet, a light on my path. Psalm 119. During my five years at USC, I dare understand how Israelites feel when they travel guided by a pillar of cloud and a pillar of fire in the wilderness. When I arrived in LA, I had no family, no friend, even no place to live. I was just overwhelmed with loneliness and fear. However, God have guided me to meet good people, to live in a good place, and finally to finish the program. I praise God for all that God have done for me. I would first like to give my deepest gratitude to my committee members, Andrew Simpson, Maria Lusia Zubizarreta, Audrey Li, and Mario Saltarelli. Especially, I don’t even have the words to thank my advisor Andrew. Andrew patiently advised and encouraged me when I was stuck and lost. His kind words raised me up when I was thinking of giving up many times. He has read, commented on and revised many drafts, papers, chapters, and abstracts. His comments were always to the point and made me find out a big picture. I am deeply grateful to Maria Luisa. It has been a great privilege for me to work closely with her. Her comments and questions were always insightful leading me where I should head to. I have benefited a great deal from comments from Audrey. Her precise comments and criticisms helped me greatly with elaborating my vague ideas. I specially thank Mario for his help even though I asked him to join my committee very late. I am also thankful to other faculty members of the department. I owe many thanks to Elsi Kaiser. Without her training and comments, I could not even start an experimental work. I will never fail to remember that she instructed me every step of conducting an experiment. I am grateful to Roumyana Pancheva, Barry Schein, Hajime Hoji, Elena Guerzoni for their helpful comments and encouragement. My thanks go to Khalil Iskarous, Jason Zevin, Mary Byram Washburn for their understanding and help when I was working as TA. My special thanks go to professors in SNU, Jae-il Kwon, Juwon Kim, Seungho Nam, Hyopil Shin, SeungJae Lee, Ho-young Lee, Jongho Jun, Minhwa Chung. When I visited SNU during vacation or when they visited LA, they warmly encouraged and supported me. They kindly helped me out to conduct my experiment in SNU. Especially, I would like to express my big gratitude to Heejeong Ko for her countless helpful comments, supports, and encouragement from my undergraduate years till now. Many thanks go to So-Young Park for her warm help and support iii in my last year at USC during pregnancy. I am grateful to Eunjeong Oh and Sanghoun Song for their interest in my work. Without my fellow students, it was impossible for me to survive at USC. Many, many thanks to wonderful people, Cynthia Lee, Monica Do, Jessica Harmon, Samantha Gordon, Thomas Borer, Andrés Benítez, Alif Silpachai, Ulli Steindl, Hayeun Jang, Ana Besserman, Saurov Syed, Haley Wei, Alfredo García Pardo, Charles O'Hara, Caitlin Smith. Special thanks to Binh Ngo, Arunima Choudhury, and Priyanka Biswas for their warm friendship, for being always there and for helping me whenever I needed help. Special thanks to Songhee Kim and Dongwoo Park for thir fruitful discussion and encouragement. I also owe gratitude to LAUC family for their endless support and pray. Special thanks to Pastor Sumi and Patricia Oh for being there when I need “mom’s love”. I thank Mingyung Shin, Narae Lee, and Yohyun Shin for being always so considerate while we were sharing a place. I thank Hyejung Shin for sharing laughs and conversations on everything. I am grateful to many friends in Korea for their help in collecting judgments and for their emotional support. Special thanks to SNU graduate students and Aeji Hong, Eunji Lim, Jayoun yoo, Jiyoon Lee, Saemina Kim. Finally, I would like to express my profound gratitude to my family, my father, my mother, my sister, my brother, and my father in-law, my mother in-law, and my sister in-law for their unfailing love and support. I believe that God sent me my husband, Yohan Jung, to show God’s greatest love, understanding, and support through him. Komawe, from the bottom of my heart. My last thanks go to my son Dan for being my happiness that I can’t put into words. iv Abstract of the dissertation COPY THEORY OF MOVEMENT AND PF CONDITIONS ON SPELL-OUT by Chorong Kang B. A., Seoul National University M.A., Seoul National University Ph.D., University of Southern California Chair: Professor Andrew Simpson In this dissertation, I investigate how agreement, movement, and pronunciation determine surface position of a phrase. The main interest of this dissertation is in the question why in some cases an element is pronounced in the position where it is interpreted while in other cases, there is a discrepancy between the position for interpretation and the position for pronunciation. To investigate this issue, I will first discuss a relation between agreement and movement. Inspired by Reverse Agree (Wurmbrand 2012), I will clarify a condition of movement. Based on the suggested relation between agreement and movement, I will propose three different types of movement: phrasal movement, parasitic phrasal movement, and parasitic head movement. The crucial difference between phrasal movement and parasitic (phrasal/head) movement is that an element does undergo movement in the case of phrasal movement, while an element does NOT undergo movement in the case of parasitic movement. Following the copy theory of movement, I assume that an element that undergoes phrasal movement leaves its copies in its base position and its destination along with other intermediate positions, unlike one that undergoes parasitic movement. Thus, PF needs to choose which copy to pronounce when there is more than one copy in a chain. Zooming in the issue of selection of copy for pronunciation, I will discuss PF constraints that play a role in copy-selection. Especially, adopting Landau (2007)’s intuition that EPP is a pure PF requirement, I will argue v that a high-copy privilege assumed in the previous studies are misled by pronunciation-wise reinterpreted EPP. Furthermore, I will argue that once we get rid of the effects of EPP, a low copy is rather preferred to be selected for pronunciation due to economy conditions. I will show how the interaction between the EPP as a PF constraint and an economy condition favoring low-copy pronunciation accounts for both (i) prevalent high-copy pronunciation and (ii) apparent lack of a high-copy privilege across languages. Based on the system developed, I will provide a typological study in two representative cases of movement: (i) subject agreement/movement and (ii) wh-agreement/movement. This system provides a new approach for the typology of in-situ subjects and in-situ wh-phrases. In the proposed system, in-situ subject/wh-phrases are the results of either parasitic movement or low-copy pronunciation in phrasal movement. An in-situ phrase generated by parasitic movement does not have a copy in a higher position, so it cannot take a high scope. Furthermore, since the phrase does not undergo movement, it is insensitive to movement constraints (e.g. island constraints). By way of contrast, an in-situ phrase generated by a low-copy pronunciation in a movement chain shows “high” behaviors in addition to sensitivity to movement constraints. I will show how the two theoretically possible in-situ subjects/wh-phrases are realized in languages. Furthermore, based on the definition of the EPP, I will propose various ways to satisfy the EPP. Instead of filling the specifier position of a functional head, I will argue that morphology can be inserted as a means of satisfying the EPP. In addition, I assume that morphology insertion as a way of satisfying the EPP includes a case of inserting an intonation morpheme, suggested by Cheng and Rooryck (2000). This additional way to satisfy the EPP accounts for a range of otherwise puzzling prosody relevant phenomena: i) acceptance of inverse scope reading in so- called rigid scope languages with the help of prosody; ii) ameliorating the intervention effects in Korean and Japanese with the help of prosody. Finally, by reviewing previous studies, I will show how the proposed system accounts for exceptional low-copy pronunciation. vi Table of Contents Chapter 1. Introduction ................................................................................................................1 1.1. Overview .......................................................................................................................................... 1 1.2. The Copy theory of movement and low-copy pronunciation ................................ 5 1.2.1. Bobaljik (2002) ............................................................................................... 6 1.2.2. Groat and O'Neil (1996) ......................................................................................................... 11 1.2.3. Nunes (1995, 2004, 2011): Chain Reduction and Linearization .............................. 13 1.2.3.1. Chain formation and feature checking ................................................................. 13 1.2.3.2. Chain formation and feature checking ................................................................. 16 1.2.4. Bošković (2002) and Bošković and Nunes (2007) ........................................................ 18 1.2.5. Pérez (2015) .............................................................................................................................. 24 1.2.6. Stjepanović (2007) .................................................................................................................. 28 1.3. Low-copy pronunciation and lack of high copy privilege ....................................................... 32 1.3.1. Low-copy pronunciation in argument domain ................................................................ 32 1.3.2. Low-copy pronunciation in non-argument domain ....................................................... 41 1.4. Towards an alternative: Organization of dissertation .............................................................. 46 Chapter 2. Agree, Movement, and Pronunciation ..................................................................48 2.1. Overview: Dissociating Agree, the EPP, and Spell-out at PF ............................................... 48 2.2. Agreement and movement .......................................................................................................... 52 2.3. EPP as a PF constraint ................................................................................................................. 65 2.4. Cyclic Spell-out and low copy privilege .................................................................................. 77 2.5. Conclusion .................................................................................................................................... 95 Chapter 3. Typological study ....................................................................................................97 3.1. Overview ....................................................................................................................................... 97 3.2. Typology in argument domain – Focusing on subject agreement ......................................... 98 3.2.1. Type A: (S)V order languages with rich agreement morphology ............................ 101 3.2.2. Type B: SV order languages without rich agreement morphology and vii Type D: VS order languages with full agreement ............................................ 107 3.2.3. Type C: VS order languages with an alternate between full/partial agreement ..... 108 3.3. Typology in non-argument domains – Implications for the typology of in-situ wh-phrases ......................................................... 115 3.3.1. Type A: Wh-movement languages with wh-expletives ................................................... 118 3.3.2. Type B: Wh-fronting languages with optional in-situ wh-phrases ............................... 126 3.3.3. Type C: Obligatory wh-fronting languages ........................................................................ 130 3.3.4. Type D: Canonical wh-in-situ languages ............................................................................... 131 3.4. Typology of in-situ wh-phrases ............................................................................................... 141 3.5. Conclusion .................................................................................................................................. 143 Chapter 4. Case Study – Interaction between prosody and low-copy pronunciation in Korean .........................144 4.1. Overview ....................................................................................................................................... 144 4.2. Case 1 – A low-copy pronunciation in wh-chain .................................................................. 146 4.2.1. Sinhala vs. Korean/Japanese ..................................................................................................... 146 4.2.2. So-called question particles are not a genuine question particle .................................. 155 4.2.3. Intonation as a realization of abstract question particle .................................................. 157 4.2.4. Pied-piping of wh-phrase and its phonetic realization .................................................... 164 4.3. Case 2 – A low-copy pronunciation in discourse-configurational chain ........................... 171 4.3.1. Background .................................................................................................................................. 172 4.3.2. Korean topic movement ...................................................................................................... 178 4.3.2.1. Hypothesis ......................................................................................................................... 178 4.3.2.2. Design and Materials ...................................................................................................... 179 4.3.2.3. Participants ........................................................................................................................ 188 4.3.2.4. Procedure ........................................................................................................................... 189 4.3.2.5. Predictions ......................................................................................................................... 189 4.3.2.6. Results ................................................................................................................................ 190 4.3.2.7. Discussion ......................................................................................................................... 193 4.3.3. Korean focus movement ...................................................................................................... 195 viii 4.4. Return to the typology of subjects: Type E languages, “discourse-configurational languages” ............................................................................................................................................ 197 4.5. Conclusion ................................................................................................................................................. 204 Chapter 5. The interaction between selection of copy and morphology/phonology ..........205 5.1. Overview ................................................................................................................................................... 205 5.2. Phonological constraint and low-copy pronunciation .................................................................. 205 5.2.1. Prosodic constraint and low-copy pronunciation .............................................................. 205 5.2.2. Non-prosodic, but phonological constraint and low-copy pronunciation .................... 217 5.3. Morphology and low-copy pronunciation ........................................................................................ 219 5.3.1. In-situ wh-phrases in Coptic Egyptian ................................................................................... 219 5.3.2. In-situ topic phrases in Korean ................................................................................................. 224 5.4. Conclusion ................................................................................................................................................. 227 Chapter 6. Conclusion ...............................................................................................................229 6.1. Summary of the discussion ................................................................................................................ 229 6.2. Toward future research ........................................................................................................................ 229 References ...................................................................................................................................234 1 Chapter 1. Introduction 1.1. Overview One prevalent phenomenon observed across languages is the discrepancy between the position where an element is pronounced and the position where the element is interpreted. This discrepancy leads linguists develop theories on “movement”. A movement operation is assumed to displace an element to a functional domain where the element is in an Agree relation leaving behind its trace in the base position (and in intermediate drop-by positions). As the Minimalist Program was introduced, Chomsky (1993) proposed the Copy theory of movement, which substitutes the movement operation with three sequential operations: Copy, Merge and Deletion. Under the copy theory of movement, there is no operation of “displacement” per se. Rather, an element is copied and one copy is merged into a functional category, all the other copies except one (or two, in some cases) which will be phonetically realized, are deleted (I will discuss the copy theory of movement in section 1.2. in detail. In this dissertation, I will keep using the term “movement” for explanatory convenience to refer to the copy-(internal) merge-deletion process). Such a change in the notion of movement paradoxically opens up the possibility that, in principle, the lowest member of a chain can be phonetically realized as if a movement has not taken place in the surface representation. Nevertheless, although nothing in principle requires that deletion must target lower copies, it seems that the highest member of a chain is more likely to be phonetically spelled-out in the majority of so-called movement chains. To the best of my knowledge, Nunes (1995, 2004, 2011) is the first work that explores this issue thoroughly. To account for the (seemingly) highest-copy privilege in pronunciation, Nunes proposes a Chain Reduction Algorithm (Section 1.2). The Chain Reduction Algorithm privileges the highest copy to be pronounced based on PF economy conditions and allows some exceptions for non-highest copy pronunciation when an extra phonological constraint blocks the highest copy from being pronounced. Criticizing Nunes’ Chain Reduction algorithm for theoretical reasons, Pérez (2015) argues for an alternative account for the high-copy privilege using Late Insertion of vocabulary (Halle and Marantz 1993). This dissertation starts from the question whether the generalization that the highest copy is privileged in pronunciation does truly exist across languages and across 2 constructions. Observing cases where a low copy is pronounced even when there is no phonological constraint that restricts the high copy to be pronounced, I argue in this dissertation that there is no universal high-copy privilege. In actuality, it seems that a movement chain by itself does not impose a privilege on any copy with respect to pronunciation. Following Chomsky (2000, 2001)’s Agree-based approach, I assume every member in a chain shares the same properties (contrary to Nunes (1995, 2004, 2011) and Potsdam and Polinsky (2012)). In other words, every member in a chain is identical in its semantic, phonetic and syntactic properties except for the positions each one sits in a structure. It implies that there is no chain- internal reason to select one copy over other for pronunciation. Rather, what determines which copy is to be pronounced is independent of PF requirements like phonological constraints or prosodic requirements. The main interest of this dissertation is in the question why in some cases an element is pronounced in the position where it is interpreted while in other cases, there is a discrepancy between the position for interpretation and the position for pronunciation. Under the copy theory of movement, we can roughly imagine two possibilities for an element to be pronounced in the position where it is interpreted: (i) the element does not move at all, so the only copy in the base position is pronounced and gets interpreted or (ii) the element moves to a position higher than its base position and the copy in the moved position is pronounced and interpreted. For the case of an interpretation-pronunciation discrepancy, it is expected that an element undergoes movement but a low copy is pronounced while a high copy is interpreted 1 . This dissertation investigates how the three possible imaginable derivations are realized in languages. Throughout this dissertation, I will cross-linguistically investigate various movement chains that are instances of each of three cases. The following five main sub-issues will be discussed throughout dissertation: • Sub-issue 1: Relation between agreement and movement and various types of movement Given that the issue of “movement” is closely related with the issue of “agreement” in a sense that agreement has been understood as giving a motivation for syntactic movement, it seems 1 There is another possible interpretation-pronunciation discrepancy case, namely reconstruction cases (a high copy is pronounced, but a low copy is interpreted). In this dissertation, I will not deal with such cases where an element is interpreted in a position lower than a position where it is pronounced. See Bobaljik (2002) for relevant discussions. 3 crucial to clarify the relation between agreement and movement. I will suggest an answer to the question why movement takes place with respect to agreement. Based on the relation between agreement and movement I will suggest, I will propose three different types of movement: phrasal movement, parasitic phrasal movement, and parasitic head movement. The crucial difference between phrasal movement and parasitic (phrasal/head) movement is that an element does undergo movement in the case of phrasal movement, while an element does NOT undergo movement in the case of parasitic movement. I will justify the existence of three types of movements with empirical data. • Sub-issue 2: Selection of copies for pronunciation As will be proposed, an element that undergoes phrasal movement leaves its copies in both its base position and its destination, unlike one that undergoes parasitic movement. Thus, under the copy theory of movement, PF needs to choose which copy to pronounce. Zooming in the issue of selection of copies for pronunciation, I will discuss PF constraints that play a role in copy- selection. Especially, adopting Landau (2007)’s intuition that the EPP is a pure PF requirement, I will argue that the high-copy privilege assumed in the previous studies are misled by a pronunciation wise reinterpreted EPP. Furthermore, I will argue that once we get rid of the effects of the EPP, a low copy is rather preferred to be selected for pronunciation due to economy conditions. I will show how the interaction between the EPP as a PF constraint and an economy condition favoring low-copy pronunciation accounts for both (i) prevalent high-copy pronunciation and (ii) apparent lack of high-copy privilege. • Sub-issue 3: Typological studies Based on the system proposed in the dissertation, I will provide a typological study in two representative cases of movement: (i) subject agreement/movement and (ii) wh- agreement/movement. This system suggests a new approach for typology of in-situ subjects and in-situ wh-phrases. I will show that the proposed system captures well the observation that it is not the case that every in-situ subject/wh-phrase patterns together. To be more specific, in the proposed system, in-situ subject/wh-phrases are the results of either parasitic movement or low- copy pronunciation in phrasal movement. When a parasitic movement takes place, a subject/wh- phrase does not undergo movement, so the subject/wh-phrase has only one copy in its base 4 position, which is phonetically realized. In that case, the in-situ subject/wh-phrase does not show any movement-sensitive patterns (e.g. island-sensitive, lack of intervention effects). By way of contrast, when a subject/wh-phrase undergoes phrasal movement and a low copy of the chain is pronounced, the in-situ subject/wh-phrase shows movement-sensitive patterns. I will show how the two theoretically possible types of in-situ subjects/wh-phrases are realized in languages. • Sub-issue 4: Various ways of satisfying the pronunciation-wise EPP requirement Based on the definition of the EPP I will provide in chapter 2, I will propose various ways to satisfy the EPP. Instead of filling the spec position of a functional head, I will argue that agreement morphology can be inserted as a means of satisfying the EPP. In addition, I assume that morphology insertion as a way of satisfying the EPP includes the case of inserting an intonation morpheme, suggested by Cheng and Rooryck (2000). This additional way to satisfy the EPP accounts for a range of otherwise puzzling prosody relevant phenomena: i) acceptance of inverse scope reading in so-called rigid scope languages with the help of prosody; ii) ameliorating the intervention effects in Korean and Japanese with the help of prosody. • Sub-issue 5: Accommodating exceptional low-copy pronunciation In chapter 1, by reviewing previous studies, I will introduce many cases where a low copy is exceptionally selected for pronunciation because high-copy pronunciation violates a (morpho-)phonological constraint. In chapter 5, I will return to this issue and account for how this exceptional low-copy pronunciation is accommodated in the proposed system. In the current chapter, I will introduce relevant background for the dissertation. In Section 1.2., I will discuss the copy theory of movement in detail. I will introduce Chomsky (1993)’s original proposal of the copy theory of movement and move to the issue of, how to decide which copy is to be pronounced, which this dissertation will focus on. Among many others, I will discuss how Nunes (1995, 2004, 2011), Bobaljik (2002) and Bošković and Nunes (2007), Pérez (2015) account for this issue. In section 1.3., I will introduce new observations that raise questions about previous accounts and explain why an alternative approach is called for. In section 1.4., I will overview the organization of the dissertation. 5 1.2. The copy theory of movement and low-copy pronunciation In the Government and Binding (GB) era, movement is understood as if a phrase is displaced from one position to another leaving a trace behind. For example, in (1) the wh-phrase which picture of himself is base-generated within VP as the object of the sentence and undergoes movement to higher functional domain for wh-feature checking. However, coreference between the subject John and the anaphor inside the moved wh-phrase indicates that the anaphor must be in the domain c-commanded by the subject in compliance with Binding Principle A. To resolve the discrepancy between the phonetically spelled-out position and the position where it gets interpreted, two possibilities were suggested: either binding takes place at Deep Structure (before movement) or the phrase including the anaphor is reconstructed to its base position at LF. (1) [Which picture of himselfj]i did Johnj sell ti ? However, when the Minimalist Program (Chomsky 1995) was introduced, the first solution is not tenable any longer because the Minimalist Program eliminates the non-interface level, DS, leaving only the interface components, LF and PF. Furthermore, the second solution is also unwelcome because the lowering operation violates the Proper Binding Condition (Fiengo 1974). Chomsky (1993) proposes the copy theory of movement to resolve the “Reconstruction effects of Ā-movement” by replacing traces with a copy of the displaced phrase. Under the copy theory of movement, the sentence in (1) has the LF representation in (2). For the wh-operator, the high copy survives deleting the low copy, but for its restriction, the low copy remains deleting the high copy. Since the low copy of the anaphor is bound by the antecedent John at LF, the binding fact can be explained without appealing to DS or Reconstruction. (2) [Which picture of himselfj] did Johnj sell [Which picture of himselfj] This is a piece of empirical evidence supporting the theoretical change from trace-theory to the copy theory of movement. However, this conceptual change in movement actually brings more important theoretical advantages to the Minimalist Program. First, eliminating “traces”, which are not primitive elements copied from the Numeration, but created in syntax, the theory satisfies 6 the “Inclusiveness” condition. Second, even though Chomsky assumes that only LF can choose which copy to survive while PF always privilege a high copy, the theory in principle allows PF to delete either a high copy or a low copy, as LF does. This point is clearly discussed in Bobaljik (2002), which will be discussed in detail in the next section. 1.2.1. Bobaljik (2002) Since both LF and PF can, in principle, reserve the right to choose which copy to delete/survive, a four-way typology of realizations of movement is predicted, as Bobaljik illustrated in (3). (3) Privileged Copies (Bobaljik 2002, (3)) a. Overt movement PF: copy1……copy 2 LF: copy 1……copy 2 b. LF movement PF: copy1……copy 2 LF: copy 1……copy 2 c. Reconstruction PF: copy1……copy 2 LF: copy 1……copy 2 d. LF movement/reconstruction PF: copy1……copy 2 LF: copy 1……copy2 Parameterizing the “deleting” operation at LF and PF can generate four different dependency phenomena that were previously thought as created by (a combination of) independent operations. The most important consequence is that, under this approach, no displacement is assumed to take place at LF. LF movement and Reconstruction, previously thought as LF operations, are actually nothing but a case where a movement takes place in overt syntax but LF and PF choose different copies. This approach seems theoretically desirable in the sense that it imposes only a selection operation to interface components assuming every displacement operation (or computation) takes place in syntax. Bobaljik argues that this theory is empirically superior to one assuming LF movement as well. Bobaljik’s argument is based on Holmberg’s Generalization. In Scandinavian languages (except some dialects), a “weak” pronominal object must undergo “object shift” as exemplified in (4). 7 (4) a. Igår läste de deni [VP inte ti] Yesterday read they it not They didn’t read it yesterday. (Swedish, Bobaljik 2002, (7)) b. Haan las {Ϸær} ekki {*Ϸær} he read them not them He didn’t read them. (Icelandic, Diesing 1996, 67, cited in Bobaljik 2002, (7)) However, when the main verb remains in VP, the otherwise obligatory object-shift is prohibited, which is well-known as Holmberg’s Generalization. There are two environments where Holmberg’s Generalization is observed: In Swedish, the verb in the embedded complements of non-bridge verbs remains in VP and in this case, object shift is prohibited, as in (5b). (5) a. Det är troligt [att de [VP läste den]] It is probable that they read it b. *Det är troligt [att de deni [VP läste ti]] It is probable that they it read It is probable that they read it. (Swedish, Bobaljik 2002, (11)) Another environment where object shift is blocked is when compound tenses or modals appear. In such structures, only the highest auxiliary verbal element raises to satisfy the V2 condition leaving the main verb inside the VP. In this case as well, the object must follow the main verb (6a) and object shift is blocked (6b). (6) a. Johann har [VP sett den]. Johann has seen it b. *Johann har deni [VP sett ti]. Johann has it seen Johann has seen it. (Swedish, Bobaljik 2002, (12)) What is common in the two object-shift-blocking environments is that, according to Bobaljik, the object intervenes between the inflectional head (Infl) and the verb, so the adjacency between Infl 8 and the verb is disrupted. The fact that Holmberg’s Generalization is not observed in the head final (OV) Germanic languages confirms that the impossibility of object shift is closely related with Infl-verb adjacency. Bobaljik argues that in Scandinavian languages, the verb within the VP must undergo a morphological merger operation with the Infl to realize the inflectional morphology on the verb stem. Crucially, this merger operation is a post-syntactic operation occurring at the PF-interface. In other words, the adjacency requirement between the verb and the Infl needs to be satisfied in PF, rather than in syntax. Bobaljik contends that the copy theory of movement can naturally explain the interaction between the lack of the object shift and the verb-Infl adjacency requirement at PF: The object shift always takes place in syntax, but PF decides which copy to pronounce according to PF constraints. Thus, the difference between (5a) and (5b) (or (6a) and (6b) as well) is not a syntactic one, but just a difference in PF- representation, as in (7). (7) a. Det är troligt [att [ IP de -te deni [VP läste deni]]] It is probable that they +past it read it [Merger successful o------------------o] b. * Det är troligt [att [ IP de -te deni [VP läste deni]]] It is probable that they +past it read it [Adjacency disrupted: o------*-----------o] It is probable that they read it. (Swedish, Bobaljik 2002, (36)) Since the merger operation is a morphological requirement, it seems reasonable that the operation engenders a PF response in the way of pronouncing the lower copy. As Bobaljik argues, if the VO order does not represent the case where the lower copy is pronounced, but the case where the object shift has not occurred in syntax, the theory must assume that there is a kind of PF filter that eliminates the illegitimate object shift case or that syntax knows the morphological constraint in a look-ahead fashion. Furthermore, the alternative has a burden to assume that the agreement that triggers object shift can be satisfied either in syntax or LF (by LF movement of the object). The copy theory of movement analysis does neither need to assume such a PF filter nor confront the look-ahead problem. Therefore, theoretically and empirically, the copy theory of movement seems to be a more desirable analysis. 9 However, one might wonder why the high copy is preferred to be pronounced when there is no violation of the verb-Infl adjacency requirement in the case of object shift. To put it in different words, why is the higher copy privileged by default? Bobaljik argues that this high- copy privilege in Scandinavian object shift comes from the economy condition, Minimize Mismatch: (8) Minimize Mismatch (Bobaljik 2002, (70)) (To the extent possible) privilege the same copy at PF and LF. The Minimize Mismatch is further supported by the observation that object shift is correlated with the information structure of the object. When the object represents old information, it is interpreted outside of VP (according do Mapping Hypothesis proposed by Diesing (1992)). Thus, the Minimize Mismatch forces the high copy to be pronounced unless there are any other phonological constraints that blocks it. In addition, the Minimize Mismatch correctly predicts that when the object represents new information, it never undergoes object shift. This indicates that the low copy is privileged at PF as parallel to its LF counterpart, which is interpreted inside of VP. This proposal seems interesting, but has both theoretical and empirical problems. Theoretically, it rests on the unusual assumption that two interface levels, LF and PF, communicate each other. It is not expected in the model where the output of syntax is transferred to the distinct interface components, LF and PF, each. Not only that, this approach assumes (though not explicitly) that LF determines which copy to interpret first and then based on that information, PF determines which copy to pronounce. This point is clearly implied in Bobaljik’s account for English expletive constructions. As represented in (9), the indefinite phrase in the raising construction can have either a wide scope reading or a narrow scope reading. Bobaljik argues that when it has wide scope, the high copy must be pronounced because the high copy pronunciation can satisfy both the EPP requirement and the Minimize Mismatch. If the low copy is pronounced at PF when the higher copy is interpreted at LF, it violates not only the EPP requirement but also the Minimize Mismatch. Thus, the low copy is not selected in PF due to the more economical alternative (i.e. the high copy) that does not violate any constraint. On the other hand, when the indefinite has narrow scope, according to Bobaljik, either copy can be 10 pronounced at PF. If the high copy is pronounced, it satisfies the EPP requirement, but violates the economy condition, Minimize Mismatch. By way of contrast, if the low copy is pronounced, it satisfies Minimize Mismatch, but requires the last resort “there” insertion for the EPP requirement. Since neither option in PF is more economical, either option can be selected. This is the reason Bobaljik provides for why (10a) has scope ambiguity while (10b) has only the indefinite-narrow scope reading. (9) [Someone from New York] seems to be [someone from New York] at the party. (10) a. [Someone from New York] seems to be at the party. a. There seems to be [someone from New York] at the party. Such OT style comparison between PF outputs based on information from LF seems to require another meta-computational level that collects information from the interface levels and makes the final decision. It does not seem clear to me how the current computational model can incorporate such a hidden computational level. In addition, this account incorrectly predicts that expletive insertion in transitives in English is possible, as in (11). (11) *There John bought a book. In English transitives, it has been assumed that the subject is base-generated in SpecvP and moves to the position of TP. Under the copy theory of movement, the copies of the subject are assumed to be in SpecvP and SpecTP. If we assume that LF chooses the low copy to be interpreted because SpecvP position is where the thematic role of the subject is assigned, as we saw in the raising construction in (9)-(10), there is no reason for PF to not choose the low copy to pronounce with “there” insertion. The high copy pronunciation violates Minimize Mismatch, but satisfies the EPP requirement. On the other hand, the low-copy pronunciation satisfies Minimize Mismatch, but requires costly the expletive insertion. Neither option is perfectly economical, so it is expected that there is optionality, contrary to facts. It does not seem clear how Bobaljik’s proposal can exclude transitive expletives in English. Therefore, although the insight that there is a correlation between the information structure of the object and the choice of copy at PF in object shift seems to be important and 11 needs to be carefully considered in any theory of movement, I suspect that Minimize Mismatch creates problems not only theory internally but also empirically 2 . So, we now turn back to our original question. Why does PF choose the high copy in object shift unless there is a (morpho-)phonological constraint that blocks it? More generally, why does it seem that high-copy pronunciation (i.e. overt movement) is more common than low- copy pronunciation (i.e. covert movement) in languages? This question is discussed in Groat and O'Neil (1996), Nunes (1995, 2004, 2011), and Pérez (2015) in depth. I will discuss these studies in the subsequent sections. 1.2.2. Groat and O'Neil (1996) Adopting the copy theory of movement, O'Neil and Groat (1996) argue that the traditionally understood overt/covert movement distinction is not equivalent to pre-vs. post spelled-out movement. As Bobaljik (2002) argues, they argue that what is traditionally called “overt” movement is the case where a high copy is pronounced while what is traditionally called “covert” movement is the case where a low copy is pronounced. They propose that only when the features to be checked are strong, the phonological feature is copied into the high member having the high copy to be phonetically realized. However, as pointed out by Nunes (1995), this account incorrectly predicts that more than one member of a chain can be pronounced if the chain has two strong features as in (12). Since who has two strong features, each of which is checked by independent functional category, one against the embedded T and the other against the matrix C. If the realization of phonological features is parasitic on strong feature checking, who is incorrectly expected to be pronounced in two positions, as in (12a). (12) [CP Who[WH] do you think [ TP who[Case] [vP who invite John?]] 2 Given the Mapping Hypothesis or similar constraints proposed in the literature, a constraint similar to Minimize Mismatch in some sense might be required in languages (see 4.2). However, if there is another option that can substitute for the LF-PF matching requirement, it would be better to take such an option due to the challenges against Minimize Mismatch, as just discussed. I will discuss the correlation between information structure and the choice of copy at PF in object shift in 4.2. In addition, scope interpretation in English raising constructions will be discussed under the system I will propose in chapter 2. 12 a. *Who do you think who invite John? b. Who do you think invite John? One might suggest that phonological features can move from the first strong feature checking site to another one. Thus, in (12), the phonological features of who move first to the embedded SpecTP and then moves to the matrix SpecCP resulting in (12b). However, Nunes points out that this causes another problem. One problem is found in wanna-constructions. The wanna- contraction has been traditionally assumed to take place in the phonological component. The derivation of (13b) is represented in (14). If the phonological feature of who moves to SpecTP and then further moves to the matrix SpecCP for another instance of strong feature checking, the embedded SpecTP is expected to have no phonological element, allowing wanna-contraction, contrary to facts. Note, since who in (13a) has not stopped by the embedded SpecTP, the wanna- contraction in (13a) is not blocked. Thus, correlating strong feature checking with the position where the element is phonetically realized seems to be incorrect. (13) a. Who do you want to/ wanna dance with? b. Who do you want to/*wanna dance with Mary? (Nunes, 1995, (71)) (14) who, did you want [ TP who, to [vp who dance with Mary ] ] Groat and O'Neil’s account is basically in line with Chomsky (1993, 1994, 1995) in that strong features trigger overt movement. Even though Groat and O'Neil depart from Chomsky by assuming that covert movement is just a case of low-copy pronunciation of the movement chain, their intuition that the distinction between overt and covert movement is rooted in the strength of features, as assumed by Chomsky. However, such a division of strength among features is arbitrary and without any independent motivation. If an alternative approach could account for copy selection for pronunciation without appealing to feature strength, it would be theoretically more attractive. In 2.2, I will propose an alternative account for the motivation of movements. I will put this issue aside for now and continue to discuss how previous studies have dealt with the (seemingly) high-copy privilege in pronunciation. 13 1.2.3. Nunes (1995, 2004, 2011): Chain Reduction and Linearization To understand Nunes’ proposal, it is necessary to discuss the feature checking system Nunes assumes. The theory of “Feature checking” has played a major role in the Minimalist Program. There are many issues entangled with it: what kind of features human languages have, where those features are located, what the role of the features have, and many others. In section 1.2.3.1, I will introduce Chomsky’s (1993, 1995) feature checking mechanism especially focusing on chain formation and Nunes’ critique of it. Nunes proposes a Chain Uniformity condition which systematically distinguishes the property of a high copy and that of a low copy in a chain. Based on Nunes’ notion of chain formation, I will discuss a Chain Reduction algorithm in section 1.2.3.2 1.2.3.1. Chain formation and feature checking In the early Minimalist Program, Chomsky (1993, 1994, 1995) provides an answer to the question why movement takes place based on the principle of full interpretation (FI): The principle FI is assumed as a matter of course in phonology; if a symbol in a representation has no sensorimotor interpretations, the representation does not qualify as PF representation…The same condition applied to LF also entails that every element of the representation has a (language independent) interpretation. (Chomsky 1995. P27) To avoid the violation of FI, any element that is not interpretable at the interface levels must be eliminated before it is transferred to the interface levels. Let me discuss this in detail. Chomsky classifies features into three types as in (15). (15) a. Phonological features: [+high], [ −back] … b. Semantic features: [ −mass], [+human] … c. Syntactic features: Case, Number, Person… 14 Phonological features and semantic features are interpretable at the interface levels whereas syntactic features are distinguished into uninterpretable features and interpretable features. For example, number features in noun phrases are interpretable, but those in T are uninterpretable. Chomsky further distinguishes among uninterpretable features by strength. Chomsky (1993) contends that strong features are visible at interfaces but weak features are not. Therefore, unlike weak features which can remain even after Spell-out, strong features must be eliminated before Spell-out to keep the derivation from crashing 3 . The way to eliminate uninterpretable features, according to Chomsky, is movement. For example, in (16a), the uninterpretable Case feature of the unaccusative subject needs to be eliminated (along with the ϕ-features on T). For uninterpretable strong features to be eliminated, the subject DP needs to check its features against T in a local (Spec-head) domain. Thus, the movement of the subject DP takes place (16b). Crucially, under the copy theory of movement, the copy left in the base position of the subject DP should have an uninterpretable Case feature as well since the high member of the chain and the low member of the chain are nothing but clones of the identical element. However, since only the high copy is in a Spec-head relation with the functional category for feature agreement, the uninterpretable feature of the low copy remains unchecked. This remaining strong feature will cause the derivation to crash, so it must be checked as well. To resolve this problem, Chomsky (1995) contends that when the head member of a chain undergoes feature checking with a functional head, the uninterpretable features of the lower members of a chain are also checked and eliminated. In our example, as a consequence of Case checking in (16b), the Case feature of the low copy inside vP is also checked and erased as in (16c). Thus, the uninterpretable features of the low member of the chain are eliminated rendering the member inaccessible to further computation. (16) a. [ TP John[Case] [vP arrived John[Case]] b. [ TP John[Case] [vP arrived John[Case]] 3 Chomsky (1995) slightly changes his view on the properties of strong features and suggests that a derivation just cannot tolerate strong features. This descriptive property of strong features does not refer to interface levels. Rather, when a strong feature is introduced into a derivation, it just needs to be eliminated as soon as possible. Chomsky discusses the conceptual reason for such a change (Chomsky 1995, p233), but there is no difference in the effect of the influence of strong features on movement. 15 c. [ TP John[Case] [vP arrived John[Case]] Nunes points out that eliminating the formal features of the lower members in a chain causes a problem. It is easily observed in many languages that a low copy is used for the computation of scope or binding. Assuming that formal features are needed for such computations, Nunes argues that eliminating and being inert of formal features of lower copies for further computation incorrectly rules out the reconstruction cases. As an alternative, Nunes proposes Chain uniformization (17) without the assumption that a low member of a chain is affected by the checking relation between a high member of the chain and a functional category. This proposal is based on the assumption that copies in a chain have basically the same property. In other words, a low member of a chain does not have a special property that it can check and eliminate its uninterpretable formal features even without being in a proper checking domain. Rather, even though a high copy can delete its uninterpretable formal features, formal features of low members remain unchecked. This asymmetry between the number of unchecked uninterpretable features violates the Feature Uniformity condition (18), so an independent operation, Chain Uniformization (17), is needed to delete the uninterpretable features of lower copies. (17) Chain Uniformization (Nunes 1995, (36)) Delete the minimal number of features of a nontrivial chain CH in order to allow its links to satisfy the Feature Uniformity Condition. (18) Feature Uniformity Condition: Given a chain CH = ( aj,..., an), every ai (1 ≥ i ≥ n) must have the same set of features visible at LF. Let’s discuss how this system works with our previous example in (16), which I repeat in (19). Since this system assumes that a low copy is not affected by the feature checking of a high copy, the two copies differ with respect to the number of undeleted uninterpretable formal features: the high copy does not have any uninterpretable feature because its uninterpretable feature, in this case, case feature, is eliminated after the checking operation it participates in. On 16 the other hand, the low copy has an uninterpretable case feature because it has not participated in any checking operation. This discrepancy violates the Feature Uniformity Condition in (18). To save this derivation, Nunes argues that Chain Uniformization (17) deletes the uninterpretable case feature of the low copy. This operation allows the derivation to satisfy not only FI (since it makes the uninterpretable case feature invisible at interface levels) but also the feature uniformity condition. Furthermore, since this operation does not eliminate the formal feature by the operation “erasing”, the feature is accessible to further computation. (19) a. [ TP John[Case] [vP arrived John[Case]] Nunes highlighted that such a change in the way of deleting uninterpretable formal features of a low copy naturally derives an asymmetry between a high copy and a low copy in a chain and this asymmetry is the key to explain high copy privilege in pronunciation. I will discuss Nunes’ Chain Reduction algorithm in the next section. 1.2.3.2. Chain Reduction and high copy privilege Nunes argues that the operation copy targets a given term M and assigns an index i to M, if M has no index, and creates a copy of the indexed term M i . The indexation device distinguishes homophonous relation (20a) from copies that are nondistinct terms (20b). (20) a John loves John (John loves the person whose name is also John). b. John i is loved John i . Elements in syntactic structures must be linearized for pronunciation. According to the Linear Correspondence Axiom (LCA), proposed by Kayne (1994), asymmetric c-command relations determine how a syntactic structure can be transformed into a linear string. For example, the sentence in (20b) has two copies of John and each copy sits in different positions in the structure as shown in (21). 17 (21) TP John i T ′ was VP loved John i The high copy of John i sits in SpecTP and it asymmetrically c-commands the copular was and the main verb loved. Based on the asymmetric c-command relation, the three elements must be linearized as <John i , was, loved> according to LCA. However, this word order contradicts the string <was, loved, John i >, which is generated by the asymmetric c-command relation with the copular/main verb and the low copy of John. The low copy of John is asymmetrically c- commanded by was and loved, so John i is expected to follow was and loved in the linear order. Since both copies of John are nondistinct, as indicated by its index, such a contradiction causes a problem in linearization. Furthermore, since the high copy of John asymmetrically c-commands the low copy of John, the higher one is expected to precede the lower one in linear word order. However, since the two copies are nondistinct, it violates reflexivity condition on linear order. Thus, to avoid such contradiction, Nunes argues that a chain is required to delete all copies except one for linearization purposes. In addition, Nunes raises a question why only the head of a chain is pronounced leaving the tail of the chain unpronounced (in general). To answer this question, Nunes proposes an economy condition on Chain Reduction: (22) Chain Reduction: (Nunes 1995, (93)) Delete the minimal number of terms of a nontrivial chain CH in order for the largest structure containing CH to yield a linear order in accordance with the LCA. As discussed in the previous section, Nunes argues that a low copy is not affected by the feature checking of a high copy against a functional category. Thus, in (21), the high copy of John deletes its case feature, so is invisible to the interface level, while the low copy of John still has an unchecked uninterpretable case feature which must be deleted by Chain Uniformization for FI and Feature Uniformity condition. However, once Chain Reduction deletes the low copy of John, 18 there is no need to apply the extra operation (Chain Uniformization) to delete the case feature of the low copy because the case feature of the low copy is already deleted along with the term. By way of contrast, if the high copy of John is deleted by Chain Reduction, Chain Uniformization needs to apply additionally to delete the case feature of the low copy independent to Chain Reduction. Thus, deleting the low copy is more economical in terms of the number of operations that are required for the derivation to be convergent. The combination of LCA and economy conditions on Chain Reduction derives high copy privilege in pronunciation, according to Nunes. High copy privilege seems to be correct as a generalization if we compare the number of overt movement with the number of covert movements observed in languages in general. However, in chapter 2, I will argue against the generalization and suggest that high copy privilege is just an illusion caused by the EPP requirements. Postponing the issue to chapter 2, in the next section, I will discuss how covert movement has been dealt with in Bošković and Nunes (2007) based on Nunes (1995). 1.2.4. Bošković (2002) and Bošković and Nunes (2007) Bošković (2002) and Bošković and Nunes (B&N, 2007) thoroughly investigate traditionally understood covert movement in terms of low-copy pronunciation under the copy theory of movement. Assuming the Chain Reduction algorithm proposed in Nunes (1995), they argue that a high copy is privileged in pronunciation. However, if there is any (morpho-)phonological constraint that blocks high-copy pronunciation, a low copy is instead chosen for pronunciation. In this section, I will review some of examples Bošković (2002) and Bošković and Nunes (2007) provide. Multiple wh-fronting. Romanian is one of the multiple wh-fronting languages. Thus, as in (23a), every wh-phrases must move to the front of the sentence. If one of the wh-phrases stays in its base position as in (23b), the sentence becomes ungrammatical. However, if the fronted wh- phrases are homophonous, the second wh-phrase cannot be fronted as in (24). (23) a. Cine ce precede? (Romanian, B&N 2007, (5)) who what precedes 19 b. *Cine precede ce? who precedes what Who precedes what? (24) a. Ce precede ce? (Romanian, B&N 2007, (6)) what precedes what b. *Ce ce precede? what what precedes What precedes what? Adopting the copy theory of movement, Bošković (2002) argues that even in (24a), the object wh-phrase ce ‘what’ moves next to the first wh-phrase in syntax as in normal multiple wh- fronting constructions as represented in (25a). However, due to the PF constraint that blocks consecutive homophonous wh-phrases, PF does not choose the high copy, but the low copy of the object wh-phrase to be phonetically realized (25b) 4,5 . 4 Bošković provides another example of a low-copy pronunciation in Romanian echo-questions. He argues that when an echo wh-phrase violates a prosodic constraint, the otherwise obligatory high-copy pronunciation of the echo wh-phrase is blocked and the low copy of the wh-phrase is phonetically realized. For more details, see Bošković (2002). 5 Bošković shows that the wh-in-situ in Romanian patterns different to wh-in-situ in English multiple wh- constructions with respect to a parasitic gap licensing. English is a single wh-fronting language, so the second wh- phrase in English multiple wh-phrase does not undergo a movement as a phrase, being unable to license a parasitic gap in (i). On the other hand, the in-situ wh-phrase in Romanian can license a parasitic gap as in (ii) indicating that the wh-phrase undergoes a phrasal movement in syntax even though it is pronounced in its base position. (i) *Who read WHAT without filing? (Bošković 2002, (62)) (ii) Ce precede ce fără să influenţeze? (Romanian, Bošković 2002, (60)) what precedes what without subj.particle influence.3p.sg What precedes what without influencing. However, Pesetsky (2000) argues that a subset of in-situ wh-phrases in English undergoes covert phrasal movement, as will be discussed in 1.3. According to Pesetsky, the in-situ wh-phrase in (i) licenses Antecedent-Contained deletion, so it should be a case where a covert wh-phrase movement takes place. I will follow Pesetsky’s proposal and leave the inability of parasitic gap licensing of in-situ wh-phrases as a problem that needs to be resolved in future works. 20 (25) a. Syntax: Cei cej cei precede cej? b. PF: Cei cej cei precede cej? The advantage of the copy theory of movement account for the in-situ wh-phrases in multiple wh-fronting languages is that there is no need to postulate exceptional wh-movement. Under this account, every wh-phrase moves to the front to satisfy syntactic requirements in multiple wh-fronting languages regardless of phonological constraints. The difference between (23) and (24) is not in syntax, but in PF. Furthermore, this account does not confront the look- ahead problem that syntax responds to PF requirements. One might wonder why PF forces the second wh-phrase to be pronounced with the lower copy, not the first wh-phrase. To answer this question, B&N (2007) state that Chain Reduction proceeds in a top-down fashion. In (23a), Chain Reduction deletes the low copy of the first chain (the subject wh-phrase) and Chain Reduction moves to the second chain (of the object wh- phrase) and also deletes also low copy because it does not trigger a violation of any PF constraints. Similarly, in (24a) as well, Chain Reduction deletes the low copy of the first chain. However, when Chain Reduction moves to the second chain, it detects that the high copy pronunciation in the second wh-chain violates a PF constraint. Thus, Chain Reduction deletes the high copy instead even though it violates economy conditions on chain reduction in order to satisfy the PF constraint for convergence. This analysis assumes that economy conditions compare only convergent derivations. Therefore, the non-convergent derivation in (24b) cannot be chosen as an optimal solution. Clitics. In Romanian, object clitics must be realized in phonetically reduced form when they precede an auxiliary clitic starting with a vowel, as in the comparison in (26). In addition, when an object clitic appears with an auxiliary clitic, clitic split is forbidden. If the object clitic appears in its base position separated by the auxiliary clitic as in (27), the sentence becomes ungrammatical. (26) a. *Îmi au ales articolul. (Romanian, B&N 2007, (24)) me.Dat have chosen article-the 21 b. Mi- au ales articolul. me.Dat have chosen article-the They have chosen my article. (27) *au ales articolul Îmi. have chosen article-the me.Dat Interestingly, there is one exception to this obligatorily adjacent clitic formation. When the object clitic is the 3person, singular, feminine clitic o, it cannot precede the auxiliary clitic (28a), but must appear as an enclitic to the verb, separated by the auxiliary clitic, as in (28b). (28) a. *O am vazut. (Romanian, B&N 2007, (25)-(26)) her have seen I have seen her. b. Am vazut-o. have seen-her Following Bošković (2001), B&N argue that the object clitic o undergoes clitic climbing in (28b) as other clitics do. Thus, in syntax, a high copy of o precedes the auxiliary clitic and a low copy of o follows the verb. However, the clitic o cannot be phonetically reduced, so the high copy of o cannot be pronounced because it sits in the position where only phonetically reduced clitics are allowed (see (26)). As the alternative, the low copy is phonetically realized, as in (28b) 6 . 6 In Romanian, the cluster of the object clitic and the auxiliary clitic can appear in a post-verb position. (i) Mânca l- ar mama! (Popescu 2000, (17c)) Eat.INF 3SG.M.ACC 3SG.COND mama He is sweet! (Lit. : Mama would eat him) However, such enclitics requires special mood of verbs, for example, swearing, affection, or imperatives. Thus, in such cases, it would be more plausible to assume verb raising over the clitics, so it does not seem to be proper to analyze such cases as low-copy pronunciation cases. 22 Another example of low-copy pronunciation in clitic movement is observed in Serbo- Croatian (SC). In SC, auxiliary clitics must precede every pronominal clitic in a clitic cluster, as in (29). Interestingly, there is one exception to this rule. The auxiliary clitic je must follow every pronominal clitic as shown in the contrast in (30). (29) a. Oni su mu ga predstavili. (Serbo-Croatian B&N 2007, (38)) they are him.Dat him.Acc introduced b. *Oni mu ga su predstavili. they him.Dat him.Acc are introduced They introduced him to him. (30) a. *Ona je mu ga predstavila. (Serbo-Croatian B&N 2007, (39)) she is him.Dat him.Acc introduced b. Ona mu ga je predstavila. she him.Dat him.Acc is introduced She introduced him to him. The clitic je in (30b) shows the ‘high’ behavior in some aspects: 1. je can survive from VP ellipsis, unlike pronominal clitics (31), indicating that je is structurally higher than pronominal clitics 7 . 2. je is compatible with not only a manner reading but also a sentential reading of the adverb (32a). This implies that the clitic je sits in a structurally higher position than the position where sentential adverbs sit. By way of contrast, pronominal clitics are compatible only with a manner reading of the adverb, as in (32b) implying that pronominal clitics must be structurally lower than sentential adverbs. This also leads to the same conclusion that the auxiliary clitic je is 7 Even though B&N do not explicitly show the derivation of the sentence in (31), je in the ellipsis clause is arguably pronounced with the high copy because the low copy is included in the ellipsis site. Since other clitics in the clitic cluster have been removed along with VP ellipsis, there is no PF constraint that blocks the high-copy pronunciation of je. However, one remaining question on this issue concerns the lack of parallelism between the antecedent clause with low-copy pronunciation of je and ellipsis clause with high-copy pronunciation. One possibility is that the parallelism does not play any role in PF. The antecedent clause and the ellipsis clause cannot be identical in their PF form due to the gap in the ellipsis clause. 23 structurally higher than the pronominal clitics even though je follows the pronominal clitics in the linear sequence 8 . (31) Ona mu ga je predstavila, a i on je mu ga predstavio. She him.Dat him.Acc is introduced and also he is him.Dat him.Acc introduced She introduced him to him and he did too. (Serbo-Croatian B&N 2007, (43)) (32) a. Jovan je pravilno odgovorio Mileni. (Serbo-Croatian B&N 2007, (44)) Jovan is correctly answered Milena.Dat Jovan did the right thing in answering Milena. Jovan gave Milena a correct answer. b. On joj je pravilno odgovorio. he her.Dat is correctly answered *He did the right thing in answering her. He gave her a correct answer. B&N argue that the auxiliary clitic je undergoes movement to a position higher than pronominal clitics as other auxiliary clitics do, but it is pronounced with a low copy due to a PF constraint that requires je must follow all the other clitics in a clitic cluster. Those examples discussed in this section share a common phenomenon: Obligatory overt movement (in a traditional sense) becomes covert movement if there is any (morpho-) phonological constraint that blocks the overt movement. This syntax-phonology interaction 8 The logic of (32b) is not clearly explained in B&N (2007). However, I suspect that the following reasoning is assumed. In (32a), there is only auxiliary clitic je without any pronominal clitics, so its phonetic realization is free from a PF constraint. Thus, it can be pronounced with either a high copy or a low copy. When the sentence in (32a) is the case of pronouncing the high copy of je, the adverb has a sentential reading, while when the sentence in (32a) is the case of pronouncing the low copy of je, the adverb has a manner reading. However, such optionality is impossible when a pronominal clitic appears. If B&N’s claim is correct, the pronominal clitic does not move and it stays inside VP. Due to a PF constraint, the auxiliary clitic je must be pronounced with a low copy which follows the pronominal clitic in linear order, as in (32b). Since the pronominal clitic and auxiliary clitic are all within VP, the adverb that follows them cannot have a sentential reading, but a manner reading only. 24 seems to be well captured by Nunes’ interpretation of the copy theory of movement: there is a high copy privilege in pronunciation due to an economy condition on feature deletion. However, if the high copy pronunciation is not convergent violating any PF requirement, a low copy is instead chosen to be phonetically realized. However, this analysis is crucially based on the assumption that feature checking takes place in a Spec-head relation. The reason why a high copy has a lesser number of unchecked features is that the high copy is in a Spec-head relation with a functional head where it can check its uninterpretable features whereas lower copies are not in a local relation with the functional head being unable to check its uninterpretable features. However, since Chomsky (2000, 2001) proposes Agree which makes agreement at a distance possible, the status of Spec-head relation as a minimal search domain has become questionable. I will discuss the probe-goal Agree system in chapter 2 in detail. However, here I want to point out that the Chain Reduction algorithm has a burden to rely on the theoretically unstable feature checking system. Furthermore, the Chain Reduction algorithm has some serious theoretical problems. In Nunes’ system, uninterpretable features in the lower copies are deleted in PF, interface, along with its term. It means that this system assumes that copies with uninterpretable features are not deleted in syntax, but transferred to interface components, not only to PF but also to LF without any problem. This assumption contradicts the common assumption that undeleted uninterpretable formal features coming into the interface levels cause a derivation to crash. In addition, how the uninterpretable features that have been transferred to LF can be deleted is not clearly explained. Along with those theoretical problems, there are some pieces of empirical evidence against high copy privilege. I will introduce some examples that show the lack of high copy privilege in section 1.4. Before moving to these issues, I will discuss two more studies that assume a high copy privilege. 1.2.5. Pérez (2015) Pérez (2015) raises an important question about Nunes’ Chain Reduction algorithm. For the Chain Reduction algorithm to work, the syntactic derivation must be transferred to PF at the very moment when the highest copy has checked its uninterpretable features, but Chain Uniformization operates, so the highest copy has “advantage” over the lower copies. Due to such 25 a strong assumption, Pérez does not follow Nunes’ system, but still assumes high copy privilege as in Nunes (1995, 2004, 2011). To account for high copy privilege in pronunciation, Pérez adopts the idea of Late Insertion (Halle & Marantz 1993) and assumes that syntax operates only with a set of formal features without any phonological information. V ocabulary Insertion takes place post-syntactically. Following Chomsky (2000, 2001), Pérez assumes that syntactic features are pairs of attribute-value <Att,Val>. Attribute represents types of features (e.g. Tense, Number…), and value is a specific feature (e.g. past, plural…) that belongs to an attribute. Agreement gives a value to uninterpretable features, as exemplified in (33). In (33a), the wh- phrase who has two unvalued uninterpretable features: one is an unvalued case feature <κ,_> and the other is an unvalued question feature <ω,_>, each of which will be valued against T and C respectively. The lowest who is external-merged from the numeration to its theta-position. The second lowest who is the copy of the first one and merged in the (defective) embedded SpecTP as an intermediate step for further movement. Since neither uninterpretable feature can get valued in this position, the two features remain uninterpretable. As the next step, another copy of who is made and merged into the matrix SpecTP where its case feature get valued. Thus, the third copy in the matrix SpecTP has a valued case feature and an unvalued question feature. Finally, the last copy is created and merged to the matrix SpecCP where it gets its question feature valued. Thus, the highest copy has both a valued case feature and a valued question feature. (33) a. [CP Who [ TP who seems [ TP who to be [who happy]]]]? b. Who{<κ,NOM>,<ω,Q>, …}… who{<κ,NOM>,<ω,_>, …} … who{<κ,_>,<ω,_>, …} … who{<κ,_>,<ω,_>, …} (Pérez 2015, (19)) Another assumption Pérez made is that only the value-part of the feature is visible at the interfaces. Thus, from the perspective of PF and LF, the derivation in (33b) is not visible, but the representation with valued-part only as in (34) is visible. (34) Who{NOM,Q, …} … who{NOM, …} … who {…}… who {…} (Pérez 2015, (20)) Since members in the same chain show a subset relation in their valued feature entry, as in (34), every valued feature in a lower copy belongs to the set of valued features of a higher copy. Pérez 26 argues that this is the way how interface levels know that members in the same chain are non- distinct items. Based on the chain formation system above, Pérez proposes a Condition on Recoverability of Deletion (35) and Earliness principle in (36). The Condition on Recoverability of Deletion derives the uniqueness of pronunciation: at least one copy must be phonetically realized for the chain to be recoverable. However, it is not allowed that more than one copy is pronounced because pronouncing non-distinct elements as few times as possible is the most economical way to comply with the Condition on Recoverability of Deletion. In addition, the Earliness principle forces V ocabulary Insertion (VI) to take place as early as possible. The “earliness” is calculated based on a linearized representation as in (34), rather than a hierarchically structured one as in (33). Thus, the leftmost member in a chain, which corresponds to the highest member, will be assigned a phonological exponent. This is why high copy privilege is obtained, according to Pérez. (35) Condition on Recoverability of Deletion (Pérez 2015, (29)) An element may be deleted (i.e., not pronounced) if it is totally determined by a structurally related syntactic constituent. (36) Earliness (Pérez 2015, (31)) If a representation Σ is a proper input for an operation OP, apply OP immediately (i.e., apply operations as soon as possible). To account for an exceptional low-copy pronunciation in Serbo-Croatian multiple wh- fronting (Bošković 2001, see 1.2.4.), some flexibility of earliness principle needs to be required. The presumed derivation of the sentence in (37), under Pérez’ account, is demonstrated in (38). When VI started from the leftmost element in linearized elements, VI targets the highest copy of the subject due to the Earliness principle (38a). Since the next element, the highest copy of the object wh-phrase, is homophone to the subject that has already been assigned a phonetic exponent, the highest copy of the object wh-phrase is deleted by PF constraint against homophony (38b). The next element is a lower copy of the subject. Since the Condition on Recoverability of Deletion requires VI as few times as possible, this copy is deleted (38c). After 27 VI targets the verb (38d), finally a low copy of the object wh-phrase is assigned a phonetic exponent (38e) to obey the Condition on Recoverability of Deletion. (37) Šta uslovljava šta? what conditions what What conditions what? (Pérez 2015, (47), cited from (Bošković 2002:364) (38) a. ŠtaSUBJ < štaOBJ < štaSUBJ < uslovljava < štaOBJ [BY VI] b. ŠtaSUBJ < štaOBJ < štaSUBJ < uslovljava < štaOBJ [BY Antihomophony] c. ŠtaSUBJ < štaOBJ < štaSUBJ < uslovljava < štaOBJ [BY Recoverability of Deletion] d. ŠtaSUBJ < štaOBJ < štaSUBJ < uslovljava < štaOBJ [BY VI] e. ŠtaSUBJ < štaOBJ < štaSUBJ < uslovljava < štaOBJ [BY VI] (Pérez 2015, (52)) Pérez’s approach is intriguing especially in a sense that it does not associate structure information to selection of copy for pronunciation. Selection of copy for pronunciation takes place based on a linearized string without referring to any information about where each copy has been positioned in syntax. Once we assume that PF is solely in charge of selecting a copy for pronunciation, this structure-blind approach seems conceptually appealing. However, it is questionable whether this is theoretically better than other approaches which associate selection of copy for pronunciation to structure information. There has often been an intuition that a certain syntactic position requires an overt element. For example, SpecTP in English needs to have an overt element and based on the intuition, the EPP has been proposed. Under Pérez’s analysis, the EPP is at best understood as a pure syntactic requirement that triggers movement but has nothing to do with pronunciation. Thus, this approach does not capture the intuition that pronunciation is associated with syntactic requirements in some cases. Putting this issue aside, there is another theoretical problem in this approach. The system assumes that selection of copy for pronunciation occurs after linearization. However, it is not clear how it avoids the linearization contradiction discussed in Nunes (1999, 2004, 2011). Nunes argues that Chain Reduction takes place for linearization because non-distinct elements scattered in different syntactic positions cause contradictions for linearization under LCA (see 1.2.3). In addition, this approach has empirical problems like Nunes’ Chain Reduction algorithm because this system 28 grants a privilege to a high copy for pronunciation. However, as we will discuss in 1.3 and throughout this dissertation, a high copy privilege does not seem to truly exist. 1.2.6. Stjepanović (2007) Regarding selection of copy for pronunciation, Stjepanović (2007) provides an interesting observation in Serbo-Croatian (SC). Stjepanović observes that the distribution of subjects in SC presents a paradox. On one hand, it seems that subjects in SC must undergo movement to a position higher than TP. When the subject follows the sentential adverb, which is adjoined to TP (Bošković 1997), the sentence is ungrammatical (39a). By way of contrast, when the subject precedes the sentential adverb, the sentence is grammatical (39b). Stjepanović argues that the contrast in (39) shows that the position for subjects in SC cannot be lower than TP. (39) a. * Vjerovatno Marko tuče Petra. (Serbo-Croatian, Stjepanović 2007, (11)) probably Marko beat Petra “Marko is probably beating Petra.” b. Marko vjerovatno tuče Petra Marko probably beat Petra “Marko is probably beating Petra.” On the other hand, in some cases, it seems that subjects are allowed to stay in a position lower than TP. In (40), the subject Jovan appears at the end of the sentence. However, despite its surface position, the sentence final subject shows “high” behavior as if it sits in an A-position higher than its spell-out position. For example, the sentence final subject Jovan can bind the reflexive in the adverb that precedes the subject. Not only that, there is no violation of weak crossover in (41), so the subject quantifier can bind the variable in the adverb. (40) Mariju je [protiv svoje volje] oborio Jovan. (Serbo-Croatian, Stjepanović 2007, (12)) Marija-Acc is [against his will] failed Jovan-Nom “Jovani failed Marija against hisi will.” 29 (41) Mariju je [protiv svoje volje] oborio svaki profesor. Marija-Acc is [against his will] failed every professor Every professori failed Marija against hisi will. (Serbo-Croatian, Stjepanović 2007, (13)) To resolve this contradiction (obligatory subject fronting in (39) and absence of it in (40)-(41)), Stjepanović argues that the sentence final subjects in (40) and (41) actually undergo movement to the position which standardly hosts subjects in SC, namely SpecAgrsP, but the lower copy of the movement chain is phonetically realized to satisfy a prosodic requirement imposed to subjects in (40)-(41). To put it in differently, in SC, subjects must move to SpecAgrsP and are preferred to be pronounced there. However, when there is a prosodic requirement that forces the lower copy to be pronounced, the lower copy left in the base position is phonetically realized deleting the higher copy instead. The prosodic requirement applied in (40) and (41) is, according to Stjepanović, the stress assigning rules. The subjects in (40) and (41) are focused elements that must receive main stress in the sentence final position due to the Nuclear Stress Rule assignment algorithm. Since the stress rules in SC assign the main stress on the lowest copy of the subject, the copy left in its base position is chosen for pronunciation to realize the main stress 910 . Furthermore, Stjepanović parameterizes the ordering between stress assignment and copy deletion in PF across languages. In SC, stress assignment precedes copy deletion, so the low copy is assigned the main stress and pronounced to realize this main stress. By way of contrast, in English, stress assignment always follows copy deletion, so there is no case where a low copy, instead of a high copy, is selected for pronunciation due to prosodic requirement, as shown by the ungrammaticality of (42). 9 Especially, Stjepanović follows the version of NSR proposed by Zubizarreta (1998). Zubizarreta argues that NSR applies not directly to syntactic structure, but to metrically interpreted syntactic structure. Since it is not the case that a metrical structure always coincides with a corresponding syntactic structure, the sentence final element does not necessarily mean “sentence final syntactic phrase” in Zubizarreta’s system. See Stjepanović (2007) for how the algorithm works in SC. 10 To complete the explanation for the word order OVS as in (40) and (41), Stjepanović assumes that the object and the verb move out of VP by object shift or verb raising respectively (Stjepanović 1999, Bošković 1997). However, I will show that object movement in (40) and (41) patterns as scrambling rather than object shift. 30 (42) a. Question: Who was chased? b Answer: *Was chased Bill. Cf. Bill was chased. However, Stjepanović also points out in a footnote that it is not impossible to have a focused subject in the front of the sentence in SC. In some cases, copy deletion can precede the stress assignment, so (43b) is also a felicitous answer for the question in (43a). In this case, the high copy of the subject is selected for pronunciation and then focus accents are assigned on the subject by the subsequent stress assignment rules, as in the English grammatical counterpart. (43) a. Who is buying a computer? (Stjepanović 2007, fn 15) b. MARIJA kupuje kompjuter. MARIJA buys computer Marija is buying computer. Based on the copy theory of movement, Stjepanović’s account nicely captures the interaction between word order and information structure (and prosody) along with the “high” behavior of the sentence final subject. Even though there are some points that are not clear to me, as will be discussed, I found Stjepanović’s observation and account for it are important to understand the copy-selection mechanism in PF. I adopt Stjepanović’s core idea that subjects in SC move out of VP to some higher position and needs to be pronounced there unless the subject is focused. When the subject is focused, the subject chain is pronounced with a low copy to get a proper stress assignment. However, I depart from Stjepanović’s analysis regarding the following points. I am skeptical about the position of the subject in (39b). SC is a (relatively) free word language with scrambling. Thus, unlike English, the sentence initial position is not quite associated with subject hood but discourse configurational function. For example, in (40) and (41), Stjepanović assumes that the objects undergo object shift. However, the objects are pronounced in a position higher than the position where a shifted object is usually situated. In usual object shift in SC, the shifted object sits in between the auxiliary verb and the main verb as in (44). Not only that, it has been argued that object shift applies to only pronominal objects in SC (Stojanovič, 1997). Optional raising of full NP objects in SC has been analyzed as scrambling 31 rather than object shift. If this is the correct generalization, then the full NP object in (40) and (41) would not be a case of object shift, but a case of scrambling for discourse configurational reason. (44) a. Juče mi je Marija [nešto]i [VP ispričala ti]. (Serbo-Croatian, Stojanovič, 1997, (12)-(13)) Yesterday me is Marija something told Marija told me something yesterday. b. Jasna mi je rekla [CP da je Petar [nešto]i kupio ti] Jasna me is told that is Peter something bought Jasna told me that Peter had bought something. I assume that the object undergoes scrambling as topicalization in (40) and (41). Stjepanović as well notes that in SC, an (direct/indirect) object can occupy AgrsP when the object is a topic, as in (45). (45) a. Petra Ivan predstavlja Mariji. (Serbo-Croatian, Stjepanović 2007, (34-35)) Petar-Acc Ivan introduces Marija-Dat Ivan is introducing Petar to Marija. b. Petru Ivan predstavlja Mariju. Petar-Dat Ivan introduces Marija-Acc Ivan is introducing Marija to Petar. Based on (45), it seems more plausible to assume TopicP, instead of AgrsP, as the position which a sentence initial topical element (including subject, indirect object and direct object) occupies. In addition, a PF constraint that requires the element in SpecTopP to be phonetically realized forces the highest copy of topic movement chain to be pronounced. If this is correct, the subject in (39), as well as the objects in (40) and (41), seems to sit in SpecTopP regardless of their subject hood. In other words, if a subject is not topic, the subject does not necessarily appear in the sentence initial position. Since the subject in (39) is a topic, it undergoes movement as topicalization, and this gives the impression that subjects in SC must undergo movement in (39), but I suspect that this is misleading. Then, does a subject in SC stay in its base position within 32 VP? My answer is no. It actually moves, but the position which it targets is different from the one suggested by Stjepanović. In (44) and (45), subjects sit below a topicalized element but above the main predicate. Once we assume, as Stjepanović assumes, that verbs undergo movement out of VP, the subjects in (44) and (45) must be assumed to be in a position higher than VP departing from their base position. For now, I do not have an argument for the concrete position of subjects that can be fully justified, but the whole point is that the subject undergoes movement independent of topic-hood and the position the subject sits in is lower than the topicalized element. If this is a case, then the subjects in (44) and (45) are cases where a high copy in the subject chain is phonetically realized. In Stjepanović’s analysis, it is not explicitly mentioned why a high copy needs to be phonetically realized. In chapter 2, I will propose an idea that a functional category (for topic agreement or subject agreement) requires an element in its specifier position to be phonetically realized by reinterpreting the EPP as a PF constraint, following McFadden and Sundaresan (2015) and Landau (2007). Based on this proposal, I will provide an alternative analysis of the distribution of SC subjects in 4.2. For now, it seems to be enough to point out two critical patterns about the distribution of subjects in SC: (i) Subjects in SC undergo movement and need to be pronounced with a high copy in the moved position; (ii) However, when the subject needs to get a sentence stress in the sentence final position as a new information focus, the lowest copy is phonetically realized due to prosodic requirements. 1.3. Low-copy pronunciation and lack of high copy privilege So far, we have discussed low-copy pronunciation cases in Bobaljik (2002), Bošković and Nunes (2007), and Stjepanović (2007). The lower copies in those examples seem to be able to be pronounced because pronunciation of the highest copy violates a PF constraint. In this section, I will introduce some cases where a low copy is pronounced even without any PF constraint that blocks high-copy pronunciation. 1.3.1. Low-copy pronunciation in argument domain Except object shift (Bobaljik 2002), covert A-movement has been rarely observed. To the best of my knowledge, Polinsky and Potsdam (2012) for Adyghe and Wurmbrand and Haddad (2014) for 33 Standard Arabic are the only works that seriously investigate (a potential) covert A-movement 11 . Potsdam and Polinsky argue that the Northwest Caucasian language Adyghe has covert subject- to-subject raising in sentences like one in (46b) and it is the covert analogue to the overt subject raising derivation in (46a). (46) a. a-xe-ri [a-xe-mei pjəsme-r a-txə-new] ø-fjeZ’a-Re-x (Adyghe) DEM-PL-ABS DEM-PL-ERG letter-ABS 3PL.ERG-write-INF 3ABS-begin-PAST-3PL.ABS ‘They began to write a letter.’ b. a-xe-ri [a-xe-mei pjəsme-r a-txə-new] ø-fjeZ’a-Re-x DEM-PL-ABS DEM-PL-ERG letter-ABS 3PL. ERG-write-INF 3ABS-begin-PAST-3PL.ABS ‘They began to write a letter.’ (Potsdam & Polinsky 2012, (1)-(2)) They argue that the case marking of the subject indicates the clause membership of the subject: When the subject is ergative case marked, it belongs to the embedded clause, while when the subject is absolutive case marked, it belongs to the matrix clause. Based on this distinction, Potsdam and Polinsky states that the subject in (46a) sits in the matrix clause whereas the subject in (46b) is in the embedded clause 12 . Interestingly, the subject in (46b) shows ‘high’ behavior as if it is in the matrix clause. Several diagnostic tests are suggested to detect the high behavior of the subject: i) locality 11 I will discuss Standard Arabic in chapter 2 in more detail. 12 Potsdam and Polinsky show that the subject in (46) is not the subject of a complex predicate incorporating the embedded verb and the matrix verb as a single clause. One piece of evidence supporting this argument is observed in adverb modification. In (i), the adverb ṭwe ‘twice’ can modify either the matrix event or the embedded event. In (ia), the adverb modifies the matrix event, while in (ib), it modifies the embedded event. This indicates that the matrix clause and the embedded clause describe an independent event each. It serves as a piece of evidence that shows that the subject is not in a single clause with a complex predicate. (i) a. [mašjəne-r depqə-m jewe-new] ṭ we χ wə-ʁe (Adyghe, Potsdam & Polinsky 2012, (18)) Car-Abs wall-Obl hit-Inf twice turn out-Past The car twice turned out to hit the wall. b. [mašjəne-r depqə-m ṭ we jewe-new] χ wə-ʁe Car-Abs wall-Obl twice hit-Inf turn out-Past The car turned out to twice hit the wall. 34 diagnosis and ii) c-command diagnosis. The first diagnosis is based on the assumption that agreement takes place in a Spec-head relation. Under this assumption, they argue that agreement between the verb and the DP signals that the DP sits in the spec of the functional head that is in charge of the agreement. In the Adyghe raising construction (46b), there is an φ-agreement between the subject DP in the embedded clause a-xe-me ‘DEM-PL-ERG’ and the matrix verb ø- fjeZ’a-Re-x “3ABS-begin-PAST-3PL.ABS”. Thus, it seems that the subject in the embedded clause is in Spec-head relation with the matrix verb at some point in the derivation. The second diagnosis consists of three independent tests: control of PRO, binding of reflexives, and scope ambiguities. The assumption is that a DP that undergoes covert A-movement can c-command everything below it from its potential landing site. In Adyghe (covert) raising construction (47a), the subject DP in the embedded clause can control PRO in the matrix clause. Similarly, in (47b), the subject DP in the embedded clause can bind the reflexive attached to the matrix clause. The last test, scope ambiguity, leads to the same conclusion. In (47c), the subject DP in the embedded clause can scope over the scope-bearing adverb in the matrix clause. (47) a. Control (Adyghe, Potsdam & Polinsky 2012, (40)) [a-š’ wəse-r ə-txə-new] a-ri [PROi jež’ə-r] fjež’aʁ. 3SG.ERG poem-ABS 3SG.ERG-write-INF DEM-ABS by_self-ABS began He began to write a poem himself. b. Reflexive binding (Adyghe, Potsdam & Polinsky 2012, (37)) [a-xe-me pjəsme-r a-txə-new] zəi-fjež’a-ʁe-x DEM-PL-ERG letter-ABS 3PL.ERG-write-INF REFL-begin-PAST-PL They began to write a letter for themselves.’ c. Scope (Adyghe, Potsdam & Polinsky 2012, (34)) [stWədentə pepč zadačə-r q’-a-ŝ ̣ ə-new] t ̣ we fjeZž’aʁ. student each(ERG) problem-ABS DIR-3PL.ERG-do-INF twice began Each student twice had a go at solving a problem.’ EACH > TWICE It happened twice that each student began to work on a problem.’ TWICE > EACH To account for the data that show that the subject in the embedded clause behaves as if it is in the matrix clause with respect to the agreement, scope or binding, Polinsky and Potsdam argue 35 for the following derivation, represented in (48): the subject is base-generated in SpecvP in the embedded clause and gets ergative case checked in the base position. And to satisfy the EPP requirement of the embedded clause, the subject makes a copy of itself and merges one into the embedded SpecTP (48a). Since the matrix verb is an unaccusative verb which cannot assign any theta role to its external argument, so cannot have an external argument, the embedded subject is recruited for the EPP satisfaction in the matrix clause. Thus, another copy of the embedded subject is created and merged to SpecTP in the matrix clause (48b). In addition, assuming that the absolutive case is checked in SpecTP in an intransitive clause, they argue that in Adyghe, it is possible for a DP to have its case checked twice (48c). Crucially they adopt Nunes’ Chain Reduction algorithm, so they assume that when a copy of a chain checks its feature, features of the particular copy are deleted not affecting other copies in the chain. Thus, the low copy of the subject has its case checked by the embedded v whereas the higher copy has its case checked by the matrix T. As a result, neither copy, the one in the matrix SpecTP or the one in the embedded SpecTP, has unchecked features. According to the Chain Reduction algorithm, PF leaves a single copy that has the least number of unchecked features and deletes every other copy for linearization and economy considerations. If this is the case, then in the Adyghe raising construction in question, Chain Reduction can optionally leave either the copy in the matrix SpecTP or one in the embedded SpecTP because they are the same in the number of unchecked copies (i.e. either of them has no unchecked features). When the higher copy is pronounced, the PF realization is reminiscent of overt movement as in (46a), while when the lower copy is pronounced, the PF realization is of covert movement as in (46b). (48) a-xe-ri [a-xe-mei pjəsme-r a-txə-new] ø-fjeZ’a-Re-x. DEM-PL-ABS DEM-PL-ERG letter-ABS 3PL. ERG-write-INF 3ABS-begin-PAST-3PL.ABS They began to write a letter. a. Copy of the subject and merge of one of the copies in the embedded SpecTP [ TP a-xe-me i [Case:ERG] [vP a-xe-me i [Case:ERG] pjəsme-r a-txə-new] DEM-PL-ERG DEM-PL-ERG letter-ABS 3PL. ERG-write-INF 36 b. Merge of one of the copies into matrix SpecTP [ TP a-xe-me i [Case:ERG] [CP[ TP a-xe-me i [Case:ERG] pjəsme-r a-txə-new]] ø-fjeZ’a]. DEM-PL-ERG EM-PL-ERG letter-ABS 3PL.ERG-write-INF 3ABS-begin-PAST c. Feature checking in the matrix clause [ TP a-xe-me i /r[Case:ERG/ABS] [CP[ TP a-xe-me i [Case:ERG] pjəsme-r a-txə-new]] ø-fjeZ’a-Re-x]. DEM-PL-ERG/ABS DEM-PL-ERG letter-ABS 3PL.ERG-write-INF 3ABS- begin-PAST-3PL.ABS However, I want to point out two potential problems in this analysis. First, it is not clear how the system chooses the copy of subject DP in the embedded TP over one in the base position (SpecvP) for pronunciation. Since Potsdam and Polinsky assume that the ergative case is valued by the embedded v, the copy in its base position has no unchecked features like other copies. Given that Adyghe is a SOV language, without any adverbs, it seems difficult to tell whether a copy in the embedded SpecTP or another in the embedded SpecvP is spelled-out in the case like (46b). I believe this part must be considered to complete the picture of their analysis. Another issue comes from scope interpretations. In (47c), the subject with the ergative case marker shows scope ambiguity. It can scope over or below the matrix scope-bearing adverb. However, the narrow scope of the subject is puzzling given the wide-spread observation that A-movement (especially in raising construction) does not undergo scope reconstruction unless the subject is indefinite (Chomsky 1993, 1995, Lasnik 1999, Boeckx 2001 among many others). As in (49), the subject that undergoes A-raising cannot have its scope from its base position. If we assume that the subject in (52) undergoes A-raising, then the subject is predicted to have only wide scope, contrary to fact. (49) Everyone seems not to be there yet. ( ∀ > ¬; ∗¬ > ∀) Cf. Someone from New York seems to be at the party. (someone > seem, seem > someone) In sum, even though it is not clear whether the subject undergoes a true A-raising, the Adyghe data show that a subject can be interpreted in a position higher than the position where it is phonetically realized. If this is a real case where a low copy of a movement chain is pronounced, as proposed by Potsdam and Polinsky, it would be a case where a low copy is 37 pronounced even without any PF constraint that blocks a high copy to be pronounced. This is problematic to Pérez (2015)’s approach due to the apparent lack of high copy privilege. Nunes’ Chain Reduction algorithm is in fact compatible with the data even though there is no high copy privilege, as explained by Potsdam and Polinsky. However, this account seems to be challenged by the somewhat puzzling reconstruction issue. Abandoning the idea of high copy privilege, I will discuss how an alternative approach proposed in this dissertation can account for the Adyghe data in 3.2.1. In Korean, we can find another case where a low copy in an A-chain is pronounced without any PF constraint that blocks high-copy pronunciation. Korean has a topic marker –(n)un. This marker can be attached to subjects (50a), objects (50b) or adverbs (50c). A phrase marked with this topic marker represents contrastive topic. However, when the topic marker is attached to a subject as in (50a), the subject can also represent a sentential (thematic) topic (This is the only case where a phrase with the topic marker has sentential (thematic) topic reading (I will discuss the reason why non-subject elements with the topic marker cannot have thematic topic reading in 4.3.). (50) a. John-un chayk-ul sa-ss-ta. John-top book-Acc buy-past-decl. John bought a book (for others, I do not have an idea though). In the morning, [John]TOPIC bought a book. b. John-i chayk-un sa-ss-ta. John-Nom book-top buy-past-decl. John bought a book (for other items, I do not have an idea though). c. Achimey-nun John-i chayk-ul sa-ss-ta. In the morning-top, John-Nom book-Acc buy-past-decl. In the morning (not in the afternoon), John bought a book. Putting aside the contrastive topic reading, I will focus on the sentential topic reading of the subject with the topic marker. Even if an object is scrambled over the subject with the topic marker, as in (51), the subject maintains its sentential topic status. (Note, in this case, the scrambled object is interpreted as focus). I will show that the subject with the topic marker behaves as if it is in a position higher than the scrambled object in (51). 38 (51) Chayk-uli John-un ti sa-ss-ta. Book-acc John-top buy-past-decl. John bought a book. Interestingly, there is a sharp distinction between a nominative case-marked subject and a topic-marked subject in their binding property and scope taking: (i) In (52), short distance-moved object Minho-lul ‘Minho-Acc’ can bind the (linearly) following subject with nominative case marker caki-emma-ka ‘self-mom-Nom’, but not one with topic marker caki-emma-nun ‘self-mom- top’. That the scrambled object can bind the subject with nominative marker in (52a) means that the object is an A-position where it c-commands the in-situ subject. Even though the object moves to a higher position in (52b) as well, it cannot bind the subject with the topic marker, which implies that the subject with the topic marker is not in the c-command domain of the scrambled object. (52) a. Minho-luli caki emma-ka Minho-luli chwuchenhay-ss-ta. Minho-Acc self mom-Nom recommend-past-decl.. b. *Minho-luli caki emma-nun Minho-luli chwuchenhay-ss-ta. Minho-Acc self mom-top recommend-past-decl. Lit. Minho’s mother recommended Minho (ii) A subject with a topic marker is free from binding C violation when a co-referential anaphor linearly precedes it, but a subject with the nominative case marker is not, as shown in (53). The binding relation in (53b) is not due to object reconstruction because the scrambled object must scope over negation (also > Neg, *Neg > also). (53b) shows that a subject with the topic marker sits higher than the scrambled object. (53) a. *Caki-casin-toi Minho-ka Caki-casin-toi pinanhay-ci anh-ass-ta self -self -also Minho-nom blame-NEG-Past-decl. b. Caki-casin-toi Minho-nun Caki-casin-toi pinanhay-ci anh-ass-ta self -self -also Minho-top blame-NEG-Past-decl. Minho did not blame himself too. (i) also > neg: Minho did not blame others as well as himself. 39 (ii) *neg > also: It is not the case that Minho blame himself as well (= Minho blamed others, but not himself). (iii) The scrambled anaphor can be bound by the following subject with a topic marker (54b), but not by a subject with a case marker (54a). This also indicates that the subject in (54b), unlike that in (54a), is in a position where it c-commands the scrambled object. (54) a. *Caki chayk-uli Minho-ka caki chayk-uli kwuiphay-ss-ta. Self book-acc Minho-top buy-past-decl. b. Caki chayk-uli Minho-nun caki chayk-uli kwuiphay-ss-ta. Self book-acc Minho-top buy-past-decl. Lit. Minho bought his book. (iv) A subject with the topic marker can scope over an object quantifier even though the subject follows the object in the linear order. Thus, the sentence in (55) is compatible with the situation in (56a), but not the one in (56b) indicating that the object quantifier cannot scope over the subject quantifier (contrary to their word order). (55) [Chayk twu kwon-to]i haksayng sey myeng-un [Chayk twu kwon-to]i sa-ss-ta. [Book two CL-also] student three CL-top buy-past-decl. Three students bought two books too. *two > three, three > two (56) a. Students were supposed to buy any two books and any four pens in a shop in a limited time. Many students did not focus on shopping, but there were three students who concentrated in their job. When their teacher gathered all the students, many students just bought four pens, but (55) b. There are two DVDs and two books on sale in a small book store. Three students came and bought a DVD and a book. Another three students came and bought the other DVD and the other book. 40 In all cases (51)-(55), a topic marked subject, not a nominative case marked subject, seems to sit in a position higher than the preceding object. I will discuss the derivation of those examples above in section 4.3, in detail. For now, let’s simply assume that the subject undergoes topic movement when it is attached with a topic marker. Thus, under the copy theory of movement, those Korean examples seem to serve as a case of pronouncing a low copy. Crucially, there is no PF constraint that blocks the higher copy pronunciation 13 . The Korean data challenges Nunes’ Chain Reduction algorithm. The Chain Reduction algorithm assumes that feature checking takes place in a Spec-head relation and feature checking only affects the copy in the local relation with the relevant functional head. Once the operation “copy” makes a copy of the subject in SpecvP and merges a copy into SpecTP, the topic feature of the high copy is checked whereas the feature of the low copy is left unchecked, as exemplified in (57a) (note. there is another movement, the focus scrambling of the object. However, to concentrate discussion on the subject movement, I will not deal with this movement. See footnote (13)). Since the Chain Reduction finds the most economical derivation, it is expected to delete the low copy of the chain because it outnumbers the high copy in the number of unchecked features as in (57b). In other words, unless there is any PF constraint that blocks the high copy to be 13 When the subject with the topic marker linearly precedes the object, the syntactic structure for the linear order is ambiguous: both subject and object are pronounced with its high copy (ia) or both are pronounced with its low copy (ib). This possibility is hinted by the relative scope taking with negation. When both chains are pronounced with its high copy, the focus particle scopes over negation, as in (ia) while when both chains are pronounced with its low copy, negation scopes over the focus particle. These data implicate that the choice of copy for pronunciation in topic movement is optional in fact. (i) a. [ TopP Minho-nun [ FocP caki-casin-to i [ vP Minho-nun caki-casin-to i pinanhay-ci anh-ass-ta]]] Minho-top self -self -also Minho-top self -self -also blame-NEG-Past-decl. Minho did not criticize himself as well (He did not criticize anyone including himself). also > Neg b. [ TopP Minho-nun [ FocP caki-casin-to i [ vP Minho-nun caki-casin-to i pinanhay-ci anh-ass-ta]]] Minho-top self -self -also Minho-top self -self -also blame-NEG-Past-decl. It is not the case that Minho criticized himself as well (He criticized others but not himself). Neg > also The last possibility that the subject is pronounced with its high copy, but the object is with its low copy is hard to tease apart from the case in (ib) in terms of scope tests. Since it is not important to the main argument, I will put this issue aside. (ii) [ TopP Minho-nun [ FocP caki-casin-to i [ vP Minho-nun caki-casin-to i pinanhay-ci anh-ass-ta]]] Minho-top self -self -also Minho-top self -self -also blame-NEG-Past-decl. 41 pronounced, the high copy is expected to be privileged for pronunciation under the Chain Reduction algorithm. This is not what happened in the Korean examples. In the Korean examples, even though there is no PF constraint that blocks the high copy pronunciation, the low copy is pronounced unexpectedly by the Chain Reduction algorithm. (57) a. [TopMinho-nun[top] [FocP caki-casin-toi [vP Minho-nun[top] caki-casin-toi pinanhay-ci Minho-top self -self -also Minho-top self-self-also blame- anh-ass-ta]]] NEG-Past-decl. b. [TopMinho-nun[top] [FocP caki-casin-toi [vP Minho-nun[top] caki-casin-toi pinanhay-ci Minho-top self -self -also Minho-top self-self-also blame- anh-ass-ta]]] NEG-Past-decl. In this section, I discussed two cases where a low copy is pronounced even without any PF constraint that blocks a high copy to be pronounced. Although Potsdam and Polinsky provide an analysis for (seemingly) subject-to-subject raising in Adyghe based on the Chain Reduction algorithm, I pointed out that the analysis is challenged by the puzzling reconstruction effects in A- movement. Furthermore, Korean topic movement seems to violate the Chain Reduction algorithm. In the next section, I will briefly introduce some cases where a low copy is pronounced in an Ā- chain even without any PF constraint that blocks a high copy to be pronounced. The data in this section and the next section call an alternative analysis for copy-selecting mechanism in PF. 1.3.2. Low-copy pronunciation in non-argument domain One example of low-copy pronunciation in a non-argument domain is observed in German. German has been known as a rigid-scope language. A sentence with two quantifiers as presented in (58a) is scope-wise unambiguous: Only the subject QP scopes over the object QP, in compliance with the linear order. However, interestingly, many have observed that this scope rigidity disappears when the sentence is pronounced with a particular intonation. If the rising-falling prosody is applied to the sentence, the inverse scope reading (i.e. the object QP wide scope reading) 42 becomes available, as in (58b) (Jacobs 1982, Féry 1993, Büring 1994, 1997b, Krifka 1998, Bobaljik & Wumbrand. 2012). (58) a. Ein Kind hat fünf Museen besucht (Neutral prosody) a child has five museums visited There is a child who visited five different museums. a > five Five museums have been visited by a potentially different student. *five > a b. Ein /KIND hat \FÜNF Museen besucht (Rising-falling intonation) There is a child who visited five different museums. a > five Five museums have been visited by a potentially different student. five > a (Krifka 1998, (15)) An existential quantifier-numeral quantifier pair in (58), is not a particularly special case showing scope inversion under a rising-falling intonation. It has been reported that other pairs of quantifiers show the same effects. For example, when the subject quantifier is weniger als fünf ‘less than five’ and the object quantifier is viele ‘many’ as in (59), we still observe that the in-situ object quantifier viele ‘many’, which follows the subject quantifier weniger als fünf ‘less than five’ can distribute over the subject quantifier when the sentence is pronounced with the rising-falling prosody (59b). Without the special prosody, the sentence has only a surface linear scope reading in which weniger als fünf ‘less than five’ scopes over viele ‘many’ (59a). (59) a. Weniger als fünf Studenten haben viele Lehrer eingeladen. (Neutral prosody) less than five students have many teachers invited For less than five students x, each of x invited many teachers. less than five > many *For many teacher x, each of x was invited by less than five students. *many > less than five b. Weniger als fünf /Studenten haben \VIELE Lehrer eingeladen. (Rising-falling) less than five students have many teachers invited For less than five students x, each of x invited many teachers. less than five > many For many teacher x, each of x was invited by less than five students. 43 many > less than five Other quantifier pairs (someone-every, at least one-every in Krifka (1998), Numeral QP-Numeral- QP in Cook et al (2006)) were also tested in German and the inverse reading becomes available with the rising-falling intonation. Thus, the possibility of inverse scope with special prosody is not a lexically-specific phenomenon; rather it requires a structural explanation. Then, what is special in the rising-falling intonation? How can the object QP scope over the subject QP that precedes it in the linear order? If we interpret the rising intonation on the subject QP as a way of expanding pitch range of the subject QP and falling intonation on the object QP as a way of de-stressing of the object QP, the occurrence of rising-falling intonation seems to make the subject QP focus and the object QP topic. If this speculation is the correct analysis for the prosodic pattern in question, the object wide scope reading in the rising-falling intonation can be understood as one of the general observation that a topic occupies the highest position in a structure, so has wide scope (Beghelli and Stowell. 1997, Jimenez-Fernandez & Miyagawa 2014, among many others). Furthermore, Krifka (1998, p105) observed that most patterns different from other quantifiers in that it always has wide scope over another quantifier regardless of the word order. He attributes this wide scope biased reading of most to its inherent topic property. Assuming this is the correct analysis, I argue that an object wide scope reading is obtained when an object undergoes topic movement to a position higher than a subject. The reason why the object QP follows the subject QP in the linear order is because the chain is pronounced with a low copy. For example, the derivation of (58b) is represented in (60). The object QP fünf Museen ‘five museums’ moves to the SpecCP for topic interpretation, but the high copy of the chain is deleted at PF. (60) [CP fünf Museen ein Kind [C ′ hat [Ein Kind fünf Museen besucht]]] five museums a child has a child five museums visited Of course, the high copy can survive. However, in that case, the high copy of the subject QP must be deleted because German V2 constraint requires that only a single phrase can be phonetically 44 realized in SpecCP (Boeckx 1998, Bošković 2001) 1415 . The sentence in (61) is the case where the high copy of the object survives. The sentence in (61) has been discussed in the name of “object scrambling” that eliminates scope-rigidity and allows the object-wide scope reading. (61) [CP fünf Museen ein Kind [C ′ hat [Ein Kind fünf Museen besucht]]] five museums a child has a child five museums visited In 4.3, I will introduce some experimental work on prosodic effects on scope interpretation in Korean. The results of the experiment imply that covert topic movement is the correct analysis for the object wide scope reading in Korean sentence corresponding to (58b). I believe that the Korean data can serve as indirect evidence for covert topic movement in German. Postponing the detailed analysis to 4.3, I hope to point out that the inverse scope reading supported by prosody in German is another case where a low copy is phonetically realized even without any PF constraints that block high-copy pronunciation. Rather, given that the object QP is hard to have a wide scope in neutral prosody (due to scope rigidity), it seems that the low copy is “allowed” to be pronounced only if a certain condition (which is realized in prosody) is satisfied. These examples cannot be accounted by the Chain Reduction algorithm. Bobaljik and Wumbrand (2012) account for the object wide scope reading with the help of prosody using the idea of Minimize (LF-PF) Mismatch (see section 1.2.1.). In chapter 4, I will show why Minimize Mismatch is not the best analysis 14 Boeckx (1998) argues that the V2 order in German is due to a PF parameter that forces the finite verb to be in the second position of its intonation phrase. This claim is supported by that a finite verb can be the third syntactic constituent as long as the verb is in the second position in its intonation phrase. Bošković (2001) also claims V2 requirement is a PF constraint that there must be one element phonetically realized before a finite verb. 15 In German, a focus element can either move to SpecCP (ia) or stay in its base position (ib). This optionality can be also accounted for under the same line of explanation. In (ia), a high copy of the focus-movement-chain survives, while in (ib), a low copy of the chain survives at PF. (i) a. [ CP Ein BUCH [ hat [ IP er gelesen]]] (Fanselow 2004, (5)-(6)) a book has he read b. [ CP Er [hat [ein BUCH gelesen]]] he has a book read He has read a book. 45 cross-linguistically by discussing a similar observation in Korean that cannot be accounted for by the Minimize Mismatch. What is important for now is that in German, it seems that a topic chain can be optionally pronounced with either a high copy or a low copy. Another example is found in English. Pesetsky (2000) proposes that languages develop two types of covert movement: covert phrasal movement and formal feature (ff) movement. The critical evidence that distinguishes the two types of covert movements comes from Antecedent- contained Deletion (ACD) licensing (62a). It has been widely assumed that in English, VP- ellipsis in ACD is licensed by QR-like covert movement that displaces the DP containing the elided VP to a position higher than VP, as in (62b). (62) a. John invited [every guest Bill did [VP Δ]] b. John [every guest Bill did [VP Δ]]i [VP invited ti] This ACD licensing of covert movement is applied to detect covert phrasal wh-movement. In English, multiple wh-questions are constrained by the Superiority condition (Chomsky 1973) as in (63). Thus, the structurally higher wh-phrase can appear in the front of the sentence (63a) while the structurally lower one cannot move over the other wh-phrase (63b). However, there is an exception. When the wh-phrases are D-linked, the lower wh-phrase can move over the higher one as shown in (64). (63) a. Who ate what? b. *What does who eat what? (64) Which book did which person buy which book? Pesetsky (2000) observes that there is an interesting correlation between superiority violation and ACD licensing. In-situ wh-phrases in contexts of superiority violations such as (64) cannot license ACD (65a), while the in-situ wh-phrase in superiority obeying configuration can license ACD as in (65b). (65) a. *I need to know which girl Sue ordered [which boy that Mary (also) did ∆] to congratulate __. 46 meaning: I need to know for which girl x and [which boy y such that Mary ordered y to congratulate x], Sue also ordered y to congratulate x. b. I need to know who can speak [which languages that Ken Hale can [VP ∆]]. (Pesetsky 2000: 29-30) Based on such an observation, Pesetsky argues that there are two types of covert movement; covert phrasal movement (in-situ wh-phrases in Superiority-obeying contexts (63a, 65b), and covert ff-movement (in-situ wh-phrases in Superiority-violating contexts (64, 65a) 16 . Therefore, a system must develop a way to distinguish two distinct covert movement operations. As Pesetsky noted, the covert wh-phrasal movement can be understood as a case where a low copy is phonetically spelled-out in an Ā-chain. One question that naturally follows is why the lower copy of the second wh-phrase, not the first one, in multiple wh-constructions in English is selected for pronunciation. If we follow Nunes, it is expected that there should be a phonological constraint that blocks the high-copy pronunciation of the second wh-phrase. The possible PF constraint is the ban of more than one wh-phrase to be pronounced in the sentence initial position. However, this possibility does not seem to be promising because such a phonological constraint is not cross-linguistically supported. In fact, many languages, including Slavic languages, Korean and Japanese, allow multiple wh-phrases to be pronounced in the sentence initial position. Rather, it seems more plausible to assume a language parametric restriction that English allows only one wh-phrase to be phonetically realized in SpecCP unlike multiple wh- fronting languages where multiple wh-phrases are allowed to be pronounced with the high copy in SpecCP. Then, a question needs to be answered is what PF constraints force English multiple wh-questions to have only one overt wh-phrase in SpecCP leaving other wh-phrases to be pronounced in their base position. The answer will be proposed in the rest of this dissertation. 1.4. Towards an alternative: Organization of dissertation In this chapter, reviewing Bobaljik (2002), Nunes (1995, 2004, 2011), Bošković and Nunes (2007), and Pérez (2015), I have discussed the benefit of the copy theory of movement in accounting for 16 Pesetsky proposes intervention effects as another diagnostic tool to detect two different types of covert movement. I will discuss intervention effects in detail in 4.2. 47 so-called covert movements. Under the copy theory of movement, there is no need to assume that the same operation, phrasal movement, takes place either in syntax or in LF. Every phrasal movement can be reduced into a syntactic operation. Not only that, this line of approach well accounts for some covert movement cases that are constrained by phonological requirement without look-ahead problems. Different mechanisms of selection of copy have been proposed (Minimize Mismatch by Bobaljik 2002, Chain Reduction algorithm by Nunes 1995, 2004, 2011, Earliness principle by Pérez 2015). However, those accounts assume high copy privilege for pronunciation in a sense, which is not compatible with the observations reported in 1.3. In this dissertation, I will propose an alternative system that can account not only for a high copy privilege in some cases but also the lack of high copy privilege in others. In chapter 2, I will propose a system that dissociates agreement, movement and pronunciation from each other. Under the system, I will propose three different types of movement and show how those different types of movement are realized in languages. After justifying the three types of movement, I will focus on the issue of how to select a copy at PF. I will suggest two PF constraints that play a role in copy selection at PF. Based on the theoretical apparatus developed so far, in chapter 3, I will investigate typological studies in subject agreement and wh-agreement. Especially, the analysis leads to a new approach to the typology of in-situ wh-phrases. In chapter 4, I will discuss two case studies in Korean and show how the system developed in the dissertation well captures an interaction between prosody and syntax in the language, which can be extended to other relevant languages. In chapter 5, I will provide an alternative analysis based on the proposed system for the issues raised in chapter 1 (i.e. a low copy is exceptionally selected for pronunciation because high-copy pronunciation violates a (morpho-)phonological constraint). Not only that, I will provide additional observations that support the current system. In chapter 6, I will conclude the discussion. 48 Chapter 2. Agree, Movement, and Pronunciation 2.1. Overview: Dissociating Agree, the EPP, and Spell-out at PF As introduced in chapter 1, in the early minimalist program, it has been assumed that movement takes place for feature checking to occur in a Spec-head relation. This way of explaining the motivation of movement was challenged by a new agreement hypothesis, namely, Agree (at a distance) (Chomsky 2000, 2001). Under the Agree system, a functional head (probe) which has an uninterpretable feature searches for an element that also has an uninterpretable feature that can be checked against the probe (the Activity Condition – both probe and goal must have uninterpretable features to be active). Crucially, the minimal search domain for agreement is defined based on the complement domain excluding Spec-head relations from the local domain for agreement. Once the probe finds its goal, agreement takes place at a distance. For example, in the unaccusative sentence in (1a), the case feature of the subject John is checked against T in the base position of John inside VP (1b-c). However, as the linear order indicates, the subject DP must move to the front of the sentence. Chomsky argues that this movement takes place to satisfy the EPP requirement, irrelevant to uninterpretable case or φ-feature checking. Due to the EPP satisfaction, the subject DP that has already checked its case feature in its base position moves to SpecTP. Under the copy theory of movement, the derivation makes a copy of the subject DP and merges it in SpecTP for the EPP satisfaction (1d). (1) a. John arrived. b. Probing (Probe)T vP arrived VP John[uCase: ] (Goal) 49 c. Agree (Probe)T vP arrived VP John[uCase:Nom] (Goal) d. The EPP satisfaction TP John[uCase:Nom] T vP arrived VP John[uCase:Nom] Thus, under the Agree system, movement is not triggered for feature checking. Rather, movement and feature checking are dissociated from each other. The most representative example that supports this idea is English there-expletives. The subject in a raising sentence can either move to the matrix SpecTP (2a) or stay in its base position (2b). Even though the indefinite NP someone from New York stays in a position lower than SpecTP as in (2b), it undergoes agreement with the matrix T. As a result of this agreement, the matrix verb is inflected by the number agreement with the postverbal NP and the NP has its case feature checked, as in its overt movement counterpart in (2a). The sentence in (2b) differs from the one in (2a) in that the NP does not move to SpecTP, but instead, there is inserted to satisfy the EPP requirement of the matrix clause. (2) a. [Someone from New York] seems to be at the party. b. There seems/*seem to be [someone from New York] at the party. (Bobaljik 2002, (72)) 50 To explain the definiteness effect and the agreement facts, some argue that the associate NP covertly moves to the expletive, as in (3) 17 . (3) [There+someone [seems to be [someone from New York] at the party]]. However, such covert raising analysis of there-associates has been rejected because the postverbal associate patterns differently from the preverbal NP with respect to scope-taking and binding (den Dikken 1995, Bošković 1997). In (2a), the indefinite someone from New York can either scope over or scope under the matrix verb seem. Thus, (2a) is scope-wise ambiguous between two readings in (4). However, (2b) has only the indefinite narrow-scope reading, corresponding to (4b). Another contrast between preverbal subjects in raising constructions and postverbal NP in there construction is found in binding. The preverbal subject in a raising construction in (5a) can be a A-binder, while the postverbal NP in (5b) cannot. These examples show that the postverbal NP does not covertly move to a position higher than its surface position. Thus, even though there seems an agreement between the matrix verb and the embedded NP, this agreement does not require the two elements to be in spec-head relation. Agreement can take place at a distance 18 . (4) a. someone > seem: There is a person from New York who seems to at the party. b. seem > someone: It seems that a person from New York is at the party. 17 Definiteness effects restricts the set of possible associates of expletives in indefinites. In other words, definite NPs cannot be the associate of there, as shown in (i). (i) a. There are dogs in the park. b. *There is the student in the classroom. 18 In the early minimalist program (Chomsky 1995), there-expletives are accounted for by formal feature movement (ff-movement) to I(nfl). Since ff-movement only targets formal features, it does not carry phonological features of the target giving an impression that the movement takes place covertly. Furthermore, ff-movement analysis well captures the “low” behavior of there-associates (2)-(5) since semantic components are left in-situ as well. Nevertheless, Chomsky (2000) abandons ff-movement in favor of Agree for theoretical reasons. Even though it seems too subtle to empirically distinguish Agree from ff-movement, Boeckx (2004) argues that an Agree-based account is superior to an ff-movement account in analyzing Hindi long distance agreement. For more information, See Boeckx (2004). 51 (5) a. Nobodyi/somebodyi’s mother was blaming hisi father. b. *There was nobodyi/somebodyi’s mother blaming hisi father. By introducing Agree, Chomsky (2000, 2001) reanalyzes (so-called) overt/covert movement using Agree. Overt movement is understood as the combination of Agree and the EPP satisfaction and covert movement is understood as if just Agree takes place. This reanalysis has a hidden assumption that when a phrase moves for the EPP satisfaction, a high copy is always pronounced. This hidden assumption then raises a question why the EPP satisfaction requires the high copy to be pronounced. This question becomes more puzzling when we take null expletives into consideration which are assumed in Romance. Null expletives are proposed as phonetically null elements that fill the SpecTP position syntactically in Romance null subject languages, as in (6). (6) (*egli/ciò) sta piovendo (Italian, Sheehan 2016, (9)) it/this is.3SG raining It’s raining. Thus, a somewhat contradictory pattern regarding the EPP is observed: In some languages, it seems that an element recruited for the EPP satisfaction is not necessarily pronounced (null expletives) while in other languages it seems that an element that satisfies the EPP must be pronounced (overt movement to SpecTP in English). This contradiction shows that the dichotomy in Chomsky (2000, 2001), Agree alone is like covert movement and Agree+EPP satisfaction causes overt movement, is too simple to capture the whole picture. In addition, another big question follows: In chapter 1, I discussed that there are many cases cross- linguistically that indicate the existence of covert phrasal-type movement. It differs from the postverbal NPs in there-construction (to which Agree without the EPP applies) in that an element is interpreted in a position higher than its spell-out position. Under an Agree approach, what would be the motivation of the movement that raises a phrase to a higher position in syntax, but has it pronounced in the lower position? Under Chomsky’s original proposal, we only have two types of movement (borrowing the terminology from Pesetsky (2000)): ff-movement that corresponds to Agree and overt phrasal movement that corresponds to Agree + EPP satisfaction. 52 However, there is no room for covert phrasal movement. Therefore, it appears that we need to further dissociate movement from choosing a copy for pronunciation: movement and choosing a copy for pronunciation are independent operations. This point is also discussed in previous studies I reviewed in chapter 1. Assuming the copy theory of movement hypothesis, previous studies have shown that movement is a syntactic operation and selection of copy for pronunciation is a PF operation. I assume this is the correct assumption. However, as I discussed in chapter 1, none of the previous analysis provides a satisfactory account for various types of low-copy pronunciation. Thus, in this chapter, I will propose an alternative principle that determines which copy is to be pronounced in a movement chain along with a system that dissociates Agree, movement and selection of a copy for pronunciation. This chapter is organized as follows: in section 2.2. I will discuss a relation between agreement and movement and propose three different types of movement. In 2.3., I will propose a PF constraint that plays a role in copy selection: the EPP as a PF constraint. In 2.4, I will propose another PF constraint with respect to copy selection. In addition, I will show how the combination of the two PF constraints captures well the varied distribution of so-called overt/covert movement. In 2.5, I will conclude this chapter. 2.2. Agreement and movement After Chomsky (2000, 2001) abandons the Spec-head relation as a local search domain for agreement, agreement under c-command has remained as the only possibility (as we discussed in 2.1). However, there are still many studies that share the intuition that movement is related with φ-feature agreement, especially in T-domain. Among many others, I will introduce Reverse Agree proposed in Wurmbrand (2012, 2014) as in (7). (7) Reverse Agree A feature F: __ on α is valued by a feature F: val on β, iff i. β asymmetrically c-commands α AND ii. There is no γ, γ distinct from β, with a valued interpretable feature F such that γ c-commands α and is c-commanded by β. 53 Under Reverse Agree, feature valuation takes place in c-command relation. This does not seem to differ from Agree system proposed by Chomsky at first glance because both systems assume that feature valuation occurs under an asymmetric c-command relation. However, the two systems differ from each other in two aspects: Chomsky assumes that interpretable features are valued but uninterpretable features are unvalued. By way of contrast, Wurmbrand, following Pesetsky and Torrego (2007), assumes that both interpretable and uninterpretable features can be either valued or unvalued. Another difference is that Chomsky assumes that Case/φ-agreement does not trigger movement by itself, whereas Wurmbrand assumes that movement is triggered by unvalued ϕ-features on a functional head, especially, T. In what follows, I will discuss how Reverse Agree works in Standard Arabic agreement by reviewing Wurmbrand and Haddad (W&H 2014). Standard Arabic has optional word order variation between VSO and SVO as in (8)-(9). In Arabic, the subject in SVO word order shows full agreement with verb. The preverbal subject agrees with the verb in person, number and gender as in (8a). Without the full agreement, the sentence becomes ungrammatical (8b). On the other hand, the subject in VSO word order shows partial agreement. In (9a), the post-verbal subject agrees with the verb in gender only. If the post verbal subject shows a full agreement pattern as in (9b), the sentence becomes ungrammatical. (8) a. l-fataja:t-u qaraʔ-na l-dars-a (Standard Arabic, W & H 2014, (1)) the-girls-Nom read-3.F.PL the-lesson-Acc The girls/They read the lesson. b. *l-fataja:t-u qaraʔ-at l-dars-a the-girls-Nom read-3.F.SG the-lesson-Acc (9) a. qaraʔ-at l-fataja:t-u l-dars-a (Standard Arabic, W & H 2014, (2)) read-3.F.SG the-girls-Nom the-lesson-Acc The girls read the lesson. b. *qaraʔ-na l-fataja:t-u l-dars-a read-3.F.PL the-girls-Nom the-lesson-Acc Wurmbrand and Haddad account for the agreement pattern in Standard Arabic using Reverse Agree. They assume that subject DP in (8) or the verb in (9) undergoes movement to be in a 54 position where it c-commands T, so values uninterpretable φ-features on T. They interpret this movement as a movement for the EPP satisfaction following Alexiadou and Anagnostopoulou (1998). The two ways of agreement are schematized in (10a) and (10b) respectively (regular arrows indicate movement; bolded arrows indicate agreement). In (10a), the subject has a full set of valued φ-features and it moves to SpecTP where it can value the full set of uninterpretable φ- features of T. Thus, in SV word order, full agreement is observed. By way of contrast, in VS word order in (10b), the subject does not move to SpecTP position. Rather, v-to-T movement takes place to value uninterpretable φ-features of T. Crucially, Wurmbrand and Haddad assume that v also bears unvalued ϕ-features, but unlike T, it only has an unvalued gender feature. When the subject is externally merged in SpecvP, it values the uninterpretable (unvalued) gender feature of v. The verb with uninterpretable, but now valued gender feature undergoes v-to-T movement and gives values to an uninterpretable gender feature of T 19 . Thus, T ends up being in an agreement relation with the subject mediated by the verb. Even after agreement between v and T takes place, however, T still has unvalued uninterpretable number and person features. Thus, default values, third person, singular, are assigned to the unvalued number and person features of T. This is why VS word order shows partial agreement (only in gender) between the subject and the verb 20 . 19 Following Pesetsky and Torrego (2007), they assume that agreement between two uninterpretable features are possible. This is contrary to Chomsky (2000, 2001), which limits possible agreement relation between an interpretable feature and an uninterpretable feature. 20 Wurmbrand and Haddad open up the possibility that the postverbal subject in (10b) further moves to SpecTP, leaving a copy behind inside vP following the copy theory of movement. They further assume that a phonological constraint in (i) forces the low copy to be pronounced. However, the motivation for the subject moving to SpecTP is not clear. Furthermore, it seems to me that the phonological rule in (i) is ad-hoc in a sense that this rule does not look like an explanation, but just a different way to describe the phenomenon. Unless other compelling evidence is provided, I assume that there is no further movement of the subject DP in (10b). (i) Agreement—word order preference: (Wurmbrand and Haddad, 2014: (50b)) *XP[ϕ] » V[ϕ:Ø] [where Ø refers to the absence of a ϕ-value] 55 (10) a. SV TP DP T’ iφ:3.F.PL T vP uφ:3.F.PL DP v’ iφ:3.F.PL V+v … uφ:F b. VS TP T vP V+v T DP v’ uφ:F uG:F iφ:3.F.PL u#:_ v … uP:_ uφ:F This approach differs from agreement under a Spec-head approach. Under the Reverse Agree system, the case feature of the subject DP is valued in the base position of the subject being c- commanded by T. The reason why a subject DP moves to SpecTP is to value uninterpretable φ- features of T irrelevant to case-valuation of the subject DP. In other words, when the subject DP moves to SpecTP, it does not bring an uninterpretable, unvalued case feature with it 21 . Rather it has an uninterpretable, valued case feature. However, the correlation between agreement patterns and word order (SVO shows full agreement while VSO shows partial agreement) is too simple to capture the whole picture of agreement pattern in Standard Arabic. One problem with this approach is realized with the overt (and covert) pronoun, as in (11). When the postverbal subject is a pronoun, not a R-expression, the verb shows full agreement (Soltan 2007a, AlAlmat 2014). According to Wurmbrand and Haddad, full agreement is the result of agreement between the moved subject DP and T and the moved phrase is forced to be pronounced in the moved position (compare to (9b)). 21 Similar ideas are proposed in Miyagawa (2010). Miyagawa argues that a DP moves to SpecTP as a consequence of φ and case agreement. The difference between Wurmbrand (2012, 2014) and Miyagawa (2010) is that Miyagawa assumes that both φ-feature and case feature agreement between T and subject DP takes place when T asymmetrically c-commands the subject while Wurmbrand assumes that φ-feature agreement takes place when the subject DP asymmetrically c-commands T and case-feature agreement takes place when T asymmetrically c- commands the subject DP. However, Miyagawa’s approach is also challenged by English there-expletives because the postverbal DP agrees with T for φ and case feature, so it is expected to undergo movement, contrary to fact. 56 (11) sharib- oo Hum al haleeb. (Standard Arabic, AlAlamat, 2014: (4)) drink- past.PL.M they the milk They (boys) drank the milk. A further complication arises if we consider that Standard Arabic is a pro-drop language. In (12), full agreement takes place even without a subject DP and we have no clue where the null subject sits. (12) (Hum) qaraʕ-uu (Hum) ʕal-dars-a (Standard Arabic, Soltan 2007a: 56) (they) read-3MP (they) the-lesson-ACC They read the lesson. Thus, I will not adopt Wurmbrand and Haddad’s particular analysis for Standard Arabic agreement. I will return to this issue and provide an alternative analysis to account for agreement patterns in Standard Arabic in 2.4. Nonetheless, the key element in the Reverse Agree system I will adopt in this dissertation seems very insightful and worthy to consider. Under Reverse Agreement system, agreement and movement takes place independently, although they are closely related with each other. Agreement takes place in asymmetrical c-command relation and by itself, it does not require any phrase to move somewhere (i.e. it does not require a specific local relation like Spec-head). However, as we discussed earlier with Activity Condition proposed by Chomsky (2000, 2001), a functional head in agreement (probe in Chomsky’s term) also has uninterpretable features which need to be valued being c-commanded by its interpretable counterpart (goal). Thus, movement of a goal takes place for the sake of valuing uninterpretable features of a probe (i.e. for the probe to be asymmetrically c-commanded by the goal) 22 . I believe that the Reverse Agree can be extended to the C-domain as well. For example, 22 It seems to be reminiscent of “Enlightened self-interest” proposed by Lasnik (1995). However, it differs from Enlightened self-interest in that movement does not take place for the sake of the goal (moving element). If this is the case, then one might wonder how the current approach can account for multiple wh-frontings in, for example, Romanian because the uninterpretable wh-feature in C can be arguably checked by the movement of the first wh- phrase eliminating motivation for the rest of wh-phrases to move. However, I assume that it is not impossible to postulate multiple uninterpretable wh-features as assuming multiple EPP requirement. Pesetsky (2000) categorizes 57 the reason why a wh-phrase moves to SpecCP in wh-fronting languages is to give values for uninterpretable wh-feature in C. I will discuss typological distinction among languages regarding agreement in C-domain in 3.3. For now, I will assume that the generalized Reverse Agree can capture agreement in general and continue the discussion how the word order variation can be explained without referring to the syntactic EPP. Slightly revising Reverse Agree, I modify the agreement condition in a way to make it looser by changing asymmetric c-command condition to simple c-command condition, as in (13). The definition of c-command I assume is in (14). (13) Condition on Agreement An uninterpretable feature α on X is valued by an interpretable (or uninterpretable, but valued) feature α on Y , iff i. Y c-commands X and, ii. there is no Z with (un)interpretable α that c-commands X and is c-commanded by Y . (α can be φ, Case, Q, wh-, focus, topic…) (14) α c-commands β if (i) α ≠ β and (ii) every γ that dominates α dominates β The first condition in (13a) implies that agreement can take place not only when Y asymmetrically c-commands X, but also when Y and X are in a sister relation. The second condition in (13b) is suggested to block (defective) intervention effects. The Condition on Agreement in (13) (and Reverse Agree proposed by Wurmbrand 2012, 2014 as well) has two important implications. First, it implies that agreement can take place in non-pair-wise fashion (Chomsky 2000, 2001). There has been a widely assumed premise that when a probe gives a value to a goal, the uninterpretable features of a probe must be valued against the goal. Such a one-to-one correspondence is generally assumed in φ-/case-agreement between T and subjects. However, the Condition on Agreement in (13) does not assume such a pair-wise agreement. Agreement is in fact basically a valuation from one element to another. Once an element with unvalued uninterpretable features enters into the derivation and it is c-commanded by another multiple wh-fronting languages into C m(ultiple)-spec languages. I believe this shares the intuition that the probe has multiple checking-requirement of the same uninterpretable features. 58 element that has a value in a corresponding attribute (and there is no element intervening between them), agreement (i.e. valuation) takes place. This opens the possibility that an element A gives a value to element B, but uninterpretable features of A get valued by a third element C. Partial agreement in VS word order in Standard Arabic is an example of this non-pair-wise agreement. In (10b), T values nominative case feature to its goal, the postverbal subject. However, the uninterpretable φ-features of T are valued, technically, against the v head, rather than the subject DP. Yet, the intuition that there is a give-and-take concept between two elements in an agreement relation seems to be still valid given that the verb shares the same values with the subject DP via previous agreement between v and the subject DP, so it appears as if T gets valued from the subject DP. In other words, a loose sense of pair-wise probe-goal relation is implemented in the Reverse Agree system (and slightly revised version in (13)). The second implication is for movement theory. Agreement takes place when a probe with unvalued uninterpretable features enters into the derivation. From the derivational perspective, the probe therefore sits in the highest position in the structure. Under the condition on agreement, the uninterpretable features of the probe will get valued by a phrase that c-commands the probe. Thus, the “value-giver” must move to a position from where it c-commands the probe, which has occupied the highest position so far. There are three potentially imaginable ways for values of a goal to be transferred to its c-commanding probe, as schematized in (15). First, if the goal YP moves to the specifier position of its probe X (15a), it can c-command its probe being able to value uninterpretable features of its probe. This is the case where common phrasal movement takes place. A second way in (15b) is that the goal phrase agrees first to a head it c-commands and the agreeing head, instead of the goal XP, moves to a position where it can give values to the probe X. Since the value of the goal YP is parasitic on the head movement, I will call this Parasitic head movement. I assume that this option is taken in Standard Arabic VSO word order following Wurmbrand and Haddad. One aspect of parasitic head movement needs to be addressed: what allows parasitic head movement? I speculate that parasitic head movement is available only if a language has a morphological apparatus that can carry the agreement information 23 . For example, parasitic head movement for subject agreement takes place when a 23 Holmberg (2000) also noted similar observations that the absence of subject-verb agreement morphology is syntactically reflected by the lack of v-to-I movement. See Holmberg (2000, pp456-457). 59 language is equipped with a full/rich set of inflectional morphologies (of person, gender, and number) on verb, so that the verb carries φ-information of the agreeing subject, as an agency. Similarly, parasitic head movement for wh-agreement is available if a language is equipped with a morphological tool to mark the wh-agreement on the hosting head. One example of such a morphological tool I argue is question particles (I will discuss parasitic head movement cases in wh-movement in 3.3). This assumption predicts that languages with a poor inflectional morphology system cannot take advantage of parasitic head movement 24 . A final way in (15c) is the phrasal movement counterpart to (15b), Parasitic phrasal movement. In Parasitic phrasal movement, the value of the goal is parasitic on movement of another phrase. A phrase KP that also needs to get its uninterpretable feature from YP gives values to a probe X. When KP finds its goal, the goal moves to a position where it c-commands the probe, KP, and agreement takes place. KP with valued features now needs to be probed by a higher probe X and KP moves to SpecXP to give values to uninterpretable features of X. Theoretically, the existence of parasitic phrasal movement in languages is plausibly imaginable given that parasitic head movement exists. Empirically, I claim that parasitic phrasal movement is an observable phenomenon in many “expletive”-constructions (e.g. there-construction, wh-expletive constructions), rather than 24 Under the current proposal, there is a subset relation between rich morphology and the possibility of parasitic head movement as in the diagram in (i). In other words, every language with parasitic head movement must have a rich set of inflectional morphologies. Since parasitic head movement generates VS word order, it predicts that every VS word order language has a rich morphology system, but not vice versa. I believe this generalization is true (see Roberts (1997) for the similar generalization). Even though I found no language with VS word order that lacks rich φ-feature agreement morphology, it is easy to find a language with rich morphology but without VS order (e.g. German). Figure 1. Subset relation between richness in morphology and parasitic head movement. Languages with rich morphology Languages with parasitic head movement 60 just one of various technical ways to resolve an empirical challenge faced by the current proposal. I will discuss that the derivation like one in (15c) is observed in many languages with (overt or covert) expletives throughout this chapter. (15) a. Goal moves to Spec of its probe XP Xprobe ZP …YPgoal … XP YP XP X ZP … YP… b. Parasitic head movement XP Xprobe … ... ZP (YPgoal) ZP Z (MPgoal) XP Xprobe … (…)+Z X ... ZP (YPgoal) ZP Z (MPgoal) c. Parasitic phrasal movement XP Xprobe … … ZP YP ZP KP ZP Z … YPgoal XP KP XP Xprobe … … ZP YP ZP KP ZP Z … YPgoal Based on the discussion so far, I claim that movement takes place for the sake of probe’s feature checking as in (16). 61 (16) Condition on Movement X substitutes a specifier of the probe Y or X adjuncts to the head of the probe Y iff (i) an uninterpretable feature of Y is valued by the movement (ii) the movement is a necessary step for some later agreement in which an uninterpretable feature of Z will be valued by a value of corresponding feature of X Wurmbrand and Haddad (2014) also propose that movement in some Standard Arabic raising construction is agreement-driven. However, they associate it with the EPP, which is rejected in the current proposal. Since Chomsky (2000, 2001) proposes the EPP as a driving-force of movement, it has been widely assumed that a subject DP moves to SpecTP to satisfy the EPP. More generally, if a functional head bears an EPP feature, the EPP feature must be checked by a phrase in specifier of the functional head or a head movement to the functional head (Alexiadou and Anagnostopoulou 1998). However, under the current proposal, as I will discuss in the next section, I argue that the EPP per se does not trigger syntactic movement. Postponing a detailed discussion to the next section, I want to point out that, under the proposed system, the EPP parameter proposed in previous studies (e.g. XP phrase in SpecTP or v-to-T movement) has nothing to do with the EPP in reality, but a parameter in the ways of feature valuation. In other words, some languages like Standard Arabic allow parasitic head movement (v-to-T movement) as a way of valuing uninterpretable features of T (Wurmbrand and Haddad 2014, cf Alexiadou and Anagnostopoulou 1998), but others do not (e.g. English). That movement is purely triggered for the purpose of feature valuation irrelevant to a traditional sense of the EPP 25 . I will discuss 25 It has been understood that Stylistic Fronting observed in Scandinavian languages or Stylistic fronting-like movement to the left periphery in German (Fanselow 2004) are instances of pure EPP-driven movement. Fanselow (2004) argues that there are two types of movement to SpecCP in German. One type is the well-known topic/focus movement to the left periphery. In that case, a phrase in SpecCP has a topic/focus reading, so this type of movement is compatible with the current proposal in a sense that movement is associated with feature checking. On the other hand, the other type of movement is not feature-driven. Fanselow argues that this movement occurs just to overtly fill SpecCP (EPP) without any association with a discourse configurational reading. The latter type of movement is constrained by a locality condition, so whatever element is most adjacent to SpecCP moves to fill it. Similarly, it is widely assumed that Stylistic Fronting in Scandinavian languages (e.g. Icelandic) moves any category indicating that the movement is irrelevant to a specific syntactic feature checking. Those examples (movement without prerequisite agreement) seem to be apparent counter-examples of the proposed system. For now, I do not have an 62 typological variation among languages in TP level (in 3.2.) and in CP level (in 3.3) under this approach. Before leaving this section, I will discuss how there-expletives can be accounted for under the proposed system. As we have discussed, in there-construction, a post-verbal DP agrees with T for φ-features and Case. Since English does not allow v-to-T movement as an option, parasitic head movement is arguably not available. I argue that the one in (15c) is what takes place in there-constructions. This means that there is base-generated somewhere lower than T and undergoes movement to SpecTP. Recently, many studies have argued that there is base generated in SpecvP, not SpecTP (Richards 2007, Deal 2009, Alexiadou and Schäfer 2011) A crucial evidence for vP-merge approach is provided by Alexiadou and Schäfer with a contrast in (17). The simple generalization that there-insertion is available in unaccusatives but not in unergatives or transitives cannot hold because only subset of unaccusative verbs allow there- insertion. Alexiadou and Schäfer argue that the theme argument of change-of-location verbs (17a) is base-generated in SpecResult(ative)P, while the theme argument of change-of-state verbs (17b) is base-generated in SpecvP. The reason why (17b) is impossible is, according to Alexiadou and Schäfer (and Deal 2009), that there and the theme argument compete for the same position, specvP, for being external-merged. (17) a. There arrived a boy (in the party) b. *There broke a window (in the classroom) explanation for it. However, I want to point out one possibility. Holmberg (2000) argues that Stylistic Fronting moves only the phonological feature matrix of a category leaving its semantic and formal features behind the base position. He further assumes that even though Stylistic Fronting only moves the phonological feature matrix of a category, the movement takes place prior to Spell-out, not at PF. What is an interesting argument in Holmberg (2000) for the current discussion is that the movement is feature-driven like other regular syntactic movement, and crucially, the movement is triggered exclusively by the needs of the target (since the moved category does not have any need to be satisfied by Stylistic Fronting). More precisely, a [P]-feature in Infl requires a phonetically overt element fill in SpecIP, so the feature triggers stylistic fronting. Since I propose a system where syntactic movement is dissociated with pronunciation, a movement driven by phonological [P] feature is not compatible with the current proposal. However, we can still take advantage of the idea that an uninterpretable feature (e.g. a categorical feature) triggers a movement (in the name of stylistic fronting) and this movement chain is pronounced with a high copy in the moved position due to an independent EPP (as a PF constraint) requirement. 63 Adopting this idea, I will assume that there is inserted in SpecvP. I further assume that there actively probes its goal due to its defective set of φ-features even though it is not a root node following Chomsky (2000, 2001) and Deal (2009). This is what happens in (15c), I elaborate in (18) with the example in (17a). There is merged in SpecvP with its defective set of φ-features (the person feature is interpretable, [iD] but others are uninterpretable) and an uninterpretable case feature (Deal 2009). It probes its goal in its c-command domain and agrees with the DP, a boy, which has a full set of uninterpretable features, but lacks a case feature. To give a value to its probe there, a boy moves to a position where it can c-command there, which is the outer SpecvP (18b) 2627 . However, since there is not a case-assigner, uninterpretable case features in both there and a boy remains unvalued. And finally, T enters into the derivation and probes its goal (18c). Crucially, I assume that T can agree with both there and a boy because both of them lack a case feature so are active as a goal and both have valued φ-features. Thus, it agrees with both DP simultaneously giving nominative case value to them. And there moves to SpecTP position where it can give a value to uninterpretable φ-features in T. This gives an impression that T in there-construction agrees with the postverbal DP. However, what actually takes place is that the parasitic φ-features on there-expletives value T 28 . 26 This cannot be done in Tucking-in fashion (Richards 1997) because it is not the case that there sits in SpecvP via movement. There sits in the inner specifier as an external argument. 27 One might argue against the existence of movement of unaccusative subjects to SpecvP because English does not allow object shift. However, the lack of object shift in English can be explained by the lack of verb movement in English (Holmberg’s generalization). As we saw in chapter 1, Bobaljik argues that object shift in Scandinavian languages always takes place, but when the verb does not move to Infl, the object chain must be pronounced with the low copy because overt realization of the high copy of the object chain interrupts an adjacency requirement between Infl and verb in PF. If this is what happens in English as well, objects in general can move to SpecvP to value uninterpretable φ-features of v, but the lower copy of the chain is forced to be pronounced due to the lack of verb raising. 28 I assume that there is pronounced with a high copy due to the EPP requirement, while the associate is with a low copy due to the lack of the EPP requirement. 64 (18) a. vP there[iD/uφ:_/uCase_] vP v ResultP a boy[iφ/uCase_] Result b. vP a boy[iφ/uCase_] vP there[iD/uφ:√//uCase:_] vP v ResultP a boy Result c. TP T[uφ:_/Nom] vP a boy[iφ/uCase:Nom] vP there vP v ResultP a boy Result d. TP there TP T[uφ:√/Nom] vP a boy vP there vP v ResultP a boy Result 65 However, there is one question not resolved in this approach. As shown in (19), why is it impossible to move a boy instead of there? (19) *A boy there arrived (in the party) I do not have an answer for this right now. I will leave this issue for future work making a note of two possibilities. One possible reason is a movement constraint. The postverbal DP a boy already undergoes movement for φ-agreement. It has been widely assumed that a phrase that undergoes movement is frozen there and cannot move further. Another possible stipulation is that T selectively chooses there over a boy because it requires a definite phrase. There is assumed to have D feature which definiteness is situated in and it is well known that postverbal DP in there construction must not be definite (Definiteness Restriction). In this section, I discussed the agreement system and the motivation of movement. The approach I follow neither requires Spec-head configuration for every occurrence of agreement nor allows a structurally lower element to be able to value a structurally higher element. Agreement strictly takes place in the configuration where a “value-giver” c-commands a “value- receiver” and movement takes place for the value-giver to be in a proper position for agreement. As I discussed, however, pronouncing a copy in a movement chain is another story. Choosing a copy for pronunciation at PF requires its own principle having nothing to do with agreement or movement. I will discuss how selection of copy for pronunciation is determined in PF in the following sections. 2.3. EPP as a PF constraint I adopt the copy theory of movement along with the agreement condition in (13). Agree can take place at a distance (contra Nunes (1995, 2004, 2011)) if it satisfies the c-command requirement, so every copy has the same property with respect to syntactic features as in (20c). (20) John arrived a. [ VP arrived John[uCase: ]] b. T[uφ] [VP arrived John[uCase: Nom]] 66 c. [ TP John[uCase: Nom] T[uφ] [VP arrived John[uCase: Nom]] Even without any arbitrary operation that unifies members in a chain by deleting features of a subset of the chain (Chomsky 1993, Nunes 1995, 2004, 2011), the uniformity of members is a natural consequence of agreement at a distance. The uniformity of chain implies that there should be no chain-intrinsic reason to choose one copy over another for pronunciation. In other words, since every member in a chain shares the same property as identical copies, there should be no reason, just by looking at copies, for a certain member to have advantage to be selected for pronunciation. Furthermore, if we assume that pronouncing a copy is determined in PF, not in Syntax, the question what determines which copy to be pronounced will be answered by investigating PF constraints. One PF criteria that plays a role in selection is (morpho-)phonological constraints discussed in Bošković & Nunes (2007) and Bobaljik (2002) (which I reviewed in chapter 1). However, when there is no such a phonological constraint, what kind of PF constraints play a role? For example, in the sentence in (20), why does English grammar choose the high copy in SpecTP to be pronounced? If we take a closer look at the chain, even though the copies are identical being copied from one another, the position each copy occupies differs. Thus, it sounds plausible that selection of a copy for pronunciation is related to the property of the position where copies occur (by looking at its neighbors). Making use of such an idea, I will argue that the EPP plays a role in selection. More precisely, I will claim that the EPP is not a syntactic constraint that triggers movement, but a PF constraint that requires a copy in a certain position to be phonetically spelled-out. The EPP requirement as a PF constraint is the reason why the subject in (20) and subjects in general are pronounced with a high copy. Furthermore, adopting the notion of generalized EPP, I will argue that pronouncing an Ā-chain is also determined by the EPP in left periphery. This induces the illusion of high copy privilege in pronunciation because many movement chains are subject to the (pronunciation-wise) EPP requirement. In other words, the seemingly high copy privilege has nothing to do with a special property in a high copy by itself. 67 In this section, I will review McFadden and Sundaresan (2015) and Landau (2007) as representative studies arguing for the EPP as a PF constraint 29 . Both studies share the intuition that the EPP is a requirement that the SpecTP position must be overtly filled. Based on them, I will propose a PF constraint Pronounce EPP later. McFadden and Sundaresan (2015) argue that the EPP has both phonological and syntactic nature. The nature as a PF constraint is observed in (21)-(23). In (21), the main verb is a “weather”-predicate. This verb does not assign any thematic role to its argument. Thus, a semantically null DP, the expletive it, is recruited to satisfy the EPP as in (21). The ungrammaticality of (21b) is explained in any theory of the EPP because there seems to be nothing in SpecTP. However, McFadden and Sundaresan cast a question why (21c), where SpecTP is syntactically filled but phonetically null, is not an option in English although this is an available derivation in other languages (e.g. Italian). McFadden and Sundaresan state that (21c) shows that English EPP requires DP in SpecTP to appear in an overt form. In other words, it is not enough to fill something in SpecTP syntactically. SpecTP must be phonetically realized. (21) a. It is raining. b. * Is raining. c. * pro is raining. Similarly, the contrast in (22a) and (22b) shows the nature of the EPP as a PF constraint. In English, sub-extraction out of a moved element is generally banned, as shown in (22a). However, sluicing seems to rescue the extraction, as in (22b). Merchant (2001) argues that the extraction of the wh-phrase which (Marx brother) out of the subject is legitimate in (22b) because in that case, the subject never moves to SpecTP, but stays in its base position inside vP as presented in (22c). Merchant assumes that the EPP is a PF requirement, so when sluicing deletes SpecTP, the EPP requirement disappears as well. Thus, the subject DP can stay in its base position allowing sub- extraction out of it. If the EPP is a syntactic requirement, (22b) would violate the EPP requirement, and so render the sentence ungrammatical, contrary to facts. Following Merchant, 29 For other studies that are in the same vein, see Holmberg 2000, Merchant 2001, Bobaljik 2002, Takahashi 2002. 68 McFadden and Sundaresan argue that the contrast between (22a) and (22b) shows that the EPP applies at PF. (22) a. * Which Marx brother is [a biography of which Marx brother] going to appear [a biography of which Marx brother] this year? b. A biography of one of the Marx brothers is going to appear this year, but I don’t know which (Marx brother). c. A biography of one of the Marx brothers is going to appear this year, but I don’t know [CP [which (Marx brother)] is [ TP going to appear [a biography of which (Marx brother)] this year]]. To account the PF characterization of the EPP, they propose the Constraint on the left edge of IntP in (23). (23) Constraint on the left edge of IntP (Prosodic EPP): The left edge of Intonational Phrase (IntP) must be marked by the presence of overtly pronounced material. (McFadden and Sundaresan 2015, (24)) Following An (2007), McFadden and Sundaresan assume that TP typically starts a new Intonational Phrases (IntP). Therefore, (23) forces that the left edge of TP, as the left edge of an intonational phrase, must be marked with an overtly pronounced material. They argue that this analysis captures the peculiar behavior of that-trace effects. When the object wh-phrase is extracted out of the subordinated clause in English, the appearance of the complementizer that is optional, as in (24a). By way of contrast, when the subject wh-phrase is extracted as in (24b-c), the complementizer must be absent. (24) a. Who do you think [(that) Beau nudged who]? b. Who do you think [who is stupid]? c. * Who do you think [that who is stupid]? 69 McFadden and Sundaresan assume that when there is no overt complementizer, the embedded TP, even if it is a finite one, does not form an intonational phrase. Thus, when the overt complementizer is absent as in (24b), the absence of an overt element in the left edge of TP is tolerable. On the other hand, if there is an overt compelmentizer as in (24c), the embedded TP forms an intonational phrase, so its specifier position must be filled to satisfy the pronunciation- wise EPP. Interestingly, that-trace effects observed in (24c) suddenly disappear if sluicing takes place as in (25a). Not only that, the that-trace effects disappear as well when an adverbial intervenes between the complementizer and the presumed position of the trace as in (25b). (25) a. John said that someone would write a new textbook, but I can’t remember who [John said that who would write a new textbook]. b. Who do you think [that, against doctor’s orders, who drank the hot sauce]? They give an explanation for the sudden amelioration of that-trace effects in (25) with the Prosodic EPP in (23). In (25a), since the embedded TP is erased in PF due to sluicing, the Prosodic EPP does not apply. In (25b), the adverb that can overtly mark the left edge of TP satisfies the Prosodic EPP instead of subject. Thus, neither case in (25) violates (23). On the other hand, McFadden and Sundaresan still assume that the EPP has a syntactic nature as well. In (26), only the sentence where the entire subject DP is pied-piped is grammatical (26a). When a part of the subject DP only moves as in (26b-d), the sentence becomes ungrammatical. Based on this observation, they argue that the movement that satisfies the EPP must target a syntactically defined constituent. (26) a. [DP1 [DP2 The man] [PP with the pointy green hat]]i seemed [ TP ti to be insane]. b. * [DP2 The man] i seemed [ TP [DP1 ti [PP with the pointy green hat]] to look insane]. c. * [N man]i seemed [ TP [DP1 [DP2 the ti ] [PP with the pointy green hat]] to look insane]. d. * [D the]i seemed [ TP [DP1 [DP2 ti man ] [PP with the pointy green hat]] to look insane]. The contrast in (27) is also provided as evidence of syntactic nature of the EPP. In (27b), the object moves to SpecTP leaving the subject in-situ violating locality conditions. The object 70 movement to satisfy the EPP violates locality condition, so the sentence becomes ungrammatical, contrary to (27a) where locality conditions are obeyed. McFadden and Sundaresan argue that the contrast in (27) shows that movement to satisfy the EPP is subject to syntactic locality conditions. (27) a. [DP1 The hungry man]i will ti devour [DP2 the soup]. b. * [DP2 The soup]i will [DP1 the hungry man] devour ti. McFadden and Sundaresan point out that a syntactic movement (the Syntactic nature of the EPP) is required to satisfy a phonological constraint (the Phonological EPP). However, they do not provide any conclusive mechanism that accounts for the counter-cyclicity of the EPP, and rather suggest some possibilities 30 . What is important in their account for the current discussion is that the EPP needs to be understood as a PF requirement. The evidence they provide to support the syntactic nature of the EPP can be interpreted in other ways: a syntactic movement that is triggered by an independent reason (not by the EPP) targets SpecTP giving an illusion that the EPP triggers movement. I will adopt the intuition that the EPP is a PF constraint but not follow the claim that the EPP has a syntactic nature. Instead, I will assume that the EPP has nothing to do with “triggering movement”, but is just a PF requirement to have an overt element in a designated position. Furthermore, to incorporate the idea of a generalized EPP (see below), I will not adopt their prosody-based analysis (the Prosodic EPP) that is restricted to the TP domain. In a similar vein, Landau (2007) argues that the EPP is a pure PF constraint. The key aspects of the EPP Landau proposes is in (28). (28) Aspects of EPP (Landau 2007, (2)) a. EPP is a selectional feature governing PF configurations. 30 Three possible mechanisms McFadden and Sundaresan suggest are summarized in (i). (i) a. Syntax communicates with PF. b. Syntax overgenerates structures according to the syntactic constraints and later PF filters them out. c. Movement to SpecTP is motivated for an independent reason leaving the EPP as a pure PF constraint. The last option (ic) is what I argue for in this dissertation. For the pros and cons of other options, see McFadden and Sundaresan (2015, section 7). 71 b. Every functional head may bear an EPP feature. Since Chomsky (2000), the second aspect is widely assumed in the name of generalized EPP. I will assume that the second aspect in (28b) is true. What needs more discussion is the first aspect in (28a). Landau argues that the first aspect derives the parasitic nature of the EPP on some independent syntactic features. Following Holmberg (2000), Landau assumes that the EPP is simply a feature [P] and a [P] bearing head selects an element that satisfies the [P] feature 31 . Crucially, the [P] feature does not take part in agreement by itself, like s-selectional features (e.g. [+animate]) rendering itself parasitic on other features. The parasitic nature of the EPP is easily found. A subject that satisfies [P] feature of T in English also checks case and φ-features against T. In wh-movement, wh-phrases that satisfies [P] feature of C also undergoes wh-feature checking against C. Landau calls the hosting feature that the [P] feature is parasitic on an “anchoring feature”. Landau claims that the particular association between [P] feature and the anchoring feature is parameterized across languages. For example, a [P] feature in T is anchored to nominative case feature in English, so SpecTP must be filled by a subject DP, whereas it is anchored to verbal agreement in null subject languages (e.g. Greek), so v-to-T movement satisfies the EPP (Alexiadou and Anagnostopoulou 1998). Or a [P] feature in T is anchored to [wh] or [focus] feature in focus-prominent languages like Japanese (Miyagawa 2001, 2005). The generalization Landau derives from this cross-linguistic variation is that checking of a [P] feature always requires another formal feature checking to attract a category to the local domain of functional head that has the [P] feature. The point that [P] features does not trigger movement is supported by the ungrammaticality in (29). In (29a), Mary moves from the indirect object to the direct object position to express the reading “We showed Mary to herself”. If we assume that a [P] feature alone can trigger movement, (29a) should be fine contrary to fact. The ungrammaticality of (29a) is accounted for with case theory. The illicit double case checking of 31 Since [P] feature is a selectional feature, Landau (2007) argues that [P] feature checking must obey conditions in (i). Discussion of syntactic conditions on selection serves the heart of the paper. However, I will not discuss this issue in depth for the sake of space. (i) Syntactic conditions on selection (Landau 2007, (3)) a. Locality: A head may only select elements in its complement/specifier. b. Headedness: A head may only select the head of its complement/specifier. 72 Mary results in the ungrammaticality. In other words, without an anchoring feature checking (this case, structural case checking), a pure [P] feature by itself cannot trigger movement. Similarly, in (29b), if we assume that a [P] feature can trigger a movement of Jane to satisfy the EPP of the matrix T independent to case checking, the sentence in (29b) is expected to be grammatical, contrary to fact. These examples show that the role of [P] feature is to make an element that moves for independent syntactic cause (anchoring feature) be overtly realized. (29) a. *We showed Mary to Mary. b. *Jane seemed to Jane to be fortunate. (Landau 2007, (59)) To explain the parasitic nature of [P] features, Landau suggests that PF is a filtering device that rules out illicit derivations that are independently generated by syntax. Thus, syntactic movement is triggered by independent syntactic features and any derivation that does not have an overt phrase in the spec position of a functional head with a [P] feature is filtered out in PF. According to Landau, this filtering system gives the impression that the EPP feature is parasitic. Following Landau, I claim that the EPP is a pure PF constraint. This view is more radical than McFadden and Sundaresan (2015)’s approach because it contradicts the widely-assumed views that the EPP triggers syntactic movement. I argue that this view is misled by the parasitic nature of the EPP. However, in Landau’ account, it is not clear how the EPP feature can be parasitic on its anchoring feature. Revising Landau’s idea, I propose the EPP feature assignment in (30) and pronunciation rule in (31). (30) EPP feature assignment EPP feature is assigned to an uninterpretable feature α on a functional head X. (31) Pronounce EPP Spell-out the value of EPP-hosting feature α on X by (i) Spelling-out the interpretable counterpart of α by pronouncing the phrase YP that values X where valuation takes place or (ii) Inserting morphology that represents values of α. 73 The intuition behind (30) and (31) is that the EPP feature does not have (morpho-)phonological form to realize, but it is inserted into derivation to make its hosting feature overtly realized. There is no intrinsic reason for a syntactic feature to be phonetically realized. For example, values of person and number feature (e.g. 3PL) on verb do not have phonetic force. If it is, it is expected that every language has inflectional morphology contrary to facts. Inflectional morphology on verbs appears in some, but not all, languages and it means that if a formal feature is phonetically realized, then the phonetic realization of a formal feature is totally caused by an external, not feature intrinsic, requirement. I suggest that the external requirement is an interface requirement. Let me elaborate on this idea further. It has been assumed that there are two types of features: interpretable features and uninterpretable features. Interpretable features are required for interpretation in LF. Thus, the existence of interpretable features is approved by the need of the LF interface. Thus, even if there is no PF-side reason for the existence of interpretable features, interpretable features must still be postulated for the purpose of interpretation. By way of contrast, uninterpretable features do not contribute to semantic component. They are even not interpretable in LF as the name indicates. In syntactic theories, therefore, uninterpretable features have been assumed to exist for solely syntactic reasons: uninterpretable features have been assumed to exist in order to motivate syntactic movement. However, in the minimalist program, it is desirable that any syntactic input must have an interface-driven motivation. Therefore, I argue that the existence of uninterpretable features in grammar must be motivated by the PF component. In other words, uninterpretable features are expected to have a PF consequence of its presence. When an uninterpretable feature comes into a derivation, grammar needs a means for the uninterpretable feature to affect PF. I claim that assigning the EPP feature is the way grammar chooses to have PF consequence of uninterpretable features. The EPP feature is assigned to uninterpretable features and Pronounce EPP forces the hosting uninterpretable feature to be overtly pronounced due to the parasitic nature of the EPP. I suggest that the values of uninterpretable features can be spelled-out by either morphology insertion or pronouncing a phrase with corresponding interpretable feature adjacent to the head with the EPP. For example, the EPP is assigned to uninterpretable φ features of T and Pronounce EPP is satisfied by morphology insertion in languages with rich morphology system (e.g. Standard Arabic, Greek) 74 while it is satisfied by spelling-out a high copy (in SpecTP) in subject chains in English 32 . Likewise, uninterpretable wh-features in C with the EPP can be spelled-out by inserting a wh- agreement morphology in some languages or by pronouncing a high copy in a wh-chain in other languages. The PF consequence of morphology insertion is obvious in a sense that uninterpretable features introduce morphemes into the overt form. Then, what would be the PF consequence of pronouncing high copy adjacent to the head with uninterpretable features? As I discussed, syntactic movement takes place for the benefit of the target. By pronouncing its agreeing goal in the moved position, otherwise invisible syntactic movement becomes visible in PF. Thus, in this case, the phonological consequence of an uninterpretable feature with the EPP is in marking movement chain visible at PF (i.e. overt movement). This is the point from which the EPP has been (mis)understood as a requirement of “filling SpecTP”. However, under the present system, the requirement of overtly filling the specifier position of a functional head is nothing but one type of Pronounce EPP satisfaction. The way of satisfying Pronounce EPP is parameterized by languages and by types of features. Let me briefly discuss how (30) and (31) couple with conditions on agreement/movement proposed in earlier work in English. In English, the EPP feature is assigned to uninterpretable unvalued φ-features of T. When T enters into a derivation, it probes its goal due to its uninterpretable features. T finds its goal, subject DP, so the subject DP moves to SpecTP to give values to uninterpretable φ-features in T. Now, there are two copies of the subject DP, one in its base position and the other in SpecTP. PF constraint, Pronounce EPP, forces the feature EPP to be parasitically pronounced on the hosting feature. Pronounce EPP is satisfied by pronouncing the high copy of the subject DP that gives value to uninterpretable features of T in SpecTP. Even though English shows the number-agreement morphology, the rest of φ-features cannot be spelled-out by morphology insertion due to the lack of inflectional morphology 32 Morphological richness is necessary for applying morphology insertion to Pronounce EPP. However, it is not the case that every language with rich morphology system allows morphology insertion as a means of satisfying Pronounce EPP. For example, French has a rich morphology inventory, but it does not allow morphology insertion as a means of satisfying Pronounce EPP. 75 available. Thus, the option that the subject DP is pronounced with its high copy must be selected to satisfy Pronounce EPP 33 . I proposed that there are three different forms of movement. Given that, one might wonder whether there is any difference in the ways of satisfying Pronounce EPP depending on types of movement. My answer is yes. It has been long attested that phrasal movement and head movement show different properties. Among many other differences, one crucial difference is that only in head movement, the probe and the target behave as one constituent as a result of head movement (Mahajan 2000, Matushansky 2006). Due to the peculiar property of head movement, Matushansky (2006) argues that when head movement takes place, morphological merge (m-merge) between the goal and the target takes place making them a morphologically single unit. Whatever the mechanism is going on, this special property of head movement has two important implications on parasitic head movement. First, when parasitic head movement takes place, a high copy must be pronounced. There must be no case where a low copy is pronounced in a head movement chain. This point predicts that there should be no occurrence of “covert head movement” across languages. This is contrary to phrasal movements given that there have been many discussions regarding covert phrasal movement across languages. As far as I acknowledge, no instance of covert head movement has been reported. A second implication is that, in parasitic head movement cases, since pronouncing a high copy is required independent to Pronounce EPP, morphology insertion is instead forced to occur to satisfy Pronounce EPP. This well accounts for the close association between head movement and morphological richness (Alexiadou and Anagnostopoulou 1998, Holmberg 2000). The current proposal associating high-copy pronunciation to Pronounce EPP well captures the intuition in Nunes (1999, 2004, 2011). Nunes’ proposal has an intuition that high 33 Maria Luisa Zubizarreta (p.c) commented that there appears to be an asymmetry between A-domain and Ā- domain in the way of satisfying Pronounce EPP. As we will see in subsequent chapters, it is easy to find Ā- agreement cases where Pronounce EPP is satisfied by lexical/intonation morpheme insertion while it is hard to find A-agreement cases where such lexical/intonation morpheme insertion satisfies Pronounce EPP across languages (note. the morphology insertion to satisfy Pronounce EPP in A-domain is realized as inflectional morpheme insertion though). In addition, the adjacency between the subject DP and its agreeing functional head T seems to be more required than the adjacency between any two elements agreeing in Ā-domain. The reason for such an asymmetry is not clear to me in the current stage. I will leave this issue for future works. 76 copy pronunciation is a consequence of agreement. More specifically, Nunes’ Chain Reduction algorithm derives high copy privilege from the assumption that agreement takes place in Spec- head relation and there is an asymmetry in feature checking among members in a chain. However, as I have discussed, the assumption that agreement takes place in Spec-head relation does neither fit the current theoretical assumption in Minimalist Program nor capture many occurrences of agreement that do not trigger overt movement, but a covert movement. Empirical data we have discussed shows that high copy pronunciation is closely related with agreement, but must be dissociated with it. By way of contrast, assuming a parasitic nature of the EPP, the current proposal still captures the intuition that high copy pronunciation is correlated with agreement without being challenged by theoretical drawbacks and furthermore well accounts for some data where agreement takes place but a low copy is phonetically pronounced. As I discussed, Pérez (2015) also proposes an analysis to capture high-copy privilege. However, since Pérez’s account assumes that pronouncing a copy is decided without any syntactic information, it does not capture the close relation between overt movement and agreement. Before closing this section, there is one point that I want to make clear. One important property of Pronounce EPP, is that it is basically an economy condition, not a convergent condition. The premise that introducing formal features into derivation needs a consequence in interface components implies that the introduction of formal features without any consequence in interface components is a non-economical step in a sense that an operation takes place that is semantically or phonetically vacuous 34 . One crucial consequence of the economy conditional 34 Output Economy proposed by Fox (1998) in (i) shares the similar intuition even though Fox proposes Output Economy for optionality between covert/overt movement. The Output Economy has two sub-economy conditions in (ii). In my view, the intuition that syntactic operations need a consequence in LF (iia) or PF (iib) is in the same spirit that Pronounce EPP is based on: when there is a PF or LF consequence, introducing formal features into Syntax is allowed. (i) Output Economy: Fox (1998: 75) Optional operations can apply only if they have an effect on outcome. (ii) a. Scope economy (Fox 1998: 75) Covert optional operations (i.e., QR and QL) cannot be scopally vacuous (i.e., they must reverse the relative scope of two non-commutative quantificational expressions). b. Word order economy (Fox 1998: 75) 77 property of Pronounce EPP is that it is violable when a derivation that obeys Pronounce EPP violates convergence conditions. In that case, a convergent, but a less economical derivation is selected. I will discuss such cases in Chapter 5. In summary, I provide an alternative answer for the question why a high copy seems to be privileged for pronunciation in general (i.e. why overt movement (in traditional sense) outnumbers covert movement in languages). The answer to this question I suggest is that a high copy is forced to be phonetically realized due to Pronounce EPP constraint. The proposal implicates that high copy privilege is not based on chain-intrinsic reasons. Thus, it opens the possibility that when Pronounce EPP is satisfied by other means (e.g. morphology insertion), the high copy privilege disappears. This implication is well compatible with the apparent lack of high copy privilege discussed in 1.3. In addition, since the selection of copy for pronunciation takes place in PF, this system does not need to assume a PF filter (contra Landau). Yet, I have to address one more issue to complete the big picture: it must be clarified how PF decides which copy to pronounce if a chain is free from the requirement of satisfying Pronounce EPP. I will address this issue in the next section. 2.4. Cyclic Spell-out and low copy privilege In this section, I will propose another pronunciation principle applied to a movement chain. The representative empirical data concerned is English multiple wh-constructions, as in (32). As I discussed in 1.3.2, in multiple wh-construction in superiority obeying word order (32a), the in- situ wh-phrase which book shows a “high” behavior as if it undergoes movement to where it can license ACD. This is the reason why Pesetsky (2000) claims that the in-situ wh-phrase in (32a) undergoes covert phrasal movement. If we understand this data under the copy theory of movement, this is the case where the movement chain of which book is pronounced with a low copy. One question naturally follows is whether it is possible for the high copy, instead of the low copy, to be phonetically realized. Since every copy in a chain is identical, if there is no external requirement, it is expected that either copy has equal chance to be pronounced in Overt optional operations cannot be string vacuous (i.e. they must reverse the relative order of the two [perhaps phonologically overt] expressions). 78 principle. However, this prediction is not turn out to be true in English multiple wh-questions. Contrary to the first wh-phrase, if the chain of the second wh-phrase is pronounced with the high copy in SpecCP, as in (32b), the sentence becomes ungrammatical. In other words, the second wh-chain must be pronounced with its low copy. (32) a. Which person read which book? b. *Which person which book (did) which person read which book? Another example of obligatory low-copy pronunciation is observed in Standard Arabic we discussed. We have discussed that in Standard Arabic, (it seems that) VSO order shows partial agreement while SVO order shows full agreement. However, surprisingly, pronoun subjects in this language always trigger full agreement on verb regardless of its relative order with verb. Furthermore, it has been argued that the preverbal subject in SVO is in Ā-position via CLLD or topic movement. Soltan (2007a) argues that SV is taken to represent Topic-comment structures, while VS order is interpreted with default, non-topic-comment structures. This topic-nature of preverbal subject in SV order is supported by the ban of indefinite nonspecific NPs in preverbal position, as shown in (33). Following Soltan (2007a) and Alexiadou and Anagnostopoulou (1998), I assume that the subject in SVO word order is not in canonical subject position, SpecTP. Rather, it is a topical element that is base-generated in Ā-position. (33) a. *walad-un kasara l-baab-a boy-NOM broke 3sgms the-door-ACC b. kasara walad-un l-baab-a broke 3sgmas boy-NOM the-door-ACC Given those observation, following Soltan (2007a), I assume that full agreement is an indication of (null) pronouns rather than a consequence of SV order. To account for agreement patterns in Standard Arabic, Soltan argues that the EPP appears only in SV, not in VS order. This is contrary to Alexiadou and Anagnostopoulou’s view of the same phenomenon. Alexiadou and Anagnostopoulou also argue that the Ā-property of preverbal subjects in SVO word order indicates that SpecTP is allowed to be phonetically empty and the EPP as requirement of overtly 79 filling SpecTP does not exist in Standard Arabic. They further argue however, that it does not mean that Standard Arabic does not have the EPP feature in T-domain. Rather, it has the EPP requirement like English, but the EPP requirement is satisfied by a different means, namely head movement of verb to inflectional domain. Revising Alexiadou and Anagnostopoulou’s intuition on the EPP according to the current proposal, I provide an alternative analysis for each agreement pattern in Standard Arabic as in (34). The generalization obtained in (34) is that full agreement is a result of agreement between (overt/covert) pronoun and T and partial agreement is a result of agreement between verb and T (in a sense proposed by Wurmbrand and Haddad). In (34a), since pro gives a value to uninterpretable φ-features of T from SpecTP position and subject is base-generated in TopP above TP, the preverbal subject cannot be understood as satisfying the EPP by filling in SpecTP. In (34b), I assume that subject that is base generated in SpecvP stays there. In this case, the verb moves to T to give partial set of values to uninterpretable φ-features of T. In (34c), the subject pronoun is base generated in SpecvP and moves to SpecTP to give values to uninterpretable features of T. It further moves to SpecTopP for topicalization. The final case (34d) shows the most interesting pattern. Since the verb shows full agreement, it is expected that the post-verbal subject pronoun gives a full set of values to uninterpretable φ-features of T. To give values, the pronoun must move to SpecTP position from where it c-commands T, due to the Condition on Agreement I proposed in (13). This means that the pronoun subject has a chain that has two members, one in SpecTP and the other is in the base position, SpecvP. Interestingly, in this case, the low copy is pronounced instead of the higher copy 35 . 35 This analysis accounts well for the ungrammaticality of full agreement in VS order and partial agreement in SV order in (8) and (9), I repeated here in (i) and (ii). (i) a. l-fataja:t-u qaraʔ-na l-dars-a (Standard Arabic, W & H 2014, (1)) the-girls-Nom read-3.F.PL the-lesson-Acc The girls/They read the lesson. b. *l-fataja:t-u qaraʔ-at l-dars-a the-girls-Nom read-3.F.SG the-lesson-Acc (ii) a. qaraʔ-at l-fataja:t-u l-dars-a (Standard Arabic, W & H 2014, (2)) read-3.F.SG the-girls-Nom the-lesson-Acc The girls read the lesson. 80 (34) Agreement in Standard Arabic a. SR-expressionVO + full agreement : pro agrees with T and Subject is inserted in TopP b. VSR-expressionO + partial agreement : verb agrees with T c. SPronounVO + full agreement : Pronoun agrees with T and undergoes topic movement d. VSPronounO +full agreement : Pronoun agrees with T without any further movement There are two parametric difference between English and Standard Arabic. One is that English does not allow parasitic head movement while Standard Arabic allows it. The other is that English does not have a rich set of agreement morphemes, contrary to Standard Arabic. These two parametric differences between English and Standard Arabic has two independent consequences. First, the lack of parasitic head movement in English means that English does not have an option that v gives a value to uninterpretable φ-features of T. Only a phrasal movement is allowed for valuation on T 36 . This blocks the possibility that Pronounce EPP is satisfied by pronouncing (a copy of) the verb on T, resulting in the lack of VS order (unless there is an independent v-to-T movement). Second, the difference in morphological richness is eventually related to the availability of morphology insertion as an option to satisfy Pronounce EPP. Due to the lack of a rich set of (inflectional) morphology in English, English cannot help pronouncing the EPP by selecting a high copy of a phrasal movement. Thus, the “value-giver”, the subject DP, b. *qaraʔ-na l-fataja:t-u l-dars-a read-3.F.PL the-girls-Nom the-lesson-Acc The ungrammaticality of (ib) is explained as follows: In SV order, the subject does not value the uninterpretable features in T. Instead, a null pronoun gives a value to T. As shown above, a pronoun always triggers full agreement. Thus, the partial agreement in SV order is ungrammatical. The ungrammaticality of (iib) is explained as follows: In VS order, the subject never moves, so to get a full agreement, a null pronoun is necessary. However, since the subject DP already occupies SpecvP, the null pronoun cannot be introduced. This is contrary to the SV order. In SV order, the subject is not base generated in SpecvP, but in SpecTopP, the pronoun can be freely generated in SpecvP. 36 It is important to distinguish parasitic head movement from other types of v-to-T movement in general. Verb raising can be triggered by many different reasons. Parasitic head movement is one type of verb raising triggered by the purpose of φ-valuation to T. However, it does not necessarily mean that every v-to-T movement language allows parasitic head movement. For example, French allows v-to-T movement, but the language does not allow parasitic head movement. 81 must be pronounced with a high copy in the position where valuation takes place 37 . In other words, the overt displacement of the subject is the only possible way to satisfy Pronounce EPP in English. On the other hand, in Standard Arabic, inflectional morpheme insertion is available as an option to satisfy Pronounce EPP 38 . It opens up the possibility that when Pronounce EPP is satisfied by morphology insertion, a moved phrase is free from the requirement of being spelled- out with its high copy. In that case, there should be no preference among copies for pronunciation in principle. Either a high copy or a low copy in principle could have a chance to be spelled-out. However, preverbal subjects are not in SpecTP, so SV order cannot be understood as a case where a high copy of subject chain is pronounced. In addition, unless it is further topicalized, a pronominal subject is pronounced postverbally as in (34d). It indicates that low- copy pronunciation is the option. Based on this observation, we can generalize that a low copy is privileged for pronunciation in Standard Arabic due to the availability of morphology insertion as a way of satisfying Pronounce EPP. This low copy privilege is somewhat surprising because a prevalent assumption among previous studies is that a high copy in a movement chain is privileged for pronunciation. Once we get rid of the noise of the EPP, the picture becomes clearer showing that, in actual fact, a low copy rather than a high copy, is forced to be pronounced. 37 One question that naturally follows is why languages including English redundantly use more than one way to pronounce the EPP. Even though I do not have a conclusive answer to this question, I believe the redundancy is the nature of expressing information about agreement. For example, some East Asian languages do not show subject- agreement inflectional morpheme on V . This means that overt realization of agreement on V is not necessary to convey information about agreement. In other words, all the information about the φ-features of subject T agrees with can be inferred from the subject DP. However, still, many languages overtly express the information about φ- agreement on the verb redundantly. The redundancy in overt expression of a feature is observed in many different ways. For example, [+focus] on wh-phrase in English is enough to be realized by wh-morpheme. However, in Japanese, a wh-phrase must bear focus accent on it (Ishihara 2002, 2007, Deguchi & Kitagawa 2002, Féry and Ishihara 2010 among many others). I will leave the issue of redundancy for future works. 38 One might wonder why then morphological insertion is chosen over the option that forces the high copy in SpecTP to be pronounced. I believe that pronouncing the high copy has a burden to violate another economy condition, Earliness under Cyclicity in (35). Thus, I stipulate that if a language allows one option to be enough to satisfy Pronounce EPP and morphology insertion is an available option, then only morphology insertion is chosen as a means to satisfy Pronounce EPP. I will elaborate this point with typological discussion in chapter 3. 82 I generalize that a low copy is privileged in pronunciation when there is no other PF requirement (for example, Pronounce EPP) that forces a high copy to be pronounced. In other words, the low-copy pronunciation is the default pronunciation. Why should it be? To answer this question, I propose Earliness under Cyclicity in (35). (35) Earliness under Cyclicity Pronounce the first copy transferred to PF. This earliness principle is based on cyclic spell-out (Chomsky 2001). The gist of Cyclic spell-out is that multiple applications of phonetic Spell-Out transfer each chunk (which is assumed as a Phase) of syntactic information to PF rather than syntax being transferred to PF all at once. Based on Phase theory, the spell-out domain is the complement of the head of the phase, which is C and transitive v according to Chomsky. Thus, spell-out domains in a mono-clause sentence are TP and VP as demonstrated in (36). (36) Cyclic Spell-out (Chomsky 2001) CP Spell-out C TP T vP Spell-out v VP Chomsky proposes the Phase Impenetrability Condition (PIC), which says that once a phase has been completed, the spell-out domain of the phase is transferred to the interfaces. Elements that have been spelled-out are not accessible to further operations. Only the phase head and its specifier (edge position in Chomsky’s term) are accessible to further operations. What is important for the current discussion is that the spell-out domain of each phase is transferred to PF cyclically as the derivation proceeds. The conceptual motivation of Earliness under Cyclicity principle in (35) is Economy. Assuming the syntactic derivation is cyclically transferred by phase from the bottom, different occurrence of copies in the same chain will be transferred to PF at different time slots. From the 83 perspective of PF, when the first copy is transferred to PF, there is no need to choose which copy to pronounce because there is simply just one copy at that time. As the derivation proceeds, if a copy of the same chain is transferred to PF, then PF needs to decide which copy to pronounce. As more spell-out takes place, PF needs more information in comparison among copies for pronunciation. Thus, pronouncing the first copy transferred to PF will simplify the computation needed to decide which copy to pronounce. Similar intuitions for syntactic derivations have been proposed in previous studies. One of them is called the “As Soon As Possible” Principle, as in (37) (Yang 1997, Collins 1999, 2001). (37) As Soon As Possible principle (Collins 2001, (28)) If it is possible for an operation to apply, then it must apply. Earliness under Cyclicity is a PF application of ASAP principle. Now, we have two economy- based pronunciation principles that play a role in deciding which copy to pronounce: Pronounce EPP and Earliness under Cyclicity. The two principles contradict each other in that Pronounce EPP forces a higher copy to be pronounced while Earliness under Cyclicity forces a lowest copy to be pronounced. Then, how does a grammar resolve this contradiction? To more concretely illustrate how the two pronunciation principles work, I will discuss the derivation of the English multiple wh-construction in (32a). The derivation of (32a) is represented in (38). As a transitive sentence, lexical items are inserted as in (38a). Since vP is a phase, the complement of v is transferred to PF once the phase has been completed, as in (38b). Since the object wh-phrase which book needs to be accessible from operations outside of the phase (for wh-valuation), it undergoes movement to the outer specifier of vP due to PIC and condition on movement (16ii). Thus, PF receives two items, read and which book. Both items are the first copy of each chain transferred to PF, so they are marked with [P], which signals advantage for pronunciation. Subsequently, T enters into the derivation and the subject wh-phrase which person moves to SpecTP to give values of uninterpretable φ-features of T, as in (38c). Since T is not a phase head, Spell-out does not occur at this stage. Finally, C enters into the derivation, and both wh-phrase undergoes movement to SpecCP. Since CP is a phase, when it has been completed, the complement of the phase head, namely TP, is spelled-out as in (38d). As a result of the spell-out, another member of the object wh-phrase-chain in SpecvP is introduced 84 into PF. Not only that, two copies of the subject wh-phrase-chain are introduced, one in SpecvP and the other in SpecTP. Since both copies of the subject chain are introduced into PF at the same time slot, both copies are qualified as a preferred candidate for pronunciation due to Earliness under Cyclicity. Finally, the root of the sentence is transferred to PF as in (38e). The elements transferred to PF will be selectively deleted in consideration of pronouncing rules, as in (38f). I have two assumptions. First, in multiple wh-constructions, C needs to get multiply valued to indicate that there is more than one wh-phrase in the sentence. Because of this, both wh-phrase undergoes movement to SpecCP to give a value for an uninterpretable feature of C. Second, however, English multiple wh-question is parameterized as if a single occurrence of overt spell- out of the value in C is enough to satisfy Pronounce EPP. Based on the parametric assumption, the deletion of copy in (38f) can be accounted for as follows: Although the low copies of the subject wh-phrase are [P] marked, the highest copy in SpecCP is selected for pronunciation due to Pronounce EPP (applied to C) and every other copy is deleted (Pronounce EPP over Earliness under Cyclicity). Since the first wh-phrase (i.e. the subject wh-phrase) satisfies Pronounce EPP, the second wh-phrase (i.e. the object wh-phrase) is free from the Pronounce EPP requirement. Therefore, the lowest copy with [P] marked is selected for pronunciation by Earliness under Cyclicity. (38) [CP Which person [which book C [ TP which person [vP which person read which book]]]]? a. vP which person vP v VP read v read which book PF: None 85 b. vP which book vP which person vP Spell-out 1 v VP read v read which book PF: [read] P [which book] P by Earliness under cyclicity c. TP which person TP T vP which book vP which person vP Spell-out 1 v VP read v read which book PF: [read] P [which book] P d. CP which person which book CP Spell-out 2 . C TP which person TP T vP which book vP which person vP Spell-out 1 v VP read v read which book 86 PF: [which person] P [which book] [which person] P [read] [read] P [which book] P e. Spell-out 3 CP which person which book CP Spell-out 2 . C TP which person TP T vP which book vP which person vP Spell-out 1 v VP read v read which book PF: [which person] [which person] did [which person] P [which book] [which person] P [read] [read] P [which book] P f. PF: [which person] [which book] did [which person] P [which book] [which person] P [read] [read] P [which book] P By Pronounce EPP and Earliness under cyclicity (16) Condition on Movement X substitutes a specifier of the probe Y or X adjuncts to the head of the probe Y iff (i) an uninterpretable feature of Y is valued by the movement (ii) the movement is a necessary step for some later Agree in which an uninterpretable feature of Z will be valued by a corresponding feature of X There is still a puzzle that needs to be resolved. English multiple wh-questions are divided into two types as I discussed above: multiple wh-questions that obey superiority and those violate superiority. In multiple wh-question in superiority violating order in (39), the in-situ 87 wh-phrase does not show “high” behavior being unable to license ACD. It seems rather to be in a lower position than SpecCP, possibly in SpecTP. Given that the interrogative C head in multiple wh-questions need to be multiply valued, it seems to be puzzling why in (39b), the in-situ wh- phrase does not undergo movement to SpeCP. I do not have a conclusive remark to account for this issue for now. I rather suggest two possible speculations. (39) Which teacher did which boy invite which teacher? One possible explanation is based on the assumption that a D-linked wh-phrase can give a value to an uninterpretable feature of C in a way different from agreement in its specifier position. Shields (2008) proposes a multiple restrictor requirement in multiple wh-question in general. In other words, regardless whether a multiple wh-construction is construed with D- linked wh-phrases or non-D-linked wh-phrases, multiple occurrence of restriction in the C domain is required. Following the syntax of specificity proposed by Enç (2003), Shields argues that D-linked wh-phrases are specific, so have a “linking index ℓ on D, which gives a set which that NP must belong to” (Shields 2008, (28)). Shields claims that a D-linked wh-phrase undergoes formal feature movement and movement of the wh-feature in D carries the linking index with it to C, so the moved linking index can provide a restriction for the wh-operator in C even without phrasal movement. For this reason, the lack of phrasal movement of in-situ D- linked wh-phrase is acceptable. This account is compatible with the current proposal if we assume that the value of uninterpretable feature C that needs to be valued from a wh-phrase is a restriction of the wh-operator. Then, the requirement of multiple occurrences of restriction in C can be satisfied by formal feature movement of D without phrasal movement. However, such a formal feature analysis looks like a strategy extremely restricted in English D-linked in-situ wh- phrases and hard to be generalized to the majority examples of in-situ wh-phrases across- languages that are not required to be D-linked even though they do not show a “high” behavior (I will discuss non-D-linked wh-in-situ cases in 2.4.). Furthermore, introducing formal feature movement into a system brings more complexity into theory. Another possible explanation is suggested by Pesetsky (2000). To explain the peculiar behavior of D-linked wh-phrase, Pesetsky assumes that even though multiple non-D-linked wh-questions require more than one wh- specifier in SpecCP position, multiple questions with D-linked wh-phrases exceptionally do not. 88 Under the current proposal, this assumption can be reinterpreted as that the interrogative C with multiple D-linked wh-phrases need not be multiply valued. This way of explanation seems to me to be just another way of stating the phenomenon, rather than an explanation. However, is seems to be a possibility unless any argument against it is suggested. Finally, in what follows, returning to Standard Arabic, I will provide a new account for agreement patterns in the language I have discussed earlier. The basic data needs to be accounted for is summarized in (40). (40) Agreement and word order in Standard Arabic a. SVO word order must show full agreement on verb. b. VSO word order must show partial agreement on verb except when the subject is a pronoun. c. Overt or covert pronouns show full agreement regardless of word order. Standard Arabic has a rich agreement morphology that can satisfy Pronounce EPP by morphology insertion. As I discussed, the subject in SVO is not in an A-position. This means that the preverbal subject is not a “value-giver” of the uninterpretable features of T. I further assume that (also following Soltan 2007, AlAlamat 2014) what gives a value to the uninterpretable features of T in SVO is null pro. The null pro is associated with the preverbal topicalized DP, in the same fashion Left Dislocated elements are linked to a resumptive pronoun in the thematic domain. The derivation of SVO with full agreement is represented in (41). The basic argument structure is in (41a). Since vP is a phase, when it is completed, its complement is spelled-out transferring two items to PF. Since each copy of both verb and object DP is newly introduced into PF, they are marked with [P]. When T enters into the derivation, T probes down and gives nominative case value to its goal, subject pro. Subsequently, the subject pro is copied and merged into the specifier of TP to give values to uninterpretable φ-features of T, as in (41b). As a VSO language, in Standard Arabic, I assume that verbs always undergo movement to a position higher than the base position of subject (cf. Benmamoun 1999) 39 . Even though verb raising 39 Benmamoun (1999) shows that verbs in perfective form undergo v-to-T movement, while verbs in imperfective form do not. This difference is realized in the word order as well. The same idiom chunk is expressed in VS word 89 always takes place, I assume that only when the raised verb gives a value to uninterpretable φ- features of T, partial agreement appears. In other words, when the raised verb does not participate in φ-agreement, then partial agreement does not appear even when the verb moves to T (I mark v raising only with an arrow without internal structure of T for keeping the tree simple). Since TP is not a phase, spell-out does not takes place. Next, C enters into the derivation and probes down for topic agreement. The subject DP is externally merged into SpecCP, as in (41c). Since CP is a phase, its complement, TP, is transferred to PF. This time, both copies of pro is newly introduced to PF, so they might have been [P] marked. However, since pro is phonetically null by nature, this [P] marker would not bring any effect in PF. Thus, I assume [P] order when the verb is in perfect form (past tense) (ia) or SV word order when the verb is in imperfective form (present tense) (ib). (i) a. Aim-u llah (Standard Arabic, Benmamoun 1999, (26)) Perf-bless-3ms-him God May God bless him. b. llah y-x-ohm-u God IMP-3m-bless-him May God bless him. However, once we assume that verbs with imperfective form never undergoes any movement to a position higher than the base position of subjects, we happen to confront a problem in accounting for (ii). In (iia) even the verb with imperfective form can precede its subject. There is no difference in grammaticality between VS (iia) and SV (iib). Given that the subject is base-generated higher than the verb, this indicates that the verb must move to a position higher than the subject although the position will not be T. Under the current approach, the verb with imperfective form is regarded as undergoing movement to T as an occurrence of parasitic head movement. (ii) a. ya-ftaH-u l-walad-u 1-baab-a (Standard Arabic, Huybregts (1991, (1)) IMP-3sm-open the-boy-Nom the-door-Acc The boy opens the door" b. 1-walad-u ya-ftaH-u 1-baab-a the-boy-Nom IMP-3sm-open the-door-Ac Thus, v-to-T raising in Arabic can take place for various reasons although the exact destination of each occurrence of verb movement may vary depending on its (im)perfect aspect. Investigating the exact position of each movement is beyond the scope of the current work. However, based on this, I would assume that verbs with perfective form undergo v-to-T movement independent of φ-agreement. 90 is not marked on pro. Finally, the root is spelled-out as in (41d), transferring the single copy of subject DP to PF 40 . (41) a. vP pro v Spell-out 1 v VP V v V Obj.DP PF: [V] P [Obj.DP] P b. TP pro TP T vP pro v Spell-out1 v VP V v V Obj.DP PF: [V] P [Obj.DP] P 40 Following Soltan (2007) I assume that the preverbal subject, which is base generated in non-A-position gets its case not by agreement, but by default (nominative) case assignment. Soltan noted that this approach has the advantage in accounting for the asymmetry between preverbal subjects and postverbal subjects in case marking. Postverbal subjects are always marked with nominative case, while preverbal subjects are marked nominative case only when there is no ECM or Raising that gives other case to the postverbal subject. This indicates that a preverbal subject does not take part in usual case-agreement relation with T. 91 c. CP Subj.DP CP Spell-out 2 C TP pro TP T vP v T pro v Spell-out1 V v v VP V v V Obj.DP PF: pro[V] pro [V] P [Obj.DP] P Spell-out 3 d. CP Subj.DP CP Spell-out 2 C TP pro TP T vP Subj. DP v Spell-out1 v VP V v V Obj.DP PF: [Subj.DP] pro[V] pro [V] P [Obj.DP] P e. PF decision: [Subj.DP] EPP pro[V] pro [V] P [Obj.DP] P In (41), pro, which is a covert counterpart of overt pronoun that has full set of φ-features, gives values to uninterpretable φ-features of T. However, pro cannot be phonetically realized due to its lack of phonetic contents. Thus, it is not spelled-out phonetically. However, this does not cause any problem in Pronounce EPP because Pronounce EPP is satisfied by morphology insertion due to the rich morphology system in this language. The verb is pronounced with the copy on T 92 because of independent perfect agreement. The subject DP is pronounced with the copy in SpecCP due to Pronounce EPP applied to topic agreement (and also this is the only copy of the subject DP). This derivation generates SVO order with full agreement. VSO word order with a non-pronominal subject (40b) has its derivations in (42). The derivation is in fact the same as what is proposed in Wurmbrand and Haddad (2014) except that Wurmbrand and Haddad assume a covert movement of the subject DP while I do not. The basic argument structure is construed as in (42a). In that case, v get valued by the subject in its base position, so equipped with valued uninterpretable gender feature. Since vP is a phase, its complement VP is transferred to PF with two items: a copy of the verb and the object DP, both with [P] marker. When T enters into the derivation, v-to-T raising takes place, as in (42b). After v-to-T movement, v gives gender value to uninterpretable φ-features of T. The rest of φ-features get default valued. In that case, the subject does not move to SpecTP due to the lack of motivation (under the Condition on Movement in (16)). Finally, CP is completed and PF selects which copy to pronounce, as in (42d). To satisfy Pronounce EPP, morphology insertion takes place, but v-to-T movement must be pronounced with a high copy on T as a case of head- movement. (42) a. vP Subj. DP v Spell-out 1 v VP V v V Obj.DP PF: [V] P [Obj.DP] P b. TP T vP v T Subj. DP v Spell-out 1 V v v VP V v V Obj.DP 93 PF: [V] P [Obj.DP] P c. Spell-out 3. CP Spell-out 2 C TP T vP v T Subj. DP v Spell-out 1 V v v VP V v V Obj.DP PF: [V] [Subj.DP] P [V] P [Obj.DP] P d. PF decision: [V] [Subj.DP] P [V] P [Obj.DP] P Lastly, the derivation of VSO order with full agreement when the subject is a pronoun is represented in (43). In the basic argument structure, the subject pronoun is merged into SpecvP. When T enters into the derivation, it searches for its goal and the goal, the subject pronoun, moves to SpecTP from where it gives a full set of values of uninterpretable φ-features of T. The verb also moves to T for perfective agreement. Once the root clause is completed, as in (43c), the final decision for spell-out takes place. In this case, Pronounce EPP is already satisfied by morphology insertion (i.e. inflectional morpheme on verb), so the subject pronoun is not forced to be pronounced in SpecTP. Thus, due to Earliness under Cyclicity, the lowest copy of the subject pronoun with [P] marker is selected for pronunciation. (43) a. vP Subj. Pronoun v Spell-out 1 v VP V v V Obj.DP 94 PF: [V] P [Obj.DP] P b. TP Subj. Pronoun TP T vP Subj. Pronoun v Spell-out1 v VP V v V Obj.DP PF: [V] P [Obj.DP] P c. Spell-out 3 CP Spell-out 2 C TP pronoun TP T vP v T Subj. Pronoun v Spell-out1 V v v VP V v V Obj.DP PF: [Subj.Pronoun] [V] [Subj.Pronoun ] P [V] P [Obj.DP] P d. PF decision: [Subj.Pronoun] [V] [Subj.DP] P [V] P [Obj.DP] P Since the current proposal separates the EPP as a pure PF constraint, it opens up the possibility that the subject DP can be spelled-out in its base position even though it undergoes a movement to SpecTP. Due to the rich morphology system in Standard Arabic, Pronounce EPP can be satisfied by morphology insertion without pronouncing the subject DP in the displaced position. This is the point where the current proposal differs from previous analyses. 95 2.5. Conclusion In summary, I propose a system that has a three-way distinction: agreement, movement and pronunciation. Agreement does not require a movement per se. Agreement can occur at a distance once a “value-giver” c-commands its “value-receiver”. Movement occurs when a value- giver needs to be displaced to be in a proper position from where it c-commands its value- receiver. I proposed three types of movement: phrasal movement, parasitic phrasal movement and parasitic head movement. If a phrase undergoes phrasal movement, it by itself is displaced to a higher position, so it shows “high” behavior. By way of contrast, if a phrase undergoes parasitic movement, the phrase by itself does not undergo movement, so it does not show high behavior. Rather, the values of its formal features are transferred to another phrase or a head and parasitically undergo movement. I argued that subject movement in English is a case of phrasal movement while that in Standard Arabic can be either phrasal movement or parasitic head movement. Furthermore, I argue that English there-construction is a case of parasitic phrasal movement. The current proposal crucially distinguishes movement from pronunciation of the movement chain. I assume that movement takes place only in syntax following the condition on movement in (16). Furthermore, following the copy theory of movement, I assume that a movement chain leaves its copies where it stops-by and selection of the copy to pronounce is a PF job independent to syntactic movement. I suggested two PF constraints that play a role in copy selection: Pronounce EPP and Earliness under Cyclicity. Combination of such PF constraints decides which copy to pronounce in a chain. Furthermore, I argued against those previous studies that assume “high-copy privilege”. I claim that the seemingly high copy privilege just one way of satisfying the phonetically reinterpreted EPP requirement. When the EPP requirement is satisfied by other means, like morphology insertion, then the high copy privilege disappears. Rather, the correct generalization in selection of a copy for pronunciation seems to be a “low-copy privilege” and it can be captured under Earliness under Cyclicity principle. The combination of phonetically reinterpreted EPP requirement and tEarliness under Cyclicity successfully captures the otherwise unexpected in-situ phenomenon. A last important point is that both PF constraints I proposed are basically economy conditions. Thus, those constraints are violable if a derivation obeying the economy condition 96 violates any convergence condition. Based on the system developed in this section, I will discuss typological analysis in the T-domain and C-domain across languages in the following chapter. 97 Chapter 3. Typological study 3.1. Overview This chapter is designed to typologically investigate how the proposed system generates various movement patterns. Since I proposed three different types of movement and two different ways of satisfying Pronounce EPP, logically five different possible movement patterns are expected as in (1). When a phrase XP undergoes phrasal movement and morphology insertion takes place to satisfy Pronounce EPP (1a), the phrase is pronounced with a low copy, but it is expected to be subject to movement constraints. This type of movement has been traditionally discussed in terms of covert movement. When a phrase XP undergoes phrasal movement and a high copy of the chain is pronounced to satisfy Pronounce EPP (1b), it patterns like traditional overt movement. When a phrase XP does not undergo movement, but instead, a parasitic head movement takes place (1c), there is expected to be a morphological marking of agreement in question. And crucially, the head Y is expected to be subject to movement constraints, while an XP phrase is expected not to be subject to movement constraints. When a parasitic phrasal movement takes place (1d/e), the sentence is expected to have an expletive-type phrase that can carry the information of XP. Since the hosting phrase undergoes movement, it is expected to be subject to movement constraints while a phrase XP is expected not to be subject to movement constraints. The difference between (1d) and (1e) is that in (1d) the hosting phrase YP is pronounced with a low copy due to morphology insertion that satisfies Pronounce EPP, while in (1e) the hosting phrase YP is pronounced with a high copy to satisfy Pronounce EPP. Unfortunately, so far, I could not find any example that shows the type (1d). Thus, I will discuss the type in (1e) only as a case of parasitic phrasal movement. In sum, we can expect four different types of movement under the current system. (1a, b, c, e). (1) a. Phrasal movement + morphology insertion [XP……. XP] + morphological marker 98 b. Phrasal movement + spell-out high copy [XP ……. XP] c. Parasitic head movement + morphology insertion [Y……. Y XP] + morphological marker Note. Head movement must be pronounced with a high copy for independent reason (See 2.3). Thus, Pronounce EPP must be satisfied by morphological marker. d. Parasitic phrasal movement + morphology insertion [YP ……YP XP] e. Parasitic phrasal movement + spell-out high copy [YP ……. YP XP] Each pattern parametrically appears in languages or in structures. In addition, it is possible that a language can use more than one strategy, like Standard Arabic, which uses parasitic head movement along with phrasal movement for subject agreement (see 2.4). This chapter especially contributes in providing new approach to in-situness. Traditionally, in-situ subjects have been assumed to not move at all. However, as shown in (1a, c, e), this system predicts that in-situ subjects can be derived by either low-copy pronunciation in a subject- movement chain (1b) or parasitic movement (1c, e). Likewise, this system provides a new approach to wh-in-situ. This approach is desirable given that many previous studies show that it is not the case that every in-situ wh-phrase patterns together. Rather, it has been assumed that there are different types of in-situ wh-phrases. I will show how the current proposal accounts for those variations among in-situ wh-phrases reported in previous studies. I will discuss typological study in subject agreement in 3.2, and wh-agreement in 3.3. I will conclude this chapter in 3.4. 3.2. Typology in argument domain – Focusing on subject agreement Based on the parameters suggested in the proposed system, languages are categorized as in (2). 99 (2) Parameters for agreement in T-domain φ-features Discourse features X head XP phrase XP phrase or X head Yes No The properties and examples of each type of languages are listed in (3)-(7) (3) A-type languages a. Uninterpretable φ-features must be checked by a phrase XP (XP-movement or Parasitic phrasal movement, No parasitic head movement). b. Pronounce EPP can be satisfied by morphology insertion (due to Rich morphology system). Thus, an XP movement chain is pronounced with a low copy (due to Earliness under cyclicity). c. Due to the lack of requirement of pronouncing XP in SpecTP, subject drop is possible. C. VS order languages with an alternate b/w full/partial agreement What kind of feature is involved in agreement Which category values uninterpretable features of T E. Topic/Focus prominent languages Is morphology insertion available for Pronounce EPP D. VS order languages with full agreement A. (S)V order languages with rich agreement morphology B. SV order languages without rich agreement morphology 100 d. List of languages: Adyghe (4) B-type languages a. Uninterpretable φ-features must be checked by a phrase XP (XP-movement or Parasitic phrasal movement, No parasitic head movement). b. Pronounce EPP must be satisfied by pronouncing XP in SpecTP. c. An XP phrase that values T must appear before verb. d. List of languages: English, French, Mainland Scandinavian languages (e.g. Swedish…), cf. Icelandic, German. (5) C-type languages a. Uninterpretable φ-features can be checked either by a phrase XP or by a head X (XP- movement, Parasitic phrasal movement, or Parasitic head movement). b. Pronounce EPP can be satisfied by morphology insertion due to rich morphology system c. Null subject is possible due to the lack of requirement of pronouncing XP in SpecTP. d. List of languages: Standard Arabic, Northern Italian dialects (Trentino, Fiorentino) (6) D-type languages a. Uninterpretable φ-features must be checked by a head X (parasitic head movement, no XP- movement or Parasitic phrasal movement). b. Pronounce EPP is satisfied by morphology insertion. c. Due to the lack of requirement of pronouncing XP in SpecTP, subject drop is possible. d. List of languages: Spanish, Italian… (cf. Greek) (7) E-type languages List of languages: Korean, Japanese, Hungarian… I will discuss each type of languages in order. 101 3.2.1. Type A: (S)V order languages with rich agreement morphology In the first type of languages, φ-features in T are valued by phrasal movement of a subject DP. This type of language can take advantage of morphology insertion as a way of satisfying pronounce EPP, so an overt subject in SpecTP is not required. For that reason, null subjects are allowed. As far as I know, the only language that belongs to the first type is Adyghe 41 . The main characteristic property of Adyghe is polysynthesis. It has “extremely rich verbal morphology including cross-referencing of arguments by pronominal affixes on the verb” (Arkadiev and Letuchiy 2009, p3) allowing morphology insertion as a means of satisfying Pronounce EPP. As predicted, it allows null subjects (and null objects) because the information of agreed φ-features is realized with morphology (i.e. pronominal affixes), as shown in (8). (8) sə-šə-t. 1SG.ABS-LOC-stand I am standing. Since the language is a head final, SOV language, it is hard to test whether verb raising occurs or not by looking at a single clause sentence. However, the (so-called) subject-to-subject raising construction discussed in Potsdam and Polinsky (2012), which I discussed in chapter 1, reveals the impossibility of parasitic head movement. In (9), the matrix verb agrees with the subject in the embedded clause. If a parasitic head movement takes place to give values on the matrix verb, the imaginable hosting head might be the embedded verb. However, as shown in (10), the embedded verb and the matrix verb can be intervened by an adverb. This indicates that the embedded verb is not morphologically merged with the matrix verb. Therefore, we can conclude that the parasitic head movement does not take place in the language. (9) a. a-xe-ri [a-xe-mei pjəsme-r a-txə-new] ø-fjeZ’a-Re-x (Adyghe) DEM-PL-ABS DEM-PL-ERG letter-ABS 3PL.ERG-write-INF 3ABS-begin-PAST-3PL.ABS ‘They began to write a letter.’ 41 Adyghe I discuss here is Temirgoy, which is described in Potsdam and Polinsky (2012) and Shapsug, which is described in Arkadiev and Letuchiy (2009). Both are close to standard and share the same property discussed here. 102 b. a-xe-ri [a-xe-mei pjəsme-r a-txə-new] ø-fjeZ’a-Re-x DEM-PL-ABS DEM-PL-ERG letter-ABS 3PL. ERG-write-INF 3ABS-begin-PAST-3PL.ABS ‘They began to write a letter.’ (Potsdam & Polinsky 2012, (1)-(2)) (10) (Repeated from (47c) in Chapter 1)) [stWədentə pepč zadačə-r q’-a-ŝ ̣ ə-new] t ̣ we fjeZž’aʁ. student each(ERG) problem-ABS DIR-3PL.ERG-do-INF twice began Each student twice had a go at solving a problem.’ EACH > TWICE It happened twice that each student began to work on a problem.’ TWICE > EACH Potsdam and Polinsky argue that the subject is base generated in the embedded clause and moves to the matrix SpecTP to satisfy the (traditional sense of) EPP. According to them, the only difference between (9a) and (9b) is that the raising subject is phonetically spelled-out in the matrix clause (9a) or in the embedded clause (9b). However, in 1.3.1, I showed that the raising analysis under the Chain Reduction algorithm confronts a problem in scope interpretation. Given the prevalent assumption that subject-to-subject raising does not reconstruct, scope ambiguity in the Adyghe example in (10) calls for an explanation. Nevertheless, there is still some evidence from control and binding that shows the subject in the construction like (9b) needs to be interpreted in the matrix clause. Instead, I argue that what happens in (9b) is parasitic phrasal movement. In what follows, I will provide an alternative analysis, which not only accounts for (i) the non-raising property (scope ambiguity) and (ii) the “high” behavior of the subject (i.e. being interpreted in the matrix clause with respect to control and binding). I assume that Adyghe has a “there”-like expletive, uninterpretable φ-features which can be valued by agreement with the subject DP. Crucially, unlike English there, I stipulate that the supposed Adyghe expletive lacks phonetic contents (i.e. null expletive) and person feature as well as other φ-features (yet it has φ-attributes, namely, [φ: ]). Thus, every φ-feature in the null expletive needs to be valued by agreement with the subject DP. It has been widely assumed that a person feature is a property of the head D and D is the locus of denotation. Specifically, following Longobardi (2008) I assume that “individuals are denoted through the Person feature” (Longobardi 2008, (40)). Due to the absence of an inherent value of the person feature, the 103 expletive in Adyghe differs from English there in a crucial point. English there has been assumed to have a valued person feature, so it cannot denote the same entity its associate DP denotes. By way of contrast, I assume that the expletive in Adyghe can denote the same entity its value-giver (i.e. the subject DP) denotes through agreement between the expletive and the subject. Given the derived referentiality, I call the expletive null cataphoric expletives. With this basic assumption, I provide an alternative derivation of (9b) as represented in (11) (For easy of explanation, I glossed in English and did not mark unnecessary details). The embedded argument structure is completed as in (11a). The subject is base-generated in SpecvP of embedded clause without case value. The embedded T enters into the derivation and probes down for its φ-feature agreement. It gives ergative case to the subject in SpecvP and the subject DP moves to SpecTP to give values to uninterpretable φ-features of the embedded T, as in (11b) 42 . As a result, the subject bears ergative case and the embedded T is valued for its uninterpretable φ-features. The clause is selected by the matrix verb and the matrix clause is projected as in (11c). Crucially, the null cataphoric expletive, EXpro, is inserted in SpecVP, like English there is inserted in SpecvP (see section 2.2.1., (18)), and gets absolutive case by v. As there is inserted in a position where theta-role is not assigned, EXpro is also inserted into a theta-role free position. This is the point by which EXpro is distinguished from other empty categories (pro, PRO, or so-called “trace”). Due to the lack of φ-features, it probes down and agrees with the embedded subject DP. To give values to uninterpretable φ-features of EXpro, the embedded subject moves to the matrix SpecvP and EXpro gets values (11d). Finally, the matrix T enters into the derivation and probes down for agreement. EXpro with uninterpretable valued φ-features is selected and EXpro moves to SpecTP, as in (11e). In PF, since EXpro is phonetically empty by nature, it has no phonetic realization. This does not cause a problem in satisfying Pronounce EPP because, due to the rich 42 There are traditionally two main approaches to explaining case assignment mechanism in ergative-absolutive languages. One approach assumes that absolutives receive their case from the functional projection that corresponds to nominative-assigning functional projection (AgrS or T), being treated like traditional nominative subjects while ergatives receive their case from AgrO (or v), being treated like objects in transitives (Murasugi 1992, Ura 2000 among many others). Another approach assumes that ergatives receive their case from AgrS or T, being treated like nominative subjects while absolutives receive their case from AgrO or v, corresponding to objects in transitives (Bobaljik 1993 among many others). I adopt the second view for this specific case being aware that actual case assignment is more complex in ergative-absolutive languages. 104 inflectional morphology, morphology insertion satisfies Pronounce EPP. In addition, the subject with ergative case marker is free from the Pronounce EPP requirement, so it is pronounced with the last copy in the chain because the last copy of the subject DP is [P] marked by Earliness under Cyclicity, as in (11f) 43 . (11) a-xe-ri [a-xe-mei pjəsme-r a-txə-new] ø-fjeZ’a-Re-x DEM-PL-ABS DEM-PL-ERG letter-ABS 3PL. ERG-write-INF 3ABS-begin-PAST-3PL.ABS They began to write a letter. a. vP they [uCase: ] vP VP v a letter write b. TP they[uCase: ERG] TP vP T they [uCase: ERG] vP VP v a letter write 43 Because Adyghe is a head final language, it is hard to test whether the high copy or the low copy is chosen for pronunciation. Since the low-copy pronunciation hypothesis can make the proposed theory consistent without any empirical challenge, I assume that the low copy is pronounced. 105 c. vP VP v EXpro[uφ: ][uCase:ABS] VP C/TP began they [uCase: ERG] TP vP T they [uCase: ERG] vP VP v a letter write d. vP they [uCase: ERG] vP VP v EXpro[uφ: ][uCase:ABS] VP C/TP began they [uCase: ERG] TP vP T they[uCase: ERG] vP VP v a letter write 106 e. TP EXpro[uφ:3PL][uCase:ABS] TP vP T they [uCase: ERG] vP VP v EXpro[uφ: ][uCase:ABS] VP C/TP began they [uCase: ERG] TP vP T they [uCase: ERG] vP VP v a letter write f. PF decision: [EXpro] [they.ERG] [EXpro] P [they.ERG] [they.ERG] P [a letter] [write] [began] Yet, it needs to be clarified why a phonetically and semantically null element is introduced in the derivation. I believe that introducing null cataphoric expletives is the Last Resort, to make the raising construction convergent. In Adyghe raising constructions, the embedded T is not non-finite, unlike in English. Thus, the subject DP, which originates in the embedded clause gets ergative case from the embedded T. This makes the subject DP inert for further movement because the subject cannot be a goal of another operation of agreement due to the lack of unvalued features. Still, the uninterpretable φ-features of the matrix T needs to be valued. Due to the unavailability of parasitic head movement and the impossibility of subject raising, the language needs to appeal to another way of valuation, namely, parasitic phrasal movement. Thus, the phonetically or semantically unmotivated phrase is introduced to the derivation. Then, how can this alternative analysis account for the “high” behavior of the subject with ergative case? I assume that the null cataphoric expletive can denote the same individual(s) denoted by the subject DP through person-agreement. Therefore, the null cataphoric expletive binds an anaphor and controls PRO in the matrix clause. The scope ambiguity in (10) is not a problem any longer under this analysis. The subject-wide scope reading comes from the scope 107 computation between the null cataphoric expletive and the matrix adverb and the subject-narrow scope reading comes between the matrix adverb and the subject in the embedded clause. This analysis also has a benefit of not assuming double case-checking of a single DP. One last issue to be addressed is the derivation of the sentence like one in (9a). In this case, the subject DP is marked with absolutive case, indicating that the subject actually moves from its base position in the embedded clause to the matrix SpecTP. I assume that the subject DP does not get its case valued against the embedded T to be active for the further operation. This is a legitimate step though. Even though it does not get its case valued from the embedded T, it will eventually get its case valued in the matrix clause obeying so-called “Case filter”. It is also not a problem that the subject DP gives values to uninterpretable φ-features of both embedded T and matrix T because the subject DP has interpretable, valued φ-features that can participate in agreement multiple times. Thus, the derivation of a sentence where the subject DP is marked with absolutive case is fairly straightforward under the current analysis. So far, I have discussed why Adyghe belongs to type A. Adyghe does not allow parasitic head movement, but allows parasitic phrasal movement. In addition, due to its rich morphology, Pronounce EPP can be satisfied by morphology insertion on a verb allowing subjects to be phonetically realized in their base position. 3.2.2. Type B: SV order languages without rich agreement morphology and Type D: VS order languages with full agreement Languages in type B are obligatorily SV languages. English, French and Mainland Scandinavian languages belong to this type. In this type of languages, uninterpretable φ-features of T must be valued by a phrase in SpecTP because parasitic head movement is not available. In addition, morphology insertion is not available to satisfy Pronounce EPP. Thus, the subject phrase in SpecTP is forced to be overtly spelled-out to satisfy Pronounce EPP. These languages take advantage of expletives in valuing uninterpretable φ-features of T when the verb has internal argument (without external argument) only. It has two consequences. First, when expletives are in charge of agreement via parasitic phrasal movement, subjects can stay in-situ as a postverbal subject. Since expletives are only available when the predicate does not take an external argument, so the transitive expletive construction is not available in these languages (cf. 108 Icelandic, German). Second, a postverbal subject in this type of languages is an associate of an expletive (contrary to postverbal subjects observed in type D languages below). Thus, Definiteness effects play a role. Such properties are well documented in many previous studies especially in English, French and Mainland Scandinavian languages (Milsark 1974, 1977, Deal 2009). Similarly, the properties of type D languages are well investigated in many previous studies. In type D languages, parasitic head movement takes place to give values of uninterpretable φ-features of T. When parasitic head movement takes place, a subject never moves to SpecTP for φ-agreement. This results in VS order unless there is a topic-movement or CLLD of the subject. Due to the lack of requirement of filling SpecTP, null subjects are available. Spanish, Italian or other Romance languages belong to type D (cf. Greek) 44 . In those languages, postverbal subjects are not an associate of a null expletive. Thus, postverbal subjects do not show definiteness effects (Alexiadou and Anagnostopoulou 1998). I simply mention here some previous studies instead of repeating the observation made in these previous studies. For more information, see Barbosa (1995), Alexiadou & Anagnostopoulou (1998), Kato (1999), Frascarelli, (2007), Sheehan (2016) among many others. Especially, Platzack (2003) provides an interesting analysis to distinguish Italian, Greek type V- raising languages from German, Icelandic type V-raising languages. Platzack argues that in Italian/Greek type, the agreement morphology is pronominal, so it must not be A-bound. Therefore, the preverbal subject is either dropped or in Ā-position. On the other hand, the agreement morphology in German/Icelandic type is anaphoric, so preverbal subjects must be in A-position from where it can c-command v-to-T complex. For details, see Platzack (2003). 3.2.3. Type C: VS order languages with an alternate between full/partial agreement The type C languages allow either phrasal movement or parasitic head movement to give values to uninterpretable φ-features of T. Thus, both SV and VS word order can be derived in principle. However, as far as I am aware, all languages that belong to this type allow morphology insertion 44 Alexiadou and Anagnostopoulou argue that Greek belongs to this type of language. However, there has been an argument that preverbal subjects in Greek are not always CLLDed. See 4.1. for an alternative analysis of subject agreement system in Greek. 109 as a means of satisfying Pronounce EPP. It means that the possibility of satisfying Pronounce EPP by pronouncing a high copy in SpecTP is not considered as an option in those languages because morphology insertion is preferred, so must be selected. The representative property of this type of languages is that agreement on verb can appear in full or in partial form, as seen in Standard Arabic. Other well-known languages that show both full agreement and partial agreement are the Northern Italian dialects, Fiorentino and Trentino. It has been known that in both languages, SV order shows a full agreement, but VS order agreement impoverishment. This pattern is reminiscent of Standard Arabic. For example, in Fiorentino, the postverbal subject in (12a) shows partial agreement whereas the preverbal subject in (12b) full agreement. (12) a. GI' è venuto delle ragazze (Fiorentino, Kinjo 2015, (16)) There-cl.3sg. is come.3sg some girls There arrived some girls. b. Delle ragazze le parlano some girls cl.3pl.F speak.3pl Some girls speak. However, the direct correlation between word order and (full or partial) agreement pattern can be misleading. In Standard Arabic, even though there seems to be a correlation between word order and agreement pattern at first glance, it turns out that full agreement is indicative of the existence of a (null) pronoun, having nothing to do with word order per se. Likewise, careful investigation into Fiorentino and Trentino data derives a similar conclusion. Interestingly, when full agreement appears, the sentence requires a subject clitic even though there is a preverbal subject, as represented with Trentino in (13). When partial agreement appears, a subject clitic does not show up (instead, expletive clitic appears (in Fiorentino, (14a)) or no clitic appears (in Trentino, (14b))). (13) a. El Mario *(el) parla (Trentino, Barbosa 1995, 28) the Mario *(he) speaks b. Ti *(te) parli you speak 110 (14) a. GI' è venuto dele ragazze (Barbosa 1995, 28) There-cl.3sg. is come some girls Some girls have come. b. E' vegnú qualche putela is come some girls The obligatory clitic requirement in full agreement and the absence of a clitic in partial agreement forms remind us of the correlation between full agreement and the existence of a (null or overt) pronoun in Standard Arabic. I claim that as in Standard Arabic, full agreement is indicative of pronominal elements, in this case, clitics, rather than a consequence of SV word order. If the full agreement is the consequence of SV word order, it is hard to explain how full agreement appears even without a subject DP, but only with a clitic, as in (15). In (15), the subject DP can be null. In other words, the full agreement appears between the clitic and the verb even without the subject DP. (15) (Delle ragazze) Le hanno telefonato. (Barbosa 1995, 30) (some girls) cl-3-Fem have called Some girls have called. As happens in Standard Arabic, I assume that the clitic understudies the subject DP in parasitic phrasal movement. Thus, the derivation of full agreement case with a clitic in (16) is similar to what we observed in Standard Arabic. The preverbal subject is in an Ā-position (Barbosa 1995), as in SV order in Standard Arabic. When partial agreement takes place, the derivation proceeds as in (17). In this case, a verb that only has number attributes gets valued by the subject DP in its base position under c-command relation. As a next step, it undergoes parasitic head movement with a valued number feature and unvalued gender/person features and gives values to uninterpretable φ-features in T, resulting in partial agreement. 111 (16) Northern Italian dialects full agreement with subject clitic TP T vP uG: u#:_ CL v’ uP:_ iφ:3.FL v … TP CL TP iφ:3.F.PL T vP uφ:3.F.PL CL v’ iφ:3.F.PL v … (17) Northern Italian dialects partial agreement without subject clitic TP T vP uG: u#:_ DP v’ uP:_ iφ:3.F.PL v … uφ:__ TP T vP V+v T DP v’ uφ:3 uG:_ iφ:3.F.PL u#:_ v … uP:3 uφ: 3 Based on the similarity between Standard Arabic and Northern Italian dialects, I would like to categorize Fiorentino and Trentino, as well as Standard Arabic, into Type D. As I discussed, partial agreement is a result of parasitic head movement while full agreement is a result of parasitic phrasal movement of clitics. However, these languages also use phrasal movement. When there is no subject clitic, a preverbal subject pronoun must appear, as in (18). This is the case where phrasal movement is required. Thus, Northern Italian dialects alternate between phrasal movement and parasitic phrasal/head movement as a means to agreement, being members of type D languages. (18) a. Standard Italian: Parli (Barbosa 1995, 27) (you) speak b. Fiorentino: * (Tu) parli c. Trentino: * (Te) parli 112 Turning back to Standard Arabic, Wurmbrand and Haddad (2014) argue that Standard Arabic has a subject-to-subject raising construction in which a low copy is pronounced. The sentence in (19) is an example. Contrary to the general pattern that VS word order with a proper noun subject shows partial agreement, the matrix verb tˁafiqu: ‘start’ in (19) bears full agreement. The agreement pattern in (19) is [matrix verb-full agreement >> embedded verb- partial agreement >> proper noun subject]. (19) tˁafiqu: jansˁarif-u l-na:s-u started.3.M.PL leave.3.M.SG-IND the-people-NOM The people started to leave. (Al-Ghalayini, 2003, p. 205, fn. 2, cited in Wurmbrand & Haddad 2014, (25))) Wurmbrand and Haddad note that they could not find any case that shows such agreement patterns in other contexts (without some exceptions), so it seems that such agreement pattern is not productive 45 . Nevertheless, some system needs to account for the existence of agreement pattern in languages. I do not agree that the subject in (19) undergoes movement to the matrix SpecTP because it is hard to find any motivation of the movement (under Wurmbrand & Haddad’s assumption, Standard Arabic does not have traditional sense of the EPP, so the movement cannot be triggered by the so-called EPP. In addition, case cannot be a reason as well given that the subject already checked its case against the embedded T). Instead, based on the proposed agreement system for Standard Arabic in 2.4, I argue that what happens in (19) is parasitic phrasal movement, similar to the raising construction in Adyghe in 3.2.1. The derivation 45 Khalil Iskarous (p.c) finds the verb tˁafiqu: ‘started’ in (19) strange to him. He suggested to use another verb bada'a ‘begin’ instead of the verb to test the sentence as in (i). He said that he does not believe that the sentence in (i) is grammatical for Arab Speakers. However, his impression is that much research on Arabic is about Arabic translations of sentences spoken by people who speak other languages, so (maybe due to the effects of other languages) most Arabs living in the US would say that the sentence is okay. This observation supports that this structure it not commonly used. (i) bada'u jansarifu alna:su began.3. M.PL leave.3. M.SG-IND the-people-NOM The people began to leave. 113 of (19) I propose is represented in (20) (For explanatory convenience, I use English glosses and did not mark the embedded CP to avoid unimportant complications in the tree). The subject is base generated as a complement of a verb as a subject of an unaccusative predicate (20a). To give gender value to v head, it moves to SpecvP. When T enters into the derivation (20b), it probes down and gives a case value to its goal, the subject DP. At the same time, the uninterpretable gender-feature of T get valued against v head which moves to T. Since the verb head cannot give values to other φ-features except gender, default value is assigned to person/number features, resulting in partial agreement in the embedded clause (20c). When the matrix verb enters into the derivation (20d), it projects to vP to which the null cataphoric expletive, EXpro is inserted. Like English there or the expletive in Adyghe raising constructions, it is inserted into a position where no thematic role is assigned. Due to the uninterpretable φ- features, EXpro probes down and the subject DP moves to a position where it c-commands EXpro and gives it values. Finally, the matrix T enters into the derivation and EXpro with a full set of uninterpretable valued φ-features moves to SpecTP and give values to T, resulting in the full agreement in the matrix clause. (20) a. vP the people[iφ:3.M.PL] vP v[uφ:__] VP V v b. Parasitic head movement TP T vP v[uφ:M] T[uφ: ] the people vP V v v VP V v 114 c. Partial Agreement on the embedded T TP T vP v[uφ:M] T[uG:M ] the people vP leave v v VP leave v d. Null cataphoric expletive, EXpro, insertion and φ-valuation vP the people[iφ:3.M.PL] vP EXpro[uφ:__] vP v VP start v start TP T vP v[uφ:M] T[uG:M ] the people vP leave v v VP leave v e. Agreement between EXpro and the matrix T TP EXpro[uφ:3.M.PL] TP T[uφ:__] vP the people vP EXpro vP v VP start v start TP T vP v[uφ:M] T[uG:M ] v VP leave v leave v leave the people 115 (21) PF decision: [EXpro] [the people] [EXpro] [start] EPP [start] P [leave] EPP [the people] P [leave] P In PF (21), the subject chain, the people, is pronounced with the lowest copy by Earliness under Cyclicity. The embedded verb is pronounced with the copy higher than the subject (due to V- raising). This derivation results in [matrix verb-full agreement >> embedded verb-partial agreement >> proper noun subject]. This alternative analysis has a benefit in that it does not need to assume unmotivated subject raising to the matrix clause. One issue I do not have an answer for is the interesting difference between regular DPs and pronouns. A pronoun in Trentino can appear without a subject clitic (18), but a regular DP obligatorily appears with the subject clitic (13). This indicates that only pronouns, but not proper nouns, can undergo phrasal movement. For now, I do not have an answer to the question of what makes the difference between pronouns and proper nouns in movement and leave this issue for future work. So far, I distinguished four types of languages among φ-agreement languages depending on the availability of different modes of feature valuation (phrasal movement, parasitic head movement or parasitic phrasal movement). The last type of language, Type E, which is named “Discourse configurational language” by É Kiss (1995) has not been discussed so far. Since the T-domain in those languages belonging to this type is highly sensitive to discourse properties like topic or focus, intonation, which is closely related with information structure plays a big role in satisfying Pronounce EPP. This type of languages will be discussed separately in chapter 3. 3.3. Typology in non-argument domains – Implications for the typology of in-situ wh-phrases The system proposed in this dissertation is in principle not limited to the T-domain (i.e. argument domain). Since the proposed system brings together the general properties of agreement, movement and pronunciation, any movement must be subject to the system. In this section, 116 focusing on wh-interrogatives, I will discuss how this system predicts different types of in-situ wh-phrases across languages 46 . The system predicts that phrasal movement, parasitic phrasal movement or parasitic head movement will be observed in wh-agreement/movement as well. Not only that, two options to satisfy Pronounce EPP are also expected to appear: morphology insertion or overt spell-out of the moved item with a high copy (in the specifier position of interrogative functional head) depending on the lexical inventory of each language, as we saw in argument domain. This system further predicts that when morphology insertion is available for Pronounce EPP, the wh- chain, even though it undergoes phrasal movement in syntax, can be pronounced with a low copy in the base position. Based on the parameters, the following typology is predicted under the system: 46 The following discussion will be focused on (short- or long-distance) direct questions. 117 (22) XP phrase X head Yes No Yes No (23) A-type: Wh-movement languages with wh-expletives a. Pronounce EPP is satisfied by pronouncing a copy of a wh-phrase in the moved position. (To the best of my knowledge, no language in this type allows morphology insertion as a means of satisfying Pronounce EPP) b. When wh-phrasal movement takes place, a wh-phrase is spelled-out in its scope position. c. When wh-expletives appear in long-distance wh-questions, wh-phrase is spelled-out in the clause where the wh-phrase is base-generated. d. German, Kashmiri, Hindi-Urdu… (24) B-type: Wh-fronting languages with optional in-situ wh-phrases A. Wh-fronting languages with wh- expletives Does a language use parasitic phrasal movement? D. Languages with canonical wh-in-situ C. Obligatory wh- fronting languages Which category values uninterpretable [uwh] feature of an interrogative head B. Wh-fronting languages with optional in-situ wh-phrases Is morphology insertion available for satisfying Pronounce EPP? 118 a. Wh-phrasal movement takes place for wh-agreement, but Pronounce EPP can be satisfied either via morphology insertion or spell-out of a high copy of the wh-chain. b. French, Coptic Egyptian (25) C-type: Obligatory wh-fronting languages a. Wh-phrasal movement takes place for wh-agreement. b. Pronounce EPP must be satisfied by spelling-out a high copy of the wh-chain. c. Languages can be further distinguished with respect to multiple wh-questions: d. Languages with a high copy pronunciation in every wh-chain: Romanian, Bulgarian… Languages with a high copy pronunciation in a single wh-chain: English… (26) D-type: Languages with canonical in-situ wh-phrases a. Wh-agreement is satisfied by parasitic head movement. b. Morphology insertion is used to satisfy Pronounce EPP c. Sinhala, Korean, Japanese What is interesting in the above typological distinction is that it distinguishes in-situ wh-phrases further based on how an in-situ wh-phrase can be licensed. This looks promising given that in- situ wh-phrases show different properties in different languages or constructions. In what follows, I will discuss each type of languages in sequence and its implication of the typology on wh-in-situ. 3.3.1. Type A: Wh-movement languages with wh-expletives The key property of this type of languages is to have wh-expletives in long-distance wh- questions, as Hindi-Urdu example in (27). For a wh-phrase within a finite subordinate clause in Hindi-Urdu to have a matrix scope, it must undergo fronting to the matrix clause (27a) or insert wh-expletive kya: in the matrix clause (27b). (27) a. Sita-ne kis-ko soca: ki Ravi:-ne kis-ko dekha:? (Hindi-Urdu, Manetta 2010, (1)-(2)) Sita-ERG who-ACC thought that Ravi-ERG who-ACC saw 119 Who did Sita think that Ravi saw? b. Sita-ne kya: soca: ki Ravi:-ne kis-ko dekha:? Sita-ERG EXPL thought that Ravi-ERG who-ACC saw Who did Sita think that Ravi saw? The derivation of the sentence in (27a) might be quite straightforward across analysis (as well as under the proposed system). What is interesting to the current discussion is the sentence with wh- expletives in (27b). I claim that the sentence in (27b) is an instance of a parasitic phrasal movement. In what follows, I will offer an analysis of the sentence in Hindi-Urdu wh-expletive construction in comparison with ones observed in other languages, e.g. German, and Kashmiri. I adopt a traditional view that kya: (and wh-expletives in other languages in general) is an expletive that is generated in the object position and needs a clausal associate with a wh-phrase in SpecCP (Fanselow and Mahajan 2000 among many others). Following Simpson and Bhattacharya (2003) and Manetta (2013), I crucially assume that wh-phrases in Hindi-Urdu undergo movement to SpecCP even though Hindi-Urdu has been assumed as a wh-in-situ language due to the surface position of wh-phrase, which is between the subject and the verb. For the subject preceding the wh-phrase, I assume that it sits in SpecTopP, following Simpson and Bhattacharya. For the base word order, I am open to the two possibilities, SVO or SOV , but for now, I assume SVO as a base order since it is better placed to account for the wh-expletive constructions 47 . 47 Simpson and Bhattacharya (2003) argue that Bengali (Bangla) is a SVO language and this analysis can be extended to Hindi. However, Bhatt and Dayal (2007) argue that Hindi-Urdu has SOV base word order and remnant VP movement creates different types of word order. Bhatt and Dayal’s analysis restricts remnant movement into VP level only. In other words, any XP level except VP cannot undergo rightward scrambling. Based on the restriction, they assume that the postverbal clausal object in (27) is derived by remnant VP movement after V-raising. If this is so, the analysis assumes that the clause is trapped in the VP. However, I am wondering how this analysis can capture the intuition that a wh-expletive can be in a relation with its associate CP, which is trapped inside a VP. In addition, I could not find any clear reason why a non-VP level phrase cannot undergo rightward scrambling. Thus, I might assume that the embedded CP undergoes rightward scrambling to the aspectual level (crucially below the matrix subject to satisfy binding/scope facts discussed in Bayer (1996) and Mahajan (1997)) if it is forced to assume that the base word order in Hindi-Urdu is SOV . 120 The suggested derivation of (27b) is represented in (28). (I use an English gloss for explanatory convenience). In the embedded clause, the argument structure is completed as in (28a). I assume that the wh-phrase in the embedded clause does not have an uninterpretable Q feature. Instead, I assume that a wh-phrase has an uninterpretable focus feature. This idea comes from the fact that in wh-expletive constructions, the wh-phrase in the embedded clause by itself does not show up in the clause where the question force operates. The matrix scope of the wh- phrase is obtained via the wh-expletive in the matrix clause. This indicates that the wh-phrase in the embedded clause does not participate in the feature checking associated with the “question- making”. Furthermore, there has been an observation that in Hindi-Urdu, the position which focus elements and wh-phrases occupy is the same position (Manetta 2010). Therefore, I would rather suggest that the wh-phrase in the embedded clause does not have a Q feature, but a Focus feature, which must be valued by the focus head. Crucially, a wh-phrase has an interpretable wh- feature by nature. Once the embedded C enters into the derivation (28b), the C head probes down and give a Focus value to the wh-phrase by c-commanding it. To give wh-value to the C head, the wh-phrase moves to SpecCP because under the condition on agreement proposed in this dissertation, valuation takes place under a c-command relation. Note, the lack of interpretable Q feature of the embedded C also shares the intuition that the sentence is not interpreted as an indirect question that forces the embedded sentence to be question marked. In short, the embedded clause has nothing to do with interrogative operation. The embedded subject Ravi moves to the higher spec of CP for independent reason 48 , as in (28c). And the matrix verb takes the clausal complement and wh-expletive, EXPL, is base-generated as an object in the matrix clause (28d). As an analogy to there-expletives in the A-domain, the wh-expletive does not have an interpretable wh-feature (nor interpretable Q-feature) (See Manetta 2010 for similar assumption). Rather, it has uninterpretable wh-feature that must be valued by a wh-phrase. Thus, the embedded CP associate, which is wh-marked by the embedded wh-phrase in SpecCP, moves to the matrix SpecvP to give values to uninterpretable wh-feature of the expletive (28e). This is how parasitic phrasal movement takes place in Ā-domain. Finally, the matrix C with an 48 I am not sure that this movement is also topic-driven because it has been assumed, at least in some languages, that topics cannot appear in an embedded clause (Saito 2006). This is a puzzle that any account that assume wh- movement to SpecCP in Hindi-Urdu must solve to be compatible with the word order. I will leave this issue for future works. 121 interpretable Q feature enters into the derivation and probes for its goal. As a result, the wh- expletive gets its uninterpretable Q feature valued and moves to SpecCP to give values to the uninterpretable wh-feature of the matrix C (28f). The matrix subject Sita moves to the higher TopP. Since Hindi-Urdu does not use morphology insertion as a way of satisfying Pronounce EPP, wh-expletive movement to the matrix CP and focus movement to the embedded SpecCP must be pronounced with the high copy. Thus, the PF form reflects the syntax transparently. The only exception is the pronunciation of the embedded CP. The embedded CP undergoes movement to give a value to the expletive phrase. I speculate that CP is pronounced with a low copy, based on the concept of Pronounce EPP. If we remind ourselves of the motivation of Pronounce EPP, the high copy of a movement chain is forced to be pronounced as a way to have a PF consequence of an uninterpretable feature valuation. In other words, the value of the uninterpretable feature of a “value-receiver” (e.g. T) is phonetically realized by pronouncing the “value-giver” (e.g. SubjectDP) adjacent to the goal. However, in this particular case, the “value- receiver”, the wh-expletive, moves to another position (the matrix SpecCP), so pronouncing the “value-giver” (the embedded CP) with the high copy in the moved position (matrix SpecvP) does not mean that the “value-giver” and “value-receiver” are phonetically adjacent. Thus, the embedded CP is free from the EPP requirement, and pronounced with its low copy due to Earliness under Cyclicity. (28) a. vP Ravi-ERG vP v+saw VP who-Acc [iwh, uF:__] b. CP who-Acc [iwh, uf:F] CP C[uwh:wh, iF] TP vP Ravi-ERG vP v+saw VP who-Acc [iwh, uF:__] 122 c. XP Ravi- ERG CP[wh] who-Acc [iwh, uF:F] CP[wh] C[uwh:wh, iF] TP vP Ravi-ERG vP v+saw VP who-Acc [iwh, uF:__] d. VP EXPL[uwh_, uQ_] VP thought XP Ravi- ERG CP[wh] who-Acc [iwh, uF:F] CP[wh] C[uwh:wh, iF] TP vP Ravi-ERG vP v+saw VP who-Acc [iwh, iF:__] e. VP CP[wh] VP EXPL[uwh_, uQ_] VP thought XP Ravi- ERG CP[wh] who-Acc [iwh, uF:__] CP[wh] C[uwh:wh, iF] TP vP Ravi-ERG vP v+saw VP who-Acc [iwh, uF:__] 123 f. CP Sita-ERG CP EXPL[wh,Q] CP C[uwh,iQ] TP vP Sita-ERG vP v+thought VP CP[wh] VP EXPL[uwh:wh, uQ_] VP thought XP Ravi- ERG CP[wh] who-Acc [iwh, uF:F] CP[wh] C[uwh:wh, iF] TP vP Ravi-ERG vP v+saw VP who-Acc [iwh, uQ:__] g. PF decision: [Sita] EPP [EXPL] EPP [Sita][thought] [CP] EPP [EXPL] [Ravi] EPP [who] EPP [Ravi][saw][who] The proposed analysis captures well the island effects in wh-expletive constructions. As shown in (29), the wh-phrase that is within the CP associate of the wh-expletive is sensitive to syntactic islands. Thus, the occurrence of a wh-phrase inside a complex NP island (29a) or adjunct island (29b) makes a sentence ungrammatical. (29) a. *[raam=ne kyaa kah-aa [ki ravii=ko [yeh baat [ki miiraa kyaa Ram.m=erg expl say-pfv.m.sg that Ravi.m=acc this fact that Mira.f what khaa-yegii] pataa hai]]]? eat-fut.f.3sg know be.pres.3sg What did Ram say that Ravi knows the fact that Mira will eat? 124 b. *raam=ne kyaa kahaa [ki siitaa bazaar ja-yegii [kyunki mohan Ram.m=erg expl say-pfv.m.sg that Sita.f market go-fut.f.3sg because Mohan.m kyaa nah˜ı˜ı lay-aa ]]? what neg bring-pfv.m.sg What did Ram say that Sita will go to the market because Mohan didn’t bring? (Hindi-Urdu, Manetta 2013, (59)) The ungrammaticality of (29) can be accounted for as follows: the wh-phrase in an island environment cannot move to the embedded C. Thus, the embedded C cannot get its uninterpretable wh-feature valued and as a chain-reaction, the embedded CP cannot give its wh- value to the expletive. Thus, the unvalued uninterpretable features crash the derivation. There has been proposed two analyses for wh-expletive constructions, as represented in (30). The first approach is proposed in Manetta (2010). In this analysis, the embedded wh-phrase does not move to SpeCP. Rather, the wh-XP in the embedded clause moves to the embedded SpecvP and checks the uninterpretable wh-feature and the EPP feature in the embedded v. In this analysis as well, it is assumed that the wh-expletive has uninterpretable features only. However, since there is no parasitic movement assumed, the matrix C agrees with the wh-phrase in the embedded clause via Agree at a distance. To make it possible, Manetta assumes that phase boundaries in-between are absorbed. I will not discuss the account in detail, but want to point out that this account does not show clearly how the uninterpretable Q-feature of the wh-XP in the embedded clause is checked (as well as uninterpretable Q feature of the embedded v). Not only that, it is not intuitively understandable to posit multiple Q features in a single wh-question. In addition, since the analysis does not assume wh-movement, the island effects in (29) causes a problem. Another previous approach is in (30b), proposed by Nunes (1995, 2004). In this account, the wh-expletive is just a PF realization of the morphologically fused copy of wh-XP with the matrix verbal head. In other word, under Nunes’ account, the same kind of wh-phrasal movement takes place in (27a) and (27b) in syntax, but the difference comes from PF process. In (27a), the high copy of the wh-chain is phonetically pronounced, while in (27b), both high and low copy of the wh-chain is phonetically pronounced and crucially the high copy is realized in the fused form with the v head. Since this analysis assumes that the wh-phrase undergoes movement over the embedded clause, it seems better able to account for the island effects than 125 Manetta’s account. However, a morphological fusion analysis cannot capture the object-wise nature of the wh-expletive, observed in Fanselow and Mahajan (2000) (e.g. incompatibility with the sentential expletives, see Fanselow and Mahajan (2000) for details). (30) a. Manetta (2010) [CP C . . . [vP wh-expl [v . . . ] [CP C . . . [vP wh-XP [v . . . wh-XP]]]]] uwh uQ uwh uQ uwh iQ EPP iwh uQ EPP b. Nunes (1995, 2004): Morphological fusion between v and wh-copy [CP C . . . [vP wh-XP [v . . . ] [CP C . . . [vP wh-XP [v . . . ]]]]] uwh uQ uwh uQ iQ iwh EPP iwh Morphological fusion takes place at PF: … wh-XP + v  kya: … The current proposal can be easily extended to other languages with wh-expletives. In many previous studies, the only difference has been noted between German/Kashmiri and Hindi with respect to wh-expletive constructions is the position where a wh-phrase and a wh-expletive sit in: In German and Kashmiri, a wh-expletive and a wh-phrase appear in SpecCP, while those in Hindi-Urdu seem to appear in a position lower than CP. However, following Simpson and Bhattacharya (2003), the current account assumes that a wh-phrase moves to SpecCP not only in German, Kashmiri but also in Hindi-Urdu. Thus, the current proposal seems to be able to uniformly account for wh-expletives in those languages. Furthermore, this analysis captures intervention effects in German wh-expletive constructions, as in (31). I will discuss intervention effects in detail in chapter 4. For now, let’s just assume that intervention effects take place when an intervener (e.g. negation in this case) sits between a wh-phrase and the question operator at LF. In my analysis, the wh-phrase in the embedded clause moves up to the embedded CP at best when there is a wh-expletive in the matrix clause which can parasitically check the matrix uninterpretable wh-feature. Thus, in (31a), the wh-phrase wen ‘whom’ is in the embedded SpecCP. Even though it does not check its 126 features against the matrix C, the wh-phrase needs to be bound for interpretation at LF because the wh-expletive by itself does not contain a wh-restriction; the wh-expletive only copies (syntactic) wh-features, but not the restriction of the wh-phrase. However, the subject negative quantifier in the matrix clause intervenes the binding relation between the matrix C and the wh- phrase in the embedded SpecCP resulting in the ungrammaticality. Such an intervention effects disappear when the wh-phrase undergoes movement all the way up to the matrix clause, as in (31b). (31) a. *Was glaubt niemand wen Karl gesehen hat? (German, Beck 2006, (8)) what believes nobody whom Karl seen has b. Wen glaubt niemand daß Karl gesehen hat? whom believes niemand that Karl seen has Who does nobody believe that Karl saw? The intervention data also provides support against Nunes’ morphological fusion analysis because in that analysis, the wh-phrase in (31a) undergoes movement to the matrix SpecCP like the one in the (31b). Thus, the analysis predicts a lack of intervention effects in (31a), contrary to facts. In this section, I argued that wh-expletive constructions in Hindi, German and other relevant languages are cases where parasitic phrasal movement takes place. Especially, this analysis can capture otherwise puzzling island sensitivity and intervention effects sensitivity of wh-phrases in wh-expletive constructions. A no movement analysis (Manetta 2010) is challenged by the island sensitivity and a movement analysis (Nunes 1999, 2004) is challenged by the sensitivity of intervention effects. By way of contrast, since the current proposal assumes a partial movement (to the embedded CP), both island sensitivity and intervention effects sensitivity are well explained. This analysis is also well compatible with the claim in previous studies that wh-expletives are base-generated as an object in the matrix clause taking a clausal associate. 3.3.2. Type B: Wh-fronting languages with optional in-situ wh-phrases 127 In the second type of languages as well, only phrasal movement is allowed for wh-valuation. However, unlike the Type A languages, it cannot take advantage of the parasitic phrasal movement. Thus, we cannot observe wh-expletive constructions in this type of languages. One characteristic property of this type of languages is that either morphology insertion or pronouncing a high copy of a movement chain can be used to satisfy Pronounce EPP. Let me discuss French as a representative language based on Cheng and Rooryck (2000). As is well-known, French is a wh-movement language. Thus, to make a wh-interrogative, a wh-phrase needs to be displaced to the front of the sentence, as in (32). A fronting wh-phrase can be either accompanied with an est-ce que phrase (32a) or not (32b). (32) a. Quel livre est-ce que Jean a achete´? which book EST-CE QUE Jean has bought b. Quel livre Jean a-t-il achete´? which book Jean has-he bought Which book did Jean buy? (French, Cheng and Rooryck, 2000, (5)) Interestingly, Cheng and Rooryck observe that French allows wh-interrogatives without wh-movement in a restricted environment: when a rising intonation is applied, wh-in-situ is available, as shown in (33). If the sentence in (33) is pronounced with a neutral intonation that is applied to questions with wh-movement, the sentence becomes ungrammatical. This is contrary to the wh-movement sentences in (32), which do not require such a specific rising intonation. (33) Jean a achete´ quoi? Jean has bought what What has Jean bought? (French, Cheng and Rooryck, 2000, (4)) Cheng and Rooryck claim that the rising intonation that makes the in-situ wh-phrase allowable is the same intonation observed in yes-no questions in French. Yes-no questions (without inversion or est-ce que phrase), exemplified in (34), are required to be pronounced with the rising intonation. 128 (34) Jean a achete´ un livre? (rising intonation) Jean has bought a book? Has Jean bought a book? (French, Cheng and Rooryck, 2000, (6)) Based on the intonational similarity, they propose that the yes-no prosody is actually a PF realization of an underspecified Q-morpheme that is represented as [Q: ]. The underspecified Q- morpheme is inserted to check Q-features and valued as [Q: y/n] as a default when there is no further operation at LF. However, if a wh-feature moves to C at LF, the Q-morpheme is valued as [Q: wh]. This is how the interrogatives with in-situ wh-phrase and yes-no questions share the same intonation pattern according to Cheng and Rooryck. They also point out that the intonation morpheme [Q: ] is inserted into the C head, as question particles are in some languages, and it is for Q(uestion)-feature checking on C head. The wh-feature movement by itself has nothing to do with Q-feature checking (Q-feature checking is already completed by the insertion of the intonation morpheme). Instead, wh-feature movement is required to disambiguate wh-questions from yes-no questions to get a proper interpretation at LF. Their analysis implies that in French, the Q-feature on C can be satisfied by two ways, either wh-phrasal movement to SpecCP (in ordinary wh-movement types) or the intonation morpheme insertion (in yes-no questions or wh- in-situ types). I will adopt the idea that a specific prosody is a PF spell-out form of a certain syntactic morpheme. This idea can be reformulated in the current approach as follows. The in-situ wh- phrase actually undergoes phrasal movement since French allows only phrasal movement to give values to uninterpretable features of C. However, since the language can use morphology insertion as a way to satisfy Pronounce EPP, the intonation morpheme [Q: ] is inserted to pronounce the EPP instead of pronouncing the wh-phrase with the high copy in SpecCP. The intonation morpheme does not need to even be identified by feature movement under the current proposal because the wh-phrase itself is already in SpecCP even though it is pronounced with the low copy in its base position. At PF, the inserted morpheme is realized with the proper intonation. This analysis is compatible with the island sensitivity and weak cross-over violation of in-situ wh-phrases. In (35), the in-situ wh-phrase is inside a complex NP island and the sentence is ungrammatical. In many canonical wh-in-situ languages, island insensitivity has been used as a 129 piece of evidence that shows the lack of wh-movement. For example, in Korean example (36), the wh-phrase in the same island does not cause any harm to the sentence. (35) *Jean aime le livre que qui a écrit? (rising intonation) Jean like the book that who has written Who is the person x such that Jean likes the book that x wrote? (French, Cheng and Rooryck 2000, (4b)) (36) John-i [nwu-ka ssun chayck-ul] cohahani? John-Nom who-Nom wrote book-acc like Who is the person x such that John likes the book that x wrote? Similar conclusions can be reached from the weak-cross over violation in (37). As like its overt counterpart in (37a), the in-situ wh-phrase violates weak cross-over effects in (37b). (37) a. *Quii sai mère a vu quii? (French, Mathieu 1999, (28)) who his mother has seen b. *Sai mère a vu quii? his mother has seen who *Whoi did hisi mother see? Thus, this island sensitivity in (35) and weak-cross over effects in (37) show that wh-in-situ in French undergoes phrasal movement, but a low copy in the base position is pronounced due to the insertion of the intonation morpheme that satisfies Pronounce EPP 49 . 49 I should admit that under the proposed system, it is hard to capture intervention effects in French in-situ wh- phrases, as represented in (i). If the in-situ wh-phrase is a low-copy pronunciation of the wh-movement chain, then it is expected to be free from intervention effects, like its overt movement counterpart in (ib). One possible solution to this apparent counter example comes from the semantic difference between overtly fronted wh-phrases and in-situ wh-phrases. According to Mathieu (2004), overtly fronted wh-phrases are associated with a specific reading while in-situ wh-phrases are associated with a non-specific reading. Given the difference in semantic interpretation, one 130 The insertion of an intonation morpheme as a means of satisfying Pronounce EPP plays a crucial role in syntax-prosody interactions observed in Korean/Japanese data, which will be discussed in chapter 4. In addition to the insertion of an intonation morpheme, there is another option of morphology insertion for Pronounce EPP satisfaction. Of course, lexical morphology can be inserted as a means of satisfying Pronounce EPP. One example is found in Coptic Egyptian. I will discuss in-situ wh-phrases in Coptic Egyptian in 5.3. 3.3.3. Type C: Obligatory wh-fronting languages The current proposal does not seem to have an important contribution on the type C languages. The main property of this type of languages is that it allows only phrasal movement (not parasitic phrasal movement or parasitic head movement) and it does not allow morphology insertion as a means to satisfy Pronounce EPP. Thus, languages in this type has neither wh-expletives nor intonation/lexical morphemes that mark wh-interrogatives. As is well known, this type of languages can be further distinguished based on their behavior in multiple wh-questions: some languages require every wh-phrase to be pronounced in the moved position (e.g. Romanian, Bulgarian…) while others require only a single wh-phrase to be pronounced in the moved position leaving other wh-phrases pronounced in their base position (English). Pesetsky (2000) and many others argue that such a difference comes from the typological difference in the wh-licensing C head. Languages with (so-called) multiple wh- might assume that the wh-phrase in (ib) is actually not in SpecCP but moves further to SpecTopP which is higher than a position for the intervener, while the in-situ wh-phrase just sits in SpecCP which is lower than the intervener, as in (ii). Kim (2002) and Beck (2006) identifies what triggers intervention effects as a focused phrase and argue that intervention effects take place when a focused phrase intervenes between a wh-phrase and its operator. It seems to be plausible to imagine that the intervener, wherever it originates, moves to FocP where focus is licensed. If the FocP is situated between TopP and CP, then the puzzle will be solved. I will leave this possibility open. (i) a. *Jean ne mange pas quoi? (French, Bošković 1998:15, cited in Mathieu 1999, (6)) Jean Neg eat not what b. Qui’est-ce que Jean ne mange pas ti? what that Jean Neg eat not What doesn’t John eat? (ii) [TopP Qui’est-ce [FocP intervener [CP quoi …]]] 131 fronting have a C head whose specifier is required to be multiply filled while languages with a single wh-fronting have C head whose specifier allows only one wh-phrase. In the current analysis, this difference is reduced to the parametric difference in pronunciation. In English-type languages, Pronounce EPP is satisfied by pronouncing a single wh-phrase with its high copy in SpecCP position, so the other wh-phrases must be pronounced with the low copy in the base position due to Earliness under Cyclicity. By way of contrast, in Romanian-type languages, pronounce EPP requires every wh-phrase to be pronounced with its high copy in SpecCP, so Earliness under Cyclicity does not play a role. The current approach captures well the property of in-situ wh-phrases in English multiple wh-questions in 2.4. The discrepancy between where the wh-phrase is phonetically realized and where the wh-phrase is interpreted is the consequence of the PF pronunciation rule, Earliness under Cyclicity. The surface in-situ wh-phrase actually undergoes the same wh-movement as its overt counterpart, but it is pronounced with a low copy since it is free from the EPP requirement which is already satisfied by the previously moved wh- phrase (cf. See 4.2.3. for an alternative analysis). 3.3.4. Type D: Canonical wh-in-situ languages The type D languages have only parasitic head movement. However, it is not the same head movement that we are familiar with (e.g. v-to-T movement in Greek). What I mean by head movement is a question particle (Q) movement. The idea originated in Kishimoto (2005). Kishimoto argues that in Sinhala, the question particle də is merged into the maximal projection where it delimits a wh-phrase. For examples, it can be merged into a DP (38a), PossessiveP (38b), PP (38c) or the immediate right of the complementizer kiyəla as in (39). In either case, the scope of the wh-phrase must be marked by the verbal ending -E for particle- predicate concord. The ordinary verbal ending -A is not allowed. (38) a. Chitra [monə potə] də gatte? (Sinhala, Kishimoto 2005, (23)) Chitra what book Q bought-E What book did Chitra buy? b. Chitra [kaa-ge amma] də daekke? Chitra who-Gen mother Q saw-E 132 Whose mother did Chitra see? c. Chitra [kauru ekkə] də kataa kəlee? Chitra who with Q talk did-E With whom did Chitra talk? (39) Ranjit [kauru aawa kiyəla] də danne/*dannəwa? (Sinhala, Kishimoto 2005, (24)) Ranjit who came that Q know-E/*know-A Who does Ranjit know came? Kishimoto assumes that də does not project any further due to its particle-status. It has properties of non-projected head in a sense that it can be either maximal projection or minimal projection (Chomsky 1995). As a minimal projection, the question particle behaves as a head, X 0 . This is how the question particle can be affixed to a complementizer on its left in PF 50 . On the other hand, as a maximal projection, it behaves as a phrasal element, so it is base-merged as an adjunction to the maximal projection as in (40a). It undergoes Ā-operator movement to SpecCP, the scope-marking position from where it binds the wh-phrase. Kishimoto argues that in (38) and (39), the movement is covert because the [+Q] is weak. When [+Q] is strong, the movement of the question particle is overt, as exemplified in (84). 50 Hagstrom (1998) assume that a question particle in Sinhala undergoes head movement, not phrasal movement. Hagstrom argues that the Head Movement Constraint (Travis 1984) is not relevant here because the movement is triggered by feature checking so movement of H will not be blocked by any intervening head which does not carry the relevant feature F. In the question particle movement, there seems to be no intervening head that has the same question feature. Departing from Hagstrom and Kishimoto, Cable (2007) argues that what undergoes movement in Sinhala examples is the entire QP, head of which is Q and complement of which is DP/PP/PossP or entire embedded CP. Thus, Cable claims that wh-questions in Sinhala appears that the entire complement (DP/PP/PossP/CP) is pied- piped along with the question particle. This is why, according to Cable, clause-final question particle is impossible in Sinhala. In other words, the question particle cannot be separated from its hosting head nor independently move to C it alone. Cable’s analysis is interesting but it does not seem to be clear to me how the question particle appears at the left periphery in a subordinated clause. In addition, I will address another problem of Cable’s system with respect to intervention effects later in this section. Thus, I will not follow Cable’s approach. A similar analysis is proposed by Morita (2014). 133 (40) a. DP/PP/PossP DP/PP/PossP də b. CP CP kiyəla ‘that’ CP də ⸽ DP/PP/PossP DP/PP/PossP də (41) Ranjit [kauru aawa də kiyəla] dannəwa. (Sinhala, Kishimoto 2005, (8)) Ranjit who came-A Q that know-A Ranjit knows who came. However, as Kishimoto observes, the occurrence of clause-final də is strictly restricted in its distribution. It cannot appear on the right of the verb in the matrix clause regardless of verbal marking (-A or -E), as in (42) 51 . (42) *Chitra monəwa gatta/gatte də? (Sinhala, Kishimoto 2005, (4)) Chitra what bought-A/bought-E Q What did Chitra buy? 51 There is one exception that allows the question particle be on the right of the matrix verb. When the question particle appears with kiidenek or kii-ak (both means) ‘how many’, the question particle can appear on the right of the matrix verb as in (i). This peculiar pattern has been regarded as a yes-no question type of sentence (Hagstrom 1998, Cable 2007) or the case where the entire TP moves to SpecCP (Morita 2014). Following Hagstrom and Cable, I assume that this peculiar case has an independent syntactic derivation, distinguished from wh-questions. (i) kiidenek potə kieuwa də? (Sinhala, Kishimoto 2005, (8b)) How many book read-A Q How many (people) read the book? 134 It can only appear within a subordinate clause and furthermore, the subordinate clause must be selected by a restricted set of verbs. Kishimoto notes that the class of verbs that allow clause- final question particles is a subset of predicates that take interrogative clauses as their complements. However, there seems to be no clear definition of the class of verbs that allow the clause-final question particle in its complement clause. I suspect that the appearance of the clause-final question particle is a consequence of a requirement of the matrix verb, independent to wh-agreement. One hypothesis is that a set of verbs requires its complement clause to have a question marker in the root (SpecCP), so that the matrix verb and the question particle are in a local domain. A similar requirement has been noticed in Greek (Manzini and Savoia 2011) 52 . If this is the case, the movement of the question particle to the right of the verb in the embedded clause has nothing to do with wh-agreement, but is a response to the selectional requirement of the matrix verb. I do not have a fully developed analysis for this issue. Instead, I will leave this issue open assuming that the clause-final question particle is irrelevant to wh-agreement. There are two important points in Kishimoto’s approach to the current discussion. First, there is a pseudo head movement, which behaves like head movement and phrasal movement simultaneously. Second, due to the pseudo head movement, the wh-phrase that is associated with the question particle remains in-situ without being dislocated. These two properties remind us of 52 Greek allows in-situ wh-phrases in general even in the embedded clause as in (i). However, when a subordinate clause is selected by a certain class of verbs (e.g. rotisan ‘asked’) that selects interrogative complements, in-situ wh- phrase is not allowed as in (ii). In this case, wh-phrases must move to the left periphery. Manzini and Savoia (2011) and Vlachos (2012) argue that when the latter type of verb selects an interrogative complement, the interrogative properties of the embedded clause must be overtly realized in C with either an interrogative complementizer or a wh-phrase in SpecCP. (i) Ipan [ C oti aghorases ti]? (Greek, Vlachos 2012, (20)) said-3pl that bought-2sg what-acc They said that you bought what. (ii) a. *Rotisan [ C oti aghorases ti]? (Greek, Vlachos 2012, (19)-(20)) asked-3pl that bought-2sg what-acc They asked that you bought what. b. Rotisan [ C ti aghorases]. asked-3pl what-acc bought-2sg They asked what you bought. 135 parasitic head movement. Slightly revising Kishimoto’s analysis, I claim what happens in Sinhala is (pseudo) parasitic head movement. I assume that, like other parasitic head movement cases, the host head (in this case, the question particle Q) is base-generated with uninterpretable features. When it merges as an adjunction to the maximal projection where it delimits a wh- phrase, the wh-phrase gives values to an uninterpretable wh-feature of the question particle, as in (43a). When the interrogative C head enters into the derivation, it searches for its goal and give values to uninterpretable Q features in its c-command domain. The question particle now with an uninterpretable valued wh-/Q-feature moves to SpecCP to give an uninterpretable wh-feature to C (43b). (43) a. DP/PP/PossP DP/PP/PossP[iwh, uQ] də[uwh:wh, uQ] wh-phrase b. CP C[uwh, iQ] də[uwh:wh, uQ:Q] Incorporation in PF ⸽ DP/PP/PossP[iwh,uQ:Q] DP/PP/PossP də Even though the movement of the question particle shows head movement property by being affixed to another phrase in PF, it behaves as phrasal movement in Syntax. Thus, unlike genuine head movement cases, it can be pronounced with a low copy (see 2.3. for obligatory high copy pronunciation in head movement). Thus, if a morphology insertion (the verbal ending -E) is an available option, it must be applied to satisfy Pronounce EPP rendering the movement chain to be pronounced with a low copy. This is why the question particle is spelled-out in its base- merged position in Sinhala. Put in other words, the impossibility of the clause-final də in (42) is a consequence of economy condition that applies morphology insertion over pronouncing a high copy of a chain for Pronouncing EPP satisfaction. Crucially, since this is not a convergent 136 condition, this preference is violable allowing some exceptions where a clause-final question particle is allowed as in (41). Another crucial point in Sinhala in-situ wh-questions is percolation of wh-restriction. This property is best illustrated with question-answer pairs. As shown in (42), when the question particle is adjoined to a simple wh-phrase, the sentence can be answered with a DP corresponding to the wh-phrase. However, when the question particle is adjoined to a bigger constituent as in (43), the same type of answer is impossible (43b). The answer to this question must include the entire phrase that is delimited by the question particle (43c) 53 . (42) a. Q: Chitra [kau də aawa kiyəla] kiiwe? (Sinhala, Kishimoto 2005, (35)) Chitra who Q came-A that said-E Who did Chitra say came? b. A: Ranjit. (43) a. Q: Chitra [kauru aawa kiyəla] də kiiwe? (Sinhala, Kishimoto 2005, (36)) Chitra who came-A that Q said-E Who did Chitra say came? b. A: ?*Ranjit. c. A: Ranjit aawa kiyəla. Ranjit came-A that That Ranjit came. The contrast between (43b) and (43c) shows that the wh-restriction bound by Q-operator in (43a) is not the simple wh-phrase kauru ‘who’, but the entire phrase delimited by the question particle, kauru aawa kiyəla ‘who came that’. Given that, I assume the syntactic structure of (43a) as in (44). The wh-restriction, along with wh-features, is percolated to the embedded CP. The question particle, adjoined to the embedded CP, gets its uninterpretable wh-feature valued and moves to 53 I will not provide a detailed semantic analysis. Instead, I assume that the percolation brings the same semantic effect when pied-piping takes place. This intuition comes from Pesetsky (1987) who argues what happens in sentences like (43) is a large-scale pied-piping including the entire island. 137 the matrix SpecCP to give values to uninterpretable features in the matrix C. If this derivation is correct, the syntactic feature and restriction of the wh-phrase are percolated into the embedded CP and syntactic features are copied into the question particle leaving the restriction to remain in the embedded CP. (44) . CP C[uwh, iQ] də[uwh:wh, uQ:Q] Incorporation in PF ⸽ CP[iwh,uQ:__] CP[iwh, uQ:_] də[uwh:wh, uQ:Q] Percolation wh-phrase[iwh,uQ:_] This analysis predicts (i) island sensitivity of an in-situ question particle (since it undergoes movement in syntax even though it is spelled-out with a low copy in PF), (ii) island insensitivity of a wh-phrase (because it never undergoes movement) and (ii) intervention effects sensitivity (because the restriction of the wh-phrase is left where the surface position of the question particle appears). These predictions are born out as in (45) - (47). When the question particle də is inside a syntactic island (complex NP island in (45a) or adjunct island (45b)), the sentence becomes ungrammatical. However, once the question particle is merged outside of the syntactic island as in (46), the sentences become grammatical even with a wh-phrase inside islands. This indicates that a wh-phrase by itself is not island-sensitive in Sinhala. Those island (in)sensitivities have been discussed in many previous studies in great details. In addition, these predictions are actually a consequence of Kishimoto’s analysis. The intervention sensitivity in Sinhala is not discussed in Kishimoto (2005). However, the currently proposed analysis, which slightly revised Kishimoto’s one, seems to capture the phenomenon well. In Sinhala, the particle -t makes a universal quantifier when it is attached to a wh-phrase. When the universal quantifier intervenes the wh-phrase with the question particle də and its operator head in C as in (47a), the sentence becomes ungrammatical. Once the wh-phrase and the particle moves over the intervener, the sentence becomes grammatical (47b). 138 (45) Island sensitivity of question particle də (Sinhala, Kishimoto 2005, (46a,c)) a. *oyaa [[Chitra kaa-ʈə də dunnə] potə] kieuwe? you Chitra who-Dat Q gave book read-E To whom did you read the book that Chitra gave to whom? b. *[Chitra monəwa də kanə koʈə] Ranjit pudumə unee? Chitra what Q ate time Ranjit surprise became-E What was Ranjit surprised when Chitra ate what? (46) Island insensitivity of wh-phrase (Sinhala, Kishimoto 2005, (47a,c)) a. oyaa [[Chitra kaa-ʈə dunnə] potə] də kieuwe? you Chitra who-Dat gave book Q read-E To whom did you read the book that Chitra gave to whom? b. [Chitra monəwa kanə koʈə] də Ranjit pudumə unee? Chitra what ate time Q Ranjit surprise became-E What was Ranjit surprised when Chitra ate what? (47) Intervention effects (Sinhala, Hagstrom 2001, (32)) a. ?? kauru-t mokak də kiiwe. who-T what Q said-E What did everyone say? b. mokak də kauru-t kiiwe. what Q who-T said-E What did everyone say? There is another example that supports the wh-restriction percolation. When the question particle takes an island clause as in (48), the wh-restriction is percolated into the root of the embedded clause. Thus, even when the potential intervener precedes the wh-phrase inside the embedded clause, no intervention effects is observed because the wh-restriction is already situated on the embedded CP. Thus, the potential intervener does not block the relation between the wh- restriction and the operator in the matrix C. 139 (48) [ kauru-t mokak kiyənə koʈə ] də Siri pudumə unee? (Sinhala, Hagstrom 2001, (33)) who-T what did when Q Siri surprised was-E Siri was surprised when everyone did what?’ To account Sinhala wh-interrogatives, Hagstrom (1998, 2001) proposes head movement of the question particle from wh-phrases to the C head. Hagstrom’s proposal is similar to the one proposed by Kishimoto (2005) except the base position of the question particle. Hagstrom assumes that the question particle is always base generated with a wh-phrase. When the wh- phrase is inside an island, the question particle “migrates” to the right of the island. However, if we follow Kishimoto, there is no need to assume the stipulation of “migrating” a question particle out of the island. Cable (2007) argues that covert pied-piping of the entire island clause along with the question particle takes place in sentences like (46) because the clause is, according to Cable, a complement of the question particle. In addition, based on English and German data, Cable proposes an intervention effect configuration in pied-piping structures as in (49). (49) [ [QP [ ... Offending Operator ... [wh-word] ...] Q]i [... t...]] (Cable 2007, (364)) Unfortunately, the combination of pied-piping analysis of Sinhala wh-questions and the intervention effect configuration in (49) wrongly predicts that intervention effects will occur in island sentences like (48) contrary to facts. In (48), the intervener precedes the wh-phrase within the complement clause selected by the question particle. This is the exact case that is expected to show intervention effects according to (49). However, this prediction is not born out in Sinhala. Thus, Cable’s analysis cannot account for intervention effects in Sinhala. So far, I have discussed how Kishimoto’s analysis with a slight revision (in accordance with the agreement/movement and pronunciation system proposed in the current work) can account for Sinhala wh-questions. I argue that Sinhala shows a type of in-situ wh-phrases that is allowed due to parasitic head movement. A particle with an uninterpretable wh-/Q-feature is adjoined to the question phrase XP. The question particle gets its uninterpretable wh-feature valued being c-command by XP and moves to the matrix CP. In PF, Pronounce EPP is satisfied 140 by morphology insertion (scope marker, -E), so the question particle is pronounced with a low copy in its base position 54,55 . If we remind of intonation morphemes proposed by Cheng and Rooryck (2002), natural question is what will happen in type D languages if an intonation morpheme is inserted to satisfy Pronounce EPP. If the insertion of intonation morpheme is an option for this type of languages, it implies that wh-in-situ in type D languages is licensed by two different means as in (50): one is lexical morpheme (i.e. lexical scope marker) insertion, which is observed in Sinhala (and Coptic Egyptian, see 5.3.1.). The other is intonation morpheme insertion, which is observed in French and Korean/Japanese as we will see in chapter 4. (50) Two possible ways of morphology insertion for satisfying Pronounce EPP (i) Intonation morpheme insertion: wh-in-situ with intonational marking 54 At this point, one might wonder what is meant by pronouncing EPP in wh-question domain. Once we recall the definition of Pronounce EPP, it is a requirement to pronounce a value that is assigned to uninterpretable features in a functional head. In T-domain, values assigned to uninterpretable φ-features in T are pronounced with agreement morphology or the subject DP that bears the value adjacent to T. In wh-questions, Pronounce EPP requires the value of uninterpretable wh-feature in C to be pronounced. I suggest that values that are assigned to an uninterpretable wh- feature in C is actually scope marking. In other words, the purpose of the EPP in the C-domain is to mark the scope of a wh-phrase overtly. This relation between Pronounce EPP and scope marking will be well demonstrated with the interaction between prosody and low-copy pronunciation, which will be discussed in chapter 4. 55 In Irish, it is well known that a long distance wh-movement changes the morphological form of intermediate complementizers. To be more specific, the default declarative complementizer goN becomes aL when a wh-phrase undergoes movement over the clause, as in (i). This has been suggested as evidence for successive cyclic movement. I would like to suggest a possible explanation of this pattern under the current proposal. If we assume that the compelementizer alternation is a result of agreement between an intermediate copy of the wh-phrase and the embedded C, the current system expects Pronounce EPP to apply to this agreement as well. This is the case when morphology insertion (or alternation) takes place to satisfy Pronounce EPP. However, the agreement between the intermediate copy and the embedded clause cannot be the case of wh-operator agreement because the wh-phrase does not have its scope in the embedded clause. Thus, the wh-chain is pronounced with a high copy to express its scope as a means of satisfying Pronounce EPP for wh-agreement. I thank Audrey Li for raising this issue. (i) Cé aL mheas tú aL chonaic tú? (McCloskey 1979, 52) who C thought you C saw you Who did you think that you saw? 141 (ii) Lexical morpheme insertion: wh-in-situ with overt scope marking Now we are equipped backgrounds to explore an interesting type of in-situ languages. Type D languages that has (50i) as an option are the main subject of the chapter 4. I will postpone the discussion how the difference in the means of morphology insertion derives any different between Sinhala and Korean/Japanese to Chapter 4. 3.4. Typology of in-situ wh-phrases I summarize here types of in-situ wh-phrases in each type of languages we have discussed in this section in (51) 56 . Wh-phrases can appear in-situ at surface realization because either it never undergoes movement with a help of parasitic (phrasal/head) movements or a low copy of its chain is phonetically spelled-out as being free from the requirement of Pronounce EPP. If an in- situ wh-phrase is licensed by parasitic movement, the wh-phrase never undergoes movement, so the wh-phrase is expected to be non-sensitive to movement constraints like islands. On the other hand, if an in-situ wh-phrase is licensed by morphology insertion that satisfies Pronounce EPP, the wh-phrase undergoes movement, but it is just pronounced with a low copy. These diverse reasons of wh-in-situness are welcome given the mixed picture in the wh-in-situ literatures (see Bayer and Cheng 2014 for the summary of various accounts of in-situ wh-phrases). 56 Following Simpson and Bhattacharya (2003) and Jayaseelan (1996), I will not categorize Bangla/Hindi simple wh-questions like (i) as an occurrence of an in-situ wh-phrase. The wh-phrase moves and the chain of it is pronounced with a high copy, even though movements of other phrases masks it. Therefore, it is actually not an in- situ wh-phrase. (i) jɔn kon boi-ʈa poɽlo (Bangla, Simpson and Bhattacharya 2003, (1)) John which book-CL read Which book did John read? 142 (51) Typology of in-situ wh-phrases Language Type Licensing of in-situ wh-phrase Property of in- situ wh-phrase PF form Type A Parasitic phrasal movement renders a wh-phrase stay in-situ. No movement [ CP wh-expl … w h-phrase] Type B Morphology insertion as a way of satisfying Pronounce EPP allows a low copy of a wh-phrase to be pronounced. Phrasal movement [CP wh-phrase…w h-phrase] +scope marker or intonation Type C In multiple wh-questions, if a language allows Pronounce EPP to be satisfied by only a single wh-phrase. Phrasal movement [CP wh-phrase1…wh-phrase2 wh-phrase1 … w h-phrase2] Type D Parasitic head movement renders a wh-phrase stay in-situ. No movement [CP…. w h-phrase+scope marker or intonation …] I do not claim that the above types of in-situ wh-phrases constitutues an exhaustive list of in-situ wh-phrases. There seem to be apparent in-situ wh-phrases that can belong to none of those types. For example, the second wh-phrase in homophonous multiple wh-constructions in Romanian we discussed in chapter 1 is a consequence of neither any parasitic movement nor absence of the Pronounce EPP requirement. Rather, an independent phonological constraint forces the wh-phrase to be pronounced with the low copy in the base position. Another example is in Greek. Roussou et al. (2014) argue that Modern Greek allows in-situ wh-phrases when a heavy context that provides an exhaustive list for the wh-restriction is given. Those otherwise impossible in-situ wh-phrases in Romanian and Modern Greek are allowed if special phonological/pragmatical contexts are provided. I will discuss such special in-situ wh-phrases in chapter 5. 143 3.5. Conclusion In chapter 2, I proposed a system designed to separate agreement, movement and pronunciation from each other as independent operations: Agreement takes place under c-command relation. Movement takes place when an element needs to be displaced to a position from where it can c- command a functional head for agreement. Under this system, the EPP does not have a syntactic function that triggers movement. Movement only takes place for the sake of feature valuation of a functional category that serves as a probe. Along with familiar phrasal movement, I suggested two other ways of movement, parasitic phrasal movement and parasitic head movement. The two parasitic movements share the point that it allows the original goal phrase to remain in its base position without being displaced. Along with different types of movement, I proposed phonological constraints that determine which copy in a chain to pronounce. Based on the theoretical apparatus, in this chapter, I investigated typological studies in subject-agreement and wh-agreement. The most important contribution of this chapter is that this system provides a new approach to in-situness. In-situ subjects or in-situ wh-phrases can be licensed by different reasons across languages. This is well compatible with the previously reported observation that it is not the case that in-situ subjects/wh-phrases pattern in the same way across languages. Furthermore, the currently proposed system successfully explains expletive constructions in T-domain as well as in C-domain using parasitic phrasal movement. Occurrences of parasitic head movement are also found in both T-domain and C-domain. These support that parasitic head movement is not as rare as one might guess. Rather, parasitic movement is widely observed phenomenon even though they have not been well noticed in syntactic theories. 144 Chapter 4. Case Study – Interaction between prosody and low-copy pronunciation in Korean 4.1. Overview It has been widely observed that prosody plays an important role in deciding word order in many languages. The vast majority of works in this topic have investigated how word order is determined to fulfill prosodic requirements (Zubizarreta 1998, Arregi 2001, Szendrői 2003 among many others). Another stream of research is represented in Cheng and Rooryck (2000), introduced in 3.3. Cheng and Rooryck argue that a certain intonation is a PF realization of an intonation morpheme. Due to the intonation morpheme, otherwise impossible word order (wh-in- situ) is allowed in French wh-interrogatives giving an impression that prosody licenses in-situ wh-phrases. I interpreted such phenomenon in compliance with the proposed system as that the insertion of the intonation morpheme satisfies Pronounce EPP making the wh-chain free from the Pronounce EPP requirement. As a result, the chain of wh-movement can be pronounced with a low copy. In this chapter, I will investigate two instances of movement in Korean where a low copy is phonetically realized due to the insertion of an intonation morpheme that satisfies Pronounce EPP. The first case is wh-movement. Many previous studies compare Sinhala and Korean/Japanese due to their similarity in having question particles (Hagstrom 1998, 2001, Cable 2007, Morita 2014 among many others). A general assumption among previous studies is that Sinhala and Korean/Japanese share the same wh-interrogative structures except the point when a question particle moves to C: a question particle in Sinhala moves at LF (i.e. covert movement) but in Korean/Japanese it moves at syntax (i.e. overt movement). However, as I will discuss, this is not the correct division between Sinhala and Korean/Japanese. Furthermore, there are other differences between Sinhala and Korean/Japanese if we take prosody into consideration. In section 4.2., I will discuss how they differ from each other and provide an analysis that distinguishes Korean/Japanese from Sinhala based on the system developed so far. As I already mentioned, insertion of intonation morpheme to satisfy Pronounce EPP plays an important role in distinguishing those languages. The introduction of an intonation morpheme in 145 analyzing Korean/Japanese wh-interrogatives will provide a nice explanation of syntax-prosody interaction observed in those languages. The second case is topic movement. It is well-known that Korean is a scrambling language and scrambling is closely associated with information structure. In other words, word order change is recruited to represent information structure correctly. However, word order change is not the only way to mark information structure. It has been widely investigated how different information structures are encoded via different intonation patterns across languages (Bolinger 1965, Jackendoff 1972, Selkirk 1984, 1995, Vallduví 1990, Vallduví and Engdahl 1996, Zubizarreta 1998, Büring 1999, 1997a, Féry 2013, among many others). The role of intonation in encoding information structure is also important in Korean as well 57 . For example, there are two possible ways of encoding topic in Korean. One way is as exemplified in (1a), moving the topic phrase (in this case, object NP, Hamlet) to the front of the sentence. The other way is exemplified in (1b). The topic phrase stays in its base position, but it is realized without any prominent accent. The prosodically non-prominency of topic phrase is also supported by the observation that topic is in many cases prosodically destressed, so could be realized in a pronominal form (Pesetsky 1987) or phonetically null in extreme cases (Huang 1982) 58 . (1) context: What about Hamlet. Is it famous in Korea? a. Hamlet-ul manhun salamtul-i ilk-ess-e. Hamlet-Acc many people-Nom read-past-decl. b. Manhun salamtul-i Hamlet-ul ilk-ess-e. many people-Nom Hamlet-acc read-past-decl. Speaking of Hamlet, many people read it. One natural question to ask is how the topic is licensed in (1a) and (1b). Miyagawa (2010) argues that a topic in (1a) agrees with T and as a consequence of this agreement, it undergoes A- 57 Another way of encoding information structure is to attach a morphological marker. I will discuss this in 5.3. 58 Different types of topics can be realized with different prosody (Frascarelli and Hinterhölzl 2007). Since the topic in (1) continues the conversation about Hamlet, the topic is familiar to both speaker and hearer. Thus, the topic in (1c) is subject to the “destress-given”, and “when given material appears in a sentence…this part of the sentence (if any) is (prosodically) compressed” (Féry and Ishihara (2010: p42)). 146 movement to SpecTP. Thus, so-called local A-scrambling in (1a) is an instance of topic movement and the topic is licensed by agreement. If we assume, as Miyagawa argues, that a topic agrees with T and moves to its specifier position as a consequence of the agreement, how can the in-situ topic (1b) be licensed? I claim that the in-situ topic in (1b) also undergoes topic movement, but the movement is invisible because the low copy is pronounced. In 4.3., I will discuss experimental work that supports the claim that topic movement takes place in sentences like (1b) as well. Based on the system proposed in chapter 2, I will argue that the topic movement chain is pronounced with a low copy since the chain is free from the Pronounce EPP requirement due to the intonation morpheme insertion. I will also discuss similar patterns observed in other languages including Japanese, German and Russian 59 . 4.2. Case 1 – A low-copy pronunciation in wh-chain 4.2.1. Sinhala vs. Korean/Japanese As I noted, many previous studies tend to combine Sinhala and Korean/Japanese in the same group of languages that show question particle movement. This idea is based on the observation that in the three languages, indefinites are created by combining a wh-phrase with a question particle (See Hagstrom 1998 for a semantic analysis of wh-indefinites and wh-phrases). However, I will introduce two major differences between Sinhala and Korean/Japanese that lead me to conclude that Sinhala and Korean/Japanese have different derivations for wh-questions. (Non)obligatory question particles. As discussed in 3.3, in Sinhala, the question particle də must appear in wh-questions. Thus, the absence of the question particle leads to ungrammaticality as in (2). Korean/Japanese has question particles as well 60 . However, in Korean/Japanese, question particles can be dropped optionally. In Korean, a wh-question can have its interrogative force no matter whether the sentence ends with a question particle -ni or 59 The extended version of the first case will be published in Kang (to appear) and a summarized version of the second case is published in Kang (2017). 60 Various question particles are used in Korean/Japanese. In this dissertation, I will use -ni (Korean) and -ka/-no (Japanese) as representatives but the choice of question particle is irrelevant to the main argument. 147 with a -e particle that can mark a declarative sentence as well (3). Similarly, a Japanese wh- question is fine without a question particle -ka when the sentence appears with a rising intonation (4). (2) Obligatory presence of question particle in Sinhala (Kishimoto 2005, (5)) Chitra monəwa *(də) gate/gatta? Chitra what *(Q) bought-E=bought-A What did Chitra buy? (3) Non-obligatory presence of question particle in Korean John-i mwuess-ul ilk-ess-e/ni? John-Nom what-acc read-past-C/Q. What did John read? (4) Non-obligatory presence of question particle in Japanese John-wa doko-ni ikimashita (ka)? John-Top where-to went (Q) Where did John go? (Yoshida and Yoshida 1996:37, cited in Ko (2005, (49a))) Ameliorating intervention effects by manipulating prosody. The second difference between Sinhala and Korean/Japanese is observed in intervention effects. I will discuss the nature of intervention effects and discuss relevant data. Beck (1996) observes a difference in the pairs of sentences in German like (5) and argues that a ‘quantificational’ intervener in the path of LF movement of a wh-phrase creates an intervention effect. The intervention configuration is schematized in (6). (5) a. *Was glaubt niemand, wen Karl gesehen hat? (German, Beck 1996:3-4) what believes nobody whom Karl seen has ‘Who does nobody believe that Karl saw?’ b. Was glaubt Luise, wen Karl gesehen hat? what believes Luise whom Karl seen has 148 ‘Who does Luise believe that Karl saw?’ (6) *[… Xi… [Q… [… ti LF … ]] Since Beck (1996), various analyses have been proposed to account for intervention effects. A syntactic explanation has been proposed by Hagstrom (1998), Pesetsky (2000), Yang (2012), a semantic one by Kim (2002, 2005), Beck (2006) and a pragmatic analysis is also proposed by Tomioka (2007). I will not discuss each approach here, but refer to Hagstrom (2007), Kang (to appear). In this dissertation, I adopt a semantic analysis, namely Focus Intervention Effects, proposed in Kim (2002, 2005) and Beck (2006), as in (7). (7) Focus Intervention Effect (Kim 2002, cited in Beck 2006:11) *[Qi [… [FocP [… wh-phrasei…]]]] A focused phrase (e.g. ‘only’+NP) may not intervene between a wh-phrase and its licensing complementizer. Kim and Beck argue that the common nature of so-called interveners is focus, because focus and wh-questions share the same interpretational mechanism in a sense that both are bringing alternatives in their semantics (Wold 1996). Due to this similarity, a wh-phrase can be bound not only by a question operator but also by a focus operator if the wh-phrase is in the domain bound by the focus operator. However, according to Kim and Beck, unlike focus, wh-phrases are taken not to have ordinary semantic values. They assume that a question operator can give both an ordinary semantic value and a focus value to a wh-phrase, while a focus operator can give only a focus value. Thus, a wh-phrase must be evaluated by a Q(uestion) operator to get both an ordinary semantic value and a focus value. Once a wh-phrase is evaluated by a focus operator, it cannot be properly interpreted due to the lack of an ordinary semantic interpretation. Based on such a theoretical assumption, the intervention effect in (8a) can be explained as in (8b). In (8b), before being c-commanded by its legitimate question operator Q2, the wh-phrase who2 is c- commanded by the focus operator ~ in IP2, which cannot assign an ordinary semantic value for 149 the wh-phrase. Thus, the wh-phrase cannot be further evaluated by Q2 and remains undefined, so lacks an ordinary semantic interpretation. This violates the principle of interpretation in (9) 61 . (8) a. *Only JOHN saw who? b. [CP Q2 [IP3 only C [IP2 ~C [IP1 JohnF1 saw who2]]]]] (Beck 2006:16) (9) Principle of Interpretation (Beck 2006: 16) An LF must have an ordinary semantic interpretation. In the current chapter, following Beck and Kim, I argue that a focus element in-between a question operator in C and an in-situ wh-phrase at LF triggers an intervention effect. The definition of “focus” I adopt is what has been understood in the alternative semantics tradition: focus indicates the presence of alternatives on the level of denotation. Based on this definition of focus, I maintain Beck (2006) and Kim’s (2002, 2005) intuition that intervention effects are the result of semantic uninterpretability of a wh-phrase. However, it is not the case that every in-situ wh-phrase is subject to intervention effects. As we discussed in chapter 2, Pesetsky (2000) distinguishes two types of covert wh-movement, covert phrasal movement and formal feature movement. According to Pesetsky, sensitivity to intervention effects is another parameter to distinguish those two types of covert movement. To be more specific, in English, an in-situ wh-phrase in a superiority violating context is sensitive to intervention effects as in (10a) while an in-situ wh-phrase in a superiority obeying context is free from intervention effects as in (10b). In other words, an in-situ wh-phrase that undergoes covert phrasal movement is free from intervention effects whereas an in-situ wh-phrase that undergoes formal feature movement is vulnerable to intervention effects. (10) a. ??Which diplomat should I not discuss which issue with which diplomat? b. Which issue should I not discuss which issue with which diplomat? (Pesetsky 2000:61) 61 In some languages, for example English, not only so-called focus phrases like ‘only, even’, but also quantifiers like ‘always, often’ trigger intervention effects. To incorporate these examples, Beck (2006) stipulates that quantifiers in such languages are actually focus elements. See Yang (2012), which distinguishes two types of intervention effects, focus intervention effects and intervention effects triggered by scope-bearing elements. 150 Based on this observation, Pesetsky proposes a syntactic account for intervention effects. Pesetsky argues that when only the ff-bundle moves to SpecCP, the restriction of the wh-phrase remains in the base-position at LF. In this situation, if there is any potential intervener situated between the moved formal feature bundle and the restriction of the wh-phrase, the intervener blocks the association of the two parts of the wh-phrase. By way of contrast, when wh-phrasal movement takes place, the restriction of the wh-phrase moves along with its formal feature bundle. Thus, the potential intervener cannot block the association between the two parts of the wh-phrase. This is the reason, according to Pesetsky, why covert phrasal movement, but not formal feature movement, is free from intervention effects. To accommodate in-situ wh-phrases as in (10b) that are free from intervention effects, the intervention configuration in (7) needs to be slightly modified, as in (11) following Pesetsky’s analysis. The only difference between (7) and (11) is that in (11) not a wh-phrase as a whole, but a wh-restriction is subject to intervention effects. Beck (2006) also suggests similar configuration (but not explicitly) to incorporate Pesetsky’s observation. (11) Intervention effects configuration *[Qi [… [FocP [… wh-restrictioni… ]]]] However, in the current system proposed in this dissertation, I do not assume formal feature movement. Rather, the in-situ wh-phrase in a superiority violating context, as in (10a), never undergoes movement because there is no reason for it to move (see 2.4). Assuming that the in- situ wh-phrase in (10a) never moves, it fits the intervention effect configuration in (11). However, in (10b), even though there is a potential intervener between the question operator in C and the in-situ wh-phrase, since the wh-phrase including its restriction moves to a higher position than the potential intervener, no intervention effect is observed. Based on this background, let’s discuss intervention effects in Sinhala and Korean/Japanese. Since we already discussed intervention effects in Sinhala, I will briefly review some relevant discussion. As I repeated in (12a), when a universal quantifier kauru-t ‘everyone’ c-commands a wh-phrase with the question particle, intervention effects arise. This intervention effect appears because the restriction of the wh-phrase and its operator in C is blocked by a potential intervener, the universal quantifier. If the wh-phrase moves over the potential 151 intervener, then the intervention effect is ameliorated as in (12b). In addition, we discussed that when the question particle də is attached to a higher phrase including the wh-phrase, the restriction of the wh-phrase percolates into the maximal phrase that is delimited by the question particle. Therefore, the restriction of the wh-phrase in (13) is on the embedded CP, so the association between the question operator in the matrix C and the wh-restriction is not blocked by the potential intervener inside the embedded clause. This is why there is no intervention effect observed in (13). (12) Intervention effects (Sinhala, Hagstrom 2001, (32)) a. ?? kauru-t mokak də kiiwe. who-T what Q said-E What did everyone say? b. mokak də kauru-t kiiwe. what Q who-T said-E What did everyone say? (13) [ kauru-t mokak kiyənə koʈə] də Siri pudumə unee? (Sinhala, Hagstrom 2001, (33)) who-T what did when Q Siri surprised was-E Siri was surprised when everyone did what?’ Intervention effects in Korean/Japanese show a similar pattern. It has been documented that, as shown in (14a) and (15a), when a wh-phrase is c-commanded, hence blocked by a potential intervener from its operator, an intervention effect appears. If the wh-phrase is scrambled over the potential intervener, the intervention effect disappears (14/15b). (14) Korean intervention effects a. *amwuto nwukwu-lul manna-ci anh-ass-ni? anyone who-Acc meet-CI not-past- CQ b. nwukwu-lul amwuto nwukwu-lul manna-ci anh-ass-ni? who-ACC anyone who-ACC meet- CI not-past- CQ Who did no one meet? 152 (15) Japanese intervention effects a. *daremo dare-o sasow-ana-katta-no? anyone who-Acc invite-Neg-Past-Q b. dare-o daremo dare-o sasow-ana-katta-no? who-Acc anyone who-Acc invite-Neg-Past-Q Who didn’t anyone invite However, this simple contrast does not give a full an exact picture of intervention effects in Korean/Japanese. It has been widely observed that when intonation is manipulated in the sentences in (14a) and (15a), the intervention effects suddenly disappear. For example, in Korean, when there is an extra accent on the wh-phrase, greater than a wh-phrase usually bears, as marked with capital letters in (16), the sentence sounds grammatical (Sohn 1995, Ko 2005, Yoon 2011, Kang to appear). (16) Amwuto nwuKWU-lul manna-ci anh-ass-ni? anyone WHO-Acc meet-CI not-past- CQ Who did no one meet? Prosodic manipulation in Japanese is somewhat different from that in Korean, but still, different prosody leads to different judgment in intervention effects. I will discuss prosodic differences in wh-questions between Korean and Japanese in the next section. For now, I will introduce Kitagawa et al. (2012)’s experiment that shows the relevant patterns in question. Kitagawa et al. conducted an experiment that reveals how prosody affects Japanese’ speakers’ judgment of intervention effects. In their experiment, three different potential interveners were used, as in (17)-(19). Dake ‘only’ and sika ‘anyone but’ have lexical accent as indicated, but daremo ‘anyone’ does not. They conducted two sets of experiments with the same population of participants: In the first experiment, participants saw visual stimuli only and in the second experiment, participants saw visual stimuli while they were hearing audio stimuli that corresponds to each visual stimulus. Audio stimuli were designed to avoid assigning prominent focal accent on the potential intervener. Thus, in the visual+auditory stimuli condition, participants responded to sentences in which prominent accent was assigned only on wh-phrases. 153 On the other hand, in visual only stimuli condition, participants were expected to respond to sentences in which both a potential intervener and a wh-phrase had prominent accent because both items were expected to have prominent accent in implicit reading as focus elements. (17) dake ‘only’ (Japanese, Kitagawa et al. 2012, (13)) a. Ma’riko-dake (’) -ga dare-o sasot-ta-no? Mariko-only-Nom who-Acc invite-Past-Q Who did only Mariko invite?’ b. dare-o Ma’riko-dake (’) -ga dare-o sasot-ta-no? who-Acc Mariko-only-Nom who-Acc invite-Past-Q Who did only Mariko invite?’ (18) sika ‘anyone but’ (Japanese, Kitagawa et al. 2012, (14)) a. Mi’nako-si (’) -ka dare-o sasow-ana-katta-no? Minako-anyone.but who-Acc invite-Neg-Past-Q b. dare-o sasow-ana-katta-no? Minako-anyone.but who-Acc invite-Neg-Past-Q Who didn’t anyone but Minako invite? (19) daremo ‘anyone’ (Japanese, Kitagawa et al. 2012, (15)) a. daremo dare-o sasow-ana-katta-no? anyone who-Acc invite-Neg-Past-Q b. dare-o daremo sasow-ana-katta-no? who-Acc anyone invite-Neg-Past-Q Who didn’t anyone invite? What they found is that the acceptability of the intervention configuration (17a, 18a, 19a) is improved in the visual+auditory stimuli condition when the intervener is either dake ‘only’ or sika ‘anyone but’ compared to the visual stimuli only condition. When the intervener is daremo ‘anyone’, improvement in acceptability was also observed, but not statistically significant. They interpreted the results as indicating that Japanese speakers assign implicit prosody in silent 154 reading and the unacceptability of the intervention configuration is the result of an ill-formed information structure, based on Tomioka’s (2007) interpretation on intervention effects. To be more specific, Tomioka argues that the wh-phrase itself naturally occurs as the focus. To be in a proper information structure, the rest of the sentence including the intervener must belong to the background. However, potential interveners are “Anti-Topic Items/ATIs” which cannot be parts of the background due to their inherent focus-nature. Thus, to save the intervention configuration, the potential interveners must be prosodically reduced to eliminate their focus property. In Kitagawa et al’s experiment, when interveners were dake ‘only’ or sika ‘anyone but’ which have lexical accent, providing audio stimuli with reduced pitch accent on the interveners improves the acceptability while such improvement was weak (but still observed) when the intervener was daremo ‘anyone’ which does not have lexical accent. They interpreted the results as indicating that intervention effects have nothing to do with semantic or syntactic constraints but are due to properties of information structure required by wh-questions. Kitagawa et al’s finding is interesting and it looks crucial to the study of intervention effects. The results show that the same word order in the so-called intervention configuration can have different acceptability depending on prosody it appears with. However, I do not agree with their interpretation of the results. There are some points not clear to me. First, if Japanese speakers do not put implicit focal accent on daremo ‘anyone’, due to its lack of lexical accent, the potential intervener is expected to be put in prosodically reduced part even without any explicit prosody in silent reading. If the information structure of wh-questions requires that every element in a sentence except the wh-phrase is in a prosodically reduced part as background, the sentence in (19a) is expected to be information structure-wise fine even in a visual-only condition. However, we can still observe intervention effects. This indirectly implicates that intervention effects are not solely a problem of information structure. Furthermore, the results show an improvement in the visual+auditory stimuli condition even when the intervener is daremo ‘anyone’ (even though the improvement is statistically not significant 62 ). It is not clear how they explain the marginal improvement in acceptability in visual+auditory stimuli condition when the intervener is daremo ‘anyone’. Another problem in their interpretation is that the information structure account is difficult to extend to intervention effects observed in other 62 Furthermore, the p-value they reported (0.11) is just above that of marginal effects. 155 languages. Intervention effects are widely spread phenomena across languages (Beck 2006) and as far as I acknowledge, such prosody sensitive intervention effects are not widely reported across languages. Not only that, if information structure is the only reason for the unacceptability of intervention effects, it seems to be hard to account for Sinhala data we just discussed in (12)- (13). Therefore, I argue against their interpretation. Rather, maintaining Kim and Beck’s focus intervention effects as a core reason of unacceptability of relevant sentences in Japanese, as well as in Korean, I will provide an alternative analysis that can accommodate the prosody effects observed in Korean/Japanese intervention effects. So far, I have pointed out two major differences between Sinhala and Korean/Japanese. In Sinhala, the presence of a question particle is obligatory and intervention effects ameliorated by intonation have not been observed. By way of contrast, in Korean and Japanese, question particles can optionally appear in wh-questions and intervention effects can disappear with the help of a proper intonation. In 3.3.4, I argued that Sinhala is a language that uses parasitic head movement for wh-agreement. Furthermore, the hosting head is the question particle də, which is base-merged as an adjunction to a maximum projection including the wh-phrase and displaced to SpecCP (and ultimately C head) to give values to uninterpretable features of C. I claim that the same type of parasitic head movement takes place in Korean and Japanese as well. What makes Korean/Japanese differ from Sinhala is (i) the phonetic realization of question particles and (ii) the availability of pied-piping of wh-phrase. I will discuss these two points in the subsequent sections. 4.2.2. So-called question particles are not a genuine question particle The first question is why the presence of so-called question particles is optional in Korean and Japanese. If they are genuine question particles like də in Sinhala, their presence is expected to be mandatory. Following Kim (2010) and Miyagawa (2012), I claim that so-called question particles in Korean and Japanese are not genuine question particles. Rather, they are agreement markers between the matrix predicate and the addressee (as a case of so-called allocutive agreement, terminology from Miyagawa (2012)). Kim (2010) argues that so-called question particles in Korean are pronominals that encode information about the person who the question is directed to. Korean has various 156 particles (-na, -ka, -ni, -nya, -kka) that are used in questions. Kim argues that if a question is directed to the speaker, and the question sentences do not seek an answer from the hearer, the particles, -na and -ka are used. On the other hand, when a question is directed to the hearer, so the hearer is asked to answer to the question, the question ends with the particles, -ni, -nya, -kka. Based on the observation that pronouns and so-called question particles in Korean are morphologically similar (or identical), Kim provides a semantic analysis that accounts for the distribution of different types of so-called question particles. I will refer to Kim (2010) for details but bring out two important points to the current discussion. Kim argues that illocutionary force (as a wh-question) comes from an abstract Q-morpheme and so-called question particles are just pronominals attached to the sentence to give information to whom the question is addressed. This proposal can be restated as that Korean has an abstract question particle that contributes wh- /Q agreement and so-called question particles are not genuine question particles but agreement markers between the predicate and the hearer. A similar observation is also made in Japanese. Miyagawa (2012) observes that the so- called question particle -ka in Japanese is allowed only when a politeness marker -mas- is on the main predicate, as show in the contrast in (20). (20) a. Dare-ga ki-mas-u ka? (Japanese, Miyagawa 2012, (15)-(16)) Who-Nom come-mas-pres Q b. *Dare-ga kuru ka? Who-Nom come Q Who will come? Based on the observation, Miyagawa argues that the so-called question particle -ka must be selected by a Speech Act head (or a bridge verb). This is based on the intuition that the hearer needs to be second person when a question contains the politeness marker. If the particle -ka is a pure question particle, there should be no reason why (20b) would be ungrammatical. The contrast in (20) clearly shows that the presence of the question particle -ka is rather something relevant to agreement for speech-act (Predicate-hearer) which needs to be licensed by the presence of the politeness marker. If this is the case, then it seems to be hard to maintain the assumption that the sentence final -ka in a question is the same lexical item that is used in wh- 157 indefinites in Japanese (e.g. dareka ‘someone’). Miyagawa notes that (20b) can be saved when the question particle is substituted by -no, another question particle, or when the question is pronounced with a rising intonation without any question particle. However, as noted in Yokoyama (2013), a rising intonation is still required even when the -ka particle appears as in (20a). In other words, the question particle and the intonation are not in complementary distribution. Rather, the rising intonation seems to be always required for question making regardless of the presence of question particles. Therefore, it seems plausible to assume that so-called question particles in Korean/Japanese are not genuine question particles based on the observation that those in Korean/Japanese are not obligatory in a question sentence coupled with the observation that such so-called question particles encode speech-act agreement. If this is so, I have to answer how wh- /Q agreement takes place in Korean/Japanese. In the next section, I will seek an answer focusing on question prosody in the two languages. 4.2.3. Intonation as a realization of abstract question particle As I already mentioned, I argue that parasitic head movement takes place in Korean/Japanese wh-agreement. I assume that Korean/Japanese has an abstract (null) question particle, which is a null counterpart of də in Sinhala. As happens in Sinhala, I assume that the null question particle Q is base-merged as an adjunction to a maximal projection including a wh-phrase and moves to SpecCP in Korean and Japanese. (21) CP C[uwh, iQ] Q[uwh:wh, uQ:Q] ⸽ DP/PP/PossP[iwh,uQ:Q] DP/PP/PossP Q In 3.3.4, I argued that Pronounce EPP of the (parasitic head) movement chain in Sinhala is satisfied by insertion of the morphology -E. The insertion of the morphology -E renders the question particle də able to be pronounced in its base position. Similar to -E insertion in Sinhala, 158 I argue that Pronounce EPP is satisfied by morphology insertion in Korean/Japanese and I further claim that the inserted morpheme is an intonation morpheme. Thus, even though the question particle is invisible due to its abstract nature, the movement of the question particle is realized in intonation indicating the scope of the wh-phrase. The intonation morpheme is phonetically realized by rising intonation with dephrasing after the wh-phrase (or post-focal F0 compression). In what follows, I will discuss why the movement of an abstract question particle, which is phonetically realized by intonation, rather than so-called question particles, dictates scope of wh- phrase in Korean and Japanese. As I already discussed, the so-called question particles -ni (Korean) or -ka/no (Japanese) are not mandatory in a question formation when there is rising intonation. Furthermore, the rising intonation is required even when there is -ni (Korean) or -ka/no (Japanese) particle. In Japanese (Tokyo-dialect), rising boundary tone (H% or LH%) is predominant (and almost required) for wh-interrogatives (Venditti 1997, Yokoyama 2013). However, rising intonation is not necessarily employed for wh-questions in Korean. In a production experiment in Jun and Oh (1998), Korean (Seoul-dialect) speakers more likely to use a rising (LH%) boundary tone than other boundary tonal patterns, yet speakers also put H% or HL% in wh-questions. Even though boundary tone could be a non-deterministic phonetic property of wh-questions, dephrasing after wh-phrase (i.e. the elimination of Accentual Phrase boundaries (Jun 1993)) or post-wh F0 compression seems to be necessarily required in both languages (Jun and Oh 1996, Yun 2012 for Korean, Ishihara 2002, 2007, Deguchi and Kitagawa 2002 for Japanese). Ishihara (2011) claims that dephrasing after the wh-phrase is not obligatory in Japanese while post-wh F0 compression is mandatory. In other words, dephrasing after a wh-phrase in Japanese is optional, but that in Korean is obligatory. I suspect that the difference is related with the difference in (non)existence of lexical accent. Japanese has lexical accent, so post-wh F0 compression might be appealed as a way to make post-wh-phrases perceptually non-salient. On the other hand, Korean does not have lexical accent, so dephrasing is emplyed to achieve the same perceptual effects. Despite such a difference, it seems to be an accurate generalization that wh-phrases in wh-interrogatives are made perceptually salient by depressing neighbor elements in Korean and Japanese. Furthermore, dephrasing or post-wh-phrase f0 reduction indicates wh-scope in the two languages respectively. In Korean, as shown in (22), the same sentence (i.e. the same lexical items in the same linear order) can be interpreted either as direct questions or indirect questions 159 depending on the domain of dephrasing (indicated with shading). If dephrasing is maintained to the end of the sentence (22a), the wh-phrase has a matrix scope while if dephrasing terminates in the embedded clause (22b), the wh-phrase has an embedded scope. A similar phenomenon is observed in Japanese as well. If f0 reduction is maintained to the end of the sentence as in (23a), the wh-phrase in the embedded clause has a matrix scope while if f0 reduction ends in the embedded clause, the wh-phrase has an embedded scope (23b). (22) Korean a. John-un Bill-i mwuess-ul sa-ss-nunci alkoissni? John-top Bill-Nom what-Acc buy-past-CompQ know Lit. What does John know Bill bought? b. John-un Bill-i mwuess-ul sa-ss-nunci alkoissni? John-top Bill-Nom what-Acc buy-past-CompQ know Does John know what Bill bought? (23) Japanese (Deguchi & Kitagawa 2002, (11), (14a)) a. [JO'hn-wa [ MA'ry-ga NA'ni-o Øera'nda-to ] i'mademo omo'tteiruØ-nO! ]? -Top -Nom what-Acc selected-Comp even.now think -Q What does John still think that Mary selected? b. [ JO'hn-wa [MA'ry-ga NA'ni-o Øera'ndaØ-ka] I'mademo sirana'i-nO! ]? -Top -Nom what-Acc selected- Comp WH even.now don't.know-Q Doesn't John know yet what Mary selected? Therefore, we can conclude that wh-scope is determined by intonation. In summary, wh- questions in Korean and Japanese (i) require a characteristic prosody (rising boundary tone and post-wh-dephrasing/focal depression) and (ii) mark scope by intonation. Based on the observation, I claim that wh-agreement in Korean/Japanese takes place via parasitic head movement of an abstract question particle, which is realized by insertion of an intonation morpheme. The analysis for Korean/Japanese so far is exactly the same as the one for Sinhala I provided except that the question particle in Sinhala is overt but that in Korean/Japanese is 160 covert and Pronounce EPP is satisfied by insertion of a lexical morpheme in Sinhala but by insertion of an intonation morpheme in Korean/Japanese. Thus, this analysis predicts that like Sinhala wh-phrases, Korean/Japanese in-situ wh-phrases are not sensitive to syntactic islands. This prediction is born out, as shown in (24)-(25). (24) Absence of Complex NP island a. John-un [[nwu-ka pilin] chayk]-ul ilk-ess-e? John-Top who-Nom rent book-Acc read-past-Comp Whox did John read the book x rented. b. Taro-wa [[dare-ga katta] mochi]-o tabemasita ka? (Japanese, Shimoyama 2006, (4)) Taro-Top who-Nom bought rice cake-Acc ate Q Whox did Taro eat rice cakes that x bought? (25) Absence of adjunct island a. John-un [nwu-ka pathi-ey wase] cip-ey ka-ss-e? John-top who-Nom party-loc come-because home-goal go-past-Comp Whox did John went home because x came to the party? b. Taro-wa [dare-ga kita-kara] kaerimasita ka? (Japanese, Shimoyama 2006, (4)) Taro-Top who-Nom came-because left Q Whox did Taro leave because x came? In addition, as shown in (22a), Korean does not show subjacency effects as well. The in-situ wh- phrase in the embedded clause with interrogative complementizer -nunci can have a matrix scope reading. There has been an argument that Japanese in-situ wh-phrases are sensitive to subjacency effects (Watanabe 1992). As shown in (26a), an in-situ wh-phrase in the embedded clause with kado'oka ‘CompWHETHER’ sounds ungrammatical to many Japanese speakers even though some judgment variation exists. This has been suggested as a piece of evidence for the claim that Japanese in-situ wh-phrases undergo some kind of movement. However, Deguchi and Kitagawa argue that such subjacency effects are an artefact of the mismatch between prosody and wh- scope. Once one pronounces the same sentence with a suitable intonation (i.e. prolonged post wh- f0 reduction to the matrix clause), represented in (26b), the sentence becomes much more 161 acceptable. Deguchi and Kitagawa claim that the subjacencey effects are masked by the improper intonation. If this is the case, subjacency is not observed in Japanese as well. (26) a.??*[Jo'hn-wa [Ma'ry-ga na'ni-o katta'-kado'oka] i'mademo siritaga'tteiru-nO! ] John-Top Mary-Nom what-Acc bought-CompWHETHER even.now want.to.know-Q What does John want to know [ whether Mary bought what]? (Deguchi & Kitagawa 2002, (16)) b. [Jo'hn-wa [Ma'ry-ga NA'ni-o katta'-kado'oka] i'mademo siritaga'tteiru-nO! ] John-Top Mary-Nom what-Acc bought-CompWHETHER even.now want.to.know-Q What does John want to know [ whether Mary bought what]? (Deguchi & Kitagawa 2002, (21)) The absence of island conditions is a natural consequence of parasitic head movement. Since in- situ wh-phrases do not move, but the abstract question particle that is adjoined outside of island (as what happened in Sinhala) moves to the matrix SpecCP, the presence of in-situ wh-phrase is insensitive to islands. This is confirmed by the answer-hood of a question with a wh-phrase inside an island. In Sinhala, I argued that the wh-restriction of a wh-phrase inside an island is percolated to the maximal projection that the question particle delimits, so an answer to the question must include the entire island phrase. Since I assume that the abstract question particle in Korean/Japanese is adjoined outside of island as in Sinhala, it is predicted that an answer to the question with an island includes the entire island phrase, due to the percolation of the wh- restriction. The prediction is born out as in (27) and (28) 63 . 63 It seems necessary to note that some Korean native speakers (especially young speakers) allow an answer like in (27a) even though it is much less natural than (27b). The acceptability increases if the answer is case-marked as in (i). As a native speaker, I feel like that the answer in (i) includes ellipsis, so that even though the overt form only contains the word corresponding to the wh-phrase, the meaning I obtain from the answer is the entire island phrase, ‘the person recommended by Suji’ (but the reading is not necessary to include the entire sentence including the matrix clause, ‘John voted for the person recommended by Suji’). I will leave this issue for future studies. (i) Suji-ka Suji-Nom. 162 (27) Q: John-un [[nwu-ka chwuchenhan] salam]-ul ppop-ess-e? (Korean) John-Top who-Nom recommend-Rel person-Acc vote-past-Comp Whox did John voted for a man x recommended. . a. A: ?? Suji b. A: Suji-ka chwuchenhan salam Suji-Nom recommend-Rel person The person recommended by Suji. (28) Q: Mary-wa [[ John-ni nani-o ageta] hito-ni] atta-no? (Japanese) Mary-Top John-Dat what-ACC gave man-Dat meet-Q For which thing x did Mary meet a man who gave x to John? a. A:*/?? Konpyuutaa desu computer Cop It is a computer. b. A: [[ konpyuutaa-o ageta] hito] desu computer-ACC gave man Cop It is the man who gave a computer (to him). (Bayer and Cheng 2015, (15)-(16)) In sum, the parasitic head movement analysis coupled with the insertion of an intonation morpheme as a means of satisfying Pronounce EPP accounts for why a certain intonation is mandatory in wh-questions and the prosody determines the scope of the wh-phrase along with the lack of island sensitivity, lack of subjacency effects and mandatory long-answer hood to a question with an island. Now, I want to reconsider English multiple wh-phrases based on the discussion so far. In chapter 2, I analyzed non-initial wh-phrases in a multiple wh-question as if they move to SpecCP (as like the first wh-phrase), but are pronounced with a low copy. I noted that in English, unlike multiple wh-fronting languages, Pronounce EPP can be satisfied by a single instance of a high- copy pronunciation of wh-chains. However, positing intonation morpheme insertion as a means of satisfying Pronounce EPP opens up an alternative account: In English multiple wh-questions, every wh-phrase undergoes movement to SpecCP to assign values of uninterpretable features in 163 C, as in multiple wh-fronting languages. However, in English, unlike in multiple wh-fronting languages, Pronounce EPP can be satisfied by intonation morpheme insertion instead of pronouncing a high-copy of a wh-chain if the wh-chain is not the first one. Therefore, even though a low-copy of a wh-chain is pronounced, Pronounce EPP is satisfied. This analysis is supported by the following evidence. It has been reported that in English multiple wh-questions, in-situ wh-phrases have focal stress (Chomsky 1995, Truckenbrodt 2013) (29a). This type of focal accent is not observed in fronted wh-phrases. Second, the focal accent of an in-situ wh-phrase is realized in the head of the wh-phrase (29b). Truckenbrodt (2013) accounts for this shift of stress from a wh-word to the head of a phrase with “F-feature percolation” (F represents focus). This resembles wh-movement in the sense that wh-movement requires moving a whole wh-phrase, not only a wh-word, as in (30). When the entire phrase is not pied-piped along with a wh-word, as in (30b), the sentence becomes ungrammatical. In addition, such a stress shift to the head of the wh-phrase is not observed in echo-questions in English which presumably do not have wh-movement. Thus, the stress shift indicates that in-situ wh-phrases and wh-movement in English are subject to the same syntactic constraints. I assume that this similarity appears because the two types of wh-phrases are both wh-movement cases, but they differ in ways of pronouncing the EPP. Since English can use intonation morpheme insertion as a means of satisfying Pronounce EPP from the second wh- chain, the insertion of an intonation morpheme is applied. Since the wh-chain is free from the requirement of satisfying Pronounce EPP, the low-copy is pronounced due to Earliness under Cyclicity. The inserted intonation morpheme is realized with a focal accent (F0 boost) on the head of the wh-phrase. (29) a. Who bought WHAT? b. Who bought [how many BOOKS]? (30) a. How many books did John buy how many books? b. *How many did John buy how many books? Truckenbrodt also reports an association between wh-movement and the lack of focal accent on the moved wh-phrase in English even though a detailed mechanism is different from mine. I will not review Truckenbrodt’s analysis in detail, but want to point out the gist of 164 Truckenbrodt analysis as follows: every wh-word needs to be connected to the Q-operator (Q- marker in Truckenbrodt’s term). For in-situ wh-phrases, F(ocus)-feature plays a role in connecting in-situ wh-phrase and Q-operator. However, when a wh-phrase moves, the wh-phrase is connected to the Q-operator in a different way (i.e. movement), so an F-feature is superfluous and deleted resulting in the lack of focal accent. Truckenbrodt analysis is based on the assumption that in-situ wh-phrases in English multiple wh-questions do not undergo movement. However, as I discussed, even though in-situ wh-phrases are pronounced in their base position, they have “high” behavior as if they undergo movement regarding ACD licensing and intervention effects. Thus, even though Truckenbrodt’s analysis share the same insight with mine, I believe the current analysis is more compatible with empirical evidence observed in English in-situ wh-phrases 64 . In this section, I discussed how wh-agreement takes place in Korean/Japanese. As in Sinhala, I argue that parasitic head movement takes place for wh-agreement in Korean/Japanese as well. The difference between Sinhala and Korean/Japanese in the (absence of) obligatory question particle is reduced to the difference in the (absence of) phonetic content of question particles. However, they share the same mechanism, parasitic head movement, as a means for wh-agreement. In the next section, I will discuss the second difference between the two types of languages, ameliorating intervention effects under a certain intonation. 4.2.4. Pied-piping of wh-phrase and its phonetic realization Now, the remaining question is why in Korean and Japanese, intervention effects are sensitive to prosody. Before moving on, I will discuss prosodic and semantic properties of in-situ wh-phrases in an intervention configuration. In Kang (to appear), I argue that in-situ wh-phrases in an intervention configuration show contrastive topic-hood both prosodically and semantically. Let me briefly review the discussion. The prosodic character that ameliorates the intervention effects is an extra prominent accent on 64 However, this overt movement analysis confronts problems in explaining island insensitivity and the impossibility of parasitic gap licensing of in-situ wh-phrases in English. I do not have an answer for this puzzling pattern for now. 165 in-situ wh-phrases. As we already discussed, dephrasing after wh-phrases is mandatory in wh- question formation in (Seoul) Korean. However, a prominent accent on wh-phrases is not obligatory. Yun (2012) conducted a perception experiment using synthesized speech. What Yun found is that prosodic dephrasing must be satisfied but F0 boost on the wh-phrase is optional in forming a wh-interrogative question. If there is no dephrasing of AP boundaries after the wh- phrase as in (31b), a wh-phrase, regardless of the presence of high F0 on it, is more likely to be interpreted as an indefinite in a yes-no question. On the other hand, if there is dephrasing of AP boundaries after the wh-phrase, the sentence is perceived as a wh-interrogative even without F0 boost on the wh-phrase as much as in the condition where both dephrasing and F0 boost were present (31a). This clearly shows that in (Seoul) Korean, prominent accents on a wh-phrase are not an obligatory prosodic feature for phonologically well-formed wh-interrogatives. (31) a. Wh-phrase interpretation: F0 boost + dephrasing b. wh-indefinite interpretation: F0 boost only (Yun 2012: 288) Thus, we can imagine the possibility that the extra prominent accent on wh-phrases is applied to indicate something other than just question formation. In fact, this prosodic cue is widely used to indicate contrastive focus in Korean (Jun and Lee 1998). If this is the case, one possible interpretation of the extra prominent focus on wh-phrases is that wh-phrases with prominent accents are interpreted as contrastive foci. Interestingly, wh-phrases in an intervention 166 configuration shows exhaustivity, which is presumably a property of contrastive focus. In Kang (to appear), I test three diagnostics suggested by Boeckx (1999, 2000, cited in Zubizarreta (2003)). These tests were used to distinguish in-situ wh-phrases from fronted wh-phrases with respect to exhaustivity in French. First, a cleft sentence as in (32) can be “an answer” to the in- situ wh-phrase (32a) or the cleft question (32b), but not to the question with the fronted wh- phrase (32c). (32) A:C’est à Marie que Pierre a parlé. It is to Marie that Pierre talk ‘It is Mary Pierre talk to.’ (Zubizarreta 2003: 360) a. Q1: Pierre a parlé à qui? Pierre talked to whom? b. Q2: C’est à qui que Pierre a parlé? It is who that Pierre saw? c. Q3: A qui est-ce que Pierre a parlé? / A qui Pierre a-t-il parlé? ‘To whom did Pierre talk?’ (Zubizarreta 2003: 360) Second, (33) is a possible answer for the fronted wh-question (32c), but not for the in-situ wh- question (32a) or the clefted wh-question (32b). (33) (Pierre n’a parlé) à personne. ‘(Pierre did not speak) to nobody.’ (Zubizarreta 2003: 360) Finally, the phrase par exemple ‘for example’ is able to modify the fronted wh-phrase (34a), while it cannot modify the in-situ wh-phrase (34b). (34) a. Qui par exemple est-ce que Pierre a invite? Whom for example did Pierre invite b. *Pierre a invite qui par exemple? Pierre invite whom for example (Zubizarreta 2003: 361) 167 All the tests show that unlike fronted wh-phrases, in-situ wh-phrases in French have the property of exhaustivity. Korean wh-phrases (both in-situ wh-phrase or fronted wh-phrase by scrambling), in general, pattern like fronted wh-phrases in French without having exhaustivity. However, wh- phrases in the intervention configuration pattern like in-situ/clefted wh-questions in French. Regarding the first diagnostic, a cleft sentence can be an answer to a wh-question which has a canonical intervention configuration (35), but is not perfectly natural for a question in non- intervention configuration (36) 65 . (35) a. John-man nwuKWU-lul chotayhay-ss-e? (Kang to appear, (33)) John-only WHO-Acc invite--Past-Comp Who is the person only John invited? b. John-man chotayha-n-saram-un Mary-ita. John-only invite-Rel-person-Top Mary-decl. It is Mary who only John invited (36) a. John-un nwukwu-lul chotayhay-ss-e? (Kang to appear, (34)) John-Top who-Acc invite-Past-Comp Who did John invite? b. ?John-i chotayha-n-saram-un Mary-ita. #John-Nom invite-Rel-person-Top Mary-decl. ‘It is Mary who John invited’ Second, (37a) is a possible answer to the non-intervention wh-question (36a) (like fronted wh- questions in French), but (37b) is not a proper answer to the wh-question in the intervention configuration (35a) (like in-situ wh-questions in French). 65 If the topic marker -nun in (36a) is interpreted as contrastive topic, the answer in (36b) becomes much more natural (almost grammatical). However, if the subject with the topic marker is interpreted as a thematic topic, the answer in (36b) is not acceptable. The contrastive topic has a different derivation from the one assumed here. Thus, I will put this acceptability with a contrastive reading of the subject aside. 168 (37) a. John-un amwuto chotayha-ci anh-ass-e. (Kang to appear, (35)) John-Top nobody invite-ci Neg-Past-Comp. John did not invite anyone b. John-man amwuto chotayha-ci anh-ass-e. John-only nobody invite-ci Neg-Past-Comp. Only John did not invite anyone Finally, it sounds fine to modify the wh-question in non-intervention configurations with ‘for example’ (parallel with fronted wh-phrases in French) as shown in (38), but the wh-question in the intervention configuration is not compatible with such modifying phrase (parallel with in-situ wh-phrases in French) (39). The order between the in-situ wh-phrases and the potential intervener does not matter. (38) a. John-un yelultule etten chayk-ul cwulo ilk-e? John-Top for example which book-Acc usually read-Comp? b. Yelultule etten chayk-ul John-un yelultule etten chayk-ul cwulo ilk-e? for example which book-Acc John-Top for example which book-Acc usually read-Comp ‘Which book, for example, does John read usually?’ (Kang to appear, (36)) (39) a. *John-man yelultule ETTEN CHAYK-UL cwulo ilk-e? John-only for example WHICH BOOK-ACC usually read-Comp? b. *Yelultule ETTEN CHAYK-ul John-man yelultule etten chayk-ul cwulo ilk-e? for example WHICH BOOK-acc John-only for example which book-Acc usually read-Comp Which book, for example, does only John read usually? (Kang to appear, (37)) The above observations show that in-situ wh-phrases in intervention configuration in Korean have exhaustivity unlike wh-phrases in non-intervention configurations. Assuming that a focus element with exhaustivity is defined as a contrastive focus, I conclude that in-situ wh-phrases in intervention configuration are wh-phrases with contrastive focus based on their semantic and prosodic properties. 169 Assuming that the wh-phrases in intervention configurations in Korean are contrastive foci and wh-phrases with contrastive focus are free from intervention effects, I suggest that when there is the specific intonation that makes otherwise illicit configuration acceptable, the wh- phrase undergoes contrastive focus movement and the movement chain is pronounced with a low copy in the base position. The relevant derivation in syntax and LF/PF representation is schematized in (40). In syntax, the wh-phrase undergoes movement due to its contrastive focus property. In Korean, contrastively focused phrases can undergo movement instances of scrambling (Choi 1996 among many others). However, crucially, the wh-phrase with a contrastive reading cannot move independently for the following reasons: In the intervention configuration, a potential intervener precedes the wh-phrase. Potential interveners attested in Korean have exhaustivity in their semantics, so they have a contrastive focus reading as well. In addition, contrastive focus movement is subject to superiority effects in Korean (See Kang (to appear) for details). Thus, there are two contradictory requirements for the wh-phrase in intervention configurations: as a contrastive focus, it is required to be in SpecCP but such movement of the wh-phrase is blocked by the preceding intervener (which is also a contrastive focus) due to superiority. To resolve this contradiction, I argue that, the wh-phrase is pied-piped along with the question particle. If this is the case, the wh-phrase is not in direct agreement relation with the probe for contrastive focus, so it does not violate the superiority condition. Nevertheless, the wh-phrase can be situated in SpecCP by being pied-piped. Therefore, this movement chain has two copies, a low copy in its base position and a high copy in SpecCP. In LF, to have a convergent semantic interpretation, the wh-phrase is interpreted with the high copy because this one is free from the intervener. However, in PF, the low copy is pronounced because Pronounce EPP is satisfied by insertion of the intonation morpheme that is realized with prominent accent (i.e. extra high F0) on the wh-phrase. Thus, the combination of pied-piping and low-copy pronunciation due to insertion of an intonation morpheme accounts for the otherwise puzzling interaction between prosody and intervention effects in Korean. (40) a. Syntax: [CP Q+wh-phrase [IP3 Intervenerc [ ~C [vP Q+ wh-phrase]]]]] b. LF: [CP Q+wh-phrase [IP3 Intervenerc [ ~C [vP Q+ wh-phrase]]]]] c. PF: [CP Q+wh-phrase [IP3 Intervenerc [ ~C [vP Q+ wh-phrase]]]]] 170 Interestingly, parasitic gap sentences provide further evidence for wh-movement indicated in prosody. In previous studies on Korean parasitic gaps, it has been argued that the antecedent of a parasitic gap must be in an Ā-position. Thus, the ungrammaticality of the sentence in (41a) has been explained with a lack of wh-movement. However, if a prominent accent is assigned to the in-situ wh-phrase as in (41b), it sounds much better to my native judgment. This also supports the claim that a wh-phrase undergoes movement that can license parasitic gap, but a low copy of the chain is pronounced with a help of insertion of an intonation morpheme. (41) a.?*Mary-nun [pgi ilk-cito anhko] mwuess-uli peli-ess-ni? Mary-top read-without what-acc throw away-past-Q What Mary threw away without reading? b. Mary-nun [pgi ilk-cito anhko] MWUESS-uli peli-ess-ni? Mary-top read-without what-acc throw away-past-Q What Mary threw away without reading? If the proposed analysis is correct, it needs to be addressed what happens in Japanese, which shows a similar pattern. It is well known that in Japanese, the prominent accent on wh- phrases is obligatory contrary to wh-phrases in Korean (Ishihara 2002, 2007, Féry and Ishihara 2010). Thus, it has been argued that wh-prosody in Japanese has focus prosody. Due to its inherent focus nature, in Kitagawa et al’s experiment, the prosody that ameliorates intervention effects seem different from what is applied in Korean. In Korean, intervention effects disappear when in-situ wh-phrases have an extra prominent accent, while in Japanese, intervention effects are ameliorated when the preceding intervener is not assigned the prominent accent which is normally supposed to be assigned to it. However, the prosodic property observed in the two languages seems to have the same goal, making an in-situ wh-phrase focused, even though specific mechanisms are not the same. For now, I have not tested whether wh-phrases in Japanese intervention configuration also show a contrastive focus property. Thus, it is premature to conclude that the analysis proposed for Korean extends to Japanese. In addition, there is actually a difference in the acceptability of prosodically manipulated intervention effects. Even though Kitagawa et al’s experiment show that intervention effects can be ameliorated by 171 prosodic manipulation, it seems much hard to ameliorate IEs with the help of prosody in Japanese compared to Korean. My two Japanese informants reported that sentences with the intervention configuration are hard to accept even though prosody is properly manipulated. I suspect there is more unrevealed difference between Korean and Japanese even if the big picture in the two languages show similar behavior. I will leave this issue for future works. My last comments must go to Sinhala. In Sinhala, as far as I am aware, no observation regarding the interaction between prosody and intervention effects has been made yet. Sinhala also allows scrambling that is associated with information structure. If a wh-phrase in the intervention configuration has a certain information structural property that triggers movement and intonation can mark such information structural property in Sinhala, this system predicts that Sinhala also shows the amelioration of intervention effects with a help of prosody. I will leave this issue for future work as well. In this section, I discussed how in-situ wh-phrases in Korean/Japanese are licensed. As type D languages, Korean/Japanese allow wh-agreement to take place via parasitic head movement, as happens in Sinhala. However, unlike Sinhala, Pronounce EPP is satisfied by insertion of intonation morpheme, not a lexical morpheme. This microvariation distinguishes Korean/Japanese from Sinhala in a way different from what is suggested by previous studies. I believe that the current account is better able to account for (i) the role of so-called question particles and (ii) obligatory wh-intonation. In addition, the possibility of pied-piping of wh- phrases due to its contrastive focus property accounts for why intervention effects in Korean (and possibly Japanese) are sensitive to prosody. In this case as well, the insertion of intonation morpheme that is phonetically realized with an extra f0 boost on in-situ wh-phrases plays a crucial role in accounting for otherwise puzzling syntax-prosody interface. 4.3. Case 2 – A low-copy pronunciation in discourse-configurational chain In this section, I will discuss the second case that allows low-copy pronunciation due to insertion of an intonation morpheme in Korean. In 4.3.1., I will first introduce the core phenomenon the current discussion is interested in: scope inversion with the help of prosody in so-called rigid scope languages. Based on this background, in 4.3.2., I will discuss an experiment conducted in Kang (2017) that shows that topic movement in Korean can be pronounced with a low copy 172 when appropriate prosody accompanies this. Extending the analysis proposed in 4.3.2, I will discuss focus movement that is reflected in prosody in 4.3.3 in Korean, German and English. And I will conclude this chapter in 4.4. 4.3.1. Background It has been widely reported that in so-called rigid scope languages, scope inversion is occasionally available when appropriate prosody occurs. In this section, I will introduce such puzzling scope inversion phenomenon observed in Japanese, German and Russian. To my knowledge, Kitagawa (1990) is the first study that observes scope inversion with respect to prosody in Japanese. Japanese has been widely assumed to show scope-rigidity when there are more than one quantifier in a single sentence. Thus, without any special intonation, the sentence in (42a) only has a subject-wide scope reading as reflected in its linear word order. However, if a prominent accent on the subject quantifier is followed by a pause, as in (42b), the same sentence suddenly happens to have an additional reading, the object-wide scope reading. (42) a. Dareka-ga daremo-o aisiteiru someone-nom everyone-acc love Someone loves everyone. Some > every, *every > some (Kitagawa 1990, (85)) b. DAREKA-ga // daremo-o aisiteiru someone-nom everyone-acc love Someone loves everyone. Some > every, every > some (Kitagawa 1990, (85)) Watanabe (2000) also briefly mentions this phenomenon in footnote 15. Watanabe additionally provides the example in (43). In (43a), the disjunctive subject scopes over the universal object, not vice versa. However, if the disjunctive morpheme ka is repeated before the nominative case marker –ga with rising intonation on the repeated disjunctive marker and falling intonation on the nominative case marker as in (43b), the sentence becomes ambiguous. (43) a. [John ka Mary]-ga daremo-o sonkeishiteiru or > every, *every > or John or Mary-nom everyone-acc admire 173 b. [John ka Mary KA]-ga daremo-o sonkeishiteiru or > every, every > or John or Mary or -nom everyone-acc admire John or Mary admires everyone. (Watanabe 2000, fn 15) While making interesting observations, the studies do not provide an account for the data they present. Similar phenomena are also observed in Russian and German as well. I will discuss each language in order. Antonyuk-Yudina (2011) conducted a production test to investigate relations between scope interpretation and prosody in Russian. Antonyuk-Yudina asked participants to read out loud a Russian SVO sentence with contexts biasing either the subject-wide scope reading or the object-wide scope reading, as in (44). (44) Target sentence: Несколько аниматоров рисовали каждую героиню. Several animators drew every heroine. a. Surface reading-biased context, SVO Когда рисуют мультфильмы, то каждый из мастеров рисует своего персонажа отдельно. В этот раз сделали исключение и парa лучших аниматоров сделали всю работу сообща. When cartoons are being drawn, each artist draws his own character separately. This time an exception was made and several of the best animators did the whole job together: b. Inverse reading-biased context, SVO Последний Диснеевский мультфильм не зря считается шедевром, в каждую деталь вложено максимум труда, количество задействованных художников просто огромно: The last Disney cartoon is considered a masterpiece for good reasons, each detail has been given maximum attention, the total number of artists used is huge: (Antonyuk-Yudina 2011, (4)) The figure in (45) shows the prosodic contour found in Antonyuk-Yudina’s experiment. The picture in (45a) is the intonation pattern of the subject-wide scope reading in surface-biasing contexts and the picture in (45b) is the intonation pattern of the object-wide scope reading in 174 inverse-biasing contexts. The crucial difference between two intonation patterns is, according to Antonyuk-Yudina, that in the subject-wide scope reading, a perceptually very strong, contrastive pitch accent was assigned on the object determiner while in the object-wide scope reading, a pitch accent on the object determiner was not present. (45) Russian intonation pattern depending on scope (Antonyuk-Yudina 2011, Fig2) a. Subject-wide scope reading b. Object-wide scope reading The results show that when there is a focus accent on object QPs, the object-wide scope reading is hard to obtain. On the contrary, when there is no focus accent on object QPs, and a down step- like pitch decrease appears, the object-wide scope reading is more likely to be obtained. Furthermore, as we look carefully into the intonation contour of the object-wide scope reading in Antonyuk-Yudina’s results (45b), the subject QP has an expanded pitch range compared to the subject QP in the surface reading condition. This looks similar to the Japanese data we just discussed. Another experimental study conducted by Ionin and Luchkina (2015) shows different parts of the same coin. In Ionin and Luchkina’s perception experiment, participants listened to audio stimuli (presented in two word orders, SVO and OVS) and saw pictures that depicted the intended readings (either the surface reading or the inverse reading). Participants chose Yes if they thought the sentence matched the picture or NO otherwise. There are two interesting findings that are relevant to the current discussion. First, in the OVS condition, if the scrambled object is prosodically focused or substituted by a focus particle po krajnej mere odin ‘at least one’, the inverse scope (i.e. the subject-wide scope) reading is more likely to be obtained. Ionin and Luchkina claim that prosodic prominence and adding a focus particle play the same role as a 175 marker of contrastive focus and when the preverbal object is contrastively marked by either prosody or adding a focus particle, the post-verbal subject is more likely to scope over the preverbal object resulting in the inverse scope reading. Second, in SVO word order, contrastive focus on the subject does not facilitate an inverse reading (i.e. the object-wide scope reading). Furthermore, SVO order with neutral prosody strongly prefers the surface scope reading (i.e. the subject-wide scope reading). They interpret this surface scope preference in SOV order as a result of topic-hood of the preverbal subject. In other words, when the preverbal QP is construed as the topic (either due to de-stressed accents or a rise-fall contour), surface scope is strongly preferred. These two studies of Russian each seem to be a half portion of a single puzzle. The inverse scope is obtained only if a preceding QP is focused (Ionin and Luchkina) and a following QP is destressed (Antonyuk-Yudina). In addition, when the subject is interpreted as topic, it always has a wide scope. Even though a specific realization of prosody (i.e. f0 range, or tonal pattern) differs, it seems that the intonation pattern that allows scope inversion in Russian and Japanese shares the same core property: the preceding QP is prosodically prominent and the following QP is de-stressed. Ionin and Luchkina associate the wide scope reading of a subject in preverbal position to its topic-hood and the narrow scope reading of an object to contrastive focus-hood. Unfortunately, however, Ionin and Luchkina (2015) and Antonyuk-Yudina (2011) do not provide a theoretical analysis for the phenomenon beyond the interesting generalization. An similar observation has been reported in German, another so-called rigid-scope language, as well. Since I already discussed relevant German examples in 1.3.2, here I will briefly review the core observation. Similar to Japanese and Russian, when a sentence occurs with quantifiers in both subject position and object position, as in (46), only a surface linear scope reading (a child scopes over five museums) is possible with neutral prosody as indicated in (46a). By way of contrast, if a rising-falling prosody is applied, the inverse scope reading (five museums scopes over a child) becomes available, as well as the surface scope reading, as shown in (46b). (46) Ein /KIND hat \FÜNF Museen besucht a child has five museums visited a. Neutral Prosody (a > five, *five > a) 176 There is a child who visited five museums. b. Special Prosody (Rising-falling intonation following Krifka (1998)) (a > five, five > a) Five museums under consideration have been visited by a potentially different student. What is interesting in the German data for the current discussion is that the rising intonation occurs on the subject QP and the falling intonation occurs on the object QP to get the object-wide scope reading. Rising intonation has been understood as topic accent and falling intonation has been understood as focus accent. Thus, at first glance, this looks contradictory to the general pattern observed in Japanese and Russian. However, if we interpret the rising intonation on a subject QP as a way of expanding pitch range of the subject QP and falling intonation on an object QP as a way of de-stressing the object QP, the German data does not look totally different. Thus, the occurrence of rising-falling intonation is just to make the subject QP prosodically prominent and the object QP prosodically destressed. In sum, Japanese, Russian and German all show the same pattern with respect to the interaction between prosody and quantifier scope interpretation: When the preceding quantifier becomes prosodically prominent by raising its pitch accent and the following quantifier becomes prosodically suppressed by lowering its pitch accent, scope inversion is available. Given that prosody manipulation is the most applicable way to indicate information structure, it seems plausible to say that otherwise impossible scope inversion with a special prosody is associated with change in information structure. Féry and Ishihara (2010) argue that prosodic effects of syntactic structure and information structure should be distinguished. Syntactic structure is suggested to affect prosodic phrasing, while information structure arguably affects F0 scaling of certain prosodic domains. Thus, focus intonation is related with the notion of “prominence” (Jackendoff 1972, Truckenbrodt 1995, Büring 2001), as in (44). Different languages may use different strategies to make a focused constituent prominent (Büring 2010). In Japanese, according to Féry and Ishihara, “the F0 register of a focus is affected such that its reference top line is raised, provoking a sudden boost of the pitch accent correlating with the focused word or exponent” (Féry and Ishihara (2010: p42). In addition to focus prominent, they propose Destress-Given in (45) for given constituents. They noted that “when given material appears in a sentence…this part of the sentence (if any) is (prosodically) compressed” (Féry and Ishihara (2010: p42)). 177 (44) Focus Prominence Focus is realized by prominence in its focus domain. (45) Destress–Given A given phrase is prosodically non-prominent The prosodic pattern that allows scope inversion in Japanese, Russian and German now can be interpreted as that the prosodically prominent subject indicates the subject is information structurally focused and the prosodically destressed object is information structurally given. I would further assume that the destressed objects in the scope inversion context are actually information structurally topic, as familiar topics (Frascarelli and Hinterhölzl 2007). One might say that it is premature to say that this givenness prosodic property indicates that the phrase is a topic phrase just based on the intonation pattern. This objection sounds plausible since it is not the case that every given element is a topic. However, what would happen if the object quantifier were to be a given phrase, so that it does not move to a higher position (by hypothesis)? One might argue that the focused subject could be reconstructed to its base position at LF. Even in that case, since the base position of the subject is higher than the base position of the object, there is no way the object could scope over the subject without object movement. This means, in order to obtain an object-wide scope reading, object movement must take place to somewhere higher than (at least) the base position of the subject, which is not quite attested in those languages in question. This being so, it is more plausible to assume that the object is a topic phrase and so undergoes topic movement to a higher position than to suggest that the object is a given (but not topic) phrase and the given phrase moves to a higher position for some unknown reason. One might wonder whether the scope inversion described here is just a result of prosodic illusion. In other words, when a quantifier is pronounced with a focus accent and another quantifier is pronounced with a suppressed accent, the quantifier with the suppressed accent takes wider scope regardless of the syntactic structure as a direct matching between prosody and interpretation. My answer to this question is no. In the next section, an experiment on Korean shows that prosody does not affect the scopal relations of every kind of quantifier. Only 178 quantifiers that are able to move in syntax can be affected by prosody in such a way. This clearly shows that prosody is not magically able to change scope interpretation regardless of syntax. Now, we have a consensus phenomenon across Japanese, Russian and German. In a sentence with two quantifiers, when the preceding quantifier is focus and the following quantifier is topic, the linearly following quantifier can scope over its preceding quantifier. In the next section, I will discuss an experiment conducted in Kang (2017) that shows similar observation in Korean. In that study, I designed the experiment to explore whether the wide scope reading of the object in SOV order is a consequence of syntactic movement. If this is the case, then we can find another example that shows low-copy pronunciation in a movement chain. 4.3.2. Korean topic movement In this section, I will introduce an experiment, which is also reported in Kang (2017). This section provides detailed information. I refer to Kang (2017) for a summarized version of this section. 4.3.2.1. Hypothesis It has been widely assumed that topic occupies a higher position than focus. For example, Beghelli and Stowell (1996) posit RefP as the highest position in a syntactic structure and identify the Specifier of RefP as the topic position. Benincà and Poletto (2004) argue against Rizzi (1997)’s lower (than focus) topic and propose that there is no topic lower than focus in Italian. Hsu (2008) also argues that TopicP is higher than FocusP in Chinese. Adopting such a prevalent assumption, I assume that a topic element moves higher than a focus element in Korean. Coupled with the structure hierarchy between topic and focus, I assume a scope assignment principle which is a slightly revised version from one proposed in Krifka (1998). Frey (1993) proposes a scope assignment principle, which is adopted in Krifka (1998) as in (46). 179 (46) If A, B are operators occurring in a sentence S, then S has a reading in which A has scope over B if and only if: a. A c-commands B, or b. A c-commands a trace of B. (Krifka 1998: 76) Since I adopt the copy theory of movement, there is no trace, but copies in the system I proposed. Thus, the scope assignment principle can be restated as in (47). (47) If A, B are operators occurring in a sentence S, then S has a reading in which A has scope over B if and only if A c-commands any copy of B. Given the two assumptions, topic over focus and a scope assignment principle under the copy theory of movement, I hypothesize that a topic phrase moves to a position higher than a focus phrase, but a low copy of the topic movement chain can be phonetically realized when an intonation morpheme is inserted to satisfy Pronounce EPP. In the next section, I will discuss an experiment designed to test this hypothesis. 4.3.2.2. Design and Materials The main purpose of the experiment is to investigate whether it is possible for prosodically destressed object QP to scope over its preceding prosodically prominent subject QP in Korean, as attested in other languages. In addition, I will investigate whether this scope inversion is a result of covert topic movement (i.e. low-copy pronunciation of a topic movement). To test the hypothesis, I need to compare two types of sentences schematized in (48). (48b) is what the experiment is interested in and (48a) is for comparison. (48) a. Subject QP = focus, Object QP = focus in SOV b. Subject QP = focus, Object QP = topic in SOV Target sentences used in the experiment were in SOV word order and both subject and object were quantifier phrases. The subject quantifier was either choisohan NP han myeng ‘at least one 180 NP’ or choisohan NP twu myeng ‘at least two NPs’ (counter-balanced) throughout the test. According to Krifka (1999), QPs with at least is a focus phrase and in SOV word order, the subject in the sentence initial position gets highest pitch in neutral intonation due to down-step wise pitch pattern in Korean. Thus, the subject QPs in target sentences were all focus elements prosodically and semantically. I manipulated two factors with three levels each: (i) quantifier types (motun ‘all/every’ NP, numeral NP, modified numeral NP (# or more)) and (ii) prosody (focus accent on quantifier, focus accent on NP, no accent), for a total of nine conditions as shown in Table 1. Table 1. Experiment Design 3 X 3 Design Quantifier types X Prosody (motun vs. # vs. # or more) (NP focus vs. Q focus vs. no accent) Prosody NP focus Q focus No accent QP types motun motun [NP]foc [motun]foc NP motun NP # [NP]foc # CL NP [# CL]foc NP # CL # or more [NP]foc # or more NP [# or more]foc NP # or more Three different quantifier types were tested, motun NP, numeral NP, and # or more NP as exemplified in (49). (49) a. motun NP Choisohan uysa han myeng-i motun kanhosa-lul chingchanhay-ss-ta. At least doctor one CL-nom motun nurse-acc compliment-past-decl. At least one doctor complimented every/all nurse(s). b. numeral NP Choisohan uysa han myeng-i kanhosa sey myeng-ul chingchanhay-ss-ta. At least doctor one CL-nom nurse three CL-acc compliment-past-decl. At least one doctor complimented three nurses. c. # or more NP Choisohan uysa han myeng-i kanhosa sey myeng isang-ul chingchanhay-ss-ta. At least doctor one CL-nom nurse three CL more-acc compliment-past-decl. At least one doctor complimented three or more nurses. 181 The quantifier motun can be translated either as every or all in English. In English and many other languages, every NP in object position is the most widely studied QP regarding scope inversion. As a distributive quantifier, every NP in English and many other languages can take wide scope relatively freely. However, all is not distributed over other quantifiers. Since it is good to investigate patterns that can be compared with other languages, motun NP was selected for the test. Furthermore, it has been reported in previous studies that a numeral NP in object position can be distributed over a subject QP, but a modified numeral NP like # or more NP cannot (Liu 1990, Beghelli & Stowell 1997, Szabolcsi 1997) in many languages. I suspect that the difference comes from their ‘topic-hood’. A topic phrase must be specific (É Kiss 1998), but modified numeral quantifiers resist a specific reading (Beghelli and Stowell 1997). The combination of these two premises predicts that modified numeral quantifiers cannot occur as a topic phrase. This prediction is compatible with the observation that modified numeral quantifiers cannot have a thematic topic reading when a topic marker is attached to it in Korean as in (50). In Korean, (as I discussed in chapter 1), the topic marker –nun can mark a thematic topic when it is attached to a subject. However, if the subject is a modified numeral quantifier as in (50a), -nun cannot cause a thematic-topic reading, but only a contrastive-list reading. This is contrary to the numeral quantifier with topic marker in (50b), which can cause a thematic-topic reading. Therefore, I assume that numeral quantifiers can be topic phrases, while modified numeral quantifiers cannot. Due to this difference, I selected those two types of quantifiers to test. If both types of quantifiers in object position can have wide scope reading when they are prosodically destressed, it implies that the wide-scope reading is irrelevant to topic-hood of quantifiers. In that case, we could think of direct prosody-scope matching or other non-syntactic mechanisms. By way of contrast, if there is difference in the possibility of the wide scope reading between the two types of quantifiers, it implies that the scope taking with respect to prosody is relevant to topic-driven movement 66 . 66 Audrey Li (p.c) pointed out that in Korean, modified numeral quantifiers can be fronted (as other topic elements are scrambled to the front of the sentence). In addition, modified numeral quantifiers in Chinese seem to be topics even though Chinese also has a specificity restriction on topic. Those points are true. However, the fronted modified numeral quantifiers in Korean pattern different from regular numeral quantifiers in the interpretation, as in (i). The interpretation indicates that modified numeral quantifiers cannot be topic even though it is fronted. 182 (50) a. Haksayng sey myeng isang-un manhwachayk-ul ilk-ess-ta. Student three CL more-top comic book-acc read-past-decl. Contrastive (list) reading: Three or more students read comic books (and four or more students read novel…) *Thematic topic reading: There are three or more students who read comic books. b. Haksayng sey myeng-un manhwachayk-ul ilk-ess-ta. Student three CL-top comic book-acc read-past-decl. Thematic topic reading: There are three students who read comic books. For prosody factors, I differentiate accents on objects in three ways. According to previous literature on Korean intonation (Jun 1993, 2005, Oh 2008, Lee 2000, Jo et al. 2003 among many others), prosodic accent makes focus more prominent on an element than its neighbors, and topic is distressed accent-wise, as in Japanese. More concretely, Jun (2005) presents Figure 1 to show how a focused word is realized with an expanded pitch range, lengthening, and large amplitude. Fig. 1. (a) is a pitch track of the sentence EmEni minari mEGnINdeyo ‘Mother eats dropwort’ in neutral intonation (no focus on any word). The F0 contour decreases (like down-step) over the course of the sentence. When the object minari ‘dropwort’ is focused in the same sentence, the pitch track (b) shows that the focused word has a raised f0 peak (precisely, on the second syllable based on LHLH pattern in AP) which is higher than the preceding word. (i) a. Haksayng sey myeng isang-ul John-i haksayng sey myeng isang-ul kaluchye-ss-ta. Student three CL more-Acc John-Nom teach-past-decl. John taught students x and |x| ≥ 3. #There are more than three students x such that John thought x. b. Haksayng sey myeng-ul John-i haksayng sey myeng-ul kaluchye-ss-ta. Student three CL-Acc John-Nom teach-past-decl. There are three students x such that John taught x. I do not have a full developed answer for the Chinese pattern, but suspect that something similar happens in Chinese as well. 183 Fig. 1. Jun (2005)’s example of focused item in Korean (Jun 2005: 18). a. b. Even though focus intonation in Korean has been studied intensively, topic intonation has been poorly studied. Lee (2000) and Oh (2008) investigate the intonation of sentence-initial topics with topic marker –(n)un, but to the best of my knowledge, no research seriously discusses sentence-medial topic intonation without the topic marker. Thus, topic intonation in sentence medial position in Korean remains undiscovered. Consequently, in this experiment, under the assumption that topic, as a given element, is de-accented (Destress-given), I assume that a sentence-medial topic constituent is assigned a lowered f0 as a consequence of down step-like lowering, as in Fig.1.(a). This prosodic contour is also comparable to the one used in Japanese, Russian, and German for scope inversion. Based on the focus/topic intonation we have discussed, I manipulated prosody conditions in three ways. Basically, two object-focus conditions and one object-topic condition were tested. The first two condition assigns a prominent accent on objects. The difference between the two condition is that prominent accent is on a noun phrase that a quantifier modifies in the first condition, but that is on a quantifier in the second condition. Since prominent accent can be assigned either on NP or quantifiers in Korean, both cases were all tested in the experiment. In the third condition, a prominent accent does not occur on any part of an object QP. Thus, such instances have the prosodic shape depicted in Fig.1.(a). Figure 2 shows samples of target audio stimuli. Each target sentence was recorded using a head-mounted microphone in a sound-attenuated booth by a male native speaker of Korean. Speakers were trained to make a clear distinction between different prosody conditions. Recording was conducted using Praat and saved as separate wave files. 184 Figure 2. Samples of target audio stimuli a. motun ‘every/all’ NP (49a) a. NP focus b. Q focus c. No accents b. numeral NP (49b) a. NP focus b. Q focus c. No accents c. modified numeral NP (49c) a. NP focus b. Q focus c. No accents Jun and Lee (1998) argue that Seoul Korean speakers manipulate pitch range difference between a focused word and the following word to enhance the prominence of a focused item. Thus, the pitch range between a focused word and the following word is larger than the pitch range between two non-focused words. Phonetic analysis of the audio stimuli was conducted based on 185 this observation. Figure 3 shows the results of phonetic analysis of audio stimuli. Since each quantifier condition requires a different word order, I present each result in separate charts. Figure 3. Changes of peak f0 value over the sentence a. The motun NP condition The pitch range between motun and the following NP is more enhanced (sudden falling) in the Q focus condition than in the no accent condition (the NP focus condition cannot be compared here because in that case, the following word is focused). The pitch range between the NP and the following verb is more enhanced in the NP focus condition than in the other conditions. b. The numeral NP condition The pitch range between the NP and the following numeral phrase is more enhanced (sudden falling) in the NP focus condition than in the no accent condition (the Q focus condition cannot be compared here because in that case, the following word is focused). The pitch range between a numeral phrase and the following verb is more enhanced in the Q focus condition than in the other conditions. C. The modified numeral NP condition The pitch range between the NP and the following modified numeral phrase is more enhanced (sudden falling) in the NP focus condition than in the no accent condition (the Q focus condition cannot be compared here because the following word is focused). The pitch range between a modified 0 50 100 150 motun NP verb Mean of F0 peak NP focus Q focus No accent 0 50 100 150 200 NP Number verb Mean of F0 peak NP focus Q focus No accent 0 50 100 150 200 NP # or more verb Mean of F0 peak NP focus Q focus No accent 186 numeral phrase and the following verb is more enhanced in the Q focus condition than in the other conditions. In summary, for all of the three quantifier conditions, when the object NPs were focused, the pitch range between the focused NP and the following word was enhanced compared to the no accent (neutral) condition. Thus, focused NPs were recorded to be prominently perceived than its neighbor phrases. Likewise, for all of the three quantifier conditions, when quantifiers were focused, the pitch range between the focused Q and the following word was significantly enhanced compared to other conditions indicating the prominence of quantifiers. Mean F0 values also show that there is clear distinction between prosody conditions. Figure 4 summarizes the mean f0 values of NPs and quantifiers in each condition. In every quantifier condition, when NPs were focused, the mean f0 of NPs was higher than in the other prosody conditions. Likewise, when quantifiers were focused, the mean f0 of the quantifiers was higher than in the other prosody conditions. Figure 4. Mean f0 value of NPs and quantifiers. Each auditory target sentence was presented with two pictures which indicated two potential readings of the sentence, a subject-wide scope reading and an object-wide scope reading, as shown in Figure 5 (the English version is in Figure 6). I will go through an example using the sentences provided in (49), repeated here in (51). For each item, the motun NP condition are presented with images like (a) in Figure 5/6 and the numeral and modified numeral conditions are with images like (b). In motun conditions, circles around the set of teachers were 0 50 100 150 200 NP Quantifier NP Quantifier NP Quantifier Motun Numeral Modified numeral Mean f0 NP focus Q focus No focus 187 added to ensure that participants interpret the image as depicting a situation where all nurses (every nurse) were complimented by one doctor. The subject-wide scope reading image (the left- side image in each condition) depicts a situation where at least one doctor complimented three nurses (every nurse in the motun condition). I used arrows to depict the events mentioned in the sentences. Participants became familiar to the meaning of the arrow in a practice session. In addition, the different characters – in this case doctor and nurse – were labeled in Korean to ensure participants knew which was which. The object-wide scope reading image (right-side image in each condition) for this item shows a situation where three nurses in the picture were complimented by at least one doctor (who could potentially be different for each nurse). Both subject-wide scope reading and object-wide scope reading images were presented for each target sentence to see whether two readings were available. Figure 5. Sample targets (used in the experiment) a. Motun NP b. numeral/modified numeral Figure 6. Sample targets (translated in English) a. Motun NP b. Numeral/modified numeral 188 (51) a. motun NP Choisohan uysa han myeng-i motun kanhosa-lul chingchanhay-ss-ta. At least doctor one CL-nom motun nurse-acc compliment-past-decl. At least one doctor complimented every/all nurse(s). b. numeral NP Choisohan uysa han myeng-i kanhosa sey myeng-ul chingchanhay-ss-ta. At least doctor one CL-nom nurse three CL-acc compliment-past-decl. At least one doctor complimented three nurses. c. Modified numeral NP Choisohan uysa han myeng-i kanhosa sey myeng isang-ul chingchanhay-ss-ta. At least doctor one CL-nom nurse three CL more-acc compliment-past-decl. At least one teacher complimented three or more students. All of the experimental pictures were matched in animacy, such that every subject NP and every object NP referred to a human (e.g. boy, student, doctor). Two left/right mirror image versions were created for each target picture. The left-right positions of the subject-wide scope reading and the object-wide scope reading were counterbalanced. I used a Latin-square design such that one participant saw one version of each picture, and each version of each picture was seen an equal number of times across participants. In addition to target trials, each participant was given 64 fillers, 32 fillers with anaphora interpretation and another 32 fillers with time adverb attachment. The fillers were designed to have ambiguous interpretations, in order to make the potentially ambiguous properties of targets less striking, thereby making it harder for participants to guess the purpose of the experiment. The fillers were presented with two pictures depicting two possible readings each. I did not include any scope bearing element in the fillers to avoid any possible interaction between targets and fillers in terms of scope. 4.3.2.3. Participants Thirty-six adult native Korean speakers participated in the experiment. None of the participants reported being bilingual in Korean and another language. The participants were living in Korea 189 at the time of testing. They received 5-10 U.S. dollars for their participation depending on the duration of testing. 4.3.2.4. Procedure The experiment was conducted in a sound-attenuated booth. Participants saw 100 Power Point slides - 36 target trials and 64 filler trials. When they clicked a speaker button at the top of each slide, they heard audio stimuli over a head-mounted headset. They were given an answer sheet and were instructed to give scores from 1 to 5 (1 is the worst and 5 is the best) based on how they thought the sentence they heard could be naturally used to describe each picture in a slide. They were instructed to give scores for each picture independently. In other words, both pictures could receive high scores or low scores or middle scores. Participants were first introduced to the task with two practice trials, one in which a given sentence could only be used for one of the images and the other in which a given sentence was not perfectly plausible, but not impossible, for both images. They were also instructed to ask the experimenter if they could not understand what each picture was intended to mean. 4.3.2.5. Predictions Based on the hypothesis, I predict (52). (52) Predictions The object QP will be more likely to have wide scope in SOV order when a. The object QP is prosodically destressed and b. The object QP is a type of QP that can undergo topic movement (numeral quantifiers, not modified numeral quantifier, not sure motun-NP) The second prediction in (52b) is particularly interesting in a sense that it shows whether the object-wide scope reading is syntax-mediated or not. 190 4.3.2.6. Results For each condition, the dependent measure was the scores participants gave. Table 2 shows the average scores of the subject-wide scope reading for each of the nine conditions. Table 3 shows that the average scores of the object-wide scope reading for each of the nine conditions. As can be seen in Figure 7, the subject-wide scope reading had higher scores regardless of QP types or prosody. To see the effects of prosody or quantifier types on the object-wide scope reading, which this experiment was primarily interested in, the acceptability scores of the object-wide scope reading for each condition were plotted in Figure 7. Here we can see that, in the numeral quantifier condition, the score is higher in the no accent condition than in the NP focus or Q focus conditions. By way of contrast, with the other quantifier types (motun and modified numeral quantifier), prosody does not seem to affect the acceptability. To assess these results statistically, I conducted analyses of variance (ANOV As). Participant and item means were z- score transformed. All ANOV As had the following two factors: Quantifier types (motun vs. numeral quantifier vs. modified numeral quantifier) and Prosody (NP focus vs. Q focus vs. No accent). ANOVA results show that regardless of prosody, the acceptance rate of the object-wide scope reading was significantly higher in the numeral condition than in the other two quantifier conditions in both by-subjects analysis and by-items analysis (main effect of quantifier types: F1(2,70) =5.405001, p<0.05, F2(2,70) =6.177798, p<0.01). There was no significant main effect of prosody. In addition, both by-subject analysis and by-item analysis show that there was a significant interaction between the two factors (F1(4,140) =2.489319, p<0.05, F2(4,140) =2.520780, p<0.05). Table2. Mean of scores (Subject-wide reading) Prosody NP focus Q focus No accent QP types Motun 4.347222222 4.340277778 4.416666667 Numeral 3.659722222 3.659722222 3.513888889 Modified numeral 3.777777778 3.756944444 3.833333333 191 Table 3. Mean of scores (Object-wide reading) Prosody NP focus Q focus No accent QP types Motun 2.25 2.375 2.548611111 Numeral 2.222222222 2.152777778 2.215277778 Modified numeral 2.145833333 2.243055556 2.159722222 Figure 7. Mean of scores Figure 8. Interaction between quantifier types and prosody in the object-wide scope reading. 0 0.5 1 1.5 2 2.5 3 3.5 4 4.5 5 Mean of scores Inverse Surface 1.5 1.7 1.9 2.1 2.3 2.5 2.7 NP_focus Q_focus No accent Mean scores of Object-wide reading motun num more 192 To better understand the source of the interaction, I conducted paired t-tests for two cases this experiment was interested in: 1. Prosody effects within the same quantifier type, 2. The effects of different quantifier types within the same prosody. The results are summarized in tables 4 and 5. Two main observations are found 67 : 1. Numeral quantifiers are the only quantifier type that is affected by prosody. In Table 4, the only significant effect is observed between the NP focus condition and the No accent condition in the Numeral quantifier condition. One marginal effect is observed between the Q focus condition and the No accent condition in the Numeral quantifier condition as well. All the other paired t-tests show no significant p-value. This implies that numeral quantifiers are affected by prosody, but not the other types of quantifiers. In addition, when there is no accent on the object quantifier, the object is more likely to have wide scope than the case where any prominent accent is on it. 2. When an object quantifier is not assigned any prominent accent, the acceptability of the object-wide scope reading is significantly different depending on the types of object quantifiers. As shown in Table 5, when the object quantifier is a numeral quantifier, then it is significantly more likely to have wide scope than when the object quantifier belongs to other type of quantifiers 68 . Table 4. Paired t-test (Prosody factors within the same quantifier type) Paired conditions P-value Motun NP focus Q focus 0.1977 Q focus No accent 0.4011 NP focus No accent 0.6576 Numeral quantifier NP focus Q focus 0.1479 Q focus No accent 0.09479 NP focus No accent 0.000677*** Modified numeral quantifier NP focus Q focus 0.3294 Q focus No accent 0.4636 NP focus No accent 0.8723 67 Bonferroni-corrected pairwise test also shows the same results. 68 When Q is focused, numeral quantifiers are significantly more likely to have an object-wide scope reading than motun NP. I do not know what this means for now, thus I will put this result aside. 193 Table 5. Paired t-test (Quantifier type factors within the same prosody) Paired conditions P-value NP focus motun Numeral 0.696 Numeral Modified numeral 0.2099 motun Modified numeral 0.08877 Q focus motun Numeral 0.03929* Numeral Modified numeral 0.1267 motun Modified numeral 0.5486 No accent motun Numeral 0.001342** Numeral Modified numeral 3.874e-06*** motun Modified numeral 0.3179 Therefore, the findings can be summarized as follows: Regardless of quantifier types or prosody, the subject-wide scope reading is preferred over the object-wide scope reading. Even though the overall acceptance rate of the object-wide scope reading is low, there is a significant difference among the conditions investigated. When the object quantifier is a numeral quantifier, it is more likely to have wide scope than when the object quantifier is motun or modified numeral quantifiers. In addition, when the object quantifier is a numeral quantifier, the acceptability of the wide-scope reading is affected by prosody (i.e. when there is no accent on the object quantifier, it is more likely to have the object-wide scope reading) while when the object quantifier is a motun or a modified numeral quantifier, prosody does not seem to play any role to improve the object- wide scope reading. Furthermore, when there is any accent on objects, the object-wide scope reading is not preferable regardless of quantifier types, while when there is no accent on objects, the object-wide scope reading is more likely to be obtained only when the object is a numeral NP. 4.3.2.7. Discussion The experiment results confirm the predictions in (52). Similar to Japanese, Russian and German, Korean also shows prosody sensitivity to scope interpretation. However, this interaction between prosody-scope interpretation is crucially quantifier-type sensitive: Only those types of quantifiers that can be topic are able to have wide scope with the help of prosody. This implies that this prosody effect is a reflection of syntactic movement, not just a case of pragmatic 194 illusion. Based on the results, I argue that object QPs that have a wide scope undergo movement over the focused subject due to topic agreement as depicted in (53). However, Pronounce EPP is satisfied by insertion of an intonation morpheme that is phonetically realized with destressd pitch accent. Thus, the movement chain does not need to be pronounced with a high copy in the moved position being free from the Pronounce EPP requirement, so the low copy is instead pronounced due to Earliness under cyclicity. Therefore, the Korean (and arguably Japanese, Russian, and German) topic movement shows another example of low-copy pronunciation in a chain due to an insertion of an intonation morpheme. (53) TopP Obj.QP TopP Top TP Subj.QP TP T vP Subj.QP vP ObjQP There is another issue I need to address regarding the results of the experiment. When the object QP is assigned prominent accent (either on NP or quantifier), the object QP consistently has narrow scope. Does it mean that the object QP stays in its base position? My answer is no. As we will see in the next section, a focus phrase that is indicated with prosodic prominence in Korean seems to undergo movement as well. I claim that those in-situ focus phrases undergo focus movement but the chain is pronounced with a low copy in the base position. If this is the case, why is it impossible for the prosodically focused object QPs in the reported experiment to have wide scope? I claim that this is because even though the object QP undergoes covert focus movement (by hypothesis), it must move to a position lower than where another focus element, the subject QP, is moved to (Tucking-in fashion proposed in Richards 1997) due to Superiority. In other words, the focus movement of the object QP is masked by another phrase that undergoes focus movement (See Miyagawa 2010 for superiority effects in multiple focus movement in 195 Japanese, which applies to Korean as well). The locality issue plays a role between two movements that are triggered for the same reason (focus agreement). However, such a locality issue does not arise when the object QP is topic. Thus, we can easily detect the topic movement, but not focus movement of the object QP by looking at relative scope interpretation. Before closing this section, I would like to point out one more interesting result in the experiment. Similar to the modified numeral QP condition, the motun-NP condition rarely allows an object-wide scope reading regardless of prosody. This implies that at least in cases where the quantifier appears in non-scrambled object positions, Korean motun patterns more likely to be the collective quantifier all in English, than the distributive quantifier every. In addition, the overall preference of the subject-wide scope regardless of prosody can be accounted for by many possible different factors: reconstruction of Ā-movement, processing cost theory (Anderson 2004) or other analyses. The detailed analysis for this pattern is beyond the scope of this work. 4.3.3. Korean focus movement In 4.2. I reported that if a wh-phrase has an additional accent than a normal wh-phrase has in Korean, the wh-phrase is free from intervention effects. To account for this prosody effect, I argued that when a wh-phrase has an extra accent, it undergoes movement due to its contrastive focus property. In this section, I will show that in-situ non-wh-focus with prominent focal accent also show “high” behavior as if it undergoes movement. In Yun (2012)’s experiment which I introduced in 4.2, another interesting observation is reported. Target sentences in the experiment had an indeterminate phrase in a conditional clause. The indeterminate phrase can be interpreted in three different ways depending on intonation, as indicated in (54). (54) I pyeng-un mwe-lul mek-umyen na-a (Yun 2012, (2)) this illness-Top what-Acc eat-if cured-INT (Lit. ‘This illness will be cured if you eat [what/something].’) i) narrow scope indefinite (if > ∃) This illness will be cured if you eat something. ii) wide scope indefinite ( ∃ > if) 196 There is a certain something such that this illness will be cured if you eat it. iii) wh-interrogative What is the thing such that this illness will be cured if you eat it? The results of the experiment show that each possible interpretation has a characteristic prosody pattern. As already discussed, the characteristic prosody for wh-interrogatives is dephrasing after the wh-phrase (see 4.2.). What is interesting for the current discussion is the distinction between the meaning in (54i) and in (54ii). When the sentence is pronounced with a neutral intonation as in (55a), the indefinite narrow scope reading (54i) is the most salient reading. On the other hand, when a prominent accent is assigned on the indefinite by F0 boost without dephrasing (55b), the indefinite wide scope reading (54ii) is the most preferred one. Based on these results, Yun argues that wh-indefinites in Korean are ambiguous in their scope configuration, but a prominent accent (i.e. f0 boost) on an indeterminate phrase makes the indefinite wide scope reading salient. (55) Target audio stimuli in Yun (2012)’s experiment (Yun 2012, 289) a. Neutral b. F0 boost on wh-indeterminate 197 I adopt this observation as a piece of evidence that shows focus movement of the in-situ indefinite. More specifically, the indefinite moves out of the conditional clause in overt syntax making two copies in its chain. In LF, it is interpreted with the high copy in the moved position, but in PF, it is pronounced with the low copy in the base position. The low-copy pronunciation is allowed because the insertion of an intonation morpheme that is phonetically realized with f0 boost satisfies Pronounce EPP instead. So far, I have discussed that in Korean, topic or focus phrase movement can be pronounced with a low copy when an intonation morpheme is inserted to satisfy Pronounce EPP. In the next section, I will discuss more about properties of topic/focus movement in Korean. Especially, I will introduce the “discourse-configurational language” type suggested in É Kiss (1995) and discuss how the current analysis accounts for discourse-configurational languages, which is categorized as type E languages in chapter 3.2. 4.4. Return to the typology of subjects: Type E languages, “discourse-configurational languages” In 3.2., I categorized Korean and Japanese into type E languages, in which discourse-features are involved in subject-agreement. I will start this section introducing the notion of discourse- configurational languages, suggested in previous studies, particularly in É Kiss (1995) as in (56). (56) Discourse-configurational languages Primary sentence articulation is motivated by discourse-semantics, rather than theta-role or case considerations. (É Kiss 1995:3) A. The (discourse-) semantic function ‘topic’, serving to foreground a specific individual that something will be predicated about (not necessarily identical with the grammatical subject), is expressed through a particular structural relation (in other words, it is associated with a particular structural position). (É Kiss 1995: 6) B. The (discourse-) semantic function ‘focus’, expressing identification, is realized through a particular structural relation (that is, by movement into a particular structural position). (É Kiss 1995: 6) 198 É Kiss suggests that the structural relation between a subject and a predicate in languages like English is actually the same as the structural relation between a topic and a predicate in discourse-configurational languages. Such an idea, especially with regard to so-called ‘Topic- prominent languages’ has been explored by a number of works (See É Kiss (1995) and references therein). Miyagawa (2010), developing this idea, proposes that as a subject agrees with T in English, a topic or a focus phrase in discourse-configurational languages agrees with T (or α, a functional head above T when there is more than one topic/focus) and undergoes A- movement to SpecTP (or SpecαP). Thus, so-called local scrambling can be reduced to focus/topic movement as a consequence of focus/topic agreement with T. This account captures the observation that the scrambling is closely related with information structure and the A- movement property of local scrambling in Korean and Japanese. In what follows, I will discuss Miyagawa (2010) focusing on topic/focus movement targeting SpecTP, an A-position. Miyagawa argues that in Japanese canonical SOV word order with neutral intonation, the subject is interpreted as topic, and moves to SpecTP. That is why the subject quantifier can scope over the negation (which is assumed to be between T and v) in (57a) but not vice versa. Miyagawa says that this indicates the lack of reconstruction in A-movement. Interestingly, if there is a local scrambling of the object over the subject quantifier as in (57c) as a consequence of topic/focus agreement, what moves to SpecTP in such cases is not the subject QP, but the object. Thus, the subject remains in its base-position, SpecvP, and we find scope ambiguity between negation and the subject QP 69 . This account captures the intuition that in canonical SOV sentences, the subject patterns as the topic of the sentence, while in a sentence with object local scrambling, the object may be interpreted as the topic of the sentence (noting that a topic feature is the default featural specification of T). (57) a. Zen’in-ga siken-o uke-nakat-ta. (Japanese, Miyagawa 2010: 73-74) all-nom test-acc take-neg-past All did not take the test.’ *not > all, all > not 69 However, it seems not clear to me how the subject-wide scope reading is available in (57b) if we assume that the subject never moves. I believe that in this case, the subject undergoes focus movement over the negation but below the object that undergoes topic movement. 199 b. Siken-oi zen’in-ga ti uke-nakat-ta. test-acc all-nom take-neg-past All didn’t take the test.’ not > all, all > not The A-movement property of local scrambling is indicated in (58a) as well. It has been assumed that a phrase in an A-position can be a binder for anaphora, while a phrase in an Ā- position cannot (Mahajan 1990, Saito 1992 among others). In (58a), Taroo-to Hanako-o ‘Taro and Hanako-acc’ undergoes local scrambling to T according to Miyagawa and serves as a binder for the anaphor otagai ‘each other’ indicating this type of movement is A-movement. On the other hand, the long-distance scrambled Tarro-to Hanako-o ‘Taro and Hanako-acc’ in (58b) cannot serve as a binder resulting in the ungrammaticality of this example. This has been discussed as a piece of evidence that long-distance scrambling in Japanese is Ā-movement. (58) a. Taroo-to Hanako-oi otagai-no sensei-ga ti suisensita. Taro-and Hanako-acc each.other-gen teacher-nom recommended Taro and Hanako, each other’s teachers recommended. b. ?*Taroo-to Hanako-oi otagai-no sensei-ga [koutyou-ga ti sikaru to] omotta. Taro-and Hanako-acc each.other-gen teacher-nom principal-nom scold C thought Lit. Taro and Hanako, each other’s teachers thought that the principal will scold. (Japanese, Miyagawa 2010: 61) Thus, in the examples above, the subject in (57a) and the object in (57b) and (58a) in Japanese seem to undergo topic movement to SpecTP. A similar pattern is observed in Korean (I will not repeat the same observations in the Korean version). If it is true that in Korean and Japanese, a topic phrase (and a focus phrase) undergoes movement to SpecTP as proposed by Miyagawa, how could we incorporate those observations into the proposed system? To put it differently, why is some topic/focus movement pronounced with a high copy in SpecTP but others with a low copy in its base position with appropriate intonation? Under the system developed in this dissertation, both types of movement (i.e. so-called overtly scrambled one and covertly moved one with intonation marker) take place in syntax. The only difference is in which copy is to be pronounced. If this is the case, it is predicted that the two types of movement show 200 the same syntactic properties. In what follows, I will show that scrambled topic/focus that is overtly realized with a high copy and the in-situ topic/focus with proper prosody actually share the same syntactic property with respect to i) creating a binding relation, ii) parasitic gap licensing and iii) weak cross over in Korean. In addition, since they share the same syntactic architecture, the two movements are expected to show interwoven locality constraints. I will discuss the issue as well. As Miyagawa mentions with (58), a topic phrase that undergoes local scrambling can create A-binding. Likewise, when a topic phrase is pronounced with a low copy, but with appropriate intonation, as in (59b), the in-situ topic can also create A-binding. If the sentence is pronounced with a neutral intonation (59a), his mother cannot be co-indexed with John’s mother. However, if there is a focal accent on the subject case marker followed by a pause as in (59b), the co-indexation between his in subject position and John in object position sounds much better 70 . (59) a. Ku i-uy sensayngnim-i John i-ul chwuchenhay-ss-e. (Korean) His teacher-Nom John-acc recommend-past-decl. Lit. His teacher (=John’s teacher) recommended John. b. ?Ku i-uy sensayngnim-I || John i-ul chwuchenhay-ss-e. His teacher-NOM || John-acc recommend-past-decl. Lit. His teacher (=John’s teacher) recommended John. 70 If we use a self-anaphor instead of a pronominal anaphor as in (ia), the intonation cannot help to create A- binding. Likewise, an overtly scrambled topic phrase also cannot create A-binding for the self-anaphor (ib). (i) a.* Caki i-uy sensayngnim-I || John i-ul chwuchenhay-ss-e. self teacher-NOM || John-acc recommend-past-decl. Lit. Self’s teacher (=John’s teacher) recommended John. b. *John i-ul caki i sensayngnim-i John i-ul chwuchenhay-ss-e. John-acc self teacher-Nom John-acc recommend-past-decl. Lit. Self’s teacher (=John’s teacher) recommended John. 201 The intonation used in (59b) is similar to the one used for scope inversion. In other words, when the subject is interpreted as focus and the object is interpreted as topic, the object antecedent can bind its linearly preceding subject anaphor. Focus movement of objects looks more complex. When a focus phrase is pronounced with a high copy (as a case of scrambling) as in (60a), it seems to be able to create a A-binding. However, when the focus phrase stays in-situ and a prominent accent is assigned on the in-situ focus phrase, it does not seem to be able to bind the linearly preceding anaphor (60b). At first glance, this seems to argue against the claim that the two movements are the same except where to be pronounced in a chain. However, I argue that the difference does not come from the focus movement per se, but from the interpretation of the subject. In (60a), the subject that linearly follows the scrambled object can be interpreted as a given phrase. However, in (60b), the sentence initial subject is strongly preferred to be interpreted as topic. Since topic sits in higher than focus in Korean (as discussed in 4.3.), the object cannot bind the anaphor even though it undergoes focus movement. Even when the preceding subject is pronounced with an extra accent to get rid of topic-hood, the binding is still impossible and this impossibility of binding is also predicted in the current proposal: When the subject is focus, the two foci move obeying superiority. Therefore, the object focus still sits lower than the subject. (60) a. John i-to ku i-uy sensayngnim-i John-to chwuchenhay-ss-e. John-also his teacher-Nom John-also recommend-past-decl. Lit. John as well, his teacher (=John’s teacher) recommended him b. *Ku i-uy sensayngnim-i John i-TO chwuchenhay-ss-e. His teacher-Nom John-ALSO recommend-past-decl. Lit. His teacher (=John’s teacher) recommended John. The two types of movement pattern similarly regarding weak cross-over effects. My informants reported that the overt scrambling version in (61a) is not perfectly fine, although it is not ungrammatical. They also reported that the sentence with a low-copy pronunciation in (61b) is worse than (61a), but still it is not ungrammatical. To my native judgment, the difference in acceptability of the two sentences quite marginal, so it seems to be hard to say one is grammatical, but the other is not. 202 (61) a. Johnj-ul [ecey proi proj mannanpon salami]-i Johnj-ul chwuchenhay-ss-e. John-acc [yesterday met person]-Nom recommend-past-decl. The person who met John criticized him (=John). b. ?Johnj-ul [ecey proi proj mannanpon salami]-I || Johnj-ul chwuchenhay-ss-e. John-acc [yesterday met person]-NOM || John-acc recommend-past-decl. The person who met John criticized him (=John). Finally, the two types of movement also pattern similarly regarding parasitic gap licensing. In (62a), the subject with focus particle is assigned a prominent accent on it (as an indication of focus) and the in-situ object QP is destressed (as an indicative of topic). In that case, the parasitic gap is licensed by the topic phrase that is pronounced with a low copy with the help of prosody. More interestingly, the sentence is scopally ambiguous. This suggests that the object QP covertly moves to SpecTopP where it licenses a parasitic gap and furthermore it scopes over the subject focus particle (based on the structure hierarchy, topic higher than focus). The other scope interpretation (also > two) is obtained when a low copy is interpreted in LF. Interestingly, when the in-situ object QP is assigned a prominent accent on it as indicative of focus, as in (62b), the sentence sounds grammatical as well. This is predicted in my account too: the in-situ object QP with prominent accent undergoes focus movement obeying superiority. Thus, the high copy of the movement chain licenses the parasitic gap. Crucially, however, the moved object QP cannot scope over the subject with a focus particle because the landing site of the movement must be lower than the subject with the focus particle due to Superiority effects. (62) a. MARY-TO [pgi ilk-cito anhko] chayk twu kwon-uli peli-ess-ta. Mary-also read-without book two CL-acc throw away-past-Q Lit. Two books, Mary threw them away without reading. i) There are two books x such that Mary also threw x away without reading x. Two > also ii) Mary also threw two books away without reading them. Also > two b. Mary-to [pgi ilk-cito anhko] CHAYK TWU KWON-ULi peli-ess-ta. Mary-also read-without book two CL-acc throw away-past-Q Two books, Mary threw it away without reading. 203 i) *There are two books x such that Mary also threw x away without reading x. Two > also ii) Mary also threw two books away without reading them. Also > two One might wonder whether the sentence in (62) is a genuine parasitic gap example because Korean allows null arguments. I would claim that the sentences in (62) do not contain null arguments because overt pronouns in the gaps make the sentence ungrammatical. If the gap is not a real gap, but a null argument, there should be no reason for the sentences with an overt pronoun in (63) to be ungrammatical. (63). a. *MARY-TO [ kukes-ul ilk-cito anhko] ecey chayk twu kwon-uli peli-ess-ta. Mary-also them-acc read-without yesterday book two CL-acc throw away-past-Q Lit. Two books, yesterday Mary threw it away without reading. b. *Mary-to [ kukes-ul ilk-cito anhko] ecey CHAYK TWU KWON-ULi peli-ess-ta. Mary-also them-acc read-without yesterday book two CL-acc throw away-past-Q Two books, yesterday Mary threw it away without reading. Similarly, a scrambling version also licenses a parasitic gap as shown in (64). The scope ambiguity observed in (62a) is also observed in (64). (64) Chayk twu kwon-ul Mary-to ecey [pgi ilk-cito anhko] chayk twu kwon-uli peli-ess-ta. book two CL-acc Mary-also yesterday read-without book two CL-acc throw away-past-Q Lit. Two books, Mary threw it away without reading. The comparison between the two types of movement shows that the two types of movement syntactically behave very similarly. I suspect the minor difference in acceptability comes from a third reason, for example, interpretation of neighboring phrases or processing. In addition, they do not show a uniform A-movement or Ā-movement pattern. In the sense that they can create A- binding and overcome weak cross-over effects, they look like undergoing A-movement. However, they also show Ā-movement property in the sense that they can license parasitic gap. I currently do not know the reason for this mixed picture of topic/focus movement. One possibility 204 is that the peculiar behavior of movement comes from the nature of discourse configurational languages. In those languages, SpecTP is not reserved for a grammatical subject, that is assigned a nominative case. Rather, any phrase can sit in SpecTP once it is topic or focus. This pattern does not match to a traditional assumption that SpecTP is a A(rgument) position where case and φ-feature agreement takes place. Thus, I suspect, the peculiar behavior of topic/focus movement that shows both A-movement and Ā-movement properties is a consequence of the fact that a non-argument sits in an argument position. Apparently, this issue needs further investigation. However, the point that the two types of movement are actually the same syntactic operation except for where they are pronounced is still valid. 4.5. Conclusion In this chapter, I have discussed two cases in Korean where a low copy is phonetically realized in a movement chain due to the insertion of an intonation morpheme. The characteristic property of this type of derivation can be restated with a traditional term, covert movement, appearing with a certain intonation. The combination of the copy theory of movement and idea of intonation morpheme insertion proposed in Cheng and Rooryck (2000) captures the syntax-prosody interaction in Korean very well. The proposed analysis can be extended to other languages (including Japanese, German and Russian) that show similar patterns. In addition, the proposed analysis provides a microparameter that can distinguish among wh-in-situ languages (Sinhala vs. Korean/Japanese). 205 Chapter 5. The interaction between selection of copy and morphology/phonology 5.1. Overview In this chapter, I will investigate how the proposed analysis can account for cases where a low copy is pronounced due to morphological or phonological constraint. As I already introduced in chapter 1, Bošković (2002) and Bošković and Nunes (2007) provide a range of examples of low- copy pronunciation when high-copy pronunciation violates an independent (morpho-)phonological constraint. In this chapter, I will discuss how the proposal developed in this dissertation can account for such phonologically/morphologically forced low-copy pronunciation. In 5.2., I will discuss how phonological constraints affect selection of copy in a chain at PF. The section 5.2. will largely discuss two issues: one is prosodically forced low-copy pronunciation and the other is phonologically (but not prosodically) forced low-copy pronunciation. In both cases, otherwise impossible (or non-preferred) low-copy pronunciation is exceptionally forced for independent PF constraints. In 5.3., I will investigate cases where morphology interacts with selection of copies for pronunciation. Especially, I will discuss cases where insertion of (lexical) morphology satisfies Pronounce EPP, so a low copy is exceptionally pronounced. 5.2. Phonological constraint and low-copy pronunciation 5.2.1. Prosodic constraint and low-copy pronunciation In chapter 4, I discussed how the insertion of an intonation morpheme can license low-copy pronunciation in Korean (and possibly in other relevant languages). In those cases, a low copy of a chain is pronounced due to Earliness under Cyclicity when the chain is free from the Pronounce EPP requirement. In this section, I will discuss another instance of low-copy pronunciation that is also correlated with prosody. However, in the cases I will discuss in this section, a low copy of a chain is pronounced even when the chain is required to satisfy Pronounce EPP. Said simply, otherwise unacceptable low-copy pronunciation (due to satisfy Pronounce EPP) is forced by a prosodic requirement violating Pronounce EPP. As I already 206 noted, since Pronounce EPP is an economy condition, it is a violable constraint when there is a contradicting convergence constraint at play. In this section, I will review Spyropoulos and Revithiadou’s (2007) analysis of subject distribution in Greek and Ortega-Santos’ (2006) analysis of the distribution of subjects in Spanish as examples of such prosodically forced low- copy pronunciation. Distribution of subjects in Greek. Spyropoulos and Revithiadou (2007) argue that preverbal subjects in Greek are distinguished into two different types depending on their syntactic position. One type is well-investigated preverbal subjects that have properties of left- dislocated topic and the other type is preverbal subjects without a topic reading as in (1). (1) Q: tí éɣine? /tí néa? (Greek, Spyropoulos and Revithiadou 2007, (7))) what-Acc happen-past.3sg/what-Nom new-PL.Nom What happened? /What’s up? A. o Jánis fílise ti maría the john-Nom kiss-Past.3SG the Mary-Acc Spyropoulos and Revithiadou argue that the latter type of preverbal subjects has not received proper attention so far in the literature because this type of preverbal subject is only preferred in transitive constructions (not in intransitive constructions). More interesting evidence for the non- CLLDed status of preverbal subjects in SVO order is observed in the prosodic phrasing. Adopting Revithiadou and Spyropoulos (2003, 2005, 2009), Spyropoulos and Revithiadou show that a CLLDed object in Greek constitutes its own prosodic phrasing resisting prosodic restructuring. This is shown in the comparison between (2) and (3). In Greek, a sentence without any CLLDed element can be prosodically phrased based on either end-based mapping or binarity-based mapping. Thus, the sentence in (2) has two different ways of prosodic phrasing: One way is to bound the entire sentence into a single long prosodic phrase as in (2a) and the other way is to group pairs of words based on binarity-based mapping. Importantly, some sandhi rules apply within the same phonological phrase. For example, one sandhi rule, nasal-stop assimilation, is observed between word final nasal /n/ in sinantúsan and the word initial stop /t/ in téseris, which is realized in [d] in (2). When there is a prosodic boundary between the two sounds that are otherwise subject to a sandhi rule, the sandhi rule does not apply. For example, s- 207 voicing before voiced sonorants takes place in-between /z/ (in déseriz) and /m/ (in maθitéz) when they are in the same phonological domain as in (2a), but it does not occur when the two words are separated by a prosodic boundary as in (2b). This application of sandhi can be used to test phonological phrasing in Greek. When an object is CLLE-ed, it must be in its own phonological phrase as in (3a), so the following words cannot be together with the CLLD-ed object in a same prosodic domain (3b). The existence of prosodic boundary is supported by the lack of sandhi rule (nasal-stop assimilation) application compared to sentences in (2). (2) /sinantúsan téseris maθités likíu/ meet-Past.3SG four student-PL.Acc high-school-GEN They were meeting four high-school boys. a. [sinandúsan déseriz maθitéz licíu]φ end-based mapping b. [sinandúsan déseris]φ [maθitéz licíu]φ binarity-based mapping (Greek, Spyropoulos and Revithiadou 2007, (14)) (3) tus amán, tus parakoluθún ta peDJá (Greek, Spyropoulos and Revithiadou 2007, (16)) the A.M.A.N-Acc them watch-Pres.3PL he kid-PL.Nom As for the A.M.A.N (group), the kids watch them. a. [tus amán]φ [tus parakoluθún ta peDJá]φ b. *[tus amándus parakoluθún]φ [ta peDJá]φ Spyropoulos and Revithiadou claim that in SVO sentences like the one in (4), preverbal subjects are not necessarily CLLD-ed. If such a subject must be CLLD-ed, it must pattern together with a CLLD-ed object in (3b) with respect to prosodic phrasing. However, as shown in (4b), the preverbal subject can occur freely with the following word in the same prosodic domain. The lack of prosodic boundary between the preverbal subject and the following verb is supported by the s-voicing rule application (i.e. /s/ becomes [z] before the fricative /ð/). (4) to fós ðíni isxí sti mixaní the light-Nom give-Pres.3SG power-ACC to-the engine-ACC The light gives power to the engine. 208 a. [to fós]φ [ðín∅ isçí]φ [sti mixaní]φ end-based mapping b. [to fóz ðíni]φ [isçí sti mixaní]φ binarity-based mapping (Greek, Spyropoulos and Revithiadou 2007, (21)) In summary, there are the following issues in need of being addressed regarding the distribution of subjects in Greek 71 . 71 Spyropoulos and Revithiadou discuss obligatory preverbal subjects in middle, generic and stative verb constructions as in (i) (Roussou and Tsimpli 2006). Crucially, such a construction does not allow a topic reading of the preverbal subjects. The obligatory preverbal subject in stative/middle/generic predicates in (i) are, according to Spyropoulos and Revithiadou, a consequence of LF-PF matching constraint (RECIPROCITY, like Minimize Mismatch proposed by Bobaljik, see 1.2.1.). Adopting Roussou and Tsimpli’s (2006) argument that the preverbal subjects in stative/middle/generic predicates must be interpreted outside of the VP domain, Spyropoulos and Revithiadou argue that the LF requirement to interpret the subjects in preverbal position forces selection of the high copy in PF as well the matching of copies selected in LF and PF. (5) a. stative verb construction i maría kséri (#i maría) tin apántisi the Mary-Nom know-PRES.3SG (#the Mary-Nom) the answer-ACC Mary knows the answer. b. middle construction ta liná plénonte éfkola (#ta liná) the linen-Nom wash-Pres.3PL easily (#the linen-Nom) Linen wash easily. c. generic statement with a stage-level predicate i fálenes íne (#i fálenes) θilastiká the whale-PL.Nom are (#the whale-PL.Nom) mammal-PL.Nom Whales are mammals. (Spyropoulos and Revithiadou 2007, (9)) Since the current proposal does not assume minimize-mismatch type LF-PF matching constraint, I need to give an alternative analysis for this pattern. Unfortunately, for now, I do not have a fully developed analysis for the obligatory preverbal subjects with certain predicates. I would rather suggest two possibilities. First, as claimed by Spyropoulos and Revithiadou, I would need to incorporate a kind of PF-LF matching constraint into the theory. However, as I noted in 1.2.1. this LF-PF matching constraint faces some challenges. The other possibility is that the preverbal subject in such constructions is not in regular subject position. When Roussou and Tsimpli discussed the obligatory preverbal subjects in such constructions, they argue that the generic reading of subjects in middle/generic 209 (5) a. The existence of CLLDed preverbal subjects with topic reading. b. The preference of non-CLLDed preverbal subjects in transitives. c. The preference of postverbal subjects in intransitives. Spyropoulos and Revithiadou provide an Optimality Theoretical analysis to explain the distribution of subjects in Greek. Their idea shares the same spirit with the current proposal in that the different distribution of subjects is the result of selection of copy in PF (under the copy theory of movement) depending on various (morpho-)phonological constraints. However, their analysis differs from the current proposal in that they basically assume high-copy privilege (I will explain my analysis later in this section). In what follows, I will briefly summarize their analysis for the distribution of subjects in Greek in (5). They did not explicitly provide an analysis for CLLDed preverbal subjects (5a) just assuming their existence. The preference of preverbal subjects without topic reading in transitives (5b) is accounted for by the higher ranking of the PRONOUNCE HIGHEST constraint that forces low copies to be silent. Finally, the preference of postverbal subjects in intransitives (5c) is accounted for by an economy condition on sentence stress assignment. Since the last point is crucial to an alternative analysis I will suggest, I will review this point in more detail. They assume that while spell-out takes place, prosodic phrasing also takes place in a way that the edges of syntactic phrases are mapped into prosodic phrase boundaries. For example, assuming spell-out takes place by phase, the first spell-out domain of an intransitive sentence in (6a) is VP, as represented in the syntactic derivation in (6b) 72 . The only element inside the VP domain is the low copy of the subject, ta peðiá ‘the child’ 73 . Thus, it forms its own prosodic phrase as in (7). When the next higher phase (CP) is spelled-out, the spell-out domain contains the higher copy of predicates and subjects in stative predicates are associated with temporal and aspectual properties of the clause (- past, -perfective). Based on their claim, we can imagine that the subjects in such constructions undergo another movement independent to subject agreement and the position is somehow related with pronunciation. I will leave this issue for future works. 72 Spyropoulos and Revithiadou assume that unaccusatives project a strong phase head in Greek (Legate 2003). 73 Spyropoulos and Revithiadou assume that the verb írθan ‘come’ moves from its base position inside VP to T as head movement. However, they assume that a head movement chain is always pronounced with a high copy, so there is no copy remaining inside VP in their analysis. I also suggested a similar requirement on head movement in 2.3. 210 the subject and the verb. Since the syntactic phrases are required to be mapped into prosodic phrase boundaries, the prosodic phrasing in the CP spell-out domain looks like what is shown in (8). (6) a. írθan ta peðiá (Greek, Spyropoulos and Revithiadou, 2007, (53)) come-Past.3SG the child-pl.Nom The children came. b. [CP [ TP ta peðiá írθan [vP írθan [VP írθan ta peðiá]]]] (7) p-phrasing in vP-phase (Greek, Spyropoulos and Revithiadou, 2007, (55)) [ta peðiá]φ (8) p-phrasing in CP-phase (Greek, Spyropoulos and Revithiadou, 2007, (56)) [ta peðiá]φ írθan Given (7) and (8), if the low copy of the subject chain is pronounced, the prosodic phrasing looks like in (9a) while if the high copy is pronounced, the prosodic phrasing looks like in (9b). The different prosodic phrasing affects the sentence stress assignment. In Greek, sentence stress is assigned to the right-most element in prosodic phrasing (or IP) (Spyropoulos and Revithiadou 2007, (50)). If prosodic phrasing appears as in (9a), the sentence stress can be assigned to the postverbal subject. On the other hand, if prosodic phrasing appears as in (9b), the sentence stress cannot be assigned because the supposedly last element, verb, has not been prosodically phrased yet. Thus, sentence stress assignment has to be postponed to the second round of prosodic phrasing for the verb to be prosodically phrased. Thus, according to Spyropoulos and Revithiadou, the prosodic phrasing as in (9a) is more economical than the one in (9b) because it does not need to postpone the application of the stress assignment. Due to this reason, postverbal subjects are preferred in intransitives. Note, in transitives, the sentence stress is always assigned to objects, which are the rightmost element in a sentence. Therefore, postverbal subjects (i.e. lower copy in a subject chain) has no advantage in stress assignment rendering the high copy preferred instead due to PRONOUNCE HIGHEST. 211 (9) a. [írθan ta peðiá] b. [ta peðiá] írθan In chapter 2, I propose that agreement per se takes place under c-command relation without any movement required. Movement of a (goal) phrase takes place for uninterpretable features of a probe to be valued. Crucially, the goal phrase does not necessarily move if there is another way for the interpretable features of the goal to be carried parasitically, namely, parasitic phrasal/head movement. The core difference between phrasal movement and parasitic movement is that the phrase would never undergo movement in the latter. Thus, in-situ elements in PF can be further distinguished by the way they can stay in-situ. When parasitic movement takes place, the element can stay in-situ, and there should be no high copy in LF as well. Thus, the element does not show “high” behavior in interpretation. For examples, English there-construction or Sinhala in-situ wh-phrases are examples. By way of contrast, when an element undergoes phrasal movement, but a low copy is pronounced, then the in-situ element is expected to show high behavior since it has a copy that can be interpreted in LF. For examples, French in-situ wh- phrases or Korean in-situ topics shows such derivation. Turning back to Greek subjects, in what follows, I will provide an alternative analysis for the distribution of subjects in Greek based on the proposed system. I claim that both phrasal movement and parasitic head movement take place in Greek. When a subject is topic (5a), the subject is base-generated in an Ā-position (as in Standard Arabic) and has nothing to do with subject agreement. Therefore, uninterpretable φ-features in T are expected to be valued by non-surface subjects. I argue that, parasitic head movement in the form of v-to-T raising takes place in this case. If this analysis is correct, it is predicted that preverbal subjects do not show a canonical A-property (Alexiadou and Anagnostopoulou 1998). By way of contrast, if a subject is free from a topic reading (5b/c), I argue that subjects are base- generated within VP and undergoes movement to SpecTP to give values to uninterpretable features in T, as in English. Furthermore, due to Pronounce EPP, a high copy in SpecTP is forced to be phonetically spelled-out. However, in intransitive sentences, such a high-copy pronunciation would lead to less economical stress assignment procedure in PF (as discussed by Spyropoulos and Revithiadou), so PF selects a low copy to be pronounced in the subject chain. This is why postverbal subjects are preferred in intransitives while preverbal subjects are 212 preferred in transitives. Crucially, the distribution of subjects is a matter of preference, not grammaticality. It is also predicted in the current analysis. Since both PF constraints, Pronounce EPP and timing of application of stress assignment, are economy conditions, they are violable. Thus, either preverbal or postverbal subjects are in principle possible in both transitives and intransitives. I want to highlight the point that otherwise non-preferred choice (low-copy pronunciation) due to Pronounce EPP is forced when another phonological constraint prefers low-copy pronunciation over high-copy pronunciation. In other words, postverbal subjects in Greek intransitives are examples of prosodically forced low-copy pronunciation. The alternative analysis proposed in this section for the distribution of Greek subjects basically share many aspects with the account proposed by Spyropoulos and Revithiadou. However, their analysis is based on the assumption that a high-copy is privileged in pronunciation (e.g. PRONOUNCE HIGHEST). However, as I discussed in length in chapter 2, the high-copy privilege in pronunciation is a misleading idea masked by the EPP requirement. Under the current system, therefore, without assuming a puzzling high-copy advantage in pronunciation, the same data can be accounted for. Distribution of subjects in Spanish/Serbo-Croatian. A second case of prosodically forced low-copy pronunciation is observed in Serbo-Croatian as discussed in 1.2.6. Before discussing the SC data, I would like to introduce Ortega-Santos (2006) which proposes prosodically driven low-copy pronunciation in Spanish. It is well-known that subjects in Spanish can appear either in preverbal position or in postverbal position and the sentence final subjects are associated with new information focus (Zubizarreta 1998). (9) Pedro le dio un libro a María. (Spanish, Ortega-Santos 2006, (5)) Pedro gave a book to María (10) a. Quién le dio un libro a María? (Spanish, Ortega-Santos 2006, (2)) Who gave a book to María? b. Le dio un libro a María PEDRO. Gave a book to María Pedro 213 Ortega-Santos tests a hypothesis that a preverbal subject in SVO order as in (9) is the phonetic realization of a high copy in the subject chain while a postverbal subject in VOS word order as in (10b) is the phonetic realization of a low copy based on the copy theory of movement. Following Zubizarreta (1998), Ortega-Santos assumes that sentence stress is assigned to the rightmost element in the sentence via Nuclear Stress Rule (NSR), so a sentence with (new information) focus must be phonetically realized in the sentence final position to be properly stress-assigned. To satisfy this prosodic requirement of information structure, the low copy in a chain is pronounced when the subject is focused. However, Ortega-Santos points out that there is an apparent difference between preverbal subjects in SVO order and postverbal subjects in VSO/VOS order as shown in (11) and (12). The contrast in (11) shows that there is an asymmetry between preverbal subjects and postverbal subjects when the subject is a Negative Polarity item. Likewise, there is difference in interpretation between the preverbal subject in (12a) and the postverbal subject in (12b). In (12a), the subject is given a categorical judgement while that in (12b) is given a thetic judgement. (11) Negative Polarity Items (Spanish, Ortega-Santos 2006, (26)) a. Ayer, nadie llegó. yesterday, nobody arrived b. Ayer no llegó nadie. (12) a. El rey ha muerto. (Spanish, Ortega-Santos 2006, (27)) the king has died b. Ha muerto el rey. has died the king The contrast between preverbal subjects and postverbal subjects in (11) and (12) calls for an explanation in the hypothesis presented above because such a contrast is not expected if the two instances of subjects belong to the same chain, but only differ in the position where they are pronounced in PF. To incorporate this observation into an analysis based on the copy theory of movement, following Gallego (2004), Ortega-Santos argues that preverbal subjects in SV order have “surface semantics” that is obtained in SpecTP and only when a subject has surface 214 semantics, it moves to SpecTP position. Otherwise, the subject moves only to SpecAgrSP. Ortega-Santos further assumes that the selection of copies for pronunciation (under the copy- theory approach) applies only to a chain that does not include any member in SpecTP. In other words, when the subject has surface semantics, it moves to SpecTP and is pronounced there as in (13). By way of contrast, when the subject does not have surface semantics, the subject moves from its base position to AgrSP and is pronounced depending on the information structure: if the subject is new information focus, the low copy is pronounced as in (15), while if the subject is not a focused phrase, the high copy is pronounced as in (14). (13) a. Pedro le dio un libro a María. SVO Pedro gave a book to María b. [ TP Pedro le dio [AgrS Pedro dio [AgrDO un libro dio [AgrIO a María dio [V1 Pedro dio [V2 un libro dio a María]]]]]] [ TP Pedro gave [AgrS Pedro gave [AgrDO a book gave [AgrIO to María gave [V1 Pedro gave [V2 a book gave to María]]]]]] (14) a. UN LIBRO le dio Pedro a María. DO-V-S-IO A BOOK gave Pedro to María b. [ TP UN LIBRO le dio [AgrS Pedro dio [AgrDO un libro dio [AgrIO a María dio [V1 Pedro dio [V2 un libro dio a María]]]]]] [ TP A BOOK gave [AgrS Pedro gave [AgrDO a book gave [AgrIO to María gave [V1 Pedro gave [V2 a book gave to María]]]]]] (15) a. El libro se lo dio a María Pedro. DO-V-IO-S the book gave to María Pedro b. [TP El libro se lo dio [AgrS Pedro dio [AgrDO el libro dio [AgrIO a María dio [V1 Pedro dio [V2 un libro dio a María]]]]]] [ TP the book gave [AgrS Pedro gave [AgrDO the book gave [AgrIO to María gave [V1 Pedro gave [V2 the book gave to María]]]]]] (Spanish, Ortega-Santos 2006, (29)-(31)) 215 The analysis proposed by Ortega-Santos predicts no difference between subjects in VSO order and VOS order in their interpretation, for example, scope taking. This prediction is born out as in (16). Regardless of word order, subjects can bind objects. The crucial example is (16b) where the subject that linearly follows the object can bind the object. This indicates that the subject moves to a position higher than the object during derivation leaving a copy there, which can be interpreted in LF even though a low copy is phonetically realized in PF. (16) a. Cuándo hablaron [tus hermanas]j [de sí mimas]j? (Spanish, Ortega-Santos 2006, (16)) when talked your sisters about themselves b. Cuándo hablaron [de sí mismas]j [tus hermanas]j? when talked about themselves your sisters. The system developed in this dissertation can be extended to the observation and analysis provided in Ortega-Santos (2006). Even though it is not clear to me what “surface semantics” means, it seems clear that movement to SpecTP (in structure assumed in Ortega-Santos) is not for subject agreement since various elements including indirect objects or direct objects can be in SpecTP, as shown in (14) and (15). Similar points are remarked on in Zubizarreta (1998). Zubizarreta suggests that the SpecTP position in Spanish is not subject-associated, but discourse- associated (topic/emphatic/focus) 74 . Thus, preverbal subjects appear for discourse-feature agreement (rather than subject agreement). Thus, we can safely conclude that preverbal subjects in Spanish undergo movement for some reasons independent of subject agreement (i.e. topic movement). I assume that the different motivation of movement leads to the different semantic effects between preverbal subjects and postverbal subjects. This is compatible with the current proposal. Since subjects in SpecTP undergo another movement for an independent reason, the subject chain in (13) and one in (14/15) are not identical. A subject in (13) is constrained by Pronounce EPP applied to the additional movement (topic movement) rather than by Pronounce 74 Zubizarreta (1998) also points out that preverbal subjects and postverbal subjects behave differently with respect to binding. When the bindee is a topicalized object clitic, which is originated in the same clause with the subject, only preverbal subjects, not postverbal subject, can bind the object clitic. This contrast supports that postverbal subjects do not undergo a phrasal movement to SpecTP. 216 EPP applied to a regular subject-chain. Thus, what Ortega-Santos argues for as the distinction between preverbal subjects and postverbal subjects can be accommodated in the current system. However, departing from Ortega-Santos’ argument, I assume that VSO order is not created by subject movement. The subject is spelled-out in its base position, so it is linearly between a verb that is raised to T and the in-situ object. On the other hand, the VOS order is created by an object movement in the way suggested by Zubizarreta (1998) (p-movement). Thus, the subject agreement uniformly takes place via parasitic head movement (v-to-T movement) across the variable word order including SVO, VSO, and VOS. To conclude, the distribution of subjects in Spanish is successfully captured under the current system providing another example of prosodically triggered low-copy pronunciation. The Spanish data remind us of the distribution of subjects in Serbo-Croatian (SC), discussed in chapter 1. In 1.2.6., I showed that subjects in SC undergo movement and need to be pronounced with a high copy in the moved position unless the subject is new information focus that needs to be assigned sentence stress at the sentence-final position. The pattern observed in SC is quite similar to the distribution of subjects in Spanish. However, there is a crucial difference between the two languages. In Spanish, when the subject is not a topic/focus element, VSO order is the basic order. By way of contrast, in SC, even when the subject is not a topic element, a subject precedes a verb indicating that the subject gives values to uninterpretable φ- features of T. In other words, in Spanish, subject agreement takes place via parasitic head movement, while in SC, subject agreement takes place via phrasal movement of the subject. Thus, extending and revising the analysis proposed for Spanish, I propose an analysis for SC as follows. The sentence initial subject is related with a discourse configurational function, as topic. It is base-generated in SpecTopP. Thus, the sentence initial subject is a result of pronouncing a copy in SpecTopP to satisfy Pronounce EPP required by topic (or other discourse) agreement. For a regular subject without a topic reading, it undergoes movement to SpecTP to give values of uninterpretable features in T and the chain is pronounced with a high copy due to Pronounce EPP. However, when the subject is focus, and needs to be assigned an appropriate sentence stress in the sentence final position, the lowest copy is instead selected for pronunciation. This is possible because Pronounce EPP is a violable economy condition. In other words, the current system allows low-copy pronunciation even when it violates Pronounce EPP if only low-copy pronunciation, but not high-copy pronunciation, can derive a convergent derivation due to 217 prosodic constraints. Therefore, we can conclude that distributions of subjects in SC show another example of prosodically driven low-copy pronunciation. This analysis implies that information structure can be encoded by selecting different copies. The general pattern observed in Spanish, and SC is that subjects with a topic reading are situated in a position higher than where a regular subject sits in and subjects with a focus reading appear in the sentence final position. This general pattern is explained as follows: in a language where the NSR applies, a focus phrase needs to be pronounced in the sentence final position. Therefore, a focus element is forced to be pronounced with a low copy in the base position. However, since a topic element is not associated with such position-sensitive stress rules, it is pronounced with a high copy in the moved position to satisfy Pronounce EPP. 5.2.2. Non-prosodic, but phonological constraint and low-copy pronunciation In 1.2., reviewing previous studies, I discussed cases where a low copy of a chain is pronounced because high-copy pronunciation violates a phonological constraint that results in a non- convergent derivation. In this section, I will introduce another example of low-copy pronunciation forced by a phonological constraint. In the previous section, we discussed that subjects in Greek undergo phrasal movement to SpecTP (unless they have a topic reading) and the subject chain is pronounced with a high copy due to Pronounce EPP unless the predicate is intransitive. However, there is a case where the low copy must be phonetically realized even when the predicate is not an intransitive. Spyropoulos and Revithiadou (2007) show that subjects in subjunctive clauses must appear in postverbal position 75 . The subjunctive clause structure assumed by Spyropoulos and Revithiadou is in (17). 75 They note that when a subject undergoes topic/focus movement, it can appear before a subjunctive particle -na, as in (i). However, in that case, the subject is not in a canonical subject position (SpecTP), but in Ā-position above the subjunctive particle. (i) a. subject in focus (Spyropoulos and Revithiadou 2007, (31)) TO PÚLMAN na fíVi the bus-Nom SUBJ leave-3SG THE BUS should go. b. to púlman || na fíVi (Spyropoulos and Revithiadou 2007, (32)) 218 (17) na (*o Jánis) féri o Jánis to vivlío SUBJ the John-Nom bring-3SG the John-Nom the book-Acc John should/may bring the book. MP M TP na Spec T′ o Jánis T vP féri o Jánis féri to vivlío (Greek, Spyropoulos and Revithiadou 2007, (34)) Following Philippaki-Warburton and Spyropoulos (1999), they assume that the subjunctive particle na and the verb undergo post syntactic morphological merger, so the particle is procliticized to the verb. Such morphological merger operation requires the two elements to be linearly adjacent not being intervened by other elements. However, if a high copy of the subject in SpecTP is pronounced, it happens to intervene between the subjunctive particle and the verb, creating an undesired environment for the morphological merger between them. Thus, the derivation will crash due to the illicit PF form with a stranded particle. To avoid this, according to Spyropoulos and Revithiadou, the low copy of the subject within VP is phonetically realized even though a high copy is preferred to be phonetically realized in general. Reviewing Bobaljik (2002), I introduced similar examples in chapter 1. In Scandinavian languages, if a high copy of the chain intervenes between a verb and an inflectional head, the verb and the inflectional head cannot be morphologically merged, so a low copy of a shifted object chain is instead phonetically realized. For more relevant examples, see 1.2. The remaining question which needs to be answered is how the current system accounts for such cases where otherwise unavailable low-copy pronunciation is forced because high-copy pronunciation violates some phonological constraints. The system developed in this dissertation is compatible with the interaction between the exceptional low-copy pronunciation and phonological the bus-Nom SUBJ leave-3SG As for the bus, it should go. 219 constraints. Throughout the dissertation, I argue that the seemingly high-copy privilege is just a consequence of the PF constraint, Pronounce EPP, not a chain-intrinsic property. For uninterpretable features to have a PF consequence, Pronounce EPP is introduced as a constraint in PF (see 2.3). Crucially, Pronounce EPP is an economy condition, not a convergence requirement. Thus, the economy condition can be violated if a more economical option violates a convergence requirement. This is why otherwise unavailable low-copy pronunciation (due to Pronounce EPP) can be allowed in a special phonological environment. 5.3. Morphology and low-copy pronunciation In chapter 4, I discussed two cases where the insertion of an intonation morpheme satisfies Pronounce EPP making low-copy pronunciation available. In this section, I will provide two examples where an insertion of a lexical morpheme satisfies Pronounce EPP so that a movement chain is pronounced with a low copy due to Earliness under cyclicity. 5.3.1. In-situ wh-phrases in Coptic Egyptian Coptic Egyptian is one of the ancient Egyptian languages with SVO word order. In what follows, I will review Reintges (2007) for relevant data regarding in-situ wh-phrases in Coptic Egyptian and show how the system developed so far can account for them. Reintges (2007) argues that in-situ wh-phrases in Coptic Egyptian are instances of low- copy pronunciation of a wh-movement chain. Let me present Reintges reasoning. In Coptic Egyptian, wh-phrases can appear either in their base position or in sentence initial position. However, there is a crucial difference between the two types of wh-questions. When a wh-phrase appears in its base position, a relative tense marker is required to appear on the verb while such a marker does not show up in wh-fronting questions. For example, when wh-in-situ appears, a relative tense marker nt- is prefixed to the verb (18a) while a wh-fronting question in (18b) lacks the marker. 220 (18) a. Adverbial wh-in-situ question with relative tense marker Awo nt-a-u-ei eβol tɔn? And REL-PERF-3PL-come PCL where From where did they come? (Apoc. 7, 13; ed. Budge, cited in Reintges 2007, (2a)) b. Adverbial wh-fronting question without relative marker eβol tɔn a-tetən-ei e-pei-ma? PCL where PERF-2PL-come to-DEM.SG.M-place From where did you come here? (Budge, Martyr. 220,8 cited in Reintges 2007, (2b)) Reintges argues that the relative tense marker -nt is not a question particle because (i) it appears in a broad syntactic environment rather than just being limited to wh- or yes-no questions. For example, temporal adjunct clauses or conditionals have such a relative tense marker even though they do not have a question force, and (ii) it appears below the complementizer C unlike other canonical question particles in Coptic Egyptian that appear in the topmost position (even before a fronted wh-phrase). Reintges claims that the relative tense marker is wh-agreement morphology. This is supported by the two following observations. First, in Coptic Egyptian, an in-situ wh-phrase can be interpreted as an indefinite (as like many in-situ languages). The interrogative reading and the indefinite reading can be distinguished by the presence of the relative tense marker. As in (19a), when the relative tense marker appears, the sentence with an in-situ wh-phrase has an interrogative interpretation as like its overt wh-fronting counterpart in (19b). However, when the relative tense marker is missing as in (19c), the sentence with an in-situ wh-phrase is interpreted as a declarative sentence with a specific indefinite phrase. (19) a. wh-in-situ with relative tense marker (Hilaria 12:29, cited in Reintges 2007, (23a) ənt-a nim špɔ na-f n-tei-hypomenɛ? REL-PERF who achieve for-3SG.M PERP-DEM.SG.F-endurance Who has achieved for himself such endurance? b. wh-fronting (KHML I3:7-8, cited in Reintges 2007, (23b)) nim a-f-ent-k e-pei-ma? Who PERF-3SG.M-bring-2G.SM to-DEM.SG.M-place Who brought you here? 221 c. wh-in-situ without relative tense marker: lack of an interrogative reading (Budge, Apocr. 134, 14-15, cited in Reintges 2007, (23c)) ša-u-eime [če a nim pə-šɛre ən-nim črɔ həm p-agɔn]. HAB-3PL-know C PERF who DEF.SG.M-child of-who win in DEF.SG.M.-contest They know that so-and-so, the son of so-and-so, won the contests. Second, the relative tense marker indicates the scope of a wh-phrase. When the relative tense marker appears in the embedded clause, the in-situ wh-phrase has embedded scope as in (20a). By way of contrast, when the relative tense marker appears in the matrix clause as in (20b), the in-situ wh-phrase in the embedded clause has matrix scope (here, the relative tense marker appears with the allomorph ere). (20) a. əmpe-f-čoo-s [če ənt-a-f-kaa-f tɔn]. NEG.PERF-3SG.M-say-3SG.F C REL-PERF-3SG.M-place-3SG.M where He did not tell where he had put it. (Apophth.Patr., Chaîne 235, 65,18, cited in Reintges 2007, (28a)) b. ere əm-mɛɛše čɔ əmmo-s [če ang nim]? REL(-PRES) DEF.PL-crowd say PREP-3SG.F C I who Who is the crowd saying that I am (Luke 9,18, cited in Reintges 2007, (29b)) Interestingly, in-situ wh-phrases do not follow patterns observed in canonical wh-in-situ languages (e.g. Chinese), but follow the pattern of their fronting counterparts. Reintges provides two pieces of evidence. First, the argument-adjunct asymmetries in the availability of in-situ wh- phrases, observed in Mandarin Chinese or Malay, so-called wh-in-situ languages, are not observed in Coptic Egyptian. As shown in (20a) and (20b), both argument and adjunct wh-phrases can stay in-situ. Second, in-situ wh-phrases in Coptic Egyptian are not sensitive to intervention effects. As we discussed in 4.2., intervention effects can be used as a diagnostic tool to distinguish wh-in-situ from wh-movement. In Coptic Egyptian, in-situ wh-phrases can be freely c-commanded by a negative tense-aspect marker (21a), a focus particle (21b) or a universal quantifier (21c). 222 (21) a. ete-mpe-čɔhəm hən aš əm-ma? REL-NEG.PERF.2SG.F-defile in what of-place In which place have you (woman) not been defiled? (Jeremiah 3,2, cited in Reintges 2007, (36a)) b. tenu k y e hɔ e-i-na-tamie u-ɛi na-I ən-tə-nau? Now PCL SELF.1SG REL(-FUT-)1SG-AUX-create INDEF.SG-house for-1SG PREP-when When will I, myself, build me a house as well? (Genesis 30, 30, cited in Reintges 2007, (36b)) c. k-nau [če ənt-a-f-sənt əm-pɛwe tɛr-u ən-aš ən-he (PRES).2SG.M-see C REL-PERF-3SG.M-establish DEF.PL-heavens all-3PL.F in-what of-manner həm pe.f-logismos] through DEF.SG.M.3SG.M-reasoning You see how He has established all the heavens through His reasoning. (Shenoute, Mingarelli 288b:1-8, cited in Reintges 2007, (36c)) The lack of intervention effects in interrogative sentences with in-situ wh-phrases in (21) support the idea that the seemingly in-situ wh-phrases actually undergo movement, but the movement chain is pronounced with a low copy in the base position of the wh-phrase. Based on the observation, Reintges provides the following analysis for in-situ wh-phrases in the language. Following Chomsky (2000), Reintges assumes that wh-phrases have an unvalued wh-feature, so every wh-phrase (regardless of whether it is overtly fronted or not) moves to a position where it gets its uninterpretable feature valued. SpecFocP, according to Reintges, is the position where wh-phrases move to. An interpretable wh-feature in C is inherited to Foc head, so Foc agrees with the wh-phrase, as in (22). When the wh-chain is pronounced with a low copy in the base position, a relative tense marker appears on the focus head as an agreement marker between the functional head FOC and the higher copy of the wh-phrase. On the other hand, when the wh-chain is pronounced with a high copy in SpecFocP, the relative tense marker does not appear for economy conditions, which prevent a redundant marking of wh- dependencies by spelling out both the relative tense marker and the high copy of the wh-chain. 223 (22) Wh-movement schema in Coptic Egyptian proposed by Reintges (2007) [C [FocP WH [Foc REL ] [FinP T … WH …]]] Feature inheritance T-to-Foc movement Reintges observation and analysis seems quite compatible with the current system developed in this dissertation. The in-situ wh-phrases in Coptic Egyptian appear to be the exact case where Pronounce EPP is realized by lexical morphology insertion rendering the wh-chain free from the requirement of Pronounce EPP. In the current system, Coptic Egyptian belongs to the type of language French belongs to: wh-phrases undergo phrasal movement, but when morphology insertion takes place to satisfy Pronounce EPP, the wh-chain does not need to satisfy the PF constraint, and the chain can be phonetically realized by the lowest copy due to another PF constraint, Earliness under Cyclicity. On the other hand, when a wh-chain by itself satisfies Pronounce EPP, the high copy in the moved position is phonetically spelled-out. Some minor points are different between Reintges analysis and mine. For example, Reintges assumes that wh- movement takes place due to an uninterpretable wh-feature in wh-phrases, but I assume that wh- movement takes place due to an uninterpretable wh-feature in the functional head, C. In addition, Reintges explains the lack of the relative tense marker in wh-fronting sentences with an economy condition that prevents redundant marking of wh-dependencies while in my system, it can be explained by a different economy condition, Earliness under Cyclicity, which forces the lowest copy to be pronounced. The two analyses seem almost the same, but they are distinguished with respect to a prediction about intermediate copies. Since the economy condition proposed by Reintges blocks double wh-marking by pronouncing both the relative tense marker and a copy of the wh-chain in the same local domain (where the wh-phrase has its scope), Reintges account predicts that pronouncing an intermediate copy is allowed since the intermediate copy is not in the scope position of the wh-phrase (and the relative tense marker appears in the scope position). By way of contrast, my account predicts that intermediate copies are not allowed to be pronounced due to Earliness under Cyclicity. Unfortunately, I have not found an example that supports either account. Nevertheless, the two accounts share the same spirit and I would like to point out that so far, the phenomenon reported in Reintges (2007) is successfully accounted for under the system proposed in the dissertation. 224 5.3.2. In-situ topic phrases in Korean In this section, I will provide an analysis for the in-situ topic phrases in Korean introduced in chapter 1 (1.3.1.) based on the proposed system. I showed that Korean has a morphological topic marker –(n)un and when this topic marker is attached to a subject, the subject has a thematic topic reading. In addition, even when the object is scrambled over the subject with the topic marker as schematized in (23), the subject seems to behave as if it is in a structurally higher position than the scrambled object in terms of binding and scope taking. I repeat a relevant example in (24). In (24a), a subject with a nominative case marker cannot follow a co-indexed self-anaphor due to Binding Principle C, but a subject with a topic marker can as in (24b). This serves as a piece of evidence supporting that the topic marked subject is structurally higher than the scrambled object. (23) [ Object-acc [ Subject-top [Object-acc … V … ]]] (24) a. *Caki-casin-toi Minho-ka Caki-casin-toi pinanhay-ci anh-ass-ta self -self -also Minho-nom blame-NEG-Past-decl. b. Caki-casin-toi Minho-nun Caki-casin-toi pinanhay-ci anh-ass-ta self -self -also Minho-top blame-NEG-Past-decl. Minho blamed himself. I argue that the subject with the topic marker in (24) actually undergoes topic movement higher than the scrambled object, but the low copy of the chain is phonetically realized due to morphology insertion that satisfies Pronounce EPP, as represented in (25). (25) [Subject-top [ Object-acc [ Subject-top [Object-acc … V … ]]] To complete this account, I need to address the exact agreement mechanism in (25). In 4.3., I argued that in Korean, a topic movement can be pronounced with a low copy when Pronounce EPP is satisfied by the insertion of an intonation morpheme. Given the two observations in Korean, we can conclude that there are two instances of low-copy pronunciation in Korean topic movement. 225 One is low-copy pronunciation with intonation morphology insertion and the other is low-copy pronunciation with lexical morphology -(n)un insertion. However, there is a difference between the two types of in-situ topics. In 1.3.1., I showed that the topic marker can induce a thematic topic reading only when it is attached to a subject. When it is attached to non-subject (e.g. objects, adverbs…), those phrases with the topic marker must have a contrastive topic reading. In other words, the lexical morphology insertion is available only when the topic is the subject in a sentence. By way of contrast, the phonologically marked in-situ topic is available across syntactic categories. In other words, not only subjects but also objects or adverbs can express topic-hood with intonation. Based on that, two questions arise: (i) Do topics in the two types of in-situ topics move to the same position? (ii) Why is lexical morpheme insertion available only for subject topics? I will give an answer to these questions in sequence. My answer to the first question is “yes”. I assume that the two types of in-situ topics actually have topic-chain that has its highest copy in the same position. I argue that the position topics target is T, following Miyagawa (2010) (see 4.4. for details about Miyagawa (2010)’s proposal). Moving to the second question, I assume that in Korean, as in agreement languages, T is responsible for nominative case assignment in addition to its role in agreement with topic. In other words, T is the head that agrees with a topic phrase and simultaneously gives nominative case value to a subject. Due to this double responsibility, we could imagine two possible derivations as in (26). When a subject is topic, T agrees with a single phrase, the subject as in (26a). On the other hand, when a non-subject is topic, T simultaneously agrees with two different phrases, one for nominative case valuation to the subject and the other for topic agreement with a non- subject phrase, as in (26b). Since two independent agreements are at play with a single functional head T, T is required to have two different morphologies inserted corresponding to subject agreement (nominative case marker) and topic agreement (topic marker). I crucially assume that only one type of lexical (agreement) morphology can at best be inserted to a single functional head. In other words, either topic morphology or nominative case morphology, but not both, can be inserted in T. Therefore, when a subject is topic, either the topic marker (without the nominative case marker) (27a) or the nominative case marker (without the topic marker) can be inserted. The inserted topic/nominative case morphology undergoes morphological merge with the subject. On the other hand, when a non-subject is topic, T agrees with two different phrases. However, in Korean, a nominative case marker cannot be dropped (unless the subject is topic-marked) unlike 226 other case markers for independent reason (Yoon 2012). Thus, a nominative case marker must be inserted in T. It blocks insertion of another lexical morphology, the topic marker, in T. Therefore, the topic movement chain must satisfy Pronounce EPP by spelling-out the high copy of the chain or by inserting an intonation morpheme resulting in sentences in (28) respectively 76 . This is why when a non-subject phrase is topic, we cannot observe the topic marker –(n)un attached to the topic phrase. In addition, since I assume that there is a single topic phrase per sentence, there should be no case where a subject is morphologically topic-marked and a non-subject phrase is prosodically topic marked (obviously, the reverse case cannot be attested because non-subject phrases cannot be attached with a topic marker). (26) a. Subject = topic T [vP SUBJ …] Case/topic agreement b. Non-subject (e.g. object) = topic T [vP SUBJ OBJ] Case agreement topic agreement (27) a. John-(*i)un sosel-ul ilk-ess-ta. John-(*Nom)-Top novel-Acc read-past-decl. b. John-i(*un) sosel-ul ilk-ess-ta. John-Nom(*Top) novel-Acc read-past-decl. John read a novel. 76 I assume that subjects with a nominative case marker do not necessarily move to SpecTP. In the proposed system, movement takes place only when the probe requires the movement. As a discourse-configurational language, I assume that Korean does not have φ-agreement between T and subject, following Miyagawa (2010). However, despite the lack of φ-agreement, case assignment is required. Since T can give nominative case to the subject in-situ under a c-command relation without subject movement, the subject does not move to SpecTP. However, if the subject with the nominative case marker has a topic reading, it moves to SpecTP due to the agreement between T and the subject DP. 227 (28) a. [ TP Sosel-ul [ TP John-i [vP John-i sosel-ul ilk-ess-ta]]]. Novel-acc John-Nom novel-acc read-past-decl. b. [ TP Sosel-ul [TP John-i [vP John-I || sosel-ul ilk-ess-ta. Novel-acc John-Nom John-Nom novel-acc read-past-decl. Lit. Regarding a novel, John read it. In this section, I investigated another type of low-copy pronunciation case in Korean. When a subject is topic, it undergoes movement to SpecTP like regular topic movement. However, this chain can be pronounced with a low copy in the base position when it is free from the Pronounce EPP requirement due to lexical morphology insertion (i.e. the topic marker –(n)un). This is why the seemingly in-situ subject with a topic marker has higher scope than its surface position. 5.4. Conclusion In this chapter, I have discussed various cases where a low copy in a movement chain is pronounced as a result of interaction between copy-selection constraints proposed in this dissertation and phonological/morphological properties in languages in question. Due to the nature of economy constraint, Pronounce EPP is violable favoring a convergent, but less-economical derivation over a non-convergent, but more economical derivation. This allows otherwise unacceptable low-copy pronunciation available in subject-chains when the subject is focus in Greek, Spanish and Serbo-Croatian. Furthermore, the violability of Pronounce EPP plays a role in many exceptional low-copy pronunciation cases discussed in Bobaljik (2002), Bošković (2002) and Bošković and Nunes (2007): if a high copy pronunciation violates some phonological/morphological constraints, a low-copy is instead phonetically realized violating Pronounce EPP. Finally, the interaction between morphology insertion and low-copy pronunciation observed in Coptic Egyptian and Korean is also successfully captured by assuming that Pronounce EPP can be satisfied by morphology insertion. Crucially, the last point is hard to explain in Nunes’ Chain Reduction Algorithm because a high copy is always expected to have fewer unchecked uninterpretable features than a low copy regardless of the additional morphology, so the low copy is forced to be deleted even in a case where special agreement morphology appears. However, in the current approach, morphology insertion and high-copy pronunciation are expected 228 to appear in a complementary distribution (because when morphology insertion takes place and satisfies Pronounce EPP, the movement chain that is free from Pronounce EPP is subject to Earliness under Cyclicity favoring low-copy pronunciation. Therefore, the impossibility of having both a relative tense marker and high-copy pronunciation in a single sentence in Coptic Egyptian is naturally explained. 229 Chapter 6. Conclusion 6.1. Summary of the discussion Throughout the dissertation, I have investigated how a combination of agreement, movement, and pronunciation determines surface position of a phrase. Inspired by Reverse Agree (proposed by Wurmbrand 2012) and the EPP as a pure PF constraint (proposed by Landau 2007), I proposed a system where agreement, movement, and pronunciation are operated by independent constraints. In this section, I will review the discussion and implication of the proposal. I assume that agreement takes place under a c-command relation, not Spec-head relation. As a consequence of the structural constraint on agreement, movement takes place when a structurally higher element needs to be valued by a structurally lower element. Crucially, I suggested three different ways of movement for such valuation, which are repeated in (1). A goal phrase by itself can move as phrasal movement or a relevant feature of a goal can be parasitically carried on to the probe by parasitic head/phrasal movement. The core difference between phrasal movement and parasitic movement is that in phrasal movement, the goal by itself undergoes movement, so it leaves a high copy in the moved position which can have a chance to be interpreted in LF. By way of contrast, in parasitic movements, the goal by itself does not move even though its features are going up. Thus, the goal does not have a high copy due to the lack of movement chain. (1) a. Phrasal movement XP Xprobe ZP …YPgoal … XP YP XP X ZP … YP… 230 b. Parasitic head movement XP Xprobe … ... ZP (YPgoal) ZP Z (MPgoal) XP Xprobe … (…)+Z X ... ZP (YPgoal) ZP Z (MPgoal) c. Parasitic phrasal movement XP Xprobe … … ZP YP ZP KP ZP Z … YPgoal XP KP XP Xprobe … … ZP YP ZP KP ZP Z … YPgoal Such distinction among movements provides a new way of looking at in-situ phrases. According to the system, there are broadly two cases where a phrase can be pronounced in its base position. One way is that due to parasitic movement, the goal does not move at all, so it is pronounced with the only copy in the base position. The other way is that a goal undergoes phrasal movement, but a low copy is pronounced in the base position giving an impression of in-situness. The two types of in-situ goals are differentiated in their interpretation. An in-situ goal with parasitic movement does not have a copy in a higher position, so it cannot be interpreted in a higher position (sensitive to intervention effects, taking a low scope) and it is insensitive to movement constraints (e.g. island constraints). By way of contrast, an in-situ goal that is a phonetic realization of a low copy in a movement chain shows “high” behaviors in addition to sensitivity to movement constraints. The latter type of in-situ phrases can be divided further by what licenses the low-copy pronunciation. One important parameter is the EPP. I basically assume that the EPP is a pure PF constraint that forces information about agreement to be overtly spelled-out. I suggested two 231 possible ways of pronouncing the EPP: morphology insertion or pronouncing a high copy of a goal, which is adjacent to a probe. These two ways of pronouncing the EPP have two implications for movement. First, the so-called high-copy privilege in pronunciation, assumed in many previous studies, is a consequence of the pronounce-wise reinterpreted EPP, not an inherent feature of a chain. Second, since pronouncing a high copy is just one way of satisfying the EPP, a high copy is able to not be phonetically realized when there is another means of satisfying the EPP. As I noted, morphology insertion can be an alternative. Thus, when an agreement morphology appears to pronounce the EPP, a movement chain is pronounced with a low copy being free from the EPP requirement, but constrained by another PF constraint, Earliness under Cyclicity. Morphology insertion is realized with various forms. In Romance subject agreement cases, the inserted morpheme is realized with inflectional morpheme. In French and Korean (and many other languages discussed in this dissertation), inserted intonation morphemes are realized with a characteristic prosody. In Korean and Coptic Egyptian, morphology insertion is realized with a lexical marker. One thing all of those cases share is that the goal phrase can be interpreted in a position higher than the position where it is spelled-out and a special morphological/intonational property occurs. Based on these theoretical backgrounds, I provided a typological study of subject agreement and wh-agreement. Obviously, the typological study has been restricted in the number of languages it has considered. However, the discussion is meaningful in the sense that the proposed system can systematically explain the typological variations among languages listed there. Furthermore, it is noteworthy that the proposed system can account for not only an argument domain (subject agreement) but also a non-argument domain (wh-movement, topic/focus movement). I also explained how the proposed system can accommodate the exceptional low-copy pronunciation cases observed in previous studies. Based on the economy conditional nature of the EPP constraint, I argue that Pronounce EPP is violable for a convergent derivation. 6.2. Toward future research There seems to be many challenging issues connecting to the proposal of this dissertation. The most outstanding issue, in my perspective, is the relation between null subjects and Pronounce EPP. 232 It has been proposed that a language with rich morphology system allows null subjects (Speas 2006). However, such an account is confronted by a problem in so-called “radical pro-drop” languages (e.g. Korean, Japanese, and Chinese). Those languages allow null subjects even without a rich set of agreement morphology. Not only that, French is another type of counter-example. French has a rich set of agreement morphology, but the language does not allow null subjects. Due to such counter-examples, there are other approaches suggested to account for the licensing mechanism of null subjects. From the perspective of the copy theory of movement, null subjects are cases where every copy in a chain is deleted (if we follow the PF-deletion tradition of ellipsis, rather than an LF copy analysis). Can we find any clue of null-subject licensing mechanism under the proposed system? In the current proposal, I assume that the EPP is a pure PF constraint that requires information about agreement to be overtly pronounced. For example, in subject movement, if there is other means to express the subject-T agreement information, a subject does not need to appear in SpecTP. Agreement morphology on a verb is employed to express subject-T agreement information in Romance languages, so the subject does not need to be phonetically realized in SpecTP. Due to this, we can assume that a copy in SpecTP can be deleted. Thus, a low copy within VP is selected for pronunciation due to Earliness under Cyclicity. If this is the case, in Romance languages, null subject cases are in fact a deletion of a low copy within VP. Bearing that in mind, let’s bring the radical pro-drop languages to our attention. Those languages are categorized in discourse-configurational languages by É. Kiss (1995). Since there is no φ-agreement between T and the subject, the subject does not move to SpecTP. In other words, unless the subject undergoes topic/focus movement, the subject has just a single copy in its base position within VP. In the two examples above, we could find a common property: a copy within VP can be deleted for some unknown reason. Let’s turn to our attention to French. French has φ-agreement between T and a subject but the language does not allow morphology insertion as a means of pronouncing the EPP. Thus, the copy in SpecTP must be spelled-out to express the agreement information. Given the generalization found in pro-drop Romance languages and radical pro-drop languages on the one hand and the impossibility of null subjects in French on the other hand, it seems that if a high copy in SpecTP is the copy that is selected to be pronounced in PF, it must be phonetically realized while if a low copy within VP is the copy that is selected to be pronounced in PF, it can be deleted. Currently, I do not have a fully developed account to explain this phenomenon. However, it seems to be worthy to investigate the issue from the perspective suggested in this dissertation. 233 There are other issues that needs to be investigated more. For example, the mixed property of in-situ wh-phrases in English multiple wh-questions is one of them. As we discussed, an in-situ wh-phrase in English multiple questions is insensitive to island effects and not able to license a parasitic gap as if it does not undergo movement. However, it can license ACD and is not sensitive to intervention effects as if it does undergo movement. Such mixed properties call for an explanation. Another issue is the interaction between syntax and prosody. In this dissertation, I have discussed how prosody differentiates the acceptability of intervention configurations or scope interpretation in Korean. It appears that prosody plays an important role in distinguishing the language from a typologically similar language, Sinhala. Thus, it would be interesting research topic how prosody plays a role in the corresponding Sinhala examples. 234 References AlAlamat, Hamed. 2014. pro and verb movement in Arabic syntax. European Scientific Journal, 10:442-457. Alexiadou, Artemis, and Elena Anagnostopoulou. 1998. Parametrizing AGR: Word order, verb- movement and EPP-checking. Natural Language and Linguistic Theory 16:491–539. Alexiadou, Artemis & Florian Schäfer. 2011. There-insertion: an unaccusativity mismatch at the syntax-semantics interface. In Online Proceedings of WCCFL 28. <https://sites.google.com/site/wccfl28pro/papers/AlexiadouandSchaefer2010.pdf?attredirect s=0> Al-Mendeel, Modi Fahad. 2012. Application of Government and Binding theory on agreement in Modern Standard Arabic and Moroccan Arabic. International Journal of Linguistics, Vol4, No.4. An, Duk-Ho. 2007. Clauses in noncanonical positions at the syntax-phonology interface. Syntax, 10:38–79. Anderson, Catherine. 2004. The structure and real-time comprehension of quantifier scope ambiguity. Doctoral dissertation, Northwestern University. Antonyuk-Yudina, Svitlana. 2011. Why Prosody Matters: Surface Scope Bias in Russian Quantifier Scope. In LSA Annual Meeting Extended Abstracts 2011. Arkadiev, Peter, M. and Alexander B. Letuchiy. 2009. The syntax and semantics of event structure and Adyghe causatives. Ms. Russian Academy of Sciences, Moscow. Arregi, Karlos. 2001. Focus and word order in Basque. Ms., Massachusetts Institute of Technology. Available at http://mit.edu/karlos/www Barbosa, Maria do Pilar Pereira. 1995. Null Subjects. Doctoral dissertation, MIT. Cambridge, MA.: MITWPL. Bayer, Josef. 1996. Directionality and Logical Form: On the scope of focusing particles and wh- in-situ. Dordrecht: Kluwer. Bayer, Josef and Lisa Lai-Shen Cheng. (submitted) 2014. Wh-in-situ. In The Blackwell Companion to Syntax, eds. M, Everaert & H, van Riemsdijk. Oxford: Blackwell. 235 Beck, Sigrid. 2006. Intervention effects follow from focus interpretations. Natural Language Semantics. 14:1-56. Beghelli, Filippo and T. Stowell. 1997. Distributivity and negation. In Ways of scope taking ed. A. Szabolcsi, 71–109. Dordrecht: Kluwer. Benincà, Paola and Cecilia, Poletto. 2004. Topic, Focus and V2: Defining the CP sublayers. In The Structure of CP and IP, vol. 3, ed. L, Rizzi. Oxford/New York, Oxford University Press, 52–75. Benmamoun, Elabbas. 1999. Arabic morphology: The central role of the imperfective. Lingua, 108:175-201. Bhatt, Rajesh and Veneeta, Dayal. 2007. Rightward Scrambling as Rightward Remnant Movement. Linguistic Inquiry, 38: 287–301. Bobaljik, Jonathan David. 1993. On ergativity and ergative unergatives. MIT Working Papers in Linguistics 19: Papers on Case and Agreement II, 45-88. Bobaljik, Jonathan David. 2002. A-chains at the PF-interface: copies and ‘covert’ movement. Natural Language & Linguistic Theory 20: 197–267. Bobaljik, Jonathan David & Susi, Wumbrand. 2012. Word Order and Scope:Transparent Interfaces and the 3⁄4 Signature. Linguistic Inquiry 43:371-421. Boeckx, Cedric. 1998. A minimalist view on the passive. UConn Working Papers Occasional Papers in Linguistics 2. Department of Linguistics, University of Connecticut, Storrs. Boeckx, Cedric. 2004. Long-distance agreement in Hindi: some theoretical implications. Studia Linguistica 58: 23–36. Bolinger, Dwight. 1965. Forms of English: Accent, morpheme, order, Harvard University Press, Cambridge MASS.: The MIT Press. Bošković, Željko. 1997. The syntax of non-finite complementation: An economy approach. Cambridge MASS.: The MIT Press. Boskovic, Željko. 1998. LF movement and the Minimalist Program. To appear in Proceedings of NELS 28. Bošković, Željko. 2001. On the nature of the syntax-phonology interface: Cliticization and related phenomena. Amsterdam: Elsevier Science. Bošković, Željko. 2002. On multiple wh-fronting. Linguistic Inquiry 33: 351-383. 236 Bošković, Željko and Jairo, Nunes. 2007. The Copy Theory of Movement: A view from PF. In The Copy Theory of Movement, ed. N, Corver and J, Nunes, 13-74. Amsterdam: John Benjamins. Büring, Daniel. 1994. Topic. In Focus and natural language processing. Vol. 2, Semantics, (ed.) P, Bosch and R, van der Sandt, 271–280. Heidelberg: IBM Germany. Büring, Daniel. 1997a. The Meaning of Topic and Focus. Routledge, London. Büring, Daniel. 1997b. The great scope inversion conspiracy. Linguistics and Philosophy 20:175– 194. Büring, Daniel. 2001. Let’s Phrase it! Focus, Word Order, and Prosodic Phrasing in German Double Object Constructions. In Competition in Syntax, eds. G. Müller and W. Sternefeld. NewYork: MoutondeGruyter, 69– 105. Büring, Daniel. 2010. Towards a typology of focus realization. In Information structure. Theoretical, typological, and experimental perspectives, eds. M, Zimmermann and C, Féry, Oxford: Oxford University Press, 177–205. Cable, Seth. 2007. The grammar of Q: Q-Particles and the nature of wh-fronting, as revealed by the wh-questions of Tlingit. Doctoral dissertation, MIT. Cambridge, MA. Cheng, Lisa L.-S. 1991. On the typology of wh-questions. Doctoral dissertation, MIT. Cambridge, MA. Cheng, Lisa Lai-Shen and Johan, Rooryck. 2002. Licensing wh-in-situ. Syntax, 3:1–19. Choi, Hye-won. 1996. Optimizing structure in context: scrambling and information structure. Doctoral dissertation, Stanford university. Chomsky, Noam. 1973. Conditions on Transformations. In A Festschrift for Morris Halle, (eds.) S. Anderson and P. Kiparsky. New York, Holt, Rinehart and Winston. Chomsky, Noam. 1993. A minimalist program for linguistic theory. In The view from Building 20, ed. K, Hale and S, J. Keyser, 1-52. Cambridge, Mass.: MIT Press. Chomsky, Noam. 1994. Bare phrase structure. In MIT Occasional Papers in Linguistics 5. Cambridge, Mass.: MIT press. Chomsky, Noam. 1995. The minimalist program. Cambridge, Mass.: MIT press. Chomsky, Noam. 2000. Minimalist Inquiries: The Framework. In Step by Step:Essays in Minimalist Syntax in Honor of Howard Lasnik, eds. R, Martin, D, Michaels, And J, Uriagereka, Cambridge, Mass.: MIT Press, 89–155. 237 Chomsky, Noam. 2001. Derivation by Phase. In Ken Hale: A Life in Language, ed. M, Kenstowicz, Cambridge, Mass.: MIT Press, 1–52. Chomsky, Noam and Howard Lasnik. 1977. Filters and Control. Linguistic Inquiry 8:425–504. Collins, Chris. 1997. Local Economy. Cambridge, Mass.: MIT press. Collins, Chris. 2001. Economy Conditions in Syntax. In The Handbook of Contemporary Syntactic Theory, ed. M. Baltin and C, Collins, Oxford: Blackwell, 45–61. Cook, Philippa and John, Payne. 2006. Information Structure and Scope in German. In Proceedings of the LFG06 Conference, eds. M, Butt and T, Holloway King, CSLI Publications. Dayal, Veneeta. 1996. Locality in wh-quantification. Dordrecht: Kluwer. Deal, Amy Rose. 2009. The origin and content of expletives: Evidence form selection, Syntax, 12:285-323. Deguchi, Masanori, and Yoshihisa Kitagawa. 2002. Prosody and wh-questions. In the Proceedings of NELS 32, University of Massachusetts at Amherst, 73-92. Diesing, Molly. 1992. Indefinites. Cambridge, Mass.: MIT press. Dikken, Marcel den. 1995. Binding, expletives, and levels. Linguistic Inquiry 26:347-354. Enç, Mürvet. 2003. The Syntax of Specificity. Ms., University of Wisconsin-Madison. Erteschik-Shir, Nomi, Gunlög Josefsson, and Björn Köhnlein. 2017. In abstract of Fanselow, Gisbert. 2004. Cyclic Phonology–Syntax-interaction: movement to first position in German. In Interdisciplinary Studies on Information Structure, eds. S. Ishihara, M. Schmitz and A. Schwarz, 1: 1–42. Fanselow, Gisbert, and Anoop K. Mahajan. 2000. Towards a Minimalist Theory of wh- expletives, wh-copying, and successive cyclicity. In Wh-scope marking, Linguistik Aktuell/Linguistics Today 37, eds. U, Lutz, G, Müller and A, von Stechow, Amsterdam: John Benjamins, 195–230. Féry, Caroline. 1993. German intonational patterns. Tübingen: Niemeyer. Féry, Caroline. 2013. Focus as prosodic alignment. Natural Language and Linguist Theory, 31:683–734. Féry, Caroline, and Shinichiro Ishihara. 2010. How Focus and Givenness Shape Prosody. In Information structure: Theoretical, typological, and experimental perspectives, ed. M, Zimmermann and C, Féry, Oxford, UK: Oxford University Press, 36–63. 238 Fiengo, Robert. 1974. Semantic Conditions on Surface Structure. Doctoral dissertation, MIT. Cambridge, MA.: MITWPL. Fox, Daniel, j. 1998. Economy and semantic interpretation – A study of scope and variable binding. Doctoral dissertation, MIT. Cambridge, MA. Frascarelli, Mara. 2007. Subjects, topics, and the interpretation of referential pro. Natural Language and Linguistic Theory, 25:691-734. Frascarelli, Mara and Roland, Hinterhölzl. 2007. Types of topics in German and Italian. In On information structure, meaning and form, eds. S. Winkler & K. Schwabe, Amsterdam/Philadelphia: John Benjamins, 87–116. Frey, Werner. 1993. Syntaktische Bedingungen für die semantische Repräsentation: Ü ber Bindung, implizite Argumente und Skopus. (Studia Grammatica XXXV.) Berlin: Akademie Verlag. Gallego, Angel. 2004. Phase Effects in (Iberian) Romance. Ms. Universitat Autònoma de Barcelona. Groat, Erich and John O’Neil. 1996. Spell-Out at the LF interface. In Minimal Ideas: Syntactic Studies in the Minimalist Framework, eds. W, Abraham, S, D, Epstein, H, Thráinsson and C. J-W Zwart, Amsterdam: John Benjamins, 113–139. Hagstrom, Paul. 1998. Decomposing questions. Doctoral dissertation, MIT. Cambridge, MA.: Hagstrom, Paul. 2001. Particle movement in Sinhala and Japanese. In Clause Structure in South Asian Languages, eds. V, Dayal & A, Mahajan. Kluwer, Dordrecht. Halle, Morris, and Marantz, Alec. 1993. Distributed Morphology and the pieces of inflection. In The view from Building 20: Essays in Linguistics in honour of Sylvain Bromberger, eds. K, Hale and S, J, Keyser, Cambridge, Mass: MIT Press, 111-176. Holmberg, Anders. 2000. Scandinavian stylistic fronting: How any category can become an expletive. Linguistic Inquiry 31:445–483. Huang, C.-T. James. 1982. Logical relations in Chinese and the theory of grammar. Doctoral dissertation, MIT. Cambridge, MA.: MITWPL Hsu, Yu-Yin. 2008. Sentence-internal Topic and Focus in Chinese. Proceedings of the 20th North American Conference on Chinese Linguistics (NACCL-20). V ol 2. eds. M, K.M. Chan and H, Kang. Columbus, Ohio: The Ohio State University, 635-652. 239 Huybregts, Riny. 1991. Allosteric Agreement in VSO Languages. Linguistics in the Netherlands, 1:81-90. Ionin, Tania and Tatiana, Luchkina. 2015. Focus on scope: information structure and quantifier scope in Russian. Presented at the 24th annual meeting of Formal Approaches to Slavic Linguistics NYU, NY . Ishihara, Shinichiro. 2002. Invisible but Audible Wh-scope marking: Wh-constructions and deaccenting in Japanese. The Proceedings of WCCFL 21, Somerville: Cascadilla Press, 180-193. Ishihara, Shinichiro. 2007. Major Phrase, Focus Intonation, Multiple Spell-Out. The Linguistic Review 24:137-167. Ishihara, Shinichiro. 2011. Japanese focus prosody revisited: Freeing focus from prosodic phrasing. Lingua 121:1870–1889. Jacobs, Joachim. 1982. Syntax und Semantik der Negation im Deutschen. Munich: Fink. Jayaseelan, Karattuparambil A. 1996. Question-word movement to focus and scrambling in Malayalam. Linguistic Analysis, 26: 63-83. Jackendoff, Ray.S. 1972. Semantic Interpretation in Generative Grammar. Cambridge, Mass.: MIT press. Jiménez-Fernández, Á ngel L. and Shigeru Miyagawa. 2014. A feature-inheritance approach to root phenomena and parametric variation. Lingua 145:276-302. Jo, Jungmin, Seok-Keun Kang and Tae-Jin Yoon. 2003. Grammatical Encoding of Focus and Topic Information in Korean: Morpho Syntactic, Semantic, and Acoustic Evidence. Presented in Michigan Linguistic Society Annual Meeting 2003, University of Michigan, Ann Arbor, MI. Jun, Sun-Ah. 1993. The Phonetics and Phonology of Korean Prosody. Doctoral Dissertation. Ohio State University, Linguistics. Jun, Sun-Ah. 2005. Intonational Phonology of Seoul Korean Revisited. UCLA Working Papers in Phonetics, 104: 14-25. Jun, Sun-Ah. and Hyuck-Joon. Lee. 1998. Phonetic and Phonological Markers of Contrastive Focus in Korean, in Proceedings of the 5th ICSLP, 4:1295-1298. Sydney. Jun, Sun-Ah and Mira, Oh. 1996. A prosodic analysis of three types of wh-phrases in Korean. Language and Speech, 39: 37-61. 240 Kang, Chorong. 2017. Some QPs, but not all, interact with prosody. In Proceedings of the WCCFL 34, 295-302. Kang, Chorong. To appear. Korean Intervention Effects are not a single phenomenon: Evidence from syntax-prosody interface. Linguistic Review. Kato, Mary Aizawa. 1999. Strong and weak pronominals in the null subject parameter. Probus, 11:1-37 Kayne, Richard. 1994. The Antisymmetry of Syntax. Cambridge, Mass.: MIT press. Kim, Chonghyuck. 2010. Korean question particles are pronominals: a transparent case of representing discourse participants in the syntax. http://ling.auf.net/lingBuzz/001157/. Kim, Shin-Sook. 2002. Focus Matters: Two Types of Intervention Effect. Paper presented at WCCFL 21, Santa Cruz. Kim, Shin-Sook. 2005. Focus Intervention Effects in Questions. Handout of Theoretical East Asian Languages 3. Harvard University. É. Kiss, Katalin. 1995. Introduction. In Discourse configurational languages, ed. Katalin É. Kiss, 3–27. Oxford: Oxford University Press. É. Kiss, Katalin. 1998. On Generic and Existential Bare Plurals and the Classification of Predicates. In Events and Grammar, ed. S. Rothstein, Kluwer Academic Publishers, Dordrecht, 145– 162. Kishimoto, Hideki. 2005. Wh-In-Situ and Movement in Sinhala Questions. Natural Langauge and Linguistic Theory, 23:1 - 51. Kitagawa, Yoshihisa. 1990. Anti-scrambling. Ms. University of Rochester Ko, Heejeong. 2005. Syntax of Why-in-situ: Merge into [SpecCP] in the Overt Syntax. Natural Language & Linguistic Theory. 23: 867–916. Koopman, Hilda. 1984. The syntax of verbs: From verb movement rules in the Kru languages to Universal Grammar. Dordrecht: Foris. Krifka, Manfred. 1998. Scope inversion under the rise-fall contour in German. Linguistic Inquiry 29:75–112. Landau, Idan. 2007. EPP Extensions. Linguistic Inquiry 38:485-523. Lasnik, Howard. 1995. Case and expletives revisited: On Greed and other human failings. Linguistic Inquiry 26:615-633. 241 Lasnik, Howard. 1999. Chains of Arguments. In Working Minimalism, eds. S. Epstein and N. Hornstein, Cambridge, Mass.: MIT, 189–215 Lee, Chungmin. 2000. Topic, contrastive topic and focus: what's on our minds. Journal of Cognitive Science 1: 21-38. Legate, Julie Anne. 2003. Some interface properties of the phase. Linguistic Inquiry 34: 506-516. Liu, Fengh-hsi. 1990. Scope dependency in English and Chinese. Doctoral dissertation. UCLA. Longobardi, Giuseppe. 2008. Reference to individuals, person, and the variety of mapping parameters. In Essays on Nominal Determination: From morphology to discourse management, eds. H, Müller, Henrik and A, Klinge, Amsterdam: John Benjamins 189– 211 McCloskey, James. 1979. Transformational syntax and model theoretic semantics: A case study in Modern Irish. D. Reidel: Dordrecht. Mahajan, Anoop. 1990. The A/A-bar distinction and movement theory. Doctoral dissertation, MIT, Cambridge, Mass. Mahajan, Anoop. 1997. Rightward scrambling. In Rightward movement, ed. D, Beerman, D, LeBlanc, and H, van Riemsdijk, Amsterdam: John Benjamins, 186–214. Mahajan, Anoop. 2000. Eliminating head movement. GLOW Newsletter 44:44–45. Malhotra, Shiti. 2011. Movement and Intervention Effects: Evidence from Hindi/Urdu. Doctoral dissertation, University of Maryland, College Park. Manetta, Emily. 2010. Wh-Expletives in Hindi-Urdu: The vP Phase. Linguistic Inquiry, 41:1-34. Manetta, Emily. 2013. Copy theory in wh-in-situ languages: Sluicing in Hindi-Urdu. JSAL 6: 3- 24. Manzini, M. Rita and Leonardo M. Savoia. 2011. Wh-in situ and wh-doubling in northern Italian varieties. Linguistic Analysis 37: 79–113. Mathieu, Eric. 1999. French WH in situ and the Intervention Effect. In UCL Working Papers in Linguistics 11, eds. C. Iten and A. Neeleman, 441-472. Matushansky, Ora. 2006. Head Movement in Linguistic Theory. Linguistic Inquiry, 37:69-109. McFadden, Thomas, and Sandhya Sundaresan. 2015. Towards resolving the countercyclicity of the EPP, Ms. Merchant, Jason. 2001. The syntax of silence: Sluicing, islands, and the theory of ellipsis. Oxford, UK: Oxford University Press. 242 Milsark, Gary. 1974. Existential Sentences in English. Doctoral Dissertation, MIT. Milsark, Gary. 1977. Toward an explanation of certain peculiarities of the existential construction in English. Linguistic Analysis 3:1-29. Miyagawa, Shigeru. 1998. Wh chains and quantifier induced barriers, ms, MIT. Miyagawa, Shigeru. 2001. The EPP, scrambling, and wh-in-situ. In Ken Hale: A life in language, ed. Michael Kenstowicz, Cambridge, Mass.: MIT Press, 293–338. Miyagawa, Shigeru. 2005. On the EPP. In Perspectives on phases, eds. N, Richards and M, McGinnis, MIT Working Papers in Linguistics 49. Cambridge, Mass.: MIT, 201–236. Miyagawa, Shigeru. 2010. Why Agree? Why Move? Unifying Agreement-based and Discourse Configurational Languages. Cambridge, Mass.: MIT Press Miyagawa, Shigeru. 2012. Agreements that occur mainly in the main clause. Main clause phenomena: new horizons, eds. L. Aelbrecht, L. Haegeman , and R. Nye Philadelphia; John Benjamins. Morita, Hisashi. 2014. Large-scale Pied-piping in Sinhala and Japanese. In Proceedings of WIGL 2014, 31-45 Murasugi, Kumiko. 1992. Crossing and Nested Paths: NP Movement in Accusative and Ergative Languages. Doctoral dissertation, MIT. Cambridge, MA.: MITWPL. Nunes, Jairo. 1995. The copy theory of movement and linearization of chains in the Minimalist Program. Doctoral dissertation, University of Maryland, College Park. Nunes, Jairo. 2004. Linearization of chains and sideward movement. Cambridge, Mass: MIT Press. Nunes, Jairo. 2011. The Copy Theory. In The Oxford Handbook of Linguistic Minimalism, ed. Cedric Boeckx, New York: Oxford University Press, 143-175. Oh, Mira. 2008. Prosody and Information Structure: Phonetic Realizations of Focus and Topic in Korean. 음성과학, 15:7-19. Ortega-Santos, Iván. 2006. On Postverbal Subjects, PF and the Copy Theory: The Spanish Case. In Proceedings of the 9th Hispanic Linguistics Symposium, 56-66. Pérez, Carlos Muñoz. 2015. Three explanatory challenges for Copy Theory. In Proceedings of ConSOLE XXIII, 2015, 340-359 243 Pesetsky, David. 1987. Wh-in-situ: Movement and unselective binding. In The representation of (In)Definiteness, eds. E. Reuland and A. ter Meulen, Cambridge, Mass.: MIT Press, 98- 129. Pesetsky, David. 2000. Phrasal movement and its kin. Cambridge, Mass: MIT Press. Pesetsky, David, and Torrego, Esther. 2007. The syntax of valuation and the interpretability of features. In Phrasal and clausal architecture, ed. by S, Karimi, V, Samiian and W, Wilkins, Amsterdam: John Benjamins, 262-294. Platzack, Christer. 2003. Agreement and Null Subjects. In Proceedings of the 19th Scandinavian Conference of Linguistics, 31.2:326-355. Popescu, Alexandra. 2000. The morphophonology of the Romanian clitic sequence. Lingua 110:773-799. Potsdam, Eric and Maria, Polinsky. 2012. Backward raising. Syntax 15:75-108. Reintges, Chris, H. 2007. Variable pronunciation sites and types of wh-in-situ. In The copy theory of movement, eds. N. Covert and J, Nunes. John Benjamins, Amsterdam / Philadelphia, pp. 220-287. Revithiadou, Anthi and Vassilios Spyropoulos. 2003. Trapped within a phrase: Effects of syntactic derivation on p-phrasing. In Proceedings of IP 2003 Prosodic Interfaces. Nantes, eds. A, Mettouchi & G, Ferré, 167-172. Revithiadou, Anthi and Vassilios Spyropoulos. 2005. The Multiple Spell-Out Hypothesis and the phonological component: Evidence from Greek. In Proceedings of NELS 35. Revithiadou, Anthi and Vassilios Spyropoulos. 2009. A dynamic approach to the syntaxphonology interface: A case study of Greek. In Inter Phases: Phase-Theoretic Investigations of Linguistic Interfaces, ed. K, Grohmann. Oxford: Oxford University Press. Richards, Mark. 2007. On object shift, phases and transitive expletive constructions in Germanic. In Linguistic Variation Yearbook 6, eds. P. Pica et al., Amsterdam: John Benjamins, 139-159. Richards, Norvin. 1997. What moves where when in which language? Doctoral dissertation, MIT. Cambridge, MA.: MITWPL. Rizzi, Luigi. 1997. The fine structure of the left periphery. In Elements of Grammar: A Handbook of Generative Syntax, ed. L, Haegeman, 281-337. Dordrecht: Kluwer. 244 Roussou, Anna and Ianthi-Maria Tsimpli. 2006. On Greek VSO again. Journal of Linguistics 42: 317-354. Saito, Mamoru. 1992. Long-distance scrambling in Japanese. Journal of East Asian Linguistics 1:69–118. Saito, M. 2006. Optional A-scrambling. In Japanese/Korean linguistics proceedings 16, 44–63. Stanford, Calif.: CSLI Publications. Selkirk, Elisabeth. 1984. Phonology and syntax: the relation between sound and structure. Cambridge Mass: MIT Press. Selkirk, Elisabeth. 1995. Sentence prosody: Intonation, stress and phrasing. In The handbook of phonological theory, ed. John A. Goldsmith, 550-569. Cambridge MA and Oxford UK: Blackwell. Sheehan, Michelle. 2016. Subjects, null-subjects and expletives in Romance. In Manual of grammatical interfaces in Romance, eds. S, Fischer & C, Gabriel, Berlin: Mouton de Gruyter. 392-362. Shields, Rebecca. 2008. What’s so Special about D-Linking? Handout of Nels 39, Cornell University, available at http://citeseerx.ist.psu.edu/viewdoc/download?doi=10.1.1.534.3676&rep=rep1&type=pdf . Shimoyama, Junko. 2006. Indeterminate phrase quantification in Japanese. Natural Language Semantics, 14:139–173. Simpson, Andrew and Tanmoy, Bhattacharya. 2003. Obligatory overt wh-movement in a wh-in- situ language. Linguistic Inquiry, 34: 127–142. Sohn, Keunwon. 1995. Negative Polarity Items, Scope and Economy. Doctoral Dissertation, University of Connecticut. Soltan, Usama, 2007a. On Formal Feature Licensing in Minimalism: Aspects of Standard Arabic Morphosyntax. Doctoral dissertation, University of Maryland, College Park. Soltan, Usama. 2007. On agree and postcyclic merge in syntactic derivations: First conjunct agreement in Standard Arabic. In Perspectives on Arabic Linguistics 19, ed.E, Benmamoun, Amsterdam: John Benjamins, pp191-213. Speas, Margaret. 2006. Economy, agreement, and the representation of null arguments. In Arguments and Agreement, eds P. Ackema and F. Weerman, Oxford: Oxford 245 University Press, 35–75. Spyropoulos, Vassilios and Anthi, Revithiadou1. 2007. Subject chains in Greek and PF processing. In Proceedings of the 2007 Workshop in Greek Syntax and Semantics at MIT, MIT Working Papers in Linguistics 57, 293-309. Stjepanović, Sandra 1999. What do second position cliticization, scrambling, and multiple Wh- fronting have in common. Doctoral dissertation, University of Connecticut, Storrs. Stojanovič, Danjela. 1997. Object shift in Serbo-croatian. In, Clitics, Pronouns and Movement, eds. J, R. Black, V , Motapanyane, Amsterdam: John Benjamins, 301-320. Stojanovič, Danjela. 2007. Free word order and copy theory of movement. In The copy theory of movement, eds. N. Covert and J, Nunes. John Benjamins, Amsterdam / Philadelphia, pp. 219-248. Szabolcsi, Anna. 1997. Strategies for Scope Taking. In Ways of Scope Taking, ed. A, Szabolcsi. Dordrecht: Kluwer. Szendrői, Krista, 2003. A stress-based approach to the syntax of Hungarian focus. Linguistic Review 20:37–78. Takahashi, Daiko. 2002. The EPP and null operator movement. In Proceedings of Linguistics and Phonetics 2002. Available at www.wat- net.com/proceedings/DTakahashi/LP2002DTakahashi.pdf. Tomioka, Satoshi. 2007. Pragmatics of LF Intervention Effects: Wh-interrogatives in Japanese and Korean. Journal of Pragmatics 39: 1570-1590. Travis, Lisa. 1984. Parameters and the effects of word order variation. Doctoral dissertation, Massachusetts Institute of Technology. Truckenbrodt, Hubert. 1995. Phonological Phrases: Their Relation to Syntax, Focus, and Prominence. Doctoral dissertation, MIT, Cambridge, Mass. Truckenbrodt, Hubert. 2013. An analysis of prosodic F-effects in interrogatives: Prosody, syntax and semantics. Lingua, 124:131-175. Ura, Hiroyuki. 2000. Checking theory and grammatical functions in universal grammar. New York: Oxford University Press. Vallduví, Enric. 1992. The Informational Component. New York: Garland. Vallduví, Enric and Elisabet, Engdahl. 1996. The linguistic realization of information packaging. Linguistics 34:459–519. 246 Venditti, Jennifer. 1997. Japanese ToBI labelling guidelines. In Papers from the Linguistics Laboratory, eds. K. Ainsworth Darnell and M. D’Imperio, The Ohio State University Working Papers in Linguistics 50, 127-162. The Ohio State University. Vlachos, Christos. 2012. Wh-constructions and the division of labour between syntax and the interfaces. Doctoral dissertation, university of patras. Watanabe, Akira. 1992. Subjacency and s-structure movement of wh-in-situ. Journal of East Asian Linguistics, 1:255-291. Watanabe, Akira. 2000. Absorption: interpretability and feature strength. Center for Excellence Report, Kanda University of International Studies. Wold, Dag. 1996. Long Distance Selective Binding: The Case of Focus. In Proceedings of SALT 6, (eds.) T. Galloway and S. Spence, 311–328. CLC, Ithaca, N.Y. Wurmbrand, Susi. 2012. The syntax of valuation in auxiliary–participle constructions. In Proceedings of the 29th West Coast Conference on Formal Linguistics (WCCFL 29) University of Arizona: Tucson. Wurmbrand, Susi. 2014. The Merge Condition: A syntactic approach to selection. In Minimalism and Beyond: Radicalizing the interfaces, eds. P. Kosta, L. Schürcks, S. Franks, & T. Radev-Bork, Amsterdam: John Benjamins, 39–177. Wurmbrand, Susi and Youssef A. Haddad. 2014. Cyclic Spell-Out Derived Agreement in Arabic Raising Constructions. In Perspectives on Arabic Linguistics XXVIII, eds Y, A. Haddad and E, Potsdam. Amsterdam: John Benjamins, 193-226. Yang, Barry C.-Y . 2012. Intervention Effects and Wh-construals. Journal of East Asian Linguistics 21: 43–87. Yang, Charles. 1997. Minimal Computation: Derivation of Syntactic Structures. Manuscript, MIT. Yoon, Suwon. 2011. A Structural Asymmetry in Intervention Effects, Lingua 121:942–962. Yoon, Junghyoe. 2012. Case drop in Korean---Its empirical and theoretical investigation. Doctoral dissertation, Indiana university. Yoshida, Keiko, and Tomoyuki Yoshida. 1996. Question Marker Drop in Japanese, ICU Language Research Bulletin 11, 37–54. Yokoyama, Tomohiro. 2013. Re-evaluating the “question” marker ka in Japanese. In Proceedings of the 2013 annual conference of the Canadian Linguistic Association. 247 Yun, Jiwon. 2012. The Deterministic Prosody of Indeterminates. Paper presented at WCCFL 29, Somerville, MA. Zubizarrreta, Maria-Luisa. 1998. Prosody, focus and word order. Cambridge MA: MIT Press.

Abstract (if available)

Abstract In this dissertation, I investigate how agreement, movement, and pronunciation determine surface position of a phrase. The main interest of this dissertation is in the question why in some cases an element is pronounced in the position where it is interpreted while in other cases, there is a discrepancy between the position for interpretation and the position for pronunciation. To investigate this issue, I will first discuss a relation between agreement and movement. Inspired by Reverse Agree (Wurmbrand 2012), I will clarify a condition of movement. Based on the suggested relation between agreement and movement, I will propose three different types of movement: phrasal movement, parasitic phrasal movement, and parasitic head movement. The crucial difference between phrasal movement and parasitic (phrasal/head) movement is that an element does undergo movement in the case of phrasal movement, while an element does NOT undergo movement in the case of parasitic movement. ❧ Following the copy theory of movement, I assume that an element that undergoes phrasal movement leaves its copies in its base position and its destination along with other intermediate positions, unlike one that undergoes parasitic movement. Thus, PF needs to choose which copy to pronounce when there is more than one copy in a chain. Zooming in the issue of selection of copy for pronunciation, I will discuss PF constraints that play a role in copy-selection. Especially, adopting Landau (2007)’s intuition that EPP is a pure PF requirement, I will argue that a high-copy privilege assumed in the previous studies are misled by pronunciation-wise reinterpreted EPP. Furthermore, I will argue that once we get rid of the effects of EPP, a low copy is rather preferred to be selected for pronunciation due to economy conditions. I will show how the interaction between the EPP as a PF constraint and an economy condition favoring low-copy pronunciation accounts for both (i) prevalent high-copy pronunciation and (ii) apparent lack of a high-copy privilege across languages. ❧ Based on the system developed, I will provide a typological study in two representative cases of movement: (i) subject agreement/movement and (ii) wh-agreement/movement. This system provides a new approach for the typology of in-situ subjects and in-situ wh-phrases. In the proposed system, in-situ subject/wh-phrases are the results of either parasitic movement or low-copy pronunciation in phrasal movement. An in-situ phrase generated by parasitic movement does not have a copy in a higher position, so it cannot take a high scope. Furthermore, since the phrase does not undergo movement, it is insensitive to movement constraints (e.g. island constraints). By way of contrast, an in-situ phrase generated by a low-copy pronunciation in a movement chain shows “high” behaviors in addition to sensitivity to movement constraints. I will show how the two theoretically possible in-situ subjects/wh-phrases are realized in languages. ❧ Furthermore, based on the definition of the EPP, I will propose various ways to satisfy the EPP. Instead of filling the specifier position of a functional head, I will argue that morphology can be inserted as a means of satisfying the EPP. In addition, I assume that morphology insertion as a way of satisfying the EPP includes a case of inserting an intonation morpheme, suggested by Cheng and Rooryck (2000). This additional way to satisfy the EPP accounts for a range of otherwise puzzling prosody relevant phenomena: i) acceptance of inverse scope reading in so-called rigid scope languages with the help of prosody

Linked assets

University of Southern California Dissertations and Theses

Conceptually similar

PDF

Towards the unity of movement: implications from verb movement in Cantonese

PDF

A reduplicative analysis of sentence modal adverbs in Spanish

PDF

The theory of empty noun in Chinese: with special reference to the right node raising construction

PDF

Agreement on the left edge: the syntax of left dislocation in Spanish

PDF

Building adjectival meaning without adjectives

PDF

The case of a person: The person case constraint in German

PDF

Perspective in Turkish complementation

PDF

Syntax-prosody interactions in the clausal domain: head movement and coalescence

PDF

Silence in answers: a study of ellipsis in Hindi

PDF

Towards a correlational law of language: three factors constraining judgement variation

PDF

Binding and scope dependencies with 'floating quantifiers' in Japanese

PDF

Development of a movement performance assessment to predict ACL re-injury

PDF

Integration of cost effective luminescent materials into organic light emitting diodes (OLEDs) and molecular approaches to improve the device efficiency

Asset Metadata

Creator Kang, Chorong (author)

Core Title Copy theory of movement and PF conditions on spell-out

School College of Letters, Arts and Sciences

Degree Doctor of Philosophy

Degree Program Linguistics

Publication Date 07/24/2017

Defense Date 05/01/2017

Publisher University of Southern California (original), University of Southern California. Libraries (digital)

Tag agreement,experimental syntax,movement,OAI-PMH Harvest,spell-out rule,syntax-prosody interface

Format application/pdf (imt)

Language English

Contributor Electronically uploaded by the author (provenance)

Advisor Simpson, Andrew (committee chair), Li, Audrey (committee member), Saltarelli, Mario (committee member), Zubizarreta, Maria Luisa (committee member)

Creator Email chorongk@usc.edu,chrkkang@gmail.com

Permanent Link (DOI) https://doi.org/10.25549/usctheses-c40-412324

Unique identifier UC11214555

Identifier etd-KangChoron-5614.pdf (filename),usctheses-c40-412324 (legacy record id)

Legacy Identifier etd-KangChoron-5614.pdf

Dmrecord 412324

Document Type Dissertation

Format application/pdf (imt)

Rights Kang, Chorong

Type texts

Source University of Southern California (contributing entity), University of Southern California Dissertations and Theses (collection)

Access Conditions The author retains rights to his/her dissertation, thesis or other graduate work according to U.S. copyright law. Electronic access is being provided by the USC Libraries in agreement with the a...

Repository Name University of Southern California Digital Library

Repository Location USC Digital Library, University of Southern California, University Park Campus MC 2810, 3434 South Grand Avenue, 2nd Floor, Los Angeles, California 90089-2810, USA