Perspective in Turkish complementation

PDF

Download

Open document

Transcript (if available)

Content PERSPECTIVE IN TURKISH COMPLEMENTATION by BETUL ERBASI A Dissertation Presented to the FACULTY OF THE USC GRADUATE SCHOOL UNIVERSITY OF SOUTHERN CALIFORNIA In Partial Fulfillment of the Requirements for the Degree DOCTOR OF PHILOSOPHY [LINGUISTICS] May 2022 Copyright [2022] [Betul Erbasi] ii ACKNOWLEDGEMENTS I am not the same person as I was when I started the PhD program at USC. Writing a dissertation not only trained my research skills, but it also transformed me personally. It helped me become the person I wanted to be, equipped with the right kind of skills. It goes without saying, it was not easy. But I have been blessed with the right people to support me relentlessly in the process. This made it a lot easier than I would have otherwise been capable of. Here are my special thanks to everyone that kept me composed and progressing. First and foremost, I don’t think I could have asked for a better advisor than Maria Luisa Zubizarreta. She has been my research guide, my protector, my biggest supporter. She always saw beyond what I was able to see in my research and opened a whole new horizon for me when I was stuck with my data, which became this dissertation. She not only gave me useful insights about my research, but she was also a great psychological support. She did not give up on me even when I wanted to give up. She believed in me when I didn’t. Maria Luisa, I thank you for all that you did for me. I really hope we can visit Turkey together soon. There is so much to see! My committee members Andrew Simpson, Roumyana Pancheva, Gabriel Uzquiano and Isabelle Charnavel have all been my sources of inspiration as I developed my dissertation. It would have been nowhere close to the consistency I was able to achieve with it without them. Their insights helped me diversify my “perspective” on various issues I discuss in my dissertation and I am grateful for that. I would like to extend special thanks to Isabelle. Not only did her insightful research help me shaped mine, but she has also provided detailed feedback at every stage of the dissertation and helped me think critically when I assumed things. I am grateful for it and I really hope we can meet in person some day. iii My family. They have been with me through thick and thin. Thank you, dad, for encouraging me to pursue and be consistent with my studies. Thank you, mom, for listening calmly to my rants and always supporting me. Thank you, brother, for always being ready to be there for me and listening to my most random conversations. Thank you, Jale (our cat), for scratching me, finishing my coconut oil, being cute and overall for just existing. I love you all. I am very blessed to have the most supporting friends one could have asked for. Nada Al Darrab was always ready to meet me when I felt down, and she knew how to cheer me up. Erum Mushtaq was always available for a deep conversation that nourished my soul. Bhamati Dash understood exactly what I was going through and made me feel that she understood me. That is why she was the best mentor for me in muy first year in the PhD program. Madhumanti Datta showed me she was with me through thick and thin, when she accepted to host me in her place for the last few months of my dissertation on a very short notice. Thank you! My informants: Songül Gündoğdu Yücel, Handan Akıl, Asude Karaaslan, Ayşe Büşra Yakut, Gülay Çevik. If you did not exist, this dissertation would not exist. Thank you for putting up with my constant requests for data discussion, each taking not less than an hour. Special thanks to Songül, who was not only an informant but also the greatest friend. She always had my back and she was as happy as I was when I defended my dissertation. The acknowledgments would be incomplete without a mention of all my PhD colleagues: Ana Besserman, Andres Benitez, Charlie O’Hara, Sarah Lee, Reed Blaylock, Sylvia Kim, Haley Farkas. All have supported me during my dissertation and still continue to do so (special shoutout to Ana, Andres, Charlie, who all helped me with job applications). Special thanks Thomas Borer, Ulli Steindl and Binh Ngo, who introduced me USC Linguistics and Los Angeles and have been some of my biggest allies. I would not have understood and loved Los Angeles at iv all if they had not made that awesome introduction. Thomas and Ulli, I still did not forget that you were kind enough to carry an extra suitcase to Los Angeles from Europe before I moved here. That still means so much! Thank you! Thanks to my cohort, Jesse Storbeck, Sarah Harper, Tanner Sorensen, Yifan Yang, for making my first year a happy time, when I was trying to adapt to the many changes that were happening in my life. Thank you also for your continuing support. How can I forget the great syntactic discussions in the syntax group that broadened my view and made me into a better syntactician? Thank you, Daniel Plesniak, Luis Miguel Toquero Pérez, Tommy Lee, Haley Wei. Haley, I will not forget our hikes together. How much of these you, Junhyo Lee and I spent discussing linguistics (also, thank you, Junhyo Lee, for bringing the philosophical aspect to the discussions!). Special thanks to Audrey Li and Stefan Keine. Both provided very insightful comments and guidance on my earlier research (and Stefan also on Chapter 3 of my dissertation). Thank you, Hajime Hoji, for discussing Language Faculty Science with me! I admire your vision! Thank you, Elsi Kaiser and Mary Byram Washburn, for introducing me into quantitative research in the most fun and productive way and Elsi for also guiding me in my first-ever quantitative data project. It opened up new horizons for me beyond my dissertation. Horizons that will probably change the direction of the rest of my life. Thank you, Barry Schein and Deniz Rudin for introducing me to semantics and helping me think critically and be creative with my thinking. Thank you, Luis Goldstein, for making Phonetics accessible. I would have had no idea what I was doing in my first year when I was already very confused if it was not for your clear teaching. Thank you, Rachel Walker and Karen Jesney, for making phonology accessible to me. Your teaching styles are unmatched. Thank you, Khalil Iskarous, for taking me to Beverly Hills in my first visit during my Open House. I got my v first iPhone thanks to you! Also thank you for helping me in my preparation for job interviews. I really appreciate it! Thank you, Guillermo Ruiz, for making my life with the official processes easy for me - as a non-American, I often had no idea what I was doing- and for having the most fun conversations with me! Thank you for helping me in the organization of syntax meetings, departmental events in my second year and the Person and Perspective Workshop in 2019. I would have failed without your help. Thank you, Lisa Jo, for handling my education records, because I would have lost track of a lot of things without you. Thank you, Prof. Klaus von Heusinger, for helping me work on German and encouraging me to speak in German in any way possible. You have been the greatest host during my stay in Germany and I am grateful. I am looking forward to the time when we will publish our findings! Thank you, Elyesa Seidel, for everything but especially for hosting me in Germany and for coming to Sequoia National Park with me. Thank you Semra Kızılkaya, Frederike Weeber, Lukasz Jedrzejowski for your warm guidance and conversations in Germany! Thank you, Jaklin Kornfilt, for providing me with valuable feedback and insights on my various research projects and for always being one of the friendliest and kindest people I have met. You have set a great example for us Turkish linguists! I have other friends I left in Turkey, but I know they were there for me in spirit. Thank you so much: Dariia Safina, Burcu Elif İriş, Ayşegül Balıkçıoğlu. They never held back any support when I needed and I am blessed to have them in my life. Same goes for my awesome roommates Zainab Anwar and Barakah Al-Ramah. Last but not least, these acknowledgements would not be complete without some thanks to Cafe Dulce at USC Village and their cappuccinos, lattes and breakfast sandwiches as well as vi countless locations of Starbucks and their Iced Pumpkin Spice Lattes (not hot) and Iced Pistaccio Lattes. Thank you, Urth Cafe at Santa Monica, for brightening my weekends. vii TABLE OF CONTENTS Acknowledgements……………………………………………………………………………….ii List of Tables……………………………………………………………………………………..ix List of Figures…………………………………………………………………………………….x Abstract……………………………………………………………………………………….......xi Chapter 1: Introduction ............................................................................................................... 1 1.1 Introduction and Background .................................................................................... 1 1.2 Complementation in Turkish..................................................................................... 9 1.3 Syntactic Comparison between –DIK and diye ....................................................... 12 1.4 Semantic Comparison between –DIK and diye ...................................................... 16 1.5 Chapters Summary ................................................................................................. 19 Chapter 2: The Role of Perspective in Turkish Complementation .............................................. 26 2.1 What does not Differentiate between –DIK and diye Clauses .................................. 31 2.2 Lowest Matrix Subject as Attitude Holder .............................................................. 38 2.3 Perspective ............................................................................................................. 43 2.4 Proposal.................................................................................................................. 71 2.5 Conclusion .......................................................................................................... 104 Chapter 3: When the Speaker is Involved .......................................................... ……………...106 3.1 Optional Matrix Subject Orientation ...................................................................... 110 3.2 Topichood in diye Clauses ................................................................................... 133 3.3 Proposal .............................................................................................................. 141 3.4 Elements that are Obligatorily Associated with Utterance Context ....................... 176 3.5 Interim Summary and Discussion ........................................................................ 178 3.6 Conclusion .......................................................................................................... 186 viii Chapter 4: Embedding Verb Types and Complementation in Turkish...................................... 188 4.1 Perspectival Patterns under Matrix Subject-Oriented Verbs .................................. 197 4.2 Proposal................................................................................................................ 213 4.3 Discussion ........................................................................................................... 233 4.4 Summary and Conclusion .................................................................................... 264 Chapter 5: Conclusion ............................................................................................................. 267 References .............................................................................................................................. 276 ix LIST OF TABLES Chapter 2 Table 1: Who evaluates embedded expressions………………………………………………….47 Table 2: Who evaluates embedded expressions, all combined…………………………………..51 Chapter 3 Table 1: Indexical Shift versus Indexiphors……………………………………………………118 Table 2: Comparison of Indexical Shift, Indexiphors, Turkish 1 st Person Agreement………...125 Table 3: pro and Epithets……………………………………………………………………....169 Table 4: Schlenker’s Classification………………………………………………………….....183 x LIST OF FIGURES Chapter 3 Figure 1: Correlation of (Non-)Optioanlity of Reporting Context and Bound-Element Property………………………………………………………………………………180 xi ABSTRACT Being considered a discourse-pragmatic phenomenon earlier, the role perspective plays in the syntax was peripheral in syntactic research until recently (e.g. Sundaresan 2012, Charnavel 2019). The primary goal of this dissertation is to demonstrate that perspective plays a vital role in syntactic domain formation and various syntactic processes based on data from Turkish complementation. I argue that the two CP-level complement clauses in Turkish differ from each other in whether they encode perspective. This difference, in turn, leads to various syntactic differences such that availability of syntactic encoding of perspective forms an opaque domain, as evidenced by the ban on binding, NPI-licensing, among other patterns. However, the opaqueness is not absolute but is rather selective (Keine 2019), as evidenced by the ability of topic-related elements to escape perspectival domains, including what was earlier considered to be indexical shift cases in Turkish. The investigation of the interaction of embedded perspectival domains with various embedding cognitive attitude verbs reveal important patterns that lead to the conclusion that Turkish makes a three-way syntactic distinction based on perspective between these verbs: speaker-oriented verbs, matrix subject-oriented verbs and perspective- underspecified verbs. I correlate perspective with (non)factivity such that factive reading is a result of syntactic encoding of the speaker’s perspective; non-factivity is a result of encoding the matrix subject’s perspective, and lack of perspective specification leads to what is known as semi-factive verbs. I show that matrix subject-oriented verbs turn embedded opaque domains such as perspectival domains into transparent domains while speaker-oriented verbs turn otherwise-transparent domains into opaque domains. I attribute this to the specific properties of the extra structure they both introduce. 1 CHAPTER 1 INTRODUCTION 1.1. Introduction & Background This dissertation is about perspective in the syntax. In contrast to traditional studies, which argue that perspective is a discourse-pragmatic phenomenon (e.g., Clements 1975, Kuno 1987), it investigates the idea that perspective is syntactically represented, building on studies such as Jayaseelan (1998), Sundaresan (2012) and Charnavel (2019), and that it can be marked overtly in certain languages. Various data on Turkish complementation are presented with the aim to demonstrate that perspective is syntactically represented and has important consequences in terms of domain formation and interaction between various types of elements within a sentence. Along the way, we will see various syntactic effects of perspectival marking and domainhood in Turkish. By the end, the reader will hopefully be convinced that overt markers syntactically delimit perspectival domains in Turkish, domains in which they interact with other elements of the clause. The main data focus of the present study is exemplified in (1): 1 (1) a. Esra [Ali-nin gel-diğ-in]*(-i) bil-iyor E A-3GEN come-DIK-3S-ACC know-IMPF ‘Esra knows [that Ali came/is coming]’ 1 The data collection process for the present study went as follows: The author set up sentences and contexts and judged them herself first. Then, she checked all data with four other native speakers of Turkish consistently, while casually involving a few others in various points. The four speakers and the author agreed on the judgments, unless variation is pointed out. The data that came up after the first round of data collection were checked with two of the four participants due to availability concerns of the participants and occasionally with a third one if there was variation. The data discussion sessions were in semi-structured interview format. 2 b. Esra [Ali gel-di diye] bil-iyor E A come-PST diye know-IMPF ‘Esra knows [diye Ali came]’ (1a) exemplifies a nominalized complement clause, nominalization being evidenced by the genitive subject, person agreement marker on the nominalizer and the obligatory accusative marking on the entire clause, assigned by the embedding verb. (1b) has a non-nominal complement clause headed by a subordinating element diye (compositionally, de+ye, i.e., say+linker). The core argument of this dissertation is that though they both include a CP syntactically (and denote a proposition semantically), diye marks a perspective domain, under which all of the embedded clause content falls, whereas the nominalized complement does not. This is accomplished in Chapter 2. Chapter 3 and 4 discuss properties of this domain and the interaction of this domain with other elements in the clause, respectively. Another focus of this dissertation is the embedding verb type; it is shown that the choice of the verb interacts with the type of the embedded clause in a significant way. The three types of embedding verbs discussed in this study are in (2). They are all cognitive attitude verbs, with the property of perspective (defined based on whether perspective is syntactically represented and if so, whose perspective is represented), which I take to be related to literature on factivity such that matrix subject’s perspective correlates with non-factive verbs and speaker’s perspective correlates with true factive verbs. The factive presupposition is a result of the structural encoding of the speaker’s perspective. 3 (2) a. Perspective-underspecified (=Factivity-underspecified) verbs: ex: bil- ‘know’ 2 or hatırla ‘remember’ b. Matrix Subject-oriented (=Non-factive) verbs: ex. düşün- ‘think’, hayal et- ‘dream’ c. Speaker-oriented (=True factive) verbs: ex. unut- ‘forget’, keşfet- ’discover’ Perspective in syntax is a relatively new domain of investigation. Traditional studies consider the effect of perspective to be a discourse-pragmatic one (e.g., Clements 1975, Kuno 1987), suggesting that perspective is not a core syntactic-semantic phenomenon. More recently, however, some syntactic effects of perspective have been brought to surface and discussed in some detail (e.g., Sundaresan 2012, Charnavel 2019), which highlights the possibility that perspective might play a bigger role in core syntax than was previously assumed. These studies highlight both the discourse-pragmatic role of perspective and its syntactic effects. For instance, Sundaresan (2012) develops a two-step system for Tamil anaphor ta(a)n, where the first step involves a syntactic relation of Agree between the anaphor and the perspective operator while the second step involves a discourse-pragmatically-sensitive relation established between this operator and the DP antecedent (discourse-pragmatic sensitivity due to speaker-intent, common ground, presupposition etc). For Charnavel (2019), each phase can be specified as being presented from an individual’s perspective through LogP, which hosts a perspectival operator and a prolog subject in its specifier. She uses this proposal to account for the syntactic similarities and differences between plain anaphors and exempt anaphors in (primarily) French. Note that exempt anaphors were sometimes previously argued to be not of a syntactic nature and were 2 bil- ‘know’ in Turkish does not refer to knowledge reports as in English. Rather, its meaning is closer to ‘be convinced’. Syntactically, it behaves differently from both verbs in (2b) and (2c), discussed in detail in Chapter 4. Note, however, that the verb root bil- is used to form the noun ‘knowledge’ in Turkish: bil-gi. This remains a puzzle. 4 argued to be logophoric, where logophoricity was taken to be a perspective-based discourse- related notion without syntactic import. This endeavor opened a new line of research based on the quest to find the specific ways in which perspective interacts with the syntax. This dissertation takes on the idea and applies it to Turkish, which not only lends support for the main intuitions of these proposals but also provides some challenges that need to be accounted for. It also takes it one step further by arguing that perspective can be overtly marked in the grammar, as perspective is argued to be encoded by covert elements in both studies. The present work contributes to this recent endeavor through a detailed analysis of Turkish complementation with a focus on its perspectival properties with an extensive database, in fact the most extensive on perspective in Turkish thus far. The present study is unique in its analysis of various types of elements that simultaneously appear in a single perspectival domain (Chapter 2, 3). It also focuses on the interaction of this domain with other elements in the sentence (i.e., embedding verbs). This provides an in-depth understanding of the inner structure of the perspectival domain and its relation to the broader context. This more comprehensive investigation of the perspectival domain reveals that its domain properties are different depending on the properties inside and outside of it. In other words, it is a domain that is not always opaque or always transparent; but rather it is selectively opaque (Keine 2019). Thus, a perspective domain does not (necessarily) correspond to a rigid definition of phases, as was argued for in previous studies (e.g Sundaresan 2012, Charnavel 2019). The selectively-opaque domain of perspective is formed by a perspective operator with a PRO subject, where the PRO subject is associated with the individual that is the perspective center, i.e., the (matrix subject) attitude holder. Crucially, the selectivity of the perspective domain 5 depends on the internal structure of the items involved in the interaction as well as the (non)factivity of the embedding verb. The analysis of individual items in the perspectival domain reveals three types of elements in the domain: (i) those that appear with an anaphoric variable, (ii) those that appear with a pronominal variable and (iii) indexical elements specified for the utterance context. These are exemplified in (3) in respective order. The relevant elements are in italics. (3) a. Context: I know that Ali had a race between 2-3 pm. John thought his race was between 5-6 pm. Ali and John are both in Los Angeles. John [Ali {5-6 arası/#2-3 arası} yarış-tı diye] bil-iyor J A 5-6 between/2-3 between compete-PST diye know-IMPF ‘John knows [diye Ali competed between 5-6/#between 2-3]’ b. Esra [pro sınıf-ta kal-dı-m diye] bil-iyor E class-LOC stay-PST-1S diye know-IMPF ‘Esra knows [diye I failed the class]’ (1SG pro=S (Speaker), MS (Matrix Subject)) c. Context: Mary is in Los Angeles. However, she thinks she is in Irvine. The speaker is in Los Angeles. Mary has an appointment in LA. Mary [toplantı burada/*orada ol-acak diye] bil-iyor M meeting here/there be-FUT diye know-IMPF ‘Mary knows [diye the meeting will be here/*there]’ 6 There is a (somewhat) systematic correspondence of the three groups, i.e., the italicized elements in (3a)-(3c), in the syntax: The first category exemplified in (3a) corresponds to the content of phrase-level elements in the syntax (e.g., VP, DP, AP, AdvP). I argue that all phrase levels in the syntax come with an anaphoric variable, which forces them to be bound by the closest antecedent, which in the present case is the closest perspectival domain. Thus, such content is reported from the first-person perspective of the closest perspective center. The second category in (3b) is topical elements, including the kind of pronominal element (pro) that introduces optional binding by perspectival domains: They can refer to matrix subjects (perspective centers) or the speaker (also a perspective center in the matrix clause; see Ross 1967, Speas & Tenny 2003, Stowell 2007, Zu 2018 for representation of speaker in the matrix clause). The third category as given in (3c) includes lexical elements such as overt pronouns, indexicals such as here that cannot be bound by a lower perspectival domain but must be associated with the utterance context, hence the speaker. This is attributed to the variable they are generated with that is specified for the utterance context (Kaplan 1989). How does this approach to data compare to previous studies? As stated before, most previous studies focus on a specific type of data. To exemplify more concretely the type of data previous studies focus on, traditional or recent, see (4), which exemplifies nominal anaphora (Ross 1967: 466): (4) Abernathyi admitted that the poison pen letter had been written by my sister and himselfi. 7 The anaphor himself in (4) is long-distance bound by the subject Abernathy. This is surprising because anaphors are ordinarily subject to Condition A of the binding theory in English: they must be bound in their local domain, as illustrated in (5) (Chomsky 1986: 166): (5) a. I told themi about [each other]i. b. *I told themi that Bill liked [each other]i. Another type of data that can be related to the notion of perspective various studies have focused on are predicates of personal taste. For example, an adjective can be reported from a non-local perspective as well as a local one and both can appear within the same clause, as shown in (6) (Anand & Korotkova 2021:2). 3 This shows that not only a perspectival element can be bound non-locally, but two different perspectives can be utilized within the same domain. (6) Pascal: Mordecai believes [that the uplifting Pascal documentary is depressing Morcedai] One study that takes a broader approach, in a similar way to the present study, is Schlenker (1999), who investigates all-purpose and matrix indexicals as well as anaphoric devices in various languages including English, Amharic, Russian and German. Eckardt (2015) develops a similar analysis for free indirect discourse in German. These studies are crucial because they attempt to take a much-needed broader look at the relevant phenomena and will be discussed in more detail in Chapter 3 in relation to the present study. However, it must be noted that though similar in spirit, the present work is syntactically-oriented while Schlenker (1999) and Eckardt 3 Anand & Korotkova use the term ‘judge’ (albeit in a theory-neutral way) rather than perspective (though the two are related) but they define predicates of personal taste as expressing a subjective perspective on a matter (p.2). Thus, their data are broadly perspective-related. 8 (2005) were semantically-oriented. This provides crucial unification that has not been sought for in the syntax of perspective but in fact has significant implications. Another important contribution of the present study regards the embedding verbs, which is discussed in Chapter 4. I restrict the analysis in Chapter 4 to cognitive attitude verbs so an in- depth understanding can be achieved, but see Chapter 5 for other non-cognitive attitude verbs for a brief discussion of differences and similarities in behavior. The main relevant distinction in the Turkish case is one of perspective, redefined in terms of the notion of (non)factivity such that factivity presupposition is a result of structural encoding of speaker’s perspective, as mentioned earlier. I argue for a more complex syntactic structure for matrix subject-oriented (i.e., non- factive) cognitive attitude verbs such as düşün- ‘think’ and speaker-oriented (i.e., true factive) verbs such as unut- ‘forget’, as opposed to perspective-underspecified (i.e., factivity- underspecified; semifactive) verbs like bil- ‘know’. The proposal that true factive verbs have extra structure is not new and has its roots in studies as early as Kiparsky & Kiparsky (1971). This corresponds to my argument for speaker-oriented verbs. The proposal that non-factive verbs have extra structure is also not new and contributes to more recent studies such as de Cuba (2007) and his related other work. This corresponds to matrix subject-oriented verbs in the present analysis. Based on Turkish data, the argument will be that both lines of research have their merits and can be unified. I will also show that factivity is default in the absence of specification otherwise, involving no extra structure, which lends some support to the recent studies in the field. In a nutshell, the present study provides a broader, more detailed investigation of the syntactic properties of perspectival domains, thus offering a unified system of perspective for Turkish grammar. The next few sections provide an introduction to complementation in Turkish. 9 1.2. Complementation in Turkish This section discusses properties of different Turkish complement clauses. This discussion is necessary because complementation in Turkish is different from English in various ways. I also discuss why –DIK-nominalizations and diye clauses are opted for in the present work. Nominalization is the main complementation strategy in Turkish. It comes in different types exemplified in (7). The nominal nature of all complement clauses in (7) is evidenced by the genitive-marked embedded subject, person agreement on the nominalizer and the case-marking assigned to the whole embedded clause by the embedding verb. The embedded verb can be any verb: inherent-case marking (7a), transitive (7b), unaccusative (7c) or ditransitive (7c’). (7) a. Esra [Ali-nin karanlık-tan kork-ma-sın]-ı ist-iyor E A-3GEN dark-ABL fear-NMLZ-3S-ACC want-IMPF ‘Esra wants [Ali to fear the dark]’ b. Esra [Ali-nin top-u getir-iş-in]-i izle-di E A-3GEN ball-ACC bring-NMLZ-3S-ACC watch-PST ‘Esra watched [Ali bring the ball]’ c. Esra [Ali-nin gel-diğ-in]-i bil-iyor E A-3GEN come-DIK-3S-ACC know-IMPF ‘Esra knows [that Ali came/is coming]’ c’. Esra [Ali-nin kitab-ı Ahmet-e ver-eceğ-in]-i bil-iyor E A-3GEN book-ACC A-DAT give-FUT-3S-ACC know-IMPF ‘Esra knows [that Ali will give the book to Ahmet]’ 10 There are also two non-nominal complementation types in Turkish. One is formed with diye as in (8a) and the other is ki as in (8b). Diye forms a pre-verbal complement clause while a ki clause must be post-verbal. In both, the embedded subject is nominative-marked, as it is in a matrix clause. Since the nominative is null in Turkish (Kornfilt 2003), I do not mark it in this study: (8) a. Esra [Ali gel-di diye] bil-iyor E A come-PST diye know-IMPF ‘Esra knows [diye Ali came]’ b. Esra bil-iyor [ki Ali top-u getir-di] E know-IMPF ki A ball-ACC bring-PST ‘Esra knows that Ali brought the ball’ The focus of this dissertation is on the nominalized complement in (7c), namely clauses formed with –DIK nominalizer, and the non-nominal complement in (8a), namely clauses headed by diye. The reason for choosing these two embedded clause types is two-fold. One is the nature of restrictions the complementizers pose in terms of the embedding verb’s sub-categorizational patterns. I start the discussion with the embedded clause types that are not the focus of the present study for the sake of completeness: –mE in (7a) tends to be selected by bouletic verbs where they are the only choice of complementation (e.g., iste- ‘want’ can only take a –mE clause as a complement). They are also possible under other verbs like izle- ‘watch’, iste- ‘believe’ and emotive verbs like ‘şaşır-’ be.surprised, in which case it exists alongside other complement types (e.g., –(y)Iş or –DIK). The important sub-categorizational pattern for present purposes, however, is that –mE does not occur (easily) with cognitive attitude verbs like bil- ‘know’ and düşün- ‘think’ (though it is possible), 11 which are the focus of the present study. Thus, -mE clauses are excluded from the present study unless necessary for contrast purposes. –(y)Iş clauses occur under all verbs other than bouletic verbs, which entails they can occur under cognitive attitude verbs as well. However, their major meaning contribution is a manner meaning (Erdal 1998, Bassac & Cicek 2013), which restricts the domain of investigation here as it is not easily tied to perspective; therefore, they are excluded from the present study. Ki is borrowed from Persian and is the only post-verbal complementation in Turkish (Turkish is consistently head-final, so all other complements occur to the left of the verb). This special status is a confound, since it does not allow a direct comparison of complements in the same position. (Moreover, Predolac (2017) argues that ki clauses are not complementation clauses via subordination or coordination in the first place but rather they are paratactical structures.) Ki clauses are therefore left out of the present study due to structural differences. As for –(y)EcEK in (7c’), –DIK and –(y)EcEK have been argued to be two forms of the same nominalizer. –DIK is the indicative, factive, non-future (i.e., past and present) or perfective form while –(y)EcEK, which is the subjunctive, future or imperfective form (Kornfilt 1984, 2003, Lewis 1967, Underhill 1976, Demirok 2019). I focus only on –DIK (non-future) in this dissertation as it allows for a wider range of meanings (e.g., past and present (including habitual) versus only future) and thus allows for a better test for understanding the role of perspective in complement clauses in Turkish. –(y)EcEK is more restricted as it only refers to future and is not particularly helpful in the discussion of perspectival patterns. This leaves a contrast between –DIK and diye as the focus point of this dissertation. -DIK and diye can be easily contrasted under very similar sets of verbs. In fact, such contrasts between –DIK and diye clauses have already been made (Özyıldız 2016, 2017). As shown below and in 12 more detail in Chapter 2, their internal structure is also comparable. This comparability is the second reason why –DIK and diye are chosen in the present study. 1.3. Syntactic Comparison between –DIK and diye The goal of this section is to show that –DIK and diye clauses are syntactically comparable in that they both have TPs and CPs. I start with their differences. The main structural difference between –DIK and diye is that the former is a nominalizer whereas the latter is a non-nominal subordinator. Thus, diye clauses cannot be case-marked (9b) whereas –DIK clauses must be (9a). Conversely, diye can be preceded by a nominal element (10b) that is case-marked by the verb whereas –DIK cannot (10a). 4 These data show that –DIK clauses satisfy the nominal argument requirement by the verb (for case-assignment purposes): (9) a. Esra [Ali-nin gel-diğ-in]*(-i) bil-iyor E A-3GEN come-DIK-3S-ACC know-IMPF ‘Esra knows [that Ali came/is coming]’ b. Esra [Ali gel-di diye](*-i) bil-iyor E A come-PST diye-ACC know-IMPF ‘Esra knows [diye Ali came]’ (10) a. *Esra bu durum-u [Ali-nin gel-diğ-in]*(-i) bil-iyor E this situation-ACC A-3GEN come-DIK-3S-ACC know-IMPF ‘Intended: Esra knows of this situation [(it is a situation) that Ali came/is coming]’ 4 Note that (10b) shows that diye can be in an adjunct position to a nominal. Indeed, diye has multiple functions in Turkish, some of which are discussed in various parts of this dissertation (also see --Gündoğdu 2017). The specific structure in (10b) is discussed in Chapter 2. 13 b. Esra bu durum-u [Ali gel-di diye] bil-iyor E this situation-ACC A come-PST diye know-IMPF ‘Esra knows this situation [diye (as) Ali came]’ Another difference is tense-aspect marking. The complement of diye can have all tense-aspect markers (indicating it has a TP), exemplified with past (11a), indirect past (11b), future (11c) and imperfective (11d). –DIK itself marks non-future, which indicates that it also has a TP, though possibly defective (Kornfilt 2003). However, -DIK clauses can host most tense-aspect markers via the copula ol- ‘be’ (12). 5 (11) a. Semra [Ali gel-di diye] bil-iyor S A come-PST diye know-IMPF ‘Semra knows [diye Ali came]’ b. Semra [Ali gel-miş diye] bil-iyor S A come-INDPST diye know-IMPF ‘Semra knows [diye Ali came (as it is said)]’ c. Semra [Ali gel-ecek diye] bil-iyor S A come-FUT diye know-IMPF ‘Semra knows [diye Ali will come]’ d. Semra [Ali gel-iyor diye] bil-iyor S A come-IMPF diye know-IMPF ‘Semra knows [diye Ali is coming]’ 5 The copula cannot attach to the past marker in Turkish (Kelepir 2001), which prevents it from occurring in a –DIK clause. But since –DIK itself can express past meaning, -DIK clauses can ultimately express past meaning. Note also that the ability of adding tense-aspect markers via a copula is available for all nominalizers, even for those that are VPs in their bare form. 14 (12) a. Semra [Ali-nin gel-miş ol-duğ-un]-u bil-iyor S A-3GEN come-INDPST be-DIK-3S-ACC know-IMPF ‘Semra knows [that Ali came/had come]’ b. Semra [Ali-nin gel-ecek ol-duğ-un]-u bil-iyor S A-3GEN come-FUT be-DIK-3S-ACC know-IMPF ‘Semra knows [that Ali will come]’ c. Semra [Ali-nin gel-iyor ol-duğ-un]-u bil-iyor S A-3GEN come-IMPF be-DIK-3S-ACC know-IMPF ‘Semra knows [that Ali is coming]’ There is evidence that –DIK and diye both possess a CP as well. For one, diye can host wh- words and question particles: 6 (13) a. Semra [kim gel-di diye] düşün-üyor S who come-PST diye think-IMPF ‘Whoi does Semra think that s/hei came?’ ‘Semra is thinking who came’ 7 b. Semra [Ali gel-di mi diye] düşün-üyor S A come-PST Q diye think-IMPF ‘Does Semra think that Ali came?’ ‘Semra is thinking whether Ali came’ 6 This depends on the verb they are embedded under. Under bil- ‘know’, a question element within a diye clause has only wide scope while under non-factive verbs like düşün ‘think’, the wh- word can have wide or narrow scope. – DIK clauses have a different pattern: they allow both wide and narrow scope with wh- words, regardless of the embedding verb. What matters for now is that diye sometimes allows question elements in its scope, which I take to be evidence for CP. The verbal distinctions will be discussed in mostly Chapter 4. 7 Notice that think is translated as in the progressive. Indeed, when a question element appears in the complement of a non-factive verb in Turkish, the interpretation is one of ‘consider/ wonder/think about’, which I tried to reflect with the progressive form. Thanks to Roumyana Pancheva for this insight. See Özyıldız (2021) for an analysis of the verb think based on eventuality types it can encode (eventive versus stative). 15 (14) a. Semra [kim-in gel-diğ-in]-i düşün-üyor S who-3GEN come-DIK-3S-ACC think-IMPF ‘Whoi does Semra think that s/hei came?’ ‘Semra is thinking who came’ b. Semra [Ali-nin mi gel-diğ-in]-i düşün-üyor S A-3GEN Q come-DIK-3S-ACC think-IMP ‘Does Semra think that Ali came?’ ‘#Semra is thinking whether Ali came’ -DIK is also the form used to form relative clauses in Turkish. For this part of the argumentation, I follow Kornfilt (2003) and use her examples (p. 145, ex. (11)-(13)) with minimal alteration. Kornfilt (2003) takes this set of data (as well as data on ability of –DIK to host wh- words) as evidence that it possesses the CP layer. To start, (15a) is an example of a non-reduced relative clause. No other nominalizer can be used in this position, as shown in (15b) with –mE. (15b) can be saved if it is embedded under a –DIK nominal, as in (15c): (15) a. Ali-nin pişir-diğ-i yemek A-3GEN cook-DIK-3S meal ‘The meal Ali cooked’ b. *Ali-nin pişir-me-si yemek A-3GEN cook-mE-3S meal ‘The meal Ali should cook’ c. [[Ali-nin pişir-me-sin]-i söyle-diğ-im] yemek A-3GEN cook-mE-3S tell-DIK-1S meal ‘The meal which I said Ali should cook’ 16 Diye can be used as complement clauses to indefinite head nouns, which I take to be an indication of a CP-level structure: (16) Ali geldi diye bir söylenti A come-PST diye a rumor ‘The rumor that Ali came’ The conclusion is that both –DIK and diye have CPs (see Kornfilt 2003 for the same conclusion for –DIK based on the evidence discussed here). In Chapter 2, this point is taken up again and discussed as to how –DIK clauses and diye clauses differ with respect to the CP layer. 1.4. Semantic Comparison between –DIK and diye In this section, it is established that both –DIK and diye clauses denote propositions. This provides the background necessary to discuss them under attitude verbs in later chapters, as these verbs take a proposition as their argument in most analyses. The deficient tense-aspect marking in –DIK clauses and their nominal nature do not prevent them from being treated as propositions. The propositional status of –DIK clauses is best observed when compared with -mE clauses, argued to lack CP-level (Kornfilt 2003) as well as tense-aspect phenomena (hence, it is a VP- nominalization). This comparison is the backbone of this section, following Demirok (2019). The first argument for the propositional status of –DIK clauses is a negative one: they cannot refer to events (17a) while –mE nominalizations can, as demonstrated in (17b). Likewise, the predicate in (18), iki dakika sür- ‘last two minutes’, makes reference to events and is therefore compatible with –mE clauses (18b) but not -DIK clauses (18a): 17 (17) a. *[Suzan-ın hata-yı bul-duğ-u] harika bir olay-dı S-3GEN mistake-ACC find-DIK-3S amazing one event-PST ‘Intended: That Susan found the mistake was an amazing event’ b. [Suzan-ın hata-yı bul-ma-sı] harika bir olay-dı S-3GEN mistake-ACC find-mE-3S amazing one event-PST ‘Susan’s finding the mistake was an amazing event’ (18) a. #[Suzan-ın hata-yı bul-duğ-u] iki dakika sür-dü S-3GEN mistake-ACC find-DIK-3S two minute last-PST ‘Intended: It took two minutes for Susan to find the mistake’ (lit. That Susan found the mistake took two minutes) b. [Suzan-ın hata-yı bul-ma-sı] iki dakika sür-dü S-3GEN mistake-ACC find-mE-3S two minute last-PST ‘For Susan to find the mistake took two minutes’ Both (17) and (18) point to the conclusion that –mE clauses refer to events while –DIK clauses do not refer to events. What does the latter refer to? The next sets of data let us conclude that – DIK clauses denote propositions. -DIK clauses can be the subject of a predicate referring to truth (19a) while –mE clauses cannot (under a truth reading, but ok if the predicate takes on a meaning that can make reference to an event) (19b). This indicates that –DIK clauses have truth values, which in turn indicates that they are propositions: (19) a. [Suzan-ın hata-yı bul-duğ-u] doğru-ydu S-3GEN mistake-ACC find-DIK-3S true-PST ‘It was true that Susan found the mistake’ 18 b. [Suzan-ın hata-yı bul-ma-sı] doğru-ydu S-3GEN mistake-ACC find-mE-3S true-PST ‘Susan’s finding the mistake was true (i.e., was the right thing to do) ‘#It was true that Susan found the mistake’ When –DIK clauses occur under cognitive verbs (i.e., verbs with propositional complements –at least complements associated with propositions (Moulton 2009, 2013, among others)-), their truth value can be challenged, indicating that they denote propositions: (20) A: Semra [Ali-nin gel-diğ-in]-i bil-iyor S A-3GEN come-DIK-3S-ACC know-IMPF ‘Semra knows [that Ali came]’ B: Ali gelmedi ki. ‘But Ali did not come.’ Diye clauses also denote propositions. They do not occur as subjects unless the embedding predicate is in the passive (e.g., biliniyor ‘is known’). But we can test them under propositional verbs and see that their truth value can be challenged (21): (21) A: Semra [Ali gel-di diye] bil-iyor S A come-PST diye know-IMPF ‘Semra knows [diye Ali came]’ B: Ali gelmedi ki. ‘But Ali did not come.’ 19 This concludes the discussion on the core similarities between –DIK and diye clauses, making them the best candidates for a comparative study under cognitive attitude verbs. In the next section, some details of the main proposals of each chapter are provided. 1.5. Chapters Summary Chapter 2 is the backbone of this dissertation as it establishes the main syntactic and interpretational properties of nominalized complement clauses formed with –DIK and diye clauses. The discussion of these clauses results in the following generalization: Generalization: Diye clauses encode a perspective center (PC) position, forcing all phrasal content in the clause to be evaluated from the point of view of the attitude holder occupying the PC position; this constitutes a single perspective domain. Only the closest matrix subject can fulfil the role of perspective center. (–DIK clauses do not encode any perspective). This chapter explores these two complement clauses under what I refer to as ‘perspective- underspecified’ verbs such as bil- ‘know’ or hatırla ‘remember’ in Turkish, though bil- ‘know’ is used predominantly for exemplification purposes. The reason for this choice of embedding verbs becomes clear in Chapter 4. Briefly stated, it is because these verbs are the only ones among the ones analyzed in the present study (i.e., cognitive attitude verbs) that do not impose syntactic or interpretational restrictions on their complements, allowing the default form and interpretation of the complement clauses to be realized. In other words, they do not introduce extra structure into the derivation, which translates into them being bare VPs in the syntax. They do not encode 20 perspectival properties unlike speaker-oriented verbs (e.g., unut- ‘forget’, keşfet- ’discover’) or matrix subject-oriented verbs (e.g., düşün- ‘think’, hayal et- ‘dream’), making them semantically plain as well. Chapter 4 shows that having extra structure significantly affects how the complement clauses are interpreted and/or behave syntactically. To account for the generalization stated above, I propose that diye is the overt realization of a perspectival operator with a PRO subject, introduced via a phrase level Pers(pective)P(hrase). PersP takes a CP complement, which corresponds to the embedded clause below diye. This is represented in (22), when the diye clause is embedded under a VP: (22) PersP under perspective-underspecified verbs combines directly with a VP and yields a syntactically opaque domain. This appears to support proposals that perspectival domains are phases (e.g Sundaresan 2012, Charnavel 2019), but in Chapter 3 it is shown that this correlation does not hold for the Turkish data in its strong form. Another implication of this proposal is that diye is the overt realization of a perspectival operator, which, to my knowledge, is the first time such proposal has been made. Perspectival operators have been assumed to be silent elements. This has sometimes called their relevance for syntax into question. The present proposal provides more direct evidence for overt encoding of 21 perspective in natural language, and therefore strengthens the position that perspective is relevant for syntax. Note also that the operator heads the PersP projection, which is unusual, given that operators are often assumed to be in specifier positions. In this study, I follow Charnavel (2019), who argued for a perspectival operator heading Log(ophoric)P(hrase), with a pro subject in its specifier. The function of this operator is to report the embedded content from the first-person perspective of the matrix subject (à la Charnavel 2019) by binding anaphoric world variables that all phrasal content in its scope comes with. This makes diye clauses strictly matrix subject- oriented. In contrast, –DIK nominalizations are syntactically transparent domains (i.e., they do not block NPI-licesing, quantifier interaction) and they do not encode perspective. Lacking a perspective operator, the anaphoric variables within them are bound by the next available perspectival domain, which is the speaker in the matrix clause. That explains why –DIK clauses must involve the speaker’s perspective. However, we will see that any embedded clause involves the matrix subject and an agreement between the perspectives of the speaker and the matrix subject is required in -DIK clauses. This can be seen in the light of studies such as Giorgi & Pianesi (2001), discussed in Chapter 2 in more detail. After Chapter 2 establishes the above-stated patterns, Chapter 3 takes on two types of exceptions: (i) elements that can be bound by the diye operator or by higher operators and (ii) elements that cannot be bound by the diye operator but must be related to the matrix operator (i.e., the utterance context). Optional binding arises due to the availability of pro in the relevant elements, which is associated with a topic function in Turkish. To account for this pattern, I follow recent studies on selective opacity (Keine 2019, especially Poole 2020) and propose that diye clauses are selectively opaque domains, their selectivity depending on the height of the 22 probe (Poole 2020). I propose that the unvalued [topic] feature in pro in the diye clause makes them want to be valued by TopicP(hrase), (one of) the highest projections in Turkish clausal structure (Şener 2010). When pro is associated with this probe, it takes the salient individual as antecedent. On a height-related account of selective opacity of diye clauses, TopicP is higher than PersP in the structural sequence, enabling it to probe into the diye clause (23). This contrasts with NPI-licensing, anaphor binding and quantifier scope, all of which happen below PersP. 8 (23) 8 There are two PersPs in (23); the higher one for the matrix clause, the lower one for the embedded diye clause. 23 Elements that must be associated with the utterance context come with a variable that indicates just that, following traditional analyses (e.g., Kaplan 1989). Thus, they are lexically specified to be evaluated in the utterance context. OPdiye cannot bind such elements. Chapter 4 moves on from inside diye clauses to the clause that embeds them, specifically the embedding verb. All in all, I show that the embedding verb is an important factor in determining the syntactic behavior of diye clauses. Both true factive (=encoding the speaker’s perspective) and non-factive (=encoding matrix subject’s perspective) verbs take an extra CP complement, which takes the embedded complement clause as its own argument. This is represented in (24a) and (24b), respectively. Under the former, diye clauses simply cannot function as arguments. Under the latter, diye clauses are no longer opaque domains and their contrast with –DIK clauses are neutralized (e.g., interpretational differences). CP matrixsubject encodes the matrix subject’s perspective and CPspeaker encodes the speaker’s perspective. Only the latter has a PRO subject. The lack of PRO in the former leads to the complement clause being a transparent domain. Under factivity-underspecified verbs as in (24c), diye clauses are opaque domains and contrast with –DIK clauses interpretationally and syntactically. 9 9 DP/PersP in (24b) and (24c) means that these verbs sub-categorize for either a –DIK clause or a diye clause. 24 (24) a. Speaker-oriented b. Matrix subject-oriented c. Perspective-Underspecified Chapter 5 concludes the dissertation with a summary of the main findings. It also addresses some loose ends that merit further investigation. One is when diye clauses are embedded by emotive and speech verbs. These verbs share some of the properties of true factives and non-factives at the same time, making it difficult to fit them into the current picture but also indicating that the 25 same kinds of factors play a role. This is promising. The other is when diye clauses are used as reason clauses. In this case, reason clauses formed with diye display some perspectival properties as opposed to reason clauses formed by için ‘for’, which do not exhibit perspectival properties. This correlation is also a promising line of research for future studies. 26 CHAPTER 2 THE ROLE OF PERSPECTIVE IN TURKISH COMPLEMENTATION This chapter and the next are about the core properties and interactions in perspectival domains in complement clauses in Turkish. The combined goal of these chapters is to establish that diye clauses correspond to a selectively opaque perspectival domain whereas –DIK clauses are transparent non-perspectival domains. Diye is a perspective marker, embodied as the overt realization of a perspectival operator in the head of Pers(pectival)P(hrase). This analysis not only adds to arguments that perspective is syntactically represented (e.g., Sundaresan 2012, Charnavel 2019), but it also asserts that the perspective marker itself can be overt. I show that the perspectival domain formed with diye has important syntactic properties. Specifically, it is established that diye forms an opaque domain. In Chapter 3, I revise this conclusion and argue that diye clauses are selectively opaque domains (Keine 2019). This conclusion is arrived at via a detailed analysis of various types of elements within the diye domain: (i) those that appear with an anaphoric variable, (ii) those that appear with a pronominal variable and (iii) indexical elements that are specified for the utterance context. The first category corresponds to the content of phrase-level elements in the syntax (e.g., VP, DP, AP, AdvP), which are discussed in this chapter. The discussion establishes that these elements must be reported from the first- person perspective of the closest perspective center, which is the lowest attitude holder. The latter two are elements that do not align with the generalization established in this chapter. These data are discussed in Chapter 3. The second group includes the data that are optionally-bound by the closest perspectival operator. This includes structures with pro in Turkish, such as first- 27 person agreement (as well as epithets, though epithets have more controversial properties). The latter are indexicals such as burada ‘here’ and overt pronouns in Turkish, which are associated with the utterance context. These data types can all appear in diye clauses simultaneously. Though they do not interact with each other directly, each reveals an important property of perspectival domains in Turkish. This broad outlook into diye clauses have not been practiced before in the Turkish literature. Previous studies either focused on diye clauses as related to a belief component (e.g., Özyıldız 2016), which I take to be most similar to the data I discuss in this chapter, namely (i), or they focused on the so-called indexical shift data (e.g., Şener & Şener 2011, Yıldırım-Gündoğdu 2017a), which correspond to category (ii), discussed in Chapter 3. Not only are these data combined together in the present study with the addition of some novel data sets, but category (iii) is also added, which I believe has not been discussed explicitly in the relevant literature on Turkish. I take this to be a major contribution of the present study, as this approach facilitates a better understanding of the syntax of the complement clauses as well as providing extensive data. Most studies on the topic of perspective in the area deal with person markers, anaphors or pronouns (e.g., Anand 2006, Charnavel & Sportiche 2017, Charnavel 2019, Bogal-Allbritten 2016), as well as other indexical elements such as temporal adverbs (e.g., Schlenker 1999), but usually not both together, with some exceptions such as Deal (2020). Some other studies focus on predicates of personal taste (Anand & Korotkova 2021), but do not include the elements mentioned earlier. All in all, most studies focus on a specific type of data, which can miss out on some important properties of perspectival domains as a whole. In other words, focusing on different types of data within a perspectival domain can bring out patterns that can otherwise be missed. For example, if the present study had followed suit and only focused on category (i) 28 above, the conclusion would have been that diye clauses are fully opaque domains. If the focus was on (ii) and/or (iii), the conclusion would have been that diye clauses are not opaque. Yet, the combination of the two sets of data allows us to reach a more detailed and telling conclusion: diye clauses are sometimes opaque and sometimes not, conditioned by certain factors. The investigation of when the opaqueness can be violated brings out valuable patterns that lead to interesting syntactic revelations. In the end, the conclusions in Chapter 2 and 3 come closest to Schlenker (1999) in attributing the difference with respect to “shifting behavior” to the indexicals themselves. However, Schlenker (1999) was semantically-oriented whereas the present study is concerned with the internal syntax of the bound elements as well as the syntactic structure and properties of perspectival domains. There are also differences as to how elements are categorized, as will be discussed in Chapter 3. A more specific contribution of Chapters 2 and 3 is that it is the first study in Turkish that focuses on the interpretation of diye clauses in this much detail and one of the first to focus on its syntax (see Yıldırım-Gündoğdu 2017a for an analysis of the syntax of diye under verbs of communication) in a way that connects the observed patterns to independent facts of the language. The proposal developed in this chapter is as follows: diye is the overt realization of a perspectival operator (OPdiye) that occupies the head of a designated phrase, PersP, which takes a CP as its complement and PRO subject in its specifier. This is represented in (1): 10 10 Recall from Chapter 1 that diye clauses possess a CP. That is why the argument to PersP is a CP and not a TP. We will also see that diye clauses can host various elements, discussed in relevant parts of the dissertation. 29 (1) OPdiye reports the embedded content from the first-person perspective of the closest matrix subject, thereby forming a complete evaluation domain corresponding to a syntactically opaque domain. A complete evaluation domain here refers to a domain that is full in interpretation: it includes what is interpreted (i.e., the proposition), the evaluator (i.e., the matrix subject) and the evaluation worlds (i.e., logophoric alternatives of the matrix subject). In effect, this definition assimilates OPdiye to an intensional predicate, which aligns the present study with previous proposals that diye has a verbal component since it is derived from the verbal root de- ‘say’ (Yıldırım-Gündoğdu 2017a; Major 2021, to appear, for a similar marker dep in Uyghur). This approach to OPdiye is adapted from Charnavel (2019), who proposes a logophoric operator to bind exempt anaphors in French. With respect to the elements bound by OPdiye, the assumption is that phrase levels such as VP, NP/DP, AdvP, AP come with a silent syntactically-represented world variable that must be bound by the closest binder (à la Percus 2000), which the operator in the closest perspectival domain can bind. To that end, I argue that diye complementation is the only embedded clause that marks perspective in Turkish. I also assume that matrix clause is also a perspectival domain, with the speaker being the perspective center (adapted from Ross 1967, Speas & Tenny 2003, Stowell 2007, Zu 2018). Other than these, there are no perspectival domains in Turkish. In other words, unlike Charnavel (2019), who argues that all phases (corresponding to phrase levels such 30 as DP, CP, AP) can form perspectival domains in French, I argue that perspective is restricted to some of the CP-level structures in Turkish. Given this, the elements that come with an anaphoric variable can only be bound in the lowest diye clause, since the lowest operator that can bind them is OPdiye. This explains why they must be reported from the perspective of the closest matrix subject. In this chapter, only perspective-underspecified (attitude) verbs such as bil- ‘know’ and hatırla- ‘remember’ are used as embedding verbs (though I stick to the former for consistency). The reason for this will be clearer in Chapter 4. For the meanwhile, it will suffice to say that perspective-underspecified verbs are the only cognitive attitude verbs without specification for perspective in their syntax or semantics. This allows us to notice patterns within the complement domain without interpretive or syntactic effects coming from the embedding verb. Also note that bil- ‘know’ in Turkish does not refer to knowledge reports as know in English. Rather, its meaning is closer to ‘be convinced’. Lastly, I use (non)factivity in discussion of general literature, as this is the term that is commonly used. In the discussion of Turkish data, I use perspective instead. To that end, keep in mind that true factive=speaker-oriented, non- factive=matrix subject-oriented, factivity-underspecified/semi-factive=perspective- underspecified. The chapter is organized as follows. In Section 2.1, some previous literature on diye and –DIK clauses is discussed. In Section 2.2, I show that the matrix subject is an attitude holder in both diye and –DIK clauses. In Section 2.3 the main set of data is presented, namely the data that need to be reported from the matrix subject’s perspective in diye clauses and require conscious evaluation of the propositional content within its immediate scope. We also show the absence of 31 such properties in –DIK clauses. Section 2.4 presents the proposal outlined above in detail and discusses how it accounts for the data patterns. Section 5 concludes the chapter. 2.1. What does not Differentiate between –DIK and diye Clauses Özyıldız (2016, 2017) discuss -DIK and –diye clauses in light of factivity alternation, exemplified in (2). In a context where we know the embedded proposition to be false, -DIK clauses are not acceptable, shown in (2a) under the (semi-)factive verb bil- ‘know’, whereas diye clauses are acceptable, as shown in (2b). (2) Context: Trump won the election, but… a. #Tunç [Bernie-nin kazan-dığ-ın]-ı bil-iyor T B-3GEN win-DIK-3S-ACC know-IMPF ‘#Tunç knows [that Bernie won]’ b. Tunç [Bernie kazan-dı diye] bil-iyor T B win-PST diye know-IMPF ‘Tunç knows [diye that Bernie won]’ Based on contrasts like in (2), Özyıldız concludes that truth is what differentiates between –DIK and diye clauses. Özyıldız (2016) argues that diye clauses introduce a secondary belief component as well as its propositional content while the knowledge component is contributed by the matrix predicate. –DIK clauses are knowledge propositions. Clauses with diye yield a non- factive interpretation and those with –DIK yield a factive interpretation (both under bil- ‘know’). 32 Neither side of this argument holds across all data. On the one side, as acknowledged by Özyıldız himself, factivity alternation between complement clauses disappears when the embedding verb is non-factive (see Chapter 4 for a more detailed discussion on this point), where factivity implies the truth of the attitude proposition 11 and non-factivity lacks such implication. This is because the overall interpretation is non-factive regardless of the complement (3): (3) Context: Trump won the election, but… a. Tunç [Bernie-nin kazan-dığ-ın]-ı düşün-üyor T B-3GEN win-DIK-3S-ACC think-IMPF ‘Tunç thinks [that Bernie won]’ b. Tunç [Bernie kazan-dı diye] düşün-üyor T B win-PST diye think-IMPF ‘Tunç thinks [diye that Bernie won]’ Conversely, the compatibility of the diye clause in (2b) with the given context indicates that it is non-factive, but diye is not anti-factive (Özyıldız 2017). It is compatible with situations where the embedded proposition happens to be true (in the actual world and/or for the speaker). This is exemplified in (4), where both diye (4a) and -DIK (4b) clauses are acceptable in a context where the embedded proposition has been established as true: (4) a. A: Ali kaçtı, biliyorsun değil mi? Ali ran away; you know that, right? B: Evet biliyorum. 11 It is not clear whether truth refers to the speaker’s knowledge state or truth in actual world in Özyıldız’s studies. However, he (personal communication) pointed out that his definition might indicate that the speaker and the hearer agree with the truth evaluation. However, this is not what he aimed to get formally, but rather truth in actual world. 33 Yes, I do. A: Başka kim biliyor? Who else knows it? B: Pelin de [Ali kaç-tı diye] bil-iyor P too A escape-PST diye know-IMPF ‘Pelin also knows [diye Ali ran away]’ b. A: Ali kaçtı, biliyorsun değil mi? Ali ran away; you know that, right? B: Evet biliyorum. Yes, I do. A: Başka kim biliyor? Who else knows it? B: Pelin de [Ali-nin kaç-tığ-ın]-ı bil-iyor P too A-3GEN escape-DIK-3S-ACC know-IMPF ‘Pelin also knows [that Ali ran away]’ Thus, neither diye clauses are always non-factive nor –DIK clauses are always factive. This calls into question whether truth is a reliable differentiating factor between the two embedded clause types. One might argue that diye clauses are neutral as to the truth of the embedded proposition, since it never presupposes the truth of its complement under cognitive verbs. However, we are forced to take –DIK clauses as neutral too since they are factive under factive verbs and non- factive under non-factive verbs. In other words, -DIK clauses also do not come with a factivity inference either. This reduces to saying that both diye and –DIK clauses are neutral as to the truth of the embedded proposition, which implies that truth is not the differentiating factor between the two clauses. In this case, we are back to question with which where we started: what differentiates between –DIK and diye clauses? 34 Another potential criterion proposed to differentiate between –DIK and diye clauses is justification, which is conditionally imposed on diye clauses but never in –DIK clauses. Özyıldız (2016, 2017) show that diye clauses require justification for the matrix subject to hold the belief that the content of the embedded clause under the (semi-)factive verb bil- ‘know’ is true. The justification in (5) (Özyıldız 2017: 18, ex. (29), with some modifications) is the prank the guards play on Tunç (Context 2). The justification makes the diye clause acceptable, as opposed to a mere belief context (Context 1). Justification can be direct or indirect evidence: (5) Context 1: NON-ACCEPTABLE Tunç is cut off from getting any information about the world news but is obsessed by Hillary and accepts no state of the world where she hasn’t won. He says to the speaker: “Hillary won”. Hillary indeed won. Context 2: ACCEPTABLE Tunç is cut off from getting any information about the world news but is obsessed by Hillary and accepts no state of the world where she hasn’t won. In addition, the guards celebrate a Hillary victory outside his door. He says to the speaker: “Hillary won”. Hillary indeed won. Tunç da [Hillary kazan-dı diye] bil-iyor 12 T too H win-PST diye know-IMPF ‘Tunç also knows [diye Hillary won]’ Although not discussed by Özyıldız, we can show that no acceptability difference between contexts arises when the diye clause in (5) is replaced by a –DIK clause, as shown in (6). The 12 The addition of the particle da ‘too’ is needed for diye to sound more natural under bil- ‘know’ in a context where the embedded proposition is true. Özyıldız did not have a statement regarding the actual truth in the context, so he did not use the particle. I added such a statement to make the contrast with –DIK clauses in (6) more direct and – DIK clauses under bil- ‘know’ have factivity inference. In other words, since I have to use a statement regarding the truth of the embedded proposition in (6), I also added it in (5). 35 intuition here is that the speaker need not know the source of the matrix subject’s knowledge, or that the matrix subject has a source/evidence for believing that the embedded proposition is true. Rather, the speaker can take at face value the statement that the matrix subject believes that the embedded proposition is true. This allows –DIK clauses to be accepted in mere belief contexts such as Context 1. 13 (6) Context 1: ACCEPTABLE Tunç is cut off from getting any information about the world news but is obsessed by Hillary and accepts no state of the world where she hasn’t won. He says to the speaker: “Hillary won”. Hillary indeed won. Context 2: ACCEPTABLE Tunç is cut off from getting any information about the world news but is obsessed by Hillary and accepts no state of the world where she hasn’t won. In addition, the guards celebrate a Hillary victory outside his door. He says to the speaker: “Hillary won”. Hillary indeed won. Tunç [Hillary-nin kazan-dığ-ın]-ı bil-iyor T H-3GEN win-DIK-3S-ACC know-IMPF ‘Tunç knows [that Hillary won]’ However, the justification requirement as a distinguishing criterion between –DIK and diye clauses runs into the same problem as the truth requirement: it does not hold across all data. 13 Another way to show the justification requirement by diye clauses and lack thereof in –DIK clauses, as suggested to me by Maria Luisa Zubizarreta, is to use them with abstract concepts like God’s existence. As the example below shows, only –DIK possible in such clauses. This also shows that diye requires tangible evidence (i.e., it cannot be (gut) feelings or abstract logic). i. Semra [Allah-in var ol-duğ-un]-u bil-iyor S A-3GEN exist COP-DIK-3S-ACC know-IMPF ‘Semra knows [that God exists]’ ii. #Semra [Allah var diye] bil-iyor S A exist diye know-IMPF ‘Semra knows [diye God exists]’ 36 When diye is embedded under a non-factive verb such as düşün- ‘think’, no justification requirement is required (Özyıldız 2016, 2017): (7) Context 1: ACCEPTABLE (for (a) and (b)) Tunç is cut off from getting any information about the world news but is obsessed by Hillary and accepts no state of the world where she hasn’t won. He says to the speaker: “Hillary won”. Hillary indeed won. Context 2: ACCEPTABLE (for (a) and (b)) Tunç is cut off from getting any information about the world news but is obsessed by Hillary and accepts no state of the world where she hasn’t won. In addition, the guards celebrate a Hillary victory outside his door. He says to the speaker: “Hillary won”. Hillary indeed won. a. Tunç da [Hillary-nin kazan-dığ-ın]-ı düşün-üyor T too H-3GEN win-DIK-3S-ACC think-IMPF ‘Tunç also thinks [that Hillary won]’ b. Tunç da [Hillary kazan-dı diye] düşün-üyor T too H win-PST diye think-IMPF ‘Tunç thinks [diye Hillary won]’ The justification requirement can also be overridden in some contexts, where the lack of justification has been established in the discourse as in (8a), with diye clauses under bil- ‘diye’, showing that it is not an inherent requirement by diye clauses: (8) Context: A and B are somehow convinced that Ali ran away, without evidence: a. A: Ali’nin kaçtığına dair kanıt yok, biliyorsun değil mi? There is no evidence that Ali ran away; you know that, right? 37 B: Evet biliyorum. Yes, I do. A: Başka kim öyle biliyor (i.e., öyle=Ali kaçtı)? Who else knows it that way (i.e., that way=Ali ran away)? B: Pelin de [Ali kaç-tı diye] bil-iyor P too A escape-PST diye know-IMPF ‘Pelin also knows [diye Ali ran away]’ b. A: Ali’nin kaçtığına dair kanıt yok, biliyorsun değil mi? There is no evidence that Ali ran away; you know that, right? B: Evet biliyorum. Yes, I do. A: Başka kim öyle biliyor (i.e., öyle=Ali kaçtı)? Who else knows it that way (that way=Ali ran away)? B: Pelin de [Ali-nin kaç-tığ-ın]-ı bil-iyor P too A-3GEN escape-DIK-3S-ACC know-IMPF ‘Pelin also knows [that Ali ran away]’ All in all, truth and justification are not enough to make a clear distinction between –DIK and diye clauses. We need to account for why we observe a difference between the two clauses as in example (2), when they are embedded under factivity-underspecified (perspective- underspecified, in my terms) verbs such as bil- ‘know’ as well as why the distinctions get lost under non-factive (i.e., matrix subject-oriented) verbs. But we have to do this in a way that does not resort to truth or justification only (but maybe a concept that includes them). This chapter will focus on the cases where the difference is observed (the pattern with non-factive verbs is discussed in Chapter 4). To achieve that, I will turn to another notion that can better explain the distinction between –DIK and diye clauses in that it brings out the inherent requirements of diye clauses and in that it explains a much wider range of data. That relevant notion, I argue, is 38 perspective, specifically attitudinal perspective. Before I start with the role of perspective in complement clauses in Turkish, however, I discuss some baseline data, where the matrix subject is established as an attitude holder in both –DIK and diye clauses. The aim is to establish a background for later sections in terms of attitude holders that are perspective centers. 2.2. Lowest Matrix Subject as Attitude Holder In this section, the lowest matrix subject will be established as an attitude holder in both –DIK and diye clauses. This might be considered as an unnecessary step, given that we are concerned with embedding attitude verbs and the environment they create is an attitude environment. The lowest matrix subject is the default attitude holder in those environments. However, the contrast between the lowest matrix subject and higher ones will be important in the analysis, especially in determining which attitude holders are also perspective centers. Therefore, the conclusion from this section serves as a useful baseline for this comparison. I clarify the notion of perspective center in detail later. However, let me note that I assume that the referent of a DP can be an attitude holder for a certain content without necessarily being a perspective center for it (but the perspective center for diye clauses must be an attitude holder). In other words, attitude holder and perspective centers are not the same thing in that not all attitude holders are perspective centers, but all attitude holders are potential perspective centers. To make sure that we are dealing with an attitude holder, I first establish whether the lowest matrix subject can be an attitude holder for the embedded proposition in – DIK and diye clauses. This will determine which one(s) of the attitude holders are also perspective centers. This approach will reveal the limitations of the notion of perspective centers. To determine whether the matrix subject is an attitude holder in a given clause, I use two tests 39 following Charnavel (2019, 2020): (i) co-referentiality with epithets and (ii) double-orientation, each of which is discussed in the next two sub-sections. 2.2.1. Co-referentiality with epithets Epithets are useful in determining the attitude holder(s) since they cannot be co-indexed with the c-commanding matrix subject of an attitude verb. They are anti-attitudinal (Dubinsky & Hamilton 1998). An example is given in (9a), where the idiot cannot refer to the matrix subject of an attitude verb. Yet the ban is not absolute in that co-referentiality works if the embedding verb is not attitudinal (9b) ((9b) from Haïk 1984: fn. 21, attributed to Haj Ross): (9) a. *[Mary]i thought that [the idiot]i passed the exam. b. [My brother]i invests in many projects that [the idiot]i thinks will make him rich. Another example from Dubinsky & Hamilton (1998) is in (10), where the epithet can be co- referential with the matrix subject only if the embedding verb is not an attitude verb (10b). This example shows that the choice of the embedding verb matters because “an epithet must not be anteceded by an individual from whose perspective the attributive content of the epithet is evaluated (Dubinsky & Hamilton 1998: 689)”. Based on this explanation, the embedding verb in (10a) has John as its attitude holder but the one in (10b) does not. 14 14 Notice that Dubinsky & Hamilton do not distinguish between attitude holders and perspective centers in a way I will adopt here. However, the epithet test is used to determine attitude contexts and attitude holders. 40 (10) a. #[John]i told us of a man who was trying to give [the idiot]i directions. 15 b. [John]i ran over a man who was trying to give [the idiot]i directions. We conclude that epithets cannot co-refer with the attitude holder matrix subject. How does this apply to the Turkish data? First, it should be kept in mind that the structures under discussion already involve an embedding attitude verb, so they are necessarily attitude contexts. The question is who is the relevant attitude holder(s) in these attitude contexts. The hypothesis, based on the above discussion, is that if an epithet is inside a –DIK or a diye clause, it should not be co- referential with the matrix subject of the embedding attitude verb; the matrix subject must be an attitude holder. This hypothesis is confirmed for both –DIK and diye clauses, as shown in (11) and (12) for subject and object epithets, respectively. Note that, these examples do not have a reading where the epithet is part of the commentary by the speaker and is still coreferential with the matrix subject (this chapter, Footnote 15): (11) a. Esrai [o salağ *i/j-ın sınav-ı kazan-dığ-ı]-nı bil-iyor E that idiot-3GEN exam-ACC pass-DIK-3S-ACC know-IMPF ‘Esrai knows [that that idiot *i/j passed the exam]’ b. Esrai [o salak *i/j sınav-ı kazan-dı diye] bil-iyor E that idiot exam-ACC pass-PST diye know-IMPF ‘Esrai knows [diye that idiot *i/j passed the exam]’ 15 As pointed out by Roumyana Pancheva, one might argue that the unacceptability of (10a) is because the relative clause is interpreted under the scope of the embedding attitude verb. On a reading where the relative clause is the comment by the speaker, (10a) would be acceptable. Although this might be true for this specific case, I will argue that this counter-argument does not apply to Turkish case I will discuss, and therefore, the anti-attitudinality of epithets holds as it is described in this section. 41 (12) a. Esrai [Ali-nin o salağ *i/j-ı gör-düğ-ü]-nü bil-iyor E Ali-3GEN that idiot-ACC see-DIK-3S-ACC know-IMPF ‘Esrai knows [that Ali saw that idiot *i/j]’ b. Esrai [Ali o salağ *i/j-ı gör-dü diye] bil-iyor E A that idiot-3GEN see-PST diye know-IMPF ‘Esrai knows [diye that Ali saw that idiot *i/j]’ We conclude that epithets in diye and –DIK clauses cannot be co-referential with the matrix subject, indicating that the matrix subject is an attitude holder for these clauses. 16 2.2.2. Double orientation The second test, again from Charnavel (2019), involves investigation of orientation in attitude contexts, namely whether an evaluative expression in the embedded clause can be evaluated by the matrix subject in addition to the speaker (hence the name double orientation). This test is one step closer to identification of perspective centers. An evaluative expression reflects an individual’s subjective evaluation of an entity, event or a situation. Charnavel uses this test to determine optionally-attitude contexts such as reason clauses, as exemplified in (13) (Charnavel 2018a). In (13b), the evaluative adjective embarrassing can be evaluated by both the speaker and Liz, suggesting that (13b) is an optional attitude context: (13) a. Lizi left the party because a picture of heri was circulating. b. Liz left the party because [an embarrassing picture of heri was circulating]. 16 Epithets will be discussed again in Chapter 3, this time with a slightly different property. 42 Since we are dealing with an attitude context headed by an attitude verb, we need not question whether the data under discussion is an optional or obligatory attitude context. It is obligatory. Therefore, I hypothesize that the matrix subject must be an attitude holder for a –DIK or diye clause if the matrix subject is an evaluator for an evaluative expression contained in their complements. This is borne out. The matrix subject is the only evaluator in diye clauses and at least one of the evaluators in –DIK clauses, along with the speaker. We return later to a more thorough discussion. Ignoring the speaker’s role in –DIK clauses for now, the important point shown in (14) is that the lowest matrix subject is an evaluator for both clause types: 17 (14) Context: Ali believes the topic Esra talked about was a good one. a. Ali [Esra-nın güzel bir konu-yu anlat-tığ-ın]- ı bil-iyor A E-3GEN nice one topic-ACC tell-DIK-3S-ACC know-IMPF ‘Ali knows [that Esra explained a nice topic]’ b. Ali [Esra güzel bir konu-yu anlat-tı diye] bil-iyor A E nice one topic-ACC tell-PST diye know-IMPF ‘Ali knows [that Esra explained a nice topic]’ Based on the anti-attitudinality of epithets and the double orientation test, I conclude that the lowest matrix subject is an attitude holder in both –DIK and diye clauses. We will now determine how attitude holder-ship relates to being a perspective center in –DIK and diye clauses and how these attitude holders are treated with respect to the speaker, who is also the default attitude holder in matrix clauses. 17 Note that these examples do not contrast the speaker’s and matrix subject’s evaluations. I will discuss contrasting perspectives in Section 2.3.1. Also, the speaker and Ali must agree in evaluation in –DIK clauses, a property I discuss in Section 2.3.4. in this chapter. 43 2.3. Perspective In this section, I discuss different perspectival elements and various interpretational contrasts in – DIK and diye clauses. The main goal is to establish whether these elements can be evaluated by the speaker, the matrix subject or both. Answering this question helps determine whether the attitude holder is also the perspective center in these clauses. Perspective is defined as the viewpoint from which the content of an evaluative expression is presented (Harris 2009). Perspective holder is the individual who evaluates this content. I adopt the definition of evaluativity in Reschke & Anand (2011: 370). For them, “Evaluativity is concerned with determining private states (e.g., judgment or emotion) that a particular evaluator bears towards a target entity, event or proposition”. This definition indicates that evaluativity is observed in the content of various levels of syntactic structures (e.g., DP- entity, VP-event, TP-proposition), an idea that will prove to be useful for the current proposal. I, however, extend this definition to adjunct content as well: a perspective holder can have private states towards elements that indicate modification for an event (content of AdvP) or property of an entity (i.e., content of AP) distinct from the event or the entity itself (see Section 2.3.1.4.). In this section, I focus on three types of non-propositional adjunct content that can be evaluated by an individual: evaluative expressions such as adjectives, temporal and locative expressions, DP and VP content (entity and event, respectively), followed by a combination of these, each in a separate section, as well as the requirement for conscious evaluation. The main conclusion from these data is that the lowest matrix subject is the only perspective center in diye clauses, forcing all elements to be evaluated from the matrix subject’s perspective. Section 2.3.1. focuses on single embedding structures and Section 2.3.2. on multiple embedding structures with 44 the same type of evaluative expressions. Section 2.3.3. will discuss de re- de dicto interpretation in –DIK and diye clauses. Section 2.3.4. summarizes the data and generalizations and discusses the structure of –DIK clauses. 2.3.1. Single embedding structures 18 In this section, I present data in contexts where the speaker and the matrix subject have contrasting perspectives with respect to certain content within the embedded clause. The contrast is used to make clear whose perspective is being utilized in which structure. This will highlight a clear and consistent contrast between diye and –DIK clauses. 2.3.1.1. Evaluative expressions The manner adverbs iyi ‘good’ and kötü ‘bad’ in the embedded clause in (15) describe a judgement with respect to the way the embedded action was conducted. İyi is an accurate description if the evaluator is the matrix subject and kötü is accurate if the evaluator is the speaker. –DIK clauses allow neither adverb (15a). Diye clauses allow evaluation only by the matrix subject and not the speaker (15b): (15) Context: I know John competed bad. Ali believes, however, that John competed well. a. Ali [John-un {#iyi/#kötü} yarış-tığ-ın]-ı bil-iyor A J-3GEN good/bad compete-DIK-3S-ACC know-IMPF 18 The data in this section are similar to those in Section 2.2.2. (Double Orientation Data) except that I now test the matrix subject as well as the speaker as evaluators. 45 ‘Ali knows [that John competed #well/#bad]’ b. Ali [John {iyi/#kötü} yarış-tı diye] bil-iyor A J good/bad compete-PST diye know-IMPF ‘Ali knows [diye John competed well/#bad]’ We observe the same pattern when these same forms are used as adjectives. In (16), iyi ‘good’ and kötü ‘bad’ are used to describe the individual (woman) the embedded subject refers to. The said woman is kötü ‘bad’ from the speaker’s perspective, whereas the matrix subject evaluates the same woman as iyi ‘good’, resulting in contrasting evaluations. (16b) shows that diye only accepts the matrix subject’s evaluation. –DIK clauses accept neither, as shown in (16a): (16) Context: I heard a bad woman came from London today. Ali was told of that same woman that she is a nice woman. We each believe what we have been told. a. Ali [{#iyi/#kötü} bir kadın-ın Londra-dan gel-diğ-in]-i A good/bad one woman-3GEN London-ABL come-DIK-3S-ACC bil-iyor know-IMPF ‘Ali knows [that a #good/#bad woman came from London]’ b. Ali [{iyi/#kötü} bir kadın Londra-dan gel-di diye] bil-iyor A good/bad one woman London-ABL come-PST diye know-IMPF ‘Ali knows [diye a good/#bad woman came from London]’ 46 2.3.1.2. Temporal expressions This section considers evaluation of phrasal temporal expressions. The context in (17) presents a contrast of how the speaker and the matrix subject judge the time of the embedded event. The matrix subject in diye clauses is the only possible evaluator for the temporal expression (17b) whereas –DIK clauses do not allow evaluation by either the speaker or the matrix subject (17a): (17) Context: I know that Ali had a race between 2-3 pm. John thought his race was between 5-6 pm. Ali and John are both in Los Angeles. a. John [Ali-nin {#5-6 arası/#2-3 arası} yarış-tığ-ın]-ı bil-iyor J A-3GEN 5-6 between/2-3 between compete-DIK-3S-ACC know-IMPF ‘John knows [that Ali competed #between 5-6/#between 2-3]’ b. John [Ali {5-6 arası/#2-3 arası} yarış-tı diye] bil-iyor J A 5-6 between/2-3 between compete-PST diye know-IMPF ‘John knows [diye Ali competed between 5-6/#between 2-3]’ 2.3.1.3. Locative expressions When the speaker and the matrix subject have different perspectives on the location of an event, diye clauses are compatible only with the expression of matrix subject’s perspective (18b) whereas –DIK clauses are compatible with neither (18a): (18) Context: I know that Ali had a race in Los Angeles. John thought his race was in New York. 47 a. John [Ali-nin {#New York-ta /#Los Angeles-ta} yarış-tığ-ın]-ı J A-3GEN N Y-LOC L A-LOC compete-DIK-3S-ACC bil-iyor know-IMPF ‘John knows [that Ali competed #in New York/#in Los Angeles]’ b. John [Ali {New York-ta/#Los Angeles-ta} yarış-tı diye] bil-iyor J A N Y-LOC/L A-LOC compete-PST diye know-IMPF ‘John knows [diye Ali competed in New York/#in Los Angeles]’ We have seen so far that evaluative expressions with adjunct content need to be reported from the perspective of the matrix subject in diye clauses while –DIK clauses allow evaluation by neither the matrix subject nor the speaker if there are contradictory evaluations. The data are summarized in Table 1. I represent the evaluation in –DIK clauses as S & MS (Speaker & Matrix Subject) though the examples in this section pointed out that neither perspective can be the sole perspective. This representation is chosen because both the speaker and matrix subject can agree on the evaluation, as in example (19). I discuss this in more detail later in this chapter. Table 1: Who evaluates embedded expressions (S: Speaker; MS: Matrix Subject) Evaluative Modifiers Temporal Expressions Locative Expressions -DIK diye -DIK diye -DIK diye S&MS MS S&MS MS S&MS MS 48 (19) Context: Ali believes the topic Esra talked about was a good one. The speaker agrees. Ali [Esra-nın güzel bir konu-yu anlat-tığ-ın]-ı bil-iyor A E-3GEN nice one topic-ACC tell-DIK-3S-ACC know-IMPF ‘Ali knows [that Esra explained a nice topic]’ 2.3.1.4. Entities and events Examples (20) and (21) present contrasting evaluations for nominal and verbal content, respectively, and show that diye clauses align with the matrix subject’s perspective and –DIK clauses align with neither the matrix subject’s nor the speaker’s perspective only. This corroborates the previous conclusion regarding whose perspective is relevant in these clauses: (20) Context: Ali and I were told that someone nice came from London today. I was told that this person was a man and Ali was told that the person was a woman. We each believe what we have been told. a. Ali [iyi bir {#kadın-ın/#adam-ın} Londra-dan gel-diğ-in]-i A good one woman-3GEN/man-3GEN London-ABL come-DIK-3S-ACC bil-iyor know-IMPF ‘Ali knows [that a good #woman/#man came from London]’ b. Ali [iyi bir {kadın/#adam} Londra-dan gel-di diye] bil-iyor A good one woman/man London-ABL come-PST diye know-IMPF ‘Ali knows [that a good woman/#man came from London]’ 49 (21) Context: I know Ali had a running competition between 2-3 pm. John knew this but he thought it was a walking competition. Ali and John are both in Los Angeles. a. John [Ali-nin 2-3 arası {#yürü-düğ-ün-ü/#koş-tuğ-un-u} bil-iyor J A-3GEN 2-3 between walk-DIK-3S-ACC/run-DIK-3S-ACC know-IMPF ‘John knows [that Ali #walked/#ran between 2-3]’ b. John [Ali 2-3 arası {yürü-dü/#koş-tu} diye] bil-iyor J A 2-3 between walk-PST/run-PST diye know-IMPF ‘John knows [diye Ali walked/#ran between 2-3]’ 2.3.1.5. Combinations The data in the previous sections showed that the matrix subject in diye clauses is the only evaluator of the embedded expressions (i.e., the speaker cannot be an evaluator) whereas neither the speaker nor the matrix subject can be the sole evaluator in –DIK clauses in cases of contrastive perspectives (one can argue they must both be evaluators, but I eliminate this option later in this chapter). This consistent pattern does not change when we combine multiple evaluative elements in a single clause. All such elements must be reported from the perspective of the matrix subject in diye clauses and they cannot be reported using a –DIK clause if the speaker’s and the matrix subject’s perspectives clash. As an example, see (22), where a temporal expression, a locative expression and an evaluative adverb are used in the same embedded clause. If they occur in a diye clause, they must reflect the matrix subject’s perspective (22b). If they occur in a –DIK clause, none of the expressions are acceptable (22a) because the speaker and the matrix subject differ where, when and how they believed the race happened: 50 (22) Context: I know Ali had a race between 2-3 in Los Angeles. John thought his race was between 5-6 in New York. Ali and John are both in Los Angeles. John also thought Ali competed well and the speaker thinks Ali was bad. a. John [Ali-nin {#5-6 arası/#2-3 arası} {#New York-ta/#Los Angeles-ta} J A-3GEN 5-6 between/2-3 between N Y-LOC/L A-LOC {#iyi/#kötü} yarış-tığ-ın]-ı bil-iyor good/bad compete-DIK-3S-ACC know-IMPF ‘John knows [that Ali competed {#well/#bad}{#between 2-3/#between 5-6}{#in New York/#in Los Angeles}]’ b. John [Ali {5-6 arası/#2-3 arası} {New York-ta/#Los Angeles-ta} J A 5-6 between/2-3 between N Y-LOC/L A-LOC {iyi/#kötü} yarış-tı diye] bil-iyor well/bad compete-PST diye know-IMPF ‘John knows [diye Ali competed {well/#bad}{between 5-6/#between 2-3}{in New York/#in Los Angeles}]’ 2.3.1.6. Interim conclusion The data so far are summarized in Table 2. As the table shows, any phrasal expression in diye clauses must be evaluated only by the matrix subject (and not the speaker). –DIK clauses must express the combined perspective of the speaker and the matrix subject (i.e., the speaker and the matrix subject must agree). 51 Table 2: Who evaluates embedded expressions, all combined: Evaluative Modifiers Temporal Expressions Locative Expressions Entities & Events Combinations -DIK diye -DIK diye -DIK diye -DIK diye -DIK diye S&MS MS S&MS MS S&MS MS S&MS MS S&MS MS 2.3.2. Multiple embedding This section shows that the higher matrix subject is an attitude holder for the most embedded clause (Section 2.3.2.1.), but not a perspective center for it (Section 2.3.2.2.). 2.3.2.1. Higher matrix subject as an attitude holder Seda, the higher matrix subject, in (23) cannot be co-referential with the epithet in the most embedded clause (in a similar way to the lowest matrix subject), providing evidence that it is a potential attitude holder for both –DIK and diye clauses: (23) a. Sedai [Alij-nin [Esrak-nın {o salağ *i/*j/*k/m}-ı gör-düğ-ün]-ü S A-3GEN E-3GEN that idiot-ACC see-DIK-3S-ACC bil-diğ-in]-i bil-iyor know-DIK-3S-ACC know-IMPF ‘Seda knows [that Ali knows [that Esra saw that idiot *i/*j/*k/m]]’ 52 b. Sedai [Alij [Esrak {o salağ *i/*j/*k/m }-ı gör-dü diye] S A E that idiot-ACC see-PST diye bil-iyor diye] bil-iyor know-IMPF diye know-IMPF ‘Seda knows [diye Ali knows [diye Esra saw that idiot *i/*j/*k/m]]’ When no explicit mention of conflicting perspectives is made in the context (hence the natural assumption is to assume agreement), 19 the higher matrix subject can be an evaluator for the evaluative adjective in the most embedded clause regardless of the complement type (24): (24) Context: Seda believes that the topic Esra talked about is nice. a. Seda [Ali-nin [Esra-nın güzel bir konu-yu anlat-tığ-ın]-ı S A-3GEN E-3GEN nice one topic-ACC tell-DIK-3S-ACC bil-diğ-in]-i bil-iyor know-DIK-3S-ACC know-IMPF ‘Seda knows [that Ali knows [that Esra explained a nice topic]]’ b. Seda [Ali [Esra güzel bir konu-yu anlat-tı diye] S A E nice one topic-ACC tell-PST diye bil-iyor diye] bil-iyor know-IMPF diye know-IMPF 19 I discuss what happens when explicit mention of conflicting perspectives is made in upcoming sections. The main point here is to show that the higher matrix subject can be an evaluator for the most embedded clause. 53 ‘Seda knows [diye Ali knows [diye Esra explained a nice topic]]’ This establishes the higher matrix subject as an attitude holder for the most embedded clause. Is she also the perspective center for it? The answer is no, as the next sections show. 20 2.3.2.2.Higher matrix subject as impossible perspective center In this section, I discuss evaluation of modifiers in multiple embedding structures. The evaluation contrast will be between the higher matrix subject versus the speaker and the intermediate embedded subject. In other words, I isolate the higher matrix subject perspectivally to understand whether it can be the sole evaluator in –DIK and diye clauses. What is crucial is that the higher matrix subject(s) differs perspectivally from the lowest matrix subject. We will confirm this to be the case. To achieve that, I utilize conflicting evaluations for the said elements, in contrast to the way the context in (24) was set up to highlight agreement in evaluations. I keep types of the complement clauses the same within a multiply-embedded clause. 2.3.2.2.1. Evaluative expressions We can change the context in (24) in a way that highlights a perspectival conflict such that the higher matrix subject is the only individual with a different evaluation in the context. Seda’s evaluation differs from that of the speaker’s and the lower matrix subject’s (Ali). In such cases, diye clauses cannot express the evaluation by the higher matrix subject (25b): 20 Only evaluative adjectives, temporal and locative expressions are discussed here but nominal and verbal content are similar. 54 (25) Context: Seda believes that the topic Esra talked about is nice. The speaker and Ali are harsher critics and think that the topic was bad (aka, uninteresting). Seda is aware of the different evaluations of hers versus the speaker’s and Ali’s. a. Seda [Ali-nin [Esra-nın {#güzel/#kötü} bir konu-yu S A-3GEN E-3GEN nice/bad one topic-ACC anlat-tığ-ın]-ı bil-diğ-in]-i bil-iyor tell-DIK-3S-ACC know-DIK-3S-ACC know-IMPF ‘Seda knows [that Ali knows [that Esra explained a #nice/#bad topic]]’ b. Seda [Ali [Esra {#güzel/kötü} bir konu-yu anlat-tı diye S A E nice/bad one topic-ACC tell-PST diye bil-iyor diye] bil-iyor know-IMPF diye know-IMPF ‘Seda knows [diye Ali knows [diye Esra explained a #nice/bad topic]]’ This is the first piece of evidence that the higher matrix subject is not a perspective center for the most embedded clause. Only the intermediate one is a potential perspective center for the most embedded diye clauses. As for –DIK clauses, these do not allow for any conflicting perspectives under any combination of attitude holders, be it between the speaker and a single matrix subject or the speaker and different matrix subjects. 55 2.3.2.2.2. Temporal expressions Diye clauses cannot express higher matrix subject’s evaluation (26b) when s/he evaluates for the embedded temporal expression differently. –DIK clauses do not accept either expression (26a): (26) Context: The speaker and John talked to Alii, who told them that hei had a race between 2-3 pm. Murat mistakenly thought the race was between 5-6 pm. Murat is aware of the different evaluations between him versus the speaker and John. They are all in Los Angeles. a. Murat [John-un [Ali-nin {#5-6 arası/#2-3 arası} yarış-tığ-ın]-ı M J-3GEN A-3GEN 5-6 between/2-3 between compete-DIK-3S-ACC bil-diğ-in]-i bil-iyor know-DIK-3S-ACC know-IMPF ‘Murat knows [John knows [that Ali competed #between 2-3/#between 5-6]]’ b. Murat [John [Ali {#5-6 arası/2-3 arası} yarış-tı diye] M J A 5-6 between/2-3 between compete-PST diye bil-iyor diye] bil-iyor know-IMPF diye know-IMPF ‘Murat knows [John knows [that Ali competed #between 5-6/between 2-3]]’ 56 2.3.2.2.3. Locative expressions We observe that the higher subject cannot be the sole evaluator for the location expression in the embedded diye clause (27b). When there is a conflict of evaluation, neither the lower nor the higher subject can evaluate the locative expression in –DIK clauses (27a): (27) Context: The speaker and John talked to Ali, who told them that Ali had a race in Los Angeles. Murat mistakenly thought the race was in New York. Murat is aware of the difference in evaluations between him versus the speaker and John. a. Murat [John-un [Ali-nin {#Los Angeles-ta/#New York-ta} M J-3GEN A-3GEN L A-LOC N Y-LOC yarış-tığ-ın]-ı bil-diğ-in]-i bil-iyor compete-DIK-3S-ACC know-DIK-3S-ACC know-IMPF ‘Murat knows [John knows [that Ali competed #in Los Angeles/#in New York]]’ b. Murat [John [Ali {Los Angeles-ta/#New York-ta} yarış-tı M J A L A-LOC/N Y-LOC compete-PST diye] bil-iyor diye] bil-iyor diye know-IMPF diye know-IMPF ‘Murat knows [John knows [diye Ali competed in Los Angeles/#in New York]]’ 57 This brings us to our generalization: 21 (28) Generalization: Diye clauses encode a perspective center (PC) position, forcing all phrasal content in the clause to be evaluated from the point of view of the attitude holder occupying the PC position; this constitutes a single perspective domain. Only the closest matrix subject can fulfil the role of perspective center. (–DIK clauses do not encode any perspective 22 ). Further evidence for diye clauses being a single perspective domain comes from the ban on de re readings, which I discuss in the next section. 2.3.3. De re –De Dicto distinction in –DIK and diye clauses The de re-de dicto distinction is important in understanding attitude contexts. I stay neutral as to how de re readings can be derived and use different types of tests used to support different analyses of de re-de dicto in the literature (See Keshet 2008 and Keshet & Schwarz 2019 for a critical review). I take de re readings to be associated with the utterance context (i.e., not the attitude context). My goal here is to establish whether de re readings, however they are analyzed, are allowed in –DIK and diye clauses at all. If they are not, I take this as evidence that de dicto interpretation is enforced in them. This, in turn, supports diye clauses being single perspective domains. With that being said, let us now start the discussion with the two sentences in (29) (Keshet 2008), both of which have a de re and a de dicto reading: 21 I do not discuss data contrasting the speaker’s and the closest matrix subject’s perspectives as this type of data was already discussed in the previous section. 22 See Section 2.3.5. for a more detailed discussion of lack of perspective in –DIK clauses and why I opt for no perspective rather than multiple perspectives. 58 (29) a. An hour ago, someone in this room was outside. b. I introduced Mary to her husband. In the de re reading in (29a), someone in this room is evaluated with respect to the actual world now whereas the rest of the sentence is evaluated with respect to the actual world an hour ago (=[Someone in this room now] was outside an hour ago). The de dicto reading leads to a reading not allowed by our world knowledge, namely that someone was both in this room and outside at the same time. Likewise, the de re reading in (29b) goes as follows: ‘I introduced Mary in some past time to [the man who is her husband now]’. De dicto reading of (29b) is again not compatible with our world knowledge, unless Mary married her husband without meeting him before marriage. The common point in the de re interpretation is that it allows multiple evaluation contexts, more specifically that the part of the sentence read de re is interpreted in a different evaluation context (in the above examples: in the now) than other parts of the sentence. This is the hallmark of de re readings. The example in (29) is an example with different temporal specifications. De re-de dicto distinction is not limited to temporality, however. We can attain the distinction via different worlds using attitude verbs (since these involve attitude verbs, they are closer to the present data) and we still arrive at the same conclusion: de re interpretation entails valuation in different worlds while de dicto readings do not. In (30a), someone in this room is evaluated in the actual world whereas the rest of the sentence is evaluated in Mary’s thought worlds. This is the de re reading of someone. The de dicto reading attributes contradiction to Mary such that she thinks someone is in this room and outside at the same time. In the de re reading in (30b), her husband is evaluated in the actual world whereas the rest of the sentence in Mary’s thought worlds. In the de dicto reading, Mary must have married her husband knowing he is a burglar. 59 (30) a. Mary thinks someone in this room is outside. b. Mary thought her husband was a burglar. One major way to test for the availability of de re readings is substitution. The idea is as follows: When an element is substituted with a co-referring term, no unacceptability should arise if de re reading is available in a given clause. If the substitution leads to unacceptability, this means that the given clause disallows de re interpretation. A famous example is given in (31), from Quine (1956), where the relevant co-referring terms are Cicero and Tullus: (31) a. Tom believes that Cicero denounced Cataline. b. Tom believes that Tullus denounced Cataline. If Tom does not know that Cicero and Tullus are the same person and he knows only that Cicero denounced Cataline, we cannot conclude (31b) from (31a). (31b) can only be read de re, since Tom is not the evaluator for Tullus but the speaker. In other words, Tullus is evaluated in the speaker’s thought-worlds while the rest of the sentence is evaluated in Tom’s thought-worlds. This indicates the availability of multiple evaluation contexts. (31a) is read de dicto because all parts of the embedded proposition are evaluated by Tom, hence a single evaluation domain. There is another way to tap into the de re-de dicto distinction: the scope of scope-taking elements (assuming the Scope Theory for the time being). See the example in (32), also discussed by Quine (1956). The scope-taking element here is the embedded subject someone. (32) can have the two readings given in (33), where (33a) is the de dicto reading and (33b) is the de re reading. When I use this test, I refer to the higher scope reading as the specific reading and 60 the lower scope reading as the unspecific reading to stay neutral as to potential analyses of such readings (see Keshet 2008 for a review): (32) Ortcutt believes that someone is a spy. (33) a. Ortcutt believes that there are spies. b. Someone is an x such that Ortcutt believes that x is a spy. We now turn to the Turkish data. The first two tests involve substitution. The first one involves two names with the same reference in the actual world, Two Names, One Person, replicating (31). This test involves two names with the same reference in the actual world. The second test is The President test, which is similar to the first test in design, with the difference that it involves a description and a name referring to the same individual. The third one is the Someone test, which tests whether someone in the embedded clause must be interpreted as wide or narrow in scope (i.e., de re-de dicto, respectively), similar to (32)-(33). 2.3.3.1. Two names, one person The context in (34) specifies Murat Ali (embedded subject) to have two names. The speaker and the matrix subject both know the referent (i.e., embedded subject) as an individual in the real world but the speaker knows both names of the referent while the matrix subject knows only one (i.e., Murat). When both the speaker and the matrix subject know (the referent of) the name, both embedded clause types are acceptable (34a-b). When there is divergence, however, diye clauses cannot contain an element that the matrix subject does not know about (34d), although the matrix 61 subject knows the referent of that name to have passed the exam (but as Murat, not as Ali). –DIK clauses do not impose any restrictions in this aspect (34b): 23 (34) Context: Murat has two names, Murat and Ali. John knows him only as Murat (i.e., he does not know that Murat has another name). Speaker knows both names (i.e., knows him as Murat Ali) but always refers to him as Ali. Murat Ali passed his exam. 24 a. John [Murat-ın sınav-ı geç-tiğ-in]-i bil-iyor aslında J M-3GEN exam-ACC pass-DIK-3S-ACC know-IMPF actually ‘John knows [that Murat passed the exam] actually’ b. John [Murat sınav-ı geç-ti diye] bil-iyor aslında J M exam-ACC pass-PST diye know-IMPF actually ‘John knows [diye Murat passed the exam] actually’ c. John [Ali-nin sınav-ı geç-tiğ-in]-i bil-iyor aslında J A-3GEN exam-ACC pass-DIK-3S-ACC know-IMPF actually ‘John knows [that Ali passed the exam] actually’ d. #John [Ali sınav-ı geç-ti diye] bil-iyor aslında J A exam-ACC pass-PST diye know-IMPF actually ‘John knows [diye Ali passed the exam] actually’ 23 Notice that this flexibility in –DIK clauses is different from the evaluative, temporal and locative expressions discussed earlier in that these expressions required agreed perspective of the speaker and the matrix subject (and thus conflicting perspectives could not be reported). In (34), the speaker can report on something that the matrix subject may not (consciously) know. Although I do not have a definitive answer to where that difference comes from, one might propose the following: In (34), the matrix subject knows the referent of the subject. So, in fact, the speaker and the matrix subject share the same core propositional knowledge, which in this case is that the person whose referent in the real world is Murat Ali passed the exam. This means there is no actual conflict here, which is different from evaluative expressions. 24 In the sentences in this example and some others below, aslında ‘actually’ is used in both –DIK and diye clauses. This is because otherwise, these types of sentences are not the most natural way of expressing the situation and it was difficult for my informants to judge. What is important to note here is that aslında ‘actually’ is used in both – DIK and diye clauses and it improves –DIK clauses but not diye clauses. 62 The conclusion from this test is that elements in the embedded proposition must be evaluated by the matrix subject only in diye clauses. This suggests that there is a single perspective domain. 2.3.3.2. The president The context in (35) is similar to the previous test except that instead of two names, there is a description and a name associated with a single person. The speaker knows both whereas the matrix subject knows only the name. As before, when the embedded subject is part of both the speaker’s and matrix subject’s knowledge state, both –DIK and diye clauses are acceptable (35a- b). When part of the speaker’s but not matrix subject’s knowledge state is used in the embedded clause, only –DIK is acceptable (35c) while diye is unacceptable (35d): (35) Context: Murat is the president of a student club and he visited the school my roommate John and I work at. John knows Murat by name (perhaps as a friend) but he does not know that Murat is the president. John also knows Murat visited the school. a. John [Murat-ın okul-a gel-diğ-in]-i bil-iyor aslında J M-3GEN school-DAT come-DIK-3S-ACC know-IMPF actually ‘John knows [that Murat came to the school] actually’ b. John [Murat okul-a gel-di diye] bil-iyor aslında J M school-DAT come-PST diye know-IMPF actually ‘John knows [diye Murat came to the school] actually’ c. John [başkan-ın okul-a gel-diğ-in]-i bil-iyor aslında J president-3GEN school-DAT come-DIK-3S-ACC know-IMPF actually ‘John knows [that the president came to the school] actually’ 63 d. #John [başkan okul-a gel-di diye] bil-iyor aslında J president school-DAT come-PST diye know-IMPF actually ‘John knows [diye the president came to the school] actually’ 2.3.3.3. Someone (36a) shows that someone in a –DIK clause can be interpreted both as de re and de dicto whereas diye enforces the de dicto interpretation (36b). The de re reading here is the specific reading of someone and the de dicto reading is unspecific. 25 (36) a. Pelin [burada biri-nin hasta ol-duğ-un]-u bil-iyor P here someone-3GEN sick be-DIK-3SACC know-IMPF ‘Pelin knows [that someone is sick here]’ (specific) ‘Pelin knows [that there is someone sick here]’ (unspecific) b. Pelin [burada biri hasta diye] bil-iyor P here someone sick diye know-IMPF ‘#Pelin knows [diye someone is sick here]’ (specific) ‘Pelin knows [diye there is someone sick here]’ (unspecific) The overall conclusion is that de re readings are not allowed in diye clauses, which provide further support that they are a single perspectival domain. 25 I leave the de dicto reading of elements that are specific for future research. 64 2.3.4. Summary and more on –DIK clauses Diye clauses require that all content in its scope be reported from the perspective of the intermediate matrix subject. The main generalization from the previous sections is in (37): (37) Generalization: Diye clauses encode a perspective center (PC) position, forcing all phrasal content in the clause to be evaluated from the point of view of the attitude holder occupying the PC position; this constitutes a single perspective domain. Only the closest matrix subject can fulfil the role of perspective center. (–DIK clauses do not encode any perspective). Before analyzing diye, I discuss –DIK clauses in detail here, as I do not discuss their internal structures any further unless they are directly relevant to a given discussion. I focus on the perspectival properties of -DIK clauses and argue that they do not encode perspective. More specifically, I argue that they are CP-nominalizations without a perspectival component. As was discussed in Chapter 1, –DIK clauses have a CP. For example, we have seen in Chapter 1 that they can host wh- words, CP-level elements, as shown in (38). See Chapter 1 for more evidence as well as Kornfilt (2003): (38) Semra [kim-in gel-diğ-in]-i bil-iyor S who-3GEN come-DIK-3S-ACC know-IMPF ‘Whoi does Semra know that s/hei came?’ ‘Semra knows who came’ 65 I take the syntactic structure of –DIK clauses to be as in (39). The nominalizer –DIK heads the DP embedding a CP, whose subject moves to SpecDP (from SpecnP) for genitive case assignment. This DP is selected by the embedding verb to saturate its internal argument position. I also assume that the CP in –DIK clauses is a defective one –recall from Chapter 1 that –DIK has limited tense-aspect marking and also has a defective TP-, resulting in it not being an opaque domain. 26 This will be discussed in more detail after I present the proposal: (39) What about the perspectival properties of –DIK clauses? (39) indicates that there is nothing that relates to perspective in these clauses. We have seen in the data so far that -DIK clauses are consistently unable to express conflicting evaluations. My informants commented that this unacceptability results because whatever perspectival element is chosen, we end up conflicting either with the speaker’s or the matrix subject’s perspective. We have also seen that when no conflicting evaluations between the speaker and the matrix subject) are indicated (hence, defaulting to agreement between them), -DIK clauses can express their agreed evaluation. This is exemplified in (40), repeated from (14a): 26 We will see evidence for defective CP in –DIK clauses in later chapters, when –DIK clauses cannot host certain CP-level elements (such as adverbs like iyi ki ‘fortunately’ while diye clauses can. 66 (40) Context: Ali believes the topic Esra talked about was a good one. Ali [Esra-nın güzel bir konu-yu anlat-tığ-ın]-ı bil-iyor A E-3GEN nice one topic-ACC tell-DIK-3S-ACC know-IMPF ‘Ali knows [that Esra explained a nice topic]’ This might be taken to mean either that -DIK clauses encode two (or more in case of multiple embeddings) perspectives or that they do not encode any. I argue for the latter for a few reasons. One is that having multiple perspective centers in a single clause (CP) is theoretically unexpected. Rather, one perspective center per CP is often assumed. 27 Koopman & Sportiche (1989), for example, argue that “At most one n-operator is allowed per Comp (p. 570)”, where n- operator refers to operators that interact with a set of logophoric pronouns in Abe. This has been assumed to be true in later studies as well (e.g., Huang & Liu 2001; Sundaresan 2012; Charnavel & Mateu 2015; Charnavel 2017; see Pancheva & Zubizarreta 2018 and Charnavel 2019 for examples of the assumption of a single perspective center per evaluation domain but where evaluation domain can be different than a full clause, i.e., CP). Another concerning point about having multiple perspectives per CP is that the multiplicity of these centers is potentially limitless. We have seen that the incompatibility arises when there is a conflict between the speaker and the matrix subject in these clauses in single embedding clauses. However, as the number of embeddings increases, there is required agreement between all matrix subjects and the speaker. (41) illustrates this point. Context 1 was used in (24) to show that one cannot isolate the higher matrix subject’s perspective and effectively use –DIK clauses under bil- ‘know’. Context 2 and 3 do the same for the lower matrix subject and the speaker, respectively. We have seen that when the context does not clearly state a perspective conflict, however, -DIK clauses are happily 27 See, for comparison, Charnavel (2018b), who argues that perspective centers can be plural in causal clauses. 67 available, as shown in (42), repeated from (25). The overall conclusion is that matrix subjects and the speaker need to agree on an evaluation for –DIK clauses to be used, which can mean an infinite number of perspective centers in –DIK clauses if one adopts a multiple perspective center approach. It must be clear to the reader that this is not a tenable and practical assumption: (41) Context 1 (higher matrix subject): Seda believes that the topic Esra talked about is nice. The speaker and Ali are harsher critics and think that the topic was bad (aka, uninteresting). Context 2 (lower matrix subject): Ali believes that the topic Esra talked about is nice. The speaker and Seda are harsher critics and think that the topic was bad (aka, uninteresting). Context 3 (speaker): Speaker believes that the topic Esra talked about is nice. Seda and Ali are harsher critics and think that the topic was bad (aka, uninteresting). Seda [Ali-nin [Esra-nın {#güzel/#kötü} bir konu-yu anlat-tığ-ın]-ı S A-3GEN E-3GEN nice/bad one topic-ACC tell-DIK-3S-ACC bil-diğ-in]-i bil-iyor know-DIK-3S-ACC know-IMPF ‘Seda knows [that Ali knows [that Esra explained a #nice/#bad topic]]’ (42) Context: Seda believes that the topic Esra talked about is nice (the speaker and Ali need not agree but can). Seda [Ali-nin [Esra-nın güzel bir konu-yu anlat-tığ-ın]-ı S A-3GEN E-3GEN nice one topic-ACC tell-DIK-3S-ACC 68 bil-diğ-in]-i bil-iyor know-DIK-3S-ACC know-IMPF ‘Seda knows [that Ali knows [that Esra explained a nice topic]]’ This was a negative piece of evidence against multiple perspectives in -DIK clauses. There is also positive evidence. –DIK clauses cannot host perspectival (CP-level) adverbs such as iyi ki ‘fortunately’ (43a) as opposed to diye clauses (43b) (see Chapter 4 for more discussion). In diye clauses, iyi ki ‘fortunately’ can be either associated with the matrix subject (i.e., ‘It is fortunate that Ali came’) or the speaker (i.e., ‘It is fortunate that Esra knows [diye Ali came]’). 28 Neither association is possible with -DIK clauses. (43) a.*Esra [Ali-nin iyi ki gel-diğ-in]-i bil-iyor E A-3GEN good that come-DIK-3S-ACC know-IMPF ‘Intended: #Esra knows [that Ali fortunately came]’ b. Esra [Ali iyi ki gel-di diye] bil-iyor 29 E A good that come-PST diye know-IMPF ‘Esra knows [diye Ali fortunately came]’ If there is no perspective center structurally encoded in –DIK clauses, where does the agreement requirement come from? Why can the world variables of these elements not be bound by the attitude verb? I do not give a definitive answer to this question, however, I will point out to a potential direction, based on Giorgi & Pianesi (2001). Giorgi & Pianesi (2001) discuss the agreed 28 This is not against diye being matrix-subject oriented, since the reading associated with the speaker is the one in which the adverb modifies the matrix verb but still appears to be in the diye clause. Presumably, there has been discourse-related movement of the embedded subject, as it appears to the left of the adverb in focus. 29 Imagine a context where the speaker and the hearer convince Esra that Ali’s arrival is good when it actually is not. 69 perspective of the speaker and the matrix subject in some attitude contexts. Though their analysis is based on temporal anchoring and its representation in the semantics of certain attitudinal verbs, their conclusion is that “if the subject-oriented assignment is such that it makes the subordinate clause true, so does the speaker-oriented assignment. On the other hand, if the clause is false under one assignment, then it is false under the other as well (p. 222)”. In other words, the subordinate clause must be true both in the matrix subject’s and the speaker’s temporal coordinates. The following examples (their (19)-(21)) exemplify this. (44) and (46) exemplify cases where there is a change since the time John said what he said, both resulting in unacceptability. This is because when the speaker utters his sentence, s/he ends up misrepresenting the situation. In (44), the speaker knows that Mary is no longer in the room while in (46), s/he knows that Mary was not in the room when John said what he said and that she entered later. The only acceptable case is (45), where Mary’s staying in the room holds both, or neither, in the speaker’s and the subject’s temporal coordinates with no change. This entails agreement between the matrix subject and the speaker. The agreement is represented as in (47) by Giorgi & Pianesi (2001, p. 222). e’ and u represent the subject’s and the speaker’s temporal coordinates, e the saying event and the double-sided arrow represent the requirement that the embedded situation needs to be true in both: 30 (44) Scenario: Mary is in the room. John: Mary is in the room. Scenario: Mary exits. Speaker: #John said that Mary is in the room. 30 Giorgi & Pianesi do not give an example of a scenario where Mary is and stays in the room in relevance to this discussion. However, since there is no change and thus no misinterpretation in such a case, my guess is that the sentence would be acceptable. 70 (45) Scenario: There is no one in the room. John: Mary is in the room. Speaker: John said that Mary is in the room. (46) Scenario: There is no one in the room. John: Mary is in the room. Scenario: Mary enters the room. Speaker: #John said that Mary is in the room. (47) ∀e[in-the-room (e) ^ theme (e, Maria)] (e o e’) (e o u) Let us see how this accounts for the requirement for agreed perspectives in –DIK clauses. Giorgi & Pianesi do not discuss perspective directly, so I only use their tool for conceptualization here. Let us assume that individuals (the speaker and the matrix subject) have their own associated temporal coordinates. The temporal coordinate of the matrix subject affects the interpretation of the embedded clause content only due to its scope. The one of the speaker, on the other hand, has scope over both the matrix clause and the embedded clause content. So, unless the embedded clause is opaque, the speaker’s temporal coordinates have access to the embedded clause at the same time as the matrix subject’s. Thus, the embedded clause has to be evaluated by the temporal coordinates of two individuals simultaneously. Such access is possible in –DIK clauses because they are transparent. How this relates to perspective is questionable, but if one can connect temporal coordinates to perspective, the above explanation can be utilized for perspective in –DIK clauses. However, if this route is taken up, one needs to account for where the differences between –DIK and diye clauses come from in terms of transparency; how one can account for potentially infinite agreement patterns discussed earlier in this section as well as why 71 –DIK clauses are overall less able to host CP-level elements such as speaker-oriented adverbs. Assuming all of this can be solved, the structure of –DIK clauses can look like this: [DP [CP OP [TP …]]], where the perspective operator is within the DP. But even then, this goes against observations that –DIK clauses are not big enough to host perspectival elements. Korotkova (2021) argues, based on her analysis of evidentiality markers in –DIK clauses, that “they lack the left periphery because the cut-off of the verbal spine happens lower (p. 220)” and that is why evidentiality markers cannot appear in them, as “evidentiality, and point-of-view in general (Speas & Tenny 2003; Sundaresan 2018; Zu 2018), is part of the left periphery (p. 220)”. Though I do not agree that –DIK clauses lack CP altogether, the defective nature of the CP in – DIK clauses is not likely to host a PersP, so in this sense I agree with Korotkova. Future work could show if these concerns are potentially solvable. However, since the focus of this dissertation is on diye clauses, I now move on to the proposal of this chapter, namely the structural properties of diye clauses. 2.4. Proposal In this section, the proposal is presented (Section 2.4.1.) and its general predications and implications are discussed with additional data (Section 2.4.2.-2.4.4). 2.4.1. The proposal I propose that diye is the overt realization of a perspective operator, OPdiye, with a PRO subject. The operator resides in the head of Pers(pectival)P(hrase), which takes a CP as its complement, 72 while the PRO subject is in SpecPersP. The PRO subject is controlled by the closest DP, which is the lowest attitude holder. I assume that the PRO is an obligatory control PRO, as it gets its meaning from another DP. As such, it must be bound in its local domain (Hornstein 1999). This explains why it is only the lowest matrix subject that can be its antecedent. The operator is a lambda abstractor that binds all the world variables that are born with each phrasal content and forces all elements in the embedded clause to be reported from the first- person perspective of the matrix subject via its PRO subject. The operator being associated with the first-person perspective of the closest matrix subject on the embedded clause content, it forms an evaluation domain. An evaluation domain is defined here as including content for evaluation (i.e., proposition in semantics and CP in syntax), an evaluator (i.e., matrix subject and PRO in syntax, marked in the operator in semantics as arguments) and evaluation worlds (logophoric alternatives, to be discussed, also marked in the operator). The operator makes its complement an opaque domain syntactically (to be detailed). The representation of a diye clause under bil- ‘know’ is in (48) (recall that hatırla- ‘remember’ is another verb in this category). Notice that the difference between –DIK in (39) and diye in (48) is that the former lacks PersP. (48) 73 Once PersP has been derived, the matrix subject is introduced in SpecVP of the matrix clause (Koopman & Sportiche 1991). The PRO has its controller antecedent and the full interpretation for the diye clause is achieved. At this stage, PersP is closed off for further syntactic operations. This indicates that diye clauses are opaque domains. Evidence that this is so comes in three ways. For one, binding of anaphors inside a diye clause by the matrix subject is not acceptable (49b), as opposed to one inside a –DIK clause (49a): (49) a. Çocuklari [Ali-nin birbirlerini-i eşleştir-diğ-in]-i bil-iyor children A-3GEN each.other-ACC match-DIK-3S-ACC know-IMPF ‘Childreni know [that Ali matched themi]’ (lit. Childreni know [that Ali matched each otheri]) b. *Çocuklari [Ali birbirlerini-i eşleştir-di diye] bil-iyor 31 children A each.other-ACC match-PST diye know-IMPF ‘Childreni know [diye Ali matched themi]’ (lit. Childreni know [diye Ali matched each otheri]) 31 One type of example to challenge this pattern and allow binding across domains would be to use an overt plural pronoun as embedded subject, as in (i): (i) *Çocuklari [onlari birbirlerini-i eşleştir-di diye] bil-iyor children they each.other-ACC match-PST diye know-IMPF ‘Intended: Childreni know [diye theyi matched themi]’ However, as will be discussed in Chapter 3, overt pronouns in Turkish can only be used for topic or contrast purposes and so, onlar in (i) cannot refer to the children. The binding can only happen between onlar and the anaphor, when onlar refers to a group of people other than the children mentioned. We can, however, use a covert pro to refer to children, as in (ii): (ii) *Çocuklari [proi birbirlerini-i eşleştir-di diye] bil-iyor children pro each.other-ACC match-PST diye know-IMPF ‘Intended: Childreni know [diye proi matched themi]’ But in my judgment, this sentence is still not grammatical. So, it looks like the type of embedded subject does not matter for the binding pattern under discussion. 74 Given that anaphors obey Principle A of the binding theory, which states that they need to be bound in their local domain (Chomsky 1981:198), diye clauses form opaque domains such that binding across them is disallowed. The PRO in SpecPersP cannot bind the anaphor because it is above CP, which I assume to be the relevant binding domain for the anaphor in a diye clause. Recall that the CP in –DIK clauses are defective, hence forming a transparent domain. That is why the binding goes through across clauses in (49a). Binding is not the only phenomena demonstrating the opaque nature of diye clauses. Another example regards NPI-licensing. The NPI kimse ‘anyone/no one’ in Turkish needs to be c-commanded by verbal negation, as the contrast in (50) shows: (50) a. Kimse gel-me-di no.one come-NEG-PST ‘Anyone did not come (i.e., No one came)’ b. *Kimse geldi NPIs cannot be licensed by matrix negation in diye clauses (51b) but can be in –DIK clauses (51a), suggesting that diye clauses are not transparent with respect to NPI-licensing under factivity-underspecified verbs while –DIK clauses are (due to their defective CP): (51) a. Ali [kimse-nin gel-diğ-in]-i bil-m-iyor A no.one-3GEN come-DIK-3S-ACC know-NEG-IMPF ‘Ali does not know [that anyone came]’ b. ??/# Ali [kimse gel-di diye] bil-m-iyor A no.one come-PST diye know-NEG-IMPF 75 ‘Intended: Ali does not know [diye anyone came]’ Another restriction imposed by diye clauses is on the interpretation of quantifiers. Assuming that de re readings are readings of specificity and specificity is related to the utterance context in some way (hence outside of the embedded context; achieved either via movement as in the Scope Theory, Montague 1974, Ladusaw 1977, among others, or indexation as in Keshet 2008, Anand & Korotkova 2021). 32 However de re readings are analyzed, what matters for present purposes is that association outside of the embedded context is not allowed in diye clauses. In (52), the embedded quantifier subject, someone, can only be interpreted as unspecific (=de dicto reading) in the diye clause (52b) while both de dicto (unspecific) and de re (specific) readings are available in –DIK clauses (52a): (52) a. Pelin [burada biri-nin hasta ol-duğ-un]-u bil-iyor P here someone-3GEN sick be-DIK-3SACC know-IMPF ‘Pelin knows [that someone is sick here]’ (specific) ‘Pelin knows [that there is someone sick here]’ (unspecific) b. Pelin [burada biri hasta diye] bil-iyor P here someone sick diye know-IMPF ‘#Pelin knows [diye someone is sick here]’ (specific) ‘Pelin knows [diye there is someone sick here]’ (unspecific) 32 I place this example in the part of the chapter dealing with syntactic evidence for the opaqueness of diye clauses because some studies argue for (covert) movement for de re elements, which I assume is syntactically relevant. 76 So far, we have established that diye is an opaque domain with respect to a number of properties 33 . It is also important to discuss what kind of an opaque domain diye clauses are. For example, can they be considered phases? If they can be, are they strong or weak phases? The data so far, however, do not help clarify this issue. The data presented and discussed in Chapter 3 will help us tease apart the distinguishing characteristic properties of diye clauses as opaque domains. Therefore, I postpone the discussion of this point to Chapter 3. It suffices now to keep in mind that diye clauses are some kind of opaque domains. Let us now recall the generalization we have established in Section 3 again: 33 The data just discussed indicate that matrix clause elements cannot interact with embedded elements in diye clauses. This suggests a ban on overt movement from the embedded clause but this is not borne out. Both–DIK (i) and diye (ii) clauses allow overt extraction of embedded elements: (i) a. Ali-nini Esra [ti kitab-ı oku-duğ-un]-u bil-iyor A-3GEN E book-ACC read-DIK-3S-ACC know-IMPF ‘Esra knows [that Ali read the book]’ b. Kitab-ıi Esra [Ali-nin ti oku-duğ-un]-u bil-iyor book-ACC E A-3GEN read-DIK-3S-ACC know-IMPF ‘Esra knows [that Ali read the book]’ (ii) a. Alii ben [ti kitab-ı oku-du diye] bil-iyor-um A I book-ACC read-PST diye know-IMPF-1S ‘I know [diye Ali read the book]’ b. Kitab-ıi Esra [Ali ti oku-du diye] bil-iyor book-ACC E A read-PST diye know-IMPF ‘Esra knows [diye Ali read the book]’ However, it should be noted that Turkish is a free word-order language and allow elements to be extracted from otherwise strong islands such as relative clauses. Recall from Chapter 1 that –DIK is the default form used in relative clause structures (iiia). It is possible to show elements can be extracted from these relative clauses, as in (iiib), where the temporal adverb bugün ‘today’ appears outside of the relative clause. Given this flexibility of the language, I will assume that the flexibility of overt movement out of diye clauses stems from the general relaxation on the rules of overt movement in the language. This will be discussed in more detail in Chapter 3: (iii) a. [Ali-nin bugün oku-duğ-u] (bir) kitap A-3GEN today read-DIK-3S a book ‘(A/The) book Ali read today’ b. [Ali-nin ti oku-duğ-u] (bir) kitap bugüni A-3GEN read-DIK-3S a book today ‘(A/The) book Ali read today’ One question regarding (ii) in this footnote is whether we can move biri ‘someone’ in (50b) to achieve a de re reading. It looks like such movement is not possible, likely due to this movement being topic-related (see Chapter 3) and the discomfort of unspecific quantifiers to move to this position. Such movement is possible in (50a), where birinin can move to the clause-initial position. I assume that is because the genitive marking on it can give it a specific reading, as genitive in Turkish marks specificity (von Heusinger & Kornfilt 2005). 77 (53) Generalization: Diye clauses encode a perspective center, forcing all phrasal content in the clause to be evaluated from that center and therefore forming a single perspective domain. Only the closest matrix subject can fulfil the role of perspective center. (–DIK clauses do not encode any perspective). The proposal in (48) accounts for the Generalization stated in (53), as the operator’s function is to report the embedded clause content from the first-person perspective of the matrix subject through its controlling of the PRO subject, forming a complete evaluation domain. This also accounts for the ban on de re elements in diye clauses via the high-scope of the operator. Since the operator scopes over all of the embedded clause, it forces all elements in it to be reported from the perspective of the matrix subject, yielding a de dicto reading. This implies that de re elements are expected to be disallowed, which was shown to be the case earlier. One last point that needs to be addressed in this section is the implication of the proposal that diye is the complement to the embedding verb. Is this the right conclusion, however? I argue that the diye clause is an argument of the cognitive attitude verbs. One reason is selectional restrictions. See, for example, (54), where the verb unut- ‘forget’ and keşfet- ‘discover’, both attitude verbs, select for a nominal complement but are incompatible with diye (see Chapter 4 for a more detailed discussion), unlike factivity-underspecified verbs that we have seen so far in this chapter. The diye clause is only acceptable with these verbs if it modifies the matrix verb to describe the cause of the forgetfulness (i.e., Esra forgot (it) because Ali came). This indicates that an adjunct reading is enforced to preserve interpretability in structures like (54c). (54) a. Esra Ali-yi unut-tu/keşfet-ti E A-ACC forget-PST/discover-PST ‘Esra forgot/discovered Ali’ 78 b. Esra [Ali-nin gel-diğ-in]-i unut-tu/keşfet-ti E A-3GEN come-DIK-3S-ACC forget-PST/discover-PST ‘Esra forgot/discovered [that Ali came]’ c. *Esra [Ali gel-di diye] unut-tu/keşfet-ti E A come-PST diye forget-PST/discover-PST ‘Intended: Esra forgot/discovered [diye Ali came]’ Another argument is based on Major (to appear). Major argues that dep clauses in Uyghur are not selected by the matrix verb and uses replacement by shundaq ‘so/like that’, which stands in for salient propositions, as argument for this proposal. Dep clauses can appear together with shundaq in Uyghur, indicating that they are not in the same position. This is exemplified in (55): (55) a. Herbir oqughuchi shundaq oyla-y-du each student so think-NONPST-3S ‘Each student thinks so’ b. Herbir oqughuichi [[Aygül-nij [tj ket-t-i]k dep] we each student A-ACC leave-PST-3S dep and Ahmet-nii [ti kel-d-i] m dep]] shundaqk+m oyla-y-du A-ACC come-PST-3S diye so think-NONPST-3S ‘Each student thinks it so that Aygul left and Ahmet came’ Diye clauses cannot appear with öyle ‘so’, indicating that they compete for the same position, namely the internal argument position: 79 (56) a. Ali öyle bil-iyor A so know-IMPF ‘Ali knows so’ b. *Ali [Esra gel-di diye] öyle bil-iyor A E come-PST diye so know-IMPF ‘Intended: Ali knows (it) so [diye Esra came]’ It is important to mention Özyıldız (2019), who argues that diye is an adjunct to the VP. One reason is that it cannot be case-marked (57b) although bare clauses can (57a) (p. 9, ex. 32b-c): (57) a. Ali gel-di-ye benz-iyor A come-PST-DAT look.like-IMPF ‘Ali looks like he arrived’ b. *Ali gel-di diye-ye benz-iyor A come-PST diye-DAT look.like-IMPF ‘Intended: Ali looks like he arrived’ The lack of case-marking on diye can be expected, since it is not a nominal element. On this explanation, however, the case-marking on (57a) is surprising. I do not have a straightforward answer as to why. Özyıldız admits in fn. 15, the structure of (57a) is probably more complex than it looks. Nevertheless, it is important to note that bare clauses can be case-marked in Turkish in other environments too (while diye can never be), indicating that the pattern of case-marking is a generalized one. 34 34 Both Özyıldız and the present study treat diye as a non-nominal element. But I believe that a clearer explanation for this issue can be possible in my account if we can posit a generalization that states “any complement that is not a PersP can be case-marked”. Only diye and matrix clauses have PersP in my account; neither –DIK clauses nor (possibly) bare clauses have it. This point can be pursued in future studies. 80 (58) a. Ali-yi falan bil-me-m A-ACC and.the.like know-NEG-1S ‘I don’t know Ali or whoever’ b. Ali gel-di-yi falan bil-me-m A come-PST-ACC and.the.like know-NEG-1S ‘I don’t know Ali came or whatever’ One thing to note, however, all environments I was able to come up with are in internal argument positions. Bare clauses cannot appear in external argument positions, neither can diye clauses. So, this argument does not rule out the possibility that position is at play in this and it might still be true that –DIK clauses are arguments while diye clauses are adjuncts. So, we look at the second piece of evidence. The second argument by Özyıldız, also forming the main type of structures focused on by Major (to appear) in Uyghur dep, is that diye can function as an adjunct clause (example from Özyıldız 2019: 10, ex. (33c)), as exemplified in (59): (59) Ayşe bir soru-yu [Ali gel-di diye] cevapla-dı A one question-ACC A come-PST diye answer-PST ‘Ayşe answered a question by saying that Ali had arrived’ Notice that there is a DP in (59) that can function as the argument to the VP, namely bir soruyu ‘one question-ACC’. The accusative marking on it indicates argumenthood. When there is already an argument, diye clauses function as adjuncts. The real question is what happens when they are the only element that can function as an argument (i.e., no other nominals are available for the internal argument position)? Now, see (60), where there are two diye clauses: 81 (60) Ali [Murat öyle de-di diye] [Ayşe git-ti diye] bil-iyor A M so say-PST diye A go-PST diye know-IMPF ‘Ali knows [diye Ayse left] [because Murat said so]’ If both of the diye clauses in (60) were both adjuncts, we would expect them to get similar interpretations. However, the one that is closer indicates the content of Ali’s knowledge, which is the default reading for propositional internal arguments, and the one that is further has an adjunct interpretation. Necessarily, the one that is further to the VP is interpreted as an adjunct, in this case, a reason clause. The one that is closer to the verb is the argument, also evidenced by its interpretation as the content of the attitude verb. I assume that when diye appears as an adjunct, it might have a similar structure to the one proposed here, namely a PersP, but this PersP is not selected by the verb directly but by a CausP, which mediates the matrix subject and the embedding event (including the propositional argument) via a cause relation. The verbs in focus here do not select for a CausP, giving these diye clauses independence from a lexically selecting head. The adjunct reading is the main focus of Major (to appear). Likewise, in (59), the diye clause might be headed by a CausP. But it turns out that structures like (59) has some more interesting implications, which I discuss below. Let us consider an example with a cognitive attitude verb (Özyıldız 2017: 13, ex. (20b)): (61) Tunç {o-nu, o durum-u} Hillary kazan-dı diye bil-iyor T it-ACC that situation-ACC H win-PST diye know-IMPF ‘Tunç believes of {that, that situation} that Hillary won’ The embedded subject in diye clauses can appear in accusative (impossible with –DIK clauses): 82 (62) Ali Esra-yı git-ti diye bil-iyor A E-ACC go-PST diye know-IMPF ‘Ali knows of Esra {diye she left}’ Major (to appear) investigates a similar structure in Uyghur (more details in the next section) and argues that the accusative is base-generated in the embedded clause and raises to the matrix clause. In other words, it is not prolepsis. However, in the Turkish diye clauses, there is evidence that these are more similar to prolepsis structures, where the accusative embedded subject is an argument of the matrix verb which controls a null pronoun in the embedded clause. Evidence that the Uyghur accusative-marked DP is generated in the embedded clause comes from the fact that NPI elements, which require c-commanding clause-mate negation as licenser, can be licensed by embedded negation (Shklovsky & Sudo 2014; Major to appear): (63) a. Héchkim ket-*(mi)-d-i nobody leave-NEG-PST-3S ‘Nobody left’ b. Ahmet [héchkim-ni ket-*(mi)-d-i] dé-di A nobody-ACC leave-NEG-PST-3S say-PST-3S ‘Ahmet said that nobody left’ In Turkish, the accusative DP needs to be licensed by matrix negation: 35 (64) a. *Ahmet kimse-yi git-me-di diye bil-iyor A no one-ACC go-NEG-PST diye know-IMPF ‘Intended: Ahmet knows of noone [diye they left]’ 35 The judgments are more variable here. (64) is aligned with my own judgment and some of the informants. 83 b. Ahmet kimse-yi git-ti diye bil-m-iyor A no one-ACC go-PST diye know-NEG-IMPF ‘Ahmet knows of noone [diye they left]’ The second argument for Uyghur not having prolepsis with accusative DPs come from idioms, which preserve meaning (65) (Major to appear). When this argument is applied to Turkish diye, we see that the idiomatic meaning is not preserved (66): (65) a. Toqquz qiz-ning tolghiq-i teng kel-d-i nine girl-GEN labor-3POSS together arrive-PST-3S ‘Times are hard (lit. Nine girls’ labor came all at once)’ b. Tursun [toqquz qiz-ning tolghiq-i-ni teng kel-d-i T nine girl-GEN labor-3POSS-ACC together arrive-PST-3S ‘Tursun said that times are hard.’ (66) a. At-ı al-an Üsküdar-ı geç-ti horse-ACC take-REL Ü-ACC pass-PST ‘It is too late (lit. That who took the horse passed Üsküdar)’ b. #Ali at-ı al-an-ı Üsküdar-ı geç-ti diye bil-iyor A horse-ACC take-REL-ACC Ü-ACC pass-PST diye bil-iyor ‘Intended: Ali knows [diye it is too late]’ We also see that there is no reconstruction because de dicto readings are not allowed (68a), unlike in Uyghur (67) (Shklovksy & Sudo 2014: 392, ex. 29). In Turkish, de dicto reading is possible with a nominative embedded subject (68b): 84 (67) Tursun tulpar-ni kel-d-i] dé-d-i, ema tulpar yoq T winged.horse-ACC arrive-PST-3S say-PST-3S but winged.horse not.exist ‘Tursun said that a winged horse arrived, but winged horses do not exist’ (68) a. #Esra hayalet-i git-ti diye bil-iyor E ghost-ACC go-PST diye know-IMPF ama hayalet diye bir sey yok but there is no such thing as a ghost ‘Esra knows of the ghost [diye it left] but there is no such thing as a ghost’ b. Esra [hayalet git-ti diye bil-iyor E ghost go-PST diye know-IMPF ama hayalet diye bir sey yok but there is no such thing as a ghost ‘Esra knows [diye the ghost left] but there is no such thing as a ghost’ The conclusion is that the accusative DP is not base-generated in the diye clause nor do they reconstruct. That means it is in a proleptic structure, where it controls an embedded null pronoun, which I represent as in (69a) for the sentence Ali Esra’yı gitti diye biliyor ‘Ali knows of Esrai diye shei left’, reminiscent of the proposal by Bruening (2001a, b) for Japanese (69b): 36 (69) a. [vP Ali [VP [Esrai-ACC [XP ti [ PersP diye [CP proi went]]] know]]] b. NP-ACCi…[CP ti […proi...] C] V 36 Also see Oğuz (2021) for a prolepsis analysis of accusative subjects in Turkish. 85 Here, the accusative-marked subject is the special subject of the diye clause, controlling the pro within the diye clause and generated in a position higher than PersP within the diye clause (a special projection which I assume is projected in proleptic structures). It moves to the matrix clause for case-assignment after generation (because SpecCP is not a case-assigning position). This accounts for the evidence so far that acusative subjects are not generated within the diye clause (rather it is outside of PersP) and the proposal thus combines pieces of previous research on the topic (I adopt the idea of raising to the matrix clause from Zidani-Eroğlu (1997) because I assume a similar movement, -although the starting point is different-. I also assume that this movement happens via a base-generated position at the edge of the embedded clause that is accessible to the matrix verb so it can get case-assigned by it because I also assume an edge position for it. The idea of an edge position for accusative subjects is similar to Şener (2008)). What this shows is that diye clauses are not selected by the main verb when there is another DP saturating the internal argument position (since the DP gets assigned accusative in the matrix clause in (69a)). Thus, my assumption is that diye clauses are arguments when there is no other element accomplishing the argument role (based on evidence from selectional restrictions, replacement by öyle ‘so’ and their obligatory content interpretation). 2.4.2. Previous syntactic accounts of diye and similar elements I now discuss two previous studies that bear striking resemblance to the present proposal. One is Yıldırım-Gündoğdu (2017a), who proposed a syntactic account of diye in Turkish and the other is Knyazev (2016), who proposed a syntactic account for giǯǝ in Kalmyk, a Western Mongolic language. I also discuss Major (to appear) on Uyghur dep intermittently in relation to these two 86 studies. None of these studies focused on perspective per se, 37 however, the syntactic patterns they propose for diye (and similar) clauses are similar to the present proposal. I drew heavily from these previous proposals in the development of my own proposal but diverged in some ways. Yıldırım-Gündoğdu (2017a) sets out with the goal of analyzing various functions of diye in Turkish and ends up proposing various structures for each of them. Here, I am only concerned with the one that appears under cognition verbs. I also only present her proposal for diye clauses that involve “indexical shift” of first-person pronouns (“indexical shift” to be exemplified soon in this chapter and detailed in Chapter 3) since the structure is easier to understand. But it should be noted that there is no significant difference between diye clauses with or without “indexical shift” for present purposes (to be clarified in Chapter 3). Yıldırım-Gündoğdu proposes (70) as the representation of diye: (70) 37 If they were concerned with perspective, I believe they would encode it into the context-overwriting operator that they propose. 87 There are several aspects of this proposal worth mentioning here. One is the decomposition of diye into de- ‘say’ and –ye, a converbial marker. Converbial markers form V+V constructions where one verb is subordinate to the other. Converbial V+Vs are different from true serial verbs in that they involve converbial suffixes whereas true serial verbs are not connected by “overt marker of coordination, subordination, or syntactic dependency of any sort (Aikhenvald 2006:1, Mufwene 1990).” Another similarity between converbial forms and true serial verbs is that in the former the verbs necessarily share their subject. These forms are common in Turkic languages. (See Straughn (2006, 2011) for more details on converbs in Turkic languages). It is common to consider diye as the converbial form of the verb de- ‘say’ formed with –(y)A (Yıldırım- Gündoğdu 2017a and the references therein). Yıldırım-Gündoğdu encodes this function of subordination on a separate functional head, namely CvbP. This phrase can be seen as the enabler of a certain subordination type. I did not use this phrase since it is largely descriptive and does not play much role in her proposal. If we disregard CvbP, one can see the similarity between the present proposal and Yıldırım-Gündoğdu’s more clearly: both have a CP and a higher phrase level (VP in hers, PersP in mine) and in both there is a PRO subject associated with the matrix subject in the specifier of the higher of the two levels. The reason why Yıldırım- Gündoğdu chooses to use a VP is due to the similarities diye displays with the verb de- ‘say’. For one, diye is derived from de- ‘say’. Another similarity is that both allow “indexical shift” as opposed to other verbs. Diye clauses allow the first-person agreement on the embedded verb to either refer to the matrix subject or the speaker (71a). When it refers to the matrix subject, this is known as indexical shift. The verb de- ‘say’ also allows, in fact enforces, this shift (71b). (71c) shows that other attitude verbs such as düşün ‘think’ cannot shift indexicals in the absence of 88 diye. This shows that diye and de- both enable shifting. Therefore, Yıldırım-Gündoğdu concludes that they have similar structures. (71) a. Ahmeti [ödev-i bitir-di-m diye] düşün-üyor A homework-ACC finish-PST-1S diye think-IMPF ‘Ahmet thinks that hei/I finished the homework.’ b. Ahmeti [ödev-i bitir-di-m] d-iyor A homework-ACC finish-PST-1S say-IMPF ‘Ahmet says that hei/*I finished the homework.’ c. Ahmeti [ödev-i bitir-diğ-im]-i düşün-üyor A homework-ACC finish-DIK-1S-ACC think-IMPF ‘Ahmet thinks that *hei/I finished the homework.’ Although this proposal is intuitive, I believe this proposal does not distinguish the VP in diye from other VPs. In other words, diye cannot be a bare VP but rather must involve an element that relates to perspective. To start, both Yıldırım-Gündoğdu (2017a) for diye and Major (to appear) for dep in Uyghur argue that these markers are morphologically derived from the verb de- ‘say’ and they carry properties of this verb. That is part of their argument that there is a VP in diye or dep. Although I agree that diye derives from de- ‘say’, I do not agree with diye carrying a VP is due to a similarity with the verb de- ‘say’. This is because the shifting behavior is different between diye and de-, suggesting that the two are not the same. This difference can be seen in (71a) versus (71b), where diye allows shifting whereas de- enforces it. Likewise, as discussed in more detail in Chapter 3, overt pronominals such as ben ‘I’ can (even must for some speakers) shift under de- but not under diye in Turkish (72): 89 (72) a. Ahmeti [ben ödev-i bitir-di-m diye] düşün-üyor 38 A I homework-ACC finish-PST-1S diye think-IMPF ‘Ahmet thinks that *hei/I finished the homework.’ b. Ahmeti [ben ödev-i bitir-di-m] d-iyor A I homework-ACC finish-PST-1S say-IMPF ‘Ahmet says that hei/*I finished the homework.’ Regardless, though, there is a common point between diye and de-: they both make indexical shift at least possible. The monster operator in (70) can be considered as the element that enables it, which has been argued for in studies on Uyghur indexical shift (e.g., Shklonsky & Sudo 2014, Major to appear) for dep as well as Turkish indexical shift in diye, as in Yıldırım-Gündoğdu (2017a) (and the assumption is that de- also has a similar operator). Yet there is evidence that the operator should be higher in the embedded clauses formed with diye, presumably in the VP in Yıldırım-Gündoğdu’s account. Notice that the embedded CP is not within the scope of the operator in Yıldırım-Gündoğdu’s account and the operator is a context-overwriting one. 39 It is similar to Major’s account. When there is indexical shift, the Uyghur dep has the structure in (73), in which the monster is in the CP: 38 A good test to tease apart the role of the main verb would be to use –DIK clauses in the complement position, since the person agreement markers in these clauses always refer to the speaker (so we could see if the verb de- ‘say’ can shift the first-person agreement marker in the -DIK clause as well). However, the verb de- ‘say’ does not take – DIK clauses as complements, so there cannot be a direct comparison with düşün- ‘think’. 39 I differ from Yıldırım-Gündoğdu in the type of the operator I propose. Since this is not related to the syntax of diye directly, I will not discuss it here. But I will oppose to in Chapter 3, where I will argue for a binding operator in diye clauses rather than a context-overwriting explicitly. 90 (73) This predicts that CP-elements themselves will not be context-overridden, which means they need not be associated with the matrix subject (e.g., it need not be evaluated by him). See (74) to that end. Context 1 makes it clear that the speaker is angry about Ali attending the party while the matrix subject is not and Context 2 reverses the emotions. The CP-level, speaker-oriented adverb iyi ki ‘fortunately’ can only reference the matrix subject’s perspective, indicating that the embedded CP falls in the scope of the operator. 40 Recall that in my account, PersP –hosting the perspective operator- lies above CP. Therefore, it takes within its scope anything in the CP, including speaker-oriented adverbs. In Yıldırım-Gündoğdu’s account, on the other hand, the operator lies below the CP, and so it does not capture this pattern. (74) Context 1: ACCEPTABLE The speaker is angry that Ali attended the party. Pelin is happy about it: Context 2: NON-ACCEPTABLE The speaker is glad that Ali attended the party. Pelin is angry about it: 40 -DIK clauses do not accept CP-level adverbs. See Chapter 3 for more detailed discussion of CP-level adverbs. 91 Pelin [Ali iyi ki parti-ye gel-di diye] bil-iyor P A fortunately party-DAT come-PST diye know-IMPF ‘Pelin knows [diye fortunately Ali came to the party]’ Moreover, in some other structures where we expect a CP, such as in the complement of a head noun as in (75), the first-person agreement marker cannot shift. Thus, it looks like the presence of a CP does not guarantee shift. This puts the argument that the CP in diye clauses host the shifting operator in doubt for Turkish or some explanation as to why that CP is special needs to be offered. In my account, the operator is in the PersP and not CP, which circumvents the issue. 41 (75) a. Ali [okul-dan kaç-tı-m diye bir söylenti] duy-muş A school-ABL escape-PST-1S diye a rumor hear-INDPST ‘Ali heard [a rumor diye *he/speaker escaped from school]’ b. Ali [okul-dan kaç-tı-m diye bir söylenti] bil-iyor 42 A school-ABL escape-PST-1S diye a rumor know-IMPF ‘Ali knows [a rumor diye *he/speaker escaped from school]’ Lastly, it is unclear how Yıldırım-Gündoğdu’s proposal can account for the opaque properties of diye. Since the highest phrase of diye is CvbP, there is no apparent reason why an embedded element should not interact with an external probe through the CvbP. In my account, this 41 This entails that diye may not always come with a PersP, for example not when it is not embedded under a verb. Indeed, diye behaves differently with nouns. For example, it only can occur with an unspecific head noun as in (75). If the head noun is specific, -DIK clauses must be used instead in the complement position. Under nouns, the complement of diye can be merely a noun as in (i), which also contradicts my proposal that diye has a propositional argument. Thus, diye that occurs with nouns require further research and thus, I leave it for future studies. (i) [Ali diye] bir adam A diye a man ‘A man called Ali’ This is actually unlike Uyghur, where dep (a comparable element to diye) always comes with the possibility of shifting (Major, to appear, Footnote 16). 42 To the extent that speakers accept söylenti ‘rumor’ and bil- ‘know’ together. 92 property is accounted for by locality constraints formed by sealing off an evaluation domain by PRO and OPdiye. Knyazev (2016) makes a similar proposal for the Kalmyk giǯǝ, which is morphologically identical to diye in that it is formed by the verb gi- ‘say’ in the language and the converbial –ǯǝ. In this account, gi- in giǯǝ corresponds to CP+VP in the syntax (rather than only VP, as is the case in Yıldırım-Gündoğdu’s account) whereas –ǯǝ corresponds to CvbP, like in Yıldırım- Gündoğdu’s proposal. I can therefore use the same counter-points as the ones I used for Yıldırım-Gündoğdu to support my own proposal. There is one other suggestion Knyazev (2016) uses: that giǯǝ forms a complex predicate with the embedding attitude verb, resulting either in a ‘say’ interpretation or a ‘belief’ interpretation depending on the meaning of the embedding attitude verb, as giǯǝ itself is semantically empty. First, diye in Turkish has clear perspectival properties and is therefore not semantically empty. The second point is that diye does not form a complex predicate in Turkish. To see why, let’s define complex predicates as in (76), from Baker (2014:20): (76) Interpret X and Y as complex predicate at LF if [and only if] X and Y form a complex head (an X 0 ). This definition implies that the two parts of the complex predicate should be inseparable, which is what is argued for by Sağ (2017) for different kinds of complex predicates in Turkish. For example, the question particle cannot come in between the two verbs in these constructions (in brackets in (77) (Sağ 2017: 2, ex. 4): 93 (77) Bulut biz-i bekle-rken [uyu-du (*mu) kal-dı] mı B we-ACC wait-while sleep-PST Q stay-PST Q ‘Did Bulut fall asleep while waiting for us?’ The question particle can break diye and the embedding verb, however, showing that diye and the embedding verb do not form a complex predicate in Turkish: (78) Pelin [Ali parti-ye gel-di diye] mi bil-iyor P A party-DAT come-PST diye Q know-IMPF ‘Does Pelin know Ali came to the party’ Major (2021, to appear) states that the Uyghur dep, some of whose properties are similar to diye, is derived from the verb de- ‘say’ and converbial (linking) suffix –I(p). The Turkish diye also derives from de- ‘say’ but not in combination with -(I)p but with –ye (which survives in diye and adverbial reduplication structures in Modern Turkish, so diye is now probably close to an isolated form with this marker). 43 Turkish has the converbial suffix –(I)p, but its properties contrast with diye. For example, presupposition triggers bile/de can appear between diye and the embedding verb (79a) but not between -(I)p and the verb it combines with (79b): 44 (79) a. Ali [Esra gel-di diye] bile/de bil-iyor A E come-PST diye even/too know-IMPF ‘Ali even/too knows [diye Esra came]’ 43 This is one reason why the analysis of Uyghur dep in Major (2021) cannot be easily applied to Turkish diye, as – ye in diye is not productively used in the language. The other reason is that there are some significant differences between diye and dep, as in the case of shifting behavior and many others, which I discuss in relevance to various discussion points throughout the dissertation. 44 This also shows that diye is not merely an assertion. 94 b. Ali gel-ip *bile/??de git-ti A come-(I)p even/too go-PST ‘Intended: Ali even/too came and went’ This not only shows that diye is not in canonical V+V (or converbial) structures in Turkish, but it also shows that the proposals for Uyghur and Kalmyk are not directly applicable to Turkish. Indeed, there are some remarkable differences between these languages and Turkish. For example, Kalmyk has multiple subordinators derived from the verb for ‘say’ whereas Turkish only has one. That is likely a major source of difference. Uyghur dep allows shifting in all environments it appears in (Major to appear, Footnote 17), whereas Turkish diye does not always allow shifting, as was discussed earlier. Due to all the reasons discussed in this section, I argue that my proposal is better than Yıldırım-Gündoğdu (2017a), Knyazev (2016) and Major (to appear) in accounting for the (i) perspectival properties and (ii) opaque properties of diye clauses as well as (iii) the relation of diye clauses to the embedding verb in Turkish. This differentiation is necessary because (i) the shifting behavior of diye and the related verb de- ‘say’ is not the same and (ii) there is evidence that the operator needs to be the highest element in the embedded clause, including elements at the CP. Therefore, my proposal is a better fit for the Turkish data. 2.4.3. Logophoricity in diye clauses Another important property of diye clauses is logophoricity in diye clauses. The logophoric nature of diye has been assumed in previous work (Özyıldız et al. 2018) but to my knowledge, has not been explicitly supported by data. In this chapter, I aim to do so. 95 Logophoric pronouns, which make reference to “the person whose speech, thoughts or perceptions are reported (Clements 1975)”, have been brought into attention in the linguistic literature through work on a number of West African languages. For example, the example in (80) from Yoruba (Pulleybank 1986) shows that the logophoric pronoun oun can only refer to the subject of the attitude verb (80b) whereas the regular pronoun o can refer to someone else (80a): (80) a. oi ri pe oj ni owo he saw that he had money b. oi ri pe ouni ni owo he saw that he(LOG) had money Turkish does not have a logophoric pronoun as in the languages discussed in the literature but diye displays certain properties of logophoric environments. For one, intolerance to insentient subject is a sign of logophoricity (Diercks 2013, Charnavel & Sportiche 2016) and diye clauses do not accept non-sentient subjects (81b) as opposed to –DIK clauses (81a): (81) a. ?Araba [pro yük-ün-ün ağır ol-duğ-un]-u bil-iyor 45 car burden-3GEN-3GEN heavy be-DIK-3S-ACC know-IMPF hareket etmiyor it is not moving ‘The car knows [that its burden is heavy] so it is not moving’ 45 The fact that bil- ‘know’ accepts a non-sentient subject in (81a) is unusual. I attribute this ability to the fact that Turkish bil- does not report knowledge reports per se, as mentioned earlier. This is an interesting puzzle to pursue a solution for. 96 b. #Araba [pro yük-ü ağır diye] bil-iyor hareket etmiyor car burden-3GEN heavy diye know-IMPF it is not moving ‘The car knows [that its burden is heavy] so it is not moving’ Insentient subjects are a theoretically neutral way of testing for logophoricity (Charnavel & Sportiche 2016). De se readings are not. However, de se examples, including de nunc and de hic, also require consciousness, in addition to other requirements (e.g., identity). Thus, I point out here that diye clauses show consistent patterns with such readings. This can be taken as further evidence that diye clauses require consciousness. An analysis of these readings of diye clauses is left for future research, though I point out to some directions that can be taken. Rather I only focus on their requirement for awareness/consciousness. 46 De se (Latin: about oneself) reading is related to an attitude holder’s awareness towards himself/herself. A famous example of a de se scenario is from Kaplan (1989: 533). Say, John sees someone’s pants burning and he thinks it is Dan’s pants burning. John yells “His pants are burning!”. Then, he smells the smoke and looks down, finally realizing that it is his own pants that are burning. This time, he yells “My pants are burning!”. This latter reading is the de se reading. The importance of these examples is that the belief John holds in these scenarios are different, yet His pants are on fire and My pants are on fire are propositionally equivalent. Thus, de se readings play a distinguishing role in the study of mental attitudes. It has been observed, however, that this differentiation does not carry over to attitude ascriptions. In other words, de se readings are not necessarily differentiated in linguistic attitude 46 The focus on the awareness component of de se readings has been practiced before. For example, Nishigauchi (2014: 173) states regarding Japanese zibun ‘self’ that “it is important to note that de se interpretation has direct relevance for the awareness requirement on the interpretation of zibun” (although zibun naturally satisfies the identity requirement too). 97 contexts. Kaplan’s scenario that we just discussed, for example, can be expressed with one and the same sentence in (82). Another example is from Anand (2006: 14) in (83): (82) Context 1: John: “His pants are on fire” (Non de se) Context 2: John: “My pants are on fire” (De se) Johni thinks hisi pants are on fire. (ok in both contexts) (83) Background: While camping last summer, John and his friends found a 100 foot cliff overhanging a deep lake. They set up a camcorder at the bottom and proceed to dive off the clifftop. John performs a very difficult dive (say, a backwards somersault). Upon reviewing the footage months later, he discovers that the height of the cliff obscured the identity of the divers. Context 1: Remembering that only he did a backwards somersault, John thinks, “I had the best dive.” (De se) Context 2: Highly impressed by the backwards somersault, John thinks, “That guy had the best dive.” (Non de se) Johni believed that hei was the best diver. (ok in both contexts) De nunc is temporal de se. The following Korean example is one that requires a de nunc reading (Deal 2020: 92). The example below shows that the matrix subject must be aware that today is Monday for the reference of onul ‘today’ to be acceptable. (84) John-i cinanwu welyoil-ey [Mary-ka onul ttenanta-ko] J-NOM last.week Monday.on M-NOM today leave-C 98 malhayssta said ‘John said last Mondayi that Mary would leave that dayi’. a. ok Last Monday John said, “Mary is leaving today”. b. #John knew that Mary was going to leave on Monday last week. On that day, John somehow thought it was Sunday rather than Monday, and said, “Mary leaves tomorrow, on Monday.” (Park 2016, ex. 29) De hic is locative de se. A Zazaki example is given in (85), from Anand (2006: 80). In this example, ita ‘here’ must refer to a location which the attitude holder identifies as hers: (85) Waxto kɛ o London-de bime Pierri va when that he London.at be-PAST P-OBL said [kɛ o ita rindɛka] that it here be-pretty-PRES When he was in London, Pierre said it is pretty there. a. ok Pierre says in London, “It is pretty here.” b. #Pierre is walking around London, which is drab and rather disappointing. He says, “I wish I were in Londres. Londres is pretty.” Other elements that require de se interpretation according to Anand (2006) (as well as Charnavel 2019) are logophors, shifted indexicals and long-distance anaphors. All have been argued to involve different types of operators governing them (e.g., Anand 2006 for all these elements), sometimes interacting with the lexical features of these elements (Charnavel 2019 for long- distance anaphora; Deal 2020 for indexical shift). 99 Now, I present de se data to test in –DIK and diye clauses. 47 The first example is in (86). It shows that when the matrix subject is not aware that he himself won the election (=non de se context), the diye clause is unacceptable (86b) (pro has to be de se): 48 (86) Context: Murat is watching the results of the election, of which he was one of the candidates, on TV. He is drunk. He sees the winner and says to me “That candidate won”. He does not realize that that candidate is himself. a. %Murati [proi seçim-i kazan-dığ-ın]- ı bil-iyor aslında M election-ACC win-DIK-3S-ACC know-IMPF actually ‘Murati knows [that hei won the election] actually’ …ama kendisi olduğunun farkına varamadı daha …but he is not yet aware that it is himself b. #Murati [proi seçim-i kazan-dı diye] bil-iyor aslında M election-ACC win-PST diye know-IMPF actually ‘Murati knows [diye hei won the election] actually’ …ama kendisi olduğunun farkına varamadı daha …but he is not yet aware that it is himself Another example is adapted from Charnavel (2018b) in (87), where the matrix subject is not aware that it is her own picture being displayed. Diye is unacceptable in such a context (87b): 47 The judgements in this section are consistently less clear, however, diye clauses are consistently worse than –DIK. 48 In all examples in this section, the continuations are crucial to grammatical judgments. They improve –DIK clauses but they do not lead to any improvement in diye clauses. Without the continuations, -DIK clauses do not sound natural either, as they are not the most natural sentence structure that would be used in these contexts. 100 (87) Context: There is a picture of a woman from the back in the exhibition. It is Esra’s picture, but Esra does not recognize herself. She thinks it is a picture of herself. a. ?Esrai [proi resm-i-nin sergilen-diğ-in]-i bil-iyor aslında E picture-3S-3GEN be.exhibited-DIK-3S-ACC know-IMPF actually ‘Esrai knows [that heri picture is exhibited] actually’ …ama kendisi olduğunun farkına varamadı …but she could not be aware that it is herself b. #Esrai [proi resm-i sergilen-di diye] bil-iyor aslında E picture-3S be.exhibited-PST diye know-IMPF actually ‘Esrai knows [diye heri picture is exhibited] actually’ …ama kendisi olduğunun farkına varamadı …but she could not be aware that it is herself The following examples in (88) and (89) show that the matrix subject must also have a de se understanding of time (de nunc) and location (de hic) of the embedded event, respectively. In (88), the time interpretation is presumably in tarih ‘date’, which refers to the present day unless otherwise specified. The location in (89) is encoded in the nominal mekan ‘place’, which refers to the present location unless otherwise specified: 49 (88) Context: It is December 31 st . John thinks it is December 30 th . John told the speaker that it was December 31 st before, but now he says it is December 30 th . 49 I avoided using bugün ‘today’ and bura ‘here’ in (88) and (89) because these elements act more like rigid indexical (especially the latter). 101 a. ?John [tarih-in 31 Aralık ol-duğ-un]-u bil-iyor aslında J date-3GEN 31 December be-DIK-3S-ACC know-IMPF actually ‘John knows [that today is NYE] actually’ …ama unuttu herhalde …but he apparently forgot b. #John [tarih 31 Aralık diye] bil-iyor aslında 50 J date 31 December diye know-IMPF actually ‘John knows [diye today is NYE] actually’ …ama unuttu herhalde …but he apparently forgot (89) Context: John, Ali and the speaker are in Istanbul. John thinks he is in Constantinople, not being aware that Istanbul and Constantinople refer to the same city. But earlier, he said to the speaker that he knew they are in Istanbul. a. ?John [mekan-ın Istanbul ol-duğ-un]-u bil-iyor aslında J location-3GEN I be-DIK-3S-ACC know-IMPF actually ‘Johni knows [that the location is Istanbul] actually’ …ama unuttu herhalde …but he apparently forgot b. #John [mekan Istanbul diye] bil-iyor aslında J location I diye know-IMPF actually ‘Johni knows [diye hei is in Istanbul]’ …ama unuttu herhalde 50 Assume this sentence is a reply to the question John bugünü ne diye biliyor? ‘What does John know today as?’ 102 …but he apparently forgot The conclusion is that diye involves some kind of logophoricity. 51 2.4.4. Semantics of diye Since the focus of this dissertation is on the syntactic properties of diye clauses, the semantics of the operator will not play a crucial role in the analysis. Moreover, since I focus on various types of elements within diye clauses, it is hard to come up with a semantics for diye that can address all nuances (for example, I put together elements that are normally not considered the same semantically: locative expressions and adjectives and nouns, all of which have different semantics in the literature). For these two reasons, I do not attempt to come up with a full semantic analysis but rather point to a possible way of putting patterns together in a generalized sense. Future research can look into the technical details of how this unification can come about. I assume that phrasal elements bound by OPdiye must come with a world variable that must be bound by the closest binder (Percus 2000), which is OPdiye in our case. This binding is roughly represented in (90): (90) a. [ PersP PRO λw’ [CP … [AdvP w’ between 4-5]…] diye] 51 The test for strict/sloppy readings is not readily applicable to diye clauses. I tried it with pro elements that are associated with the first-person agreement marker and both strict and sloppy readings are possible in this case (i). However, this marker itself is flexible in reference, as will be discussed in Chapter 3. This makes the test less conclusive. The same goes for anaphors such as kendi or kendisi ‘himself/herself’. (i) Murat [pro sınıf-ta kal-dı-m diye] bil-iyor. Ali de öyle bil-iyor M class-LOC stay-PST-1S diye know-IMPF Ali also knows so. Strict: ‘Murati knows [diye Ii failed the class]. Ali also knows [diye Murat failed the class]’ Sloppy: ‘Murati knows [diye Ii failed the class]. Ali also knows [diye Ali failed the class]’ 103 b. [ PersP PRO λw’ [CP[DP w’ the nice woman]…[AdvP w’ between 4-5]…[VP w’ came]] diye] The function of OPdiye is, informally speaking, to report the embedded content in the CP from the first-person perspective of the matrix subject, following the definition in Charnavel (2019) of logophoric operators for exempt anaphors in French and some other languages. I define the operator introduced by diye as in (91a), adapting the proposal in Charnavel (2019) and the embedding verb as in (91b) semantically. The individual argument x, of type e, is saturated by PRO in the specifier of the phrase the operator is base-generated (PersP) and PRO itself is controlled by the closest attitude holder (i.e., the lowest matrix subject). The proposition argument is saturated by the embedded CP content and it returns a truth value iff it is compatible with the logophoric alternatives of the individual x. This way, all of the embedded proposition gets evaluated from the first-person perspective of the closest matrix subject. I assume that logophoric alternatives encompass attitudinal alternatives, namely those that are compatible with what the attitude holder believes to be true in the actual world. Like the logophoric operator in Charnavel (2019), the definition in (91a) assimilates diye into an intensional predicate, which captures intuitions that diye derives from the verb de- ‘say’ (though not marked as a verb). (91) a. [[OPdiye]] c,g = λp. λxe. λws.∀<w’, x’>:<w’, x’>∈LOGx,w, p(w’)(x’)=1 where LOG x,w = {<w’, x’>: w’ is compatible with x’s mental state in w and x’ is the individual in w’ that x identifies as himself} b. [[know]] c,g = λp. λxe.know(p)(x) 104 Diye clauses produce a truth value for the embedded proposition from the perspective of the matrix subject. By the time the derivation of these clauses is complete, a truth value for the embedded proposition has been produced from the perspective of the matrix subject. This suggests that question elements within the embedded proposition should not be possible, as questions have the inference that the truth of the proposition has not been established. This is borne out as shown in (92). Only –DIK clauses allow the relevant narrow scope reading (92a): (92) a. Semra [kim-in gel-diğ-in]-i bil-iyor S who-3GEN come-DIK-3S-ACC know-IMPF ‘Whoi does Semra know that s/hei came?’ ‘Semra knows who came’ b. Semra [kim gel-di diye] bil-iyor 52 S who-3GEN come-PST diye know-IMPF ‘Whoi does Semra know [diye s/hei came]?’ ‘*Semra knows [diye who came]’ 2.5. Conclusion This chapter investigated the role of perspective in –DIK and diye complement clauses in Turkish. It provided extensive data showing that all elements in diye clauses need to be reported from the matrix subject’s perspective. A logophoric operator encoding a perspective center that is the closest matrix subject was designated the job of achieving this result: the operator enforces all phrase level content to be evaluated in the logophoric alternatives of the perspective center by binding the world variables that come with these expressions. This proposal was used to account 52 See Chapter 3 for an explanation of how the wide scope of question elements is achieved in diye clauses. 105 for various syntactic patterns such as ban on NPI-licensing from the matrix clause and ban on association with the utterance context in the interpretation of quantifiers. The syntactic restrictions were tied to the opaque nature of diye clauses. The next chapter will focus on elements that need not be reported from the matrix subject’s perspective or refer to him/her. These elements, combined with the findings from this chapter, will bring forth the importance of the nature of the bound elements and give a complete picture of the interaction between the operator introduced by diye and the bound elements. 106 CHAPTER 3 WHEN THE SPEAKER IS INVOLVED Chapter 2 discussed the role of perspective in diye clauses and concluded that the operator introduced by diye, OPdiye, enforces all phrasal content in the embedded clause to be reported from the perspective of the matrix subject, thereby forming a single evaluation domain that syntactically corresponds to an opaque domain. This chapter aims to further investigate the properties of this opaque domain by investigating data that do not fit this generalization: (i) elements that are optionally associated with the matrix subject (hence the embedded context) and (ii) elements that are obligatorily associated with the utterance context. The analysis of these data reveals that the opaque nature of diye does not correspond to a phase, which are one kind of opaque domains, but to a domain that is opaque for some elements and transparent for others, i.e., selectively opaque domains (Keine 2019). This contributes to recent studies arguing for selectivity of opaque domains with respect to movement and/or agreement and case phenomena (e.g., Rizzi 2001, 2004, Abels 2012, Keine 2019, Poole 2020). I specifically argue that the optionally bound elements are silent pronominal elements that display specific discourse- dependent properties in that they are associated with topichood via an unvalued topic feature they are born with. Discourse–related properties have been shown to circumvent opaque domains such as islands (e.g., Jiménez Fernández 2009), with topics being specifically more able to circumvent islands (e.g., Rizzi 2004) and only D(iscourse)-linked elements being able to establish interactions non-locally (Cinque 1990, Rizzi 1990, 2001). I also explore the correlation between clause size and transparency based on Poole (2020), who, adapting the Williams Cycle 107 (2003), proposes that syntactic interactions such as case-assignment happen if the landing site of movement or licenser of case is in a higher position than the clause the target element resides in. If the probe is lower than the target in the functional sequence, the interaction cannot be applied. Applied to Turkish, this means that the topic position (TopicP) is higher than the full size of the diye clause, which is a PersP, so the pro inside the diye clause can be associated with topic probes outside of the diye clause. All others, e.g., negation, are below PersP in the functional sequence and are thus not predicted to interact on this account, which is borne out. By adopting an argumentation along these lines, this study also contributes to the studies arguing that discourse properties interact closely with the syntax of Turkish (see Şener 2010 for a detailed implementation of this stance). This chapter discusses two optional elements: pro associated with first-person agreement, and epithets, with the former constituting the main focus of the discussion. Pro refers to an attitude holder (not necessarily a perspective center) while epithets are evaluated by an attitude holder. The quirkiness of these data is that they are independent of the perspective center in diye clause. I argue that this dissociation is due to structural reasons briefly mentioned before, since there is a unifying common property, namely topichood (so it is not about whether they refer to or are evaluated by an individual, but rather it is about the position of the probe). The goal of this chapter is, therefore, to provide a unified and coherent view of the data in diye clauses. In other words, I provide a coherent description for both phrasal content (VP, DP, etc.), which is sensitive to locality, and topichood, which is not. Previous studies focused on one of these only (e.g., Özyıldız 2016, 2017 on the first, Şener & Şener 2011 on the second). This also applies to the general literature, which also focused on a single type of data (as discussed in Chapter 1). 108 Indeed, we see at the end of this chapter that such a unification is feasible from a syntactic point of view (although more challenging for the semantics, but see Schlenker 1999, Eckardt 2015). The discussion of the optionality of the first-person agreement marker in Turkish shows that the optional pattern should not be categorized as indexical shift phenomenon, unlike what was argued in previous research on the topic in Turkish (Şener & Şener 2011; Özyıldız 2012). I also compare the Turkish data with what Deal (2018, 2020) termed as ‘indexiphors’, a class of in-between elements -between indexicals and logophors-. Indexiphors are person agreement markers that can have variable reference but differ from indexical person markers in a number of ways. Although there is some evidence that the Turkish pro resembles indexiphors in some ways, I establish that Turkish pro is also not an indexiphor. Eventually, I argue that the first- person agreement marker in Turkish and the related pro are special due to their topichood. pro is a pronominal element with an unvalued [topic] feature, which participates in agreement with a topic probe. This aligns with findings about the topichood association of pro in Turkish in constructions other than complementation (e.g. Öztürk 1999, 2001). In other words, by resorting to an account of the specialness of pro in diye clauses based on topichood, I do not introduce new tools into the Turkish system, but rather I use already-existing tools. This is desirable, since we avoid structure-specific formulations. I also discuss epithets and show that they exhibit similar properties while also introducing some complications that must be left for future research. The epithets are interesting because, unlike pro, they not only refer to but are also evaluated by an individual. Yet, they are strikingly similar to pro in binding patterns (i.e., optionality). Thus, the comparison brings out some interesting parallels in addition to various complications about which I can only provide some preliminary thoughts. 109 This chapter also includes the discussion of data that cannot be bound by the diye operator, for example locative indexicals such as burada ‘here’, which must be bound in the utterance context. I attribute this to the lexical encoding of association with the utterance context, following the classical Kaplanian treatment (1989). The conclusion is that elements that are lexically specified for the utterance context are exempt from being bound by OPdiye. 53 Basically, these elements have a rigid (non-variable) denotations. Taken together, these data show that diye clauses are selectively opaque domains that require all point-of-view related phrasal content to be bound by OPdiye, a restriction that does not apply to topics, and indexicals cannot be bound at all. In this chapter, I divide the relevant data into several sections. In Section 3.1, I present the data on “indexical shifting” in diye clauses with discussion. Of indexical shift and indexiphors in general literature. In Section 3.2, data on the behavior of topics in diye clauses are presented. In Section 3.3, I present the proposal, along with a discussion of epithets as it supports the proposal in some ways. In Section 3.4, I discuss the elements that are obligatorily anchored to the utterance context. Section 3.5 presents the general discussion where I combine all these data in a coherent way and connect the selective opacity of diye clauses to the relevant literature. Section 3.6 concludes the chapter. 53 Ben ‘I’ can (must) shift in the complement of the verb de- ‘say’. This is unexpected if ben is lexically associated with the utterance context. The present conclusion that ben is lexically specified for utterance data holds for cognitive attitude verbs as embedding verbs. Perhaps, de- ‘say’ in Turkish introduces a context-overwriting operator that overwrites the utterance-context specification within the overt pronoun. If this is on the right track, it would corroborate my argument that de- ‘say’ and diye in Turkish are not the same element, contra what Major (2021) argues for Uyghur. The latter includes a binding operator whereas the former a context-overwriter. This remains a speculation at this point and further research is needed. I thank Isabelle Charnavel for pointing this out to me. 110 3.1. Optional Matrix Subject Orientation In this section, I discuss the pro associated with the first-person subject agreement marker on the verb in Turkish, and their relation to the literature on indexical shift. The first-person agreement data that I discuss in this section looks like indexical shift and indeed has been referred to as such (e.g., Şener & Şener 2011, Özyıldız 2012, Özyıldız et al. 2018). However, I show that the optionality in the first-person agreement marker in Turkish is not due to indexical shift. It resembles in some ways what Deal (2018) termed as indexiphors but differs from them too. To be able to see the differences and similarities Turkish pro has with indexical shift and indexiphors, I first discuss these phenomena with examples from the general literature (Section 3.1.1.) and then make a comparison with Turkish pro (Section 3.1.2.). Then, I discuss pro and overt pronouns in Turkish (Section 3.1.3.-3.1.4.). 3.1.1. Indexical shift and indexiphors in general literature The embedded subject I in (1) necessarily refers to the speaker and not to John: (1) John says that I am a hero. Kaplan (1989) represents indexicals as in (2) and argues that indexicals must be utterance- context related and cannot be shifted. In (2), the referents of the lexical items are determined solely by their characters, which is their relation to the context of utterance, which makes them rigid designators. A context in this model c is a tuple of the form <individual, individual, time, 111 location, world>, where the first two individuals refer to the speaker and the hearer (time is the time of speech, location is location of speech and world is the world of speech): (2) a. [[I]] c,i,g =Author(c) b. [[here]] c,i,g =Location(c) It is clear by now, however, that this is not universally true. The first-person marker in some languages need not refer to the speaker in embedded contexts but it can also refer to the matrix subject when embedded under an attitude verb. First person pronouns in Zazaki, for example, can refer to the speaker or the matrix subject, as in (3) (Anand & Nevins 2004): (3) Hɛsen-ij (mɨk-ra) va kɛ ɛzj/k dɛwletia H-OBL I.OBL-to said that I rich.be-PRES ‘Hesen said that I am rich’ (ISG pronoun=S(peaker), MS (Matrix Subject=Hesen)) Amharic is another example where first-person agreement can refer to the speaker or the matrix subject, as shown in (4) (Schlenker 2003: 68): 54 (4) John [pro dʒägna nä-ññ] yɨ-l-all J hero COP-1S.SUB 3SM.S-say-AUX.3SM.S ‘John says that I am a hero’ (ISG pro=S, MS) 54 First person indexicals are not the only indexicals that shift. Second person, temporal and locative indexicals can also shift. The interested reader is referred to Anand (2006) and Deal (2020). Here, I focus on first-person agreement in Turkish, so I stick to the discussion of person indexicals in other languages unless needed for contrast. 112 Schlenker (1999) terms cases such as (4) as indexical shift cases, since the reference of the indexical element in the embedded clause is determined by the context of reported speech rather than the utterance context, shifting the evaluation context from the utterance context to the attitude context. When the embedded indexical refers to the matrix subject, this is the shifted reading. If the referent is the speaker of the utterance context, this is the unshifted reading. One property of indexical shift is that the elements that so shift in a clause Shift Together. 55 (5) is an example from Slave that shows the Shift Together constraint (Anand & Nevins 2004). The only available readings are those in which the two indexical elements ɛz ‘I’ and to ‘you’ either shift together or neither shifts, first and second readings, respectively. The readings where one of them refers to the speaker and one to the attitude context participant are not allowed, namely the third and fourth readings. This indicates that the relevant pronouns are either both interpreted relative to the utterance context or the attitude context: (5) Vɨzeri Rojda Bill-ra va kɛ ɛz to-ra miradiʃa yesterday R B-to said that I you-to angry-be-PRES. ‘Yesterday, Rojda said to Bill, “I am angry at you”’ ‘Yesterday, Rojda said to Bill, “AUTH(c) is angry at ADDR(c)”’ ‘*Yesterday, Rojda said to Bill, “AUTH(c) am angry at you”’ ‘*Yesterday Rojda said to Bill, “I am angry at ADDR(c)”’ Anand (2006) states the relevant generalization as follows (p. 100): (6) SHIFT TOGETHER Constraint All shiftable indexicals within an attitude-context domain must pick up reference from the same context. 55 Shift Together as a constraint for indexical shift has been questioned recently. I discuss this point in this chapter. 113 This requirement is often attributed to the availability of a context-shifting operator. 56 What the context-shifting operator does is to rewrite the context parameter with the index parameter, where the index has a parallel definition to the context. The overwriting is represented in (7) (simplified from Anand (2006), along the lines of Deal (2020)). Contextual values being altered, all indexicals in the altered context shift, namely they all shift together. This is how the context- over-writing approach accounts for Shift Together. (7) [[OP α]] c,i,g =[[α]] i,i,g It has become increasingly clear though that the Shift Together constraint does not always hold. For example, Sundaresan & McFadden (2019: 4, ex. (6)) cites (8) as an exception to Shift Together in indexical shift in Tamil because ennæ ‘me’ stays unshifted in the presence of the first-person agreement that diagnoses a silent shifted first-person indexical in the same CP. (8) Ramani [CP taan{i,*j} enn-æ paartt-een-nnŭ] ottŭɖɳ-aan R self-NOM me-ACC saw-1S-C admitted-3MSUB ‘Lit: Raman admitted [CP that self had seen me].’ Reading 1: ‘Ramani admitted that he{i,*j} had seen mespeaker’ Reading 2: ‘*Ramani admitted that he{i,*j} had seen mei’ There have been different approaches to solutions depending on which data a given study is focused on. Some researchers tried to account for exceptions in various ways while still 56 Other analyses are also available, but the shifty-operator analysis is most relevant to the present proposal. See Schlenker (2003) for an account that attributes the patterns to the pro-forms themselves rather than an operator. The present study aligns with Schklenker (2003) in attributing the differences to the forms themselves. 114 considering them as undergoing indexical shift (e.g., Sundaresan & McFadden 2019). I do not adopt this approach, as I argue that the Turkish data are not indexical shift. Others have pointed out that some elements that might look like undergoing indexical shift are not real indexical shift-y elements (Deal 2018). One such element in indexiphors (Deal 2018). An indexiphor displays agreement like an indexical as shown in (4). However, it also displays patterns such as De Re Blocking and Locality, neither of which canonical indexicals obey. In this sense, they are more like logophors. De Re Blocking is defined in (9) (Deal 2018: 60; adapted from Anand 2006). Anand’s De Re Blocking hypothesis follows from Rule H, proposed by Fox (2000: 111) as a principle of the Binding Theory, as in (10): (9) De Re Blocking Every bound de se element must be de re free. (10) Rule H: A variable, x, cannot be bound by antecedent, α, in cases where a more local antecedent, β, could bind x and yield the same semantic interpretation. An example of De Re Blocking effects come from dream-reports in English, where the only possible reading is the one where the dream-self c-commands the actual self (11a). As reference to actual self is a case of de re, we conclude that the appearance of a de re element in a c- commanding position to a de se element in the same clausal domain (in this case, the dream self) leads to an unacceptable reading. These data were originally discussed in George Lakoff (1972) and later discussed in more detail in Percus & Sauerland (2003) and Anand (2006): (11) a. I dreamt that I was Brigitte Bardot and IB.B kissed meG.L b. #I dreamt that I was Brigitte Bardot and IG.L kissed meB.B 115 Anand (2006) discusses De Re Blocking in Yoruba. The pronoun o can be read de re in Yoruba and therefore blocks co-reference of òun with the matrix subject, as shown in (12b) (Adesola 2005). This is similar to a case where a de re element blocks a de se element. This effect washes away when the intervening pronoun is not de re (12c). Possession phrases do not act as interveners due to lack of c-command (12d): (12) a. Adéi so pé ouni ti r’ɨ ‘ɨwé réi/j A say that oun PERF see book o-GEN ‘Adei said that hei has seen hisi/j book’ b. Olui so pé o *i/j r’ɨ bàbá òuni O say that o see father oun-GEN ‘Olui said that he *i/j has seen hisi father’ c. Olui so pé oi r’ɨ bàbá òuni O say that o see father oun-GEN ‘Olui said that hei has seen hisi father’ d. Olui so pé bàbá réi/j ti r’ɨ ‘ɨyá òuni O say that father o-GEN see mother oun-GEN ‘Olui said that hisi/j father has seen hisi mother’ Back to Amharic. As we have seen in (4), repeated in (13), Amharic has what looks like an indexical (Schlenker 1999, 2003; Anand 2006; Deal 2018). The first-person agreement and the related pro can refer to the speaker or the matrix subject: (13) John [pro dʒägna nä-ññ] yɨ-l-all J hero COP-1S.SUB 3SM.S-say-AUX.3SM.SUB 116 ‘John says that I am a hero’ (ISG pro=S, MS) However, when there are two such elements in a clause, one being the subject and one being the object, the only reading is one where the subject refers to the matrix subject and the object to the speaker (Anand & Nevins 2004, Deal 2018). When the subject refers to the speaker, the sentence becomes unacceptable due to De Re Blocking: (14) John [prosubj pro obj al-ɨttazzəzə-ññ] alə J [ NEG.1S-obey-MKIMPERF-1.S.OBJ say-PERF.3.S.M ‘Johni says hei will not obey me’ ‘#Johni says I will not obey himi’ Anand (2006) argues that the de se element, namely the agreement marker, need to be bound by a logophoric operator and the de re element acts as an intervener for this binding relation, which correctly rules out the unattested reading in (14). More evidence that this is a blocking effect comes from cases where the subject functions as a de re element but does not c-command the de se indexical, as shown in (15). Such subjects do not trigger a blocking effect due to lack of c- command, which is proof that the established relation is binding: (15) John [[pro lɨj-e]subj proobj al-ɨttazzəzə-ññ] alə J [ son-my NEG.3S-obey-MKIMPERF-1.S.OBJ say-PERF.3.S.M ‘Johni says hisi son will not obey himi’ ‘Johni says hisi son will not obey me’ ‘Johni says my son will not obey himi’ ‘Johni says my son will not obey me’ 117 Another difference is that indexical shift is not clause-bound. Consider the Korean example in (16), in which first-person na can refer to John, which is two clauses up (Park 2016, Deal 2018). (16) John-i Seoul-eyse [CP Bill-i yeki-eyse [CP Mary-ka na-lul J-NOM S-at B-NOM here-at M-NOM I-ACC cohahanta]-ko malhayssta]-ko malhayssta like-C said-C said ‘Johnj said in Seoul that Bill said {here/in Seoul} that Mary likes himj’ (lit. Johnj said in Seoul that Bill said here that Mary likes mej) Indexiphors are clause-bound, as shown in (17) for Amharic (Anand 2006). The grammatical reading in (17) is one where both pros have the speaker as their antecedent. However, neither of the pros can refer to Mary, which is two clauses up. 57 In Deal (2018, 2020)’s analysis, the unavailable readings are unavailable because (i) the element intended to refer to Mary (who is the higher subject) is an indexiphor while the other pro is an indexical and (ii) indexiphors cannot have an antecedent two clauses up. If both elements were indexicals, this pattern would be unexpected, as indexicals are not clause-bound and should be able to have antecedents in any clause higher up. (17) Mary [CP John [CP [DP pro lɨj-e]subj proobj ay-ɨttazzəzə-ññ] M J son-1S NEG-3SUB-obey.IMPF-1.S.OBJ 57 We already know that these pros can refer to John, the lower subject, as the available readings in (15) show. 118 yɨSəllɨg-all aləCC think.IMPF-3S. say.PERF.3.SF ‘Mary said John thinks my son will not obey me’ ‘#Marym said John thinks herm son will not obey me’ ‘#Marym said John thinks my son will not obey herm’ In conclusion, indexical shift shifts all elements (of the same category, Anand (2006)) together, at least in some languages (cf. Sundaresan & McFadden 2019) and it is not clause-bound. Indexiphors are subject to De Re Blocking and locality: they must find their antecedents in the closest superordinate domain. Table 1 summarizes the data: Table 1: Indexicals versus Indexiphors Indexicals Indexiphors Shift Together Yes, but not always No De Re Blocking No Yes Locality No Yes In the next section, I explore the Turkish data with regards to the Shift Together, De Re Blocking effects and sensitivity to locality. If the target elements do not obey Shift Together but show De Re Blocking and locality effects, it will be concluded they are like indexiphors. If this is the case, it indicates that the operation involved is a binding operation rather than context-over-writing. If they do not obey both De Re Blocking and locality but tend to shift together, I take the elements to undergo indexical shift, in which case the operator in diye must be a context-overwriter. The conclusion for Turkish is not simple, as the Turkish data shares similarities with both. I will 119 argue that Turkish data is neither an example of indexical shifting nor of indexiphors. Rather, they have an independent, language-internal explanation: topichood. 3.1.2. First-person agreement in Turkish: Indexical shift or indexiphors? Turkish first-person agreement marker (and the pro licensed by it) does not have to refer to the speaker in diye clauses but can also refer to the matrix subject (18b), which seems to suggest that we are dealing with an indexical shift phenomenon (Şener & Şener 2011; Özyıldız 2012). –DIK clauses require the first-person agreement and the associated pro to refer to the speaker, possibly due to the lack of a perspective operator in these clauses (i.e., in the embedded context) (18a): 58 (18) a. Esra [pro sınıf-ta kal-dığ-ım]-ı bil-iyor E class-LOC stay-DIK-1S-ACC know-IMPF ‘Esra knows [that I failed the class]’ (1SG pro=S (Speaker), #MS (Matrix Subject)) b. Esra [pro sınıf-ta kal-dı-m diye] bil-iyor E class-LOC stay-PST-1S diye know-IMPF ‘Esra knows [diye I failed the class]’ (1SG pro=S (Speaker), MS (Matrix Subject)) Now, consider the example in (19), from Özyıldız (2012): 58 I do not discuss –DIK clauses further in this chapter, as they are strictly speaker-oriented in this aspect due to the closest operator being the one related to the speaker (see Section 3, this chapter, for representation of the speaker). 120 (19) Tunç Ayşe-ye [pro sen-i nere-ye götür-eceğ-im] de-miş 59 T A-DAT you-ACC where-DAT take-FUT-1S say-INDPST ‘Where did Tunç say to Ayşe that I would take you?’ ‘Where did Tunçi say to Ayşej that hei would take herj?’ ‘Where did Tunçi say to Ayşe that hei would take you?’ ‘??Where did Tunç say to Ayşej that I would take herj?’ We see here that a third reading is available, namely one where embedded first-person and second-person indexicals take their reference from different contexts (utterance and attitude context, respectively). Thus, Shift Together is not required, casting doubt on this data being a case of indexical shift. I discuss the unavailability of the fourth reading later. 60 Shift Together is not obeyed with cognition verbs, either, as in (20). In (20), there are two possibly shifty elements: prosubj (licensed by first person agreement) and proposs, both licensed by person agreement markers in their domain. The readings where both are reported from the utterance or the attitude context are accepted, namely the first and the second readings in (20), as expected by Shift Together. However, the reading where the subject pro refers to the matrix subject and the possessor pro refers to the speaker, the third reading, is also acceptable, which goes against Shift Together. Lastly, the reading where the subject pro refers to the speaker and the possessor pro to the matrix subject, the fourth reading, is not available, as expected: 59 The examples use the question where to control for the example not being direct quotation. Also notice that I use Özyıldız’s example as it is. There are some differences: Özyıldız uses a speech verb as embedding verb while I focus on cognition verbs. This is also the first time we see an overt pronoun versus an agreement marker, which I have not discussed before (but the distinction of overt versus covert pronouns will be important later in this chapter). 60 Özyıldız attributes this to covert movement of elements such that the unshifted forms (the object) covertly move above the context-rewriting operator, keeping the shifty-operator account intact for Turkish. The unavailable reading is unavailable due to a similar explanation to ban on Shift Together. 121 (20) Murat [prosubj [proposs kardeş-im]-i tanı-yor-um diye] bil-iyor 61 M sibling-1S-ACC know-IMPF-1S diye know-IMPF ‘Murat knows [diye I know my sibling]’ ‘Murati knows [diye hei knows his sibling]’ ‘?Murati knows [diye hei knows my sibling]’ ‘#Murati knows [diye I know hisi sibling]’ This might lead us to conclude that the phenomenon of optional reference by person markers in Turkish is not indexical shift. However, disobedience to Shift Together has been argued not to hold cross-linguistically (Sundaresan & McFadden 2019), and therefore is not a strong argument against indexical shift. Thus, I investigate De Re Blocking and clause-boundedness in the interpretations of examples like (19) and (20). There is some evidence that De Re Blocking is observed by this construction, as shown below (cf. (19) and (20)). Notice that in the case of the unavailable reading, the fourth reading, there is a c-commanding de re element blocking the embedded pronominal direct object (21) or the pro that comes with the first-person agreement inside the direct object (22) from referring to a matrix attitude holder: (21) Tunç Ayşe-ye [pro sen-i nere-ye götür-eceğ-im] de-miş T A-DAT you-ACC where-DAT take-FUT-1S say-INDPST ‘Where did Tunç say to Ayşe that I would take you?’ ‘Where did Tunçi say to Ayşej that hei would take herj?’ ‘Where did Tunçi say to Ayşe that hei would take you?’ 61 Notice that I use an overt pronoun in object position in (19) to compare with a pro subject while I use two pro elements in (20). This is due to three reasons. First is that (19) is directly from the examples in Özyıldız, so I did not change his example. Second, speech verbs and cognitive verbs differ in some ways. Generally, speech verbs allow shifting of both pro and overt pronouns while cognitive attitude verbs allow shifting of only pro (see Chapter 5). I discuss the distribution of pro and overt pronouns in Turkish later in this section. For now, it suffices to assume that they are not different realizations of the same element. 122 ‘??Where did Tunç say to Ayşej that I would take herj?’ (22) Murat [prosubj [proposs kardeş-im]-i tanı-yor-um diye] bil-iyor M sibling-1S-ACC know-IMPF-1S diye know-IMPF ‘Murat knows [diye I know my sibling]’ ‘Murati knows [diye hei knows his sibling]’ ‘?Murati knows [diye hei knows my sibling]’ ‘#Murati knows [diye I know hisi sibling]’ If we assume a standard analysis of subject pro-drop, in which a silent pro sits in SpecTP, subject pro c-commands kardeşimi ‘my sibling-ACC’. This accounts for De Re Blocking. Another set of data that shows the De Re Blocking effects, albeit in a weaker way, is in (23). The overt pronominal ben ‘I’ in the embedded clause blocks reference of the pro to the matrix subject (23b), which is possible in the absence of ben (23a). This may not be the strongest argument for De Re Blocking, as ben and pro both have the same referent/theta role. However, I will later argue that ben and pro are not directly related (via movement or agreement) in the Turkish syntax. Thus, it is worth mentioning here: (23) a. Seda [pro hayalet gör-dü-m diye] bil-iyor S ghost see-PST-1S diye know-IMPF ‘Seda knows [diye I saw a ghost]’ (1SG pro =S, MS) b. Seda [ben hayalet gör-dü-m diye] bil-iyor S I ghost see-PST-1S diye know-IMPF ‘Seda knows [diye I saw a ghost]’ 123 (1SG pro =S, *MS) Regarding (23), how do we make sure that the lower element, i.e., pro associated with first- person agreement, is indeed read de se? When pro refers to the matrix subject, it must be read de se. An example from Akkuş (2019: 10) is in (24), which shows that pro associated with the first- person agreement cannot refer to the matrix subject in a non-de se context. This shows that this element is read de se by default. (24) Context: Leyla took an exam, and later saw the top 10 scores with the scorer’s student ID numbers. She forgot her own ID number, so did not know who was who. Looking to the top score, she thinks: “This student definitely passed!”. But it turned out that she was that student. Leyla [pro sınav-ı geç-ti-m] san-ıyor L exam-ACC pass-PST-1S believe-IMPF ‘#Leyla believes that she passed the exam’ The generalization is that there should not be an intervening de re element in diye clauses if we want to associate pro with the matrix subject. Thus, pro is subject to De Re Blocking, like indexiphors. In fact, Akkuş (p. 19) proposes a similar generalization (25) based on similar facts in Turkish and other languages. The present study applies it to cognitive attitude verbs: 62 (25) NO UNSHIFTED INTERVENER: An indexical pronoun may only shift if there is no unshifted pronoun intervening between the indexical pronoun and the context operator in the CP domain. 62 Notice that Akkuş assumes context over-writing (i.e., shifting), unlike the present study, which argues for binding. 124 Though Turkish first-person agreement aligns (somewhat) with the behavior of indexiphors when it comes to De Re Blocking (though the evidence involves some potential confounds), it does not obey locality restrictions. The first-person agreement in the embedded clause in (26) can refer to the lowest matrix subject, John, and to the higher one, Esra. Thus, the relevant marker can refer to an antecedent two clauses up. (26) Esra [John pro sınav-ı geç-ti-m diye] bil-iyor diye E J exam-ACC pass-PST-1S diye know-IMPF diye hatırl-ıyor remember-IMPF ‘Esra remembers that John knows [diye I passed the exam]’ (1SG pro=Esra, John, speaker) Below is a summary of the results so far. I take evidence for De Re Blocking to exist but to be weak especially because of the vagueness in judgements for (22), though the tendence is in the right direction, and because of the confound in (23) that both the overt pronoun and pro refer to the same individual (though in my account, the overt pronoun -in SpecCP- structurally c- commands the pro -in SpecTP-). For now, my conclusion is that Turkish pro is neither indexical shifty nor an indexiphor. 63 I will argue that it is topichood that makes pro special in Turkish clauses, neither indexical shift nor indexiphoric nature. Table 2 summarizes the results. 63 But if De Re Blocking is proven to hold for this data, Turkish pro would be more similar to indexiphors. In that case, we would need an account of why locality is not obeyed in this structure. Topichood would come to the rescue again. So, if Turkish pro is indeed an indexiphor, my argument based on topichood would still hold as it is. 125 Table 2: Comparison of Indexical Shift, Indexiphors, Turkish 1 st Person Agreement Indexical Shift Indexiphors Turkish 1 st Person Agreement Shift Together Yes, but not always No No De Re Blocking No Yes Yes? Locality No Yes No 3.1.3. pro versus overt pronouns in Turkish (27) shows that pronominal subjects in Turkish can be overt as in (27a) or realized as pro as in (27b), suggesting that the overt realization of the pronoun is optional: (27) a. Ben gel-di-m I come-PST-1S ‘I came’ b. pro gel-di-m come-PST-1S ‘I came’ However, it has been argued that the choice of overt pronouns and pro is governed by discourse- related functions. The overt pronoun is realized if the subjects are contrasted (28), if the subject is questioned (29) (Erguvanlı-Taylan 1986, Öztürk 2001). It establishes new topics (30a, b versus 30c) (Öztürk 1999, 2001). pro is used when the subject is not the focus (30B’) and when the reference is to an already-established topic (30a, 30c): 126 (28) a. Ben gel-di-m. Ama sen gel-me-di-n I come-PST-1S But you come-NEG-PST-2S ‘I came But you did not come’ b. Ben gel-di-m. #Ama pro gel-me-di-n I come-PST-1S But come-NEG-PST-2S ‘I came But you did not come’ (29) A: Bu soru-yu kim sor-du this question-ACC who ask-PST ‘Who asked this question?’ B: Ben sor-du-m I ask-PST-1S ‘I asked’ B’: #pro sor-du-m ask-PST-1S ‘I asked’ (30) a. Beni ev-e gel-di-m. proi kitap oku-du-m. I home-DAT come-PST-1S book read-PST-1S proi televizyon seyret-ti-m. television watch-PST-1S ‘I came home, did some book-reading, watched TV.’ b. Beni ev-e gel-di-m. #Ben kitap oku-du-m. I home-DAT come-PST-1S I book read-PST-1S 127 #Ben televizyon seyret-ti-m. I television watch-PST-1S ‘I came home. I did some book-reading. I watched TV.’ c. Beni ev-e gel-di-m. proi kitap oku-du-m. I home-DAT come-PST-1S book read-PST-1S Sen televizyon seyret-ti-m. You television watch-PST-1S ‘I came home, did some book-reading. You watched TV.’ The above indicates that both overt pronouns and pro have discourse-related functions. Both are born with topic-related functions: overt pronouns establish new topics and pro continues an established topic (i.e., overt pronouns=new topic, pro=old topic). This is an important point to keep in mind as the analysis develops. The next section discusses a more syntactically-oriented piece of evidence that suggests that overt pronouns are not directly associated with verbal agreement, strengthening their discourse-related functions. 3.1.4. ECM in Turkish This section discusses ECM structures in Turkish, which suggest that overt pronouns are not directly related to person agreement after all (while pro is). The discussion here ties the topichood properties of overt pronouns and pro in Turkish with possible De Re Blocking effects. Although the evidence for De Re Blocking is somewhat weak (especially in the second set of data with overt pronouns and pro, because overt pronouns and pro have the same reference, so it 128 is unclear whether they actually block co-reference), I pursue some speculations in this section as to how the analysis would look like if De Re Blocking is proven to hold in Turkish data. An example of ECM in Turkish is given in (31). The embedded subject is realized in the accusative rather than the nominative, and the associated verbal agreement is optional. At face value, this can be taken as direct evidence that overt pronouns do not require subject agreement but rather are compatible with it and if so, they can be dissociated from agreement: 64 (31) a. Ben sen-i git-ti(-n) san-ıyor-du-m I you-ACC go-PST(-2S) think-IMPF-PST-1S ‘I thought you were gone’ b. Ben sen-i git-ti(-n) diye bil-iyor-du-m I you-ACC go-PST(-2S) diye know-IMPF-PST-1S ‘I thought (lit. knew diye) you were gone’ I would argue, however, that this evidence does not work in the way it is intended but further investigation still indirectly supports the hypothesis that overt pronouns are not directly associated with verbal agreement (while pro is). Let us first establish that nominative overt pronouns are indeed generated in the embedded clause, as opposed to accusative overt pronouns, which are generated in the matrix clause, indicating that the two subjects should be dissociated; i.e., one is not derived from the other via movement, contra Zidani-Eroğlu (1997). To that end, more detailed examples of ECM structures in Turkish are given in (32). When the embedded subject is in the accusative, as in (32a), the agreement on the embedded verb is optional. When it 64 Only finite complements or diye complements allow ECM in Turkish, but not –DIK clauses. 129 is in the nominative, it requires verbal agreement (32b). (32b) is the case we discussed with respect to De Re Blocking. pro also requires verbal agreement (32c). (32) a. Ben sen-i git-ti(-n) diye bil-iyor-du-m I you-ACC go-PST(-2S) diye know-IMPF-PST-1S ‘I thought (lit. knew diye) you were gone’ b. Ben [sen git-ti*(-n) diye] bil-iyor-du-m I you go-PST-2S diye know-IMPF-PST-1S ‘I thought (lit. knew diye) you were gone’ c. Ben [pro git-ti*(-n) diye] bil-iyor-du-m I go-PST-2S diye know-IMPF-PST-1S ‘I thought (lit. knew diye) you were gone’ It was argued in Chapter 2 that accusative-marked embedded subjects are in fact generated in the matrix clause and tied to the embedded clause with a proleptic structure. The evidence was based on the lack of idiomatic reading (33b), lack of licensing by embedded negation (34a), and lack of de dicto readings (35a). In all these tests, the nominative-marked subject yields the opposite pattern (33c, 34c, 35b). These contrasts indicate that the nominative subject must be generated in the embedded clause, in addition to the fact that nominative subjects require agreement while accusative subjects are only compatible with it, as was shown in (32). (33) a. At-ı al-an Üsküdar-ı geç-ti horse-ACC take-REL Ü-ACC pass-PST ‘It is too late (lit. That who took the horse passed Uskudar)’ 130 b. #Ali at-ı al-an-ı Üsküdar-ı geç-ti diye bil-iyor A horse-ACC take-REL-ACC Ü-ACC pass-PST diye know-IMPF ‘Intended: Ali knows [diye it is too late]’ c. ?Ali [at-ı al-an Üsküdar-ı geç-ti diye] bil-iyor A horse-ACC take-REL Ü-ACC pass-PST diye know-IMPF ‘Ali knows [diye it is too late]’ (34) a. *Ahmet kimse-yi git-me-di diye bil-iyor 65 A no one-ACC go-NEG-PST diye know-IMPF ‘Intended: Ahmet knows of no one [diye they left]’ b. Ahmet kimse-yi git-ti diye bil-m-iyor A no one-ACC go-PST diye know-NEG-IMPF ‘Ahmet knows of no one [diye they left]’ c. Ahmet [kimse git-me-di diye] bil-iyor A no one go-NEG-PST diye know-IMPF ‘Ahmet knows [diye no one left]’ (35) a. #Esra hayalet-i git-ti diye bil-iyor E ghost-ACC go-PST diye know-IMPF ama hayalet diye bir sey yok but there is no such thing as a ghost ‘Esra knows of the ghost [diye it left] but there is no such thing as a ghost’ 65 With regards to accusative-marked NPI data such as in (34), there is variation among speakers, as Şener notes. It turns out that Şener’s informants and mine do not speak the same dialect in this sense. However, there was also variation within my informants too. This is the dominant pattern. 131 b. Esra [hayalet git-ti diye bil-iyor E ghost go-PST diye know-IMPF ama hayalet diye bir sey yok but there is no such thing as a ghost ‘Esra knows [diye the ghost left] but there is no such thing as a ghost’ If nominative lexical subjects are generated in the embedded clause, we might expect nominative pronominal subjects to also be generated in the embedded clause. I assume that indeed to be the case, although applying the tests in (33)-(35) is hard with a pronominal subject. Instead, let us focus on one particular property: nominative pronominal subjects require verbal subject agreement. That is compelling evidence that they must be in the embedded clause, probably in SpecTP. We also know from the previous section that overt pronouns have a discourse-pragmatic import, not a clear TP-related property. As a reminder, the examples in (36) indicate that overt nominal pronominal subjects are seemingly optional (as per discourse-pragmatic functions). (36) a. Ben gel-di-m I come-PST-1S ‘I came’ b. pro gel-di-m come-PST-1S ‘I came’ To combine the conflicting requirements, it can be suggested that overt pronouns do not sit nor originate in SpecTP, but instead are generated in SpecCP. This would mean that they are not 132 case-marked (nominative is null in Turkish, so it is hard to distinguish nominative marking from no marking). This would yield the structures in (37a), representing embedded clauses with overt pronoun subjects, and (37b) the ones with pro subjects. In the former, the overt pronoun, ben ‘I’ is base-generated in SpecCP. pro is in SpecTP in both (37a) and (37b), as it always requires agreement. This means that overt pronouns never directly agree with the verb, rather it is pro that does it. This would account for the De Re Blocking imposed by overt pronouns (if this is indeed proven to hold for Turkish with stronger evidence), since they c-command pro in the same structure in (37a). See the sentence in (38) as an example. The overt pronoun ben is in SpecCP, c-commanding the pro in SpecTP, both being in the same embedded clause. Here, it is the pro that agrees with the first-person agreement marker on the embedded verb. (37) a. [CP ben [TP pro [VP] [T +1S]] C] b. [CP [TP pro [VP] [T +1S]] C] (38) Seda [ PersP [CPben [TP pro [VP hayalet gör]-dü-m] C] diye] bil-iyor S I ghost see-PST-1S diye know-IMPF ‘Seda knows [diye I saw a ghost]’ (ISG pro=S, *MS) One might wonder whether it could be argued that ben ‘I’ moves from SpecTP to SpecCP. This would not work if we want to account for De Re Blocking effects. This explanation would not yield a c-commanding relation necessary to establish De Re Blocking, as the two elements would be connected via movement. I will sometimes refer to this c-commanding relation throughout the dissertation, as I believe that this is a possible route to take for Turkish (given that there is some evidence for De Re Blocking), however, I do not definitively argue for it. 133 3.2. Topichood in diye Clauses In this section, I discuss various data showing that topichood is a consistent factor in allowing interaction between the matrix clause and the embedded diye clause in that various types of topics yield the same exceptional pattern. This is important to show because later it is argued that what causes ‘exceptional’ or ‘optional’ behavior in elements within diye clauses is not due to the pronominal nature of pro, rather it is because pronominal elements are specified with a [topic] feature in Turkish. Section 3.2.1. discusses topichood of pro, building on data from the previous section. Section 3.2.2. discusses other data showing the same exceptional behavior to the generalization established in Chapter 2. Together, these sections show that topichood is the umbrella concept that captures the optional or exceptional data within diye clauses. It is useful to recall that diye clauses are opaque. Consider again the following examples from Chapter 2, (39b) exemplifies the impossibility of NPI-licensing; (40b) the lack of specific readings of quantifiers, and (41b) the lack of binding from the matrix clause. Recall furthermore that –DIK clauses lack these properties. These data indicate that interactions between the matrix clause and an element inside the diye clause are not allowed: (39) a. Ali [kimse-nin gel-diğ-in]-i bil-m-iyor A no.one-3GEN come-DIK-3S-ACC know-NEG-IMPF ‘Ali knows [that no one came]’ b. ??/#Ali [kimse gel-di diye] bil-m-iyor A no.one come-PST diye know-NEG-IMPF ‘Intended: Ali knows [diye no one came]’ 134 (40) a. Pelin [burada biri-nin hasta ol-duğ-un]-u bil-iyor P here someone-3GEN sick be-DIK-3SACC know-IMPF ‘Pelin knows [that someone is sick here]’ (specific) ‘Pelin knows [that there is someone sick here]’ (unspecific) b. Pelin [burada biri hasta diye] bil-iyor P here someone sick diye know-IMPF ‘#Pelin knows [diye someone is sick here]’ (specific) ‘Pelin knows [diye there is someone sick here]’ (unspecific) (41) a. Çocuklari [Ali-nin birbirlerin-ii eşleştir-diğ-in]-i bil-iyor children A-3GEN each.other-ACC match-DIK-3S-ACC know-IMPF ‘Childreni know [that Ali matched each otheri]’ b. *Çocuklari [Ali birbirlerin-ii eşleştir-di diye] bil-iyor children A each.other-ACC match-PST diye know-IMPF ‘Childreni know [diye Ali matched each otheri]’ 3.2.1. Topichood of pro Recall that pro refers to old topics, as opposed to overt pronouns, which establish new topics. In this section, this point is discussed in detail. First, recall that pro has been argued to be associated with established topics in Turkish (Öztürk 1999) with some examples from earlier: (42) a. Beni ev-e gel-di-m. proi kitap oku-du-m. I home-DAT come-PST-1S book read-PST-1S proi televizyon seyret-ti-m. television watch-PST-1S 135 ‘I came home, did some book-reading, watched TV.’ b. Beni ev-e gel-di-m. #Ben kitap oku-du-m. I home-DAT come-PST-1S I book read-PST-1S #Ben televizyon seyret-ti-m. I television watch-PST-1S ‘I came home. I did some book-reading. I watched TV.’ In (42a), the speaker establishes herself as the salient individual with the overt pronoun ben ‘I’, and pro picks up the reference of the established salient entity. Using ben in consecutive sentences is not acceptable (42b). Another example is (43). In Context 1, none of the three individuals in the context are highlighted while Ali is highlighted through Ahmet’s question in Context 2. In this case, the overt pronouns in the two parts of the adjunct clause refer to different individuals (any of them) in discourse, indicating they do not require an established salient entity. Pro must refer to the one salient individual in context (43b): (43) Context 1: Individuals in discourse: Alii, Muratj, Esrak. All are in the same room, as is Betul. There was some noise in the room. Ahmet walks in and asks what the noise was about. Betul says: Context 2: Individuals in discourse: Alii, Muratj, Esrak. All are in the same room, as is Betul. There was some noise in the room. Ahmet walks in and asks what the noise Ali was making was about. Betul says: a. Bilmem. Oi/j/k konuş-ur-ken oi/j/k gül-üyor-du I don’t know. S/he talk-AOR-while s/he laugh-IMPF-PST ‘While s/hei/j/k was talking, s/hei/j/k was laughing’ 136 (Context 1: ok if two pronouns do not have same reference; Context 2: #) 66 b. Bilmem. proi/*j/*k konuş-ur-ken proi/*j/*k gül-üyor-du I don’t know. talk-AOR-while laugh-IMPF-PST ‘I don’t know. While hei was talking, hei was laughing’ (Context 1: # unless Betul makes someone salient by pointing; Context 2: ok) The subject pro in diye clauses is similar in that pro picks up the individual who is highlighted in the context. 67 In (44a), the question is asking about Esra’s exam results, making Esra the salient individual. In (44b), pro refers to the speaker because it is the speaker’s exam results that is asked. In other words, pro refers to whoever is made salient in previous context in diye clauses: (44) Context: Esra and I both studied hard for the class. We both took the exam. a. A: Esrai’nın kendi sınav sonucundan haberi var mı? Does Esrai know her exam results? B: proi [pro sınıf-ta kal-dı-m diye] bil-iyor class-LOC stay-PST-1S diye know-IMPF ‘No. Shei knows [diye I failed the class]’ (ISG pro=#S , MS (Esra)) b. A: Esrai’nın senin sınav sonucundan haberi var mı? Does Esrai know your exam results? 66 This can be accepted if the two overt pronouns refer to people in the context other than Ali. But in that case, the connection of the response to the question is not tenable (i.e., why would the answer talk about different individuals if Ali was asked about?). 67 Presumably, the topichood association holds true for pro in any position, as this is a stable part of Turkish grammar. However, I focus only on subject pro here. 137 B: proi [pro sınıf-ta kal-dı-m diye] bil-iyor class-LOC stay-PST-1S diye know-IMPF ‘No. Shei knows [diye I failed the class]’ (ISG pro=S, #MS (Esra)) The conclusion is that pro has topic properties via saliency, referring to old topics. It shares the requirement of topichood with overt pronouns 68 , though the specific type of topic they are associated with is different (overt pronouns=new topic; pro=established, salient topic). Now we have all the tools we need to propose an internal structure for pro. Since pro does not refer on its own but rather picks up an individual to associate with, I assume it is a φP, 69 not a DP (hence, not a barrier for the complement NP to be associated with external probes). pro sits in the complement position of the φP. I suggest that pro is born with an unvalued and uninterpretable topic feature [udiscourse-anaphoric] that needs to be valued by a probe that carries an interpretable [discourse-anaphoric] feature: 68 Any overt pronoun: in any position, with any case marking. However, I focus here on subject overt pronouns for simplicity. Another example is below, with an accusative-marked overt pronoun that is a topic: Murat [sen-i proj ti tanı-yor-umj diye] bil-iyor M you-ACC know-IMPF-1S diye know-IMPF ‘Murat knows [diye I know my sibling]’ (ISG pro=S, *MS) 69 For Déchaine & Wiltschko (2002) φPs “simply spell out φ-features (p. 411)” and it is “a cover term for any intermediate functional projection that intervenes between N and D (p. 410)”. 138 (45) The discourse-anaphoric feature relates to a Topic probe, which will be detailed in the proposal. Suffice it to say for now that the probe for this feature is a topic-related projection that is specified for discourse-anaphoric topics, which are construed as referring to the current discourse topic (Neeleman et al. 2008). They refer to given/backgrounded information or familiarity topics (e.g., Frascarelli & Hinterhölzl (2007) and de Cat (2007)). Şener (2010) states that “pronominals are prototypical examples of DAs (p. 25)”, where DA stands for Discourse Anaphoric Elements. 3.2.2. Other topic-related exceptions As was mentioned in Footnote 32 in Chapter 2, diye allows overt scrambling of the elements inside it into the matrix clause. In (46b), the embedded object kitabı ‘the book’ appears at the edge of the matrix clause, ex-situ to its base-generation position within the embedded VP: (46) a. Alii ben [ti kitab-ı oku-du diye] bil-iyor-um A I book-ACC read-PST diye know-IMPF-1S ‘I know [diye Ali read the book]’ 139 b. Kitab-ıi Esra [Ali ti oku-du diye] bil-iyor book-ACC E A read-PST diye know-IMPF ‘Esra knows [diye Ali read the book]’ I refer to such topics as aboutness topics, which are different from the ones discussed earlier, which are discourse-anaphoric. Aboutness topics can be tested with the tell me about X test (Neeleman et al. 2008, a la Reinhart 1981). An example in Turkish is in (47a), from Şener (2010: 33, ex. 26). Such topics must be the leftmost element in the clause (cf. 47b): (47) Yeni arabandan bahsetsene biraz. Tell me a bit about your new car. a. Arabai-yı geçen hafta bir arkadaş-ım-dan ti al-dı-m car-ACC last week a friend-1S-ABL buy-PST-1S ‘I bought the car from a friend of mine last week’ b. #Geçen hafta arabai-yı bir arkadaş-ım-dan ti al-dı-m last week car-ACC a friend-1S-ABL buy-PST-1S ‘Intended: I bought the car from a friend of mine last week’ Another piece of evidence for the exceptional behavior of topic-related interactions in diye clauses comes from NPIs. Recall that NPIs in diye clauses are not licensed by matrix negation: (48) ??/# Ali [kimse gel-di diye] bil-m-iyor A no.one come-PST diye know-NEG-IMPF ‘Intended: Ali knows [diye no one came]’ 140 This is true with the de dicto reading of the embedded clause containing the NPI, as in (48). It is, however, possible to make the NPI subject salient, in which case, NPI-licensing is possible. The example in (49) is from Şener (2008: 13, ex. (27)). The NPI kimse is primed/questioned in the previous question via the subject biri(si) ‘someone’. This makes the NPI salient in the discourse and this saliency allows the NPI to be licensed by the matrix negation. The same exception applies in diye clauses, as in (50): (49) A: Biri(-si) mi gel-di san-dı-n? Someone Q come-PST think-PST-2S ‘Did you think that someone came?’ B: pro [kimse gel-di] san-ma-dı-m anyone come-PST think-NEG-PST-1S ‘I did not think anybody came’ (50) A: Ali biri(-si) mi gel-di diye bil-iyor? A someone Q come-PST diye know-IMPF ‘Does Ali know [diye someone came]?’’ B: Ali [kimse gel-di diye] bil-m-iyor A anyone come-PST diye know-NEG-IMPF ‘He does not know [diye anyone came]’ This wraps up various data that pertain to topichood, and the data overall show that topichood, regardless of its type, allows interaction between the matrix clause and an element inside the diye clause, breaking the otherwise attested generalization that diye clauses are opaque domains. The 141 question is why and how is topichood special. In the next section, I present the proposal of how this exceptional behavior can be modelled. 3.3. Proposal This section is divided into two sub-sections: Section 3.3.1. presents the proposal. Section 3.3.2. discusses an important implication of the proposal for opaque domains and the exceptional behavior of topics. Section 3.3.3. discusses epithets in light of the patterns observed with pro and its implications for the proposal. 3.3.1. The proposal Recall from Chapter 2 that diye has the structure in (51): (51) I proposed in Chapter 2 that PersP is an opaque domain. In this chapter, I revise this proposal slightly and argue that a diye clause is a selectively opaque domain (hence not a phase) in that it 142 is opaque for NPI-licensing, binding, specific interpretation of quantifiers but not opaque for topic-related interactions. Selective opacity is a term defined in Keine (2019: 13) as in (52): (52) Selective Opacity A syntactic domain ∆ is selectively opaque for α-extraction if ∆ prohibits α-extraction but allows β-extraction out of it, where α and β are different types of extraction. Sensitivity to different types of elements within (opaque) domains has been discussed earlier as well (e.g., Rizzi 2001, 2004). Topicalization is known to escape islands (e.g., Cinque 1990, Abels 2012). Both will be supported by the proposal below. Before I present this proposal, let us recall a crucial working assumption, namely that the speaker is syntactically represented in the matrix clause (Ross 1967, Speas & Tenny 2003, Stowell 2007, Zu 2018) and it is associated with another binding operator (OP matrix) (Heim 1991, Pearson 2013, Charnavel 2018b), which also has a PRO subject that is controlled by the speaker. Let us assume for simplicity that the relevant phrase is SpeakerP and it immediately embeds the matrix PersP. This is represented in (53): (53) 143 This is part of the left-periphery of the Turkish clause. The rest will be adapted from Şener (2010), whose main goal is to establish that all discourse-related interactions are syntactically represented and have syntactic effects in Turkish (and hence, that there is no scrambling in Turkish). He argues for the structure in (54). ForceP hosts operators like the relative clause operator. The C-/A-Topic in TopicP stands for Contrastive/Aboutness Topic. DaP1 and DaP2 are discourse-anaphoric elements, with the difference between them being one of contrast. IntP is the Interrogative Phrase, which hosts the question probe. FocusP the focused elements: (54) I assume that both TopicP and DaP1 are topic-related projections. TopicP in Şener’s analysis stays the same in the present analysis. It is for aboutness (and contrastive) topic. 70 I encode this 70 I do not distinguish between contrastive and aboutness topics here, since such a distinction does not play a role in the present study. I use aboutness as a term that covers both. 144 via the feature [aboutness]. I also label DaP1 as TopicP but specify it with the feature [discourse- anaphoric]. The latter is the probe for pro and the former for overt pronouns and other ex-situ elements that move from the base-generated position within the diye clause for topic reasons (i.e., DPs that are aboutness and contrastive topics, which will be discussed shortly). Given what was proposed above, the structure in (55) constitutes the left periphery of matrix clauses in Turkish: 71 (55) 72,73 71 I ignore focus-related projections in Şener (2010) as they are not immediately relevant for the present study. 72 I assume that pro does not move since I do not have overt evidence for it, so the two TopicPs have different behaviors in my proposal: The one that encodes [aboutness] agrees and forces overt movement while the one that encodes [discouse-anaphoric] only agrees. 73 Note that I reversed the order of the two topic projections from Şener’s study to fit my need to have the SpeakerP high enough in the clause (following suggestions on the syntactic representation of the speaker, e.g., Zu 2018) but also under TopicP {discourse-anaphoricity}. 145 How does pro end up referring to the speaker? I assume that TopicP {discourse-anaphoricity} enters into multiple agreement relations. Both the speaker in SpeakerP and pro come with the same unvalued feature, in which case the same probe can agree with all of them. In fact, this is the assumption in Şener (2010), who states (p. 216): “In other words, Agree by a single Probe with multiple Goals for exactly the same features can take place without a locality problem”. This is represented as in (56). Notice that pro is bound by a non-local binder, i.e., the speaker instead of the matrix subject. (56) 146 It was shown earlier that pro (and the first-person agreement) in Turkish does not impose locality restrictions. I argue that the difference in locality is because of the [udiscourse-anaphoric] feature that pro is born with in Turkish. In other words, feature [udiscourse-anaphoric] allows Turkish pro to search for a non-local binder. For completeness’ sake, when the binder is the closest matrix subject, i.e., the local subject, the derivation is as in (57): (57) 147 What happens when there are different types of topics in the same sentence? The two topic phrases, with the feature [aboutness] and [discourse-anaphoric], target different elements, as it is possible to have a pro as well as an element that carries the aboutness topic feature. An example is given in (58), where sınavı ‘exam-ACC’ is the aboutness topic and pro still refers to the established salient entity (old topic), i.e., the speaker or the matrix subject (i.e., discourse- anaphoricity): 74 (58) a. A: Esrai’nın kendi sınav sonucundan haberi var mı? Does Esrai know her exam results? B: Sınav-ı Esra [pro geç-ti-m diye] bil-iyor exam-ACC E pass-PST-1S diye know-IMPF ‘Shei knows [diye I failed the class]’ (ISG pro=#S, MS (Esra)) b. A: Esrai’nın senin sınav sonucundan haberi var mı? Does Esrai know your exam results? B: Sınav-ı Esra [pro geç-ti-m diye] bil-iyor exam-ACC E pass-PST-1S diye know-IMPF ‘Shei knows [diye I failed the class]’ (ISG pro=S, #MS (Esra)) 74 This also indirectly and vaguely supports the left periphery I adopted for Turkish, as aboutness topic can lead to overt movement while discourse-anaphoric topic does not. In (58), the latter has a clearer connection to the discourse than the former and only the former has overt syntactic result (i.e., overt movement). 148 We have also seen that overt pronouns, which are also targeted by the same probe as aboutness topics, can occur in the same clause as pro, targeted by the TopicP with the [discourse- anaphoric] specification (59a). In fact, I have argued that they do occur in the same clause, the proposed structure is repeated in (59b): (59) a. Seda [ben pro hayalet gör-dü-m diye] bil-iyor S I ghost see-PST-1S diye know-IMPF ‘Seda knows [diye I saw a ghost]’ (ISG pro=S, *MS) b. [CP ben [TP pro [VP] [T +1S]] C] One concern about (59) is that the referents of both topic phrases are the same individual, namely, the speaker, while the two topic phrases targeted different individuals/entities in (58). However, it is possible to have a sentence where the overt pronoun and pro do not co-refer, as given in (60). Here, seni is the new or contrasted topic and while pro is the old topic. Thus, the case in (59) is accidental co-reference where both TopicPs happen to pick up the same individual in discourse: (60) Murat [sen-i pro ti tanı-yor-um diye] bil-iyor M you-ACC know-IMPF-1S diye know-IMPF ‘Murat knows [diye I know my sibling]’ (ISG pro=S, *MS) 149 Before ending this section, I eliminate one potential alternative analysis: Can we attribute the association of pro with the speaker to movement? Namely, can the association be achieved via movement of the full diye clause to the matrix clause, as was argued for de re tense interpretation by Stowell (2007)? 75 I argue that we cannot. One reason is that diye clauses can simultaneously contain pro that is associated with the speaker and another element that is associated with the matrix subject. The other is that de re reading of nominals is not allowed when there is a pro. Let us briefly review Stowell’s approach to de re and de dicto readings. Consider the following sentences that have a relative clause inside the complement clause: (61) a. Terri thought [that a woman [who lent me her car] was in your office]. b. Sam claimed last week [that Max gave me some money to a panhandler [who was sitting on the sidewalk (this afternoon)]]. In both cases, de re interpretation of the past marker in the relative clause is achieved when the tense interpretation in the relative clause is independent, i.e., relative clause event time is subsequent to the main clause event time. De dicto interpretation is when the relative clause tense is interpreted as dependent, i.e., it is either past-shifted or simultaneous relative to the event time of the matrix clause. Stowell attributes the de re reading to the position of the relative clause outside of the CP complement (hence in the matrix clause. If it stays inside the complement CP, the de dicto interpretation arises. To see how this explanation works for Turkish diye, I use both pro and elements that need to be reported from the matrix subject’s perspective, namely certain phrasal content in the 75 This discussion is not directly relevant for the association of pro inside a diye clause with higher matrix subjects. However, it will still show that diye clauses do not move as a whole to the main clause (unless the whole clause is topicalized) and I will assume that they do not move to intermediate clauses either. 150 same clause, drawn from Chapter 2. If the diye clause (presumably the CP, not the PersP) moves to the matrix clause so the optional element can get bound by the speaker, then we would expect all other elements in the clause to be reported from the speaker’s perspective. Yet, we see example (62), where pro refers to the speaker but the temporal adverbial is reported from the perspective of the matrix subject. This shows that the diye clause cannot move as a whole, since the temporal adverbial needs to be bound by the diye operator to be associated with the attitude context (i.e., bound by the diye operator): (62) Context: John believes that the speaker walked between 4-5 pm while the speaker in fact walked between 5-6 pm. John [{pro}UC {4-5 arası}AC yürü-dü-m diye] bil-iyor 76 J 4-5 between walk-PST-1S diye know-IMPF ‘John knows [diye IUC walked {between 4-5}AC]’ Lastly, de re elements are not allowed even in the presence of pro in diye clauses, as in the case of the nominal başkan ‘president’ in (66). If the whole diye clause moved to the matrix clause, we would expect de re elements to be allowed. (63) Murat is the president of a student club and he visited the school my roommate John and I work at. John knows Murat by name (perhaps as a friend) but he does not know that Murat is the president. John also knows Murat visited the school with me. #John [pro başkan-la okul-a gel-di-m diye] J president-COMM school-DAT come-PST diye 76 UC stands for utterance context and AC is attitude context in this example. 151 bil-iyor aslında know-IMPF actually ‘John knows [diye I came to the school with the president] actually’ cf. John [pro Murat-la okul-a gel-di-m diye] J Murat-COMM school-DAT come-PST diye bil-iyor aslında know-IMPF actually ‘John knows [diye I came to the school with Murat] actually’ I conclude that the diye clause does not move to the matrix clause. The next section discusses the implications of the proposal for the literature on selective opacity and the exceptional behavior of topics in various languages. 3.3.2. Selective opacity and topicalization Keine (2019) defines selective opacity as follows: (64) Selective Opacity A syntactic domain ∆ is selectively opaque for α-extraction if ∆ prohibits α-extraction but allows β-extraction out of it, where α and β are different types of extraction. 152 The best studied type of selective opacity is the distinction between A and A’ movement. For example, a finite clause allows A’-extraction (65a) but not A-extraction (65b): (65) a. Whoi do you think [CP t1 eats oatmeal for breakfast]? b. *John/Whoi seems [CP t1 eats oatmeal for breakfast]? Rizzi (2004) discussed various A’ elements and how they interact with each other in allowing movement across them, eventually proposing the following categorization of such elements: 77 (66) a. Argumental: person, number, gender, case b. Quantificational: Wh, Neg, measure, focus... c. Modifier: evaluative, epistemic, Neg, frequentative, celerative, measure, manner, .... d. Topic This classification divides A’ elements into three classes: quantificational (66b), modifier (66c) and topic (66d). (66a) are traditional phi-features. Each of these classes have a common feature that define the category. Each element within a category, therefore, acts as an intervener for the purposes of Relativized Minimality (Rizzi 1990), defined in (67): (67) Y is in a Minimal Configuration (MC) with X iff there is no Z such that (i) Z is of the same structural type as X, and (ii) Z intervenes between X and Y 77 Abels (2012) considers the quantificational and modifier categories as related, where the latter can block the former but not vice versa. Argumental and Topicalization are an independent category there as well. 153 The important point I focus on is (67i), which was defined as in (68) in Rizzi (1990) and redefined as in (69) in Rizzi (2004): (68) “same structural type” = (i) head or Spec and, in the latter class, (ii) A or A’ (69) “same structural type” = Spec licensed by features of the same class in (67) The idea is that an element cannot be moved across an element of the ‘same structural type’, forming selective domains. For example, domain X can be sensitive to interactions between certain elements and not others. Members of the same group are interveners for each other. The classification finds support in various interactions in languages. Obenauer (1994) observed that DP-specifier alone can be wh-moved in French, as in the movement of combien ‘how many’ in (70), but the movement is subject to locality conditions (71). The movement of the whole DP is not restricted in the same way (71a): (70) a. {Combien de problème}i sais-tu résoudre ti? ‘How many of problems can you solve?’ b. Combieni sais-tu résoudre [ ti de problèmes]? ‘How many can you solve of problems?’ (71) a. ?{Combien de problème}i sais-tu comment résoudre ti? ‘How many of problems do you know how to solve?’ b. *Combieni sais-tu comment résoudre [ti de problèmes]? ‘How many do you know how to solve of problems?’ 154 The locality restriction is based on the availability of another intervening clause in (71). However, negation (72b) and quantificational adverbs (73b) also act as interveners, as expected from the classification in (66b). Non-quantificational adverbs do not act as interveners (74) (Rizzi 2004): (72) a. {Combien de problèmes}i ne sais-tu pas résoudre ti ? ‘How many of problems can’t you solve?’ b. *Combieni ne sais-tu pas résoudre [ti de problèmes]? ‘How many can’t you solve of problems?’ (73) a. {Combien de livres}i a-t-il beaucoup consultés ti? ‘How many of books has he a lot consulted?’ b. *Combieni a-t-il beaucoup consulté ti de livres? ‘How many has he a lot consulted of books?’ (74) a. {Combien de livres}i a-t-il attentivement consultés ti ? ‘How many of books did he carefully consult?’ b. Combieni a-t-il attentivement consulté [ti de livres] ? ‘How many did he carefully consult of books?’ Likewise, Italian adverb rapidemente ‘rapidly’ can normally be extracted (75b) but not across an adverb of the same class (75c) (Rizzi 2004): (75) a. I tecnici hanno (probabilmente) risolto rapidamente il problema ‘The technicians have probably resolved rapidly the problem’ 155 b. Rapidamentei, i tecnici hanno risolto ti il problema ‘Rapidly, the technicians have resolved the problem’ c. *Rapidamentei, i tecnici hanno probabilmente risolto ti il problema ‘Rapidly, the technicians have probably resolved the problem’ d. Probabilmentei, I tecnici hanno ti risolto rapidamente il problema ‘Probably, the technicians rapidly resolved the problem’ If the adverb is focused, it can move across probabilmente ‘probably’ (76a), but not if there is intervening negation (76b), as focus and negation are both quantificational and intervene with each other’s movement: (76) a. RAPIDAMENTE i tecnici hanno probabilmente risolto il problema (non lentamente) ‘RAPIDLY the technicians have probably solved the problem (not slowly)’ b. RAPIDAMENTE i tecnici (*non) hanno risolto il problema ‘RAPIDLY the technicians have (not) solved the problem’ So far, data support Rizzi’s classification in (66) and the proposal that members of one group are interveners for each other. On the assumption that the classification in (66) is a form of selective opacity, this can be taken as evidence that opaque domains are feature-sensitive and selective opacity applies in a more specific way than the difference in A versus A’-movement. Indeed, Keine (2019) applies it for long-distance agreement in Hindi and provides a more detailed understanding of different types of interactions and their different behavior. The goal of Keine (2019) is to demonstrate that opaque domains can be feature-dependent and one opaque domain may not be opaque for all interactions, based on agreement data as well as movement data. He 156 states (p. 15): “Interestingly, selective opacity poses a challenge to standard approaches to syntactic locality, which have historically tended to treat locality as a binary distinction: a given domain is either transparent or opaque, without sensitivity to the type of extraction”. This is not to say phases do not exist (Keine 2019). Rather, there are different types of opaque domains, some of which can be defined based on certain features and classes of elements. Keine (2019) shifts the locus of opacity to the probe rather than the goal to account for the patterns in focus. I argue that diye clauses are a kind of selectively opaque domain (so, not a phasal domain) by focusing on the types of interactions it allows. To be able to make that explicit, it is important to understand two points: the special status of topics and the role of clause size. Topics have often been argued to be special in that they do not act as interveners and generally may not obey domain restrictions. For example, Rizzi (2004) showed that the adverb rapidemente ‘rapidly’ can undergo topic-movement across a topic as in (77a). In this sense, topics do not act as interveners for each other although they share a common feature according to (66). A similar example can be constructed with English, as in (77b), where the matrix clause topic moves from its embedded clause position across an embedded topic: (77) a. Rapidamente, penso che, questo problema, non lo possiate risolvere ‘Rapidly, I think that, this problem, you could not solve it b. {The book}i, John thinks [that {to Mary}k, Daniel gave ti tk] We have seen that topics are special in Turkish diye clauses in a similar way. For one, they are the elements that break the otherwise-established generalization that diye clauses are opaque domains. Second, two types of topics discussed, i.e., aboutness topics and discourse-anaphoric 157 topics, do not intervene with each other. What I argue now is that this special status derives from the height of the topic probe, following Poole (2020). Poole (2020) argues that selective opacity is not restricted to agreement (or movement), rather it is also observed in case dependencies. In a simple transitive case in Finnish, the external argument (subject) bears the nominative case and the internal argument (object) bears the accusative (78a). If the object is passivized (78b), if the object appears in an imperative (78c) or if the subject bears lexical Case, the object bears the nominative Case (78d): (78) a. Pekka osti kirja-n P.NOM bought book-ACC ‘Pekka bought the/a book’ b. Kirja oste-ttiin book.NOM buy-PASS.PST ‘The book was bought’ c. Osta kirja! buy.IMP book.NOM ‘Buy the/a book’ d. Minu-n täytyy osta-a kirja I-GEN need buy-INF/TA book.NOM ‘I have to buy the book’ In Finnish, adjunct clauses also enter into case calculation. If the subject is marked with the nominative, the object and the adjunct both bear the accusative (79a). If the subject bears lexical 158 Case, the object bears the nominative and the adjunct the accusative (79b) ((85b): Maling 1993:57): (79) a. Liisa muisti matka-n vuode-n L.NOM remembered trip-ACC year-ACC ‘Liisa remembered the trip for a year’ b. Liisa-n täytyy muista-a matka vuode-n L-GEN need remember-INF/TA trip.NOM year-ACC ‘Liisa has to remember the trip for a year’ Finnish has MA-infinitives, where the heading verb bears a case marker assigned by the verb (e.g., illative in (80a)) (Koskinen 1998:329). They are structurally TPs. The crucial point is that the embedded object enters into case calculation with the matrix subject. When the latter is in the nominative, the embedded object is accusative-marked (80b). When the latter is non-existent, the embedded object is marked with the nominative (80c): (80) a. Minä autoin Jukka-a [TP kirjoitta-ma-an Marja-lle] I.NOM helped Jukka-PTV write-INF/MA-ILL Marja-ALL ‘I helped Jukka [to write to Marja]’ b. Hän kävi [TP avaa-ma-ssa ove-n] s/he.NOM went open-INF/MA-INE door-ACC ‘S/he went to open the door’ c. Käy [TP avaa-ma-ssa ovi] go.IMP open-INF/MA-INE door.NOM ‘Go open the door’ 159 Unlike (80), adjunct clauses do not enter into case calculation with MA-clauses. We see this especially when the subject is absent. When there is no adjunct clause, the matrix object and the embedded object are both in the nominative (81a) (Vainikka 1989:268). This shows that matrix object affects case assignment. When there is an adjunct clause, this clause does not affect the nominative assignment on the embedded object (81b) (Maling 1993:66). The adjunct clause itself bears the accusative as a result of case-dependency on the matrix subject: (81) a. Pyydä Jukka [TP luke-ma-an kirja] ask.IMP J.NOM read-INF/MA-ILL book.NOM ‘Ask Jukka to read the book’ b. Pyydä Jukka [kolmanne-n kerra-n] [TP luke-ma-an kirja] ask.IMP J.NOM third-ACC time-ACC read-INF/MA-ILL book.NOM ‘Ask Jukka for the third time to read the book’ Poole (2020) summarizes these patterns in a generalization as in (82) (p. 21) and proposes to account for this pattern using the Williams Cycle (WC) (Williams 2003), phrasing it as in (83): (82) Finnish Case Generalization In Finnish, a matrix subject can license dependent case across an embedded TP boundary, but a matrix object and a matrix adjunct cannot. (83) Williams Cycle (strong version) Within the current XP, a syntactic operation may not target an element across YP, where Y is higher than X in the functional sequence. 160 When applied to the Finnish case, “in terms of the WC: a DP in [Spec, TP] can license dependent case on another DP across a TP, because T>T, but a DP in [Spec, vP] or [Spec, VP] cannot do so because T>v and T>V (p. 26)”. The author applies a similar implementation to movement. This was the original data in Williams (2003). “The core analytical intuition behind the WC is that one and the same node is a barrier to some movement types, but not to others, and that this distinction correlates with the structural height of the landing site in the functional sequence (p. 1)”. According to this idea, CP is a barrier for movement to TP but not to CP. That is why A’- movement does not feed A-movement (84a) but the reverse pattern is possible (84b). Likewise, TP is a barrier for movement to vP but not to TP and CP. (84) a.*Alexi seems [CP ti ate the nattoo] b. Whati did Kyle expect [TP Alex to eat ti]? Let us go back to the Turkish case with this background in mind. Recall the structure of diye clauses, repeated in (85): (85) Recall also that diye clauses do not allow binding, NPI-licensing or specific reading of embedded quantifiers. The relevant examples are repeated below: 161 (86) *Çocuklari [Ali birbirlerin-ii eşleştir-di diye] bil-iyor children A each.other-ACC match-PST diye know-IMPF ‘Childreni know [diye Ali matched each otheri]’ (87) ??/# Ali [kimse gel-di diye] bil-m-iyor A no.one come-PST diye know-NEG-IMPF ‘Intended: Ali knows [diye no one came]’ (88) Pelin [burada biri hasta diye] bil-iyor P here someone sick diye know-IMPF ‘#Pelin knows [diye someone is sick here]’ (specific) ‘Pelin knows [diye there is someone sick here]’ (unspecific) The potential licenser of the NPI, i.e., matrix negation, is below AspP/TP, as evidenced by the order of the morphemes in (86) (i.e., negation is to the left of the tense-aspect marker, hence lower). I assume that the specific reading in (87) would have been achieved in the CP if it was possible. The antecedent for the anaphor in (88) is the matrix subject in SpecTP. Notice that all of these probes have a position at or below CP. We know by now that diye clauses are bigger than CPs in that they also have a PersP that takes a CP as its complement, as shown in (85). Based on WC, this predicts that elements that are below PersP in the hierarchy of phrases (e.g., VP, AspP/TP, CP) cannot interact with a PersP as probes. In other words, matrix VP, AspP/TP, CP cannot interact with the PersP in the embedded clause because PersP>VP, AspP/TP, CP hierarchically. This would account for why elements in VP, AspP/TP (e.g., negation, subjects) or CP (e.g., specific-read elements) in the matrix clause cannot come into interaction with PersP, i.e., diye clauses. 78 78 One could ask how the said interactions are possible in –DIK clauses, which I argued to be CP nominalizations (i.e., DP>CP). In this case, the elements in matrix Asp/TP (and even CP, presumably, since there is a higher 162 How about topics? I have extensively argued that pro has a topic-related feature as they associate with the salient individual in the discourse, namely the old topic. I have also showed that diye clauses allow overt topic movement with aboutness topics. If we adopt an explanation based on height, the prediction is that the topic positions must be higher than PersP (i.e., TopicP>PersP). Indeed, it is known that topics are the left-most elements in Turkish (Erguvanlı 1984, Şener 2010) (also see Rizzi 2004). This is what I assumed for Turkish, repeated in (89): (89) projection (i.e., DP) in –DIK clauses), should not be expected to interact with the –DIK clause but they do. This is because –DIK clauses do not form an opaque domain and the height-explanation holds for opaque domains only. 163 Then, the pattern in Turkish, i.e., that diye clauses are not opaque for topichood, follows from WC as adapted by Poole (2020) because they are higher than PersP and therefore can look into the diye clause. This is represented as in (90): (90) It is not only topics that are higher than PersP. The obligatory matrix scope of question elements in diye clauses can be explained through the height of IntP as well. Since IntP is higher than PersP, it can probe into it, resulting in a higher scope reading of question elements. The relevant example is in (91): 164 (91) Semra [kim gel-di diye] bil-iyor S who-3GEN come-PST diye know-IMPF ‘Whoi does Semra know [diye s/hei came]?’ ‘*Semra knows [diye who came]’ This analysis has some important implications. First, although the exact implementation of selective opacity differs (Rizzi 2004; Abels 2012, Keine 2019, Poole 2020), the intuition and the patterns behind these analyses remain the same: opaque domains are not opaque in a binary way. Rather, it may be feature-driven (Rizzi 2004) or height-driven (Keine 2019, Poole 2020). I consider the proposal in this chapter to be an addition to the conclusions from these studies: diye clauses are opaque for NPI-licesing, wide scope of quantifier elements and binding but they are transparent for topics. One difference of the present proposal is that it extends this explanation to various operations at the same time while previous studies focused on a specific type of operation (agreement and movement for Keine 2019 and case-licensing for Poole 2020). Notice that diye clauses are opaque for different types of interactions: licensing (negation), (covert) movement (specific readings) and binding (anaphors). This indicates that, as has been pointed out in Keine (2019) and Poole (2020), selective opacity has a much broader application than previously assumed and it encompasses different types of operations. Diye provides more direct evidence for this position, since one and the same domain shows restrictions on different types of operations while allowing others. The next section discusses epithets, which surprisingly show very similar behavior to pro but brings in different types of challenges. 165 3.3.3. Evaluation of epithets Another type of element that can be optionally associated with the attitude context and the utterance context is epithets. Epithets share with pro the property of (i) being subject to De Re Blocking and (ii) not being clause-bound. Additionally, epithets are also context-dependent, though in a different way than pro. These similarities are surprising given a major difference between pro and epithets: Unlike pro, which can optionally refer to the matrix subjects or the speaker, epithets can be evaluated by matrix subjects and the speaker (and in fact cannot refer to them, as was shown in Chapter 2). The difference is due to the fact that epithets already have an established referent in the discourse (i.e., they behave line names in that respect) while pro does not. In that, we could expect epithets to pattern with (some of) the data in Chapter 2 that are DPs and need to be bound within the embedded clause (i.e., entities). But this is not borne out. Since two different types of elements, one that refers to the perspective center and one that is evaluated by it show similar patterns, this eliminates a potential argument that the optional pattern is activated when one needs to refer to a perspective center. This is important because earlier research on this type of optionality only focused on optionality in pro, namely elements that refer. That is only part of the bigger picture. The present discussion indicates that there is a more generalized pattern that includes elements that refer to and elements that are evaluated by the perspective center and that unification is based on the height of the probe. Now let us see the similarities between pro and epithets. The first similarity is the optional pattern. Epithets can be evaluated either by the speaker or the matrix subject in diye clauses (92b) (In –DIK clauses, they are not accepted by all speakers when there is conflict with the speaker’s evaluation (95a)): 166 (92) Context 1: I am a teacher. I think my student Alii is very smart. His classmate Murat thinks Alii is an idiot. Murat also thinks Alii won the running competition. Context 2: I am a teacher. I think my student Alii is an idiot. His classmate Murat thinks Alii is very smart. Murat also thinks Alii won the running competition. a. Murat [o salaği-ın yarış-ı kazan-dığ-ın]-ı bil-iyor 79 M that idiot-3GEN competition-ACC win-DIK-3S-ACC know-IMPF ‘Murat knows [that that idiot won the competition]’ (Context 1: %, Context 2: ok) b. Murat [o salaki yarış-ı kazan-dı diye] bil-iyor M that idiot competition-ACC win-PST diye know-IMPF ‘Murat knows [diye that that idiot won the competition]’ (Context 1: ok, Context 2: ok) (92a) and (92b) both would be unacceptable in a non de dicto context where Murat is not aware of who ‘that idiot’ is. (93) is a case in point. Since the matrix subject is not aware who the idiot refers to in (93), s/he cannot be an evaluator for it in neither diye nor –DIK clauses. (93) Context: Murat knows that Alii won the race. Murat is not aware that Alii is an idiot. (Speaker knows both that Alii won the race and that Alii is an idiot). a. Murat [o salaği-ın yarış-ı kazan-dığ-ın]-ı bil-iyor M that idiot-3GEN competition-ACC win-DIK-3S-ACC know-IMPF ‘Murat knows [that that idiot won the competition]’ 79 The divergence between my informants was that the –DIK clause can perhaps be ok in Context 1 for some and not for others. The majority did not like it, however, saying that the speaker must agree with the evaluation. 167 (Evaluator: S, #MS) b. Murat [o salaki yarış-ı kazan-dı diye] bil-iyor M that idiot competition-ACC win-PST diye know-IMPF ‘Murat knows [diye that that idiot won the competition]’ (Evaluator: S, #MS) I argued that pro is in SpecTP, which is a higher position than an object like o salağ-ı ‘that idiot- ACC’, which is in the VP. If De Re Blocking holds for epithets, this predicts that o salağ-ı should be evaluated only by the speaker when the pro refers to the speaker, the speaker being the de re element. This is borne out (94a). When the reference of pro is the matrix subject, the epithet object can be evaluated both by the matrix subject and the speaker (94b). (94) Context 1: The speaker, but not Murat, saw Alii. Murat [pro o salaği-ı gör-dü-m diye] bil-iyor M that idiot-ACC see-PST-1S diye know-IMPF ‘Murat knows [diye pro saw that idioti]’ (Evaluator: S, #MS) Context 2: Murat, but not the speaker, saw Alii. Murat [pro o salaği-ı gör-dü-m diye] bil-iyor M that idiot-ACC see-PST-1S diye know-IMPF ‘Murat knows [diye pro saw that idioti]’ (Evaluator: S, MS) 168 The third similarity between the epithets and pro is the lack of locality effects. When there are multiple embeddings with diye clauses, epithets can be evaluated by a matrix subject two clauses up, i.e., there are no locality constraints: 80 (95) Context: I am a teacher. I and Esra think my student Alii is very smart. His classmate Murat thinks that Alii is an idiot and that Alii won the running competition: a. #Murat Esra-nın [o salaği-ın yarış-ı kazan-dığ-ın]-ı M E-3GEN that idiot-3GEN competition-ACC win-DIK-3S-ACC bil-diğ-in-i bil-iyor know-DIK-3S-ACC know-IMPF ‘Murat knows Esra knows [that that idiot won the competition]’ b. Murat Esra [o salaki yarış-ı kazan-dı diye] M E that idiot competition-ACC win-PST diye bil-iyor diye bil-iyor know-IMPF diye know-IMPF ‘Murat knows diye Esra knows [diye that idiot won the competition]’ In conclusion, epithets show optionality in that they can be evaluated by any of the matrix subjects or by the speaker. They show De Re Blocking effect but no locality effects, similar to pro. 81 This is summarised in Table 3. For ease of comparison, I add the evaluation of phrasal 80 –DIK clauses are bit harder to judge here, as my participants commented, primarily due to a stronger requirement of agreement with the speaker. 81 Assuming the weak evidence for De Re Blocking in pro to hold. 169 content from Chapter 2 so the difference between optional (i.e., associated either with matrix subjects or speaker) and obligatory elements (associated with lowest matrix subject) are clear. Table 3: pro and epithets Phrasal Content Turkish 1 st Person Agreement Epithets Optionality No No No De Re Blocking No Yes Yes Locality Yes No No This begs the question whether epithets are also discourse-dependent like pro, if they are similar in other ways just mentioned? The answer is yes, but with a different type of discourse dependency. It turns out that the discourse-relatedness of epithets derives from distance- specification via the pronominal/demonstrative contained within them. Epithets in Turkish must include a demonstrative, as shown in (96): (96) Murat [*(o/bu) salak yarış-ı kazan-dı diye] bil-iyor M (that/this) idiot competition-ACC win-PST diye know-IMPF ‘Murat knows [diye that *(that/this) idiot won the competition]’ I first investigate the pronominal/demonstrative contained within the epithet to highlight its important role in the interpretation of epithets. Turkish has the proximal ([+prox]) and distal ([+distal]) demonstratives as well as a middle one ([-prox]/[-distal]) (97), all measured according to the speaker’s stance by default and this distinction is observed under attitude verbs in diye clauses as well as –DIK clauses (98)-(99). Notice that the relevant nominal is not an epithet in (98) but it includes a pronominal/demonstrative: 170 (97) a. o 82 adam that (lit. it) man ‘that [+distal] man’ b. bu adam this man ‘this [+ prox] man’ c. şu adam that man ‘that [-prox]/[-distal] man’ (98) Context: Alii is far away from the speaker. a. John [o/#bu adami-ın kazan-dığ-ın]-ı bil-iyor J that/this man-3GEN win-DIK-3S-ACC know-IMPF ‘John knows that [that/#this man won]’ b. John [o/#bu adami kazan-dı diye] bil-iyor J that/this man win-PST diye know-IMPF ‘John knows [diye that/#this man won]’ (99) Context: Alii is near the speaker. a. John [#o/bu adami-ın kazan-dığ-ın]-ı bil-iyor J that/this man-3GEN win-DIK-3S-ACC know-IMPF ‘John knows that [#that/this man won]’ 82 The demonstrative for distal has the same overt form as the third person pronoun. 171 b. John [#o/bu adami kazan-dı diye] bil-iyor J that/this man win-PST diye know-IMPF ‘John knows [diye #that/this man won]’ When we specify the position of the individual in context (i.e., Ali) with respect to the matrix subject, however, some speakers report a difference between –DIK and diye clauses such that only the latter allows the reading where distance is evaluated from the matrix subject’s perspective: 83 (100) Context: Alii is near the speaker but far away from John. 84 a. #John [o adami-ın kazan-dığ-ın]-ı bil-iyor J that man-3GEN win-DIK-3S-ACC know-IMPF ‘John knows that [that man won]’ …ama adamiın yanımda olduğunun farkında değil …but he is not aware that the mani is by me b. John [o adami kazan-dı diye] bil-iyor J that man win-PST diye know-IMPF ‘John knows [diye that man won]’ …ama adamiın yanımda olduğunun farkında değil …but he is not aware that the mani is by me 83 Interestingly, when we reverse the situation such that the man is far away from the speaker and near John, both – DIK and diye clauses with bu adam ‘this man’ are not allowed. I do not know why, but it might be due to a difference in the demonstrative element. Perhaps bu ‘this’ has a stronger connotation in some way than o ‘that’. The fact that o is also a pronominal element (i.e., 3 rd person singular) might indicate that o is more pronominal than bu. Epithets mostly use o ‘that’. 84 These data produce slightly more variable judgments. A full-fledged analysis of how exactly are epithets context- dependent is left for future research. Based on the discussion in this section, I still assume that there is some kind of distance-dependency, which is related to context-dependency. 172 Now, let us look at epithets. Since epithets must include a demonstrative, the epithet in diye clauses shows the same type of optionality, as shown in (101)-(103): (101) Context: Alii is far away from the speaker. a. John [o/#bu salaği-ın kazan-dığ-ın]-ı bil-iyor J that/this idiot-3GEN win-DIK-3S-ACC know-IMPF ‘John knows that [that/#this man won]’ b. John [o/#bu salaki kazan-dı diye] bil-iyor J that/this idiot win-PST diye know-IMPF ‘John knows [diye that/#this idiot won]’ (102) Context: Alii is near the speaker. a. John [#o/bu salaği-ın kazan-dığ-ın]-ı bil-iyor J that/this idiot-3GEN win-DIK-3S-ACC know-IMPF ‘John knows that [#that/this idiot won]’ b. John [#o/bu salaki kazan-dı diye] bil-iyor J that/this idiot win-PST diye know-IMPF ‘John knows [diye #that/this idiot won]’ (103) Context: Alii is near the speaker but far away from John. a. #John [o salaği-ın kazan-dığ-ın]-ı bil-iyor J that idiot-3GEN win-DIK-3S-ACC know-IMPF ‘John knows that [that idiot won]’ 173 …ama Aliinin yanımda olduğunun farkında değil …but he is not aware that Alii is by me b. ?John [o salaki kazan-dı diye] bil-iyor J that idiot win-PST diye know-IMPF ‘John knows [diye that idiot won]’ …ama Aliinin yanımda olduğunun farkında değil …but he is not aware that Alii is by me This indicates that the demonstrative inside the epithet plays a crucial role in yielding the pattern observed because the demonstrative yields the same pattern regardless of the nominal head it is attached to. Interestingly, it is also the demonstrative element that causes (potential) De Re Blocking (recall that there is a confound in the following examples such that both the overt pronoun and the pro refer to the same individual, and yet there is a structural c-commanding relation between them). We can show this with the following example, where the overt pronoun acts as an intervener for evaluation from the matrix subject’s perspective if the DP (with an epithet or not) has a demonstrative (104a/c) but not if it does not (104b/d): (104) Context: Alii is near the speaker but far away from John. a. #John [ben o adami-ı gör-dü-m diye] bil-iyor J I that man-ACC see-PST-1S diye know-IMPF ‘John knows [diye I saw that man]’ …ama adamiın yanımda olduğunun farkında değil …but he is not aware that the mani is by me 174 b. John [ben adami-ı gör-dü-m diye] bil-iyor J I man-ACC see-PST-1S diye know-IMPF ‘John knows [diye I saw the man]’ …ama adamiın yanımda olduğunun farkında değil …but he is not aware that the mani is by me c. #John [ben o salaği-ı gör-dü-m diye] bil-iyor J I that idiot-ACC see-PST-1S diye know-IMPF ‘John knows [diye I saw that idiot]’ …ama Aliinin yanımda olduğunun farkında değil …but he is not aware that Alii is by me d. John [ben salaği-ı gör-dü-m diye] bil-iyor J I idiot-ACC see-PST-1S diye know-IMPF ‘John knows [diye I saw the idiot]’ …ama Aliinin yanımda olduğunun farkında değil …but he is not aware that Alii is by me So far, we have concluded that pro and epithets are both optional in similar ways; they are both context-dependent (although in different ways - but pro is dependent on salience and epithets on distance-) and the demonstrative/pronominal inside the epithet brings out the observed patterns. What is the structure of epithets? Since epithets already refer to someone in the discourse (see any example above, for example (104c))), I assume that they are DPs. But that is not all there is. They must be a special type of DP, as they include a demonstrative that is related to proximity. Since the demonstrative is sensitive to distance relations, I suggest that the DP in epithets 175 encodes deictic properties. I argue that (105) represents the internal structure of epithets adapted (simplified) from Patel-Grosz (2020), who follows Ihsane & Puskàs (2001) and Laenzlinger 2005). Epithets are a special type of DP, DP deix. DPdeix is a DP that hosts a demonstrative in Patel-Grosz (2020) and is more contextually dependent. In (105), following this, the demonstrative is in SpecDP deix and the specific contextual dependency is encoded as an unvalued feature on the head Ddeix. This feature is valued by the DP whose referent’s proximity is made salient in the context. The NP is the pronominal-like part (like in pro in NP) and the D (deixis) is the name-like part that encodes context-dependency: (105) Crucially, however, this account does not account for the fact that pro refers and epithets are evaluated. Therefore, it remains speculative. At best, I can attribute this to the difference in the internal structure of pro and epithets, specifically the former being a φP and the latter a DP. However, it is still unclear how this difference plays a role. I leave this for future research. Before closing this section, let me point out that the De Re Blocking effect observed in epithets, indicates that the relevant contextual dependency (i.e., distance) is not merely a pragmatic requirement. Rather, it has syntactic consequences. If it was pragmatic, we would not 176 expect intervention-like effects as to who can be the evaluator, but we do observe these effects, which indicates there are syntactic processes involved. 3.4. Elements that are Obligatorily Associated with Utterance Context There are some elements that cannot be associated with the matrix subject (i.e., embedded context) and must be associated with the speaker or hearer (i.e., utterance context) in diye clauses. We have already seen one of these elements, namely overt pronouns. Recall the example in (110), where the overt pronoun ben blocks the association of the first-person agreement marker (and the pro) with the matrix subject. That is because ben refers to the speaker, a coordinate of the utterance context. Likewise, the overt pronoun object in (111) must be interpreted as referring to the addressee of the utterance context. (106) a. Seda [pro hayalet gör-dü-m diye] bil-iyor S ghost see-PST-1S diye know-IMPF ‘Seda knows [diye I saw a ghost]’ (ISG pro=S, MS) b. Seda [ben pro hayalet gör-dü-m diye] bil-iyor S I ghost see-PST-1S diye know-IMPF ‘Seda knows [diye I saw a ghost]’ (ISG pro=S, *MS) (107) Murat [pro sen-i tanı-yor-um diye] bil-iyor M you-ACC know-IMPF-1S diye know-IMPF 177 ‘Murat knows [diye I know my you]’ (ISG pro=S, *MS; you=H) Another example is locative indexicals such as burada ‘here’, which must refer to the speaker’s location: 85 (108) Context 1: Mary is in Los Angeles. However, she thinks she is in Irvine. The speaker is in Los Angeles. Mary has an appointment in LA. Mary [toplantı burada/#orada ol-acak diye] bil-iyor M meeting here/there be-FUT diye know-IMPF ‘Mary knows [diye the meeting will be here/#there]’ Context 2: Mary is in Los Angeles. However, she thinks she is in Irvine. The speaker is in Irvine. Mary has an appointment in LA. Mary [toplantı #burada/orada ol-acak diye] bil-iyor M meeting here/there be-FUT diye know-IMPF ‘Mary knows [diye the meeting will be #here/there]’ 85 Interestingly, temporal elements are not as strict, as it can be associated with the matrix subject. (i) Today is Friday. Last week on Friday, John said Mary would leave the following day, i.e., on Saturday last week. Geçen Cuma John [Mary yarın gid-ecek last Friday J M tomorrow go-FUT diye] bil-iyor-du diye know-IMPF-PST ‘Intended: Last Friday, John knew [diye Mary would go the following day]’ 178 These elements can be analyzed along the lines of Kaplan (1989), Sundaresan & McFadden (2019), among others, and encode dependence on the utterance context in the lexical element itself. In other words, Group 3 elements are lexically specified to refer to the utterance context, therefore only the matrix operator can bind them. Because they are lexically encoded for the utterance context. Though I leave development of a semantic denotation that is in line with the present proposal for future work, Kaplan (1989)’s analysis of lexically utterance-context oriented elements needs to be mentioned. His analysis of such elements is as in (109): (109) a. [[here]] c,i,g =Location(c) b. [[now]] c,i,g =Time(c) How this can be formalized semantically in the present analysis, which uses a lambda-abstractor binding operator in diye clauses is a task for future research. However, an interesting observation is that attribution of the utterance context association in the elements in this section finds support in the fact that these elements are the only ones that must be single, overt words. In other words, they are overt lexical items. This differentiates them from the phrasal content that needs to be evaluated by the closest matrix subject discussed in Chapter 2; from pro, which is a covert element, and epithets, which are also phrasal. Thus, lexical encoding is possible because they are born with the said association in the lexicon. 3.5. Interim Summary and Discussion This section combines the proposals in Chapter 2 and 3 and discusses their combined implications. The proposals in these chapters can be combined in the following way: 179 (110) a. diye is the overt realization of a perspectival operator in the head of PersP, which takes a CP as complement. The operator reports the embedded content from the first-person perspective of the closest matrix subject. This makes diye the overt realization of a perspective operator. b. Phrasal content such as evaluative adjectives, phrasal temporal and lexical elements, among others, in the diye clause come with an anaphoric world variable that needs. to be bound by the diye operator in the immediate clause. This results in phrasal content (without a pronominal element) being reported from the first-person perspective of the lowest matrix subject. c. First-person agreement and the related pro (and epithets) are optionally bound by the speaker or any of the matrix subjects. I argued that the pronominal inside these elements are born with an unvalued, uninterpretable topic feature that keeps the binding domain unrestricted. This is because topic probes are higher than PersP in the clausal spine on the selective opacity account of Keine (2019) and Poole (2020). d. Some elements such as overt pronouns and locative expressions like burada ‘here’ must be associated with the utterance context. This is due to the lexical properties of these elements, following the traditional Kaplanian account. The fact that these elements are overt words and do not include phrases (unlike the phrasal content that needs to be evaluated by the closest matrix subject; pro, which is a covert element; and epithets, which are also phrasal) might be construed as supporting evidence for 180 the lexical encoding in these elements. More precisely, they are both phrasal and heads (cf. Chomsky on clitic pronouns). Taken together, these points point to the systematicity of diye clauses in Turkish. Diye consistently has the same structure, involving the same single operator in the same position. The differences derive from the bound elements. Anaphoric elements, which are all phrasal in syntax, come with a variable that needs to be bound in its immediate clause, in the attitude context. Optional elements come with a pronominal element carrying a topic feature that require that it be bound by the salient individual in the context and valued by a topic probe in syntax, which allows it the flexibility of either being bound in the utterance context or the attitude context depending on which one has a salient subject. The elements that are strictly associated with the utterance context, which are all single words in the lexicon, are lexically encoded for this association. This is summarized in Figure 1: Figure 1: 86 Correlation of (Non-)Optionality of Reporting Context and Bound-Element Property (AC: Attitude Context; UC: Utterance Context) By grouping elements in this way, I have utilized independent properties of the bound elements (e.g., their structural make-up, their requirements outside of diye clauses etc –e.g., pro requiring 86 This figure suggests a continuum, which is purposeful. It is possible that there are elements that fall in between these categories in Turkish. I want to leave that possibility open. I thank Isabelle Charnavel for pointing this out. Bound in AC Bound in AC or UC Bound in UC Phrasal Expressions pronominal Lexical Expressions 181 saliency outside of diye clauses and inside them) to account for their varying behavior in diye clauses, which is a desirable result as it leads to a unification and brings out the systematicity in the Turkish grammar. Let us remember examples from all three groups here for clarity, given in (111)-(113). The relevant elements are in italics. Group 1: Elements that must be bound by the operator in diye (111) Context: I know that Ali had a race between 2-3 pm. John thought his race was between 5-6 pm. Ali and John are both in Los Angeles. John [Ali {5-6 arası/#2-3 arası} yarış-tı diye] bil-iyor J A 5-6 between/2-3 between compete-PST diye know-IMPF ‘John knows [diye Ali competed between 5-6/#between 2-3]’ Group 2: Elements that can be bound by the operator in diye or the ones in higher clauses (112) Esra [pro sınıf-ta kal-dı-m diye] bil-iyor E class-LOC stay-PST-1S diye know-IMPF ‘Esra knows [diye I failed the class]’ (ISG pro=S, MS) Group 3: Elements that must be bound in the matrix clause (i.e., temporal deixis) (113) Context: Mary is in Los Angeles. However, she thinks she is in Irvine. The speaker is in Los Angeles. Mary has an appointment in LA. Mary [toplantı burada ol-acak diye] bil-iyor M meeting here be-FUT diye know-IMPF 182 ‘Mary knows [diye the meeting will be here]’ The three elements can be put together in combinations in diye clauses and they behave in their own unique way despite appearing under the same diye clause. The phrasal temporal expression 4-5 arası ‘between 4-5’ must be reported from the perspective of the matrix subject, which I subbed as AC (Attitude Context). pro can refer to the speaker when there is 4-5 arası ‘between 4-5’ evaluated in the attitude context in the same clause (114a). Reversely, burada ‘here’ cannot be reported from the matrix subject’s perspective. While it stays associated with the utterance context, the optional element pro can refer to the matrix subject (114b): (114) Context 1: John believes that the speaker walked between 4-5 pm while the speaker in fact walked between 5-6 pm. a. John [{pro}UC {4-5 arası}AC yürü-dü-m diye] bil-iyor J 4-5 between walk-PST-1S diye know-IMPF ‘John knows [diye IUC walked {between 4-5}AC]’ Context 2: Johni believes that hei walked here while hei in fact walked there. b. John [{pro}AC {burada}UC yürü-dü-m diye] bil-iyor J here walk-PST-1S diye know-IMPF ‘John knows [diye IAC walked {here}UC]’ Given that both clauses in (114) are diye clauses and the operator is constant, the differences must derive from the elements themselves. This account makes the present account similar to Schlenker (1999), as he also argued to account for differences based on lexical items. He proposes the following classification based on data from various languages: 183 Table 4: Schlenker’s Classification Indexical Devices Anaphoric Devices All-purpose Matrix Adverbs in two days the day after tomorrow two days later than iti Pronouns Amharic ‘I’ English ‘I’ English ‘he’ Tense Russian Present Tense English present tense English past tense [in SOT contexts] An indexical in Schlenker’s account is an element whose “denotation depends on the context of speech (p. 19)”, where context of speech refers to the tuple of the speaker, the hearer, the time and world of utterance. In Table 4, all purpose indexicals are those that “can be evaluated with respect to any context (actual or reported) (p. 26-27)” and matrix indexicals are those that “can only be evaluated with respect to the context of the actual speech act (p. 27)”. Anaphoric devices require some more discussion. For that, see the examples below (Schlenker 1999: 27, ex. (31)): (115) a. John has told me repeatedly over the years that he would give me my money back two days later b. *John has told me repeatedly over the years: ‘I’ll give you your money back two days later’ c. I met John a week ago. Two days later he gave me my money back. It is not possible to use two days later in a direct speech context to mean ‘the day after tomorrow’, as shown in (115b). It requires a salient antecedent in the discourse or speech situation (over the years in (115a) and a week ago in (115c)). Thus, two days later is anaphoric to the time in the matrix clause. Since later is a comparative, it has a concealed pronominal 184 argument: two days later than iti. Likewise, in the sentence John says that he is a hero, where he refers to John, this is an anaphoric device. One can notice the parallels between the present study and Schlenker (1999): (i) both attribute differences to the lexical items themselves, (ii) both have somewhat similar groups of elements: all purpose indexicals can be likened to the optional elements in the present study and matrix indexicals to lexically indexical elements. Anaphoric devices are somewhat different: in the present study, an anaphoric variable that comes with phrasal content needs to be bound within the immediate attitude context while for Schlenker, saliency is important for this category. There are some other differences, too. For example, for the present study, saliency is important for what is the equivalent of Schlenker’s all purpose indexicals. An important contribution is that the present study and Schlenker (1999) are both able to account for differences in syntactic/semantic behavior of different elements without positing different operators (e.g., Anand 2006, Deal 2020) or positing different evaluation domains (Charnavel 2019), thus keeping the intuition that diye is a single element that corresponds to one operator and the argument that they form an interpretationally complete domain that is syntactically selectively opaque. If diye clauses were made up of different perspectival domains, for example, we would have a harder time accounting for why they also behave as a single unit. The present study also finds systematicity in the types of bound elements that accounts for the grouping in Turkish (Figure 1). This insight is, I believe, an important contribution to understanding the roots of the patterns we find in language. Another similar approach is Eckardt (2005)’s analysis of Free Indirect Discourse. Though diye is not an example of Free Indirect Discourse but rather an overt grammatical marker of 185 perspective, they share the property of being ways of reporting perspective in different domains of grammar. (116) is an example, about which Eckardt states (117) (p.3): (116) The hair was curled, and the maid sent away, and Emma sat down to think and to be miserable. –It was a wretched business, indeed! Such an overthrow of every thing she had been wishing for. –Such a development of every thing most unwelcome!-Such a blow for Harriet!-That was the worst of all. Austen, Emma, p.103 (117) Free indirect discourse is characterized by the fact that pronouns and tense in the sentence, and other types of perspectivising elements do not single out the same person as the speaker. For instance, the pronoun in (121) (BE: her (5)) suggests that Emma is not the speaker of these sentences-she is refered to with third person pronouns-whereas all emotional and evaluative words and constructions in the passage are most naturally interpreted as showing Emma’s perspective. Eckardt proposes to evaluate sentences in Free Indirect Discourse relative to a pair of contexts <C,c>, where C is the narrator’s context and c is the protagonist’s context. Specific words can be specified where they opt for this type of context or just one of the two members, as “the analysis also predicts that shiftability is a lexical property of words, not of words-in-utterances. For instance, the word alas/leider has a shiftable meaning once and for all (p. 44)”. Both Schlenker and Eckardt stand out with their large coverage of data, as only then can we see the overall systematicity. For instance, Eckardt (2005: 46) states the following: (118) Large coverage: A meaningful linguistic analysis of free indirect discourse in 186 literary texts has to provide analyses for a wide range of phenomena. Only then can we see how different factors conspire to trigger interpretation in the free indirect discourse mood <C,c>. These factors include • Context-dependence of tense and aspect • Context-dependence of temporal and local adverbials • Context-dependence of emotive expressions, exclamatives, epistemic modals, discourse particles and other expressions that refer to the speaker and/or addressee of an utterance Admittedly, the differences between the present study and that of Schlenker and Eckardt are large: not only are they semantically focused, unlike the present study that focuses on the syntax of perspective, but the data covered are also somewhat different. What is common, however, is the large-coverage approach taken and the aim in taking this approach. For example, while Eckardt tried to unify various aspects of free indirect discourse by taking a broader look at data, I did so with various element types in diye clauses with the goal of understanding their syntactic properties and behavior. Therefore, I believe the present study retains the valuable aspects of the said studies, namely their comprehensive approach to data coverage, while utilizing these data to investigate the syntactic properties of the data. 3.6. Conclusion In this chapter, I discussed the correlation of optional binding with contextual dependency (e.g., through saliency) in diye clauses. I argued that the optional elements possess a pronominal element that needs to pick up the salient entity in discourse (old topic), achieved through topic 187 agreement. This makes diye a selectively opaque domain, opaque for various interactions between the matrix clause and the embedded clause such as NPI-licensing, binding, wide scope interpretation but transparent for topic-related interactions. This is evidence for studies arguing that opaque domains in syntax are not opaque in a binary fashion (Rizzi 2001, 2004, Abels 2012, Keine 2019) and there is a correlation with the size of the domain (Poole 2020). More specifically, diye clauses are transparent for topic-related functions because topics are higher than PersP, the highest projection in a diye clause. I also discussed elements that must be associated with the utterance context, which I argued lexically encode this association. I tied together data from Chapter 2 and discussed that the properties of the bound elements determine what kind of binding we get in diye clauses: anaphoric binding by diye with phrasal elements, pronominal binding by diye with pro and no binding by diye in lexical elements. My account brings out the systematicity in the Turkish grammar in a way that has not been observed in the Turkish or general literature, which is a significant contribution to the literature, which aims to bring out cross-linguistic systematicity. 188 CHAPTER 4 EMBEDDING VERB TYPES AND COMPLEMENTATION IN TURKISH This chapter shifts its focus from the embedded clause to the attitude verb that embeds them. The goal is to discover syntactic interactions between the embedding verb and their complement clause and to uncover what these interactions entail for the grammatical system of Turkish. This endeavor is valuable since the effect of the syntax of the embedding verb types has been brought up in various studies in relation to (non)factivity. The present study takes a more in-depth look into how the syntax relates to factivity distinctions in embedding verbs by their effects on their complement clauses. Recall that I define factivity based on perspective: (i) non-factive verbs= matrix subject-oriented verbs (e.g., düşün- ‘think’, hayal et- ‘imagine’), (ii) true factive verbs 87 = speaker-oriented verbs (e.g., unut- ‘forget’, keşfet- ‘discover’), (iii) factivity underspecified/semi- factive verbs=perspective underspecified verbs (e.g., bil- ‘know’, hatırla- ‘remember’). I sometimes use them interchangeably in this chapter. But when I use (non)factivity in the discussion of general literature and perspective in the discussion of Turkish data and analysis. Overall, the intuition here is that factivity presupposition arises as a result of structural encoding of the speaker’s perspective. The syntactic representation of the distinction for Turkish is in (1): 87 As opposed to default factivity, which is discussed in this chapter as well. 189 (1) a. Matrix Subject-oriented b. Speaker-oriented c. Perspective-Underspecified This categorization based on factivity is not new and has been known at least since Kiparsky & Kiparsky (1971) (K & K). However, evidence is provided in this chapter that this distinction has crucial syntactic consequences with regards to domainhood and various other implications of this classification will be pointed out in the course of the discussion. I argue that both speaker- 190 oriented (=true factive) and matrix subject-oriented (=non-factive) verbs select for a bigger complement, CPspeaker and CP matrixsubject, respectively, as opposed to factivity-underspecified (=perspective-underspecified) verbs, which combine directly with the complement (i.e. no CPspeaker or CP matrixsubject). CPspeaker/CP matrixsubject, in turn, combine with the complement clause. Both CPspeaker and CP matrixsubject host an operator defined in similar ways to OPdiye in that it reports the embedded content from the first-person perspective of an individual. The correlation between perspective and (non)factivity is as follows: The complement to matrix-subject oriented verbs encode the perspective of the matrix subject, which is where their non-factivity results from (i.e. not established in the actual world by the speaker). The complement to speaker-oriented verbs encode the speaker’s perspective, which is where the factivity results from. Their similarity is that they both have a perspective operator in their complement. Thus, (non)factivity and perspective are closely related terms and I have therefore adopted arguments from the literature on (non)factivity in discussion of perspective. Speaker-oriented and matrix subject-oriented verbs contrast with perspective-underspecified verbs, which are verbs that do not encode any perspective (and no factivity specification), do not combine with CPspeaker or CP matrixsubject but rather directly with the complement clause, and are thus structurally smaller. Another difference between CPspeaker and CP matrixsubject is as follows: the operator in CPspeaker has a PRO subject in its specifier (1b) whereas CP matrixsubject does not (1a), resulting in differences in the nature of their complement. This will prove to be an important factor in determining whether the complement clause is an opaque domain or not: the existence of PRO indicates an opaque domain. This proposal serves to combine the two sides in the debate on the structure of (non-)factivity, discussed below. Like more recent studies, it argues that non-factive (=matrix subject-oriented) predicates require extra structure in their complements. Like traditional studies, it argues that 191 factivity (=speaker-orientedness) is also encoded structurally in equally complex terms. It, however, points out to some important differences between them based on the Turkish data, which is the unique contribution of the present chapter. In other words, I argue that the difference between true factive (=speaker-oriented) and non-factive (=matrix subject-oriented) verbs is not the quantity of added structure but rather what is encoded in the added structure and the elements within the extra structure. I also show that there is a default factivity. Traditional studies on the structure of factivity (e.g., K & K) argue that factive predicates are headed by the noun fact (2a), which non-factives lack (2b) (updated to a more current form of tree structure in de Cuba (2007) as in (3)). In this account, the former is structurally more complex than the latter because factive predicates always select for an NP with CP complement as their own complement while non-factive predicates select for a CP complement directly. (2) a. b. (3) a. b. 192 Although this pattern of structural complexity has long been assumed and argued for (e.g., Zubizarreta 2001), some recent studies take the opposite stance. For example, Nichols (2001) provides a semantic analysis of factive and non-factive predicates and eventually argues that factivity is the default interpretation and non-factivity requires an operator because it has a special interpretation. She states that “…a plausible default interpretive strategy is the presupposition of the truth of a proposition unless it is marked otherwise (p. 204)” because “… the evaluation set containing the actual world can be considered the default evaluation context (p. 126)”. Haegeman (2006) takes up a more structural account and argues, like Nichols, that factivity is the default interpretation and therefore lacks any operator. In her account, factive predicates are default because “their content, not being asserted, or related to the speaker, is, as it were, ‘taken for granted’ (p. 1665)”. Non-factive predicates are characterized by the availability of a projection for speaker deixis (and a projection for topic and focus phrases), which anchors a proposition to the speaker. She states (p. 1665) that “the ‘factive’ interpretation of such clauses could be seen as default reading: their content, not being asserted, or related to the speaker, is, as it were, ‘taken for granted’.” Haegeman argues that this is what can account for, for example, why factive predicates do not allow speaker-oriented adverbs (4a) but non-factives do (4b). (4) a. *John regrets that Mary probably/obviously/unfortunately did not attend the meeting. b. John believes that Mary probably/obviously/unfortunately did not attend the meeting. de Cuba (2007) argues for the same point that non-factives are structurally more complex with evidence from a number of languages with V2. For example, Swedish allows V2 in embedded clauses only if the embedding verb is non-factive. In (5) and (6), the (a) examples represent the neutral word order. The (b) examples exemplify V2, where the auxiliary var appears to the left of 193 the negation inte. Notice that this inversion is possible only with sa ‘say’, a non-factive verb, but not with ångrade ‘regret’, a factive verb. (5) a. Rickard sa at than inte var hemma R said that he not was home b. Rickard sa att han var inte hemma R said that he was not home ‘Rickard said that he was not home’ (6) a. Rickard ångrade att than inte var hemma R regretted that he not was home b. *Rickard ångrade att han var inte hemma R regretted that he was not home ‘Rickard regretted that he was not home’ Based on data such as these, de Cuba proposes that non-factive predicates select for cP, extender of the embedded CP -which cP takes as its own complement-, whereas factive predicates select for the CP directly. The cP hosts an operator that eliminates the speaker as an evaluator for the embedded clause. This adds up to proposing that non-factives have a bigger structure than factives and that non-factives have an operator whereas factives lack it. Other studies such as Miyagawa (2012) argue for the same: that non-factives have a more complicated structure, but they also argued that factives still have an operator. In Miyagawa (2012), the extra structure in non-factives pertain to discourse participants and relations. Although the present study is focused on the syntax of (non)factivity (=perspective), one should be aware that there is a huge literature on the semantics of (non)factivity. To that end, I 194 briefly discuss some of the main points and questions. But it should be kept in mind that the present work is focused on their syntax and does not make definitive assertions regarding their semantics. To start, factive verbs, e.g., know, regret, discover (in English), are those that trigger the “presupposition by the speaker that the complement of the sentence expresses a true proposition (K & K: 43)”. Presuppositions have projective meaning, meaning that they survive under the scope of entailment-canceling operators (Simons et al. 2010). One such case is (7b), where the factivity inference survives negation, proving that it is a presupposition (>: presupposition). However, the factivity inference can be canceled in story-telling contexts (8) (cf. (2)): (7) a. John knows that Mary loves him. >Mary loves John. b. John doesn’t know that Mary loves him. >Mary loves John. (8) Then, if the observer can get notification from the owner, the observer knows that the owner has still stayed in the net. If not, the observer knows that the observer has already disappeared since the reset occurred. 88 Other factive verbs are even looser with respect to their factivity inference because their inference may not survive the conventional entailment-canceling environments. For instance, the conditional can cancel the inference in realize, discover (9b) but not that of regret (9a) (Karttunen 1974): (9) a. If I regret later that I have not told the truth, I will confess it to everyone. b. If I {realize, discover} later that I have not told the truth, I will confess it to everyone. 88 The lack of factivity inference in this example is not necessarily due to the existence of the conditional. As discussed later in this chapter, conditionals retain the factivity inference of know. 195 The discussion on factivity-cancelation is therefore not at a place to make definitive conclusions on how to group verbs based on factivity in English, specifically based on where factivity inference survives and where it does not. There seems to be a good amount of variation and flexibility, leading to different categorizations (e.g., soft triggers vs. hard triggers in Abusch 2002, 2010; pragmatic presupposition vs. semantic presupposition in Dahlman 2015; categorization of the verbs whose factivity inference can be canceled in entailment-canceling environments as the semi-factive verbs, whose nature is inherently complex in this aspect). That is one major reason why I opt to go for a more syntactic route to investigate whether syntax can help us tease these verbs apart more clearly. I am not the first one to do so, as the previous discussion showed. And indeed, at the end of this chapter, we will see that starting with syntactic distinctions might be more helpful in understanding the semantics of the said verbs. McCloskey (2005) is different from the studies discussed so far in that it most directly combines the semantics with the structure in this aspect (though he did not investigate (non)- factivity per se, but his points can be transferred to (non)-factivity). McCloskey investigates inversion and adjunction in embedded clauses under different kinds of predicates in the Irish dialect of English. He shows that both inversion and adjunction are possible if the embedding predicate is [+wh] (i.e., a question predicate), as in (10), and it is not if the embedding predicate is a ‘resolutive’ predicate, a group which includes what I call factive and factivity-underspecified predicates (11). (10) a. I wondered was he illiterate. b. ?He asked me when I got home if I would cook dinner. 196 (11) a. *The police discovered who had they beaten up. b.*The police established while we were out who had broken in to our apartment. He proposed to account for this difference by positing an extra structure in the embedded clause of [+wh] verbs. The higher CP with the C1 head in (12) is only available to [+wh] verbs: (12) The extra CP layer in (12) introduces a QUEST operator, which takes sets of propositions (in lower CP) and returns an interrogative speech act (in higher CP). This kind of operator can only occur with predicates that indicate that the issue (in the embedded clause) has not been resolved. This includes negation and questions. That is why a negated or questioned resolutive predicate allows inversion: (13) a. *I remember was Henry a Communist b. ?I don’t remember was Henry a communist? c. Do you remember was he a communist? 197 This chapter is organized as follows. Section 4.1 presents the core data. Section 4.2 presents the proposal and the analysis. Section 4.3 discusses the proposal, bringing together data from earlier chapters with the data in this chapter. Section 4.4 concludes the chapter. 4.1. Perspectival Patterns under Matrix Subject-Oriented Verbs In this section, various data sets with –DIK and diye clauses when they are embedded under a matrix subject-oriented verb are presented. The data sets in this chapter replicate the ones in Chapter 2 in that they test evaluation of various phrasal elements (including de re-de dicto distinction) as well as whether conscious evaluation of the embedded content is required, which was taken to be evidence for logophoric alternatives in Chapter 2. After presenting these data, data testing the opaque nature of diye clauses, namely, binding, NPI-licensing, and specific readings of quantifier elements, are presented. It is shown that (i) evaluative adjectives, phrasal temporal and locative expressions as well as nominal and verbal content in the embedded complement is evaluated from the matrix subject’s perspective under matrix subject-oriented verbs. On the other hand, neither –DIK clauses nor diye clauses allow de re elements under matrix subject-oriented verbs and they both impose a de dicto reading) (Section 4.1.1.), (ii) Matrix subject-oriented verbs require conscious evaluation of contents in their complements regardless of the complement type –i.e., in diye as well as –DIK complements (Section 4.1.2.). Section 4.1.3. provides a summary of the data supporting these conclusions. Section 4.1.4-4.1.6 show that anaphor binding across clauses, NPI-licensing across clauses and specific readings of quantifier elements are allowed in diye clauses as well as –DIK clauses under matrix subject- oriented verbs. Section 4.1.7. summarizes the findings of the last three sections. 198 4.1.1. Evaluation of evaluative adjectives, phrasal temporal and locative expressions Recall that diye clauses require that evaluative adjectives and adverbs be reported from the perspective of the matrix subject and not that of the speaker’s under perspective-underspecified verbs such as bil- ‘know’ (14) and that –DIK clauses cannot report either the speaker’s or matrix subject’s perspective under the same (Chapter 2, example 15): (14) Context: I know John competed badly. Ali believes, however, that John competed well. a. Ali [John-un #iyi/#kötü yarış-tığ-ın]-ı bil-iyor A J-3GEN good/bad compete-DIK-3S-ACC know-IMPF ‘Ali knows [that John competed #well/#bad]’ b. Ali [John iyi/#kötü yarış-tı diye] bil-iyor A J good/bad compete-PST diye know-IMPF ‘Ali knows [diye John competed well/#bad]’ When the embedding is the matrix subject-oriented verb düşün- ‘think’, as in (15), the matrix subject is the sole evaluator (i.e., the speaker cannot be an evaluator) for embedded content regardless of the complement type (cf. (15a) and (14a)): (15) Context: I know John competed badly. Ali believes, however, that John competed well. a. Ali [John-un iyi/#kötü yarış-tığ-ın]-ı düşün-üyor A J-3GEN good/bad compete-DIK-3S-ACC think-IMPF ‘Ali thinks [that John competed well/#bad]’ 199 b. Ali [John iyi/#kötü yarış-tı diye] düşün-üyor A J good/bad compete-PST diye think-IMPF ‘Ali thinks [diye John competed well/#bad]’ Temporal expressions like 5-6 arası/2-3 arası ‘between 5-6/between 2-3’ (16) and locative expressions like New York’ta/Los Angeles’ta ‘in New York/in Los Angeles’ (17) must be reported solely from the matrix subject’s perspective regardless of the complement: (16) Context: Ali had a race between 2-3 pm. John thought the race was between 5-6 pm. Ali and John are both in Los Angeles. a. John [Ali-nin 5-6 arası/#2-3 arası yarış-tığ-ın]-ı düşün-üyor J A-3GEN 5-6 between/2-3 between compete-DIK-3S-ACC think-IMPF ‘John thinks [that Ali competed between 5-6/#between 2-3]’ b. John [Ali 5-6 arası/#2-3 arası yarış-tı diye] düşün-üyor J A 5-6 between/2-3 between compete-PST diye think-IMPF ‘John thinks [diye Ali competed between 5-6/#between 2-3]’ (17) Context: Ali had a race in Los Angeles. John thought the race was in New York. a. John [Ali-nin New York-ta/#Los Angeles-ta yarış-tığ-ın]-ı J A-3GEN N Y-LOC L A-LOC compete-DIK-3S-ACC düşün-üyor think-IMPF ‘John thinks [that Ali competed in New York/#in Los Angeles]’ 200 b. John [Ali New York-ta/#Los Angeles-ta yarış-tı diye] düşün-üyor J A N Y-LOC/L A-LOC compete-PST diye think-IMPF ‘John knows [diye Ali competed in New York/#in Los Angeles]’ Likewise, nominal and verbal content are evaluated from the matrix subject’s perspective under düşün- ‘think’ no matter what the complement is. In (18), the nominal content kadın ‘woman’ is contrasted with adam ‘man’ and in (19), the verb yürü- ‘walk’ is contrasted with koş ‘run’: (18) Context: Ali and I were told that someone nice came from London today. I was told that this person is a man and Ali was told that the person is a woman. We each believe what we have been told. a. Ali [iyi bir kadın-ın/#adam-ın bugün Londra-dan A good one woman-3GEN/man-3GEN today London-ABL gel-diğ-in]-i düşün-üyor come-DIK-3S-ACC think-IMPF ‘Ali thinks [that a good woman/#man came from London today]’ b. Ali [iyi bir kadın/#adam bugün Londra-dan gel-di A good one woman/man today London-ABL come-PST diye] düşün-üyor diye think-IMPF ‘Ali thinks [that a good woman/#man came from London today]’ (19) Context: Ali had a running competition between 2-3 pm. John knew the race was between 2-3 pm but he thought it was a walking competition. Ali and John are both in 201 Los Angeles. a. John [Ali-nin 2-3 arası yürü-düğ-ün-ü/#koş-tuğ-un-u düşün-üyor J A-3GEN 2-3 between walk-DIK-3S-ACC/run-DIK-3S-ACC think-IMPF ‘John thinks [that Ali walked/#ran between 2-3]’ b. John [Ali 2-3 arası yürü-dü/#koş-tu diye] düşün-üyor J A 2-3 between walk-PST/run-PST diye think-IMPF ‘John thinks [diye Ali walked/#ran between 2-3]’ When multiple elements are combined under düşün ‘think’, the same pattern emerges such that all perspectival elements need to be reported from the perspective of the matrix subject regardless of the complement. (20) combines a temporal adverb 5-6 arası/2-3 arası ‘between 5- 6/between 2-3’, a locative adverb New York’ta/Los Angeles’ta ‘in New York/in Los Angeles’, and a manner adverb iyi/kötü ‘good/bad’, all of which must be reported from the perspective of the matrix subject in both –DIK and diye clauses: (20) Context: Ali had a race between 2-3 pm in Los Angeles. John thought his race was between 5-6 pm in New York. Ali and John are both in Los Angeles. John also thought Ali competed well and the speaker thinks Ali was bad. a. John [Ali-nin 5-6 arası/#2-3 arası New York-ta/#Los J A-3GEN 5-6 between/2-3 between N Y-LOC/L Angeles-ta iyi/#kötü yarış-tığ-ın]-ı düşün-üyor A-LOC good/#bad compete-DIK-3S-ACC think-IMPF ‘John thinks [that Ali competed well/#bad between 5-6/between 2-3 #in Los Angeles/#in New York]’ 202 b. John [Ali 5-6 arası/#2-3 arası New York-ta/#Los Angeles-ta J A 5-6 between/2-3 between N Y-LOC/L A-LOC iyi/#kötü yarış-tı diye] düşün-üyor well/bad compete-PST diye think-IMPF ‘John thinks [diye Ali competed well/#bad between 5-6/#between 2-3 in New York/#in Los Angeles]’ The generalization from the data so far is that matrix subject-oriented predicates neutralize the interpretational distinctions between their complement clauses such that all content of the embedded clause must be reported from the perspective of the matrix subject regardless of the complement type. This is stated in the generalization in (21): (21) Generalization 1: All elements in the embedded clause under matrix subject-oriented verbs must be reported from the perspective of the matrix subject regardless of the complement type. 89 Recall that when diye combines with bil ‘know’, de re elements are not allowed in them whereas they are allowed when –DIK combines with the same verb, as shown in the example in (22), repeated from Chapter 2 (example (34)). This was used as evidence that diye clauses under these verbs are a single evaluation domain associated with the matrix subject’s perspective, as de re elements are associated with the utterance context. 89 The higher matrix subject cannot be the sole evaluators under matrix subject-oriented verbs, similar to perspective-underspecified verbs in Chapter 2. 203 (22) Context: Murat has two names, Murat and Ali. John knows him only as Murat (i.e., he does not know that Murat has another name). Speaker knows both names of Murat (i.e., knows him as Murat Ali) but always refers to him as Ali. Both know Murat Ali passed his exam. a. John [Murat-ın sınav-ı geç-tiğ-in]-i bil-iyor aslında J M-3GEN exam-ACC pass-DIK-3S-ACC know-IMPF actually ‘John knows [that Murat passed the exam] actually’ b. John [Murat sınav-ı geç-ti diye] bil-iyor aslında J M exam-ACC pass-PST diye know-IMPF actually ‘John knows [diye Murat passed the exam] actually’ c. John [Ali-nin sınav-ı geç-tiğ-in]-i bil-iyor aslında J A-3GEN exam-ACC pass-DIK-3S-ACC know-IMPF actually ‘John knows [that Ali passed the exam] actually’ d. #John [Ali sınav-ı geç-ti diye] bil-iyor aslında J A exam-ACC pass-PST diye know-IMPF actually ‘John knows [diye Ali passed the exam] actually’ When the embedding verb is düşün- ‘think’, however, both complements allow de re elements, shown in (23) (23c/d versus 23a/b): (23) Same as (22) a. John [Murat-ın sınav-ı geç-tiğ-in]-i düşün-üyor aslında J M-3GEN exam-ACC pass-DIK-3S-ACC think-IMPF actually ‘John thinks [that Murat passed the exam] actually’ 204 b. John [Murat sınav-ı geç-ti diye] düşün-üyor aslında J M exam-ACC pass-PST diye think-IMPF actually ‘John thinks [diye Murat passed the exam] actually’ c. John [Ali-nin sınav-ı geç-tiğ-in]-i düşün-üyor aslında J A-3GEN exam-ACC pass-DIK-3S-ACC think-IMPF actually ‘John knows [that Ali passed the exam] actually’ d. John [Ali sınav-ı geç-ti diye] düşün-üyor aslında J A exam-ACC pass-PST diye think-IMPF actually ‘John thinks [diye Ali passed the exam] actually’ The permission for de re elements in (23) is indication that elements that are associated with different contexts are allowed in diye clauses in this case. This is a challenge to the generalization just established in (21). I come back to this issue later in this chapter. 4.1.2. Conscious evaluation Recall that diye clauses under perspective-underspecified verbs such as bil- ‘know’ need a sentient subject (24b) but not –DIK clauses (24a) (Chapter 2, example (81)). When the same is applied to a matrix subject-oriented verb, we see that conscious evaluation is required to evaluate both complement types (25): (24) a. ?Araba [pro yük-ün-ün ağır ol-duğ-un]-u bil-iyor car burden-3GEN-3GEN heavy be-DIK-3S-ACC know-IMPF hareket etmiyor it is not moving 205 ‘The car knows [that its burden is heavy] so it is not moving’ b. #Araba [pro yük-ü ağır diye] bil-iyor hareket etmiyor car burden-3GEN heavy diye know-IMPF it is not moving ‘The car knows [that its burden is heavy] so it is not moving’ (25) a. #Araba [pro yük-ün-ün ağır ol-duğ-un]-u düşün-üyor car burden-3GEN-3GEN heavy be-DIK-3S-ACC think-IMPF hareket etmiyor it is not moving ‘The car thinks [that its burden is heavy] so it is not moving’ b. #Araba [pro yük-ü ağır diye] düşün-üyor hareket etmiyor car burden-3GEN heavy diye think-IMPF it is not moving ‘The car thinks [that its burden is heavy] so it is not moving’ De se contexts were used to test for consciousness in Chapter 2 (example (86)) and we noticed that de se interpretation is imposed on the embedded clause when diye is embedded under bil- ‘know’: (26) Context: Murat is watching the results of the election, of which he was one of the candidates, on TV. He is drunk. He sees the winner and says to me “That candidate won”. He does not realize that that candidate is himself. a. ?Murat [seçim-i kazan-dığ-ın]-ı bil-iyor aslında M election-ACC win-DIK-3S-ACC know-IMPF actually ‘Murati knows [that hei won the election] actually’ 206 …ama kendisi olduğunun farkına varamadı daha …but he is not yet aware that it is himself b. #Murat [seçim-i kazan-dı diye] bil-iyor aslında M election-ACC win-PST diye know-IMPF actually ‘Murati knows [diye hei won the election] actually’ …ama kendisi olduğunun farkına varamadı daha …but he is not yet aware that it is himself When the embedding verb is matrix subject-oriented, de se interpretation is required no matter what the embedded clause type is. The de se interpretation is imposed on various elements such as on pronominal elements (27), on temporal expressions (28) and locative expressions (29): 90 (27) Context: Murat is watching the results of the election, of which he was one of the candidates, on TV. He is drunk. He sees the winner and says to me “That candidate won”. He does not realize that that candidate is himself. a. #Murati [proi seçim-i kazan-dığ-ın]-ı düşün-üyor aslında M election-ACC win-DIK-3S-ACC think-IMPF actually ‘Murati thinks [that hei won the election] actually’ …ama kendisi olduğunun farkına varamadı daha …but he is not yet aware that it is himself b. #Murati [proi seçim-i kazan-dı diye] düşün-üyor aslında M election-ACC win-PST diye think-IMPF actually 90 In (28) and (29), the words tarih, which is interpreted as referring to today, and mekan, which is interpreted as referring to the current location. 207 ‘Murati thinks [diye hei won the election] actually’ …ama kendisi olduğunun farkına varamadı daha …but he is not yet aware that it is himself (28) Context: It is December 31 st . John thinks it is December 30 th . John told the speaker that it was December 31 st before, but now he says it is December 30 th . a. #John [tarih-in 31 Aralık ol-duğ-un]-u düşün-üyor aslında J date-3GEN 31 December be-DIK-3S-ACC think-IMPF actually ‘John thinks [that today is NYE] actually’ …ama unuttu herhalde …but he apparently forgot b. #John [tarih 31 Aralık diye] düşün-üyor aslında 91 J date 31 December diye know-IMPF actually ‘John thinks [diye today is NYE] actually’ …ama unuttu herhalde …but he apparently forgot (29) Context: John, Ali and the speaker are in Istanbul. John thinks he is in Constantinople, not being aware that Istanbul and Constantinople refer to the same city. But earlier, he said to the speaker that he knew they are in Istanbul. a. #John [mekan-ın Istanbul ol-duğ-un]-u düşün-üyor aslında J location-3GEN I be-DIK-3S-ACC think-IMPF actually ‘Johni thinks [that the location is Istanbul] actually’ 91 Assume this sentence is a reply to the question John bugünü ne diye biliyor? ‘What does John know today as?’ 208 …ama unuttu herhalde …but he apparently forgot b. #John [mekan Istanbul diye] düşün-üyor aslında J location I diye think-IMPF actually ‘Johni thinks [diye hei is in Istanbul]’ …ama unuttu herhalde …but he apparently forgot This leads to another generalization, stated in (30): (30) Generalization 2: Matrix subject-oriented verbs require conscious evaluation by the matrix subject on their complement clauses, regardless of complement type. 92 4.1.3. Interim summary Matrix subject-oriented verbs in Turkish impose on their complements (i) evaluation from the perspective of the matrix subject and (ii) conscious interpretation for the matrix subject. They also allow de re elements in the embedded clause (i.e., de dicto interpretation is not required) regardless of whether their complement is a –DIK clause or a diye clause. The next three sections discuss data testing whether diye clauses are opaque domains under matrix subject-oriented verbs. These data are anaphor-binding (Section 1.4.), NPI-licensing (Section 1.5.) and specific readings of quantifiers (Section 1.6.). The conclusion is that diye clauses are not opaque under matrix subject-oriented verbs, but they are under verbs like bil- ‘know’. 92 As pointed out to me by Isabelle Charnavel (p.c), this contradicts the de re-de dicto facts, which is that de re elements are allowed under matrix subject-oriented verbs. I have a syntactic explanation for this later in this chapter. 209 4.1.4. Anaphor binding Recall that licensing of embedded anaphors from the matrix subject is allowed only in –DIK clauses under bil- ‘know’, as in (31) (Chapter 2, example 49). When the matrix verb is matrix subject-oriented, however, this binding process is allowed regardless of the complement type, as in (32): (31) a. Çocuklari [Ali-nin birbirlerin-ii eşleştir-diğ-in]-i bil-iyor children A-3GEN each.other-ACC match-DIK-3S-ACC know-IMPF ‘Childreni know [that Ali matched each otheri]’ b. *Çocuklari [Ali birbirlerin-ii eşleştir-di diye] bil-iyor children A each.other-ACC match-PST diye know-IMPF ‘Childreni know [diye Ali matched each otheri]’ (32) a. Çocuklari [Ali-nin birbirlerin-ii eşleştir-diğ-in]-i düşün-üyor children A-3GEN each.other-ACC match-DIK-3S-ACC think-IMPF ‘Childreni think [that Ali matched each otheri]’ b. ?Çocuklari [Ali birbirlerin-ii eşleştir-di diye] düşün-üyor children A each.other-ACC match-PST diye think-IMPF ‘Childreni think [diye Ali matched each otheri]’ 4.1.5. NPI-licensing Recall that NPIs in diye clauses are not licensed by matrix verb negation when the matrix verb is perspective-underspecified such as bil- ‘know’ (33b) while –DIK clauses under the same verb 210 allows the licensing (33a) (Chapter 2, example (51)). The negation on matrix subject-oriented verbs is able to license the NPI both in the –DIK clause and the diye clause (34): (33) a. Ali [kimse-nin gel-diğ-in]-i bil-m-iyor A no.one-3GEN come-DIK-3S-ACC know-NEG-IMPF ‘Ali knows [that no one came] (lit. Ali does not know that anyone came)’ b. ??/# Ali [kimse gel-di diye] bil-m-iyor A no.one come-PST diye know-NEG-IMPF ‘Intended: Ali knows [diye no one came]’ (lit. Ali does not know [diye anyone came]) (34) a. Ali [kimse-nin gel-diğ-in]-i düşün-m-üyor A no.one-3GEN come-DIK-3S-ACC think-NEG-IMPF ‘Ali thinks [that no one came] (lit. Ali does not know that anyone came)’ b. ?Ali [kimse gel-di diye] düşün-m-üyor 93 A no.one come-PST diye think-NEG-IMPF ‘Ali thinks [diye no one came]’ (lit. Ali does not think [diye anyone came]) 93 Exactly how this is interpreted is a bit vague. The options are that kimse has narrow scope (i.e., Ali thinks that [no one came]) and that kimse is interpreted as concealed partitive (e.g., Ali thinks that {none (of the people specified before} came). The latter seems easier to retrieve, which might weaken the argument since it is context (specifically, topic) related and not related to the embedding verb (since topichood is exceptional under factivity-underspecified verbs too). If the former is allowed here, this provides support for the effect of the embedding verbs, since this reading is not allowed under factivity-underspecified verbs. 211 4.1.6. Specific readings of quantifiers Diye+bil does not allow the specific reading of the quantifier embedded subject someone (35b) while –DIK+bil does (35a). (Chapter 2, example 52). Matrix subject-oriented düşün- ‘think’ allows this reading of the quantifier with both –DIK and diye (36): (35) a. Pelin [burada biri-nin hasta ol-duğ-un]-u bil-iyor P here someone-3GEN sick be-DIK-3SACC know-IMPF ‘Pelin knows [that someone (specific) is sick here]’ (specific) ‘Pelin knows [that there is someone sick here]’ (nonspecific) b. Pelin [burada biri hasta diye] bil-iyor P here someone sick diye know-IMPF ‘#Pelin knows [diye someone (specific) is sick here]’ (specific) ‘Pelin knows [diye there is someone sick here]’ (nonspecific) (36) a. Pelin [burada biri-nin hasta ol-duğ-un]-u düşün-üyor P here someone-3GEN sick be-DIK-3SACC think-IMPF ‘Pelin thinks [that someone (specific) is sick here]’ (specific) ‘Pelin thinks [that there is someone sick here]’ (nonspecific) b. Pelin [burada biri hasta diye] düşün-üyor P here someone sick diye think-IMPF ‘Pelin thinks [diye someone (specific) is sick here]’ (specific) ‘Pelin thinks [diye there is someone sick here]’ (nonspecific) Now, we reach our third generalization, stated in (37): 212 (37) Generalization 3: Anaphor binding, NPI-licensing and specific readings of quantifiers are allowed in both –DIK and diye clauses when they are embedded under a matrix subject-oriented verb. 4.1.7. Interim summary and brief remarks The data in the last three sections indicate that matrix subject-oriented predicates (in contrast with perspective-underspecified) allow (i) binding of embedded anaphors, (ii) licensing of NPIs, both across clauses, and (iii) specific readings of quantifier elements with both –DIK and diye complement clauses. The contrasting generalizations with embedding verb type are in (38): (38) Generalization; Perspective-underspecified: Diye clauses form a single perspective domain, with the closest matrix subject as the perspective center. This domain is syntactically (selectively) opaque. (–DIK clauses do not encode any perspective and are transparent). Generalization; Matrix subject-oriented: Diye clauses form a single perspective domain, with the closest matrix subject as the perspective center. This domain is syntactically transparent under matrix subject-oriented verbs. (–DIK clauses do not encode any perspective and are transparent). The overall generalization is that various syntactic operations are allowed between elements in the diye clause and the matrix clause when they are embedded under matrix subject-oriented verbs. This means that diye clauses are no longer (selectively) opaque domains under such verbs, 213 unlike their selectively opaque nature under perspective -underspecified verbs. An account for this change in behavior is needed, which I will provide in the next section. 94 4.2. Proposal In this section, I present the proposal (Section 4.2.1.), followed by a discussion of how it accounts for the patterns observed (Section 4.2.2.). 4.2.1. The proposal I propose that matrix subject-oriented verbs such as düşün- ‘think’ take an CP matrixsubject as a complement, as in (39), where CP matrixsubject is a phrase that hosts a perspectival operator, OP matrixsubject, on its head. The operator is similar to OPdiye in function in that it reports the matrix subject’s perspective on the embedded content, but they are different in that OP matrixsubject does not have a subject PRO in its specifier of the phrase it heads. 95 I assume that subjects are introduced in the Spec of VP (or vP) (Koopman & Sportische 1991) across the board, so the matrix subject is introduced in SpecVP. The –DIK or diye clause is the complement to CP matrixsubject. The structure I propose for matrix subject-oriented predicates and their complement in Turkish is given in (39a) and (39b), when they embed a diye clause and a –DIK clause, respectively: 94 –DIK clauses are always –i.e., under any type of verb under focus- transparent in that they allow all interactions in focus between the matrix clause and the embedded clause. I assume that they do not form an opaque domain in the first place because they do not have a PersP. Thus, I do not include them in the analysis here. 95 OPmatrixsubject reports the first-person perspective of an individual despite lacking a PRO. I discuss this in more detail later, but this is an important point to keep in mind in the development of the analysis. 214 (39) a. diye+think b. –DIK+think Notice that when matrix subject-oriented verbs combine with diye, we end up with two operators with similar functions in the complement clause. This introduces redundancy on the surface. However, there is one purpose the double-operator structure serves in the derivation. On the assumption that the derivation goes bottom-up, let us assume that the derivation of the diye clause, i.e., PersP, is completed, so now we have one of the two operators, namely OPdiye. The reference of the individual argument of the operator within this PersP, i.e., OPdiye, also the antecedent of PRO, needs to be introduced for complete interpretation, so the system waits for the next phrase level for the antecedent. Under perspective-underspecified verbs, the next phrase level is VP, which introduces the antecedent of PRO (i.e., the matrix subject) and the interpretation for the diye clause is completed there, resulting in an (selectively) opaque domain. The next phrase level up in the case of matrix subject-oriented verbs, however, is the CP matrixsubject, which introduces a similar operator, OP matrixsubject, but without an element relating to an individual because OP matrixsubject does not have a PRO subject. So, the derivation moves onto 215 the next phrase level to find a DP for the PRO in PersP to refer to, which it finds in the subject of the matrix VP. When the derivation is complete for the embedded diye clause, the interfaces and the syntax interpret only the higher operator (OP matrixsubject) and not the lower one (OPdiye), since they serve similar functions in the same complement clause domain (reporting an individual’s first-person perspective). 96 This makes the latter syntactically inactive, which means that it is not considered to be relevant for syntactic operations. This is what gives the transparent nature of diye clauses under non-factive verbs, which I discuss in more detail later. For now, let us assume that a probe in matrix CP sees the embedded clause as an CP matrixsubject rather than a PersP. This is similar to de Cuba (2007), for whom the additional layer of cP serves as a domain-extender in non-factive predicates. However, I diverge from de Cuba and agree with den Dikken (2007a, b) in that the said extra layer extends the opaque domain but causes the original phase to lose its phasal status. Thus, I follow a mixture of the proposals in den Dikken ((2007a, b) and de Cuba (2007) in that I take this extension to be a widening of the edge of an opaque domain (phase-edge extension in their account). But I also argue that diye loses its opaque status when this extension happens, following den Dikken (2007a, b). In a nutshell, when the diye clause appears in the complement to CP matrixsubject, CP matrixsubject is the new complement domain and due to lack of PRO at the edge, it is a transparent domain. We have seen data supporting this conclusion in Section 1, discussed in this light in the next section. 97 96 Though this process does not involve movement, this can be likened to PF interpreting the higher copy in movement chains. 97 Notice that this account assumes that although diye clauses are not phases, as opaque domains, they are still subject to a penetrability condition similar to the one proposed for phases, as in (i), unless extended and unless the probe is higher in the functional hierarchy (Chapter 3). (i) Phase Impenetrability Condition (Chomsky 2001: 14) The domain of a head X of a phase XP is not accessible to operations at ZP (the next phase); only X and its edge are accessible to such operations. 216 4.2.2. Transparency of diye clauses Recall the third generalization, repeated in (40): (40) Generalization 3: Anaphor binding, NPI-licensing and specific readings of quantifiers are allowed in both –DIK and diye clauses when they are embedded under a matrix subject-oriented verb. The suggestion that the system pays attention to the OP matrixsubject rather than OPdiye is important for (40) because there is an important difference between the two operators: the former has an empty specifier while the latter has a PRO subject in its specifier. Since the specifier of CP matrixsubject, the higher of the two, is empty, the edge of the domain that includes the complement clause is empty, making the diye clause transparent. This transparency allows binding across domains regardless of the complement type under matrix subject-oriented verbs: (41) a. Çocuklari [Ali-nin birbirlerin-ii eşleştir-diğ-in]-i düşün-üyor children A-3GEN each.other-ACC match-DIK-3S-ACC think-IMPF ‘Childreni think [that Ali matched each otheri]’ b. ?Çocuklari [Ali birbirlerin-ii eşleştir-di diye] düşün-üyor children A each.other-ACC match-PST diye think-IMPF ‘Childreni think [diye Ali matched each otheri]’ 217 Likewise, NPI-licensing by matrix negation is allowed under matrix subject-oriented verbs, as in (41). This is also possible because the diye clause is no longer an opaque domain: 98 (42) a. Ali [kimse-nin gel-diğ-in]-i düşün-m-üyor A no.one-3GEN come-DIK-3S-ACC think-NEG-IMPF ‘Ali thinks [that no one came] (lit. Ali does not know that anyone came)’ b. ?Ali [kimse gel-di diye] düşün-m-üyor A no.one come-PST diye think-NEG-IMPF ‘Ali thinks [diye no one came]’ (lit. Ali does not think [diye anyone came]) The availability of specific readings of quantifiers also has a similar explanation. Assuming that such readings require association with the utterance context (no matter how it is established), these quantifiers must be able to interact with this context. This is made possible because the diye clause is no longer an opaque domain. The relevant examples are repeated below: (43) a. Pelin [burada biri-nin hasta ol-duğ-un]-u düşün-üyor P here someone-3GEN sick be-DIK-3SACC think-IMPF ‘Pelin thinks [that someone (specific) is sick here]’ (specific) ‘Pelin thinks [that there is someone sick here]’ (nonspecific) b. Pelin [burada biri hasta diye] düşün-üyor P here someone sick diye think-IMPF ‘Pelin thinks [diye someone (specific) is sick here]’ (specific) ‘Pelin thinks [diye there is someone sick here]’ (nonspecific) 98 Also see Kayabaşı & Özgen (2018) for a phase-based account of NPI-licensing, namely that NPI-licensing is allowed when there is no phase boundary in Turkish. 218 These interactions are allowed because the diye clause waits for the VP to introduce the antecedent individual in its specifier so that it has a referent for its PRO subject. Once the individual is introduced, full interpretation is achieved but the diye operator is no longer syntactically active. The luck here is that the higher operator has an empty specifier, which keeps the embedded clause domain open and results in a transparent domain. If there was a PRO subject with OP matrixsubject, this would lead to an inaccessible embedded clause domain. We will see this happening with true factives. Thus, this proposal accounts for the last three generalizations posited. What about the first three? These are addressed in the next section. 4.2.3. Phrasal content in diye clauses under matrix subject-oriented verbs Recall that in Chapter 2, it was suggested that the phrasal elements in the embedded clause come with a variable that needs to be bound by the closest operator. This is represented in (44) for the case when diye is under bil- ‘know’: (44) a. [VP [ PersP PRO [CP … [AdvP w’ between 4-5]…] λw’ diye] V] b. [VP [ PersP PRO[CP[DP w’ the nice woman]…[AdvP w’ between 4-5]…[VP w’ came]] λw’ diye] V] Since the closest operator is OPdiye in these cases and there are no other operators, we do not have a problem with choosing the right operator to bind these variables. In the case when diye appears under düşün ‘think’, however, it is not as clear. The closest operator is OPdiye, but I also suggested that it is OP matrixsubject that is relevant for structural purposes as the higher of the two 219 similar operators and OPdiye is syntactically inactive. This time, I suggest that it is not the closest operator in terms of distance but rather it is the structurally active operator that is relevant for binding the variables. The system skips OPdiye because it is inactive and forces the binding of the variables by OP matrixsubject. This provides a unified account of why all elements in both –DIK and diye clauses must be reported from the perspective of the matrix subject (as opposed to the case under perspective-underspecified verbs, where –DIK clauses did not need to be reported from the perspective of the matrix subject only). It is because CP matrixsubject scopes over both the entire – DIK clause (45b) and the entire diye clause (45a): (45) a. [CPmatrixsubject Spec [ PersP PRO [CP … [AdvP w’ between 4-5]…] OPdiye] λw’ C matrixsubject] b. [CPmatrixsubject Spec [DP Spec [CP … [AdvP w’ between 4-5]…] –DIK] λw’ C matrixsubject] Now, onto the generalizations in (21) and (30), repeated below: (46) Generalization 1: All elements in the embedded clause under matrix subject-oriented verbs must be reported from the perspective of the matrix subject regardless of the complement type. Generalization 2: Matrix subject-oriented verbs require conscious evaluation by the matrix subject on their complement clauses, regardless of complement type. Generalization 1 is accounted for in the following way: OP matrixsubject scopes over the whole of the embedded clause and reports all of it from the first-person perspective of the matrix subject. It binds all the said variables in the embedded clause and forces all to be reported from the matrix subject’s perspective. Generalization 2 speaks to the logophoric nature of the matrix subject operator as it requires a sentient subject. This is all as expected. However, recall that de re 220 elements are allowed in diye clauses under matrix subject-oriented verbs, which contradicts Generalization 1 & 2. The de re element in (47) is Ali. Although John does not know that Ali refers to the same individual as Murat, the use of Ali in the embedded clause is allowed. (47) Context: Murat has two names, Murat and Ali. John knows him only as Murat (i.e., he does not know that Murat has another name). Speaker knows both names (i.e., knows him as Murat Ali) but always refers to him as Ali. Murat Ali passed his exam. a. John [Ali-nin sınav-ı geç-tiğ-in]-i düşün-üyor aslında J A-3GEN exam-ACC pass-DIK-3S-ACC think-IMPF actually ‘John knows [that Ali passed the exam] actually’ b. John [Ali sınav-ı geç-ti diye] düşün-üyor aslında J A exam-ACC pass-PST diye think-IMPF actually ‘John thinks [diye Ali passed the exam] actually’ De re reading is related to the utterance context. This can be argued to be achieved via covert movement of the de re element (known as the Scope Theory, which is still adopted in some recent works such as Woods 2016). If this account is adopted, the de re reading is made possible in the present account because the diye clause is no longer opaque under non-factive verbs due to the availability of CP matrixsubject, which does not provide the antecedent individual for the PRO in diye clauses. This incomplete process makes the movement out of diye clauses possible so that full interpretation can be achieved. However, some recent analyses challenged the Scope Theory (e.g., Keshet 2008, Anand & Korotkova 2021) and argued that de re elements are specified for the utterance context inside of them. The following is an example of a de re element from Anand & Korotkova 2021): 221 (48) Mary thinks an adorable dog is ugly. They state (p. 8) that “If an adorable dog were read de dicto, the PPTs adorable and ugly would be evaluated at the same index, the embedded i1. This would mean that they would be evaluated with respect to the same world and the same judge (by assumption, Mary), yielding a contradiction. However, if an adorable dog were read de re, adorable would be evaluated with respect to i0 and the matrix judge (by assumption, the speaker). The resulting interpretation would be non-contradictory”. In their account, “de re interpretation is done by non-local binding of an index variable (p.8)”, where index includes a judge and a world coordinate following Lasersohn (2005, 2017). If this account is adopted, it would mean that the matrix operator would be the non-local binder for the de re element in diye clauses under matrix subject-oriented verbs and this binding is made possible again because diye clauses are not opaque domains in this environment for de re elements. I remain neutral as to the choice between these accounts. But Generalization 1, together with this exception, means that embedded content is represented from the perspective of the matrix subject unless it is marked de re, in which case it can escape the diye domain. This also speaks to Generalization 2: all elements that are associated with the operator in the embedded clause must be read de dicto. But due to transparency, elements now can escape that domain and be read as de re. The difference, therefore, lies in the ability of de re elements to escape the diye domain under non-factive verbs and their inability to do so under factivity-underspecified verbs. This difference results from the availability of CP matrixsubject in the former and the lack of it in the latter. Again, CP matrixsubject, not providing an antecedent individual for the PRO in diye clauses, makes diye clauses transparent. 222 Another type of de re element discussed were overt pronouns and indexicals like burada ‘here’ and I argued that they are lexically specified for the utterance context in their own domains. Recall the following examples (Chapter 3, example 112): (49) Context 1: Mary is in Los Angeles. However, she thinks she is in Irvine. The speaker is in Los Angeles. Mary has an appointment in LA. Mary [toplantı burada/#orada ol-acak diye] bil-iyor M meeting here/there be-FUT diye know-IMPF ‘Mary knows [diye the meeting will be here/#there]’ Context 2: Mary is in Los Angeles. However, she thinks she is in Irvine. The speaker is in Irvine. Mary has an appointment in LA. Mary [toplantı #burada/orada ol-acak diye] bil-iyor M meeting here/there be-FUT diye know-IMPF ‘Mary knows [diye the meeting will be #here/there]’ The same holds when the embedding verb is matrix subject-oriented: (50) Context 1: Mary is in Los Angeles. However, she thinks she is in Irvine. The speaker is in Los Angeles. Mary has an appointment in LA. Mary [toplantı burada/#orada ol-acak diye] düşün-üyor M meeting here/there be-FUT diye think-IMPF ‘Mary think [diye the meeting will be here/#there]’ Context 2: Mary is in Los Angeles. However, she thinks she is in Irvine. The speaker is in Irvine. Mary has an appointment in LA. 223 Mary [toplantı #burada/orada ol-acak diye] düşün-üyor M meeting here/there be-FUT diye think-IMPF ‘Mary thinks [diye the meeting will be #here/there]’ These are different from the de re in (49) because they show the same pattern regardless of the embedding verb, or rather regardless of whether the diye clause is opaque or not. Thus, it is clear that they are lexically specified for the utterance context. That means that OP matrixsubject can only bind elements that are not specified as relating to another context lexically within their immediate projection and the binding of these elements happen within their immediate projection (i.e., not the matrix operator). 99 4.2.4. Semantics of matrix subject-orientation operator Admittedly, proposing a semantic denotation for OP matrixsubject is complicated due to compositional reasons. Although OP matrixsubject and OPdiye have similar functions in reporting the first-person perspective of an individual, they cannot have the same semantic denotation. For one, the former lacks a PRO subject that the latter has, which requires a different way of handling the individual argument. Second, the latter only takes a CP as its argument while the former can take a DP or a PersP. Therefore, OP matrixsubject needs special handling, which I leave for future research. However, a few points are worth noting regarding the world argument in this operator, which I assume is a part of any semantic definition we propose for OP matrixsubject. 99 Another potential explanation of this fact is that maybe there is co-reference rather than binding in examples like (50). This is possible, however, I stick to the binding account to keep the consistency across types of data and the correlations I posited in Chapter 3 (i.e., elements bound in embedded context=phrasal content, elements bound either in embedded or utterance context=pro, elements bound in the utterance context=lexical items). 224 Let us now specify a bit further what the function of CP matrixsubject is. de Cuba (2007) argues for a cP in the same position and suggests that the function of the operator in cP is to remove the speaker from responsibility for the truth of the embedded proposition. In his account, the operator in this position makes sure that the actual world is not included in the evaluation of the truth value of the embedded clause. He does not clarify how we can make sure that the actual world is not included in the evaluation worlds, however. Woods (2016) takes a more direct stance on this and suggests the following (for verbs of communication) (p. 172): All items in the [embedded] clause are interpreted according to the perspectival centre and, though this was not explicitly mentioned earlier, the embedded world parameter. As such the only reading available is that which is said or asked in the original discourse; a reading with respect to the actual world is unavailable because the actual world parameter has been swapped out. Nichols (2001) is a bit more specific. She takes a more generalizable route and refers to the set of evaluation worlds that do not include the actual world as <-w>. She states the following (p. 129): One might even reduce the evaluation set to a single variable <w>, where non-factive attitude predicates supply a value –w= ‘(truth) not (evaluated) in the actual world. If we follow Woods and Nichols, matrix subject-oriented(=non-factive) predicates come with a specification that their complement proposition is evaluated in worlds that do not include the actual world. If this operator can be formulated in a way to refer to logophoric alternatives of an 225 individual, as was the case with the operator in diye clauses, this might be a way of positing the non-inclusion of the actual world. I leave this for future research. 4.2.5. Question elements and McCloskey (2005) Another interesting contrast that points out to some potential unification of why the stated syntactic patterns are observed (e.g., why NPI-licensing differences) arises with question elements inside embedded clauses. Narrow scope reading of question elements in diye clauses under perspective-underspecified verbs is not allowed (51a) but they are allowed under the matrix subject-oriented düşün ‘think’ (51b): 100 (51) a. [Ali mi gel-di diye] hatırla-dı-n/bil-iyor-du-n A Q come-PST diye remember-PST-2S/know-IMPF-PST-2S ‘Did you remember/know that it was Ali that came?’ ‘*/??You remembered/knew whether Ali came’ b. [Ali mi gel-di diye] düşün-dü-n A Q come-PST diye think-PST-2S ‘Did you think that Ali came?’ ‘You thought whether Ali came’ In Chapter 3, the availability of high scope reading in (51a) was attributed to the question probe (IntP) being higher than the highest projection in the entire diye clause, on an account of 100 -DIK clauses always require wide scope reading with the question particle regardless of the embedding verb. This indicates that the inability of letting narrow scope with these particles has to do with the internal structure of –DIK clauses rather than having to do with the CPmatrixsubject. Since there is no contrast with these clauses, I do not attempt to explain it. Rather I stick to diye clauses. 226 selective opacity based on height (Poole 2020). The lower scope interpretation is not allowed in (51a). This is not the case with matrix subject-oriented embedding verbs, as in (51b). I suggest that the narrow scope reading of the question elements under matrix subject-oriented verbs is allowed due to the operator in CP matrixsubject because CP matrixsubject and question elements have a feature in common. Recall that OP matrixsubject quantifies over the logophoric alternatives of an individual to evaluate the embedded proposition in these alternatives, which indicates the proposition has not been evaluated in the actual world. Namely, the issue at hand is not resolved. Questions and negation have this same presupposition, which puts them all under the group of elements that indicate ‘unresolvedness’. This is the core intuition in McCloskey (2005). It is possible that a syntactic feature (e.g., [+/- resolved]) combines these elements and allow their interaction such that CP matrixsubject in matrix subject-oriented predicates values the unvalued feature on the question element as the probe for ‘unresolvedness’. This kind of interaction is especially noteworthy because these elements have already been shown to interact in the syntax. McCloskey (2005) argues that inversion and adjunction in embedded clauses depends on the meaning of the embedding predicate as well as the elements the embedding predicate combines with. He shows that both inversion and adjunction are possible if the embedding predicate is [+wh] (i.e., a question predicate), as in (52). However, it is not possible if the embedding predicate is a ‘resolutive’ predicate, 101 a group which includes what I call speaker-oriented and perspective-underspecified predicates (53). (52) a. I wondered was he illiterate. b. ?He asked me when I got home if I would cook dinner. 101 The term ‘resolutive predicates’ was introduced by Ginzburg and Sag (2000) to refer to verbs that do not embed true questions and as such presuppose that “the embedded question is resolved (p. 65, fn. 10)”. 227 (53) a. *The police discovered who had they beaten up. b. *The police established while we were out who had broken in to our apartment. He also noticed that it was not the case that resolutive predicates never allowed inversion and adjunction. Specifically, they allow inversion when negated (54b) or questioned (54c). Likewise, adjunction becomes possible when the embedding predicate is negated or questioned (55): (54) a. *I remember was Henry a Communist b. ?I don’t remember was Henry a communist? c. Do you remember was he a communist? (55) a. I was never sure when he went to England should I go with him. b. Do you remember when they were in Derry did they live in Rosemount? McCloskey proposes to account for these patterns via an extra level of CP available to [+wh] predicates, as in (56). Negation and questions are also compatible with the double CP in this account. This allows for a position through which adjunction and inversion can happen. This proposal is also aligned with the adjunction prohibition, stated in (57) (Chomsky 1986: 6), since adjunction is now separated from the lexical V head by the added CP. 228 (56) (57) Adjunction to a phrase which is s-selected by a lexical (open class) head is ungrammatical. Inspired by Krifka (1999), McCloskey proposes that the double CP in this account refers to the ‘higher interrogative type’, similar to root questions, while the single CP is the ‘lower interrogative type’, which is sets of propositions. The former hosts a QUEST operator, which takes sets of propositions (in lower CP) and returns an interrogative speech act (in higher CP). [+wh] verbs are compatible with such an operator. Resolutive predicates lead to a clash of presuppositions. If such verbs could combine with the QUEST operator, it would “simultaneously entail and presuppose that the issue defined by the embedded question is resolved for the experiencer and entails or presupposes that it is not (p. 35)”. Negation “entails/asserts that the issue defined by the complement is not resolved (for the referent of the experiencer argument). As a consequence, the question act is felicitous in its context (p. 35)”. A 229 similar explanation goes for matrix questions, as in (54c). The question indicates that the issue in the embedded clause has not been resolved and therefore can combine with the QUEST operator. Although a full semantic analysis of this class of elements is beyond the scope of the present study and was also not developed by McCloskey, we now have an idea of what can allow interaction between CP matrixsubject and embedded questions: both of these elements indicate that the issue in the embedded clause has not been resolved due to their matrix subject-orientedness (non-factivity) and question nature, respectively. On the assumption that the syntax marks this via a feature, we may suggest that they share a feature such as [+/- resolved] that allows them to interact in the syntax (and/or LF). It is via this feature that CP matrixsubject licenses embedded questions that are otherwise not allowed under perspective-underspecified verbs. If this account is pursued, we would have the following distinction between verbs based on the whether they are marked as [+resolved] or [-resolved] (which I take to be encoded in their complement CP matrixsubject): 102 (58) [-resolved] = matrix subject-oriented, negated verbs, questioned verbs [+resolved] = speaker-oriented, perspective-underspecified (?) In fact, this might also explain why NPIs are licensed under non-factive predicates. Although I attributed NPI-licensing to lack of an opaque domain in the diye clause under non-factive verbs rather than to specific features of CP matrixsubject, it is possible that a feature on C matrixsubject indicating unresolvedness licenses an embedded NPI, since negation is an indicator of unresolvedness. 102 The distinction in (58) is exemplary. There might be more verb types in each category. 230 On a side note, McCloskey did not discuss matrix subject-oriented, non-interrogative verbs in the same class as [+wh] verbs, negation and questions. The Turkish data, however, imply that they could indeed be considered in the same group. It is yet to be seen whether similar patterns of adjunction and inversion arise with matrix subject-oriented, non-interrogative verbs in the Irish dialect of English but I take this possibility as a promising extension of his insightful proposal. 4.2.6. Optional elements First-person pro can refer to either the speaker or the matrix subject in diye complements but not in –DIK complements both under perspective-underspecified verbs (59) (Chapter 3, example 18) and matrix subject-oriented verbs (60): (59) a. Esra [pro sınıf-ta kal-dığ-ım]-ı bil-iyor E class-LOC stay-DIK-1S-ACC know-IMPF ‘Esra knows [that I failed the class]’ (ISG pro=S, #MS) b. Esra [pro sınıf-ta kal-dı-m diye] bil-iyor E class-LOC stay-PST-1S diye know-IMPF ‘Esra knows [diye I failed the class]’ (ISG pro=S, MS) (60) a. Esra [pro sınıf-ta kal-dığ-ım]-ı düşün-üyor E class-LOC stay-DIK-1S-ACC think-IMPF 231 ‘Esra thinks [that I failed the class]’ (ISG pro=S, #MS) b. Esra [pro sınıf-ta kal-dı-m diye] düşün-üyor E class-LOC stay-PST-1S diye think-IMPF ‘Esra thinks [diye I failed the class]’ (ISG pro=S, MS) Recall that a c-commanding overt pronoun can block reference of pro to the matrix subject in diye clauses under bil- ‘know’ (61), a pattern that was (weakly) attributed to De Re Blocking. The same pattern emerges with a matrix subject-oriented embedding verb (62): (61) Esra [ben sınıf-ta kal-dı-m diye] bil-iyor E I class-LOC stay-PST-1S diye know-IMPF ‘Esra knows [diye I failed the class]’ (ISG pro=S, #MS) (62) Esra [ben sınıf-ta kal-dı-m diye] düşün-üyor E I class-LOC stay-PST-1S diye think-IMPF ‘Esra knows [diye I failed the class]’ (ISG pro=S, #MS) I have also shown that pro in diye clauses is not restricted by locality under factivity- underspecified verbs (63). Again, the same pattern is observed with non-factive verbs (64): 232 (63) Esra [John pro sınav-ı geç-ti-m diye] bil-iyor diye E J exam-ACC pass-PST-1S diye know-IMPF diye hatırl-ıyor remember-IMPF ‘Esra remembers that John knows [diye I passed the exam]’ (ISG pro =Esra, John, S) (64) Esra [John pro sınav-ı geç-ti-m diye] düşün-üyor diye E J exam-ACC pass-PST-1S diye think-IMPF diye hayal ed-iyor dream do-IMPF ‘Esra imagines that John thinks [diye I passed the exam]’ (ISG pro =Esra, John, S) I take this parallel behavior in pro to be supporting evidence for the proposal that OP matrixsubject and OPdiye are of similar natures, namely that they are both binding operators. 103 The explanations offered for OPdiye in Chapter 3 are therefore applicable to the data discussed in this section. I leave a study of epithets under matrix subject-oriented verbs for future work. 103 The only difference being in whether they have a PRO subject in their specifier or not. 233 4.3. Discussion This section discusses the implications of the proposal for the literature on (non-)factivity of predicates and highlights its contributions. Recall that factive=speaker-oriented and non- factive=matrix subject-oriented in the present study. I will use (non)factivity in the discussion of the literature. The discussion is presented in two sections. Section 4.3.1. discusses the structure of the complements of factive and non-factive predicates and discusses where the present proposal stands in this discussion. Section 4.3.2. discusses default factivity. Section 4.3.3. makes a novel contribution by providing evidence from Turkish that there is a three-way structural difference between the complements of speaker-oriented, matrix-subject-oriented and perspective-underspecified predicates. 4.3.1. Factive versus non-factive predicates That factive and non-factive predicates behave differently in grammar has been known at least since K & K. K & K noticed different behaviors between factive and non-factive verbs, where they defined factivity as presupposing the truth of their sentential complements. One difference is that only factive verbs can take the noun fact as a complement: (65) a. I want to make clear the fact that I don’t intend to participate. b. *I assert the fact that I don’t intend to participate. Only factive verbs can have the pronoun it as its object: 234 (66) a. Bill resents it that people are always comparing him to Mozart. b. *Bill claims it that people are always comparing him to Mozart. Gerunds can be objects of factive verbs only: (67) a. I regret having agreed to the proposal. b. *I believe having agreed to the proposal. To be able to account for these differences and others, K & K propose that factive predicates have the deep structure in (68a) while non-factives were represented as in (68b), reflecting the semantic difference between them. de Cuba (2007) updated the structures in (68) in current terms as in (69a) and (69b), respectively. For K & K, factive predicates are structurally more complex in that they are headed by the noun fact, which non-factives lack. This indicates that factives are special in that they are marked whereas non-factivity is the default. Factive predicates produce the factivity presupposition due to the availability of the noun fact inside them. The syntactic differences are accounted again with the noun fact and a number of transformations. (68) a. b. 235 (69) a. b. More recently, however, the argument has gone the opposite way, namely that factives are the default interpretation whereas it is the non-factive predicates that are special and marked by grammar. Nichols (2001: 204), for instance, states that “…a plausible default interpretive strategy is the presupposition of the truth of a proposition unless it is marked otherwise” because “… the evaluation set containing the actual world can be considered the default evaluation context (p. 126)”. Her analysis goes as follows in brief: CPs have tuples of context variable sets that include the speaker, (the hearer), time and world. When the value for the first coordinate is supplied by the current speaker, the actual world is necessarily included in the evaluation set. What is important is that factive predicates do not supply a speaker value for their complements, therefore, the first coordinate of the context tuple in the CP defaults to the current speaker. This is different in non-factives, which can supply a different speaker value than the current speaker because they come with an assertive operator. Haegeman (2006) takes a more structural take on the factivity issue. She argues that non- factives have an extra CP-layer for speaker deixis (also including projections like TopicP and Focus), which factives lack, resulting in the factivity presupposition in the latter. In other words, she argues that factive predicates are reduced finite structures, with the reduction resulting from the lack of speaker deixis, and that non-factives are structurally more complex. That is why factives do not allow various elements such as speaker-oriented adverbials (70a) and main clause 236 phenomena such as topicalization (70b). Non-factive verbs do not yield the same kind of restrictions (71): 104 (70) a. *John regrets that Mary probably/obviously/unfortunately did not attend the meeting. b. (%)*John regrets that this book Mary read. (Maki et at. 1999:3) (71) a. John believes that Mary probably/obviously/unfortunately did not attend the meeting. b. John thinks that this book Mary read. Based on these and other observations, Haegeman concludes that “presupposed complements are those propositions that do not encode anchoring to the speaker. The ‘factive’ interpretation of such clauses could be seen as a default reading: their content, not being asserted, or related to the speaker, is, as it were, ‘taken for granted’ ”. de Cuba (2007) argues that non-factives are marked and structurally more complex. For example, Swedish allows V2 in embedded clause only if the embedding verb is non-factive. In (72) and (73), the (a) examples have the neutral word order. The (b) examples exemplify V2, where the auxiliary var appears to the left of the negation inte. This inversion is possible with the non-factive sa ‘say’, but not with the factive ångrade ‘regret’: (72) a. Rickard sa at than inte var hemma R said that he not was home 104 The data in (71) and (72) can be due to the complements of factive predicates being more nominal (to be discussed), hence a smaller structure. See Korotkova (2021) for a similar argumentation for Turkish and other languages. I will in fact also argue that Turkish non-factives do not encode speaker-orientation despite being structurally more complex, hence I will argue against Haegeman (2006) in that aspect. 237 b. Rickard sa att han var inte hemma R said that he was not home ‘Rickard said that he was not home’ (73) a. Rickard ångrade att than inte var hemma R regretted that he not was home b. *Rickard ångrade att han var inte hemma R regretted that he was not home ‘Rickard regretted that he was not home’ Hungarian azt ‘it-ACC’ occurs when the embedding verb is non-factive (74a) but not when it is factive (74b). Since there is accusative marking on azt, it must have originated in the internal argument position of the embedding verb (de Cuba 2007; first noted in de Cuba & Ürögdi 2001): (74) a. Azt hiszem hogy Mari okos it-ACC I.think that M smart.is ‘I think that Mary is smart’ b. (*Azt) sajnálom hogy Mary okos it-ACC I.regret that M smart.is ‘I am sorry that Mary is smart’ If azt is not present in a non-factive clause, a factive reading results: (75) a. Azt mondta Péter, hogy késön kezdödik a meccs that-ACC said P that late begins the match ‘Péter said that the match will begin late’ (but we do not know if this is true) 238 b. Mondta Péter, hogy késön kezdödik a meccs said P that late begins the match ‘Péter told (me) that the match will begin late’ (and in fact it will) Based on data such as these, de Cuba proposes that non-factive predicates select for cP, extender of the embedded CP (76b), whereas factive predicates are select for the CP directly (76a): 105 (76) a. b. cP head hosts an operator, which, according to de Cuba, removes the speaker from responsibility for the truth of the embedded content (p.9). In the case of Swedish, V2 under the non-factive predicate in (76b) is possible because the C head is empty, to which the embedded verb moves. Since the complement of the factive verb is only a CP with no empty position, the embedded verb has no position to move to, so V2 is not possible. One might ask whether SpecCP might be argued to be a position where the embedded verb can move to. But the following example shows that movement over the complementizer (in C) is not allowed in Swedish, implying that SpecCP 105 De Cuba later revises the division between as being between those with familiar complements and novel complements, to which factive and non-factive verbs belong respectively. 239 is not available for movement. de Cuba attributes this to [+Fin] and [+EPP] features both being checked by the complementizer itself, leaving no attractor in the CP: (77) *Rickard ångrade [CP Dani [C att] [TP ti [NegP inte] [vP ti var hemma]]] R regretted D that not was home ‘Intended: Rickard regretted that Dan wasn’t home’ The Hungarian azt is allowed under non-factive verbs because there is an extra position in the structure where it can appear (cP), which is not possible with factive predicates. Since factive predicates select for a CP, whose head is filled with the complementizer, there is no position for an embedded element to move to. Miyagawa (2012) is another study arguing that only non-factives comes with extra structure, which in his case is a speech act projection based on Japanese data. Japanese has two complementizers, to, which appears under non-factive predicates and koto, which appears under factive predicates (Kuno 1973, Miyagawa 2012). The language also has a politeness marker mas that is used in formal contexts, as exemplified in (78a). Lack of it indicates colloquial speech (78b) (Miyagawa 2012: 86, ex. (14)): (78) a. Peter-wa hataraki-mas-i-ta P-TOP work-MAS-PST ‘Peter worked’ b. Peter-wa hatarai-ta P-TOP work-PST ‘Peter worked’ 240 The clausal complement of non-factive predicates can accommodate mas (79a) (Harada 1976, ex. (102b)) while that of factive predicates cannot (79b). Some verbs can take both to and koto. When they take the latter, they cannot accommodate mas (79c): (79) a. Taroo-wa [Hanako-ga ki-mas-i-ta to] it-ta T-TOP H-NOM come-MAS-PST CNONFACT say-PST ‘Taro said that Hanako came’ b. Taroo-wa [Hanako-ga kita/*ki-mas-u koto]-o hitei-sita T-TOP H-NOM came/come-MAS-PRS C FACT-ACC deny-PST ‘Taro denied that Hanako will come’ c. Taroo-wa [Hanako-ga kita/*ki-mas-i-ta koto]-o hookokusi-ta T-TOP H-NOM came/come-MAS-PST C FACT-ACC report-PST ‘Taro reported the fact that Hanako came’ Since the politeness marker indicates the speaker’s intention towards the hearer, it involves projections relating to the speaker and hearer. Miyagawa therefore proposes to adapt the proposal by Speas & Tenny (2003) to subordinate clauses in Japanese such that non-factive predicates select for the Speech Act Projection (saP, SAP shell) whereas factive predicates select for a regular CP, i.e., a CP without saP/SAP. saP/SAP in Japanese is represented as in (80): 241 (80) An important note on Miyagawa (2012), which will play a role in the discussion regarding Turkish factive verbs, is that it also argues that there is a factivity operator in factive complements to account for ban on topicalization in these clauses via operator movement. This operator, however, is within the CP, thus factive clauses still have a smaller structure than non- factive clauses, which have a saP embedding a CP. Miyagawa adapts the proposal by Haegeman (2007) as in (81) for Japanese factive clauses: (81) [CP [OPi C…[ FP ti [TP …]]] Although there are various proposals on the structure of non-factive predicates, the cited recent studies agree on one point: non-factives involve more structure than factives. Now the question is where the present study fits. Two points from the literature are addressed: (i) the defaultness of factivity and (ii) reduced structure in factive predicates. I conclude that factivity can indeed be the default reading and therefore also structurally reduced, but true factive verbs select for a 242 structure as complex as non-factives in Turkish. In other words, there are two ways factivity can arise: by default and by structure. Before I move onto this discussion, however, it is important to clarify that although I assume extra structure in non-factives (=matrix subject-oriented verbs) in Turkish, this is not related to the availability of the speaker in these verbs, contra Haegeman (2006) and Miyagawa (2012). The argumentation goes as follows: İyi ki ‘fortunately’ is a speaker-oriented adverb in Turkish. It expresses the epistemic stance of the speaker towards the matrix assertion in matrix clauses: (82) Ali iyi ki gel-di A good that come-PST ‘Fortunately, Ali came’ In embedded contexts, the matrix subject is the sole evaluator of this adverb, indicating that no speaker-related projection is available in embedded clauses in Turkish. When the embedded clause is formed with –DIK, it does not allow evaluation of iyi ki ‘fortunately’ by the matrix subject regardless of the embedding verb type, 106 hence the unavailable readings in (83). The only evaluation is by the speaker but not towards the embedded content. The adverb rather modifies the matrix verb in this case, as indicated by the available reading in (83): (83) a. Esra [Ali-nin iyi ki gel-diğ-in]-i bil-iyor E A-3GEN good that come-DIK-3S-ACC know-IMPF ‘#Esra knows [that Ali fortunately came]’ (matrix subject evaluation of iyi ki) ‘It is fortunate that Esra knows [diye Ali came]’ (speaker evaluation of iyi ki) 106 There is some variation in the judgments for these data, but this is the tendency. 243 b. Esra [Ali-nin iyi ki gel-diğ-in]-i düşün-üyor E A-3GEN good that come-DIK-3S-ACC think-IMPF ‘#Esra knows [that Ali fortunately came]’ (matrix subject evaluation of iyi ki) ‘It is fortunate that Esra knows [diye Ali came]’ (speaker evaluation of iyi ki) If the embedded clause is a diye clause, iyi ki can be evaluated by the matrix subject when the embedded content is modified by the adverb. When the speaker is the evaluator, iyi ki must modify the matrix verb, similar to the case in –DIK clauses. 107 This pattern again holds regardless of the type of the embedded verb: (84) a. Esra [Ali iyi ki gel-di diye] bil-iyor E A good that come-PST diye know-IMPF ‘Esra knows [diye Ali fortunately came]’ (matrix subject evaluation of iyi ki) 108 ‘It is fortunate that Esra knows [diye Ali came]’ (speaker evaluation of iyi ki) b. Esra [Ali iyi ki gel-di diye] düşün-üyor E A good that come-PST diye think-IMPF ‘Esra thinks [diye Ali fortunately came]’ (matrix subject evaluation of iyi ki) ‘It is fortunate that Esra thinks [diye Ali came]’ (speaker evaluation of iyi ki) Although the embedded clause types yield some difference, what is common between (83) and (84) is that the speaker evaluation is not available for embedded content (i.e., the speaker can only evaluate the matrix verb). I take it to mean that speaker projection is not available in 107 I assume in these cases, the embedded subject and the adverb are both in the matrix clause. The embedded subject probably undergoes a discourse-related movement such as topicalization. 108 Imagine a context where the speaker and her friend convinced Esra that Ali’s arrival is a good thing while in actuality, it is not. Or adding a previous context like the following helps to judge this reading: “The speaker is angry that Ali attended the party. Esra is happy about it.” See Chapter 2, example (74). 244 embedded clauses, arguing against saP in Miyagawa (2012) and also against the speaker-deixis analysis by Haegeman (2006) for Turkish. To complete the argumentation, I will show that the same pattern holds with what Miyagawa argues are true speaker-oriented adverbs based on Amano (1999). Amano differentiates between ‘attitudinal’ adverbs such as apparently, certainly, definitely, evidently etc. and ‘style’ adverbs such as frankly, truthfully, honestly (In this classification, iyi ki falls under the former group, so it is not a truly speaker-oriented adverb). Amano argues that attitudinal adverbs indicate assertions and they are available just in the environments that allow root transformations in Hooper & Thompson (1973) (namely, their Class A, B and E verbs): 109 (85) a. Carl told me that this book certainly has the recipes in it. (Class A) b. Bill believes that certainly, John will lose the election. (Class B) c. *I doubt Kissinger certainly is negotiating for peace. (Class C) d. *I regret that I unfortunately attended the concert. (Class D) e. I know that Santa certainly has lost a lot of weight. (Class E) Style adverbs indicate the speaker’s style of expression (Greenbaum 1969) and they need not modify an assertion (Amano 1999). Amano argues that this group of adverbs are only compatible with Class A verbs of Hooper & Thompson: 110 109 I have two sources of suspicion for the data in (85) and (86): some native speakers do not agree with the judgments. Frankly leads to a semantic oddity and with honestly, some of these options (e.g., (85d-e)) and some find (85c) and (85f) acceptable. The other suspicion is the use of a first-person subject in some and third person in others. This might affect the judgments in non-trivial ways, given that the target item is speaker-oriented. Yet, I will take the results in their face value for sake of argumentation and accept the division Amano made as it is. 110 Notice in (86) that style adverbs in English are available only in the environment where the Japanese politeness marker –mas- is available, indicating that saP is available with these adverbs just as it is with –mas-. This is because these adverbs are argued to be the only adverbs that are strictly speaker-oriented because they have the speaker as one of their arguments (Bellert 1977). 245 (86) a. She said that honestly she does not know anything about their plans. (Class A) b. *Bill believes that honestly, John will lose the election. (Class B) c. *I doubt Kissinger frankly is negotiating for peace. (Class C) d. *I regret that I frankly attended the concert. (Class D) e. *I know that Santa honestly has lost a lot of weight. (Class E) It turns out that in the Turkish case, style adverbs are always speaker-oriented, and therefore need to modify the matrix verb (i.e., they cannot modify the embedded content) regardless of the complement clause type as well as the embedding verb type ((87)-(88)). This strengthens the previous conclusion that speaker is not represented in either of the embedded clauses we are concerned with in the present study: 111 (87) a. Esra [Ali-nin açıkçası gel-diğ-in]-i bil-iyor E A-3GEN frankly come-DIK-3S-ACC know-IMPF ‘#Esra knows [that Ali frankly came]’ (matrix subject evaluation of açıkçası) ‘Frankly, Esra knows [diye Ali came]’ (speaker evaluation of açıkçası) b. Esra [Ali-nin açıkçası gel-diğ-in]-i düşün-üyor E A-3GEN frankly come-DIK-3S-ACC think-IMPF ‘#Esra knows [that Ali frankly came]’ (matrix subject evaluation of açıkçası) ‘Frankly, Esra knows [diye Ali came]’ (speaker evaluation of açıkçası) (88) a. Esra [Ali açıkçası gel-di diye] bil-iyor E A frankly come-PST diye know-IMPF 111 A question that arises with respect to how açıkçası ‘frankly’ appears at all in a diye clause given that the word order seems to be neutral and thus açıkçası ‘frankly’ is in the embedded clause. This is a problem because as was argued in Chapter 2 and 3 that diye clauses are strictly matrix-subject-oriented. But notice that the meaning of the adverb is such that it modifies matrix content. Thus, I will assume that the adverb is actually not in the embedded clause but is generated in the matrix clause. The embedded subject must have undergone some movement such as topic movement to the matrix clause over the adverb. This yields a neutral-looking structure but one where some elements have been displaced over others. 246 ‘#Esra knows [diye Ali frankly came]’ (matrix subject evaluation of açıkçası) ‘Frankly, Esra knows [diye Ali came]’ (speaker evaluation of açıkçası) b. Esra [Ali açıkçası gel-di diye] düşün-üyor E A frankly come-PST diye think-IMPF ‘#Esra thinks [diye Ali frankly came]’ (matrix subject evaluation of açıkçası) ‘Frankly, Esra thinks [diye Ali came]’ (speaker evaluation of açıkçası) 4.3.2. Factivity can be default Recall that nominalized clauses do not have a factivity specification, as they adopt the factivity specification of the verb that embeds them. This is unlike diye clauses, which are non-factive regardless of the embedding verb. This is exemplified below (Özyıldız 2016): (89) Context: Trump won the election, but… a. #Tunç [Bernie-nin kazan-dığ-ın]-ı bil-iyor T B -3GEN win-DIK-3S-ACC know-IMPF ‘#Tunç knows [that Bernie won]’ b. Tunç [Bernie kazan-dı diye] bil-iyor T B win-PST diye know-IMPF ‘Tunç knows [diye that Bernie won]’ (90) Context: Trump won the election, but… a. Tunç [Bernie-nin kazan-dığ-ın]-ı düşün-üyor T B -3GEN win-DIK-3S-ACC think-IMPF ‘Tunç thinks [that Bernie won]’ 247 b. Tunç [Bernie kazan-dı diye] düşün-üyor T B win-PST diye think-IMPF ‘Tunç thinks [diye that Bernie won]’ We see in (89) that the embedding verb bil- also does not have a specification for factivity, as it has different interpretations depending on its complement, which is unlike a verb like düşün- ‘think’, which is always non-factive (=matrix subject-oriented). Neither –DIK nor bil- ‘know’ is specified as factive in (89a), and yet the resulting interpretation of their combination is factive. I take this to mean that when no specification is available for (non)factivity, we get a factive interpretation (i.e. speaker-orientedness) by default. This provides support for the literature on (non)-factivity arguing that factivity is the default interpretation and therefore has reduced structure (e.g., Nichols 2001, Haegeman 2006, de Cuba 2007, Miyagawa 2012). This type of factive interpretation is structurally reduced in that there is no extra CP encoding this reading. Next section, however, will show that factivity can be marked in some cases. 4.3.3. Factivity can be marked In this section, I argue that: (91) a. both non-factive and true factive verbs select for the same amount of structure, as opposed to factivity-underspecified verbs, which are bare VPs -no extra structure-. True factivity refers to structural encoding of factivity, as opposed to default factivity, which was discussed in the previous section. b. However, the properties of the extra structure are slightly different. 248 In arguing for extra structure in both factive and non-factive predicates, I bring together the arguments in traditional studies with more recent observations. 4.3.3.1. What are factive verbs? Factive verbs are those that trigger the “presupposition by the speaker that the complement of the sentence expresses a true proposition (K & K: 43)”. Presuppositions have projective meaning. That is, they survive under the scope of entailment-canceling operators (Simons et al. 2010). (92a) presupposes that the embedded proposition is true. (>: presupposition), which survives under entailment-canceling operations such as negation (92b), questions (92c), conditionals (92d) and modals (92e): (92) a. John knows that Mary loves him. >Mary loves John. b. John doesn’t know that Mary loves him. >Mary loves John. c. Does John know that Mary loves him? >Mary loves John. d. If John knows that Mary loves him, we are in trouble. >Mary loves John. e. John might know that Mary loves him. >Mary loves John. However, factivity has long been a controversial concept. The factivity inference in most factive verbs can be canceled. For example, Beaver (2004) showed that although the factivity inference of know cannot be canceled in conventional entailment-canceling contexts, it can be canceled in story-telling contexts (93) (cf. (2)): (93) a. Then, if the observer can get notification from the owner, the observer knows that the owner has still stayed in the net. If not, the observer knows that the observer has already disappeared since the reset occurred. 249 b. One day Chuang Tzu and a friend were walking by a river. “Look at the fish swimming about” said Chuang Tzu, “They are really enjoying themselves.” “You are not a fish,” replied the friend, “So you can’t truly know that they are enjoying themselves.” Other factive verbs are looser in that the conventional entailment-canceling environments can cancel their factivity inference. For instance, the conditional can cancel the inference in realize, discover (94b) but not that of regret (94a) (Karttunen 1974): (94) a. If I regret later that I have not told the truth, I will confess it to everyone. b. If I{realize, discover}later that I have not told the truth, I will confess it to everyone. The verbs whose factivity inference can be canceled in entailment-canceling environments such as negation, conditionals, questions etc. are known as the ‘semi-factive’ verbs, which include verbs like realize, discover. Regret and know are taken to be more ‘factive’ in that their factivity inference survives these environments. However, just like in the case of know in (92), the factivity inference of regret can be canceled in some other environments (95) (Karttunen 1974). (95) John didn’t regret that he had not told the truth. How could he have done that when he knew that what he had said was true? What seems to be the case is that the factivity inference of some verbs are harder to cancel than others (harder to cancel=survives negation, conditions, questions and modals). The easier-to- cancel ones are semi-factives and the harder ones are (true) factives. At the end of the day, however, they can all be canceled and that is why some studies considered all cognitive factives as cancelable. The discussion on factivity-cancelation is therefore not at a place to draw 250 definitive conclusions on how to group verbs based on factivity in English, specifically based on where factivity inference survives and where it does not. The present study is helpful at this point because it aims to account for interpretive properties based on structural differences. This is a newer approach (following studies such as de Cuba 2007, Miyagawa 2012, among others). I show how this approach works after I discuss true factive verbs in Turkish, specifically how they differ from factivity-underspecified verbs, which led to the most debate in the literature. 4.3.3.2. Speaker-oriented (True Factive) Verbs in Turkish Speaker-oriented verbs in Turkish are an independent class of verbs based on their semantic and syntactic patterns. They differ from perspective-unspecified verbs in a number of ways. For one, their factivity inference, derived from speaker-orientedness, cannot be overridden. The factive inference in know, a perspective-underspecified verb in Turkish that does not report a knowledge report, can be overridden by factors such as focus in embedded clause whereas the one that arises with forget cannot be. To see this in example, contrast (96) and (97). If the main focus is on the embedding verb, factivity inference arises under know+DIK (96a), but not when an element in the embedded clause is focused (96b). 112 In other words, the factivity inference of know does not always project, indicating that it is not a lexical property of the verb. I argued in previous sections that it arises as the default reading. In other words, this inference is not structurally encoded. Forget and discover do not participate in this alternation (97), as they are always factive regardless of focus. This indicates that their factivity is encoded somewhere in the structure these verbs occur in and this differentiates them from the default factive reading. This 112 Focus in embedded clause introduces alternatives, which indicates truth is not established, which conflicts presumably with the factivity inference of the embedding verb. 251 structural endocing makes them true factives in Turkish. The following examples are from Özyıldız (2017). Capital letters indicate the locus of main stress. (96) a. Dilara [Ali-nin sigara iç-tiğ-in]-i BIL-IYOR D A-3GEN cigarette smoke-DIK-3S-ACC know-IMPF ‘Dilara knows [that Ali smokes cigarettes]’ b. Dilara [Ali-nin SIGARA iç-tiğ-in]-i bil-iyor D A-3GEN cigarette smoke-DIK-3S-ACC know-IMPF ‘Dilara knows [that Ali smokes cigarettes]’ (97) a. Dilara [Ali-nin sigara iç-tiğ-in]-i UNUT-TU/KEŞFET-TI D A-3GEN cigarette smoke-DIK-3S-ACC forget-PST/discover-PST ‘Dilara forgot/discovered [that Ali smokes cigarettes]’ b. #Dilara [Ali-nin SIGARA iç-tiğ-in]-i unut-tu/keşfet-ti D A-3GEN cigarette smoke-DIK-3S-ACC forget-PST/discover-PST ‘Dilara forgot/discovered [that Ali smokes cigarettes]’ This is similar to the observation for English effect of focus on factivity alternation is not unique to Turkish. When the verb discover is focused in English (98b), it is implied that the student is guilty (i.e., factive inference). When the complement is focused (98a), there is no such implication (Beaver 2004, Abrusán 2011). This does not hold for know (99), in which case factivity inference stays regardless of the position of focus: (98) a. If the TA discovers that [your work is plagiarized]F, I will be [forced to notify the Dean]F. 252 b. If the TA [discovers]F that your work is plagiarized, I will be [forced to notify the Dean]F. (99) a. If the TA knows that [your work is plagiarized]F, I will be [forced to notify the Dean]F. b. If the TA [knows]F that your work is plagiarized, I will be [forced to notify the Dean]F. Subcategorization is another important difference. Both true factive verbs like regret and semi- factives like realize take that-complements in English. In Turkish, however, verbs like forget, discover do not combine with diye while they freely combine with –DIK (100). Verbs like know, remember combine with diye to yield a non-factive meaning whereas they combine with –DIK to yield a factive (presuppositional) reading (101). Finally, verbs like think, dream combine with diye and –DIK to yield non-factive readings in both cases (102). Since there is no overlap in distribution, all belong to different categories: (100) a. Esra [Ali-nin gel-diğ-in]-i unut-tu/keşfet-ti E A-3GEN come-DIK-3S-ACC forget-PST/discover-PST ‘Esra forgot/discovered [that Ali came]’ b. *Esra [Ali gel-di diye] unut-tu/keşfet-ti 113 E A come-PST diye forget-PST/discover-PST ‘Intended: Esra forgot/discovered [diye Ali came]’ (101) a. Esra [Ali-nin gel-diğ-in]-i bil-iyor/hatırl-ıyor E A-3GEN come-DIK-3S-ACC know-IMPF/remember-IMPF ‘Esra knows/remembers [that Ali came] (>Ali came)’ 113 The only available reading here is if the diye clause is interpreted as an adjunct. See Chapter 5. 253 b. Esra [Ali gel-di diye] bil-iyor/hatırl-ıyor E A come-PST diye know-IMPF/remember-IMPF ‘Esra knows/remembers [diye Ali came] (>Ali came)’ (102) a. Esra [Ali-nin gel-diğ-in]-i düşün-üyor/hayal ed-iyor E A-3GEN come-DIK-3S-ACC think-IMPF/dream do-IMPF ‘Esra thinks/dreams [that Ali came] (>Ali came)’ b. Esra [Ali gel-di diye] düşün-üyor/hayal ed-iyor E A come-PST diye think-IMPF/dream do-IMPF ‘Esra thinks/dreams [diye Ali came] (>Ali came)’ Thus, speaker-oriented verbs are the only ones that cannot combine with diye among cognitive attitude verbs. This suggests that they form a group with their own unique properties. This is even more highlighted when we ask why speaker-oriented verbs are not compatible with diye. Is it due to a meaning clash between diye and these verbs? It might be a factor, but the real reason seems to have to do with the structure of the complement clause. Specifically, speaker-oriented verbs in Turkish do not take non-nominal complements. To understand this better, see the data set in (103), which shows another necessarily postverbal lexical complementizer-like item, ki ‘that’, which can combine with perspective-underspecified verbs (102b) and matrix subject- oriented verbs (103c) but not with speaker-oriented verbs (103a). Ki is the only post-verbal complementizer-like element in Turkish: 114 (103) a.*Ali unut-tu/unut-uyor ki Esra gel-di A forget-PST/forget-IMPF that E come-PST 114 But see Predolac (2017), who argues that ki clauses are not true complementation structures. Even if they are not true complements to verbs, however, the distinction between the three verb classes in (103) still remains. 254 ‘Intended: Ali forgot/keeps forgetting that Esra came’ b. Ali bil-iyor ki Esra gel-di A know-IMPF that E come-PST ‘Ali knows that Esra came’ c. Ali düşün-üyor ki Esra gel-di A think-IMPF that E come-PST ‘Ali thinks that Esra came’ Unut ‘forget’ is compatible with other nominalized complements (less acceptable with –mE than –(y)Iş but this is true for all three types; -mE clauses do not easily combine with cognitive verbs): (104) a. Bu ülke… Özal gibi Amerikan dost-lar-ın-ın… [savaş-ın this country Ö like American friend-PL-3S-3S war-3GEN eşiğ-in-e gel-me-sin]-i unut-muş değil 115,116 edge-3S-DAT come-mE-3S-ACC forget-INDPST not ‘This country has not forgotten that American allies like Özal came to the edge of war’ (lit. … American allies like Özal’s coming to the edge of war) 115 from https://www.evrensel.net/haber/125386/turkiye-yi-savasa-cekmeye-geliyor) 116 Another example (https://www.hepsiburada.com/altus-al-611-d-900w-dik-elektrikli-supurge-pm- HB0000044JMR-yorumlari): (Kargo biraz geç gel-di ama ürün güzel ol-unca) [kargo-nun geç gel-me-sin]-i unut-tu-m bile [package-3GEN late come-mE-3S-ACC forget-PST-1S even (The package arrived a bit late but since the product was nice) I forgot that the package came late’ (lit. package’s coming late) 255 b. Ali Esra-nın gel-iş-in-i unut-tu A E-3GEN come-(y)Is-3S-ACC forget-PST ‘Ali forgot (the way) Esra came’ The restriction posed by these verbs is therefore due to the structural properties of their complement clause rather than being related to the interpretation of diye. More specifically, speaker-oriented predicates only accept nominal complements. This is not a new finding. Factive predicates have been argued to have more nominal character than non-factives. The initial analysis by K & K, for example, already indicates that with the added head noun fact. Moreover, K & K observed that that-complementizer drop is allowed only in non-factives in English (105(a)-(b)) while it is obligatory in factives (105(c)-(d)): (105) a. John thinks (that) Mary stole the money. b. Bill said (that) Max was the author of the review. c. Jane regrets that/*Ø Bill won the lottery. d. It bothers Joe that/*Ø Mabel wants to tag along all the time. Webelhuth (1992) argues that this difference arises due to different categorizations of factive and non-factive predicates. Based on the argument that a clause headed by that has nominal properties due to the etymological relationship between the complementizer and the demonstrative that and the observation that only factives require the complementizer, he takes factive clauses to be nominal and non-factive clauses to be verbal (at least when they do not have the complementizer). Nichols (2001) provides cross-linguistic evidence that factive predicates are often marked nominally, as opposed to non-factive predicates. An example from Korean is in (106) (Nichols 256 2001: 45, ex. (56-57)). Korean factive predicates take a clausal complement either headed by the external noun head kes ‘thing’ (106a) or the nominalizer –um (106b) while non-factive predicates take a complement headed by non-nominal finite complementizer –ko (106c): (106) a. Chelswu-ka miswuk-i pokkwen-ey taŋchemtoy-n kes-ul C-NOM M-NOM lottery-at win-PST-SUBORD thing-ACC a-n-ta know-PRES-DECL ‘Chelswu knows that Miswuk won the lottery’ b. Chelswu-ka miswuk-i pokkwen-ey taŋchemtw-ayss-um-ul a-n-ta C-NOM M-NOM lottery-at win-PST-NMLZ-ACC know-PRES-DECL ‘Chelswu knows that Miswuk won the lottery’ c. Chelswu-ka miswuk-i ton-ul hwumchy-ess-ta-ko(*-lul) C-NOM M-NOM money-ACC stole-PST-DECL-C(-*ACC) sayŋkakhanta thinks ‘Chelswu thinks that Miswuk stole the money’ A less direct but still illustrative example comes from Hausa. Complement clauses of non-factive predicates in Hausa are derived from verbs (107(a)-(b)) whereas the complement clauses of factive predicates is headed by dà (107c), which is also used in nominal-headed relative clauses (107d) (Newman 2000, Nichols 2001): 257 (107) a. yârân sun tsayà: cêwa: sun maido: child-PL 3.PL.COMPLET insist C 3.PL.COMPLET repay manà dà kudì-nmù with money-our ‘The children insisted that they had paid us back our money’ cêwa: =’saying’ b. na: sâ râi wai zâi zo: 1.S.COMPLET expect C 3.M.FUT come ‘I expect that he will come’ wai’’it is said that’ c. mun yi farin cikì: dà kukà zo: 1.PL.COMPLET do pleasure C 2.PL.PRET come ‘We are happy that you came’ d. ya:rinyàr ~ dà ta faɗ:ì girl REL 3.F fall ‘The girl who fell’ Kastner (2015) argues that presuppositional predicates, a sub-part of which includes factive predicates, have complements whose highest level is a DP with a covert definite determiner. His discussion of the data revolves around the classification in (108) (Cattell 1978, Hegarty 1990), defined in (109) (Hegarty 1992: fn 12): 258 (108) a. Non-Stance: regret, know, remember, realize, notice etc. 117 b. Response Stance: deny, accept, agree, admit, confirm, verify etc. c. Volunteered Stance: think, suppose, assume, claim, suspect etc. (109) Non-Stance: Presuppose, as a matter of conventional meaning, that the content of the complement is part of the body of background beliefs. This class includes most standard factive verbs. Response Stance: The content of the complement is under consideration for inclusion in the body of background beliefs. This class includes the non-factives confirm and verify, and the negative verbs deny and doubt. Volunteered Stance: The content of the complement is being offered by the speaker for inclusion in the body of background beliefs. Kastner argues that volunteered stance verbs take a CP complement. Non-stance and response stance verbs take a DP complement with a covert D head, which in turn takes the embedded CP as complement. Since the head of this DP is filled with a covert determiner, presuppositional verbs induce weak islands while non-presuppositional verbs are not islands at all (Hegarty 1990) (weak islands are those where arguments can be extracted but adjuncts cannot): 118 (110) a. What do you think (that) John stole ____? b. Why do you think John stole the cookies ____? c. What do you remember/deny that John stole ____? d. #Why do you remember/deny that John stole the cookies _____? 117 Notice that the class of Non-Stance verbs include both factives and semi-factives, which I have shown behave differently in Turkish. But Kastner’s arguments are relevant in that true factives in Turkish require nominal arguments. True factives fall in the category (a) of (108), which Kastner also argues have nominal arguments. 118 Notice that the data in (110) and (111) also suggest that verbs like think have a larger complement through which elements can be extracted and topicalized. This is largely in line with the proposal in this chapter that the complement to these verbs are not opaque domains due to the extra structure available in their complement clauses. 259 Likewise, fronting is possible only with non-presuppositional verbs, as presuppositional verbs induce weak islands due to the existence of the DP layer above the embedded CP: (111) a. John thinks that [this book, Mary read] b. *John regrets that [this book, Mary read] Overall, then, there is good evidence that factive predicates, however they are categorized, subcategorize for nominal complements. Going back to Turkish, I suggest that speaker-oriented predicates in Turkish select for CPspeaker, which introduces an operator, OPspeaker (as in Miyagawa 2012) on its head that has a PRO as its subject in SpecCPspeaker coindexed with the speaker, reporting the embedded clause from the perspective of the speaker. The availability of extra CP in both speaker-oriented and matrix subject-oriented predicates accounts for the generalization in (91a) that they both have an equal amount of added structure. But this time, the perspective holder is the speaker and the actual world is therefore necessarily included in the evaluation worlds (Nichols 2001). This extra CP takes a DP as its complement. This is represented in (112), defined in (113). 260 (112) (113) [[OP speaker]] c,g =λp. λxe. λws. p(w)=1. ∀<w’, x’>:<w’, x’>∈FORGET x,w. p(w’)(x’)=1 where x is the speaker and FORGET x,w = {<w’, x’>: w’ is compatible with x’s mental state in w and x’ is the individual in w’ that x identifies as himself} The operator is similar to OP matrixsubject and OPdiye in that it introduces a perspective. OP matrixsubject, OPspeaker and OPdiye are all perspective operators, differing in (i) whose perspective they take as arguments (OP matrixsubject and OPdiye, the matrix subject versus OPspeaker, the speaker) and (ii) whether they have a PRO in their specifier (yes for OPdiye and OPspeaker and no for OP matrixsubject). Going back to the correlation between speaker-oriented predicates and nominal complements, I assume that it is the structural encoding of the actual world in OPspeaker that requires a nominal complement. How that requirement comes about is a topic for future research, however, perhaps 261 it has to do with the fact that the actual world is a specific world, and this might correspond to a DP in the syntax, as DP is related to specificity in the syntax as well. Future research can confirm or refute if this intuition is on the right track as well as how to formulate this semantically. One implication of this proposal is that the complements of true factives must be opaque, since their complement domain is marked with an operator with a PRO subject, just like diye clauses under factivity-underspecified verbs. The prediction is borne out, since true factives do not allow binding across domains (114) –for some speakers including me-, NPI-licensing (115) across domains and specific reading of embedded quantifiers (116): (114) %Çocuklari [Ali-nin birbirlerin-ii eşleştir-diğ-in]-i unut-tu children A-3GEN each.other-ACC match-DIK-3S-ACC forget-PST ‘Childreni forgot [that Ali matched each otheri]’ (115) #Ali [kimse-nin gel-diğ-in]-i unut-ma-dı A no.one-3GEN come-DIK-3S-ACC forget-NEG-PST ‘Ali did not forget [that anyone came]’ (116) Pelin [burada biri-nin hasta ol-duğ-un]-u unut-tu P here someone-3GEN sick be-DIK-3SACC forget-PST ‘#Pelin forgot [that someone is sick here]’ (specific) ‘Pelin forgot [that there is someone sick here]’ (unspecific) Now notice how the present study can contribute to the clarification of the distinctions between factive and semi-factive verbs, an area which has to proved to be the most challenging. The study makes a clearer distinction between these verbs because it focuses on the structural differences first and connects the interpretative differences to the structural differences. For example, true 262 factives and factivity-underspecified verbs in Turkish are different because (i) they have different selectional properties but also (ii) different structural properties such as their complements being opaque domains (i.e., –DIK clauses are not opaque under factivity-underspecified verbs and are opaque under true factive verbs). Both of these follow if we rely on the meaning of these verbs, i.e., that true factives encode evaluation in the actual world by the speaker, reporting the speaker’s perspective on the embedded clause. Semi-factives (i.e., factivity-underspecified verbs) do not have the parameters to establish evaluation domains, hence their differences from true factives. Taken together, I make a three-way distinction between cognitive verbs in Turkish, represented in (117): 263 (117) a. Speaker-oriented (e.g., forget) b. Matrix subject-oriented (e.g., think) c. Perspective-Underspecified (e.g., know) (117c) represents perspective-underspecified verbs such as bil- ‘know’, which do not have an extra CP encoding perspective. These verbs are bare VPs and they get interpreted as factive by default (see Nichols 2001, among others, for factivity being the default interpretation) unless there is another element or factor contributing (non)-factivity (e.g., diye clauses, focus etc). Speaker-oriented (true factive) verbs such as unut- ‘forget’ as in (117a) and matrix subject- 264 oriented verbs such as düşün- ‘think’ as in (117b) both have evaluation worlds specified in different ways in their complement clauses via the extra CP in their complements. This proposal indicates that both perspective is always marked, whether it is that of the speaker’s (=factivity) or of the matrix subject (=non-factivity), albeit in somewhat different ways. This is a way of reconciling facts discussed in traditional studies indiciating extra structure in factive predicates (e.g., K & K) and those in more recent studies arguing for extra structure in non-factive verbs (e.g., de Cuba 2007, Kastner 2015). The differences that emerge with respect to the opaqueness of the complement clause in true factive and non-factive predicates are attributed to the nature of the extra CP: CPspeaker has a PRO subject, forming an opaque complement domain, whereas CP matrixsubject has an empty specifier, resulting in a transparent domain. All in all, this account can handle both the differences and the similarities between true factives and non-factives. 4.4. Summary and Conclusion This chapter has discussed diye and –DIK clauses when they are embedded under matrix subject- oriented verbs like düşün- ‘think’, as opposed to perspective-underspecified verbs like bil- ‘know’, and later also contrasted with speaker-oriented verbs like unut- ‘forget’ in Turkish. I have shown that the differences between diye and –DIK clauses are neutralized under matrix subject-oriented verbs such that both report evaluation from the matrix subject’s conscious perspective. I attributed this neutralization to an extra structure such verbs select for that serves to mark an evaluation domain associated with the matrix subject. Since the extra structure is higher than all of the complement clause, it forces it to be interpreted from the matrix subject’s perspective. This is unlike verbs like bil- ‘know’ and hatırla- ‘remember’, which lack this 265 structure, and matrix subject’s perspective is enforced only if the embedded clause marks it as such, as in the case of diye clauses. I have also argued that diye clauses are not opaque domains under matrix subject-oriented verbs (unlike when they are under perspective-underspecified verbs) due to the existence of this structure, which extends the domain but keeps the edge empty for interaction between an external probe and the embedded clause. After I discussed the structure of matrix subject-oriented verbs in Turkish, I put together the findings about fperspective-underspecified verbs like bil- ‘know’ and hatırla- ‘remember’ in Turkish as well as speaker-oriented verbs like keşfet- ‘discover’. This allowed us to see a three- way distinction between cognitive verbs in Turkish based on perspective: (i) speaker-oriented verbs, whose complements mark evaluation in the actual world (e.g., keşfet ‘discover’) related to the speaker, (ii) matrix subject-oriented verbs, whose complements mark evaluation in the logophoric alternatives of the matrix subject (e.g., düşün ‘think’) and (iii) verbs underspecified for perspective, which are bare VPs and do not mark perspective (e.g., bil- ‘know). The complement to the last group of verbs gets evaluated in the actual world by default, unless the complement marks a different specification for evaluation. That is why –DIK clauses under these verbs get interpreted as factive although –DIK clauses themselves are not factive. That is also why diye clauses under these verbs can get a non-factive interpretation. As for the first group, I argued that they also come with the same extra structure as in matrix subject-oriented verbs, but with a different evaluation domain specification: while the extra structure under matrix subject- oriented verbs mark the logophoric alternatives of the matrix subject as the evaluation domain, speaker-oriented verbs mark the actual world as the evaluation world and the speaker as the evaluator. This difference might be a way to account for their requirement for a nominal complement in Turkish (and possibly cross-linguistically). I also argued that one difference 266 between the extra structure in speaker-oriented verbs and the one in matrix subject-oriented verbs is that only the former has a PRO subject in its specifier. This accounts for the fact that only the complement to the former forms an opaque domain, and in fact it turns an otherwise transparent domain of -DIK clauses into opaque domains. I argued that this proposal combines two competing lines of study in the structure of (non-)factivity: traditional studies arguing for extra structure marking factivity in factive verbs (e.g., Kiparsky & Kiparsky 1971) and more recent studies arguing for extra structure in non-factive verbs (e.g., de Cuba 2007, Kastner 2015) but it also retains that these two verb classes come with their differences. 267 CHAPTER 5 CONCLUSION This dissertation undertook the endeavor of discovering the role of perspective in the syntax of Turkish via a detailed analysis of complement clauses formed with the nominalizer –DIK and the non-nominal diye in light of recent studies arguing that perspective plays a role in syntactic derivation and interactions (e.g., Sundaresan 2012, Charnavel 2019) rather than being a pure discourse-pragmatic phenomenon (Clements 1975, Kuno 1987). What differentiates the present study is its broad focus on the perspectival domain. In other words, the present study focuses on elements of different natures in the perspectival domain instead of a single type of element. This helped bring out various properties of the perspectival domain that have shed light on phenomena that are otherwise difficult to explain. The conclusions are summarized below. Please keep in mind that I define (non)factivity based on perspective: non-factive=matrix subject-oriented (i.e. matrix subject’s perspective), factive=speaker-oriented (i.e. speaker’s perspective), factivity-underspecified/semi-factive=perspective-underspecified (i.e. no perspective): (1) a. diye is the overt realization of a perspectival operator in the head of Pers(pective)P, which reports the embedded content from the first-person perspective of the closest matrix subject. Incidentally, the complement of PersP is the embedded CP. SpecPersP hosts PRO, which saturates the individual argument of the diye operator via its association with the matrix subject. 268 b. Each phrasal content (e.g., adjectives in AP, temporal and lexical elements in PP) in the diye clause with comes with an anaphoric world variable that needs to be bound by the closest operator (à la Percus 2000), which is OPdiye. c. diye clauses under perspective-underspecified verbs such as bil- ‘know’ in Turkish form an opaque domain and as such they do not allow interactions between the matrix clause and the embedded clause such as NPI-licensing cross clauses. This is unlike – DIK clauses, which do not form opaque domains. d. pro that comes with the first-person agreement (and presumably epithets) are optionally bound by the speaker or any of the matrix subjects due to the pronominal inside them that come with a topic feature that needs to be valued by the topic probe in the main clause. I argue that these exceptions to the opaqueness of diye clauses are systematic and diye clauses are in fact selectively opaque (Keine 2019, Poole 2020). The systematicity is due to topichood. Since the topic probes are higher than PersP, the full size of diye clauses, on the selective opacity account of Keine (2019) and Poole (2020), it can probe into the diye clause and into the pro with the unvalued topic feature within the diye clause, as can other higher probes such as Int(errogative)P. e. Some elements such as overt pronouns and locative expressions such as burada ‘here’ must be associated with the utterance context regardless of whether they appear in a – DIK or diye clause due to the lexical encoding for such within these elements, following the traditional Kaplanian account. Since these elements are single words, 269 this might be evidence for the lexical encoding. f. When the embedding verb is matrix subject-oriented, phrasal elements in both –DIK and diye clauses must be reported from the perspective of the matrix subject and the optionality pattern and the pattern with elements that are lexically encoded for utterance context remains the same. However, diye clauses are no longer opaque domains under such verbs. I attributed this to the extra structure that matrix subject- oriented verbs select for, following studies such as de Cuba (2007) and McCloskey (2005). This structure is CP matrixsubject, which takes PersP as its complement. CP matrixsubject hosts an operator much like the diye operator but with one difference: CP matrixsubject has an empty specifier and does not form an opaque domain. This allows interactions between the matrix clause and the embedded clause. g. Speaker-oriented verbs cannot combine with diye. They require nominal complements. –DIK clauses under them are opaque domains, unlike the case under perspective-underspecified verbs. I suggested that this is because speaker-oriented verbs also come with extra structure, CPspeaker, with a similar operator. This time, however, the specifier is filled with PRO that takes the speaker as its antecedent. This leads us to have a complete evaluation domain and an opaque domain in syntax. These points point to the systematicity of diye clauses in Turkish as well as in the structure of embedding verbs. Diye consistently has the same structure, involving the same single operator. The differences derive from the bound elements and embedding verbs. 270 There are many remaining questions. For one, how can the semantics capture these generalizations? Since I investigated elements that are not semantically combinable in any easy way, coming up with a single semantics for the diye clause in all these situations is a challenging task. For present purposes, I assumed a generalized operator, but future research should look into what the implications of this proposal can be. Other questions arise when we extend the data. I only focused on cognitive attitude verbs. What happens with, for example, emotive verbs? Here are some sample data: Like perspective- underspecified verbs, emotives verbs allow alternation in the interpretation of their complements. When they combine with –DIK, the inference is that the embedded proposition is true (2a). When they combine with diye, this inference is not available (2b). (2) a. Ali [Esra-nın gel-diğ-in]-e üzül-dü A E-3GEN come-DIK-3S-DAT be.sad-NEG-PST ‘Ali was not sad [that Esra came]’ >Esra came …#ama Esra gelmedi …#but Esra did not come b. Ali [Esra gel-di diye] üzül-dü A E come-DIK diye be.sad-PST ‘Ali was sad [diye Esra came]’ >Esra came …ama Esra gelmedi …but Esra did not come However, unlike perspective-underspecified verbs (3a) and like speaker-oriented verbs (3b), emotive verbs (3c) seem to force their complement clauses into opaque domains: 271 (3) a. Ali [kimse-nin gel-diğ-in]-i bil-m-iyor A no.one-3GEN come-DIK-3S-ACC know-NEG-IMPF ‘Ali knows [that no one came]’ b. ??Ali [kimse-nin gel-diğ-in]-i unut-ma-dı/unut-m-uyor A no.one-3GEN come-DIK-3S-ACC forget-IMPF/forget-NEG-IMPF ‘Intended: Ali forgot/keeps forgetting [that no one came]’ c. ??Ali [kimse-nin gel-diğ-in]-e üzül-me-di/üzül-m-üyor A no.one-3GEN come-DIK-3S-DAT be.sad-NEG-PST/be.sad-NEG-IMPF ‘Intended: Ali was sad/is sad [that no one came]’ Thus, emotive verbs are not like any of the verb categories I have looked into (but diye behaves similar under them). Their behavior therefore requires further detailed analysis. It is likely that diye under emotive verbs has a causative meaning, as the causation can be challenged. Note that I use a clear causative marker için ‘for’ in the continuation: (4) Ali [Esra gel-di diye] üzül-me-di A E come-DIK diye be.sad-NEG-PST ‘Ali was not sad [diye Esra came]’ …Merve git-tiğ-i için üzüldü …but he got sad because Merve left Another category of verbs is speech-verbs, which I have touched briefly in Chapter 3. De- ‘say’, which diye derives from, and the other verb for ‘say’ in Turkish, söyle, do not combine with diye if diye is immediately next to the verb (5a). A diye clause can appear further from the verb (5b), in which case the element that is next to the verb is focused: 272 (5) a. *Ali [Esra gel-di diye] de-di/söyle-di A E come-PST diye say-PST/say-PST ‘Intended: Ali said [diye Esra came]’ b. [Esra gel-di diye] Ali de-di/söyle-di E come-PST diye A say-PST/say-PST ‘It was Ali who said [diye Esra came]’ Other speech verbs combine with diye, where the diye clause expresses the content of speech: (6) Ali [Esra gel-di diye] bağır-dı A E come-PST diye shout-PST ‘Ali shouted [diye(=saying) Esra came]’ Speech verbs also report the embedded content as the matrix subject says, similar to first-person perspective reporting under cognitive verbs. But they have other different patterns than cognitive verbs. For example, as discussed briefly in Chapter 2 based on Yıldırım-Gündoğdu (2017a), they enforce pro to refer to the matrix subject (7b): (7) a. Ahmeti [ben ödev-i bitir-di-m diye] düşün-üyor A I homework-ACC finish-PST-1S diye think-IMPF ‘Ahmet thinks that hei/I finished the homework.’ b. Ahmeti [ben ödev-i bitir-di-m] d-iyor A I homework-ACC finish-PST-1S say-IMPF ‘Ahmet says that hei/*I finished the homework.’ 273 Thus, diye, when combined with speech verbs, deserves separate treatment as well. I refer the reader to Yıldırım-Gündoğdu (2017a) for this combination. When diye combines with nouns, it is difficult to analyze it as a perspective marker. Evidence that it behaves differently than under cognitive verbs comes from (8c), where the pro must refer to the speaker: (8) a. Ali diye bir adam A diye one man ‘A man named Ali’ b. Ali gel-di diye bir dedikodu A come-PST diye one gossip ‘A gossip [diye Ali came]’ c. Ali [pro gel-di-m diye] bir dedikodu duy-muş Ali come-PST-1S diye one gossip hear-INDPST ‘Ali heard a gossip [diye *he/I came]’ The overall pattern in these data is that diye heavily relies on the embedding verb for meaning. What it generally provides is a perspective different from that of the speaker’s. For example, it reports the matrix subject’s perspective on the causation. Under speech verbs, it reports what the matrix subject said, which the speaker uses to indicate that it is not what s/he himself/herself said. When there is no immediate embedding verb, as in the case when they appear with a noun head, they might have a different structure, as the noun argument would not saturate the diye operator I proposed. Indeed, diye under nominals behave radically different. How (or whether) these differences can be integrated into a single analysis is a topic for future research. 274 A note on when diye appears as a reason clause. There are two reason clauses in Turkish: one formed with diye and one formed with için ‘for’. The two behave differently, however. For example, only diye allows optional reference for pro: (9) a. Ali [hasta-yım diye] doktor-a git-ti A sick -1S because doctor-DAT go-PST ‘Ali went to the doctor [because/saying that he/I is/am sick]’ (I=S, AH) b. Ali [hasta ol-duğ-um için] doktor-a git-ti A sick be-DIK-1S for doctor-DAT go-PST ‘Ali went to the doctor [because/saying that I/*he is sick]’ (I=S, *AH) Diye seems to have a similar function as a reason clause in that it reports the perspective of the closest individual despite the lack of an immediate embedding verb. I believe that this is not alarming, however. The main function of the verb in my account was to provide the individual argument. If we assume that reason clauses are CAUS(ITIVE)P(HRASE)s with an individual argument for who evaluates the cause and the embedded content and they can take PersP as their complement (i.e., CAUSP>PersP), then CAUSP would do what VP does in my account, where VP provides the antecedent for the individual argument. Için cannot allow optionality because it is a CAUSP that does not combine with PersP. This type of analysis would be similar to Charnavel (2018) for English and French and therefore has a precedent for other languages. This line of proposal can be tested for Turkish in future research and be refuted or confirmed. The potential parallels are, however, promising. 275 Diye is a widely-used element in Turkish, so it is difficult to combine all of its functions under the same analysis. But I believe this overview provides some hopes that some common elements can be found and a unifying account of diye in Turkish is possible. 276 REFERENCES Abels, Klaus. 2012. The Italian left periphery: A view from locality. Linguistic Inquiry, 43(2), 229-254. Abrusán, Márta. 2011. Predicting the presuppositions of soft triggers. Linguistics and Philosophy, 34(6), 491-535. Abusch, Dorit. 2002. Lexical alternatives as a source of pragmatic presuppositions. In Semantics and Linguistic Theory, 12, 1-19. Abusch, Dorit. 2010. Presupposition triggering from alternatives. Journal of Semantics, 27(1), 37-80. Adesola, Oluseye. 2005. Pronouns and Null Operators. Rutgers University (Doctoral dissertation). Akkuş, Faruk. 2019. Un-shifting indexicals. Unpublished Manuscript, University of Pennsylvania. Aikhenvald, Alexandra. Y. 2006. Serial verb constructions in typological perspective. Serial Verb Constructions: A Cross-Linguistic Typology, 1-68. Amano, Masa-chiyo. 1999. Gengoyoso-no ninka: Doshi, meishiku, fukushi (On the licensing of linguistic elements: Verbs, noun phrases, and adverbs). Kenkyusha, Tokyo. Anand, Pranav. 2006. De De Se. MIT (Doctoral Dissertation). Anand, Pranav. 2007. Dream report pronouns, local binding, and attitudes 'de se'. Proceedings from Semantics and Linguistic Theory, 17, 1-18. Anand, Pranav. & Natasha Korotkova. 2021. How to theorize about subjective language: a lesson from ‘de re’. Linguistics and Philosophy, 1-63. Anand, Pranav. & Andrew Nevins. 2004. Shifty operators in changing contexts. In Semantics and Linguistic Theory, 14, 20-37. Baker, Mark. 2014. Pseudo noun incorporation as covert noun incorporation: linearization and crosslinguistic variation. Language and Linguistics, 15(1), 5-46. Bassac, Christian. & Mehmet Cicek. 2013. On the derivation of nominals in-(y)Iſ and-mE in Turkish. Lingua, 133, 230-261. Beaver, David. I. 2004. The optimization of discourse anaphora. Linguistics and Philosophy, 27(1), 3-56. Bellert, Irena. 1977. On semantic and distributional properties of sentential adverbs. Linguistic Inquiry, 8(2), 337-351. 277 Bogal-Allbritten, Elizabeth. A. 2016. Building meaning in Navajo. UMASS (Doctoral dissertation). Bruening, Benjamin. 2001. Syntax at the Edge: Cross-Clausal Phenomena and the Syntax of Passamaquoddy. MIT (Doctoral dissertation). Bruening, Benjamin. 2001. QR obeys Superiority: Frozen scope and ACD. Linguistic Inquiry, 32(2), 233-273. De Cat, Cécile. 2007. French dislocation: Interpretation, syntax, acquisition, Vol 17, Oxford University Press. Cattell, Ray. 1978. On the source of interrogative adverbs. Language, 61-77. Charnavel, Isabelle. & Victoria Mateu. 2015. The clitic binding restriction revisited: Evidence for antilogophoricity. The Linguistic Review, 32(4), 671-701. Charnavel, Isabelle. & Dominique Sportiche. 2016. Anaphor binding-What French inanimate anaphors show. Linguistic Inquiry, 47(1), 35-87. Charnavel, Isabelle. 2017. Exempt anaphors and logophoricity in French. In Ruth E. V. Lopes, Juanito Ornelas de Avelar and Sonia M. L. Cyrino (Eds.), Romance Languages and Linguistic Theory 12: Selected papers from the 45th Linguistic Symposium on Romance languages (LSRL45), Campinas, Brazil, 15-28. John Benjamins Publishing Company. Charnavel, Isabelle. & Dominique Sportiche. 2017. Simplex yet local. In NELS 47: Proceedings of the 47th annual meeting of the North East Linguistic Society (1), 157-172. Amherst: University of Massachusetts, Graduate Linguistic Student Association. Charnavel, Isabelle. 2018a. Long-distance binding of French reflexive soi: First-person oriented logophoricity. In Lori Repetti and Francisco Ordóñez (Eds.) Romance Languages and Linguistic Theory 14, Selected Papers from the 46 th Linguistic Symposium on Romance Languages (LSRL46), Stony Brook, NY, 21-34. John Benjamins Publishing Company. Charnavel, Isabelle. 2018b. Perspectives in causal clauses. Natural Language & Linguistic Theory, 37(2), 389-424. Charnavel, Isabelle. 2019. Locality and Logophoricity: A Theory of Exempt Anaphora. Oxford University Press. Charnavel, Isabelle. 2020. Logophoricity and locality: A view from French anaphors. Linguistic Inquiry, 51(4), 671-723. Chomsky, Noam. 1981. On the Representation of Form and Function. 1(1), 3-40. https://doi.org/10.1515/tlir.1981.1.1.3 Chomsky, Noam. 1986. Barriers (Vol. 13). MIT Press. 278 Chomsky, Noam. 2001. Derivation by phase. In M. Kenstowicz (Ed.), Ken Hale: A Life in Language, 1-52. Cambridge, MA: MIT Press. Cinque, Guglielmo. 1990. Types of Ā-dependencies. MIT press. Clements, George. N. 1975. The logophoric pronoun in Ewe: Its role in discourse. Journal of West African Languages, 10, 141-177. De Cuba, Carlos. F. 2007. On (non) factivity, clausal complementation and the CP-field. State University of New York at Stony Brook (Doctoral Dissertation). De Cuba, Carlos. F. & Barbara Ürögdi. 2001. The role of factivity in the syntax and semantics of embedded clauses. Unpublished Manuscript. Stony Brook University, NY. Dahlman, Colonna. R. 2015. Studies on Factivity, Complementation, and Propositional Attitudes. Lund University (Doctoral dissertation). Deal, Amy. R. 2018. Indexiphors: Notes on embedded indexicals, shifty agreement, and logophoricity. UMOP, 40, 59-86. Deal, Amy. R. 2020. A theory of indexical shift: meaning, grammar, and crosslinguistic variation (Vol. 82). MIT Press. Déchaine, Rose-Marie. & Martina Wiltschko. 2002. Decomposing pronouns. Linguistic Inquiry, 33(3), 409-442. Demirok, Ömer. 2019. A semantic characterization of Turkish nominalizations. In Proceedings of WCCFL (Vol. 36), 132-142. Somerville, MA: Cascadilla Proceedings Project. Diercks, Michael. 2013. Indirect agree in Lubukusu complementizer agreement. Natural Language & Linguistic Theory, 31(2), 357-407. Dikken, Marcel den. 2007a. Phase Extension Contours of a theory of the role of head movement in phrasal extraction, 33(1), 1-41. De Gruyter Mouton. Dikken, Marcel den. 2007b. Phase Extension: A reply. Theoretical Linguistics. 33. 133-163. Dubinsky, Stanley & Robert Hamilton. 1998. Epithets as antilogophoric pronouns. Linguistic Inquiry, 29(4), 685-693. Eckardt, Regine. 2015. The semantics of free indirect discourse: How texts allow us to mind- read and eavesdrop. Current Research in the Semantics/Pragmatics Interface, 31. 1-283. Erdal, Marcel. 1998. On the verbal noun in-(y)Iş. Doğan Aksan Armağanı, 53-68. Erguvanlı, Eser. 1984. The Function of Word Order in Turkish Grammar, 106. University of California Press. 279 Fox, Danny. 2000. Economy and semantic interpretation (Vol. 35). MIT press. Frascarelli, Mara. & Roland Hinterhölzl. 2007. Types of topics in German and Italian. In Kerstin Schwabe and Susanne Winkler (Eds.) On information structure, meaning and form: Generalizations Across Languages, 87-116. Ginzburg, Jonathan & Ivan. A. Sag. 2000. English interrogative constructions. Stanford Monographs in Linguistics. CSLI Publications, Stanford, CA. Giorgi, Alessandra., & Fabio Pianesi. 2001. Tense, attitudes and subjects. In Semantics and Linguistic Theory, 11, 212-230. Greenbaum, Sydney. 1969. Studies in English Adverbial Usage. London: Longman. Haegeman, Liliane. 2006. Conditionals, factives and the left periphery. Lingua, 116(10), 1651- 1669. Haegeman, Liliane. 2007. Operator movement and topicalisation in adverbial clauses. 41(3-4), 279-325. Mouton de Gruyter. Haïk, Isabelle. 1984. Indirect binding. Linguistic Inquiry, 15(2), 185-223. Harada, Shin-Ichi. 1976. Honorifics. In Masayoshi Shibatani (Ed.) Syntax and Semantics 5: Japanese generative grammar, 499-561. New York, NY: Academic Press. Harris, Jesse. A. 2009. Epithets and perspective shift: experimental evidence. In M. Biezma & J. A. Harris (Eds.), University of Massachusetts Occasional Papers: Papers in Pragmatics, 39, 49-76. GLSA Publishing. Amherst, MA. Hegarty, Michael. 1990. On adjunct extraction from complements. In Cheng, L.L.S., Demirdash, H. (Eds.) MIT Working Papers in Linguistics, 13, 101-124. Hegarty, Michael. 1992. Familiar complements and their complementizers: On some determinants of A’-locality. Unpublished Manuscript. University of Pennsylvania. Heim, Irene. 1991. Articles and definiteness. Published in German as ‘‘Artikel und Definitheit.’’ In Arnim von Stechow and Dieter Wunderlich (Eds.) Semantics: An International Handbook of Contemporary Research, 487–535. Berlin: de Gruyter. Von Heusinger, Klaus. & Jaklin Kornfilt. 2005. The case of the direct object in Turkish: Semantics, syntax and morphology. Turkic Languages, 9, 3-44. Hooper, Joan. B., & Sandra A. Thompson. 1973. On the applicability of root transformations. Linguistic Inquiry, 4(4), 465-497. Hornstein, Norbert. 1999. Movement and control. Linguistic Inquiry, 30(1), 69-96. 280 Huang, C. T. James., & C. S. Luther Liu. 2001. Logophoricity, attitudes, and ziji at the interface. Long-Distance Reflexives 33, 141-195. Ihsane, Tabea. & Genoveva Puskás. 2001. Specific is not definite. GG@G 2, 39-54. Jayaseelan, Karattuparambil. A. 1998. Blocking effects and the syntax of Malayalam taan. In R. Singh (Ed.), The Yearbook of South Asian Languages and Linguistics, 11-27. Jiménez Fernández, Ángel. L. 2009. On the composite nature of subject islands: A phase-based approach. SKY Journal of Linguistics, 22, 91-138. Kaplan, David. 1989. Demonstratives. In Joseph Almog, John Perry & Howard Wettstein (Eds.), Themes from Kaplan, 481-563. Kastner, Itamar. 2015. Factivity mirrors interpretation: The selectional requirements of presuppositional verbs. Lingua, 164, 156-188. Karttunen, Lauri. 1974. Presuppositions and linguistic context. Theoretical Linguistics, 1, 181- 194. Kayabaşı, Demet. & Murat Özgen. 2018. A phase-based account of NPI-licensing in Turkish. Poznan Studies in Contemporary Linguistics, 54(1), 83-113. Keine, Stefan. 2019. Selective opacity. Linguistic Inquiry, 50(1), 13-62. Kelepir, Meltem. 2001. Topics in Turkish Syntax: Clausal Structure and Scope. MIT (Doctoral Dissertation). Keshet, Ezra. R. 2008. Good Intensions: Paving Two Roads to a Theory of the De Re/De Dicto Distinction. MIT (Doctoral dissertation). Keshet, Ezra. & Florian Schwarz. 2019. De re/de dicto. The Oxford Handbook of Reference, 167- 202. Kiparsky, Paul. & Carol Kiparsky. 1971. Fact. In Danny Steinberg and Leon Jacobovits (Eds.) Semantics, An interdisciplinary reader in philosophy, linguistics and psychology, 345–369. Cambridge: Cambridge University Press. Knyazev, Mikhail. 2016. Complementizers in Kalmyk. In Complementizer Semantics in European Languages, 665-690. Mouton de Gruyter. Koopman, Hilda. & Dominique Sportiche. 1989. Pronouns, logical variables, and logophoricity in Abe. Linguistic Inquiry, 555-588. Koopman, Hilda. & Dominique Sportiche. 1991. The position of subjects. Lingua, 85 (2-3), 211-258. 281 Kornfilt, Jaklin. 1984. Case Marking, Agreement and Empty Catagories in Turkish. Harvard University (Doctoral dissertation). Kornfilt, Jaklin. 2003. Scrambling, subscrambling, and case in Turkish. Word Order and Scrambling, 125-155. Krifka, Manfred. 1999. Quantifying into question acts. In Tanya Matthews and Devon Strolovitc (Eds.) SALT IX, Department of Linguistics, Cornell University, CLC Publications. Krifka, Manfred. 2001. Quantifying into question acts. Natural Language Semantics, 9(1), 1-40. Korotkova, Natasha. 2021. The embedding puzzle: constraints on evidentials in complement clauses. Linguistic Inquiry, 52(1), 210-226. Koskinen, Päivi. 1998. Features and Categories: Non-finite Constructions in Finnish. University of Toronto (Doctoral Dissertation). Kuno, Susumu. 1973. The Structure of Japanese Language. The MIT Press. Kuno, Susumu. 1987. Functional syntax: Anaphora, Discourse and Empathy. Chicago: University of Chicago Press. Ladusaw, William. A. 1977. Some problems with tense in PTQ. Texas Linguistic Forum, 6, 89- 102. Laenzlinger, Christopher. 2005. French adjective ordering: perspectives on DP internal movement types, Lingua, 115, 645-689. Lakoff, George. 1972. Semantics of natural language. Chapter Linguistic and Natural Logic. 545-655. D. Reidel. Lasersohn, Peter. 2005. Context dependence, disagreement, and predicates of personal taste. Linguistics and Philosophy, 28(6), 643-686. Lasersohn, Peter. 2017. Subjectivity and Perspective in Truth-Theoretic Semantics. Oxford: Oxford University Press. Lewis, Geoffrey. L. 1967. Turkish Grammar. Oxford: Clarendon Press. Major, Travis. 2021. On the Nature of “say” Complementation. UCLA (Doctoral dissertation). Major, Travis. to appear. Revisiting the syntax of monsters in Uyghur. Linguistic Inquiry. Maki, Hideki., Kaiser, Lizanne. & Ochi, Masao. 1999. Embedded topicalization in English and Japanese. Lingua, 109, 1-14. 282 Maling, Joan. 1993. Of nominative and accusative. In Andres Holmberg and Urpo Nikanne (Eds.) Case and Other Functional Categories in Finnish Syntax, 49-74. Berlin: Mouton de Gruyter. McCloskey, James. 2005. Questions and questioning in a local English. In Raffaella Zanuttini, Héctor Campos, Elena Herburger and Paul H. Portner (Eds.) Cross-linguistic Research in Syntax and Semantics: Negation, Tense and Clausal Architecture. Georgetown University Press. Miyagawa, Shigeru. 2012. Agreements that occur mainly in the main clause. In Lobke Aelbrecht, Liliane Haegeman, Rachel Nye (Eds.) Main Clause Phenomena: New Horizons, 79-111. John Benjamins. Montague, Richard. 1974. The proper treatment of quantification in ordinary English, in Formal Philosophy: Selected Papers of Richard Montague, Yale University Press, New Haven. Moulton, Keir. 2009. Clausal Complementation and the Wager-Class. In Anisa Schardl, Martin Walkow, and Muhammad Abdurrahman (Eds.), Proceedings of the Thirty-Eighth Annual Meeting of the North East Linguistic Society, 2. (NELS 38), GLSA: Amherst, MA. 165–178. Moulton, Keir. 2013. Not moving clauses: Connectivity in clausal arguments. Syntax, 16(3), 250-291. Mufwene, Salikoko. S. 1990. Serialization and subordination in Gullah: toward a definition of serialization. Working Papers in Linguistics, 39, 91-108. Ohio State University. Department of Linguistics. Neeleman, Ad., Elena Titov., Hans van de Koot. & Reiko Vermeulen. 2008. A syntactic typology of topic, focus and contrast. Manuscript, University of College London. Newman, Paul. 2000. The Hausa language, an encyclopedic reference grammar. New Haven: Yale University Press. Nichols, Lynn. 2001. The syntax and morphosyntax of propositional attitude complementation. Harvard University (Doctoral Dissertation). Nishigauchi, Taisuke. 2014. Reflexive binding: Awareness and empathy from a syntactic point of view. Journal of East Asian Linguistics, 23(2), 157-206. Oğuz, Metehan. 2021. Proleptic analysis of accusative subjects in Turkish. Unpublished Manuscript. University of Southern California. Öztürk, Balkız. 1999. Turkish as a Non Pro-Drop Language. Boğaziçi University (Unpublished MA Thesis). Öztürk, Balkız. 2001. Turkish as a non pro-drop language. In Taylan, E. E. (Ed.), The Verb in Turkish, 239-259. Amsterdam: John Benjamins Publishing Company. 283 Özyıldız, Deniz. 2012. When I is not me: A preliminary case study of shifted indexicals in Turkish. Manuscript, UCLA. Özyıldız, Deniz. 2016. Knowledge reports without truth. Proceedings of the European Summer School in Logic, Language, and Information (ESSLLI), 184-196. Özyıldız, Deniz. 2017. Attitude reports with and without true belief. In Semantics and Linguistic Theory, 27, 397-417. Özyıldız, Deniz. 2018. Dissertation prospectus. Unpublished Manuscript. UMASS. Özyıldız, Deniz. 2019. Embedded clauses in Turkish: Different paths to composition. Talk at RelNompComp, 19, University of Toronto. Özyıldız, Deniz., Major, Travis., & Maier, Emar. 2019. Communicative reception reports as hear-say: Evidence from indexical shift in Turkish. In Proceedings of the 36th West Coast Conference on Formal Linguistics, 296-305. Somerville, MA: Cascadilla Proceedings Project. Özyıldız, D. 2021. The Event Structure of Attitudes. UMASS (Doctoral dissertation). Pancheva, Roumyana., & Zubizarreta, Maria. L. 2018. The person case constraint. Natural Language & Linguistic Theory, 36(4), 1291-1337. Park, Yangsook. 2016. Indexical shift and the long-distance reflexive caki in Korean. Manuscript, UMASS. Patel-Grosz, Pritty. 2020. Pronominal typology and the de se/de re distinction. Linguistics and Philosophy, 43(5), 537-587. Pearson, Hazel. 2013. A judge-free semantics for predicates of personal taste. Journal of Semantics, 30(1), 103-154. Percus, Orin. 2000. Constraints on some other variables in syntax. Natural Language Semantics, 8(3), 173-229. Percus, Orin. & Ulli Sauerland. 2003. Pronoun movement in dream reports. In Makoto Kadowaki and Shigeto Kawahara (Eds.) Proceedings of NELS, 265-284. GLSA, UMASS. Poole, Ethan. 2020. Improper case. Manuscript, University of California, Los Angeles. Predolac, Esra. 2017. The Syntax of Sentential Complementation in Turkish. Cornell University (Doctoral dissertation). Pulleyblank, Douglas. 1986. Clitics in Yoruba. In The Syntax of Pronominal Clitics -Syntax and Semantics, 19, 43-64. 284 Quine, Willard. V. 1956. Quantifiers and propositional attitudes. the Journal of Philosophy, 53(5), 177-187. Reinhart, Tanya. 1981. Pragmatics and linguistics: An analysis of sentence topics. Philosophica, 27, 53-94. Reschke, Kevin., & Pranav Anand. 2011. Extracting contextual evaluativity. In Proceedings of the Ninth International Conference on Computational Semantics (IWCS 2011). Rizzi, Luigi. 1990. Relativized Minimality. The MIT Press. Rizzi, Luigi. 2001. On the position “Int (errogative)” in the left periphery of the clause. In Guglielmo Cinque and G. Salvi (Eds.) Current Studies in Italian Syntax: Essays Offered to Lorenzo Renzi, 59, 287-296. Amsterdam: Elsevier. Rizzi, Luigi. 2004. Locality and left periphery. In Adriana Belletti (ed.) Structures and Beyond: The Cartography of Syntactic Structures, 3, 223-251. Oxford/New York: Oxford University Press. Rizzi, Luigi. 2013. Locality. Lingua, 130, 169-186. Ross, John. R. 1967. Constraints on Variables in Syntax. MIT (Doctoral Dissertation). Sağ, Yağmur. 2017. Complex predication via phrasal adjunction to a head category. In 34th West Coast Conference on Formal Linguistics, 431-438. Cascadilla Proceedings Project. Schlenker, Philippe. 1999. Propositional Attitudes and Indexicality: A Cross Categorial Approach. MIT (Doctoral dissertation). Schlenker, Philippe. 2003. A plea for monsters. Linguistics and Philosophy, 26(1), 29-120. Shklovsky, Kirill., & Yasutada Sudo. 2014. The syntax of monsters. Linguistic Inquiry, 45(3), 381-402. Şener, Serkan. 2008. Non-canonical case licensing is canonical: Accusative subjects of CPs in Turkish. Manuscript, University of Connecticut. Storrs. Şener, Serkan. 2010. (Non-)peripheral Matters in Turkish Syntax. University of Connecticut (Doctoral Dissertation). Şener, Gültekin N. & Serkan Şener. 2011. Null subjects and indexicality in Turkish and Uyghur. Proceedings of Workshop on Formal Altaic Linguistics [WAFL 7], University of Southern California, Los Angeles. Sharvit, Yael. 2008. The puzzle of free indirect discourse. Linguistics and Philosophy, 31(3), 353-395. 285 Simons, Mandy., Judith Tonhauser, David Beaver. & Craige Roberts. 2010. What projects and why. In Semantics and Linguistic Theory, 20, 309-327. Speas, Margaret, & Carol Tenny. 2003. Configurational properties of point of view roles. In A.- M. Di Sciullo (Ed.) Asymmetry in Grammar, 315-344. Amsterdan: John Benjamins. Stowell, Tim. 2007. The syntactic expression of tense. Lingua, 117(2), 437-463. Straughn, Christopher. 2006. Noun incorporation and case: Evidence from Sakha. In Annual Meeting of the Berkeley Linguistics Society, 32(1), 375-385. Straughn, Christopher. 2011. Evidentiality in Uzbek and Kazakh. Northeastern Illinois University (Doctoral Dissertation). Sunderasan, Sandhya. 2012. Context and (Co)reference in the Syntax and its Interfaces. University of Tromsø and University of Stuttgart (Doctoral Dissertation). Sundaresan, Sandhya. 2018. Perspective is syntactic: Evidence from anaphora. Glossa: A Journal of General linguistics, 3(1). Sundaresan, Sandhya. & Thomas McFadden. 2019. Reducing pro and pro to a single source. The Linguistic Review, 35(3), 463-518, USC Job Talk. Taylan, Eser. E. 1986. Pronominal versus zero representation of anaphora in Turkish. In Studies in Turkish linguistics, 209-233. John Benjamins. Underhill, Robert. 1976. Turkish Grammar (Vol. 460). Cambridge, MA: MIT press. Vainikka, Anne. M. 1989. Deriving Syntactic Representations in Finnish. UMASS (Doctoral Dissertation). Webelhuth, Gert. 1992. Principles and Parameters of Syntactic Saturation. Oxford University Press. Williams, Edwin. 2003. Representation Theory. MIT Press. Woods, Rebecca. 2016. Investigating the Syntax of Speech Acts: Embedding Illocutionary Force. University of York (Doctoral dissertation). Yıldırım-Gündoğdu, Hilal. 2017a. The Structure of Diye Clauses in Turkish. Bogazici University (Unpublished MA Thesis). Yıldırım-Gündoğdu, Hilal. 2017b. Against diye clauses as complements of verbs of communication. IULC Working Papers, 18(2). Zidani-Eroğlu, Leyla. 1997. Exceptionally case-marked NPs as matrix. Linguistic Inquiry, 28, 219-230. 286 Zu, Vera. 2018. Discourse Participants and the Structural Representation of the Context. NYU (Doctoral dissertation). Zubizarreta, M. L. 2001. The constraint on preverbal subjects in Romance interrogatives: A minimality effect. In: Aafke Hulk, Jean-Yves Pollock (Eds.), Subject Inversion in Romance and the Theory of Universal Grammar. Oxford Studies in Comparative Syntax, Oxford University Press, Oxford and New York.

Abstract (if available)

Abstract Being considered a discourse-pragmatic phenomenon earlier, the role perspective plays in the syntax was peripheral in syntactic research until recently (e.g. Sundaresan 2012, Charnavel 2019). The primary goal of this dissertation is to demonstrate that perspective plays a vital role in syntactic domain formation and various syntactic processes based on data from Turkish complementation. I argue that the two CP-level complement clauses in Turkish differ from each other in whether they encode perspective. This difference, in turn, leads to various syntactic differences such that availability of syntactic encoding of perspective forms an opaque domain, as evidenced by the ban on binding, NPI-licensing, among other patterns. However, the opaqueness is not absolute but is rather selective (Keine 2019), as evidenced by the ability of topic-related elements to escape perspectival domains, including what was earlier considered to be indexical shift cases in Turkish. The investigation of the interaction of embedded perspectival domains with various embedding cognitive attitude verbs reveal important patterns that lead to the conclusion that Turkish makes a three-way syntactic distinction based on perspective between these verbs: speaker-oriented verbs, matrix subject-oriented verbs and perspective-underspecified verbs. I correlate perspective with (non)factivity such that factive reading is a result of syntactic encoding of the speaker’s perspective; non-factivity is a result of encoding the matrix subject’s perspective, and lack of perspective specification leads to what is known as semi-factive verbs. I show that matrix subject-oriented verbs turn embedded opaque domains such as perspectival domains into transparent domains while speaker-oriented verbs turn otherwise-transparent domains into opaque domains. I attribute this to the specific properties of the extra structure they both introduce.

Linked assets