* Your assessment is very important for improving the workof artificial intelligence, which forms the content of this project
Download Finite control in Korean - Iowa Research Online
Macedonian grammar wikipedia , lookup
Chinese grammar wikipedia , lookup
Portuguese grammar wikipedia , lookup
Esperanto grammar wikipedia , lookup
Swedish grammar wikipedia , lookup
Antisymmetry wikipedia , lookup
Kannada grammar wikipedia , lookup
Ancient Greek grammar wikipedia , lookup
Yiddish grammar wikipedia , lookup
Old English grammar wikipedia , lookup
Latin syntax wikipedia , lookup
Georgian grammar wikipedia , lookup
Spanish grammar wikipedia , lookup
Icelandic grammar wikipedia , lookup
Scottish Gaelic grammar wikipedia , lookup
Serbo-Croatian grammar wikipedia , lookup
Polish grammar wikipedia , lookup
Lexical semantics wikipedia , lookup
English clause syntax wikipedia , lookup
University of Iowa Iowa Research Online Theses and Dissertations Fall 2009 Finite control in Korean Kum Young Lee University of Iowa Copyright 2009 Kum Young Lee This dissertation is available at Iowa Research Online: http://ir.uiowa.edu/etd/394 Recommended Citation Lee, Kum Young. "Finite control in Korean." PhD (Doctor of Philosophy) thesis, University of Iowa, 2009. http://ir.uiowa.edu/etd/394. Follow this and additional works at: http://ir.uiowa.edu/etd Part of the Linguistics Commons FINITE CONTROL IN KOREAN by Kum Young Lee An Abstract Of a thesis submitted in partial fulfillment of the requirements for the Doctor of Philosophy degree in Linguistics in the Graduate College of The University of Iowa December 2009 Thesis Supervisor: Professor William D. Davies 1 ABSTRACT This thesis explores finite control in Korean. An overview of the previous studies of control shows that the mainstream literature on control has consistently argued that referential dependence between an overt matrix argument and an embedded null subject is characteristic of non-finite clauses which contain a PRO subject. Moreover, although some evidence for finite control involving pro in several languages has been presented, a PRO analysis of finite control has been firmly established in the literature. This thesis, however, argues that the currently established approach to Obligatory Control (OC), which is confined to PRO, cannot account for OC in Korean, and provides an empirical and theoretical analysis of finite control containing a pro subject in Korean. Although finite OC in Korean differs from non-finite OC in other languages in that the former can allow an overt NP coreferential with a matrix argument in the null subject position, finite OC in Korean displays the same properties of OC which are widely employed as the criteria for defining OC in non-finite clauses. This thesis adopts the formal approach to finiteness in which finiteness is defined as an ability of licensing nominative subjects. However, reviewing the cross-linguistic data in the literature reveals that the feature determining finiteness should not be restricted to just Tense and Agreement, as the formal approaches have argued, and that languages may vary in determining finiteness. It also explores the relevance of Mood and Modality as the manifestation of finiteness in Korean. Based on this, this thesis argues for the CP status of finite OC in Korean and a pro analysis of the null subject in the constructions. 2 Through an investigation of six complementation types that have or have not been grouped under the types of control in the literature along with ninety matrix predicates which are classified into nine different categories based on their semantic class, this thesis further argues that OC in Korean cannot be explained by a solely syntax-based or semantically-based theory. OC in Korean is mainly the result of multiple semantic factors, but syntactic and pragmatic factors can also play a role in determining control. Abstract Approved: _________________________________________________________ Thesis Supervisor _________________________________________________________ Title and Department _________________________________________________________ Date FINITE CONTROL IN KOREAN by Kum Young Lee A thesis submitted in partial fulfillment of the requirements for the Doctor of Philosophy degree in Linguistics in the Graduate College of The University of Iowa December 2009 Thesis Supervisor: Professor William D. Davies Graduate College The University of Iowa Iowa City, Iowa CERTIFICATE OF APPROVAL __________________________ PH.D. THESIS ____________ This is to certify that the Ph.D. thesis of Kum Young Lee has been approved by the Examining Committee for the thesis requirement for the Doctor of Philosophy degree in Linguistics at the December 2009 graduation. Thesis Committee: ________________________________________ William D. Davies, Thesis Supervisor ________________________________________ Alice L. Davison ________________________________________ Stanley Dubinsky ________________________________________ Sarah Fagan _________________________________________ Roumyana Slabakova To my parents who gave me love and the Korean language. ii ACKNOWLEDGMENTS As I look back on my years at the University of Iowa, I realize that I have had many invaluable experiences both academically and personally. My genuine interest in linguistics has blossomed under the tutelage of the professors and courses in the graduate program of the linguistics department at the University of Iowa. First, from the bottom of my heart I would like to thank William D. Davies, my academic and thesis adviser. Throughout my MA and PhD courses, he has given clear and even elegant lectures on Syntax. He is truly responsible for my growth as a syntactician and linguist. Without his guidance and support I could not have completed this thesis. I will always remember his steadfast work, and genuine care and understanding for his students. I would also like to thank the other committee members of my thesis: Stanley Dubinsky, Alice L. Davison, Roumyana Slavacoba, and Sarah Fagan. They read my dissertation thoroughly, and gave much valuable input. I am also tremendously thankful for all the professors who have broadened my perspectives on linguistics: William D. Davies, Alice L. Davison, Roumyana Slavacoba, Catherine Ringen, Jill Beckman, Elena Gavruseva, and Jerzy Rubach. I owe a debt of gratitude to my professors at Chungnam National University in Korea where I have spent my college and graduate school years. In particular, Professor Han Young-Mok, Toh Soo-Hee, and Kim Cha-Kyun among many other professors have provided great encouragement and support during my study in Iowa. My gratitude further extends to my fellow students and friends both within and without the University of Iowa. I would particularly like to thank to Marta Tryzna, Ivan Ivanov, Tomomasa Sasa, Jay-Young Shim, John Njue and Ann DenUyl for their iii friendship and assistance with my arguments and ideas. I will always cherish their friendship. My special thanks go to Ann DenUyl who has helped me in various ways during my stay at Iowa. She was always there for me as a genuine friend and my thanks to her is certainly more than I am able to express here. My family and friends in Korea are certainly not without my thanks. They always believed in me, cared for me endlessly, and have waited for me all these years. I am truly blessed to have such devoted and supportive family and friends. Most of all, I thank God for walking with me all these years at Iowa. Without Him, none of this would have been possible and most certainly would not have been as meaningful. iv TABLE OF CONTENTS LIST OF ABBREVIATIONS …..…………………………………………………… viii CHAPTER 1. INTRODUCTION .…………………..………………………………………….. 1 1.1 General Goals of This Study.….………………….……………………….….. 1 1.2 Previous Studies on Control and the Main Issues of this Study ...…………… 4 1.2.1 Syntactic Theory of Control in Generative Grammar .……….………….. 4 1.2.1.1 Transformational Grammar and Government & Binding (GB) Frame Work ……..………………...……………………………….... 4 1.2.1.2 Minimalist views of OC …………………….…………………..….. 10 1.2.1.2.1 The Case-theoretic Account …………….……………….……... 10 1.2.1.2.2 The Movement Theory of Control ..….……………..………….. 12 1.2.1.2.3 The Agree-based Account ....……….….………………………. 13 1.2.2 Syntactic/Semantic Analysis of Control in Non-Derivational Grammar .. 16 1.2.2.1 The Lexicalist Accounts ....………………………………………….. 16 1.2.2.1.1 Lexical-Functional Grammar (LFG) ....….….…………………... 16 1.2.2.1.2 Head-driven Phrase Structure Grammar (HPSG) .….…………… 17 1.2.2.2 Semantic Accounts of Control ....……………………………………. 19 1.2.3 Finite Control …………………………………………………………… 21 1.3 Methodology and Data Sources …………………………………………….. 27 1.4 Structure of This Study ….…..……………………………………………… 28 2. FINITENESS AS A NOMINATIVE SUBJECT LICENSING AND MOOD AND MODALITY AS FINITENESS ...……………….………….……….….. 30 2.1 Finiteness as Nominative Subject Licensing .…………….………................ 31 2.1.1 Finiteness and Subject Licensing in Generative Grammar ...…….…….. 31 2.1.2 Language Variation in Nominative Subject Licensing ..…….…………. 39 2.1.2.1 Tense or Agr as a Nominative Case Assigner ……………………… 39 2.1.2.2 Mood, Modality, or Aspect as a Nominative Case Assigner ….……. 49 2.2 Mood and Modality as Finiteness in Korean .….……….…….…………….. 58 2.2.1 Defectiveness of Tense and Agr in Nominative Case Assignment …….. 59 2.2.2 Mood and Modality on Verbs and Complementizers and Their Projections ...…………………………………………………………… 69 3. FINITE CONTROL AND SENTENTIAL COMPLEMENTATION IN KOREAN .....…..……………………………………………………………….. 85 v 3.1 Classification and Properties of Obligatory Control …………..…………… 86 3.2 Finite Control as OC in Korean ..………………………………..………….. 91 3.2.1 Finiteness of Control Complement Clauses …………………………….. 92 3.2.2 OC Properties of Structures ……………………………………………. 108 3.3 Types of Sentential Complements and Finite Control …….…….………… 115 3.3.1 Nominalization and Control ….…………….……………….................. 118 3.3.2 Adnominalization and Control …..…………………………………….. 123 3.3.3 Interrogative Complements and Control .……………………………… 130 3.3.4 Quotative Complements and Control .…………………………………. 134 3.3.5 Intentive Complements and Control ......…………….….……………… 138 3.3.6 Resultative Complements and Control .…...…………………………… 141 4. SYNTACTIC PROPERTIES OF CONTROL COMPLEMENTS IN KOREAN ……….………………………………………………………..……………… 151 4.1 The Structure of Finite Control Complements and pro in Korean ….……… 152 4.2 Syntactic Properties of the Null Subject in Finite Control ....……….……… 162 4.2.1 Case on a Controllee ...…………………………………………………. 162 4.2.1.1 Quantifier agreement …………………………………..….……..… 163 4.2.1.2 Honorific suffix agreement ...….…………………..………….…… 165 4.2.1.3 Occurrence of an Emphatic Subject Pronoun in Control Constructions ..………………………………………………….…. 167 4.2.1.4 Availability of a Controllee Conjoined with a Lexical DP .……….. 168 4.2.2 Agree Features of Controlee …..………………………………………. 171 4.2.3 Pro Different From the LD reflexive Caki and a Pronoun .….………. . 175 4.2.3.1 Different Distributions of the LD reflexive Caki and Pro ..…..…… 176 4.2.3.2 Different References and Capability of LD Control ….…..…….…. 179 4.2.3.3 Sloppy vs. Strict reading and de re vs. de se interpretation. .…..….. 184 4.3 The Status of Serial Verb Constructions As Non-Finite Control ..……….… 189 4.3.1 Monoclausal Status of Serial Verb Constructions .………………….…. 190 4.3.1.1 Clausemate Condition on NPIs ..…………………………………… 193 4.3.1.2 Placement of Sentential Adverbs ....…….……………………….…. 198 4.3.1.3 Scope of Wh-Adverbs ..…..………………………………………… 203 4.3.2 VP Analysis of Serial Verb Constructions ….………………………….. 207 5. CONTROLLER CHOICE IN OBLIGATORY CONTROL IN KOREAN …… 226 5.1 Problems for Syntactic Analyses of Obligatory Control ……………………. 227 5.1.1 Locality and C-Command ...…………………………………………….. 227 5.1.2 Tense, and Agreement …………………………………………………... 232 vi 5.2 Roles of Syntax, Semantics, and Pragmatics in Controller Choice …………. 244 5.2.1 Roles of Matrix Predicates and Complementizers ……………………… 245 5.2.1.1 Predicates and Complementizers Inducing Inherent Control ...…….. 246 5.2.1.2 Control-neutral Predicates ......……………………………………… 253 5.2.1.3 Combination of a Predicate and a Complementizer ...…….…….….. 264 5.2.2 Control Effects of Mood and Modal Markers in -Ko complements ...….. 271 5.2.3 Control Effects of Syntactic and Semantic Properties of Embedded Clauses and Pragmatics ….…..………………………………………….. 276 5.2.4 Control Shift………….………………………………………………….. 282 6. CONCLUSION ….…………………………………………………………….... 291 REFERENCES ….….……………………………………………………………… 296 vii LIST OF ABBREVIATIONS ACC ADV ADN AH AOR ASP C/COMP CONJ COP DAT DET DC DR ERG EXH/PROP FS FUT HON HUM IMP IND INF GEN LOC M MP MS NEG NOM NOML OBL PASS PERF PL PROG PROM PRS PRT PST Q RET SH S/SG Accusative Case Adverb Adnominalizer Addressee Honorific Aorist Aspect Complementizer Conjecture Copular Dative Case Determiner Declarative Direction Ergative Case Exhortative/Proposative Feminine Singular Future Honorific Humble Imperative Indicative Infinitive Genitive Case Locative Case Masculine Masculine Plural Masculine Singular Negative Nominative Case Nominalizer Oblique Case Passive Perfect Plural Progressive Promissive Present Particle Past Question Retrospective Speaker Honorific Singular viii SUBJ TOP VOL Subjunctive Topic Volitional ix 1 CHAPTER 1 INTRODUCTION 1.1. General Goals of This Study Over the past 40 years, much attention has been paid to control constructions such as (1) in linguistic research. (1) a. Pati tried [eci/*j to study hard]. b. Pati persuaded Johnj [ec*i/j/*k to study hard]. Although there is no overt subject in the embedded clauses in (1), intuitively, we know that the embedded subjects in (1a) and (1b) refer to the matrix dependent Pat and John, respectively. That is, the matrix subject Pat in (1a) is also semantically the subject of the embedded verb, while the matrix object John in (1b) is also semantically the subject of the embedded verb. Within the generative paradigm, it is assumed that the embedded clauses in (1) include an empty category ‘ec’, which has come to be referred to as “PRO”, and the matrix arguments are said to “control” the reference of the PRO. That is, “control” indicates the relation of co-reference between the null subject of to infinitival complements and one of the matrix arguments. Thus, the former is called subject control and the latter object control. The co-indexation of the null subject with other than one of the matrix arguments renders the sentence ungrammatical. As will be discussed in detail in the next section, these constructions have long been examined, touching on many different topics of the grammar, such as Binding theory, Case theory, and Finiteness. Recently, in particular, control constructions have 2 been at the center of linguistic debates with respect to the nature of the empty category in the constructions. That is, there have been attempts to integrate control and raising constructions (Hornstein 1999, 2003; Boeckx & Hornstein 2007) within the Minimalist Program, replacing PRO with an NP trace of A-movement, as represented in (2). (2) [Johni [VP ti tried [TP ti [to [VP ti study hard]]]]]. Regardless of whether the null subject is PRO or trace, within the generative grammar, the lack of an overt embedded subject in control constructions is generally attributed to the lack of projections or features associated with finiteness that license the structural nominative Case to the subject of a clause. That is, standard approaches to control have focused on the correlation between non-finiteness and the nature of the null subject. However, cross-linguistic research such as Landau (2004) has revealed that control can also exist in a finite context, challenging the correlation between the null subjects of non-finite constructions and the matrix arguments. (3) Gil1 hivtiax [še-ec1 yitna’heg yafe]… Hebrew Gil promised that-ec will-behave.3SG.M well ‘Gil promised to behave…’ (Landau 2004) Compared to the to infinitives in (1), the embedded clause in (3) is considered to be finite in that it is inflected for Tense and Agreement, and thus can license nominative Case on the null subject. 3 However, the empty category in (3), like those in (1), obligatorily refers to a matrix argument (here, the matrix subject Gil), and no overt lexical NP coreferential with the matrix subject can occur in the position. Thus, Landau argues that the empty category must be PRO. Korean also shows the existence of finite control, as follows. (4) a. Johni-i Billj-eykey [ec i/*j/*k ttena-keyss-ta]-ko yaksokha-yess-ta. J-NOM B-DAT leave-VOL-DC-C promise-PST-DC ‘Johni promised Bill that hei would leave.’ b. Johni-i Billj-eykey [kui/*j/*k/cakii/*j/*k-ka ttena-keyss-ta]-ko yaksokha-yess-ta. J-NOM B-DAT he /self-NOM leave-VOL-DC-C promise-PST-DC ‘Johni promised Bill that hei/selfi would leave.’ The empty category in (4a) is obligatorily controlled by the matrix subject John and no antecedent outside of the matrix clause is permitted. However, finite control in Korean is very different from nonfinite control in (1) and finite control in (3) in that an overt NP coreferential with one of the matrix arguments can occur in the null subject position. In (4b), ku 'he' and caki 'self' stand in the same relationship to the matrix clause as does the null subject in (4a); all are obligatorily coreferential with John, the matrix subject. This suggests that finite control in Korean is not solely a property of a specific syntactic category PRO. This thesis provides an empirical account of finite control constructions in Korean through an investigation of sentential complements and the correlation between 4 finiteness and control in Korean. In particular, this thesis mainly argues that the currently established approach to Obligatory Control (OC), which is confined to PRO, is insufficient to account for OC in Korean and that controlled complements in Korean are finite clauses with null pronominal subjects. In addition, this thesis argues that OC in Korean cannot be accounted for solely in a syntax-based or a semantically-based theory, providing evidence that OC in Korean is derived from very complex factors. To be more precise, it is determined primarily by multiple semantic factors and syntactic or pragmatic factors to a lesser extent. Compared to significant amounts of research on control in English, there are comparatively few studies on control in Korean. Considering the fact that crosslinguistic research has revealed that languages vary in many ways and that not all languages make use of the same syntactic structure to display the same linguistic phenomenon, an attempt to explore control constructions in languages other than English will provide not only a better account for the facts of control, but also better insight to the nature of language. In particular, considering the very existence of finite control in Korean, the investigation of syntactic and semantic properties of control constructions in Korean will extend the notion of control to ‘finite control’ and shed light on some current issues on control. 1.2. Previous Studies on Control and the Main Issues of this Study In this section, I will review the previous studies of control, pointing out the major issues throughout the history of the study of control, which are directly related to the main issues of this study, and then briefly address the main issues in Korean OC constructions that will be examined in this thesis. 5 1.2.1. Syntactic Theory of Control in Generative Grammar 1.2.1.1. Transformational Grammar and Government & Binding (GB) framework The first major generative analysis of control was Rosenbaum (1967), in which sentences with control were derived through the application of a transformation referred to as Equivalent Noun Phrase Deletion (Equi-NP Deletion). Equi NP Deletion is a transformational rule that deletes the subject of subordinate clauses under identity with the subject or object of the next higher clauses. Rosenbaum (1967) attempts to predict controller choice in terms of his Minimal Distance Principle (MDP), which states that the NP closest to the subject of the complement is its controller under the transformation of Equi-NP Deletion. The MDP correctly predicts the controllers in (5). (5) a. Pat tried to study hard. b. I persuaded John to come. Compared to (5a), in which the matrix subject Pat is the controller, the controller in (5b) is the matrix object John, not the matrix subject I since the matrix object is the NP closest to the subject of the complement. This object control is derived by the application of Equi NP Deletion, as shown in (6). (6) I persuaded [NP John [S [NP John] [VP come]] ↓ I persuaded [NP John [S for [NP John] [VP to come]] ↓ Base structure Complementizer Placement 6 I persuaded [NP John [S for [VP to come]] Equi NP Deletion 1 ↓ I persuaded [NP John [S [VP to come]] Complementizer Deletion However, the MDP makes the wrong predication about the controller in the sentence with the matrix verb promise. 2 (7) John promised Pat to leave. In (7) the controller is John not Pat, although Pat is the NP closest to the subject of the complement. In the framework of Government & Binding (Chomsky 1981, 1986), Equi-NP deletion is abandoned and the analysis of control phenomena is based on the empty category PRO, which has no overt counterpart. PRO is different from the other empty categories, namely pro, trace, and variable, in terms of the features that define it. (8) PRO: pro: [+anaphoric, +pronominal] [–anaphoric, +pronominal] NP-trace: [+anaphoric, –pronominal] variable: [–anaphoric, –pronominal] 1 In Rosenbaum (1967), it was referred to ‘Identity Erasure Transformation’. 2 Rosenbaum (1967) treats subject control in promise as a lexical exception. 7 These empty categories are subject to the Binding Principles stated in (7). (9) Binding Principle A: An anaphor must be bound in its governing category. Binding Principle B: A pronominal must be free in its governing category. Binding Principle C: An R-expression must be free. However, PRO’s features give it an ambiguous property. As a pronominal it must be free in its governing category, while as an anaphor it must be bound within its governing category. Therefore, its reference cannot be determined on the basis of the Binding Principles, as it must simultaneously obey two contradictory Principles A and B. Chomsky (1981) proposes to resolve the contradiction by means of what came to be known as the PRO Theorem: PRO must be ungoverned. The distribution of PRO in ungoverned positions prevents PRO from bearing Case and from occurring in finite clauses. Therefore, PRO is excluded from subject and object position in finite clauses. That is, PRO is in complementary distribution with overt DPs, which must be Casemarked, as shown in (10). (10) a. Bill thinks [Jane/*PRO is alone]. b. Bill thinks [John should meet Jane/*PRO]. PRO accounts for controlled subjects in a similar way as Equi-NP deletion does: there are two different nominals at deep structure, each of which receives one different theta role. Being exempt from the Binding Theory, the interpretation of PRO is handled by 8 Control Theory, which is a distinct module of the Grammar that is responsible for the choice of controller in OC. In contrast to Chomsky (1981), Bouchard (1984) assumes that PRO is either a pronominal or anaphor. Locally controlled PROs are bound anaphors, and long-distance controlled PROs and arbitrary PROs are pronominals. Thus, the theory of control does not exist. According to Bouchard (1984:34), the notion of government 3 plays a crucial role in binding and must be incorporated in the definition of binding as in (11). (11) Binding: α binds β if and only if α governs β and α assigns its R-index to β. The formulation of binding in (11) has immediate consequences for the relation between an antecedent and an anaphor. That is, it forces the relation to have the four basic properties of anaphors; the antecedent-anaphor relation is obligatory, unique, local, and has structural conditions, such as c-command. Bouchard assumes that anaphors are functionally defined as elements bound by an antecedent in the sense of (11), and pronominals are elements freely indexed at surface structure. For example, (12) a. John tried [S PRO to leave]. b. John knows [S' how [S PRO to behave himself/oneself]]. 3 Government In the structure [γ...β...α...β...], α governs β if and only if (i) α is an immediate constituent of γ (ii) where ψ is a maximal projection, if ψ dominates β, then ψ dominates α. (where maximal projections are NP, PP, AP,S'(-Vmax)) 9 In (12a), S' deletion takes place since the position of COMP is empty. Thus, PRO is governed by the subject of the higher clause and gets its R-index from the subject. Therefore, PRO in (12a) is functionally determined to be an anaphor. However, if the COMP is filled as in (12b), S' deletion is blocked and consequently, S' blocks government by a Binder. Since PRO gets no R-index by a Binder, PRO is freely indexed at S-structure. Therefore, the PRO has a pronominal interpretation. In contrast to the analyses within GB framework above, Williams (1980) relates control to the notion of predication, i.e. the relation between subject and predicate. In this analysis, the structure of OC complements is [PRO VP] which is a case of S in position of predication, as shown in (13). (13) Johni promised Bill [S PRO [VP to leave]]i The embedded clause in (13) is defined to be a complex predicate with PRO as the “predicate variable” and the matrix subject NP is indexed with it, not with a PRO. The coindexing of predicates and their antecedents derives Predicate Structure (PS) from surface structure and PS is subject to the c-command restriction on indexing. 4 Williams (1980) argues that the c-command requirement distinguishes Obligatory Control from Non-Obligatory Control (NOC), and provides the properties of OC, as follows: (14) a. Lexical NP cannot appear in the position of PRO. b. The antecedent precedes the controlled PRO. 4 The C-Command Condition on Predication If NP and X are coindexed, NP must c-command X or a variable bound to X. (Williams 1980:205) 10 c. The antecedent c-commands the controlled PRO. d. The antecedent is thematically or grammatically uniquely determined. e. There must be an antecedent. [PRO VP] in NOC is not in position of predication such as (15a) and thus, S is not coindexed with an NP in PS and it is marked arb(itrary), as shown in (15b). (15) a. [PRO to leave] is nice (for Bill). b. [PRO to leave]arb is nice (for Bill). Therefore, for Williams (1980) the notion of c-command plays a central role in explaining why to infinitives exhibit OC in certain contexts and NOC in others. This notion of c-command as well as MDP, in fact, becomes fundamental to syntactic theories of control. 1.2.1.2. Minimalist views of OC 1.2.1.2.1. The Case-theoretic Account To account for the problematic distribution of PRO within the GB framework, Chomsky & Lasnik (1993) and Martin (1996) propose a Case-theoretic account of PRO in which PRO is marked for a special type of Null Case, which is restricted to PRO and is licensed by nonfinite INFL. However, Martin (2001) points out that Chomsky & Lasnik (1993) have no account of Exceptional Case Marking (ECM) and raising cases of infinitives, as shown in (16). 11 (16) a. *John believes [PRO to have solved the problem]. b. John believes [Mary to have solved the problem]. c. *John seems [PRO to leave]. d. John seems [t to leave]. If nonfinite INFL always assigns Null Case, the PRO should occur under an ECM and a raising verb. However, as can be seen in (16a), PRO is not compatible with an ECM verb and (16b) shows that an ECM verb has something other than Null Case that needs to be checked against an NP. Also (16c) and (16d) show that Null Case will not work for a raising verb. To solve this problem, Martin (2001) adopts Stowell’s (1982) proposal in which control infinitives, in contrast to ECM and raising verb infinitives, are specified for Tense, and proposes that control verbs select infinitival complements headed by [+tense, –finite] INFL, while raising verbs select infinitival complements headed by [–tense, –finite] INFL. Therefore, in Martin’s proposal, only a control infinitival can check Null Case and a raising infinitival does not check Case. However, Hornstein (1999) criticizes the Null Case theory as being stipulative – Since PRO is the only lexical item able to check or bear the Null Case and only nonfinite I can check or assign the Null Case, Null Case itself winds up being definitional and not externally motivated. Moreover, there are empirical problems in the Null Case theory in that there appear to be many instances of Case-marked PRO, as addressed in Sigurðsson (1991), Landau (2003), and Cecchetto & Oniga (2004) among others. 12 1.2.1.2.2. The Movement Theory of Control The Movement theory of control (O’Neil 1995; Hornstein 1999, 2003; Boeckx & Hornstein 2007) eliminates PRO by reducing OC to A-movement effects and PRO to an NP-trace. In order to do this, Hornstein (1999, 2003) also dispenses with the Theta Criterion stipulation that each argument bears only a single theta role, and analyzes theta-roles as features that are provided by verbs and checked by DPs. In his analysis, the embedded subject must move to the matrix subject position to be Case-licensed; thus, control structures are basically the same as the standard raising structures. The difference between control and raising is whether the landing site of NP-movement is to a theta position or non-theta position. That is, in raising, the embedded subject checks the theta feature of the embedded verb by initial Merge and then moves to the matrix subject position, which is a non-theta-position, in order to check its Case. In control structures, however, the embedded subject moves into a theta position to check its Case, checking two theta features through the movement. These are illustrated in (17) and (18). (17) a. John seemed to leave soon. θ b. [TP John [T [VP (John) seem [TP (John) [to [VP (John) leave soon]]]]]]. NOM (18) a. John tried to leave. θ θ b. [TP John [T [VP (John) try [TP (John) [to [VP (John) leave]]]]]]. NOM 13 In short, the Movement analysis of control unifies control and raising as a single syntactic phenomenon. However, it has been extensively criticized in the recent literature (Culicover & Jackendoff 2001, Jackendoff & Culicover 2003, Landau 2003, Kiss 2005), arguing for a control module separate from raising. 1.2.1.2.3. The Agree-based Account In contrast to the Movement theory of control, Landau (1999, 2004) claims that PRO does cross-linguistically exist in OC constructions and that raising and control are substantially different. However, unlike the Null Case theory, Landau’s analysis does not rely on Case to account for control. Instead, OC is analyzed as an instantiation of the operation Agree (Chomsky 2000, 2001) holding between a matrix Probe and an embedded anaphoric element. In Landau’s analysis, control verbs select a CP complement with a distinct Co, which may carry both [Agr] and [T] features, and the Co in turn selects an IP with distinct [T] features. The [T] feature on the head of IP can be selected or free. Complements with free Tense will carry no [T] feature on their Co, whereas selected Tense can be further divided into two types of Tense: i) anaphoric Tense, which must be identical to the matrix Tense, and ii) dependent Tense, which is dependent on the matrix Tense but does not need to be identical to it. Landau (2004:839) proposes that the selection of the right type of tense on the embedded Io must be mediated by matching of feature values of Co interfacing with the matrix verb, as shown in (19). (19) V.......[CP C[±T] [IP I[±T] VP]] selection checking 14 Landau (2004:839) assumes that the criterion for [+T] is semantic, not morphological: A clause may be tensed without carrying any Tense morphology (e.g., Partial Control infinitives) or untensed despite its Tense morphology (e.g., controlled subjunctives in the Balkan languages). By contrast, the feature [Agr], which is a bundle of φ-features, is purely morphological. Thus, [+Agr] is present iff there is Agreement morphology. This means that the Io of infinitives is [–Agr], whereas the Io of subjunctives and indicatives is [+Agr]. Co, on the other hand, can be specified as [+Agr] when it heads a clause with dependent Tense. Following Reinhart and Reuland (1993) and Reuland and Reinhart (1995), Landau (2004:841) also assumes that DPs capable of independent reference, such as a lexical DP and pro, are endowed with the feature [+R], whereas anaphoric DPs, such as PRO, are [–R]. That is, Landau suggests that PRO in OC is a null SEanaphor of sorts, lacking any inherent specification of φ-feature. In this analysis, [+R] is assigned to Io and Co only when they are specified as [+Agr, +T], and any other combination results in [–R]. Note that no [R] is assigned when there is no [T] or [Agr] feature. In this analysis, derivations of Exhaustive Control (EC) infinitives and Partial Control (PC) infinitives are illustrated in (20) below. (20) a. [CP DP....F.. [CP Co[–T] [IP PRO[–R] [I’Io[–T, –Agr, –R] [VP tPRO..]]]]] Agree Agree[+Agr] Agree[–T] Agree[–Agr, –R] b. [CP DP....F.. [CP Co[+T, +Agr, +R] [IP PRO[–R] [I’ Io[+T, –Agr, –R] [VP tPRO..]]]]] Agree[+Agr, +R] Agree[+Agr, +R] Agree[+T, ±Agr] Agree[–Agr, –R] 15 Landau argues that the complements in EC constructions, such as (20a), are untensed and lack agreement. In (20a), the specification [–T] feature on the Io guarantees that the [R] feature of Io is [–R], making the Agree relation between tPRO and Io possible. [–T] on the Io then checks the Agree relation with Co. In addition, PRO specified as [–R] establishes an Agree relation with the matrix F(unctional head) which in turn is in an Agree relation with the matrix subject. On the other hand, since PC infinitives such as (20b) bear dependent tense, both Io and Co bear [+T] and Co also bears [+Agr]. In the derivation in (20b) PRO and the matrix F check off two opposing [R] values. To eliminate its [+R] feature, Co enters a second Agree relation with the matrix head F, which inherits [+R] from the DP argument. According to Landau (2004:848), the fact that Co bears [+Agr] does not stop this feature from entering Agree with [–Agr] of Io since [Agr] on both heads is semantically uninterpretable and phonologically null. Therefore, the resulting chain of Agr-coindexation ensures that DP controls PRO. In summary, since Rosenbaum (1967), transformational grammar has adopted a sentential (or propositional) account of control. In GB (cf. Manzini 1983), the control complement is CP with PRO, compared to a raising complement which contains an NP trace in subject position. However, the Movement theory of control takes control complements to be an IP and reduces PRO to a trace of a moved argument. This has given rise to recent debates on the nature of the null subject in control complements. Syntactic approaches to control generally focus on deriving the properties of OC vs. NOC control constructions and do not explain the roles of predicates in determining control, except as it concerns their argument structure, wherein the MDP is invoked. 16 1.2.2. Syntactic/Semantic Analysis of Control in Non-Derivational Grammar 1.2.2.1. The Lexicalist Accounts 1.2.2.1.1. Lexical-Functional Grammar (LFG) LFG (Bresnan 1982, 2001; Kaplan & Bresnan 1982; Mohanan 1983; Dalrymple 2001) has long made a distinction between functional control and anaphoric control. In functional control, the controlled element is the SUBJ(ect) function of either XCOMP(lement) or XADJ(unct). Since XCOMP and XADJ lack their overt SUBJ in C(onstituent)-structure, the SUBJ of XCOMP and XADJ must be controlled by some Grammatical Function (GF) which is referentially identical in order to satisfy the completeness condition, which states that an F-structure must contain values for the GF that is subcategorized by the predicate. The control relation is captured by the sharing of the same F-structure by the controller and the controlled element. Functional control constructions include not only controlled PRO constructions, but also ECM and NP-movement constructions. Thus, the differences between (subject) raising and control are assumed to be semantic in nature. That is, the structural difference between rasing and control is the thematic structure of the matrix verb. In this respect, LFG’s functional control is much like Hornstein’s Movement theory of control. The structural difference between rasing and control is represented in the Fstructure equations in their lexical entries and in their semantics, such as (21). (21) a. seem V (↑ PRED) = ‘seem <(↑ XCOMP)> (↑ SUBJ)’ (↑ SUBJ) = (↑ XCOMP SUBJ) 17 b. try V (↑ PRED) = ‘try <(↑ SUBJ), (↑ SUBJ)>’ (↑ SUBJ) = (↑ XCOMP SUBJ) The angle brackets of the PRED contain the thematic arguments. Unlike the raising verb seem which does not select for a thematic subject and thus has the thematic subject outside the brackets, the control verb try selects for a thematic subject and thus has the thematic subject inside the brackets. Aanaphoric control is different from functional control in that it does not involve the sharing of F-structure and requires only identity of reference between controller and controlee. In contrast to functional control, anaphoric control allows for cases where there is no overt controller within the same sentence. That is, the controller of anaphoric control need not be in a local argument position which corresponds to NOC in syntactic accounts of control. Therefore, In LFG, anaphoric control is subject to semantic (thematic) constraints, while functional control is stated in terms of a hierarchy of grammatical functions (Sag & Pollard 1991:105). Farkas (1988) and Sag & Pollard (1991) argue that Bresnon’s (1982) control theory fails to predict the fundamental correlation between type of semantic relation and controller choice. 1.2.2.1.2. Head-driven Phrase Structure Grammar (HPSG) HPSG (Pollard & Sag 1987, 1991, 1994) proposes an integrated theory of the syntax and semantics of complement control. Similar to other lexicalist frameworks, such as LFG, control in HPSG (Pollard & Sag 1987, 1991, 1994) involves structure sharing only of indices, which are parts of the syntactic elements involved in the control relation (Asudeh 2005:466). 18 (22) a. Bill persuaded John to attend the meeting. category b. persuade: ...∣HEAD∣VFORM inf ARG-ST <NP, NP[1], VP ...∣SUBJ <NP[1]> > As can be seen in (22b), in control constructions, there is structure sharing not of whole arguments, but only of their INDEX values. That is, the VP[inf] has its own argument SUBJ<NP[1]> which share the INDEX value ‘[1]’ with the argument of the verb persuade. In this grammar, like LFG, OC constructions contain an infinitival VP in which there is no structural position for a syntactic subject. Thus, the complements are assumed to be syntactically non-sentential but semantically propositional. However, unlike LFG, HPSG argues that the principles which identify the controller argument must make reference not to grammatical functions, but to the thematic roles which are assigned to the arguments of predicates in the semantic feature structure (Grover 1995:49). Pollard & Sag (1994) proposes the classification of types of relations in which all control predicates belong to one of these semantic classes, as shown in (23). (23) relation influence persuade appeal cause commitment promise ... intend try ... orientation ... want hate expect ... 19 In summary, LFG and HPSG have treated Obligatory Control in terms of structure sharing, challenging the PRO analysis of Obligatory Control. In their proposals, the syntactic structure of OC infinitives is a VP, lacking a Tense projection and resulting in subjectless (i.e., PRO-less) infinitives. Also, it is worth pointing out that with the exception of (23), the LFG/HPSG approaches are notational variants of Equi NP Deletion. 1.2.2.2. Semantic Accounts of Control Since Jackendoff (1972) first suggested that controller assignment is determined by semantic (or thematic) roles rather than by purely syntactic factors (e.g. structural configurations or grammatical relations), as has been frequently assumed (Sag & Pollard 1991:64), there have been several different types of semantic accounts of control in the literature: the predicational approach (Chierchia 1983, 1984; Dowty 1985; and Culicover & Wilkins 1986), the thematic relation-based approach (Jackendoff 1972; Nishigauchi 1984; and Culicover & Jackendoff 2006) and others. Chierchia (1983, 1984), Dowty (1985), and Culicover & Wilkins (1986) base their analysis of control relations on a theory of predication. They assume that controlled complements are VPs and that these have the semantic type property rather than proposition. In fact, the predicational approach to control has been proposed by Williams(1980) within the GB framework. However, unlike Chierchia (1983, 1984), Williams argues that the controlled complements are syntactically clauses, while they are semantically property-like. Chierchia (1983, 1984) denies the relevance of syntactic control and proposes that the control effect arises as a result of the entailment relation that can be captured in terms of meaning postulates, such as the following: 20 (24) try’ (P) (x) →□j P (x) cf. □j = a context dependent modal operator (Chierchia 1984:34) (24) states that whenever x tries to bring about P, then in all the contextually relevant situations (namely those where what x tries actually succeeds) x does P. On the other hand, Jackendoff (1972), Nishigauchi (1984), and Culicover & Jackendoff (2006) predict the controller on the basis of thematic relations. In particular, Culicover & Jackendoff (2006) argue that regardless of any locality, control can be accounted for in terms of thematic roles. In their analysis, control is a semantic relation between two arguments at Conceptual Structure (CS) that can be realized syntactically through a non-finite construction (which lacks a subject), as in English, or through a finite clause with an overt or an empty subject, as in languages such as Greek and Hebrew (Culicover & Jackendoff 2006:135). For instance, the control phenomena in the following examples can be accounted for in terms of the CS-binding relation meeting thematic conditions. (25) a. Robini intended to ileave. b. Robini’s intention to ileave. The predicate INTEND which is a two-place relation; one of whose arguments is an animate entity, the intender, and the other is an action, selects actional complements. The Actor of the action argument of INTEND is necessarily bound to the intender. Thus, 21 as a consequence of the inherent binding within the predicate INTEND, any verb that contains this predicate as part of its meaning will have a control equation in which the intender uniquely controls the actional complement (Culicover &Jackendoff 2006:140). In summary, semantic accounts of control have argued that controller assignment is determined by semantic roles rather than by purely syntactic factors, basically capturing semantic properties of control predicates and their relation with the matrix arguments. However, whether the structure of OC is bigger than VP or not is still a matter of controversy. 1.2.3. Finite Control Although the mainstream literature in control studies has consistently argued that referential dependence between an overt matrix argument and an embedded null subject is characteristic of infinitival or non-finite clauses, some evidence for finite control in several languages has been presented, including Korean (Yang 1982, 1985; Borer 1989; Gamerschlag 2007; Lee 2007; Madigan 2008), Spanish (Suñer 1984), Serbo-Croation (Zec 1987), Persian (Hashemipour 1989; Ghomeshi 2001), Greek (Terzi 1992, 1997), Japanese (Uchibori 2000), Balkan languages (Landau 2004), and Brazilian Portuguese & Finnish (Modesto 2007). That is, it has been proposed that OC, where obligatory coreference is established between an overt matrix argument and an embedded null subject, can also obtain in finite clauses. As for as Korean, Yang (1982, 1985) first noted that there are control phenomena in finite contexts. 22 (26) a. Johni-i Billj-eykey [ec*i/j/*k ttena]-tolok seltukha-yess-ta. J-NOM B-DAT leave-C persuade-PST-DC ‘John persuaded Bill to leave.’ b. Johni-i Billj-eykey [ec i/*j/*j ttena-keyss-ta]-ko malha-yess-ta. J-NOM B-DAT leave-VOL-DC-C say-PST-DC ‘John told Bill that he would leave.’ c. Johni-i Billj-eykey [kui/*j/*k/cakii/*j/*k-ka ttena-keyss-ta]-ko malha-yess-ta. J-NOM B-DAT he /self-NOM leave-VOL-DC-C say-PST-DC ‘John told Bill that he/SELF would go.’ Compared to (26a) which is often called nonfinite control in the literature, the examples in (26b) and (26c) are considered to be finite control. Yang (1985) argues that even overt lexical items can be locally controlled in Korean, as shown in (26c), where ku ‘he’ or caki ‘self’ is controlled by the matrix subject John, and that the locally controlled empty category should be treated as pro rather than PRO. Yang claims that the local control in Korean is entirely lexically determined. That is, according to him, the control effect in (26a) is triggered by the semantics of the control predicate seltukhata ‘persuade’ and control effects in (26b, c) by the volitional modality of the embedded clause keyss. On the other hand, Borer (1989) argues that in the case of OC in Korean, the INFL node in an infinitive clause which is [–tense, +Agr] is an anaphoric Agr. In other words, control effects are manifested when the embedded Agr is anaphoric. According to her, all empty categories are subject to the I-identification requirement which states that all 23 empty categories must be coindexed with an I-identifier, where I-identifier means a member of a well-defined set of coindexed antecedents with a set of sufficiently rich inflectional features (Borer 1989:71). Thus, the anaphoric Agr which cannot I-identify pro must inherit features from its binding antecedent, resulting in Obligatory Control. Finite control has been analyzed in a number of other languages, particularly Balkan languages. Zec (1987) propose that controlled complements in Serbo-Croation are finite clauses with null pronominal subjects. (27) Petar je P pokušao da Aux tried dodje. Comp come (Pres) ‘Petar tried to come.’ The controlled complement to the matrix verb pokušao has an overt complementizer, and so is clearly a CP. Zec (1987) argues that there is demonstrably a null pronominal subject in the embedded clause since it is fully inflected for tense and agreement information and that the control relation is reducible to a highly constrained anaphoric relation, specifically that of a bound anaphor and its antecedent. However, the distribution of pro in finite control in Balkan languages has been hotly debated. For example, Terzi (1997), Landau (2004, 2005), and Kapetangianni & Seely (2007) among others argue that finite control in Balkan languages should be analyzed as involving PRO, not pro. Terzi (1997) proposes a PRO analysis for finite control in Greek and provides a licensing mechanism for PRO in finite contexts. She argues that the defective Tense of subjunctives checks Null Case and licenses PRO in 24 the Balkan languages. However, for Landau (2004, 2005), ‘defective Tense’ does not play a crucial role in control. (28) a. Ion1 vrea Ion [ca Dan/pro1/2 să rezolve problema]. wants that Dan/pro PRT solve the-problem ‘Ion1 wants Dan/him2/PRO1 to solve the problem.’ b. Hem1 kivu še-atem2/pro2 Romanian telxu (Landau 2005) ha-bayta mukdam. they hoped that-you (pl)/pro will-go.2PL home Hebrew early ‘They hoped that you (pl) would go home early.’ c. Gil1 hivtiax [še- ec1 yitna’heg yafe]… Gil promised that-ec will-behave.3SG.M well ‘Gil promised to behave…’ Hebrew (Landau 2004) According to Landau, the subjunctive clauses in (28) carry dependent Tense [+T] (e.g., irrealis), but only (28c) licenses PRO, unlike the examples in (28a, b), which license a lexical DP or pro. As can be seen in the coindexation in the English translation (28a, b), when the null subject is pro, it is not controlled. That is, the pro in (28a, b) can refer to the matrix subject or some other person. In contrast, when the null subject is PRO, as in (28c), it is controlled by the matrix subject Gil. According to Landau, Hebrew allows pro only for 1st and 2nd person and PRO only for a 3rd person null subject. To account for this distributional difference, Landau (2004:846) proposes the following mechanisms. 25 (29) a. [CP DP....F.. [CP Co[+T, +Agr, +R] [IP [I’ Io[+T, +Agr,+R] [VP DP/pro [ –R],1st/2nd..]]]]] Agree Agree[+T,+Agr,+R] Agree[+Agr, +R] b. [CP DP....F.. [CP Co[+T, +Agr, +R] [IP PRO[–R] [I’ Io[+T,+Agr,+R] [VP tPRO..]]]]] Agree[+Agr, –R ] Agree[+Agr] Agree[+T,+Agr,+R] Agree[+Agr] (29a) is the derivation of No control corresponding to (28b) and (29b) is the derivation of OC corresponding to (28c). As can be seen in (28), PRO contrasts with pro in terms of the specification of [R]. That is, the anaphoric nature of PRO is captured by the specification of [-R], while lexical DP or pro are specified for [+R]. This featural contrast is encoded in the position of Co and thus yields/cancels the control possibilities. Similarly, Kapetangianni & Seely (2007) proposes that temporal independence does not always correlate with Non-Obligatory Control, as shown in (30). (30) hthes o Yanis entharine ti Maria na yesterday the-NOM John-NOM encourage-3SG/PST the-ACC Mary-ACC NA erthi avrio sta genethlia tu come-3SG/PRS tomorrow to the-ACC birthday-ACC his ‘Yesterday, John encouraged Mary to come to his birthday party tomorrow.’ Kapetangianni & Seely (2007:139) propose that the na clause in Modern Greek in (20) is temporally matrix-independent with [+Agr], but it is associated with necessary OC with a PRO subject. Kapetangianni & Seely (2007) suggest that certain predicates select phi-defective I which cannot check nominative Case, while others select phi-complete I 26 which can check nominative Case and that consequently, the former causes the OC properties, while the latter causes NOC properties. However, there have been proposals that controlled subjects in Modern Greek are not always PRO (cf. Philippaki-Warburton & Catsimali 1999; Spyropoulos & Philippaki-Warburton 2001; Philippaki-Warburton 2004; and Spyropoulos 2007). These proposals in general capture the finite properties of the complement clauses, such as nominative Case on the null subjects, as an argument for pro and against the possibility of controlled PRO in this environment. Modern Greek is not the only language for which controversial proposals regarding the nature of the null subject in control complements have been made. The recent studies of control in Korean (Gamerschlag 2007; Madigan 2008), unlike Yang (1982, 1985) and Borer (1989), argue that the null subject in control complements is PRO. As discussed above, the cross-linguistic research reveals that non-finiteness is not a necessary requirement for OC and Obligatory Control can occur in a finite context. Moreover, the review of the previous studies of control shows that the structural property of the complement clauses and the syntactic nature of the null subject in finite control are still controversial. This thesis provides an empirical account of finite control in structures containing a pro subject in Korean through exploring the concept of finiteness and the finite properties of control complements in Korean. As for control complements, this thesis attempts to investigate a full range of complementation types including serial verb constructions that have or have not been grouped under the types of control in the literature. Furthermore, this thesis provides ample data indicating that OC in Korean cannot be explained by a solely syntax-based or semantically-based 27 theory and that OC is mainly the result of multiple semantic factors, but syntactic or pragmatic factors also play some roles in determining control. 1.3. Methodology and Data Sources For the purpose of this thesis, a large amount of Korean data has been collected from Korean corpora and from natural speech, and native speakers of Korean have been consulted in order to assure the grammaticality and acceptability of the data included in this thesis. As for the corpora, most of the data in this thesis come from two major Korean corpora: one is the Korean National Corpus, the Sejong corpus, which contains more than 89 million words. I use the raw corpus of modern Korean, which includes around 64 million words from newspapers, magazines, academic works, literary works, quasi-spoken data and other sources. The other corpus is the KAIST corpus, which is a large-scale Korean corpus, containing about 70 million eojeol 5. Ninety matrix predicates have been compiled for this thesis and classified into nine different categories based on their semantic class: 1) Implicative, 2) Aspectual, 3) Directive/Manipulative, 4) Desiderative, 5) Utterance, 6) Propositional, 7) Perception, 8) Factive, and 9) Interrogative predicates. The data involving matrix predicates which are followed by a Modal auxiliary, such as -l su iss ‘can’ and ya hata ‘must/should’, are not selected in this thesis under the assumption that the modal auxiliary may change the structure of the predicates and influence the control effect that may be triggered by the lexical meaning of the predicates. The data collected based on the predicate types have been analyzed with respect to six complementation types: 1) Nominalizations, 2) Adnominalizations, 3) Quotative complements, 4) Interrogative complements, 5) Intentive complements, and 6) 5 An eojeol is a Korean spacing unit which may consist of more than one morpheme. 28 Resultative complements. I look first at the linking of the lexical semantics of matrix verbs and its relevance for control. The results will show how particular classes of verbs behave with regard to control phenomena. Secondly, I examine the correlation between semantic verb class and complementation patterns in order to find out whether there are specific syntactic or semantic restrictions in regards to control effects. 1.4. Structure of This Study In this section, I outline the remainder of this thesis. Chapter 2 provides a brief review of the formal approaches to finiteness, which has been assumed to consist of Tense and Agreement, which assign nominative Case to subjects of clauses. In addition, I will review the cross-linguistic data in the literature revealing that languages may vary in determining finiteness and that the features determining finiteness may not be restricted to just Tense and Agr. Pointing out the problems of Tense and Agr as licensing features of a nominative subject in Korean, I will further explore the relevance of Mood and Modality as the manifestation of finiteness in Korean. Chapter 3 presents a brief review of the classification of OC and the diagnostics for OC that have been previously employed in the literature. Then, I provide evidence for finiteness of control complement clauses in Korean and discuss the existence of OC in a finite clause. This chapter also discusses syntactic and semantic properties of different types of sentential complements in some detail and examines whether there is a correlation between the meaning of matrix predicates and the complement type in determining control. Chapter 4 provides empirical evidence showing that OC complements in Korean are CPs that can allow pro or an overt controlled lexical NP. In this chapter, it is further 29 argued that pro in OC complements is different from the long distance reflexive caki and from a pronoun, providing some evidence supporting the argument. Additionally, this chapter provides some evidence showing that some serial verb constructions in Korean that are claimed to involve non-finite control consist simply of VPs involving argument sharing and argues that those constructions should not be analyzed as OC constructions. Chapter 5 discusses some syntactic issues regarding OC constructions in Korean that can be problematic for the purely syntax-based analyses of control in the literature and argues for the key role of semantics in determining the controller in OC constructions in Korean. However, unlike previous studies arguing that OC is determined by the lexical semantics of the matrix predicates involved, it is argued that OC in Korean cannot be completely explained by the lexical properties of a selecting predicate alone. Rather, it is derived through more complicated semantic factors as well as syntactic and pragmatic factors. Chapter 6 concludes the thesis and summarizes the main points. It further discusses the contribution of this thesis and addresses remaining issues. 30 CHAPTER 2 FINITENESS AS NOMINATIVE SUBJECT LICENSING AND MOOD AND MODALITY AS FINITENESS As shown in the previous chapter, the correlation between non-finiteness and control has been challenged and the existence of finite control in several languages, such as Korean and Balkan languages, has been proposed. The issue of whether an embedded clause in control constructions is finite is crucial since it gives rise to different accounts of the null subject in the constructions. In other words, the controversial issue of whether the syntactic nature of the controlled subject is a PRO or pro is dependent on the finiteness of the embedded clauses. Finiteness has been employed in analyses of syntactic domains and structural Case licensing in subject position in traditional as well as formal literature. For example, in most formal approaches to finiteness, finiteness is defined as the feature licensing nominative Case on subjects of clauses, and Tense and Agreement (Agr) features are considered to be the elements responsible for the assignment of nominative Case. However, the cross-linguistic concept of finiteness in general is still a matter of controversy, and the picture is complicated by the fact that there is no clear mapping from the notion of finiteness to any single morphological or syntactic expression. In fact, there is no clear consensus as to what determines finiteness and languages may vary by employing different morpho-syntactic features that are responsible for finiteness, as will be discussed in the following sections. In this chapter, I will briefly review the formal approaches to finiteness, which have been assumed to consist of Tense and Agr that assign nominative Case to subjects 31 of clauses. In addition, I will review the cross-linguistic data in the literature revealing that languages may vary in determining finiteness and that the feature determining finiteness may not be restricted to just Tense and Agr. Pointing out the problems of Tense and Agr as licensing features of a nominative subject in Korean, I will further explore the relevance of Mood and Modality as the manifestation of finiteness in Korean (and possibly other languages). This analysis basically holds the conventional notion that nominative Case is correlated with finiteness. However, unlike most formal approaches, in this analysis finiteness is also assumed to be closely associated with Mood or Modality as reflected in morpho-syntactic expressions. 2.1. Finiteness as Nominative Subject Licensing 2.1.1. Finiteness and Subject Licensing in Generative Grammar In traditional grammar, the notion of finiteness has been employed to characterize a clause which displays propositional independence and such propositional independence has been argued to correlate with a subject in nominative Case. This is based on the observation that morphological Case marking on NPs denotes grammatical relations and the distribution of Cased NPs is tied to particular positions in the structure: (1) a. She /*Her likes him. b. She likes him/*he. c. She sang for him/*he. (2) a. John believes that she loves him. b. She tried (*she) to finish the work. 32 (1a) and (1b) show that the external argument of the verb likes takes only nominative Case, while the internal argument of the verb takes only accusative Case. (1c) shows that accusative Case is also assigned to NPs that are complements of prepositions. Traditional grammar proposes that the root clauses in (1) are all finite since the verb forms bear morphological Tense and Agreement marking. The traditional view that only nominative subjects can occur in the subject position of finite clauses has been driven by the contrast in subject licensing in the embedded clauses, as shown in (2). Unlike (2a), in which the embedded subject occurs in nominative Case, the subject of the to infinitival clause in (2b) cannot appear overtly. Traditional grammar accounts for the contrast in terms of the finite vs. non-finite distinction. The embedded clause in (2a) is finite since the verb bears Tense and Agr marking, while the to infinitive in (2b), which bears no Tense and Agr marking on the verb, is non-finite. This leads to the relation of subject licensing and finiteness in traditional grammar. That is to say, finite verbs which are inflected for Tense and Agr are associated with a nominative subject, while nonfinite verbs are associated with lack of a nominative subject. The GB framework took the traditional notion of morpho-syntactic finiteness and developed it as a property of a functional category independently projected. That is, the GB framework proposed that the head of the clause, INFL or I, is the syntactic element that can be specified for the Tense and Agr features of verbs. Thus, the distinction between finite and non-finite clauses can be explained by the feature composition of I. To be more specific, the I of a finite clause contains the feature specifications [+Tense, +Agr], whereas the I of a non-finite clause contains the contrast feature [–Tense, –Agr]. Under this dichotomy of I, the occurrence of nominative Case on a subject is correlated 33 with the positive specification for Tense and Agr. This correlation correctly accounts for the grammaticality contrast in the following examples. (3) a. It is likely that John will be sick. b. *Johni is likely that ti will be sick. c. She tried PRO/*she to finish the work. d. John believes her to love Bill. The embedded clauses in (3a) and (3b) are finite so that the subjects of the clauses must be nominative Case-marked. Compared to (3a), in which the overt subject in nominative Case occurs in the embedded clause, (3b) is ungrammatical since the embedded subject of (3b) already gets Case in the embedded clause and has no motivation to move, and having moved gets Case twice. Contrary to (3a) and (3b), the embedded subject cannot be assigned nominative Case in (3c). Contrary to (3a) and (3b), the embedded subject cannot be assigned nominative Case in (3c). Since a to infinitive is not finite, it cannot assign nominative Case to its subject so that the overt nominative subject of the infinitive in (3c) leads the construction to be ungrammatical. 6 As already discussed in 1.2.1.1, the GB theory of Case assumes that there is a PRO subject in non-finite clauses where Case is not licensed. That is, the restricted distribution of PRO into ungoverned positions prevents PRO from bearing Case and from occurring in finite clauses. This leads us to assume that lexical nominals and PRO appear in complementary distribution. 6 This is a violation of ‘the Case Filter’ which is a requirement that every overt NP must be assigned abstract Case. 34 However, not all non-finite clauses take PRO subjects. (3d) shows that some non-finite clauses can allow the overt subjects in accusative Case to appear. In order to explain the Case assignment to subjects of infinitive clauses as in (3d), the GB framework assumes that non-finite I is not a Case assigner and that infinitival IP is not a barrier to outside government. 7 Under this approach, the example in (3d), which is often referred to as Exceptional Case-Marking (ECM), is explained in terms of the subject of the to infinitive clause being assigned accusative Case by the matrix verb believe. In the GB theory of Case, therefore, structural Cases are assigned under the notion of government. That is to say, a verb governs its canonical object, finite I governs the subject of the clause, and an ECM verb governs the subject of its IP complement, as illustrated in (4). (4) a. [VP V NP] b. [IP NP VP] c. [VP V [IP NP . . .] However, within the GB theory of Case, it is not clear exactly what feature is responsible for the nominative Case on the subject of a clause, as already pointed out by many linguists, such as Stowell (1982), Reuland (1983), and Haegeman (1991). If it is the case that the type of clause is determined by the feature specification of I in terms of [Tense] and [Agr], logically four types of I are possible, as given in (5). The notion of “government’ is defined in such a way that “α is governed by β iff α is c-command by β and no barrier intervenes between α and β.” 7 35 (5) a. [+Tense, +Agr] I b. [+Tense, –Agr] I c. [–Tense, +Agr] I d. [–Tense, –Agr] I Given this typology of I, it is apparent that different assumptions as to what feature is responsible for the nominative Case on the subject of the clause will make different predications about nominative Case assignment to the subjects governed by I’s such as in (5b) and (5c) and in turn, make different predictions about finiteness of the clause. If we assume that Tense is the licenser for nominative Case, the I in (5b), but not (5c), assigns nominative Case to its subject and thus (5b), but not (5c), can be classified into finite clauses. However, if we assume that Agr is the licenser for nominative Case, it will bring the opposite result. Chomsky (1980) claims that [+Tense] I assigns nominative Case to the subject of the clause. However, in his later work (1981, 1986), he claims that the [+Agr] feature, not the [+Tense] feature, assigns nominative Case. Also, Case might be assigned only when both features [+Tense, +Agr] are present such that only (5a) would assign it. Indeed, there have been other controversial proposals over the years with respect to what feature assigns nominative Case, as will be discussed in the sections to come. In the early version of Minimalist syntax (Chomsky 1993, 1995), IP is split into TP and AgrP 8 and Case is assumed to be a uniformly uninterpretable syntactic feature which must be checked in a Spec-head configuration. That is, a Case feature on an NP must be checked by the corresponding feature on a Case-licensing head in order for the 8 The split-IP hypothesis is initiated by Pollock (1989). 36 derivation to converge. Under this framework, nominative Case is checked by virtue of matching the Case feature between DP and T. Unlike the GB framework, the most influential early Minimalist proposal concerning PRO (Chomsky & Lasnik 1993, 1995; Watanabe 1993; Bošković 1997; Martin 2001) argues that PRO has Null Case which is checked by nonfinite T. Martin (2001) argues that the ability to check Case on an NP is a property of [+Tense]. Adopting Stowell (1982) and Bresnan (1972), in which [+Tense] in control infinitives is assumed to be in some sense future-oriented, he proposes that control infinitives, like finite clauses, but contrary to ECM and raising infinitives, are specified for Tense. (6) a. [ +Tense, +Finite] checks nominative Case: finite clauses b.[ +Tense, –Finite] checks Null Case: Control infinitives c. [–Tense, –Finite] checks no Case: ECM and raising infinitives (6) shows that Martin associates finite clauses with nominative Case and nonfinite clauses with Null Case or no Case, based on its temporal properties. Under this analysis, finite clauses license a nominative subject, control infinitives license a PRO subject, and other infinitives license other types of subjects, such as a DP-trace or a lexical subject. However, Martin does not discuss what kind of Case [–Tense, +Finite] can check in the mechanism. Although he claims that Case checking of nonfinite T depends crucially on its temporal properties, it is not clear whether or not the same property affects Case checking of finite T. If Martin’s proposal for the Case licensing is on the right track, what we can assume here is that the nominative Case licensing feature is not necessarily 37 a Tense feature. In fact, there has been a claim that Agr is the relevant finiteness parameter questioning the nature of Tense as a finiteness parameter in some languages, such as European Portuguese and Turkish (as will be discussed in the next section). The more recent version of Minimalist syntax (Chomsky 2001, 2006; Landau 2001; Pesetsky and Torrego 2004) proposes that Case checking is satisfied by Agree which is an operation of feature-valuation between Probe and Goal, capturing the relationship between the featural specification of C and I domain. That is, this analysis assumes that C bears some syntactic features associated with the features of T. This turns out to be a crucial factor in terms of whether embedded subject DPs are licensed by the complement clauses or are active for Agree operations with relevant heads in the matrix clauses. According to Chomsky (2001), control clauses, which fall together with finite CPs, can assign Case to their subjects, while raising clauses, which are finite TPs, cannot assign Case and thus the subject NPs are active for Agree operation. 9 However, the connection of Tense and Agr feature to nominative Case has not always been clear and what feature values the nominative Case in minimalist syntax is still controversial. For instance, Chomsky (2001) assumes that nominative Case is not a feature of T, but is valued under Agreement with a phi-complete T probe, where phicomplete T is taken to depend on a phi-complete C. That is, the nominative Case of D itself is not matched, but deletes under matching of phi-features. On the other hand, Pesetsky and Torrego (2004) argue that C, T and DP all have uninterpretable Tense features and that nominative Case is an instance of an uninterpretable Tense feature on 9 In many ways, using CP as a means by which to control “transparency” of the embedded clause to outside factors is analogous to S’ deletion of early Transformational Grammar, but is different in that S’-deletion creates ECM clauses. 38 D which can serve as a goal for an uninterpretable Tense feature on C just as well as T can. Moreover, in Landau (2001, 2004, 2006), finiteness is not necessarily reduced to just Tense and Agreement, but just Tense and Agreement are relevant for the licensing of different kinds of subjects (Adger 2007:40). That is, Landau employs referential features of DPs, [±R], whose specification crucially depends on the specification of [Tense] and [Agr] features, (as discussed in 1.2.1.2.3) and argues that the licensing of PRO vs. overt NPs/pro in embedded clauses 10 depends on the interaction of Tense, Agr, and referential features on the functional heads, C and I. Unlike Martin (2001), Landau (2006) argues that PRO bears standard Case, relying on the phenomenon of Case concord in Icelandic and Balkan languages, as shown in (7). (7) a. Strákarnir vonast til [að PRO the boys.NOM hope for vanta ekki alla PRO.ACC to-lack not all.ACC in the-school ‘The boys hope not to be all absent from school.’ b. Anangasan tin Eleni í skólann]. [PRO na milisi (Icelandic) afti i idhja]. Forced.3PL the Eleni.ACC PRO.NOM PRT speak.3SG she herself.NOM ‘They forced Helen to speak herself.’ c. Ion a ajutat-oi [PROi (Greek) să ajungă eai prima]. John has helped-her.ACC PRO.NOM PRT arrive she.NOM the-first ‘John has helped her to arrive first.’ 10 (Romanian) Similar to Chomsky (2001), Landau assumes that ECM and raising infinitives lack a CP layer. 39 As shown in the examples above, the type of Case on PRO can vary across languages. PRO in an infinitival clause (7a) can bear quirky accusative Case and PROs in subjunctive clauses (7b) and (7c) bear nominative Case. Landau argues that PRO in nominative can occur even in finite clauses that can be specified for [+Tense] and [+Agr]. This implies that finiteness of embedded clauses plays no prominent role in explaining the occurrence of overt and empty subjects of the clauses. As discussed so far, the formal approaches to finiteness characterize finiteness as an ability of licensing nominative subjects, capturing the fact that overt subjects with nominative Case are restricted to finite clauses in many languages. However, there are still controversial issues as to what feature determines finiteness and what the relevant category is hosts finiteness. More importantly, the question of whether finiteness correlates with nominative subjects is essential for control theory, considering that the role of nominative Case in accounting for the overt distribution of DPs plays a central role in accounts of the distribution of subjects in embedded clauses. If Landau’s (2001, 2004, 2006) proposal that PRO in nominative Case can occur in finite contexts is right, the distinction between PRO and pro in terms of finiteness can no longer be maintained. However, the identification of PRO and pro in finite contexts is still a controversial issue that may rely on the notion of finiteness. 2.1.2. Language Variation in Nominative Subject Licensing 2.1.2.1. Tense or Agr as a Nominative Case Assigner As discussed above, both earlier and more recent approaches to finiteness in general have defined finiteness as a licenser of nominative Case on subjects and considered Tense and Agr features to be the relevant features for finiteness. If 40 nominative Case on subjects is an indication of finiteness, a finite vs. non-finite distinction in terms of Tense and Agr features may not be the same cross-linguistically since some non-finite structures with one or the other of (or both) Tense or Agr can assign nominative Case to their subjects. Indeed, it has been already claimed in the literature that languages might have a parametric choice of Tense and Agr with respect to nominative Case assignment. 11 First, consider the languages in which a [+Tense] feature is claimed to assign nominative Case to the subject of the clause. Well-known languages of this sort include Modern Greek (Iatridou 1993), Romanian (Alboiu 2004), and West Flemish (Haegeman 1985). First consider the following Greek data. (8) a.vlepo ton Kosta see na tiganizi psaria. DET K.ACC SUBJ fry.3SG fish ‘I see Kostas fry fish.’ b. *Idha/vlepo ton Kosta na tiganize psaria. (I) saw/see DET K.ACC SUBJ fry.3SG.PST fish ‘I saw/see Kosta fried fish.’ (9) a. elpizo o Kostas na tiganizi psaria hope DET K.NOM SUBJ fry.3SG fish ‘I hope Kostas fries fish.’ 11 Hwang (1997:33) parameterizes nominative Case checking in such a way that T with the [+Tense] feature checks off the nominative Case feature in English-type languages, while T with phi-features removes such a feature in Portuguese-type languages. According to Iatridou (1993), in Classical Greek, [+Agr] feature is the relevant element for nominative Case assignment. 41 b. elpizo o Kostas na tiganise psaria hope DET K.NOM SUBJ fry.3SG.PST fish ‘I hope Kostas fried fish.’ As is the case for most Balkan languages, Modern Greek lacks embedded infinitival complements and generally employs subjunctive clauses in their place. As shown in (8) and (9), the overt subjects of na-subjunctive clauses in Modern Greek can occur in either nominative or accusative. The examples in (8) corresponding to ECM constructions in English bear subject-agreement makers on their verbs and the subjects are marked in accusative Case. When the embedded verbs are marked for past Tense, the clause becomes ungrammatical, as shown in (8b). The embedded verbs in (9) also agree with their subjects in person and number. However, unlike the examples in (8), the examples in (9) show that the subjects of the embedded clauses can be marked in nominative Case and the embedded clause can host past Tense-marking. This indicates that the [+Tense] feature, not the [+Agr] feature, assigns nominative Case to the subject of a clause. Iatridou (1993:178-181) provides further examples supporting the correlation between the [+Tense] feature and nominative Case. (10) a. tha prospathisi na erthi. will tries.3SG comes.3SG ‘He will try to come.’ b. iposchethika ston Kosta na dhiavaso promised.1SG to Kosta afto to vivlio. read.1SG this book 42 ‘I promised Kosta to read this book.’ c. *echi/iche tin kalosini na efige. has/had.3SG the kindness left.3SG.PAST ‘He has/had the kindness to leave.’ The sentences in (10) are examples of subject control in Modern Greek. The embedded verbs in the na subjunctive clauses are marked for Agreement and the subjects of the subjunctive clauses are always coreferential with the matrix subject. Contrary to the examples in (9), there are no overt subjects in the subjunctive clauses in (10). Iatridou argues that the lack of a nominative subject is due to the [-Tense] property of the embedded clause. This is borne out since the embedded verbs of those constructions cannot be marked for past Tense, as can be seen in (10c). Therefore, it is reasonable to think that the contrast between overt and non-overt subjects in (9) and (10) is derived from the opposite value of Tense features. That is, the nominative subjects of the embedded clauses in (9) are due to the [+Tense] features of the embedded clauses. A similar pattern is found in Romanian (Alboiu 2004:57, 63). (11) a. Am reuşit [să plec (*mȋine)]. AUX.1SG managed [SUBJ leave.1SG tomorrow] ‘I managed to leave (*tomorrow).’ b. Victor ȋncearcă (*Mihai) să Victor try.3SG cȋnte (*Mihai) SUBJ sing.3SG ‘Victor is trying (*Mihai) to sing.’ 43 c. Am vrut AUX.1SG wanted [(ca Mihai) să plece mȋine]. [(that Mihai) SUBJ leave.3SG tomorrow] ‘I wanted for Mihai to leave tomorrow).’ The examples in (11a) and (11b), which are the typical examples of subject control, trigger an obligatory Tense dependency between the să subjunctive and the matrix clause. However, (11c), which is not an Obligatory Control structure, shows that the subjunctive clause allows for a Tense domain distinct from that of the matrix clause. That is, unlike (11a) and (11b), (11c) can take the independent semantic Tense denoted by the temporal adverb mȋine ‘tomorrow’. Alboiu (2004:63) takes the absence of independent semantic Tense in control subjunctives to be related to the failure to value nominative Case on the embedded subject DP. The availability of complementizer ca ‘that’ and a lexical subject, contrary to (11b), further shows that the subjunctive complement is specified for Tense. Note that as in Modern Greek, the embedded verbs in the Romanian data agree with their subjects in person and number. This provides evidence that Tense features, not Agr features, play a crucial role in licensing overt subjects in Romanian. Another instance of a language displaying a correlation between a [+Tense] feature and nominative Case assignment is West-Flemish. Haegeman (1986:134) proposes that mee nominative-cum-infinitive (NCI) clauses in West-Flemish 12 are to be 12 According to Haegeman (1986:125), West-Flemish allows a nominative subject to appear with a te-infinitive in the environment of certain prepositions such as mee ‘with’, voor ‘for’, and deur ‘by’, although like Dutch, West-Flemish infinitivals normally take PRO or accusative NP subjects. 44 related to finite tensed clauses and that they have an abstract INFL in COMP, which is specified for [+Tense], but unspecified for [±Past] and [-Agr]. One of the similarities between finite and mee-NCI clauses is found in tet focus constructions, as given in (12). 13 (12) a. Da-tet Valère morgen weggoat. that-tet Valère tomorrow away goes ‘that Valère goes away tomorrow.’ b. Mee tet Valère weg te goan. with Valère away to go ‘because Valère goes away.’ c. *Mee tet PRO weg te goan. with PRO away to go ‘because PRO goes away.’ d. *Dan-k gisteren that-I tomorrow tet Valère zagen weggoan. Valère saw leave ‘that I saw Valère leave tomorrow.’ As shown in the above examples, the focus marker tet can occur in Comp position in both finite (12a) and mee-NCI constructions (12b), while it cannot occur in normal infinitives in which the subject is PRO (12c) or accusative (12d) in form. Thus, Haegeman argues that the NCI in (12b) is finite in the sense that the focus marker tet 13 Haegeman (1986) does not give English translations for the examples in (12) through (16). Therefore, if there is any mistake in the translations, it is mine. 45 signals the presence of a finite COMP. She provides further evidence that mee-NCI constructions pattern together with pure finite clauses. (13) a. Mee dan-k da gisteren gezeid heen goa-se with that-I that yesterday said zie morgen weg. have goes-she she tomorrow away ‘Because I said that yesterday she would go away tomorrow.’ b. Mee dan-k gisteren nie moisten werken heen-k gekust. with that-I yesterday not had-to work have-I cleaned ‘Because I had not to work yesterday I have cleaned.’ c. Mee dan-k morgen moeten werken heen-k gisteren with that-I tomorrow have-to work gekust. have-I yesterday cleaned ‘Because I have to work tomorrow I cleaned yesterday.’ (14) a. Mee ik da gisteren te zeggen hee-se with I that yesterday to say dat hus gekocht. has-she that house bought ‘Because of my saying that yesterday she had bought that house.’ b. Mee ik da gisteren te zeggen goa-se with I that yesterday to say dat hus kopen. goes-she that house buy ‘Because of my saying that yesterday she would buy that house.’ c. Mee ik tnoaste joar weg te goan heen-k dat hus with I next year away to go verkocht. have-I that house sold ‘Because of my going away next year I have sold that house.’ 46 The NCI examples in (14) show that the embedded clauses can select temporal adverbials denoting a time location independent of that of the matrix clauses, just as the pure finite embedded clauses in (13) can. However, non-finite clauses cannot be temporally independent of the matrix clauses and the time reading of the non-finite clause always depends on that of the matrix clause. This is illustrated in the following examples. (15) a. *K-heen vee geld verdiend mee PRO gisteren gazetten te verkopen. I-have much money earned with PRO yesterday newspaper to sell ‘I have much money earned with PRO yesterday newspapers to sell.’ b. K-verdienen vee I-earn geld mee PRO gazetten te verkopen. much money with PRO newspaper to sell ‘I earn much money with PRO newspapers to sell.’ Based on the examples above, Haegeman claims that the INFL of the NCI constructions has Tense just like that of the pure finite clauses and the Tense property assigns nominative Case to the subject. However, Haegeman (1986:127) provides some examples suggesting that unlike pure finite clauses, INFL of NCI constructions with a lexical subject lacks Agreement. (16) a. Mee dan-k da gezeid heen. with that-I that said have ‘Because I have said that.’ 47 b. *Mee-k da te zeggen with-I that to say ‘Because of my saying that’ c. Mee ik da te zeggen with I that to say ‘Because of my saying that’ Contrary to the pure finite clause in (16a), the NCI clause in (16b) does not allow the subject clitic -k. Instead, NCI clauses must use the full pronoun form, as in (16c). Haegeman hypothesizes that the unavailability of cliticization in the NCI constructions may be derived from lack of Agr which can be a nominative Case assigner, assuming that the clitic is a lexical realization of Agr. If Haegeman is on the right track, it appears that the Agr feature has nothing to do with nominative Case assignment in NCI constructions. In contrast to the languages discussed so far, there are some languages in which nominative Case assignment is claimed to be dependent on Agr rather than Tense features. Two well known languages are European Portuguese and Turkish. 14 European Portuguese infinitives exhibit subject Agreement, but do not exhibit Tense distinctions, as shown in (17). (17) a. Será difícil will.be difficult 14 [eles/pro aprov-ar-em a proposta]. they.NOM/pro approve-INF-3PL the proposal In addition to the languages above, Brazilian Portuguese (Lightfoot 1991), Catalan and Italian (Picallo 1984), Irish (McCloskey 1986), and Latin (Cecchetto & Oniga 2001) have been assumed to license nominative case by Agr. 48 ‘It will be difficult for them/pro to approve the proposal.’ b. Será difícil [PRO aprov-ar (*-em) will.be difficult a proposta] PRO approve-INF-3PL the proposal ‘It will be difficult (*for them) to approve the proposal.’ The example (17a) shows that when an infinitival clause is inflected for Agreement, it can take a lexical NP or pro subject, just as a normal finite clause. However, the Agreement morphology is not compatible with a PRO subject. Note that the embedded subject in (17a) contains a lexical subject in nominative Case. This indicates that lack of Tense does not render the structure non-finite. Raposo (1987) argues that in European Portuguese the Agr feature of INFL assigns nominative Case to its subject only if it is itself specified for Case. 15 Another instance of a language for which it is reported that Agreement features are a necessary condition for nominative assignment is Turkish (i.e. George and Kornfilt 1981; Kornfilt 1984, 2007). 15 In fact, Raposo (1987:96) argues that Agr alone cannot assign case, and there must be a sufficiently local Case assigner that is accessible to Agr in order for Agr to receive Case and to assign nominative Case. (i) a. *Eles estão ansiosos [pro votar-em a proposta]. they are anxious pro to vote-3PL the proposal b. Eles estã ansiosos de/por [pro votar-em a proposta]. According to Raposo, (ia) is ungrammatical since adjectives do not assign Case and thus there is no source of Case for the nominal INFL in the complement clause. In contrast, the subject of the inflected infinitive, pro, in (ib) is assigned Case since Agr in (ib) is Case-marked by the dummy preposition de/por and activates its Case-assigning abilty. 49 (18) a. Ahmet [biz-i viski-yi iç-ti] san-iyor. Ahmet.NOM we-ACC whisky-ACC drink-PST believe-PRS ‘Ahmet believes us to have drunk the whisky.’ b. Ahmet [biz viski-yi iç-ti-k] san-iyor. Ahmet.NOM we-NOM whisky-ACC drink-PST-1PL believe-PRS ‘Ahmet believes we drank the whisky.’ The embedded clauses in (18) are inflected for past Tense, but (18a) differs from (18b) in that the former is not inflected for Agr features and has an accusative subject, while the latter is inflected for Agr features and has a nominative subject. Absence of person Agreement on the verb in (18a) correlates with ECM, but the appearance of the Agr morpheme in (18b) makes a nominative subject possible. That is, the examples in (18) seem to demonstrate that Tense cannot determine finiteness without the presence of Agreement, supporting the theoretical claim that Agr, not Tense, assigns nominative Case in Turkish. However, Aygen (2002) provides counter-arguments to the claim and proposes that Mood and Epistemic Modality features are the nominative Case assigner in Turkish. This will be discussed in the next section. 2.1.2.2. Mood, Modality, or Aspect as a Nominative Case Assigner Unlike the general view that Tense or Agr features are the nominative Case assigner, there have been a few proposals that Mood, Modality, or Aspect features are intimately associated with nominative Case assignment. For instance, Aygen (2002) proposes that Mood and epistemic Modality features, not Tense or Agr features, account 50 for nominative Case-licensing in Turkish. First, Aygen (2002:212) provides counterarguments to the claim that Agreement is the relevant finiteness parameter in Turkish. (19) a. Kürşat [biz kek-i Kürşat.NOM we-NOM ye-di-k] san-ıyor. cake-ACC eat-PST-1PL think-PROG ‘Kürşat thinks us to have eaten the cake.’ b. Kürşat [biz-i kek-i ye-di-(k)] san-ıyor. Kürşat.NOM we-ACC cake-ACC eat-PST-(1PL) think-PROG ‘Küşat thinks that we have eaten the cake.’ (19b) shows that Agr can occur in the ECM complement with no impact on the accusative Case assigned to the embedded subject. That is, its optionality in ECM does not affect the unavailability of nominative Case and thus challenges the Agr-based account for finiteness. Aygen (2002) claims that nominative Case in those environments is licensed neither by Tense nor Agr features, but via a composite feature consisting of a Mood feature on C and a epistemic Modal feature on F(inite) head. Consider the following examples. (20) a. Kürşat gel-di. Kürşat.NOM come-PERF/PST.3SG ‘Küşat came/has come.’ b. Ben [Kürşat gel-di] san-dı-m. I. NOM Kürşat.NOM come-PERF/PST.3SG think-PERF/PST.1SG 51 ‘I thought Küşat came/has come.’ c. Ben [Kürşat-ı gel-di] san-dı-m. I. NOM Kürşat.ACC come-PERF/PST.3SG think-PERF/PST.1SG ‘I thought Kürşat came/has come.’ (20a) is a root clause and the examples in (22b) and (22c) are complex sentences involving the root clause. As far as the verbal inflection is concerned, the embedded clause in (20b) and the ECM complement in (20c) are identical to the root clause in (20a). However, unlike (20b), (20c) does not assign nominative Case, but accusative Case. This is problematic for regarding Tense and/or Agr as a nominative Case licenser. Based on the assumption that Agreement morphology in clausal predicates is a manifestation of a Mood feature on C and Lyons (1977)’s proposal that Tense is a kind of Modality, 16 Aygen (2002) argues that the root clause in (20a) and the embedded clause in (20b) are finite in that they both bear [+indicative Mood] via the unmarked Agreement marker and epistemic Modality (denoting the highest possible probability) via the Aspect/Tense marker di. 17 However, the ECM complement clause in (20c) is 16 Lyons (1977) defines Tense as a specific kind of modality as follows: (i) Present is a product of factivity and non-remoteness. (ii) Past is a product of factivity and remoteness. (iii) Future is a product of non-factivity and non-remoteness. 17 According to Aygen (2002:178), the Modal properties of this morpheme have been discussed extensively in the literature. He proposes kinds of Modality and the corresponding conventional terms of ‘tense’ and conditionals, as follows: Epistemic Modality Factivity [+Remote] Past [-Remote] Present Non-Factivity [+Remote] Cfs [-Remote] Future 52 nonfinite since it lacks epistemic Modality. The argument for the absence of epistemic Modality in the ECM complement clause in (20c) is based on the distributional properties of epistemic Modal morphemes in the clauses, which are discussed below. (21) a. Ben [Kürşat gel-iyor-du /miş-ti] /ecek-ti I.NOM Kürşat.NOM come-PROG-PERF/PST/FUT-PERF/PST/PERF-PST-3SG san-dı-m. think-PERF/PST.1SG ‘I thought that Küşat was coming/was going to come/had come.’ b. Ben [Kürşat gel-ebil-ir] I.NOM Kürşat.NOM come-epistemic modality/ability-AOR san-dı-m. think-PERF/PST.1SG ‘I thought that Küşat might/could come.’ c. *Ben [Kürşat gel-meli/meliydi/ebilir/ebilirdi] san-dı-m. I.NOM Kürşat.NOM come-deontic obligation think-PERF/PST.1SG ‘I thought that Küşat might/could come.’ The examples in (21) are finite complement clauses which occur in indicative Mood. As can be seen in (21a), the complex Aspect/Tense markers expressing past events which in turn encode epistemic Modality can occur in the complement clauses. Also, (21b) shows that the Modal marker ebil which can denote epistemic Modality or a deontic 53 Modal of ability can occur in the finite complement clause. 18 However, (21c) shows that finite complement clauses do not allow Modal markers or complex inflectional forms which denote a deontic Modal of obligation. In these respects, ECM complement clauses pattern differently from finite complement clauses. (22) a. *Ben [Kürşat-ı gel-iyor-du /ecek-ti /miş-ti] I.NOM Kürşat-ACC come-PROG-PST/FUT-PST/PERF-PST-3SG san-dı-m. think-PERF/PST.1SG ‘I thought that Küşat was coming/was going to come/had come.’ b. Ben [Küşat-ı gel-di/ecek/iyor/miş/ir/meli/ebil-ir] I.NOM Küşat.ACC come-ASP /deontic modality-AOR san-dı-m. think-PERF/PST.1SG ‘I considered Küşat to have come/to be coming/to have to (to be required to) come/to be able to come.’ The example in (22a) shows that the complex Aspect/Tense markers expressing past events cannot occur in the complement clause. Also, (22b) shows that the ECM clause does not allow epistemic Modality, but allow only Aspect/Tense and deontic Modality morphemes. On the basis of this, Aygen argues that both [+Mood] and [+Epistemic Modality] are required to license nominative Case. Therefore, even when there is 18 According to Aygen (2002: 171), Modal morphemes in Turkish are ambiguous among various types of Modality, unless disambiguated by a Modal adverb. 54 [+Mood], lack of epistemic Modality makes the structure ECM. Aygen extends this analysis to languages other than Turkish, examining ECM constructions in English, European Portuguese, and Japanese. Aygen (2002:236) argues that the redefinition of finiteness accounts for Turkish as well as languages where subject Case varies with Mood (Arabic, Navajo), languages with no Tense morphology but only epistemic Modality (Native American languages studied by Chafe and Hockett), and languages with nominative subjects in infinitivals (European Portuguese, Italian) in a uniform way along with English. In a similar vein, Singh (1994:150) provides problematic data for the Tense or Agrbased analysis of nominative Case assignment and proposes the dependency between Case assignment and the Modal auxiliaries or Aspect in Hindi. 19 (23) a. Use bahut pi:ta: she-ACC very gaya:. beat-PERF-MS go(lit.)-PERF-MS ‘She was beaten very much.’ b. *vah bahut she-NOM very pi:ti: gayi:. beat-PERF-FS go(lit.)-PERF-FS 19 Singh (1994:156-157) excerpts some Palestinian Arabic data from Mohammad (1988:233) and argues that Tense and Agr do not trigger nominative Case in the language. i) raɁytu l-walad-a/*u yadurusu fi l-maktabat-i saw-1S the-boy-ACC/NOM study-3SM in the-library-GEN ‘Lit. I saw the boy studies in the library.’ Adopting Mohammad’s analysis that unlike English Subject-to-Object Raising constructions, the verb of the embedded clause in the above example is fully inflected with Tense, Singh concludes that although the embedded clause is fully inflected for Tense and Agr, the embedded subject NP is accusative and thus the inflectional features cannot assign nominative Case to the subject. 55 ‘She was beaten very much.’ (24) a. laRki: ko dauRna: tha:. girl-FS-OBL ACC run-INF-MS be-PST-MS ‘The girl had to run.’ b. *laRki: dauRni: thi:. girl-FS-NOM run-INF-FS be-PST-FS ‘The girl had to run.’ The examples in (23) show that although there is a possibility of agreement on the verbs, which may assign nominative Case, the subjects do not appear in nominative Case and do not show Agreement with the verb. That is, in the passive sentences in (23), the subject occurs in the accusative Case, not in the nominative Case and the Agreement morphology on the verbs has nothing to do with the subject. This implies that Agr on the verbs does not assign nominative Case to the subject. Similarly, the examples in (24) show that an accusative subject, but not nominative subject, occurs, although Tense and Agr are marked on the verbs. These examples clearly demonstrate that Tense and Agr are not the nominative Case assigners. Instead, Singh (1994) proposes that Aspect and Modality can assign nominative Case. 20 (25) a. laRke kita:b paRhte the. boy-MP-NOM book-FS-ACC read-IMP-MP be-PST-MP 20 In fact, Singh (1994) argues that there are some other factors, such as verbal compounding and transitivity, affecting Case assignment in Hindi and that intransitivity is closely related to nominative Case assignment. However, for the purposes of this thesis, I’ll not address this issue. 56 ‘The boys read/were reading the book.’ b. laRkō ne kita:b paRhi: thi:. boy-MP-OBL ERG book-FS-ACC read-PERF-FS be-PST-FS ‘The boys did read the book.’ c. laRke kita:b paRh sake the. boy-MP-NOM book-FS-ACC read can-PERF-MP be-PST-MP ‘The boys could read the book.’ d. *laRkō ne kita:b paRh saki: boy-MP-OBL ERG book-FS-ACC read thi:. can-PERF-FS be-PST-FS ‘The boys could read the book.’ e. laRkō ne kita:b paRhi: thi:. boy-MP-OBL ERG book-FS-ACC read-PERF-FS be-PST-FS ‘The boys (did) read the book.’ The subject in (25a) takes nominative Case which is phonologically null, while the subject in (25b) takes ergative Case which is realized as ne. Singh (1994:122) accounts for the contrast in the Case assignments with respect to the different Aspect markers on the verbs. That is, in (25a) the nominative Case is triggered by the imperfect Aspect, while in (25b) the ergative Case is triggered by the perfect Aspect. 21 Furthermore, Singh 21 Although the verb in (23a) is marked for the perfect Aspect, the subject does not assign ergative case. This has to do with the passive structure. Singh (1994) argues that Case in Hindi is assigned at d-structure. Based on this, the subject cannot be assigned ergative case at s-structure. 57 argues that a Modal auxiliary can assign nominative Case to its subject. 22 Although (25c) is a perfect transitive clause, the subject takes nominative Case, not ergative Case. According to Singh (1994:190), the presence of a Modal auxiliary in (25c) blocks the assignment of ergative Case, and the argument which is assigned ergative Case elsewhere is assigned nominative in the context of a Modal auxiliary. The ungrammaticality of (25d), contrary to (25e), shows that ergative Case is not compatible with a Modal auxiliary. To sum up, cross-linguistic data in the literature suggest that the solely Tense or Agr-based analysis of finiteness cannot fully account for a variety of nominative Case assignment and that the primary factor in determining finiteness that licenses a nominative subject may vary from language to language. In the next section, I will propose that nominative subjects in Korean cannot be licensed in a way that depends on 22 This is suggestive for the question as to what assigns nominative case to the subjects in English mandative subjunctives. (i) a. It is essential that he be/*is told immediately. b. Mary insisted that he meet/*met her. c. I require that he (*do) not arrive late. d. I require that under no circumstances he do that. As can be seen in (ia) and (ib), the uninflected verb morphology in the subjunctive clauses casts a doubt on nominative Case assignment in terms of Tense and Agr features. Roberts (1993:323-324) proposes that there is a null Modal operator in English mandative subjunctives, based on some syntactic properties of the subjunctives, such as the lack of Do-insertion (ic), the placement of a negative adverb phrase with auxiliary inversion (id). If Roberts is on the right track, it might be possible to assume that the null Modal operator can assign nominative Case to the subject of the clauses in an analogous way in which a Modal auxiliary in Hindi can assign nominative Case. Also, it is worthwhile noting that there have been some functional approaches defining finiteness in terms of grammatical specification for Tense and Modality on the verb on the basis of English verb forms (cf. Langacker 1991; Halliday 1994). 58 the formal features T and/or Agr, but Mood and Modality, 23 adopting the position that finiteness is not a property solely associated with Tense or Agr, but is a property of the C system that may interact with the inflectional domain. 2.2. Mood and Modality as Finiteness in Korean As discussed the previous sections, in generative grammar, finiteness has been identified as a property of a clause that assigns nominative Case to its subject. In formal grammar, Tense and Agreement have been mainly considered to be the overt morphosyntactic expressions of finiteness and not much attention has been paid to Mood and Modality as a property of finiteness. However, functionally oriented approaches to finiteness in the literature often take Mood and Modality into account as syntactic and semantic categories defining finiteness. In fact, in the functional approaches, finiteness has been defined in terms of presence or absence of various syntactic, semantic, and pragmatic factors, such as Tense, Agreement, Modality, illocutionary force, and politeness (Holmberg et al. 1993; Klein 1998; Vincent 1998; Bisang 2007). Some researchers have defined finiteness as a scalar phenomenon which expresses the degree of integration of a clause into its immediate clausal environment (cf. Givón 1990; Dik 1997). The structural reduction of a clause involves the loss of verbal properties (such as Tense, Aspect, Mood, or person/number marking) and the acquisition of nominal properties (such as Case, co-occurrence with determiners, omission of verbal arguments or their encoding as possessors or obliques) on the part of the verb describing the dependent event (Nikolaeva 2007:7). In short, the functional 23 A Mood or Modality-based analysis might offer a new insight into the finite properties of subjunctive clauses in many languages including Balkan languages, and well account for some languages like Chinese and Vietnamese where Tense and Agreement are not lexically realized. 59 approaches in general suggest that there is no uniform relation between finiteness and its expression both cross-linguistically and within individual languages. 24 One important thing to note, however, is that there seems to be general agreement between the formal and functional approaches that absence of certain syntactic or semantic properties is correlated with the inadmissibility of overt subjects and the lack of independent clausehood; however, there is little agreement with respect to what factor indicates finiteness. In the following sections, I will provide some problems for the Tense or Agr-based approaches to finiteness in Korean, and propose that Mood and Modality on verbs and complementizers, depending on clause type, are the crucial factors determining finiteness in Korean. This proposal is mainly based on the formal approaches in which finiteness is defined as capacity to license nominative Case to the subject of a clause. However, the proposal is also in line with the functional approaches, arguing that the category defining finiteness is not restricted to the morphosyntactic feature Tense and Agreement and actually involves various syntactic and semantic factors which are language-specific and construction-specific. 2.2.1. Defectiveness of Tense and Agr in Nominative Case Assignment In the literature on nominative Case assignment in Korean, it is generally held that Agr is responsible for the assignment of nominative Case (c.f. H.-S. Han 1989; Ahn & Yoon 1989; Ryu 1994; Yang 1996). Agr in Korean is realized as the subject honorific morpheme, -si, which is marked on the verb and is limited to contexts where the referent can be honored by the speaker. Unlike English, in which Tense and Agr never occur separately, Agr in Korean occurs independently of Tense, as illustrated in (26). 24 Cristofaro (2007) provides a detailed survey of functional approaches to finiteness. 60 (26) a. Mina-nun [nay-ka ku M.-TOP il-ul kkuthnay-(*ess)]-tolok towa-cwu-ess-ta. I-NOM the job-ACC finish-PST-C help-give-PST-DC ‘Mina helped me to finish the job.’ b. Mina-nun [kyoswu-nim-i M.-TOP ku il-ul kkuthnay-si-(*ess)]-tolok professor-HON-NOM the job-ACC finish-HON-PST-C towa tuli-ess-ta. give-PST-DC ‘Mina helped the professor to finish the job.’ c. Sensayng-nimi-kkeyse Minaj-eykey [tangsini-i teacher-HON-NOM.HON M.-DAT ki-o ttena-si-(*ess)]- self.HON-NOM leave-HON-PST- yaksokha-yess-ta. NOML-DR promise-PST-DC ‘The teacher promised Mina to leave.’ The embedded clauses in (26) have been taken to be typical examples of non-finite clauses in the literature. This is because the constructions are rendered ungrammatical if the embedded verb takes the past Tense or perfect Aspect morpheme -ess. In (26a) the honorific marker does not appear. However, in (26b-c) the subjects of the embedded clauses are to be honored and thus the honorific marker si does occur in agreement with the subject. The fact that these are analyzed by many as nonfinite clauses and the subjects take nominative Case suggests that Agr features, not Tense features, assign nominative Case in Korean. 61 However, the subject honorific marker does not always co-occur with a nominative subject. Consider the following examples. (27) a. Emenim-eykey-nun isiptay-ka Mother-DAT-TOP kacang hayngpokha-si-ess-ta. twenties-NOM most be happy-HON-PST-DC ‘Her twenties was the happiest time for the mother.’ b. Sensayngnim-uy son-i khu-(*si)-ta. teacher.HON-GEN hand-NOM big-HON-DC ‘The teacher’s hands are big.’ c. Na-eyekey ce I-DAT meli-ey iss-nun emenim-i poi-(*si)- that far away-LOC to.be-ADN mother.HON-NOM seen-HON- n-ta. PRS-DC ‘The mother who is far away is seen to me.’ d. Apecii-kkeyse halapenimj-kkey [tangsini-i cip-ey fatehr-NOM.HON grandfather.HON-DAT.HON self.HON-NOM home-LOC ka-(*si)-keyss-ta]-ko yaksokha-si-ess-ta. go-HON-VOL-DC-C propmise-HON-PST-DC ‘Father promised the grandfather to go home.’ The honorific marker in (27a) agrees with the non-nominative subject emenim ‘mother’, not with the nominative subject isiptay ‘twenties’. One might assume that the NP in nominative Case in (27a) triggers the honorific marking on the verb since the non- 62 nominative NP who deserved to be honored by the speaker is in an inalienable possession relation with the NP in nominative Case. In fact, Korean people often overuse the subject honorific agreement in a similar situation. For example, in (27b) the occurrence of the honorific marker is, strictly speaking, ungrammatical since the nominative subject son ‘hand’ is not the referent who can be honored by the speaker. However, many Korean people consider the sentence acceptable because the possessor of the son ‘hand’ is the genitive subject sensayngnim ‘teacher’ who deserves to be honored. Therefore, some studies (Suh 1977; Yoon 1989) claim that -si is not a grammatical agreement marker, but is pragmatically oriented. However, there are some Cases in which the honorific marker -si cannot occur, even when the nominative subject is directly honored by the speaker. The examples in (27c) and (27d) show that the occurrence of the honorific marker makes the sentences ungrammatical, although it agrees with the nominative subject apenim ‘mother’ and tangsin ‘self’, respectively. This clearly demonstrates that the honorific marker need not be present for the assignment of nominative Case in Korean. Also, an analysis that Tense assigns nominative Case is not tenable in Korean. As already shown in (26), a non-finite complement clause is not compatible with the overt past Tense marker -ess, but the nominative subject is still possible. Yim (1985) claims that nominative Case in a non-finite clause is assigned by an abstract Tense element. This is in line with the common view that some English infinitives have their own ‘semantic Tense’ (Bresnan 1972; Stowell 1982; Bošković 1997; Landau 2000; Martin 2001; Pesetsky and Torrego 2004). Landau (2000:57-59) motivates this semantic Tense 63 in terms of the presence vs. absence of a temporal adverbial that is in conflict with another temporal adverbial modifying the matrix clauses, as follows. (28) a. *Yesterday, John began to PRO solve the problem, tomorrow. b. *Today, John managed to PRO have finished his duties yesterday. c. Yesterday, John hoped to PRO solve the problem, tomorrow. The examples in (28a) and (28b) show that the event time of the complement clause is simultaneous with the matrix event and thus, precludes the occurrence of a temporal adverb different from the matrix event. In contrast, the example in (28c) shows the possibility of conflicting temporal adverbs between the matrix clause and the complement clause and so the event time of the complement clause is not the same as the matrix clause. According to Landau, the aspectual and implicative infinitives in (28a) and (28b) are untensed, whereas the desiderative infinitive in (28c) is tensed – specifically, specified for irrealis Tense. 25 Landau (2004:822) calls the Tense in (28c) dependent Tense which describes a situation where the Tense of the embedded clause is constrained by (though, crucially, not necessarily identical to) the matrix Tense. 26 Similar to the desiderative infinitive in English, non-finite complements in Korean can carry their own semantic Tense, as follows. 25 There is some disagreement with respect to the classification of tensed vs. tenseless infinitives. Contrary to Landau (2000), Bošković (1997) and Martin (2001) argue that implicative complements are tensed. 26 According to Landau, whether the embedded tense is dependent or independent is determined by the semantic class of the matrix verb. 64 (29) a. Ecye, Minai-ka Pata-eykey [nayil yesterday, M.-NOM P-DAT ki-lo (kunyei-ka) cenyek-ul sa]- tomorrow she-NOM dinner-ACC buy- yaksokha-yess-ta. DNOML-DR promise-PST-DC ‘Yesterday, Mina promised Pata to buy dinner tomorrow.’ b. Ecye, Minai-ka Pataj-eykey [nayil yesterday M.-NOM P.-DAT (kunyej-ka) ttena]-tolok tomorrow she-NOM leave-C seltukha-yess-ta. persuade-PST-DC ‘Yesterday, Mina persuaded Pata to leave tomorrow.’ The examples in (29) show that the complement clauses can take the temporal adverb nayil ‘tomorrow’ which is in conflict with the temporal adverb modifying the matrix clause ecey ‘yesterday’. This may suggest that the dependent Tense independently assigns nominative Case to the embedded subjects. However, if these examples are tensed, but still non-finite, it is not clear why the tensed non-finite clauses in Korean can take overt nominative subjects, while the English tensed infinitives cannot. Moreover, the following examples show that not all (of the so-called) non-finite clauses can take dependent Tense. (30) a.*Ecye, Minai-ka [nayil kunyei-ka cikcep ku kakey-lul yesterday, M.-NOM tomorrow she-NOM in.person the store-ACC 65 wunyengha]-ki-(lul) sicakha-yess-ta. run-NOML-ACC begin-PST-DC ‘*Yesterday, Mina began to run the store in person tomorrow.’ b.*Ecye, Minai-ka [nayil cakii-ka ku mwuncey-lul phwul]-leyko yesterday M.-NOM tomorrow self-NOM the problem-ACC solve-C sitoha-yess-ta. try-PST-DC ‘*Yesterday, Mina tried to solve the problem tomorrow.’ As shown in (30), the complement clauses selected by the aspectual verb sicakhat‘begin’ and the implicative verb sitoha- ‘try’ cannot host semantic Tense distinct from the matrix clauses, just as the English infinitives in (28a, b). However, note that unlike the English infinitives, nominative subjects can still occur in the complement clauses. This is shown in the following examples. (31) a. Minai-ka [kunyei-ka cikcep ku kakey-lul wunyengha]-ki-(lul) sicakha- M.-NOM She-NOM in.person the store-ACC run-NOML-ACC begin- yess-ta. PST-DC ‘Mina began to run the store in person.’ b. Minai-ka [cakii-ka ku mwuncey-lul phwul]-leyko sitoha-yess-ta. M.-NOM self-NOM the problem-ACC solve-C ‘Mina tried to solve the problem.’ try-PST-DC 66 The examples in (31) show that the sentences in (30) are grammatical if the conflicting temporal adverbs do not appear and that nominative subjects can occur in the complement clauses. This suggests that nominative Case in complement clauses is assigned by something other than Tense features. Note further evidence that Tense cannot assign nominative Case. (32) a. Mina-ka hakkyo-ey ka-*(ta). M.-NOM school-LOC go-DC ‘Mina went to school.’ b. Mina-ka hakkyo-ey ka-ss-*(ta). M.-NOM school-LOC go-PST-DC ‘Mina went to school.’ c. Ney-ka hakkyo-ey ka-(*ss)-*(la). you-NOM school-LOC go-PST-IMP ‘You go to school.’ d. Wuli-ka hakkyo-ey ka-(*ss)-*(ca). we-NOM school-LOC go-PST-PROP ‘Let’s go to school.’ Unlike root clauses in English, which must be headed by a tensed element, root clauses in Korean cannot end with a bare Tense morpheme and are obligatorily headed by a Mood suffix which indicates the speaker’s illocutionary force. As shown in the above 67 examples, the Mood suffixes are not tensed and appear outside the position of Tense morphology. Note that the verb form in (32a), unlike that in (32b), is temporally or aspectually unspecified, but denotes a past event. This temporal interpretation is gained by the declarative Mood marker -ta which asserts the event as an objective fact. Also note that nominative subjects can appear in imperative sentences (32c) and in proposative sentences (32d). Imperative and proposative sentences are commonly thought of as nonfinite clauses in that compared to declarative sentences they tend to show reduced Tense and Agreement morphology and show no overt subject. For example, the English sentences corresponding to the Korean imperative and proposative sentence in (32c) and (32d) cannot take an overt nominative subject. According to Hwang (1997) who assumed that nominative Case is checked by a [+Tense] feature, imperative sentences in English are [–Tense, +Agr] and thus their lexical subjects cannot be marked with nominative Case, but bear unmarked accusative Case. He supports the claim, relying on accusative Case marking of conjoined DP subjects in imperatives, as shown in (33). (33) a. You and him/her/them/*he/*she/*they make a deal! b. Don’t you and him/her/them/*he/*she/*they fight again! Considering that imperative and proposative sentences in Korean are also not tensed, but nominative Case is still assigned to the subject in those sentences, it suggests that the Mood suffixes are responsible for assigning nominative Case. This is further supported by the following data. 68 (34) a. John-i [Mina-ka/lul chencay-(yess-ta)/la-ko] mit-ess-ta. J-NOM M-NOM/ACC genius-(PST-DC)/DC-C believe-PST-DC ‘John believed that Mina was/is a genius.’ b. John-i [Mina-*ka/lul chencay-lo] mit-ess-ta. J-NOM M-NOM/ACC genius-as believe-PST-DC ‘John believed Mina to be a genius.’ The matrix verb mit- ‘believe’ has been widely accepted as an ECM verb in the literature. However, the embedded clause in (34b) can be analyzed as a small clause in that its predicate is a Prepositional Phrase without verbal morphology such as Tense or Mood, and its subject bears only accusative Case. On the other hand, the embedded clause in (34a) can bear past Tense and the declarative Mood suffix ta/la, and its subject can bear either nominative or accusative Case. These differences seem to be derived from the fact that the small clause does not bear any Mood suffix, contrary to the embedded clause in (34a) where the declarative Mood marker occurs. Note that ECM in Korean shows very unique characteristics in that it is possible to assign Case into finite clauses and it is optional: the subject is not obligatory ECMed. 27 To be more specific, Korean ECM differs from English ECM in that the former bears Tense morphology and hosts the complementizer -ko, while the latter does not bear any Tense morphology and hosts a to infinitive. In addition, Korean ECM is similar to the Turkish ECM structure in (20c) in that the embedded clause can bear Tense morphology. However, compared 27 Also, it is well known that Korean ECM constructions are confined to contexts where the embedded predicate is a member of the adjectival or copular subset of stative predicates (Yoon 1989). 69 to Turkish ECM which Aygen (2002) considers non-finite because its subject is marked with accusative Case due to lack of epistemic Modality, Korean ECM is analyzed as a finite clause in that it takes a declarative Mood suffix. Therefore, the embedded verb is capable of assigning nominative Case to its subject, as it in the case of the occurrence of nominative subject in (34a). Then, a natural question arises as to how accusative Case is licensed on the subject in ECM constructions. I assume that the DP receives a nominative Case in the embedded subject position, while the DP receives an accusative Case from the matrix verb in Spec of CP of the lower clause. DP-movement of the embedded subject to Spec, CP in (34a) is motivated by the need to have a [+Focus] feature checked off. 28 This suggests that the contrast in the Case marking of the embedded subjects in (34a) does not result from the difference in finite properties of the embedded clause. Since the presence of Tense or Agr does not necessarily result in the assignment of nominative Case on the subject of clauses in Korean, it is necessary to propose a third mechanism for nominative Case-assignment in Korean. I propose that Mood or Modality licenses nominative Case on the subject in Korean. 2.2.2. Mood and Modality on Verbs and Complementizers and Their Projections Korean verb stems cannot stand alone and are followed by a series of verbal endings which perform various grammatical functions, such as Tense, Aspect, sentence type, speech level, and conjunction. The combination of verbal endings is very complex and each ending interacts closely with others in realization of its grammatical functions. 28 J.-M. Yoon (1989) and J.-S. Lee (1991) argue for a similar analysis in more depth. 70 The following are examples of two of the possible largest verbal paradigms and a rough representation of the paradigms. (35) a. cap-hi-si-ess-ess-keyss-sup-ni-kka? catch-PASS-SH-PST-PERF-CONJ-AH-IND-Q ‘Do you think he/she might have been caught?’ b. cap-hi-si-ess-ess-keyss-te-la-(ko) catch-PASS-SH-PST-PERF-CONJ-RET-DC-C ‘(It looked like he/she) had been caught.’ c. Verb stem-Causative/Passive-Subject Honorific-Tense-Aspect-ModalAddressee.Honorific-Modal-Sentence type-Complementizer As can be seen in (35), the verb stem cap- ‘catch’ can be followed by complex verbal endings: the derivational suffix ‘passive/causative’, the subject honorific suffix -si, the complex Tense/Aspect suffixes ess-ess, denoting past perfect, the Modal suffix -keyss, denoting the speaker/hearer’s conjecture, the retrospective Modal suffix -te, the addressee honorific suffix -sup, the indicative Modal suffix -ni, sentence type suffixes, and the quotative complementizer -ko. Mood, denoting illocutionary force/sentence type, is marked on the verbs by a sentence final ending, and Modality can be marked on the verbs by a prefinal ending. 29 The sentence final Moods include declarative, imperative, 29 In general, Modality is considered to be a semantic domain, and Mood is treated as a grammaticalization of Modality and thus is expressed in the verbal paradigm. However, Mood and Modality are defined and classified in various ways and frequently overlap in the literature. In this thesis, I will distinguish Mood from Modality in that the former is represented by a sentence final ending marking illocutionary force and the latter is represented elsewhere. 71 interrogative, and proposative and their marking is obligatory in main clauses, while Modality includes injective, retrospective, indicative, and so on, and its marking is optional. Thus, the optional Modality marking on verbs is not relevant to the issue of finiteness in this thesis. As briefly discussed in 2.2, formal approaches to finiteness often treat Mood, which is associated with different speech acts, as a subcategory of finiteness. In fact, from a perspective of a functional approach, Bisang (2007:129) proposes that Korean finite clauses are determined by a set of suffixes occurring in the final slot of the verbal paradigm. These suffixes are characterized by the interaction of illocutionary force and politeness, as shown in table (36) cited from Bisang. (36) Sentence Enders 30 Declarative Plain po-n-ta see-IND-DC Intimate po-a see-INF Familiar po-ney see-DC Blunt po-o see-BL Polite po-a-yo see-INF-POL Deferential po-p-ni-ta see-AH-IND-DC 30 Interrogative Imperative Proposative po-ni see-Q po-a see-INF po-na see-Q po-o see-BL po-a-yo see-INF-POL po-p-ni-kka see-AH-IND-Q po-a-la po-ca see-INF-IMP see-PROP po-a po-a see-INF see-INF po-key po-sey see-IMP see-PROP po-o __ see-BL po-a-yo po-a-yo see-INF-POL see-INF-POL po-si-p-si-o po-p-si-ta see-SH-AH-RQ-IMP see-AH-RQ-PROP The sentence enders are based on Sohn (1999:355-358) and the list of the sentence enders in (36) does not include all possible sentential endings, such as the apperceptive suffix -kwun and the promissive -ma. 72 It is beyond the scope of this thesis to give a full account of how the illocutionary force and politeness interact with respect to finiteness, examining the full Mood and Modality systems in Korean. What is important here, however, is that neither Tense nor Agr is the crucial factor in determining finiteness. Since the sentence final endings, which denote illocutionary force as well as politeness, are the distinctive factors for finiteness in Korean root clauses, they are also most likely the crucial factor for finiteness in the subordinate clauses where they can occur. (37) a. Na-nun [Kiho-ka tomangka-ss-ta]-nun I-TOP sasil-ul al-ass-ta. K.-NOM run.away-PST-DC-ADN fact-ACC know-PST-DC ‘I knew that Kiho ran away.’ b. Na-nun Mina-eykey [Kiho-ka tomangka-ss-ta]-ko I-TOP M.-DAT malha-yess-ta. K.-NOM run.away-PST-DC-C tell-PST-DC ‘I told Mina that Kiho ran away.’ c. Mina-ka [Kiho-ka tomangka-ss-e-yo]-lako malha-yess-ta. M.-NOM K.-NOM run.away-PST-INF-POL-C tell-PST-DC ‘Mina said, “Kiho ran away”.’ As can be seen in (37), the sentence-final Mood markers can occur in a subordinate clause, although the occurrence of the politeness forms, except the plain forms, are restricted to direct quotative sentences (37c). 73 Now, consider the following subordinate clauses in which Mood markings do not occur on the verbs. (38) a. Mina-ka [Inho-ka ttena-(*yess)]-ki-lul kangyoha-yess-ta. M.-NOM I.-NOM leave-PST-NOML-ACC force-PST-DC ‘Mina forced Inho to leave.’ b. Na-nun [ku-ka I-TOP mwusahi tochakha-yess]-ki-lul he-NOM safely pala-n-ta. arrive-PST/PERF-NOML-ACC want-PRS-DC ‘I hope that he arrived/has arrived safely.’ c. Na-nun [imi I-TOP ku-ka ttena-ss]-um-ul al-ass-ta. already he-NOM leave-PST/PERF-NOML-ACC know-PST-DC ‘I knew that he already left/had left’ In Korean, there are two types of sentential nominalizers -um and -ki. The motivation for the analysis of -ki and -um as nominalizers is that they can be followed by Case, such as accusative Case -(l)ul, and behave as noun-like elements. Compared to the sentential nominalizer -um, which is compatible with the past Tense or perfect Aspect suffix -ess, the nominalizer -ki shows some restrictions in taking the suffix. The distributional difference between the two nominalizers has long been a controversial issue in the literature. However, it is generally argued among Korean linguists that the nominalizers carry inherent Modalities; -um carries realis Modality, while -ki carries irrealis. Thus, the contrast in their Modalities has been argued to cause the 74 complementary distribution; -um clauses describe factual events, while -ki clauses often denote non-factive events (this will be discussed in detail in section 3.3.1.) Note that unlike the following Turkish nominalization in (39a, b) cited from Kornfilt (2007:316-317), overt subjects in nominative Case are possible in Korean nominalization. sınav-ı (39) a. [Sen-in/*φ geç-eceğ-in]-i bil-iyor-um /bil- you-GEN/NOM test-ACC pass-FUTFN-2SG-ACC know-PRSPR-1SG/knowiyor-du-m. PROG-PST-1SG ‘I know/knew that you will/would pass the exam.’ sınav-ı b. [Sen-in/*φ geç-me-n]-i isti-yor-um /iste-di- you-GEN/NOM test-ACC pass-NFN-2SG-ACC want-PRSPR-1SG/want-PSTm/isti-yeceğ-im. 1SG/want-FUT-1SG ‘I want/wanted/will that you should pass the exam.’ c. [Sen sınav-ı geç-ti-n] san-ıyor-um. you-NOM test-ACC pass-PST-2SG believe-PRSPR-1SG ‘I believe you passed the test.’ (39a) is an example of factive (= indicative) nominalized embeddings, which are temporally independent from the matrix clause, and (39b) is an example of non-factive (= subjunctive) nominalized embeddings which are dependent on the matrix temporally. 75 Compared to the verbal embedded clause in (39c), in which the subject bears nominative Case, the subjects in the nominalized clauses are genitive. Kornfilt (2007) claims that the Agreement markers on the clauses give rise to syntactic finiteness and it licenses nominative subjects in verbal embedded clauses, while it licenses genitive subjects in nominal clauses. That is, overtness of subject is closely related to finiteness in Kornfilt’s analysis of Turkish. What is more interesting here is that the nominalizers occur in the slot for a Tense morpheme in a verbal clause. Kornfilt claims that temporal independence is expressed by the nominalization marker itself, just as in fully verbal clauses. This holds true of Korean nominalizers, although they do not appear in the slot for Tense or Aspect marker as do Turkish nominalizers, as can be seen from the fact that a Tense/Aspect suffix can occur with them, as in (38b, c). To be more specific, the irrealis Modality on the nominalizer -ki denotes a non actual event and thus provides an unrealized future interpretation, while the realis Modality on the nominalizer -um denotes an actual event and thus by implication provides a non-future Tense interpretation. In fact, there are many languages that show no morphological difference between irrealis Modality and future Tense. Therefore, in (38a), the -ki nominalizer encodes that the event of the nominalized clause is not actual and is unrealized with respect to the matrix verb. However, the -ki nominalizer in (38b) does not encode that the event of the nominalized clause is unrealized. Rather, it denotes the speaker’s attitude toward the actuality of the event. That is, the past Tense or perfect Aspect marker implies that the event of ‘arriving’ (has) happened, but the -ki nominalizer encodes the speaker’s unsureness about whether the event (has) happened in a safe manner or not. Thus, the -ki 76 nominalizer with the irrealis Modality can be compatible with the past Tense or Aspect marker. In addition, Irish is another language demonstrating that Tense distinctions can be seen across a whole range of complementizer-like elements, including relative particles, negative complementizers, and complementizers signaling temporal relations (Adger 2007:34). Consider the following Irish examples from Adger. (40) a. Rith leat sula bhfeicfear tú run with before see.PASS you ‘Run along before you are seen.’ b. D’éag sé sular thánig an sagart. die.PST he before.PST come.PST the priest ‘He died before the priest came.’ The complementizers sula and sular which head adverbial clauses show a past vs. nonpast distinction in Irish, although Tense can also be marked on the verbs. A similar phenomenon is found in Korean adnominalized sentences modifying a following noun phrase by means of the adnominal endings, ‘-(u)n, -(nu)n, and -(u)l’. 31 (41) a. Na-nun [Kiho-ka o-(*ss)]-l I-TOP kes-ul al-ass-ta. K.-NOM come-PST-ADN thing-ACC know-PST-DC ‘I knew that Kiho would come’ 31 In the literature on Korean, the categories of the nominalizers and adnominalizers are often treated as complementizers. 77 b. Na-nun [Kiho-ka o-(*ss)]-nun I-TOP kes-ul al-ass-ta. K.-NOM come-PST-ADN thing-ACC know-PST-DC ‘I knew that Kiho was coming.’ c. Na-nun [Kiho-ka o-(*ss)]-n I-TOP kes-ul al-ass-ta. K.-NOM come-PST-ADN thing-ACC know-PST-DC ‘I know that Kiho came.’ (42) a. Na-nun [Kiho-ka keki-ey I-TOP ka-ss]-ul kes-ulo sayngkakha-n-ta. K.-NOM there-LOC go-PST/PER-ADN thing-DR think-PRS-DC ‘(lit.) I think that probably, Kiho went/have gone there’ b. Na-nun [Kiho-ka keki-ey I-TOP ka-(*ss-ess-te)]-nun kes-ul mol-ass- K.-NOM there-LOC go-PST/PER-RET-ADN thing-ACC forget-PST- ta. DC ‘I didn’t know that Kiho was going/goes there’ c. Na-nun [Kiho-ka keki-ey I-TOP ka-ss-ess-te]-n kes-ul ic- K.-NOM there-LOC go-PST-PER-RET-ADN thing-ACC forget- ess-ta. PST-DC ‘I forgot that Kiho had gone there’ The examples in (41) show that no past Tense suffix can occur in the adnominalized clauses. Instead, the adnominalizers -(u)n, -(nu)n, and -(u)l show contrast in Tense; -(u)l denotes future Tense, -(nu)n present Tense, and ‘-(u)n past Tense. However, the 78 examples in (42) provide more information about the adnominalizers. Compared to -nun in (42b), which is not compatible with any Tense, Aspect, or Modal suffix, -ul in (42a) can take the past Tense or perfect Aspect suffix -ss. Despite the appearance of the suffix -ss, the adnominalized clause in (42a) expresses that the speaker does not know that the actual event of ‘going’ has happened at the moment of speech. That is, the suffix -ss implies that the event of going (has) happened, but the irrealis Modality of -ul overrides the actuality of the event and conveys the speaker’s uncertainty about whether the event really happened. The other adnominalizer -(u)n can take the retrospective Modal suffix -te or its combination with the past Tense and the perfect Aspect suffix, as in (42c). Note that without any Tense or Aspect suffix, the adnominalized clause in (42b) can express the habitual action of ‘going’ or an ongoing action that is happening at the time of speech event. This tells us that the adnominalizer carries certain Modality. Indeed, it is generally argued that the adnominalizers as well as the nominalizers in Korean encode certain Modalities and that the Modalities can be interpreted from the viewpoint of Tense or Aspect. This is illustrated in Table (43). (43) Modality on Nominalizers and Adnominalizers 32 Modality Irrealis/Non-factive Realis/Factive 32 Nominalizers -ki -(u)m Adnominalizers -(u)l -nun -(u)n Tense/Aspect Future Present (Progressive) Past Alternatively, it is often argued in the literature that -nun and -(u)l denote indicative and prospective Mood/Modality, respectively. 79 So far, it has been argued that the nominalizers and adnominalizers in Korean carry certain Modality and that certain Tense distinctions on the adnominalizers are derived from their inherent Modalities. Note that the nominalizers and adnominalizers are attached to the verb forms as sentence final endings and that in those clauses Tense morphology is not allowed or is restricted to certain contexts, yet the occurrence of a nominative subject, which is one of the important criteria for finiteness, is allowed. This suggests that the Modality on the nominalizers or adnominalizers determines the finiteness of the clauses. That is, when a Mood ending, which encodes illocutionary force and politeness, becomes unavailable in an embedded clause, the Modality on a nominalizer or adnominalizer takes over the role of the Mood ending and determines finiteness of the clause. This analysis is based on two basic assumptions. First, the nominalizers and adnominalizers in Korean occur in a C position. Second, Mood and Modality are an inherent property of the clause, adopting Frajzyngier’s (1995) analysis in which a complementizer is considered to be a component of some types of clauses on the condition that a complementizer is understood as representing the Modality of the clause. 33 In a similar vein, I propose that in the Cases of subordinate clauses, where no Mood endings can occur, a complementizer takes the role of the Mood endings. That is, a complementizer encodes a certain Modality of a clause; i.e. the speaker’s attitude toward the proposition determines finiteness of the clause. Consider the following examples. 33 Frajzyngier argues that occurrence of a complementizer is necessary as a Modality marker of a clause, but a main clause does not need to have a complementizer if inherently marked. To be specific, he argues that indicative sentences have an inherently defined epistemic Modality and in non-indicative main clauses the Modalities are encoded by syntactic devices such as word order, and morphological changes on the verb. 80 (44) a. Minho-nun [caki-ka M.-TOP ku sang-ul tha-(*ss)]-leyko ayssu-ess-ta. self.-NOM the prize-ACC win-PST-C endeavor-PST-DC ‘Minho endeavored to win the prize.’ b. Emeni-nun [Mina-ka na-wa hammkkey ka-(*ss)]-tolok helakha-si-ess-ta. Mother-TOP M.-NOM I-with together go-C allow-HON-PST-DC ‘Mother allowed Mina to go with me.’ c. Swuni-ka ka-(*ss)-ya S.-NOM Minho-to ka-n-ta. go-PST-only.if M.-also go-PRS-DC ‘Minho goes only if Swuni goes.’ (44a) and (44b) are examples of complement clauses and (44c) is an example of an adverbial subordinate clause. None of the subordinate clauses above allow the past Tense suffix -ss to appear, but overt nominative subjects are still available. Interestingly, the complementizers in the examples above carry an inherent lexical meaning which can be interpreted as a Modality. The complementizer -leyko expresses the speaker’s intention or purpose which is related to an intentional or desiderative Modality, and tolok expresses an accomplished situation that the speaker commands or insists the addressee do and thus is related to an obligative or necessitative Modality. The subordinating marker -ya ‘only if’ carries a conditional Modality. That is, the semantic Modalities of the complementizers anchor a proposition in a possible world and establish their relation to finiteness of the clause. In conclusion, I hypothesize that the presence of Mood endings on verbs or Modalities on complementizers licenses a 81 nominative subject in Korean. 34 This analysis is in line with the proposed theory that C domains are responsible for the realization of finiteness (cf. Platzack & Holmberg 1989; Rizzi 1997; Krapova 2001). 35 Rizzi (1997) proposes an articulated CP domain of clause structure given in (45). (45) [ForceP [TopicP [FocusP [TopicP [FinP [IP…]]]]]] The highest C projection, ForceP, is related to illocutionary force, while the lowest C domain, FinP, can be specified for Mood, Tense, and Agreement and encodes finiteness of a clause. In Rizzi’s system, the prepositional complementizer for in English can be situated in the head of FinP and the complementizer that which selects a finite clause is located in the head of ForceP. Roussou (2000) adapts Rizzi’s version of the CP domain 34 Since a Modality is encoded by a considerable range of syntactically quite diverse item in Korean, such as Modal auxiliaries, Modal adverbs, and non-final Modal suffixes on verbs, questions necessarily arise as to how they interact to each other with respect to their relation to finiteness and whether there are some specific modalities which determine finiteness of a clause. Recall that Aygen (2002) argues that an epistemic Modality, not a deontic Modality is required to license nominative case in Turkish. I do not offer an answer to these questions hear, but leave them to future research. One point to make, however, is that a modality which determines finiteness of a clause is closely related to the sentence final position. Recall that Mood endings which encode illocutionary acts obligatory occur in the final slot of the verbal paradigms, as shown in (35). Similarly, the presence of a complementizer carrying a certain modality is obligatory in a subordinate clause, while non-final verbal endings which carry certain modalities are not obligatory. In this respect, finiteness is defined as a structural primitive occurring at the highest clausal domain. 35 In particular, Platzack & Holmberg (1989) argue that finiteness is an operator in C which has to be realized in order to license the nominative case. Also, there are suggestions that finiteness may be signaled by a close interaction of I and C (Kayne 1994; Vincent 1997). 82 and proposes three basic C positions; the highest C is for ‘subordination’, the middle C clause-typing, and the lower C Modality. This is represented in (46) (46) [C Subordinator .. [COP Clause-typing .. [CM Modal …]]] 36 In a similar vein, I postulate two basic C positions in Korean; the highest C is for a subordinator and the lowest C for a Mood head, 37 yet the resulting structure differs from the previous proposals above: finiteness, which licenses a nominative subject, involves both C heads in Korean, while the previous proposals appear to consider FinP or ModalP as the relevant projection carrying information about finiteness. 38 The highest C hosts a complementizer which simply connects the subordinate clause to some constituent of a higher clause or to a complementizer which carries a certain Modality that can determine the finiteness of a clause in the absence of a Mood ending. The 36 The full version of the C structure containing relevant morphemes, which Roussou (2000:79) proposes, is as follows: [C pu [Topic/Focus [COp oti/an/na/as [Neg δen/min [CM θa/tna/as [I cl + V]]]]]] Subordination Clause-typing Modal 37 Bhatt & Yoon (1992:49) also argue that the category known as ‘Comp’ should be decomposed into Mood and subordinators. 38 Roussou (2000) does not directly discuss what projection is responsible for finiteness in Greek with the reason that in languages like Greek the +/– finite distinction barely holds since there are no infinitives. However, she argues that -na clauses in Greek subsume the functions of the subjunctive and optative moods as well as of the infinitive and is always finite. 83 lowest C, MoodP, corresponds to Rizzi’s ForceP and Roussou’s clause-typing projection. 39 Under this analysis, the following tree structure is postulated. (47) CP Spec C’ MP Spec C M’ TP Mood This analysis is based on the empirical data that a sentence final Mood marker can occur in combination with a complementizer. (48) a. Na-nun Mina-eykey [Kiho-ka tomangka-ss-nya]-(ko) mul-ess-ta. I-TOP M.-DAT K.-NOM run.away-PST-DC-C ask-PST-DC ‘I asked Mina whether Kiho run away.’ b. …[CP [MP Kiho-ka [TP [VP [VP tomangka- V] v] -ss T] -nya M] -ko C]… As can be seen in (48a), the interrogative Mood suffix -nya is followed by the typical complementizer -ko. 40 This suggests that Mood suffixes and complementizers should be 39 The CP system can be more articulated for Topic and Focus. However, this thesis does not aim at providing a full analysis of clause structure and a theoretical analysis as to how finiteness is licensed in the structure. This thesis concerns only with the morphological and syntactic expressions of Mood and Modality and their syntactic relevance to finiteness. 40 The complementizer -ko can often be omitted. In this respect, -ko differs from the other Modal complementizers which obligatorily occur in subordinate clauses. However, 84 syntactically independent in Korean. (48b) shows the proposed structure for the embedded clause in (48a). In this chapter, I briefly reviewed the formal approaches to finiteness in which finiteness has been defined as the feature licensing nominative Case on subjects of clauses and Tense and Agreement features are widely accepted as the elements responsible for the assignment of nominative Case. This thesis basically holds the conventional notion that nominative Case on subjects is correlated with finiteness. However, unlike most formal approaches, in this thesis finiteness is assumed to be closely associated with Mood or Modality on a C structure. This is based on the fact that the cross-linguistic data in the literature reveal that the features determining finiteness may not be restricted to just Tense and Agr and finiteness may be associated with variety of factors, such as Mood/Modality and Aspect, which the functionalist approaches to finiteness commonly take into account as syntactic and semantic categories defining finiteness. In fact, this thesis pointed out the problems of Tense and Agr as licensing features of a nominative subject in Korean and argued that Mood on verbs and Modalities on complementizers determine finiteness of the clauses, which license nominative subjects. it is argued that they both occupy the highest C head since they both function as a subordinator and occur in the same slot of the verbal paradigms. 85 CHAPTER 3 FINITE CONTROL AND SENTENTIAL COMPLEMENTATION IN KOREAN The main purpose of this chapter is to examine whether finite control exists in Korean, through the investigation of complementation types and their relation to Obligatory Control (OC)/Non-obligatory Control (NOC). In this thesis, finite control refers to control in a finite context that licenses a nominative subject. There have been some studies arguing that Korean exhibits finite control, as discussed in chapter 1 (section 1.2.3). However, the previous studies fail to cover all sentential complements that may show Obligatory Control. Moreover, non-finite control which has been taken for granted in the previous studies shows similarities to finite control in terms of syntactic properties of finiteness. This thesis attempts to examine different complement constructions which may exhibit Obligatory Control. A careful investigation of those constructions may give a clear account for the syntactic and semantic properties of OC constructions in Korean, and may reveal whether the control effects are due to structural constraints such as argument identification and locality or due to semantic constraints in terms of the situation denoted by the matrix predicate or some other semantic factors. The organization of this chapter is as follows. In section 3.1, I will present a brief review of the classification of OC and discussion of the diagnostics for OC that have been previously employed in the literature. In section 3.2, I will show some evidence for finiteness of control complement clauses in Korean and discuss the existence of OC in a finite clause. In section 3.3, I will investigate syntactic and semantic properties of different types of sentential complements in some detail and examine whether there is a 86 correlation between the meaning of matrix predicates and the complement type in determining control. 3.1. Classification and Properties of Obligatory Control It has been widely accepted that control constructions can be classified into several types based on the degree to which the controllee depends on the controller for its reference. Some of types of control excerpted from Landau (2000:3) are shown below. (1) a. Johni tried [eci to leave early]. b. John wanted (for Mary) [ec to leave early]. Obligatory control (OC) Non-obligatory control (NOC) c. The chairi managed [eci to gather the committee at 6]. d. The chairi preferred [eci+ to gather at 6]. Exhaustive control (EC) Partial control (PC) e. Billi persuaded Maryj [eci+j to leave together]. Split control (SC) f. It is dangerous for babies [ecarb to smoke around them]. Arbitrary control g. It was difficult [ec to leave]. Implicit control Contrary to Obligatory control (OC) in (1a), in which the controller and the infinitive must be clausemates, Non-obligatory control (NOC) shows that the infinitive need not have a clausemate controller. Hence the empty category in (1b) can refer to John, or it can refer to Mary, if ‘for Mary’ is inserted in the sentence. The null subject in (1c) must be identical to the controller, while that in (1d) includes more referents than the matrix controller. These are called Exhaustive control (EC) and Partial control (PC), respectively. Split control (SC) in (1e) indicates that the null subject of the infinitive is 87 controlled jointly by two matrix arguments. Finally, compared to Arbitrary control in (1f), where no argumental controller occurs, Implicit control in (1g) indicates that the controller exists but is not syntactically expressed. Although most linguists agree that the characterization of OC vs. NOC is an important prerequisite for the analysis of complement control, the classification of OC vs. NOC is subject to some debate. Williams (1980) excludes partial and split control as well as implicit control from the notion of OC, based on his rather strict characterization of OC: i) Locality, ii) Uniqueness of antecedent, iii) C-command, and iv) Precedence, as shown in chapter 1 section 1.2.1.1. Hornstein (1999) proposes to capture the strict locality conditions on OC proposed by Williams via the Minimal Link Condition (MLC: Chomsky 1995). 41 In addition, he uses the sloppy vs. strict reading and de re vs. de se interpretation as criteria for the OC vs. NOC distinction, as will be discussed in detail below. According to Hornstein, OC PRO must carry a sloppy interpretation under ellipsis of the verb complement and de se interpretation. He considers only EC to be an OC since it can be accounted for under his movement analysis, while all non-exhaustive control readings cannot be so. However, this analysis has been challenged in recent proposals (Landau 1999, 2000; Fuji 2006; Madigan 2006, 2008), which argue that PC and SC are Obligatory Control. Landau (2000), in particular, analyzes all instances of local control as OC, thus including EC, PC, SC, and implicit control. 41 MLC is an updated version of the Minimal Distance Principle (MDP) proposed by Rosenbaum (1967). That is, the MLC excludes movement of an element α to a position K if there is another element β of the same type which is closer to K. 88 On the other hand, there have been some proposals 42 that offer different classifications of control, deviating from the standard classification of control above. Stiebels (2007) and Gamerschlag (2007) propose a Structural vs. Inherent control distinction. Structural control 43 is induced by the structure of a clausal complement, while ‘Inherent control’ occurs in the constructions where the matrix predicate requires Obligatory Control readings independent of the instantiated structure of complement clauses. These controls are classified as Control-inducing constructions, and the structures where no specific factor requires Obligatory Control reading are classified as control-neutral constructions. In this thesis, I will basically follow Landau’s classification of OC, but also adapt Stiebels’ (2007) classification of control in order to show that finite control in Korean is mostly triggered by semantics of predicates or other factors, such as complementizers. I will also employ the following criteria, which are widely accepted in the literature as the criteria for defining OC in non-finite clauses, in order to define the distribution of OC versus NOC in Korean. (2) The OC/NOC Categories a. Arbitrary control is impossible in OC, but possible in NOC. b. Long-distance control is impossible in OC, but possible in NOC. c. Strict-reading of PRO in gapping constructions is impossible in OC, but possible in NOC. 42 Also, Jackendoff & Culicover (2003) distinguishes between free, nearly free and unique control. In contrast to unique control, which is generally called OC in the literature, free and nearly free control allow generic/arbitrary and split control. Free control includes non-local control and control by discourse or speech act participants, which is not allowed in nearly free control. 43 This is similar to ‘Syntactic control’ in Cormack & Smith (2004). 89 d. De re reading of PRO is impossible in OC (only de se), but possible in NOC. Arbitrary control is the Case where no argument in the sentence is understood as the controller of the null subject. (3) a. Johni tried [PROi/*arb to study hard]. b. It is possible [PROarb to solve the problem]. As shown in (3b), the identity of the null subject in the construction is implicit and can be anyone, depending on the context. That is, the potential antecedent for the null subject is free in its reference. On the other hand, in OC structures, such as (3a), the identity of the null subject is the matrix argument and arbitrary control is disallowed. Similarly, the following examples illustrate that Long-distance (LD) control is disallowed in OC, but allowed in NOC. (4) a. Johni believes that Maryj promised him PRO*i/j to leave the town. b. Maryi knew that it damaged John [PROi to perjure herself]. In example (4a), the only possible controller for PRO is the subject argument of the clause immediately containing the infinitive, Mary, and the argument John, which is a long-distance argument, cannot be the controller for PRO. However, (4b) shows that the long-distance subject Mary can be a possible controller for PRO in Non-Obligatory Control constructions. 90 In addition, OC vs. NOC constructions show differences in the interpretation of the null subjects in verbal-gapping constructions, as follows. (5) John tried to study hard, and Mary did too. a. *Johni tried PROi to study hard, and Maryj tried PROi to study hard. (Strict reading) b. Johni tried PROi to study hard, and Maryj tried PROj to study hard. (Sloppy reading) (6) John thinks that feeding himself will be difficult, and Bill does too. a. Johni thinks that PROi feeding himself will be difficult, and Billj thinks that PROi feeding himself will be difficult. (Strict reading) b. Johni thinks that PROi feeding himself will be difficult, and Billj thinks that PROj feeding himself will be difficult. (Sloppy reading) As shown in the examples above, OC constructions exclude the strict reading where PRO is strictly identical to John (5a) and allow only the sloppy reading (5b), while NOC constructions allow both strict (6a) and sloppy reading (6b). The last well known diagnostic of OC vs. NOC structures is the de se/de re contrast in the structures. Consider the following examples in (7) and (8) which are originally provided by Horstein (1999:pp35-36). (7) “(There is) a war hero who suffers from amnesia and remembers nothing of his wartime experiences. Suppose this person (hereafter, “the unfortunate”) sees a TV 91 program describing his own exploits, and is impressed with the courage exhibited by that person, who he does not know is himself.” (8) a. The unfortunate expects that getting a medal would be boring. b. The unfortunate expects to get a medal. Given the scenario in (7), (8a) can be true, but (8b) cannot since (8b), unlike (8a), requires a belief of self. In other words, contrary to (8b) in which only de se reading is possible, the antecedent for the null subject of the gerund phrase in (8a) can be ambiguous in a way captured by the de se and de re reading. That is, (8a) can be either interpreted by the de se reading where the unfortunate is aware of himself as the person who is getting a medal, or by the de re reading where the unfortunate is not be aware of it and thus, expects that someone else will get a medal. In sections 3.2.2, I will examine whether finite control constructions in Korean display the same diagnostics for defining OC in non-finite clauses. 3.2. Finite Control as OC in Korean The existence of OC in finite clauses has been largely discussed in the literature on control in Balkan languages 44. However, finiteness of control complement clauses and the nature of the null subject in the clauses are still debatable. This section examines whether Obligatory Control is possible in a finite context in Korean and further shows that finite control in Korean is quite different from what has been documented in other languages. Although some evidence for finite control in Korean has been presented in literature on Korean control (e.g. Yang 1982, 1985; Borer 1989; 44 See Landau (2004) for a pan-Balkan overview of these constructions. 92 Madigan 2006, 2008), finiteness of control complement clauses in Korean has not been clearly accounted for. 3.2.1. Finiteness of Control Complement Clauses Consider the following examples in which embedded clauses are headed by the complementizer -ko. (9) a. Apecii-kkeyse Pataj-eykey [eci/*j hakkyo-ey ka-(si)-keyss-ta]-ko yaksokha- father-HON.NOM P.-DAT school-LOC go-HON-VOL-DC-C promise- si-ess-ta. HON-PST-DC ‘The father promised Pata that he would go to school.’ b. Minai-ka Pataj-ul [ec*i/j ttena-la]-ko M.-NOM P.-ACC seltukha-yess-ta. leave-IMP-C persuade-PST-DC ‘Mina persuaded Pata that she should leave.’ (10) Minai-ka Pataj-eykey [eci/*j/k hakkyo-ey M.-NOM P.-DAT ka-ss-ta]-ko malha-yess-ta. school-LOC go-PST-DC-C say-PST-DC ‘Minai told Pataj that eci/*j/k went to school.’ The complement clauses in (9) are analyzed as finite constructions since they bear a Mood marker, declarative -ta in (9a) and imperative -la in (9b). In addition, (9a) shows that -ko complement clauses can take an agreement marker -si- and a volitional Modal marker -keyss- which can denote future Tense. This shows that -ko complement clauses can be fully inflected for Tense and Agreement. In fact, the complement clause in (10) 93 shows that -ko complement clauses can be specified for past Tense. Therefore, -ko complement clauses should be treated as finite clauses. However, note that the examples in (9), contrary to the example in (10), which shows non-obligatory control, trigger the same coreference restrictions as typical non-finite control constructions. That is, the null subject in (9a) must be co-indexed with the matrix subject apeci ‘father’ and the null subject in (9b) must refer to the matrix object, Pata. Any other co-reference would be ungrammatical. Consequently, (9a) and (9b) trigger subject and object control, respectively. Also, it is worth noting that, like a typical non-finite control construction, the event time of the embedded clauses in (9) is unrealized or future with respect to that of the matrix due to the volitional Modal -keyss- in (9a) and the imperative Mood marker -la in (9b). The following examples further show that the complement clauses in OC constructions, such as (9) above, are finite as the null subject position can be filled with an overt DP in nominative Case. (11) a. Apecii-kkeyse Pataj-eykey [tangsini/*j/*k father-HON.NOM P.-DAT (si)-keyss-ta]-ko /*Wujin-i hakkyo-ey ka- he/himself. HON W-NOM school-LOC go- yaksokha-si-ess-ta. HON-VOL-DC-C promise-HON-PST-DC ‘The father promised Pata that he/*Wujin would go to school.’ b. Minai-ka Pataj-lul [kunye*i/j/*k/*Enju-ka ttena-la]-ko M.-NOM P.-ACC she seltukha-yess-ta. E.-NOM leave-HON-IMP-C persuade-PST-DC ‘Mina persuaded Pata that she/*Enju should leave.’ 94 c. Minai-ka Pataj-eykey [kunyei/*j/*k/cakii/*j/*Enju-ka M.-NOM P.-DAT she self ttena-keyss-ta]-ko E.-NOM leave-VOL-DC-C yaksokha-yess-ta. promise-PST-DC ‘Mina promised Pata that she/self/*Enju would leave.’ (12) Minai-ka Pataj-eykey [kunyei/*j/k/cakii/*j/Enju-ka hakkyo-ey M.-NOM P.-DAT she self ka-ss-ta]-ko E.-NOM school-LOC go-PST-DC-C malha-yess-ta. say-PST-DC ‘Mina told Pata that she/self/Enju went to school.’ The examples in (11) show that the null subject position can be filled by a phonetically realized DP, other than a referential lexical DP. That is to say, only a null subject, ec, or an overt pronoun or reflexive is compatible with the complement subject position in OC constructions. (11a), which corresponds to (9a), shows that the null subject can alternate with an honorific 3rd person pronoun/reflexive, tangsin, which refers to the matrix subject. (11b) shows that the null subject can alternate with a normal 3rd person pronoun, kunye, since the matrix subject is not a person who needs to be honored. Also, (11c) shows that a reflexive pronoun, caki or casin, can occur in the null subject position as well as a regular pronoun. Note that the overt DP pronoun in the embedded clauses has nominative Case. Also, it is important to note that the null subject and the overt DP in (11) display exactly the same reference possibilities. This is contrary to (12) which is considered to be a NOC construction. The example in (12), corresponding to (10), 95 shows that the overt pronoun can refer to entities other than the matrix argument and a referential lexical DP can occur in the subject position. The referential property of the overt pronoun in (11) is quite distinct from that of overt pronouns in subjunctive constructions in some languages, such as Hebrew and Balkan languages. In those languages, unlike the Korean data (11), an overt pronominal subject in subjunctive clauses triggers obviation. Consider the following Hebrew examples cited from Landau (2004:p.813). (13) a. Himlacti le-Gil1 še-ec1/*2 yearšem I-recommended to-Gil la-xug that-ec will-register.3SG.M to-the-department le-balšanut. to-linguistics ‘I recommended to Gil to register to the linguistics department.’ b. Himlacti le-Gil1 še-Dani2/hu2 yearšem I-recommended to-Gil la-xug that-Dani/he will-register.3SG.M to-the-department le-balšanut. to-linguistics ‘I recommended to Gil that Dani/he should register to the linguistics department.’ c. Himlacti I-recommended le-Gil1 še-ec2 terašem la-xug to-Gil that-ec will-register.2SG.M to-the-department le-balšanut. to-linguistics ‘I recommended to Gil that you should register to the linguistics department.’ 96 According to Landau (2004), (13a) displays OC into a finite complement- the embedded null subject must refer to the matrix object Gil, while (13b) with a lexical subject and (13c) do not. (13b) shows subject obviation – the pronoun in the embedded subject position is bound by an antecedent outside of the matrix clause – which is a typical subjunctive property in the language. (13c) shows that the null subject is not controlled since it is not 3rd person. Based on the empirical data above, Landau argues that finite control is found only in null 3rd person subjects and that the null subject in (13a) is PRO, while the null subject in (13c) is pro. However, this is not the case in Korean finite control. Unlike the Hebrew data, the Korean data in (11) show that the null subject can be replaced with an overt pronoun in nominative Case with the same coreference relation between the embedded subject and one of the arguments of the matrix clause. That is, the available interpretations are the same regardless of whether the overt pronoun or null subject appears in the complement clause. These examples cannot be accommodated in a scale of finiteness such as that proposed by Landau (2004), mainly because an overt DP subject is licensed in OC constructions, where Landau’s typology predicts that only PRO can be licensed. This implies that control can be attested in cases where no PRO can be licensed. Related arguments will be discussed in detail in chapter 4. Korean is not the only language that exhibits the availability of an overt DP in the null subject position of control constructions. It has been noted that OC constructions in some languages can have an overt DP in the complement subject position, as follows: 45 45 Madigan (2008) further reports that Zapotec (Boeckx, Hornstein and Nunes 2005) and Japanese (Kuno 1976, Iida 1991) allow overt material to be controlled. 97 (14) a. Ana je naterala Marijui Ana Aux forced ∅i doddje da Marija (ACC) COMP Serbo-Croatian come ‘Ana forced Marija that she should come.’ b. Ana je naterala Marijui Ana Aux forced da onai doddje Marija (ACC) COMP she come (Zec 1987) 46 ‘Ana forced Marija that she should come.’ (15) Anangasan tin Eleni na milisi afti i idhja Greek forced-3PL the Eleni-ACC SUBJ speak-3SG she herself-NOM ‘They forced Helen to speak herself.’ (16) (Victor) încearcă (Victor) să V try. 3SG V cînte (Phipippaki-Warburton & Catsimali 1999) (Victor) la trombon (Victor) Romanian SUBJ sing.3SG V ‘Victor is trying to play the trombone.’ at trombone V (Alboiu 2004) The examples in (14b) and (15) show that the complement clause of OC constructions can have an emphatic pronoun which bears nominative Case – the nominative Case is unmarked on the emphatic pronoun in Serbo-Croatian and marked in Greek. In fact, 46 According to Zec (1987), unlike the OC constructions (14b), ‘try’ constructions do not allow the overt subject to appear in the embedded clause, as shown below. Petari je pokušao da ∅i / *oni doddje Petar(Nom) Aux tried Comp he come ‘Peter tried to come.’ Compared to Serbo-Croation, Korean OC does not block the occurrence of overt pronouns in the controllee position. Even though in some cases, such as (18a) below, the pronoun makes the sentence somewhat unnatural, it is generally acceptable if it is stressed like an emphatic pronoun. 98 Philippaki-Warburton & Catsimali (1999) and Spyropoulos (2007) among others use nominative Case on the null subject as an argument for ‘pro’, which occurs in a finite context, and against the possibility of controlled PRO in this environment. Also, the Romanian example in (16) shows that the referential lexical subject can appear in a number of legitimate slots, including the embedded clause, although only one instantiation of the DP subject is permitted (Alboiu 2004:56). This example demonstrates that the missing arguments in OC constructions actually occupy a position that is independently theta-marked and Case-assigned. Given the fact that overt subjects in nominative Case are restricted to finite clauses, as discussed in the previous chapter, the fact that an overt subject in nominative Case can be licensed instead of the null subject suggests that the complement clauses we are dealing with are finite. In sum, it is argued that the embedded clauses of OC constructions in Korean and some other languages have finite properties. This suggests that the standard theory of Control restricting OC to only non-finite clauses and (PRO controlees) needs to be refined. Next, consider the following examples which are considered to be typical nonfinite control in the literature on Korean control. (17) a. Kim kyoswui-nim-i K. [eci ku tayhak-ulo olmki-e ka-(si)]-lyeko professor-HON-NOM the university-DR move-E go-HON-C nolyekha-si-ess-ta. endeavor-HON-PST-DC ‘Professor Kim endeavored to move into the university.’ 99 b. Apecii-kkeyse Pataj-eykey [eci/*j cenyek-ul Father-HON.NOM P.-DAT sa-(si)-(*ess/keyss)]- dinner-ACC buy-(HON)-(PST/VOL)- ki-lo yaksokha-si-ess-ta. NOML-DR promise-HON-PST-DC ‘Father promised Pata to buy dinner.’ c. Minai-ka sensayng-nimj-ul [ec*i/j ttena-si-(*ess/keyss)]-tolok seltukha-yess-ta. M.-NOM teacher-HON-ACC leave-HON-(PST/VOL)-C persuade-PST-DC ‘Mina persuaded the teacher to leave.’ The above examples trigger an Obligatory Control reading. That is, the missing subject in the embedded clauses must be co-referential with a local matrix antecedent. However, compared to the finite control constructions in (9), which can take both the honorific agreement suffix -si- and some Tense/Modal morphemes, the complement clauses in (17) can take only the honorific agreement suffix. Thus, the constructions in (17) are rendered ungrammatical if the embedded verb takes any Tense morphemes. This is the crucial reason why the complement clauses in (17) have been assumed to be typical examples of non-finite OC in the literature (Yang 1985; Borer 1989; Madigan 2006, 2008). 47 However, regarding the definition of finiteness discussed in the previous chapter, the complement clauses in (17) as well as (9) should be considered finite. That is, the complementizer -lyeko and -tolok in (17a, c) and the nominalizer ki in (17b), respectively, function as Modal complemetizers which license a nominative subject and render the complement clause to be finite. 47 Borer (1989) assumes that certain kinds of Korean infinitives, such as (18), are characterized as [-Tense, +Agr], and that the null subject in the non-finite is obligatorily controlled by the matrix argument. 100 The evidence for the finiteness of the complement clauses in (17) derives from the fact that an overt DP in nominative Case can appear in the null subject position of the complement clauses, as follows. (18) a. Minai-ka [kunye?i/*j/cakii /*Pata-ka ttena]-lyeko nolyekha-yess-ta. M.-NOM she self P.-NOM leave-C endeavor-PST-DC ‘Mina tried for her/self/*Pata to leave.’ b. Apecii-kkeyse Pataj-eykey [tangsini/*j father-HON.NOM P.-DAT si]-ki-lo /*Wujin-i cenyek-ul sa- he/himself.HON W.-NOM dinner-ACC buy yaksokha-si-ess-ta. HON-NOML-DR promise-HON-PST-DC ‘The father promised Pata that he/*Wujin would buy dinner.’ c. Minai-ka Pataj-lul [kunye*i/j/*k/*caki/*Enju-ka ttena]-tolok seltukha-yess-ta. M.-NOM P.-ACC she self E.-NOM leave-C persuade-PST-DC ‘Mina persuaded Pata that she/*self/*Enju leave.’ As shown in the above examples, an overt pronoun, but not a referential lexical NP, can appear in the embedded subject position, just as in the finite OC constructions in (11). But, the overt pronouns in (18) display exactly the same reference possibilities as when the embedded subject is null. 48 This indicates that the constructions in (9) and (17) should be treated as the same, that is, as finite control constructions. Otherwise, non- 48 The overt reflexive caki in (18c) cannot occur in the construction since the reflexive in Korean is subject oriented. 101 finite analyses of the constructions must explain the complete availability of lexical DPs in the contexts. In addition, it is worth noting that the matrix and the complement clauses in (17) can host conflicting temporal adverbs, as illustrated in (19). (19) a. Olhay cho Kim kyoswui-nim-i this.year beginning K. [eci naynyen-ey ku tayhak-ulo professor-HON-NOM next.year the university-DR olmki-e ka-si]-lyeko nolyekha-si-ess-ess-ta. move-E go-HON-C endeavor-HON-PST-PERF-DC ‘Early this year Professor Kim had endeavored to move into the university next year.’ b. Ecye, apecii-kkeyse Pataj-eykey [eci/*j nayil yesterday, father-HON.NOM P.-DAT si-ki-lo cenyek-ul sa- tomorrow dinner-ACC buy- yaksokha-si-ess-ta. HON-NOML-DR promise-HON-PST-DC ‘Yesterday, the teacher promised Pata to buy dinner tomorrow.’ b. Ecye, Minai-ka sensayng-nimj-ul [nayil ec*i/j ttena-si]-tolok seltukha- yesterday M.-NOM teacher-HON-ACC tomorrow leave-HON-C persuade- yess-ta. PST-DC ‘Yesterday, Mina persuaded the teacher that (s)he would leave tomorrow.’ 102 The example in (19a) shows that the future-oriented adverb naynyen ‘next year’ does not conflict with the past-oriented adverb olhay cho ‘early this year’ and with the past perfect Tense of the matrix verb. Similarly, the examples in (19b) and (19c) show that the matrix clauses can take the past-oriented adverb ‘yesterday’ and the embedded clauses the future-oriented adverb ‘tomorrow’. That is, the embedded clauses in (19) include their own Tense. This temporal property of the clause has been widely accepted as ‘semantic/dependent Tense’ which renders the clause ‘tensed’, as discussed in 2.2.1. Given this and the fact that the embedded clauses also can take the agreement marker si, the embedded clauses in (19) can be analyzed as being specified for [+T, +Agr] which have been considered to be the crucial features licensing a nominative subject in the literature. In this respect, the so-called non-finite constructions in Korean are quite different from non-finite control constructions which are analyzed as establishing a PRO subject in many other languages. Consider the following examples. (20) a. *Yesterday, Johni began to PROi solve the problem, tomorrow. b. Yesterday, John hoped to PRO solve the problem, tomorrow. (21) a.*Včera zabravix [PRO da zamina yesterday forgot-1SG utre]. Bulgarian DA leave-1SG tomorrow ‘*Yesterday, I forgot to leave tomorrow.’ b. Včera rešix [utre yesterday decided tomorrow pro da ne puša poveče]. DA not smoke-1SG anymore ‘Yesterday I decided that tomorrow I would give up smoking.’ (Krapova, 1999) 103 c. Mariai a încercat [PROi/*j/*Ion Mary has tried să plece]. Romanian PRO /*John PRT leave ‘Mary tried to leave.’ d. Ioni vrea ca Dan/proi/j să Ion wants that Dan/pro resolve problema. PRT solve the-problem ‘Ion1 wants Dan/him2 to solve the problem.’ (Landau 2005) The English examples in (20) (repeated from (28) in section 2.2.1) show that some infinitives can carry semantic Tense, while some cannot. According to Landau (2000), EC constructions, such as aspectual and implicative complements, cannot carry semantic Tense and thus are specified for [–Tense], while PC constructions, such as desiderative and interrogative complements, can carry semantic Tense and thus are specified for [+Tense]. However, Landau considers those constructions non-finite since they are not fully inflected for Tense and Agr, resulting in PRO subjects. Similarly, the Bulgarian examples in (21a) and (21b) excerpted from Krapova (1999) show a contrast in the availability of semantic Tense in the complement clauses. Krapova (1999) argues that these Tense distinctions determine the nature of the controlled subject. She proposes that while the subjunctive clause in (21a) is specified as [–T, +Agr] and checks Null Case, that in (21b) is specified as [+T, +Agr] and thus checks Nominative Case. Consequently, the former licenses a PRO subject and the latter pro. The same phenomenon is found in the Romanian examples in (21c) and (21d), which correspond to implicative and desiderative infinitives in English, respectively. According to Landau (2005), (21c) is specified as [–T, +Agr], which licenses PRO, while (21d) is specified as 104 [+T, +Agr], which licenses pro or a lexical subject. These empirical data suggest that the so-called non-finite constructions in Korean are more likely finite clauses in that they are fully specified for Tense, Agreement and Modality, and have the ability to license a nominative subject. In addition, note that the so-called non-finite OC constructions in Korean do not differ from finite OC constructions with respect to having different temporal adverbs, as illustrated in (22). (22) a. Ecye Minai-ka Pataj-eykey [eci/*j nayil yesterday M.-NOM P.-DAT hakkyo-ey ka-keyss-ta]- tomorrow school-LOC go-VOL-DC- ko yaksokha-yess-ta. C promise-PST-DC ‘Yesterday, Mina promised Pata that she would go to school tomorrow.’ b. Ecye Minai-ka sensayng-nimj-ul [ec*i/j nayil yesterday M.-NOM teacher-HON-ACC ttena-si-la]-ko tomorrow leave-HON-IMP-C seltukha-yess-ta. persuade-PST-DC ‘Yesterday, Mina persuaded the teacher that he should leave tomorrow.’ The unrealized or future-oriented event time of the embedded clauses in (22) is imposed by the volitional Modal -keyss- and the imperative Mood marker -la, and so the adverbial nayil ‘tomorrow’ can be licensed. This is the same as the event time of the 105 embedded clauses in (17) being determined by the Modal meaning of the complementizers -leyko, -ki and -tolok, respectively. Lastly, consider another type of complement clause in Korean that exhibits an OC reading - a type that has rarely been addressed in the literature on control in Korean. Here the embedded clause is marked by an adnominal suffix, -l, -nun or -(u)n. (23) a. Ecye Minai-ka Pataj- eykey [eci/*j nayil yesterday M.-NOM P.-DAT kes-ul cenyek-ul sa]-l tomorrow dinner-ACC buy-ADN yaksokha-yess-ta. thing-ACC promise-PST-DC ‘Yesterday, Minai promised Pata that shei would buy dinner tomorrow.’ b. Minai-ka Pataj- eykey [kunyei/*j/cakii/*j/*Enju-ka cenyek-ul M.-NOM P.-DAT kes-ul she self sa]-l E.-NOM dinner-ACC buy-ADN yaksokha-yess-ta. thing-ACC promise-PST-DC ‘Yesterday, Mina promised Pata that she/self/*Enju should buy dinner tomorrow.’ (24) a. Sensayng-nimi-kkeyse [eci ku ai-tul-ul teacher-HON-HON.NOM kaluchi-(si)]-nun kes-ul the child-PL-ACC teach-HON-ADN thing-ACC cungtanha-si-ess-ta. stop-HON-PST-DC ‘The teacher stopped teaching the children.’ 106 b. Sensayng-nimi-kkeyse [eci kongpwuha-(si)-te]-n teacher-HON-HON.NOM kes-ul cungtanha-si- study-HON-RET-ADN thing-ACC stop-HON- ess-ta. PST-DC ‘The teacher stopped studying.’ c. Sensayng-nimi-kkeyse [tangsini- /*Enju-ka ku ai-tul-ul teacher-HON-HON.NOM he/himself.HON E-NOM the child-PL-ACC kaluchi-(si)]-nun kes-ul cungtanha-si-ess-ta. teach-HON-ADN thing-ACC stop-HON-PST-DC ‘The teacher stopped teaching the children.’ Just like the other types of complement clauses discussed so far, all the null subjects in the clauses in (23) and (24) necessarily refer to the matrix subject, and no other reading is possible. However, the null subject can alternate with an overt pronoun that displays the same reference as the null subject; no other referential lexical NP is possible. This suggests that these types of complement clauses should be treated as finite OC constructions. As already discussed in 2.2.2, -l, -nun and -(u)n contrast with one another in terms of Modality or Tense and the selection of an adnominalizer is determined by the semantics of the matrix predicate. While the verb yaksokha- ‘promise’ in (23) selects the irrealis Modal adnominalizer -l, the verb cungtanha- ‘stop’ in (24) selects the realis Modal adnominalizer -nun or-(u)n . In particular, (24b) shows that the retrospective Modal suffix -te-, which denotes ‘past (progressive)’ Tense, can appear in the 107 embedded clause. This indicates that the event time of the embedded clauses in (24) is not “irrealis”. Therefore, unlike (23), in which the embedded clause can host a future Tense adverb, the embedded clauses in (24) exclude the future Tense adverb. In fact, the event time of the embedded clauses in (24) is non-distinct from the matrix event. This is expected, considering the semantic properties of the selecting predicates. For example, the event time of the embedded clause in (24a) is linked to that of the matrix clause through the realis/factive Modal adnominalizer -nun and is interpreted as a past action, although the embedded clause originally lacks a Tense marker. The examples in (24) are very interesting in that they do not display the typical Tense property of nonfinite OC constructions: the event time of the infinitival complement is “hypothetical or unrealized” (Bresnan 1972; Stowell 1982). However, this does not seem to be a peculiar phenomenon of Korean. Landau (2000:58) shows that factive and propositional complements typically involve a realis Tense preceding the matrix Tense, as in “Today, John claimed to have lost his car keys last week”. This might imply that the typical Tense property is not a necessary requirement for OC constructions, particularly for finite OC. As discussed thus far, control constructions in Korean whose complement clauses have been analyzed as finite or non-finite in the literature are in fact finite in that the constructions can allow an overt lexical NP in nominative Case to appear in the null subject position and can be also specified for Mood/Modality, Tense and Agreement. However, just as with OC in infinitives, finite control constructions in Korean exhibit an obligatory referential dependence between the null subject of the complement clause and some argument of the matrix clause. 108 3.2.2. OC Properties of Structures One might question the OC status of the finite control constructions in Korean discussed in the previous section. However, in addition to the referential dependence, the finite control constructions display the basic properties of OC which are employed as the criteria for defining OC in non-finite clauses (see (2) in 3.1). First of all, arbitrary control or control by a distant antecedent that is not in the immediately dominating clause (LD control) is not possible in the finite control structures, as shown in (25). (25) a. Johni-i [Minaj-ka Patak-eykey [ec*i/j/*k/*arb cenyek-ul J.-NOM M.-NOM P.-DAT sa-keyss-ta]-ko dinner- ACC buy-VOL-DC-C yaksokha-yess-ta]-ko malha-yess-ta. promise-PST-DC-C say-PST-DC ‘John said that Mina promised Pata that she would buy dinner.’ b. Johni-i [Minaj-ka Patak-lul [ec*i/*j/k/*arb ttena-]-tolok seltukha-yess-ta]-ko J.-NOM M.-NOM P.-ACC leave-C persuade-PST-DC-C mit-nun-ta. believe-PRS-DC ‘John believes that Mina persuaded Pata to leave.’ c. Johni-i [Minaj-ka [ec*i/j/*arb hakkyo-ey J.-NOM M.-NOM yess-ta]-ko cungtanha- school-LOC go-ADN thing-ACC stop- sayngkakha-n-ta. PST-DC-C think-PRS-DC ka]-nun kes-ul 109 ‘John thinks that Mina stopped going to school.’ In (25a), the null subject in the most deeply embedded clause refers to Mina, in the immediately higher clause, but cannot have arbitrary reference or take the LD controller, John, as its controller. The examples in (25b) and (25c) illustrate the same point. This contrasts with NOC constructions in Korean. (26) a. Johni-i [Minaj-ka Patak-eykey [eci/j/*k/arb hakkyo-ey J.-NOM M.-NOM P.-DAT ka-n-ta]-ko school-LOC go-PRS-DC-C malha-yess-ta]-ko mit-nun-ta. say-PST-DC-C believe-PRS-DC ‘John believes that Mina told Pata that she/he goes to school.’ b. Johni-i [Minaj-ka [eci/j/arb hakkyo-ey J.-NOM M.-NOM ka]-ki-lul pala-n-ta]-ko school- LOC go-NOML-ACC want-PRS-DC-C mit-nun-ta. believe-PRS-DC ‘Johni believes that Minaj wants himi/herj/someone else to go to school.’ c. Johni-i [Minaj-ka [eci/j/arb hakkyo-ey J.-NOM M.-NOM ka]-nun kes-ul pala-n-ta]- school-LOC go-ADN thing-ACC want-PRS-DC- ko mit-nun-ta. C believe-PRS-DC ‘John believes that Mina wants him/her/someone else to go to school.’ 110 As shown in (26), the null subject of the most deeply embedded clauses can refer to either Mary or the LD subject, John. Also, it can have an arbitrary interpretation, depending on the context where the sentence is spoken. Next, in gapping constructions, finite control constructions only admit a ‘sloppy reading’ interpretation of the controlee, as in (27), just as non-finite control constructions in English. (27) a. Minai-ka Pataj-eykey [eci/*j cenyek-ul M.-NOM P.-DAT sa-keyss-ta]-ko dinner-ACC buy-VOL-DC-C yaksokha-yess-ta. promise-PST-DC Wujink-to kuli-ha-yess-ta. W.-also so-do-PST-DC i) ‘Minai promised Pata that shei would buy dinner and Wujink promised Pata that hek would buy dinner.’ ii) *‘Minai promised Pata that shei would buy dinner and Wujin promised Pata that shei would buy dinner.’ b. Minai-ka Pataj-lul [ec*i/j ttena]-tolok seltukha-yess-ta. M.-NOM P.-ACC leave-C Wujink-to kuli-ha- persuade-PST-DC W.-also so-do- yess-ta. PST-DC i) ‘Minai persuaded Pataj that shej should leave and Wujink persuaded Pataj that shej should leave.’ ii) *‘Minai persuaded Pataj that shej should leave and Wujink persuaded Mina to persuade Pataj that shej should leave.’ 111 c. Minai-ka [eci hakkyo-ey M.- NOM to ka]-nun kes-ul cungtanha-yess-ta. Wujin- school-LOC go-ADN thing-ACC stop-PST-DC W.- kuli-ha-yess-ta. also so-do-PST-DC i) ‘Mina stopped going to school and Wujin stopped going to school.’ ii) *‘Mina stopped going to school and Wujin stopped Mina’s going to school.’ In the gapped sentences in (27), ‘Wjuin-to kuri-ha-yess-ta’, the empty subject cannot take the ‘strict reading’: that is, it cannot refer to the same antecedent as in the first clause, Mina. For example, in (27a) the only possible interpretation is that Wujin promises that he, Wujin, will buy dinner, the sloppy reading; it cannot mean the Wujin promises that Mina will buy dinner, the strict reading. The same holds for (27b) and (27c). On the other hand, NOC constructions in Korean can have both sloppy and strict reading, as illustrated in (28). (28) a. Minai-ka Pataj-eykey [eci hakkyo-ey M.-NOM P.-DAT ka-n-ta]-ko malha-yess-ta. school-LOC go-PRS-DC-C say-PST-DC Wujink-to kuli-ha-yess-ta. W.-also so-do-PST-DC i) ‘Minai told Pata that shei goes to school and Wujin told Pata that shei goes to school.’ ii) ‘Minai told Pata that shei goes to school and Wujink told Pata that hek goes to 112 school.’ b. Minai-ka [eci hakkyo-ey ka]-ki-lul M.-NOM para-yess-ta. Wujink-to kuli-ha- school-LOC go-NOML-ACC want-PST-DC W.-also so-do- yess-ta. PST-DC i) ‘Minai wanted to go to school and Wujin wanted Mina to go to school.’ ii) ‘Minai wanted to go to school and Wujin wanted to go to school.’ c. Minai-ka [eci hakkyo-ey M.-NOM ka]-nun kes-ul para-yess-ta. Wujink-to school-LOC go-ADN thing-ACC want-PST-DC W.-also kuli-ha-yess-ta. so-do-PST-DC i) ‘Mina wanted to go to school and Wujin wanted Mina to go to school.’ ii) ‘Mina wanted to go to school and Wujin wanted to go to school.’ Here, both interpretations are possible. Thus, in (28a) Wujin can be saying either that Mina is going to school, the strict reading, or that he, Wujin, is going to school, the sloppy reading. (28b) and (28c) behave similarly. Finally, finite control structures can only receive a de se interpretation. Consider the following scenario, (29), from Moulton (2005:5). (29) Fred is the president of a small nation. He is watching Fred (himself) deliver an address on television, but he is so drunk that he doesn’t recognize himself. He thinks that the guy he is seeing is a terrible president. (Nevertheless, Fred wouldn’t 113 knowingly think such things about himself.) (30) a. Fredi-nun [eci/*j te-isang nappun taythonglyeng-i toy-ci anh-keyss-ta]-ko F.-TOP anymore bad president-NOM become-NEG-VOL-DC-C yaksokha-yess-ta. promise-PST-DC ‘Fredi promised that hei/*j would not be a bad president.’ b. Fredi-nun [eci/*j te-isang nappun taythonglyeng-i toy-ci anh]-ki-lo F.-TOP anymore bad president-NOM become-NEG-NOML-DR yaksokha-yess-ta. promise-PST-DC ‘Fredi promised that hei/*j would not be a bad president.’ c. Fredi-nun [eci/*j te-isang nappun taythonglyeng-i toy-ci anh]-ul F.-TOP kes-ul anymore bad president-NOM become-NEG-ADN yaksokha-yess-ta. thing-ACC promise-PST-DC ‘Fredi promised that hei/*j would not be a bad president.’ The null subjects in (30) can only be given a de se reading. That is, each of the examples in (30) display the strong implicature of Fred’s awareness of himself as the terrible president, and thus the sentences are false since Fred does not have such a belief about himself in the scenario. However, in NOC constructions, de re reading as well as de se reading is possible. 114 (31) a. Fredi-nun [eci/j aphulo F.-TOP cohun taythonglyeng-i toy-li-la]-ko in.the.future good president-NOM become-CONJ-DC-C mit-ess-ta. believe-PST-DC ‘Fredi believed that hei/j would become a good president.’ b. Fredi-nun [eci/j aphulo F.-TOP cohun taythonglyeng-i toy]-ki-lul in.the.future good president-NOM become-NOML-ACC pala-yess-ta. want-PST-DC ‘Fredi wanted that hei/j would become a good president.’ c. Fredi-nun [eci/j aphulo F.-TOP ul cohun taythonglyeng-i toy]-l in the future good kes- president-NOM become-ADN thing- kitayha-yess-ta. ACC expect-PST-DC ‘Fredi expected that hei/j would become a good president.’ As shown in (31), the null subject of the embedded clause can refer to an antecedent which is outside of the construction. As a result, the constructions are not OC and de re reading is possible. That is, the examples in (31) cannot be falsified by de re reading since the examples can be interpreted in a way that Fred is not aware of his true identity as the president. In sum, these data above clearly display all the typical properties of Obligatory Control in non-finite control constructions, even though the embedded clauses are 115 demonstrably finite. That is, finite control constructions in Korean can instantiate OC. Consequently, this analysis suggests that finiteness is actually hardly relevant to OC constructions in Korean. This is similar to the analysis of finite control in Hebrew and Balkan languages that has been proposed by Landau (2000, 2004). However, unlike Landau’s analysis, in which the null subjects of finite OC constructions as well as of non-finite are assumed to be PRO, the null subjects of finite OC constructions in Korean are best analyzed as pro. This will be motivated in chapter 5 in detail. Given the existence of finite control, as discussed so far, the crucial distinction between OC and NOC in Korean may depend on two factors: 1) whether a matrix argument obligatorily binds the complement subject argument and 2) whether Rexpressions (which are intrinsically referentially indepenedent) can replace a controllee. While no antecedent is necessary and a lexical DP including R-expressions can alternate with a controllee for NOC, the controllees in OC constructions must have an antecedent and they cannot be replaced by R-expressions. 3.3. Types of Sentential Complements and Finite Control A few studies (cf. Yang 1985, Madigan 2008) have argued for finite control in Korean. However, none of the studies addressed the complete range of finite control constructions, focusing on only quotative complement clauses as finite control constructions. Moreover, they differentiated the so-called non-finite control constructions from other finite control constructions, missing the fact that both structures display the properties of finiteness in that an overt subject in nominative Case can appear in the complement clause and the complement clause can be specified for Tense, Agreement and Modality, which are widely accepted as factors determining 116 finiteness in the literature. As syntactic factors appear to be unable to account for these control facts, perhaps the answer lies in semantic factors. In this section, I will examine whether there is a correlation between the meaning of matrix predicates and the complement type in determining control. There have been very few studies on this issue (cf. Gamerschlag, 2007), although it is a very crucial issue to investigate in order to account for the characteristics of control. Six basic types of complements that are examined in this thesis are 1) Nominalizations, 2) Adnominalizations, 3) Interrogative complements, 4) Quotative complements, 5) Intentive Complements, and 6) Resultative complements. These complements are headed by different types of elements, as illustrated in (32). (32) Elements Selecting the Complement Clauses. Complement types Elements heading complements Nominaliation Nominalizer -ki and -(u)m Adnominalization Adnominalizer -(u)n, -nun, and -(u)l Interrogative complements Quotative complements Intentive complements Resultative complements Interrogative endings -ka/kka and -na/nya Adnominalizer -nun/(u)l + the defective noun ci Quotative complementizer -ko Intentive complementizer -(u)lyeko and -koca Resultative complementizer -tolok and -key Each complement type will be analyzed with respect to the lexical properties of the selecting predicate and their control relation. Ninety matrix predicates have been 117 compiled to form the basis of this thesis and classified into 9 different categories, 49 based on their semantic class, as shown below. (33) A List of the Predicates Implicative Aspectual Directive/ Manipulative Desiderative Utterance Propositional Perception Factive Interrogative sidohata ‘try’, noleykhata/ayssuta ‘endeavor’, 50 silphayhata ‘fail’, kyeulihata/soholhata, ‘neglect’, samkata ‘refrain’, phihata ‘avoid’ sichakhata ‘begin/start’, keysokhata ‘continue’, kkutnayta ‘finish’, memchwuta/cwungtanhata/cwungcihata ‘stop/cease’, kumantwuta ‘quit/stop’ kangyouhata ‘force’, seltukhata ‘persuade’, mantulta ‘make’, sikhita ‘let’, cwumwunhata ‘order’, hyeppakhata/wihyephata ‘threaten’, myengleynghata/cihwihata ‘command’, pangcihata/makta ‘prevent/ stop’, heyonghata/helakhata ‘allow’, ceyanhata ‘propose/suggest’, coluta ‘coax’, pwutakhata/tangpwuhata ‘ask’, yochenghata ‘request’, kanchenghata ‘beg’ wenhata/parata ‘want’, hyimanghata ‘hope’, sowonhata ‘wish’, senhohata ‘prefer’, kalmanghata/kalkwuhata ‘desire’, tonguyhata ‘agree’, yaksokhata ‘promise’, kyeyhoykhata ‘plan’, cwunpihata ‘prepare’, uytohata ‘intend’, keylcenghata ‘decide’, kepwuhata ‘refuse/decline’, kitayhata ‘expect’, keylsimhata ‘determine/resolve’, yokwuhata ‘ask/demand’, senthaykhata/thaykhata/koluta/chwihata ‘choose’ malhata/yaikihata ‘say’, pokohata ‘report’, potohata ‘report (in a mass media)’, palphyohata ‘announce’ chwucenghata/chwuchukhata ‘assume’, cwucanghata ‘claim’, mitta ‘believe’, pwuinhata/pwucenghata ‘deny’, kacenghata ‘suppose’, sayngkakhata ‘think’, uysimhata ‘doubt’, yesanghata/yechukhata ‘anticipate’ pota ‘see’, tutta ‘hear’, kwukeynghata ‘watch’, nukkita ‘feel’ hwuhoyhata ‘regret’, miwehata/silhehata ‘hate’, cohahata ‘like’, hyemohata ‘loath’, icta/mangkakhata ‘forget’, kiekhata ‘remember’, alta ‘know’ mwutta ‘ask’, palkyenhata/chacta ‘find out’, phaakhata ‘grasp’, ihayhata ‘understand’, kwungkumhata ‘wonder’ 49 There are no uniform verb classes in the literature. The categories employed in this thesis are based on Landau (2000:38) and Stiebels (2007:79). 50 noleykhata/ayssuta ‘endeavor’ can be often replaced with sidohata ‘try’. 118 3.3.1. Nominalization and Control As discussed in 2.2.2, there are two sentential nominalizers in Korean: -ki and -um which are basically in complementary distribution. The nominalizer -ki tends to appear in the complements of most of the aspectual, desiderative, directive/manipulative, and implicative predicates in (33), while it tends not to appear in the complement of the remaining types of the predicates. On the other hand, -(u)m is found in the complements of factive, propositional, and utterance predicates, while it is not found in the complements of the predicates where -ki can appear. The following data illustrate some examples of the contrast in the distributions of both nominalizers. (34) a. Minai-ka [eci/*j khephi-lul M.-NOM masi]-ki/*m sicakha-yess-ta. coffee-ACC drink-NOML start-PST-DC ‘Mina started drinking coffee.’ b. Minai-ka naj-eykey [ec*i/j/*k kecis-ul malha]-ki/*m-(l)ul kangyoha-yess-ta. M-.NOM I-DAT lie-ACC tell-NOML-ACC force-PST-DC ‘Mina forced me to tell a lie.’ (35) a. Minai-ka [eci/*j ke chayk-ul M.-NOM ilk-ci anh-ass]-m/*ki-(l)ul hwuhoyha-yess-ta. the book-ACC read-NEG-NOML-ACC regret-PST-DC ‘Mina regretted reading the book.’ b. Mina-kai [eci/j cha-ey M.-NOM kilum-ul chaywu-ess]-m/*ki-(l)ul ic-ess-ta. car-DAT gas-ACC fill-PST-NOML-ACC ‘Mina forget that she/someone else filled the car with gas.’ forget-PST-DC 119 The examples above show that the aspectual predicate sicakha- ‘start’ (34a) and the directive predicate kangyoha- ‘force’ (34b) can select only -ki complements, while the factive predicate hwuhoyha- ‘regret’ (35a) and icta ‘forget’ (35b) can select only -(u)m complements. This illustrates that -um occurs in cases when the event has already occurred, while -ki is used when the event has not occurred. Note that unlike the -(u)m complements in (35), the -ki complements in (34) show OC reading. However, the following examples show that the choice of a nominalized complement might not be the factor which triggers the OC reading. (36) and (37) exemplify the fact that the same matrix predicate can sometimes select -ki complements and at others select -(u)m complements. Wujin-ul manna]-ki-lo 51 (36) a. Minai-ka [eci/*j tabang-eyse M.-NOM yaksokha-yess-ta. coffee.shop-LOC W.-ACC meet-NOML-DR promise-PST-DC ‘Mina promised that she would meet Wujin in a coffee shop.’ b. Minai-ka Minswuj-eykey [eci/i+j ttena]-ki-lo M.-NOM M.-DAT yaksokha-yess-ta. leave-NOML-DR promise-PST-DC ‘Mina promised Minswu to leave/that they (Mina and Minswu) would leave.’ c. Wulii-nun [eci/*j comte We-TOP khun cip-ulo a little more big isaka]-ki-lo home-LOC move-NOML-DR kyelcengha-yess-ta. decide-PRS-DC ‘We decided to move into a little bigger house.’ 51 Some researchers consider kilo one morpheme since the directional marker -lo cannot be dropped, contrary to the accusative case which can be dropped, depending on the matrix predicates. 120 (37) a. Wujini-i [eci/*j hoycang-ulose-uy yekhwal-ul chungsilhi ha-l kes-i]W.-NOM m-ul president-as-GEN role-ACC faithfully do-FUT-COP- yaksokha-yess-ta. NOML-CC promise-PST-DC ‘Wujin promised that he would play a president’s role faithfully.’ b. Cwungkwuki-un [cwungkwuk hyekmyeng-i Chinese-TOP naaka-ya ha]-m-ul Chinese twu kaci kil-lo revolution-NOM two kind way-DR kyelcengha-yess-ta. go.forward-should-NOML-ACC decide-PST-DC ‘The Chinese decided that the Chinese revolution should go forward in two ways.’ In (36a, b) the predicate yaksokha- ‘promise’ selects -ki complements, but in (37a) it selects an -(u)m complement. Likewise, kyelcengha- ‘decide’ selects a -ki complement in (36c), but an -(u)m complement in (37b). As discussed in 2.2.2, the generalization about the complementary distribution of the nominalizers is not uncontroversial since there are empirical examples, such as (36) and (37), showing an exception to the generalization. However, even in the case of the exception, some predicates show a preference for one over the other type of nominalized complement. In the above examples, the matrix predicates are much more compatible with -ki complements than (u)m complements. This is attributed to the contrast in the Modality of both nominalizers. Recall that the event time of control complements is generally irrealis. Since -ki, contrary to -(u)m, denotes irrealis Modality, -ki complements are more 121 compatible with the matrix predicates than -(u)m complements. A more important thing to note, however, is that both types of nominalized complements do not always exhibit OC. In the literature on Korean and other languages, the verb yaksokha- ‘promise’ has been treated as a subject control verb. Indeed, data involving the predicate in the Korean corpora predominantly show subject control. However, the example in (36b) with the -ki nominalizer shows non obligatory control. This suggests that the OC in (34) is induced by factors other than the -ki nominalizer. As a matter of fact, a prominent factor that perhaps triggers OC in (34a) is the lexical meaning of the aspectual predicate, sicakha- ‘start’. That is, the person who does the act of starting in (34a) must simultaneously do the act of drinking coffee, by the lexical semantics of sicakha- ‘start’. Thus, obligatory subject control is triggered. However, the directive verb kangyoha‘force’, a non-aspectual verb, in (34b) does not always trigger OC, as will be discussed in section 3.3.6. The following examples further show that the choice of a -ki nominalized complement does not necessarily trigger OC. (38) a. Minai-ka [eci/j cip-ey M.-NOM ka-ki]-lul para-n-ta. home-LOC go-NOML-ACC want-PRS-DC ‘Mina wants (someone else) to go home.’ b. Kim kyoswui-ka K. [eci/j ku chayk-ul professor-NOM kitayha-yess-ta. expect-PST-DC kaycengha-l swu iss-ki]-lul the book-ACC revise-can exist-NOML-ACC 122 ‘Professor Kim expected that he could revise the book.’ c. Ku-nun [anay-ka he-TOP cip-ey tolao-ki]-lul para/kitayha- wife-NOM home-LOC come.back-NOML-ACC want/expect- n-ta. PRS-DC ‘He wants/expects his wife to come back home.’ In the examples in (38), the null subjects are not obligatorily coreferential with the matrix subject, Mina, but can refer to someone else outside of the sentences. The admissibility of the overt embedded subject instead of the null subject in (38c) shows again that a -ki nominalized complement does not always trigger OC. In conclusion, the data show that the choice of a nominalized complement does not ensure OC, while the lexical meaning of some matrix predicates, such as the aspectual verb sicakha- ‘start’, can trigger OC. Following Stiebels (2007) and Gamerschlag (2007), I will call the former case ‘control neutral’ and the latter case ‘Inherent control’. ‘Inherent control’ occurs in the constructions where some semantic factors, such as the lexical semantics of matrix predicates, require Obligatory Control readings independent of the instantiated structure of complement clauses. ‘Controlneutral’ applies to the structures where no specific factor requires an Obligatory Control reading and thus, a lexical NP referring to other than the matrix arguments can be overtly realized. Therefore, nominalized complementation in Korean is analyzed as a control-neutral structure. 123 3.3.2. Adnominalization and Control As discussed in 2.2.2, adnominal complementizers, -(u)n, -nun, and -(u)l, contrast with one another in terms of Modality or Tense. The choice of the adnominal endings in Korean basically depends on the Tense of the embedded sentence and the type of the embedded predicate. 52 There are two kinds of complementation formed by the adnominal complementizers in Korean: 53 direct and indirect. 54 In the former case, the adnominal endings, -(u)n, -nun, and -(u)l, replace the final endings of the embedded predicates, while in the latter case, they are simply attached to the embedded predicates without deleting the final ending of the predicates. This is illustrated in the following examples. (39) a. Kui-nun [eci ku kos-ey he-TOP ka]-l/nun kes-ul kepwuha-yess-ta. the place-LOC go-ADN thing-ACC refuse-PST-DC ‘He refused to go to the place.’ b. Emenii-nun atulj-eykey [ec*i/j cip-ul mother-TOP son-DAT ttena]-l/nun kes-ul home-ACC leave-ADN thing-ACC 52 The realis/present adnominalizer for a verb is -nun, while that for an adjective is -n. Also, adjectives show some restrictions on the selection of the adnominalizers, which will not be of concern in this thesis. 53 The complements headed by the adnominal endings can be divided into two types, relativization and adnominalization, depending on whether or not the head noun is an element of the complement clauses. The head noun in the latter case, unlike in the former case, cannot be conceptually postulated within the complement clause (as will be shown below) and it is coreferential to the whole complement clause. Only the latter type is considered to be relevant to the issues of this thesis. 54 Indirect complementation is called “indirect’ because it incorporates indirect quotes (Lee & Ramsey, 2000:204). 124 helakha-si-ess-ta. allow-HON-PST-DC ‘The mother allowed her son to leave home.’ c. Sonyei-nun [eci chayk-ul girl-TOP ilk-te]-n/*nun/*l kes-ul book-ACC read-RET-ADN memchwu-ess-ta. thing-ACC stop-PST-DC ‘The girl stopped reading a book.’ (39) shows examples of direct complementation. As shown in the examples, direct complementation can involve all the adnominal endings. The embedded clauses with the adnominal endings modify the dummy noun kes which is the head noun of the clauses. 55 In (39a) and (39b) the adnominalizer -(u)l or -nun can follow the verb stem ka- ‘go’ and ttena- ‘leave’, respectively. However, in (39c) only the adnominalizer -n can appear because the verb stem is followed by the retrospective Modal suffix -te, which denotes ‘past (progressive)’ Tense. Unlike direct complementation, indirect complementation can involve only the adnominalizer -nun, as can be seen in (40). (40) a. Kunyei-nun [eci naj-eykey pimil-ul she-TOP 55 I-DAT malha-yess-ta]-nun kes-ul secret-ACC tell-PST-DC-ADN thing-ACC In addition to numerous free nouns, such as sasil ‘fact’ and il ‘matter’, there are many defective or dummy noun, such as kes ‘thing’, ci ‘whether’, and swu ‘possibility’, which can function as the head noun of the adnominal complements. Since what kind of the head noun is selected is not a central issue to the discussion of this thesis, I will give only the examples of kes complements for the adnominal complementation. 125 pwuinha-yess-ta. deny-PST-DC ‘She denied that she told me a secret.’ b. John-un [kunye-ka J.-TOP she-NOM alumtap-ta]-nun kes-ey tonguyha-n-ta. beautiful-DC-ADN thing-LOC agree-PRS-DC ‘John agrees on a fact that she is beautiful.’ In the above examples, the adnominalizer -nun is attached to the embedded predicates without deleting the sentence-final ending of the predicates. The distribution of -(u)n complements is restricted to a few types of predicates, such as factive, perception, utterance and propositional predicates. This complementation generally shows no control. However, when it involves the retrospective Modal -te- directly followed by the adnominal ending -(u)n and is selected by an aspectual predicate, such as in (39c), it shows Obligatory Control. Compared to -(u)n complements, -nun and -(u)l complements are compatible with almost every type of predicate in (33) and in some cases either one can occur, as shown in (39a) and (39b). 56 However, they do show some contrast in their distribution: -nun complements, unlike -(u)l complements, are incompatible with many of directive predicates and -(u)l complements, unlike -nun complements, are incompatible with aspectual and many implicative predicates. The following are some examples of the distribution. 56 Also, in many cases, both adnominal complements can be replaced with -ki nominalization. 126 (41) a. Nai-nun kisaj-eykey [ec*i/j cha-lul I-TOP driver-DAT seywu]-l/*nun kes-ul car-ACC stop-ADN yokwuha- thing-ACC ask- yess-ta. PST-DC ‘I asked the driver to stop the car.’ b. Chani-i [eci kose-lul C.-NOM mou]-nun/*l kes-ul sicakha-yess-ta. old.book-ACC collect-ADN thing-ACC start-PST-DC ‘Chan started to collect old books.’ c. Ku-tuli-un [eci Tongi-lul cwuki]-nun/*l kes-un he-PL-TOP T.-ACC kill-ADN silphayha-yess-ta. thing-TOP fail-PST-DC ‘They failed to kill Tongi.’ As can bee seen in (41a), most directive predicates, such as yokwuha- ‘ask’, select -(u)l complements, but do not select -nun complements. In contrast, aspectual predicates and many implicative predicates, such as sicakha- ‘start’ and silphayha- ‘fail’, do not select -(u)l complements, but select -nun complements, as in (41b) and (41c). These restrictions seem to be imposed by the lexical semantics of the matrix predicates. For example, the directive predicate yokwuha- ‘ask’ in (41a) denotes that the event or action denoted by the embedded verb is going to be carried out by the object referent kisa ‘driver’. Thus, the adnominalizer -l, which denotes an irrealis or prospective Modality, can be selected, but not the adnominalizer -nun, which denotes a realis or indicative Modality. 127 Just looking at the examples in (39) through (41), it seems that ‘direct complementation’, unlike ‘indirect complementation’, triggers control. However, the following examples show that the former, as well as the latter, is actually controlneutral. (42) a. Kui-nun [nwui-ka he-TOP cip-ey o]-nun kes-cocha kepwuha-yess-ta. sister-NOM home-DAT come-ADN thing-even refuse-PST-DC ‘He even refused to let his sister come home.’ b. Yang sacangi-un [eci mikwuk-ulo ttena]-l Y.president-TOP USA.-DR kes-ul kyelcengha-yess-ta. leave-ADN thing-ACC decide-PST-DC ‘President Yang decided to leave for USA.’ c. Yueyni-un [1948 nyen-ulo yengkwuk-uy thongchi-ka kkuthna]-l kesUN-TOP ul 1948 year-DR England-GEN reign-NOM end-ADN thing- kyelcengha-yess-ta. ACC decide-PST-DC ‘The UN decided that the reign of England would end in 1948.’ The examples in (39a) and (42a) employ the same matrix predicate, kepwuha- ‘refuse’ and the same adnominal ending -nun. However, unlike (39a) in which subject control occurs, (42a) displays no control at all. The examples in (42b) and (42c) show the same pattern. That is, (42b) shows subject control, while (42c) shows no control. In addition, the most of the adnominal complements selected by propositional, utterance, and perception predicates show no control, as some examples are shown in (43). 128 (43) a. Kunye-nun [oppa -ka she-TOP cacin-ul towa cwu]-l kes-ul older.brother-NOM self-ACC help give-ADN thing-ACC mit-ess-ta. believe-PST-DC ‘She believed that her old brother would help her.’ b. Ceca-nun [wuli-ka pyenhwaha]-l kes-ul author-TOP we-NOM change-ADN cwucangha-n-ta. thing-ACC claim-PRS-DC ‘The author claims that we should change ourselves.’ c. Enlon-un [taycwung-i ku pepan-ul chansengha]-nun/l kes- mass.media-TOP the masses-NOM the bill-ACC consent-ADN thing- ulo potoha-yess-ta. DR report-PST-DC ‘The mass media reported that the masses consent/will consent to the bill.’ d. Ku ai-ka [Kiho-ka tomangka]-nun kes-ul po-ass-ta. he child-NOM K.-NOM run.away-ADN thing-ACC see-PST-DC ‘The child saw Kiho running away.’ The examples in (43a) and (43b) show that the -l adnominal complements selected by the propositional predicates mit- ‘believe’ and cwucangha- ‘claim’ show no control at all. Similarly, the examples in (43c) and (43d) show that the adnominal complements selected by the utterance predicate potoha- ‘report’ and the perception predicate po‘see’ display no control. These tell us that adnominalization, like nominalization, is 129 control-neutral. However, just like nominal complements selected by aspectual predicates, when adnominal complements are selected by aspectual predicates, they show obligatory subject control, as shown in (39c) and (41b). That is, as already discussed in the previous section, aspectual predicates always trigger Inherent control. The following are further examples showing that adnominal complement clauses selected by some implicative predicates require Obligatory Control reading. (44) a. Cwumongi-un [eci/*j kot C.-TOP talana]-l/*nun/*n kes-ul soon run.away-ADN kyelsimha-yess-ta. thing-ACC resolve-PST-DC ‘Cwumong resolved to run away soon.’ b. Taychii-nun [eci/*j Tongcin-eykey teisang T.-TOP T.-DAT malul ke]-nun/*l/*n kes-ul anymore speak.to-ADN thing-ACC samka-ss-ta. refrain-PST-DC ‘Taychi refrained talking to Tohngcin anymore.’ c. Kutuli-un [eci/*j ku-lul they-TOP kes-un sipcaka-ey maytal-ass-ciman cwuki]-nun/*l/*n he-ACC cross-LOC hang-PST-although kill-ADN silphayha-yess-ta. thing-TOP fail-PST-DC ‘They fail to kill him, although they hung him on the cross.’ d. Kui-nun [eci/*j saylowun calyo-lul he-TOP new swucipha]-nun/*l/*n kes-ul data-ACC collect-ADN thing-ACC 130 keyuliha-yess-ta. neglect-PST-DC ‘He neglected to collect new data.’ As shown in the examples above, implicative predicates show some restrictions on their choice of complements. That is, the matrix predicate kyelsimha- ‘resolve’ can select -l complements, but cannot select -nun and -n adnominal complements. On the other hand, the matrix predicate samka- ‘refrain’, silphayha- ‘fail’, and keyuliha- ‘neglect’ can select -nun complements, but cannot select the other types of adnominal complements. However, the more important thing to note here is that they always trigger subject control. That is, the null subjects in (44) cannot refer to a referent different from the matrix subject, and thus an overt lexical NP other than the matrix subject cannot occur in the subject position of the complement clauses. That is, these predicates, like aspectual predicates, always trigger subject control regardless of the types of the complement that they select. This will be discussed in detail in chapter 5 (section 5.2.1.1). 3.3.3. Interrogative Complements and Control Interrogative complements are formed in two ways: one is by attaching the interrogative endings, -ka/kka and -na/nya, to the verb and the other is by an adnominalization headed by the defective noun, ci. The general characteristic of these complements is that they occur with a question word, such as nwuka ‘who’ or woy ‘why’. Interrogative complements are compatible with mostly propositional, perception, 131 utterance, factive, and interrogative predicates and with some of the desiderative predicates, such as keylcengha- ‘decide’ and senthaykha- ‘choose’. Observe the following examples which contain the interrogative ending -ka/kka. (45) a. Nai-nun [eci/*j ku ttye way cip-ul I-TOP ttena-ss-te-n-ka]-(lul) 57 the time why house-ACC leave-PST-RET-IND-Q-ACC hwuhoyha-yess-ta. regret-PST-DC ‘I regretted why I left home at that time.’ b. Nai-nun [cikum-kkaci-uy na-uy hayngtohn-i I-TOP elmana kyengmangha-yess- now-until-GEN I-GEN behavior-NOM how te-n-ka]-(lul) light-PST- hwuhoyha-yess-ta. RET-IND-Q-ACC regret-PST-DC ‘I regretted how imprudent my behavior was.’ c. Kui-nun [Tongi-ka eti-ey he-TOP sal-ass-te-n-ka]-(lul) ic-ess-ta. T.-NOM where-LOC live-PST-RET-IND-Q-ACC forget-PST-DC ‘He forgot where Tongi was living.’ d. Sengsehakca-tuli-i [etten kes-i wenmwun-ey kakkawu-l- Biblicists-PL-NOM what.kind.of thing-ACC original.text-LOC close-FUT- 57 Based on the admissibility of the case markers in (45) and (46) below, there has been a claim that the embedded question morphemes –ka/kka and –na/nya should be analyzed as sentential nominalizers (cf. Kim, 1984). However, considering the distributions of the question morphemes as verbal endings in root clauses, they will be treated as interrogative Mood endings, not nominalizers, in this thesis. Whether they are nominalizers or not is not central to the discussion above. 132 kka]-lul kyelcengha-yess-ta. Q-ACC decide-PST-DC ‘The Biblicists decided which one would be close to the original text.’ As can be seen in (45), some factive predicates, such as hwuhoyha- ‘regret’ and ic‘forget’, and some desiderative predicates, such as kyelcengha- ‘decide’, select interrogative complements headded by -ka/kka. The interrogative complements can take Tense/Aspect or Modal suffixes imposed by the selecting predicates. That is, in (45a) and (45b) the complement clauses can take the retrospective modal suffix -te-, which denotes a past (progressive) event, due to the lexical meaning of the matrix predicate hwuhoyha ‘regret’. In contrast, the complement clause in (45d) selected by the matrix predicate kyelcengha- ‘decide’ can take the future Tense suffix -l . Note that unlike (45a) which shows subject control, the examples in (45b-d) show no control at all. The same pattern can be found in the interrogative complements headed by the interrogative ending -nya/na, as follows. (46) a. Kui-nun [eci/*j ku il-ul he-TOP kyesokha-l ke-nya ma-l ke-nya-](lul) the work-ACC continue-FUT-Q don’t-FUT-Q-ACC kyelcengha-yess-ta. decide-PST-DC ‘He decided whether or not he would continue the work.’ b. Kui-nun Minkyuj-eykey [ec*i/j/k ettehkey ku he-TOP M.-DAT how ton-ul mo-ass-na]- the money-ACC save-PST-Q 133 mul-ess-ta. ask-PST-DC ‘He asked Minkyu how he saved the money.’ Similar to the -ka/kka interrogative complements in (45), the -nya/na interrogative complements in (46) can take Tense or aspectual specification imposed by the selecting predicates. Also, the examples in (46) demonstrate that -nya/na interrogative complements cannot trigger Obligatory Control; (46a) shows subject control, but (46b) shows Non-Obligatory control. The other type of interrogative complement shows the same control phenomenon. (47) a. Swunii-nun [casini-/ku-ka mwusun il-ul S.-TOP self/he-NOM what wenha]-nun/l ci-(lul) job-ACC want-ADN whether-ACC al-ass-ta. know-PST-DC ‘Swuni knew what job she/he wanted/ what job she/he would want.’ b. Na-nun [anay-ka I-TOP eti-lo ka-ss]-nun /ka]-l ci-(ka) wife-NOM where-DR go-PST-ADN/go-ADN whether-NOM kwungkumha-yess-ta. wonder-PST-DC ‘I wondered where my wife went/would go.’ c. Halapecii-kkeyse [nwuka hoysa-lul kyengyengha]-*nun/l grandfather-HON.NOM who.NOM company-ACC manage-ADN 134 ci-lul kyelcengha-si-ess-ta. whether-ACC decide-HON-PST-DC ‘The grandfather decided who would manage the company.’ The examples in (47) include adnominalized complements headed by the defective head noun ci ‘whether’. 58 Similar to the adnominal complements discussed in the previous section, different types of adnominalizers can be selected depending on the lexical meaning of the matrix predicates. That is, the factive verb al- ‘know’ in (47a) and the interrogative verb kwungkumha- ‘wonder’ in (47b) can select either -nun or -l, while the predicate kyelcengha- ‘decide’ can select only -l. More importantly, however, the examples show that ci interrogative complements usually show no control. In (47a) the reflexive pronoun casin refers to the matrix subject Swuni, but a referent other than the matrix subject, ku ‘he’, can appear. Likewise, the examples in (47b) and (47c) show no control. In conclusion, the examples above show that interrogative complements are control-neutral, just as the other complement clauses discussed so far and that no specific predicates selecting interrogative complements show Inherent control. 3.3.4. Quotative Complements and Control A quotative complement is a verbal complement construction that is followed by the complementizer -ko. In general, the complementizer -ko is called a quotative particle in the literature (cf. Sohn 1995), which marks reported speech or someone’s thoughts. 58 Considering the fact that the examples in (47) involve an adnominalization headed by the defective noun ci, these examples can be classified as adnominalized complements. However, since they occur with a question word and thus form an interrogative sentence they are treasted as interrogative complements. 135 The predicates in -ko complements can bear the full range of verbal affixes found with the matrix predicates, so some matrix predicates subcategorizing for a -ko complement can permit Tense and Modal suffixes, as well as Mood suffixes, to be attached to the embedded verb, as shown below. (48) a. Minai-ka Pataj-eykey [eci/*j/k hakkyo-ey ka-ss-ta]-(ko) malha-yess-ta. M.-NOM P.-DAT school-LOC go-PST-DC-C say-PST-DC ‘Mina told Pata that she/someone else went to school.’ b. Nai-nun [K-ka I-TOP ku-uy uykeyn-ey pantayha-li-la]-(ko) sayngkakha-n-ta. K-NOM he-GEN opinion-LOC object-FUT-DC-C think-PRS-DC ‘I think that K will object to his opinion.’ c. Miswuni-i [eci/j ku il-ul M.-NOM ha-nun kes-i etteha-nya]-(ko) mul- the job-ACC do-ADN thing-NOM how-Q-C ask- ess-ta. PST-DC ‘Miswun asked how it would be if she/someone else does the job.’ As shown in the examples in (48), the quotative particle -ko, which can be omitted, follows the Mood markers, such as declarative -ta (48a, b) and interrogative -nya (48c). The predicats of the -ko complements can take a Tense marker, such as past -ss- and future -li-, as shown in (48a) and (48b). A -ko complement can be selected by many of the predicate types in (33), although it is not compatible with aspectual predicates or some of the implicative and factive predicates, such as samka- ‘refrain’ and al- ‘know’. 136 The above examples also demonstrate that -ko complements do not generally induce an OC reading. However, when a -ko complement takes some specific Mood or Modal suffixes, Obligatory Control can arise. (49) a. Minai-ka Pataj-eykey [eci/*j cip-ey M.-NOM P.-DAT ka-keyss-ta]-ko malha-yess-ta. home-LOC go-VOL-DC-C tell-PST-DC ‘Mina told Pata that she will go home.’ b. Minai-ka Pataj-eykey [ec*i/j cip-ey M.-NOM P.-DAT ka-la]-ko home-LOC go-IMP-C malha-yess-ta. tell-PST-DC ‘Mina told Pata to go home.’ c. Minai-ka Pataj-eykey [eci+j cip-ey M.-NOM P.-DAT ka-ca]-ko malha-yess-ta. home-LOC go-PROP-C tell-PST-DC ‘Mina said to Pata that they (Mina and Pata) should go home.’ Contrary to (48), the examples in (49) show Obligatory Control. Note that the -ko complements in (49) and (48a) are selected by the same matrix predicate malha- ‘tell’, which is generally considered to be an NOC predicate. But the control relations are different in each, as shown by the contrast in the coindexation of the null subject. Note that the examples in (49) differ solely in terms of Mood and Modal markers. Clearly the different types of Obligatory Control in (49) are triggered by the different types of Mood and Modal markers. That is, while (49a) involves subject control induced by the volitional marker -keyss-, (49b) and (49c) involve object and split control induced by 137 the imperative Mood marker -la and the proposative Mood marker -ca, respectively. This shows that -ko complementation does not trigger Obligatory Control, but some Mood or Modal markers in the -ko complements can determine control relations, regardless of the lexical meaning of the matrix predicates. Further evidence is found in the following examples. (50) a. Ku kanswui-ka coyswu-tulj-eykey [ec*i/j kkwuleanc]-ki-lul the jailer-NOM prisoners-PL-DAT sit.on.one’s knees-NOML-ACC kangyoha-yess-ta. force-PST-DC ‘The jailer forced the prisoners to sit on their knees.’ b. Emenii-nun Sungilj-eykey-man [ec*i/j kongpwuha]-l kes-ul mother-TOP S.-DAT-only study-ADN kangyoha-- thing-ACC force- si-ess-ta. HON-PST-DC ‘The mother forced only Sungil to study.’ (51) a. Ku namcai-ka naj-eykey ton-ul he man-NOM I-DAT ul se-keyss-ta]-ko yokwuha-mye [eci/*j sacin money-ACC ask-and modeyl- picture model- kangyoha-yess-ta. ACC stand-VOL-DC-C force-PST-DC ‘The man forced me that he would be a model of my picture, asking me for some money.’ 138 b. Minai-ka Pataj-eykey [ec*i/j John-ul manna-la]-ko kangyoha-yess-ta. M.-NOM P.-DAT J.-ACC meet-IMP-C force-PST-DC ‘Mina forced Pata to meet John.’ c. Minai-ka Pataj-eykey [eci+j swul-ul M.-NOM P.-DAT masi-ca]-ko kangyoha-yess-ta. liquor-ACC drink-PROP-C force-PST-DC ‘Mina forced Pata to have a drink together.’ The predicate kangyoha- ‘force’ is generally considered to be an object control predicate, and this is indeed the case in (50), in which the null subjects of the -ki nominalized complement and the adnominalized complement refer to only the matrix object. However, the examples in (51) present a different set of facts. Here we find subject control (51a), object control (51b), and split control (51c), and the only difference is again the Modality/Mood markers. This indicates that the Modal and the Mood suffixes can override the ability of the matrix predicate to induce object control and thus, creat different types of OC readings. That is, the Modal suffix -keyssobligatorily triggers subject control (51a), whereas the Mood suffix -la and -ca always trigger object (51b) and split control (51c), respectively, regardless of the type of the matrix predicate. The control effect of Mood and Modal markers of -ko complement clauses will be further discussed in section 5.2.2. 3.3.5. Intentive Complements and Control Intentive complements are clauses with Modal complementizers denoting the meaning of ‘intention’ or ‘purpose’ – specifically the complementizers -(u)yeko and 139 -koca. In the literature, -koca complements have not been reported as participating control constructions. However, they pattern similarly to -(u)lyeko complements, which have been associated with control as can be seen in the following examples. (52) a. Minai-ka [eci/*j ku mwuncey-lul M.-NOM haykeylha]-lyeko noleykha-yess-ta. the problem-ACC solve-C endeavor-PST-DC ‘Mina endeavored to solve the problem.’ b. Kui-nun [eci/*j tasinun kulen silphay-lul ha-ci anh]-ulyeko kyelsimha-yess-ta. he-TOP again such failure-ACC do-NEG-C resolve-PST-DC ‘He resolved not to do such a failure again.’ c. Cei-nun [eci/*j ku yenghwa-lul tanswunha-ciman caymi iss-key mantul]I.POL-TOP the movie-ACC simple-but fun exist-C make- lyeko uytoha-yess-ta. C intend-PST-DC ‘I intended to make a simple, but funny movie.’ -(U)lyeko 59 complements show very limited distributions: they are compatible with only some implicative predicates, such as noleykha- ‘endeavor’ in (52a), and some desiderative predicates, such as kyelsimha- ‘resolve’ in (52b) and uytoha- ‘intend’ in (52c). The intentive complements cannot take any Tense or aspectual marking due to the inherent meaning of the complementizers. That is, the complementizers constrain 59 Some researchers divide -leyko into -lye- which is an intentive Mood marker, and -ko which is itself considered a quotative particle since in many cases -lye- can be used without -ko. However, -ko of -leyko differs from the quotative particle -ko in that the former cannot function as a quotative particle in these constructions. Therefore, in this thesis, -lyeko is differentiated from -ko complements. 140 the occurrence of Tense and Aspect in the embedded clauses since they encode future Tense information as an inherent semantic feature. Recall that the complementizer -leyko expresses the speaker’s intention or purpose, which is related to an intentional or desiderative Modality, as discussed in 2.2.2. As shown in the above examples, the complements headed by the complementizer -lyeko always trigger obligatory subject control regardless of the type of matrix predicate. The same pattern can be found in -koca complements. (53) a. Kui-nun [eci/*j sinsilhan kitokkyoin-i he-TOP faithful toy]-koca noleykha-yess-ta. Christian-COP become-C endeavor-PST-DC ‘He endeavored to be a faithful Christian.’ b. Nai-nun [eci/*j hyensil-lopwuthe melli talana]-koca kyelsimha-yess-ta. I-TOP reality-from far run.away-C resolve-PST-DC ‘I resolved to run away from reality.’ c. Kim kyoswui-ka [eci/*j silhem K. professor-NOM kyelkwa-lul cocakha]-koca uytoha-yess-ta. experiment result-ACC manipulate-C intend-PST-DC ‘Professor Kim intended to manipulate the laboratory results.’ d. Pwumo-tuli-i [eci/*j elin parents-PL-NOM canye-tul-eykey yenge-lul kaluchi]-koca young children-PL-DAT English-ACC teach-C wenha-yess-ta. want-PST-DC ‘Parents wanted to teach English to their children.’ 141 The -koca complements in (53a)-(53c) show the same distributions as the -lyeko complements in (52). However, -koca complements show compatibility with more desiderative predicates than -(u)lyeko, such as wenha- ‘want’ in (53d). As can be seen in the above examples, -koca complements, like -lyeko complements, always trigger obligatory subject control, regardless of the type of matrix predicate. Note that no referent other than the matrix subject can be allowed in the null subject position in (53d), although wenha- ‘want’ is generally considered an NOC predicate. This indicates that the intentive complementizer is the key factor which triggers Obligatory Control in this construction. In this respect, the intentive complements are different from other types of complements discussed so far. However, it is more important to note that the key factor determining Obligatory Control in the intentive complements is the semanctic property of the complementizers rather than syntactic property of the complement. Since the complementizers -(u)lyeko and -koca posses the meaning of ‘intention’ or ‘purpose’, the null subject must be controlled by an individual capable of having an intention or purpose, and thus inducing subject control. Therefore, these complementizers are treated as a factor triggering Inherent control. 3.3.6. Resultative Complements and Control Resultative complements, which indicate the result or effect of a preceding cause, are formed by attaching a resultative or causative complementizer, such as -tolok or -key, to a verbal stem. With the exception of the subject honorific -si, no other affix can precede the complementizers. Consequently, neither Tense nor Mood markers can appear in the resultative complements. -Tolok complements are mostly restricted to the 142 directive/manipulative predicates, although they can be selected by some implicative, desiderative, and utterance predicates. (54) a. Minai-ka sensayng-nimj-ul [ec*i/j/*k ttena-si]-tolok seltukha-yess-ta. M.-NOM teacher-HON-ACC leave-HON-C persuade-PST-DC ‘Mina persuaded the teacher to leave’ b. Sinpwui-nun kutulj-eykey [ec*i/j/*k casin-ul priest-TOP they-DAT ttalu]-tolok kangyoha-yess-ta. self-ACC follow-C force-PST-DC ‘The priest forced them to follow him.’ c. Kimkwui-ka naj-lul/eykey [ec*i/j/*k ku cali-eyse K.-NOM I-ACC/DAT ttena]-tolok hyeppakha- the position-LOC leave-C threaten- yess-ta. PST-DC ‘Kimkwu threatened me that I would have to leave the postion.’ (55) a. Minai-ka ecj [sensayng-nimj-kkeyse M.-NOM ttena-si]-tolok seltukha-yess-ta. teacher-HON-HON.NOM leave-HON-C persuade-PST-DC ‘Mina persuaded the teacher to leave’ b. Sinpwui-nun ecj [ku-tulj-i priest-TOP casin-ul ttalu]-tolok kangyoha-yess-ta. he-PL-NOM self-ACC follow-C force-PST-DC ‘The priest forced them to follow him.’ c. Kimkwui-ka ecj [nayj-ka ku cali-eyse K.-NOM ttalu]-tolok hyeppakha-yess-ta. I-NOM the position-LOC leave-C threaten-PST-DC ‘Kimkwu threatened me that I would have to leave the postion.’ 143 (54) and (55) show the examples of -tolok complements selected by the directive/manipulative predicates which are often treated as object control verbs. As shown in (54), the null subject of the -tolok complements is coreferential with the matrix object. That is, they seem to trigger object control. Note that the matrix object in the constructions can bear either accusative or dative Case, depending on the lexical meaning of the matrix predicate. In some cases, both Cases are equally possible with no particular contrast in meaning, as in (54c). The data in (55) instantiate what has been referred to as backward control in the literature (e.g. Monahan 2003). Here the embedded subject controls a null object argument in the matrix clause. 60 While the -tolok complements in (54) show obligatory object control, -tolok complements selected by predicates other than directive/manipulative predicates, such as nolekykha- ‘endeavor’ and yaksokha- ‘promise’, do not necessarily give rise to object control, as illustrated in (56). (56) a. Cehuy hoysai-nun [eci/*j te we-POL.TOP company-TOP 60 cohun aphathu-lul cis]-tolok more good apartment-ACC build-C Within the generative framework, backward control on Korean has been analyzed in several ways: the movement based analysis (Monahan, 2003), and the semantic based analysis (Cormack & Smith, 2004; Choe, 2006). The previous studies on backward control show that although backward control might be possible within current theoretical assumptions, it is not clear what mechanism constrains the distribution of backward control and what the nature of the empty category in control constructions is. In this thesis, a detailed discusson of backward control will be left aside for the future research. However, considering the properties of finite forward control, as will be more discussed in the next chapters, it is assumed that the overt DP is best analyzed as being co-indexed with a pro argument in both forward and backward control constructions and that the relationship between a controller and a controllee in the constructions is best accounted for in the frame of semantics rather than syntax. 144 nolyekha-keyss-sup-ni-ta. endeavor-VOL-AH-IND-DC ‘Our company will endeavor to build better apartments.’ b. Nayi-ka [kulen il-i eps]-tolok yaksokha-li-ta. I-NOM such matter-NOM not.exist-C promise-VOL-DC ‘I will promise that such a thing won’t happen.’ c. Nai-nun [Yuna-eykey mescin I-TOP Y.-DAT salam-i nathana]-tolok sowenha-yess-ta. splendid man-NOM show.up-C wish-PST-DC ‘I wished that a splendid man will show up for Yuna.’ (56a) shows subject control. This seems to be due to the semantic property of the matrix predicate, nolyekha- ‘endeavor’, which often triggers subject control. (56b) and (56c) show no control at all. Consequently, I propose that -tolok complements are controlneutral and Obligatory Control in this type of complements is basically triggered by the matrix predicates. Next, consider -key complements. When selected by directive/manipulative and some implicative predicates, -tolok can often be replaced with a complementizer -key, although in some cases -key complements might sound less natural than -tolok complements for some Korean native speakers. In the following examples, either -tolok or -key can occur. (57) a. Sacangi-un cikwen-tulj-eykey/ul [ec*i/j/*k os-ul president-TOP employee-PL-DAT/ACC cayuloi ip-ul dress-ACC freely put.on-ADN 145 swu iss]-key/tolok heyongha-yess-ta. way exist-C allow-PST-DC ‘The president allowed his employees to dress freely.’ b. Kyocangi-un kyosa-tulj-ul [ec*i/j/*k haycik principle-TOP teacher-PL-ACC kyosa hwuwenkum-ul dismissal teacher support.money-ACC nay ci mos ha]-key/tolok mak-ass-ta. make NEG do-C stop-PST-DC ‘The principle stopped the teachers from making support money for dismissing teachers.’ The above examples illustrate that -key constructions generally pattern together with tolok constructions. The examples in (57) are the typical examples of the -key complements which show obligatory object control. That is, the null subject of the -key complements selected by the directive matrix predicates heyongha- ‘allow’ (57a) and mak- ‘stop’ (57b) must refer to the matrix object. However, the following examples show that -key complements, just as -tolok complements, do not always trigger obligatory object control. (58) a. Apecii-nun atulj-eykey [casini/j-i father-TOP son-DAT te khu-ko kangha-key nukki]-key/tolok self-NOM more big-and strong-ADV feel-C mantul-ess-ta. make-PST-DC ‘The fatheri made his son feel bigger and stronger/The fatheri made his son feel 146 that hei is bigger and strongger.’ b. Cey-ka [yelepwun-tul-i tolaka-l swu iss]-key/tolok I.POL-NOM you.HON-PL-NOM go.back-ADN way exist-C noleykha-keyss-sup-ni-ta. endeavor-VOL-AH-IND-DC ‘I will endeavor to make it happen that you can go back.’ (58) is an example of -key complements selected by the directive verb mantul- ‘make’. In this example, the reflexive pronoun of the complement clause can refer to either the matrix subject apeci ‘father’ or the matrix object atul ‘son’. In addition, (58b) which is an example of -key complements selected by the implicative verb noleykha- ‘endeavor’ shows no control at all. This indicates that -key complements, just as -tolok complements, do not always trigger obligatory object control. Thus, these complementizers are treasted as a control-neutral factor. With respect to resultative complements, there are some other types of -key constructions to note. First, consider the periphrastic causative constructions formed by the light verb ha- ‘do’ with a -key complement clause. (59) a. Ciswui-ka Kangsanj-ul/eykey [ec*i/j/*k hakkyo-lul C.-NOM K.-ACC/DAT yess-ta. PST-DC ‘Ciswu caused Kangsan to quit school.’ kumantu]-key/tolok ha- school-ACC quit-C do- 147 b. Ciswu-ka [Kangsan-i hakkyo-lul C.-NOM K.-NOM kumantu]-key/tolok ha-yess-ta. school-ACC quit-C do-PST-DC ‘Ciswu caused Kangsan to quit school.’ Similar to the resultative complements discussed so far, the periphrastic causative constructions show object control (59a) and backward control (59b). Also, the complementizer -key in (59) can be replaced with -tolok with a subtle difference in meaning: -tolok constructions produce weaker causal reading than -key constructions. Pointing out the correlation between causative -key and -tolok constructions above, Madigan (2008) argues that -tolok complements might be better classed with causatives. However, -tolok complements as well as -key complements generally designate the resulting state of the matrix event or action, which might lead to the causative reading of complements. In fact, some of the -tolok and -key complements, such as (56) and (58b), lack the causative meaning, but express the result of the matrix event or action. Moreover, even if the resultative constructions can be analyzed as causative constructions, in principle, it does not exclude the possibility that the resultative constructions should be analyzed as control constructions. The following represent another instance of -key constructions that has been analyzed as control. (60) a. Sunii-ka ppallayj-ul [ecj hayah]-key ppal-ass-ta. S.-NOM laundry-ACC white-C ‘Suni washed the laundry clean.’ wash-PST-DC 148 b. Sunii-ka emenij-lul [ecj S.- NOM mother-ACC alumtap]-key sayngkakha-yess-ta. beautiful-C think-PST-DC ‘Suni thought her mother beautiful.’ (61) a. *?Suni-ka [ppalay-ka hayah]-key ppal-ass-ta. S.-NOM laundry-NOM white-C wash-PST-DC ‘Suni washed the laundry clean.’ b. *Suni-ka [emeni-ka alumtap]-key sayngkakha-yess-ta. S.-NOM mother-NOM beautiful-C think-PST-DC ‘Suni thought her mother beautiful.’ Wechsler (1997) divides resultative constructions into control resultatives and ECM resultatives. For him, control resultatives are subject to semantic restrictions imposed by the verb, while ECM resultatives lack this type of restriction. That is, the null subject in control resultatives is a semantic argument of the matrix verb, while the null subject in ECM resultatives is not a semantic argument of the matrix verb. According to this classification, unlike (60b), which can be analyzed as an ECM resultative, (60a) can be analyzed as a control resultative since ppallay ‘laundry’, which is the semantic subject of the predicate hayah- ‘white’, is a semantic argument of the matrix verb ppal ‘wash’. On the other hand, Kang (2001) argues that transitive resultatives such as (60a) are control, while complement small clauses such as (60b) are ECM/raising. However, the examples in (60) are not best analyzed as control constructions. Note that the embedded subject in the nominative Case renders the sentence ungrammatical (61), unlike other resultative complements, which allow an overt nominative subject in the control 149 complements, resulting in backward control. Moreover, the -key phrase in (60a) is actually not a verbal complement but an adjunct phrase. 61 The omission of the -key phrase in (60a) does not affect the grammaticality of the sentence. The sentence without the -key phrase is a complete sentence, while the omission of the resultative complement in control constructions causes the sentence to be ungrammatical. Unlike (60a), the -key phrase in (60b) functions as a complement. However, it is not a control construction, but a small clause. The -key construction in (60b) is different from the resultative constructions that can create Obligatory Control in that the -key complement does not denote a resulting state of the matrix event or action. As discussed in the previous sections, the event time of the embedded clause in control constructions is usually unrealized. However, the -key complements in (61b) is not related to an unrealized future event, and the embedded subject is simply predicated to be in the state of possessing ‘beauty’. To sum up, it is observed that all complement types, except intentive complements, are basically control-neutral and thus, do not trigger Obligatory Control independently. However, certain combinations of a matrix predicate and complement type, such as tolok complements selected by directive predicates, can create Obligatory Control. Moreover, in the case of -ko complements, Modal or Mood suffixes play a key role in determining control. This suggests that Obligatory Control in Korean is determined by various semantic factors rather than by purely syntactic factors. This will be discussed in detail in chapter 5. 61 In fact, the suffix -key can be analyzed as an adverbializer (Sohn 1999:396). (i) Kam-i [acwu pulk-key] ik-ess-ta. Persimmon-NOM very red-ADV ripe-PST-DC ‘The persimmons are red ripe.’ 150 In this chapter, I provided some evidence for finiteness of control complement clauses in Korean and discussed the existence of OC in finite clauses. Unlike the previous studies arguing that Korean exhibits both finite and non-finite control, I proposed that there is no non-finite control in Korean, providing empirical data exhibiting that the so-called non-finite control constructions show similarities to finite control constructions in terms of syntactic properties of finiteness. The far-ranging investigation of complementation types with respect to their relation to OC vs. NOC revealed that there are some semantic factors, such as the lexical meaning of aspectual predicates and intentive complementizers, which always trigger Inherent control, and that there is an interaction between the meaning of matrix predicates and the complement type in determining control. In addition, it has been discussed that certain Modal or Mood suffixes in -ko complements can trigger Obligatory Control. 151 CHAPTER 4 SYNTACTIC PROPERTIES OF CONTROL COMPLEMENTS IN KOREAN The aim of the present chapter is to identify the nature of null subjects in control complements in Korean. The question of what syntactic structure OC complements occurs in is a central issue with respect to whether the syntactic nature of the controlled subject is a PRO, pro or a trace. It has long been argued that there is a close correlation between non-finiteness and the nature of PRO. However, it has also been noticed that OC is attested in not only non-finite clauses but also in other syntactic structures, such as subjunctives and finite embedded clauses. There are basically three approaches to the structure of control complements in the literature: 1) the CP analysis, 2) the IP/TP analysis, and 3) the VP analysis. Standard approaches to control, such as Chomsky (1995), Pesetsky (1992) and Landau (2000, 2004), 62 have claimed that control infinitives are CP, containing PRO subjects. On the other hand, some researchers, such as Bošković (1997), propose that control infinitives have properties typical of IPs/TPs rather than CPs. Evidence for this is generally based on the absence of a complementizer and a Tense marker in control infinitives. Under the IP/TP analysis of control structures, there have been two different proposals with respect to the nature of the null subject in the control complements. One is a PRO analysis of the null subject position and the other is a DP trace analysis under movement. Furthermore, some researchers, such as Bresnan (1982), Chierchia (1984), and Wurmbrand (2001, 2002), challenge the PRO analysis of Obligatory Control, arguing that at least some instances 62 Their CP analyses, though, are different in terms of Tense properties of some control infinitives. For example, Landau (2000:74) assumes that implicative complements are semantically untensed, whereas Pesetsky (1992) assumes that the complements are tensed CP’s containing a phonetically (but not semantically) null tense morpheme. 152 of control involve structure sharing. In their proposals, the syntactic structure of OC infinitives is a VP, lacking a Tense projection and resulting in subjectless (i.e., PROless) infinitives. In this chapter, I will show that in Korean, the control complements in finite OC constructions are full clauses in terms of having complementizers and of the availability of overt subjects, and thus are analyzed as CPs taking pro subjects rather than PRO. On the other hand, it will be argued that some serial verb constructions in Korean, which are claimed to involve non-finite control, consist simply of VPs. However, the serial verb constructions are not grouped with OC constructions, but simply involve argument sharing. 4.1. The Structure of Finite Control Complements and pro in Korean Standard approaches to control have claimed that control infinitives are CP with a PRO subject. However, the CP lacks an overt complementizer and has a [-finite] TP, as shown in (1a). This is based on empirical data in which the presence of a complementizer leads the sentence to be ungrammatical, as shown in (1b). (1) a. Pati wants [CP Ø [IP PROi to [VP leave early]]]]. b. *Pat wants [CP for [IP PROi to [VP leave early]]]]. c. Bill thinks [CP that [IP Jane/*PRO is alone]. In the framework of GB, the ungrammaticality of (1b) is explained by the PRO theorem: PRO must be ungoverned. In other words, PRO is assured by the absence of a governor, such as a complementizer. Therefore, PRO is excluded from occurring in finite clauses, 153 as shown in (1c). This in turn suggests that PRO is in complementary distribution with overt DPs which must be Case-marked. 63 However, within the Minimalist Program, the notion of government has been dispensed with and Case-theoretic accounts of PRO have been proposed. For instance, Bošković (1997) claims that control complements without any complementizer are IPs, arguing that [+Tense] non-finite clauses assign Null Case to PRO subjects. Bošković provides some evidence that control complements behave differently from finite clauses, as shown below. (2) a. *[C Ø He would buy a car] was believed at that time. b. *Mary believed Peter finished school and Bill [C Ø Peter got a job]. (3) a. [C Ø PRO to buy a car] was desirable at that time. b. Mary tried to finish and Peter [C Ø PRO to get a job]. The ungrammaticality in (2) is related to the ECP (Empty Category Principle). That is, the examples in (2) are ungrammatical since the empty complementizers are not properly governed by verbs. Contrary to (2), the examples in (3) are grammatical. Therefore, Bošković takes this evidence for the claim that control complements cannot be a CP and thus, do not contain an empty complementizer. 63 Schütze (1997:31-35) claims that PRO and lexical subjects are not in complementary distribution, based on some empirical data, such as below: a. John/PRO leaving early would be rude. b. PRO/him not picking up the kids on time, John had upset Mary. c. Nobody/PRO move! You/PRO put your hands up! Schütze (1997:273) claims that the presence of an event binder licenses subjects and PRO is permitted where Tense contrasts cannot be marked. 154 The movement analyses of control (Hornstein 1999; Manzini & Roussou 2000; Alboiu 2004) also take control complements to be an IP. However, the analyses reduce PRO to a trace of a moved argument. In the movement analyses, the embedded subject must move to a matrix argument position to be Case-licensed. This motivates the movement analyses to crucially depend on a non-CP status of control complements since the CP structure of control complements under the movement analyses would cause a theoretical problem. For instance, in Hornstein’s analysis, a derivation begins with a DP merging with a verb, thereby checking the verb’s theta-role and then the DP moves into the embedded [Spec, IP] to check the D-feature of the IP. The DP further moves into the matrix [Spec, IP] to check its Case and the D-feature of the IP via the the matrix [Spec, VP] in order to check the theta-role of the matrix verb. Thus, if control complements take a CP projection, A-movement from an embedded clause to the matrix clause necessarily passes through the intermediate Spec CP of the embedded clause. This violates the Phase-Impenetrability Condition (Chomsky 2000) stating that movment of a DP out of a phase (i.e., vP and CP) is not permitted unless the DP first moves to the edge of the phase. This is illustrated in (4). 64 Therefore, movement analyses of control necessitate the reduced structure of control. (4) [IP John [I [VP (John) try [CP (John) [IP (John) [to [VP (John) leave]]]]]]. merge [Case, D] 64 [θ2] [?] [D] [θ1] This movement also can be interpreted as a violation of the Chain Uniformity Principle: “A chain must be uniform with regard to phrase structure status.” (Chomsky 1994:18) 155 In addition, Hornstein (1999) claims that null subjects in NOC constructions are analyzed as pro and are licensed in [Spec, IP] of CP complements. This is different from the standard approaches to control that treat null subjects in non-finite constructions as PRO regardless of the type of control construction, i.e., OC or NOC. As can be seen in (5a), the availability of an overt DP in the null subject position indicates that the position must have its Case and/or theta features checked. Therefore, the null subject does not need to move up in order to check Case, as it does in OC constructions. As a result, movement from the position is prohibited, as shown in (5b). (5) a. It is believed that Bill’s/pro shaving is important. b. *Bill’s is believed that shaving is important. (Hornstein 1999:p.92) Note that the previous approaches to the structure of control complements are all based on the typical properties of non-finite control complements: i.e., lack of a complementizer and lack of overt subject and Tense morphology. Given the analyses of non-finite control above, it is obvious that finite control complements in Korean must be analyzed as CP structures in which a complementizer selects a finite clause. 65 65 Recall that not only complementizers but also nominalizers and adnominalizers can head sentential complements in Korean and that the latter are often considered complementizers in the literature. Contrary to this, it has been pointed out in the literature (Philippaki-Warburton 1987; Terzi 1992; Rivero 1994; among others) that a subjunctive marker in Balkan languages is not a complementizer, but part of the inflectional system. One piece of evidence for this is that the subjunctive particle can co-occur with elements that are unambiguously complementizers (examples from Terzi 1997:351). (i) Jani do [CP qё Maria [MP tё hajё] (Albanian) John want-3SG COMP Mary PRT eat-3SG ‘John wants Mary to eat.’ 156 (6) a. Minai-ka Pataj-eykey [senmwul-un eci/*j sacwu-keyss-ta]-ko M.-NOM P.-DAT present-TOP yaksokha- buy.give-VOL-DC-C promise- yess-ta. PST-DC ‘Mina promised Pata that she would buy (her) a present.’ b. Minai-ka Pataj-eykey [Con-man ec*i/j manna-la]-ko seltukha-yess-ta. M.-NOM P.-DAT John-only meet-IMP-C persuade-PST-Q ‘Mina persuaded Pata to meet only John.’ c. Minai-ka Pataj-eykey [eci+j mwues-ul M.-NOM P.-DAT ha-ca]-ko ceyanha-yess-ni? what-ACC do-PROP-C suggest-PST-Q ‘What did Mina suggest Pata that they should do?’ (7) a. Minai-ka [ku chayk-un eci/*j phal]-leyko noleykha-yess-ta. M.-NOM the book-TOP sell-C endeavor-PST-DC ‘Mina endeavored to sell the book (but not others).’ b. Minai-ka Pataj-eykey [Con-man M.-NOM P.-DAT ec*i/j/*k manna]-tolok seltukha-yess-ta. John-only meet-C persuade-PST-DC ‘Mina persuaded Pata to meet only John.’ As shown in the above examples (and discussed in 3.2), an OC reading is available in finite control complements which are headed by a complementizer. Intuitively, the presence of a complementizer gives the complement the flavor of a distinct and separate (ii) Doresc [CP ca pe Ion [MP sa-l examineze Popescu] (Rumanian) wish-1SG COMP pe John PRT-him examine-3SG Popescu ‘I wish for John to be examined by Popescu. 157 clause, a CP. Further motivation for the CP status of finite control complements in Korean is derived from the presence of a Mood particle that projects ForceP. The examples in (6) show that finite control complements can host the declarative Mood ta, imperative -la, or proposative -ca. In addition, the examples in (6) show that control complements can contain an element that projects a CP layer, such as a TopP or a FocP. 66, 67 For instance, (6a) contains the object contrastive topic senmwul-un and (6b) contains the object contrastive focus Con-man. 68 Similarly, the examples in (7) show that the contrastive topic or focus can appear inside the so-called non-finite complements, although the complements do not contain any Mood or Tense morpheme on the verb. This evidence demonstrates that the structure of finite control in Korean cannot be reduced to an IP/TP. In conclusion, finite control complements in Korean should be analyzed as a CP structure. Consequently, the CP phasal nature of the complements rules out the movement theory of control. With respect to the type of the empty category in finite control complements, the CP status of finite control complements suggests that the empty category of the embedded clauses is pro. Since the embedded clauses in finite control are CPs, it is 66 It has been proposed that a contrastive topic and a wh-phrase undergo movement into Spec of CP at LF (see Hoji 1985 and Cho 1998 for the analysis of a contrastive topic). If it is the case, it can strongly support the CP status of the finite control complements. However, it is beyond the scope of this thesis to address the syntactic mechanism that licenses those elements. The important thing to note with respect to the structure of finite control complements at issue is that the structure of the control complements should be higher than an IP/TP since the complements contain such an information structure. Bošković (1995) demonstrates that topicalization, while permissible in a CP structure, is impossible in an infinitival. 67 68 Contrastive topic in Korean is marked with morphological marker -(n)un which is usually called as the topic marker. Also, the contrastive focus marker -man ‘only’ is sometimes called the ‘exhaustive focus’ maker. 158 expected that two independent subjects can be licensed, both in the matrix and in the embedded clause. In fact, this holds true, as shown in (8) in which the embedded complements have an overt pronominal subject kunye ‘she’ and already discussed in 3.2.1. (8) a. Minai-ka Pataj-eykey [senmwul-un kunyei/*j/*kka sacwu-keyss-ta]-ko M.-NOM P.-DAT present-TOP she-NOM yaksokha- buy.give-VOL-DC-C promise- yess-ta. PST-DC ‘Mina promised Pata that she would buy a present.’ b. Minai-ka Pataj-eykey [Con-man kunye*i/j/*k-ka manna]-tolok seltukha-yess-ta. M.-NOM P.-DAT John-only she-NOM meet-C persuade-PST-DC ‘Mina persuaded Pata that she should meet only John.’ As in the corresponding sentences in (6) and (7), an OC reading is still required despite the overt lexical pronouns in the embedded clauses in (8). Given the typical property of OC constructions that overt subjects are banned in the subject position of the complement clauses, the fact that an overt pronoun in nominative Case can occur in the embedded clause supports the claim that what is involved in finite control in Korean is pro and not PRO. 69 However, finiteness or the CP status of control complements might 69 Similar claims for other languages have been made in the literature. See PhilippakiWarburton & Catsimali (1999), Spyropoulos & Philippaki-Warburton (2001), Philippaki-Warburton (2004), and Spyropoulos (2007) for Modern Greek, and Zec (1987) for Serbo-Croatian. Additionally, Suñer (1984) and Hashemipour (1985) allow not only PRO, but also pro to be controlled, in order to account for controlled finite complements in Spanish and Persian, respectively. 159 not necessarily guarantee that the null subject in control complements is pro, considering the claim that PRO can be involved in finite OC constructions, such as in Landau (2000) discussed in 3.2.1. The following Hebrew example shows that a nominative Case-marked PRO can occur in a finite CP. (9) Himlacti le-Gil1 [še-PRO1/*2/ Dani2/hu2 yearšem I-recommended to-Gil la-xug that-PRO/Dani/he will-register.3SG.M le-balšanut]. to-the-department to-linguistics ‘I recommended to Gil to register to the linguistics department.’ ‘I recommended to Gil that Dani/he should register to the linguistics department.’ (Landau 2004:p.813) (10) a. [CP DP...F.. [CP Co[+T, +Agr, Agree +R] [IP[I’ Io [+T, +Agr, +R] [VP DP/pro[+R],1st/2nd...]]]]] Agree[+T, +Agr, +R] Agree[+Agr, +R] b. [CP DP....F.. [CP Co[+T, +Agr, +R] [IP PRO[–R] [I’ Io[+T, +Agr, +R] [VP tPRO..]]]]] Agree[+Agr, -R] Agree[+Agr] Agree[+T, +Agr, +R] Agree[+Agr] (Landau 2004:p.846) Note that the null subject in (9), PRO in Landau’s analysis, is co-referent with a matrix argument, while an overt DP in the same position has a different referential interpretation. Thus, as illustrated in (10), the distribution of PRO and a lexical DP/pro in his analysis is regulated by their inherent capability of independent reference ([±R] feature), not by the syntactic features, i.e. [+T, +Agr] on Io and Co. However, the 160 referential difference between in a null subject and a lexical DP/pro does not obtain in OC in Korean. In fact, the following data as well as (8) indicate that a null subject in a finite context patterns together with a lexical pronoun in Korean. (11) a. Nayi-ka Nej-eykey [eci/*j/*k/nay-ka/*casin-i I-NOM you-DAT ttena-keyss-ta]-ko malha-yess-ta. I-NOM self-NOM leave-VOL-DC-C tell-PST-DC ‘I told you that I would leave.’ b. Nayi-ka Nej-eykey [ec*i/j/*k/ney-ka/*casin-i I-NOM you-DAT ttena-la]-ko malha-yess-ta. you-NOM/self-NOM leave-IMP-C tell-PST-DC ‘I told you that you should leave.’ c. Nayi-ka Nej-eykey [eci/j/k/nay/ney/ku(nye)-ka/*casin-i ttena-li-la]-ko I-NOM you-DAT I/you/(s)he-NOM/self-NOM leave-CONJ-DC-C malha-yess-ta. tell-PST-DC ‘I told you that I/you/(s)he/*self might leave.’ The embedded clauses in (11) are structurally identical in that they are headed by the same complementizer selected by the same matrix predicate and they even show the same event time, which should be unrealized at the time of the utterance. The only difference in the clauses is the modal markers which are the factor that triggers the different control relations in the constructions. Unlike (11a) and (11b) in which the volitional modal suffix -keyss- and the imperative Mood suffix -la- trigger obligatory subject and object control, respectively, (11c), which takes the conjecture modal suffix - 161 li-, shows NOC. The empty category in (11c) can refer to the speaker ‘I’, the hearer ‘you’, or a 3rd person salient from the context. Note that if the null subjects in (11a,b) are PROs as in Landau’s analysis for the Hebrew data, we would not expect them to be able to have the same reference as a lexical pronoun, and it would not be expected that the null subject could obligatorily refer to any other person than a 3rd person. That is, the fact that an overt pronoun in nominative Case can be licensed instead of the null subject with the same referential property in Korean suggests that Obligatory Control can be attested in cases where no PRO can be licensed. In addition, the examples in (11) show that unlike the personal pronoun, the monomorphemic reflexive casin 70 cannot occur in this context. This suggests that null subjects in Korean OC constructions are pronominal, but not anaphoric in nature in that they are A-free in their local domain, while anaphors must be locally bound. The examples in (12) show that finite control complements can stand as independent clauses with the same referential properties of an empty category and a lexical pronoun as in (11). (12) a. proi/*j/*k/nay/*ney/*ku(nye)-ka/*casin-i ttena-keyss-ta. I/you/(s)he-NOM/self-NOM leave-VOL-DC ‘proi/I/*you/*(s)he/*self will leave.’ b. pro*i/j/*k/*nay/ney/*ku(nye)-ka/*casin-i ttena-la. I/you/(s)he-NOM/self-NOM leave-CONJ-DC ‘(proj/*I/You/*(s)he/*self) leave! 70 Unlike caki which is restricted to 3rd person DPs, casin can refer to the 1st and 2nd person as well as the 3rd person. Also, it is well-known that casin can function as a long-distance reflexive. 162 c. proi/j/k/nay/ney/ku(nye)-ka/*casin-i ttena-li-la. I/you/(s)he-NOM/self-NOM leave-CONJ-DC ‘proi/j/k/I/you/(s)he/he might leave. This also indicates that the empty category in finite control complements is not PRO, but pro, which has the same distribution as an overt pronoun. Considering that Korean is a pro drop-language in which a pronoun can be dropped when it is in some sense pragmatically inferrable, the empty category in finite control complements can be analyzed as pro. In what follows, I will more thoroughly investigate the syntactic nature of the empty subject at issue. 4.2. Syntactic Properties of the Null Subject in Finite Control It has been proposed in the past that control in Korean can be established with pro elements (cf., Yang 1984; Borer 1989). However, the recent approaches to control in Korean, such as Madigan (2006, 2008), argue for a PRO analysis for finite control. Moreover, the distribution and the syntactic properties of a controlled pro in Korean have not been well addressed. In this section, I will show that the null subject in finite control complements in Korean is pro, whose Case and formal features are checked in an independent locality domain, i.e. CP, which is fully specified for Tense, Agreement, and Mood. Thus, the null subject in a finite OC construction is compatible with a pronominal nature, whether it is obligatorily controlled or not. 4.2.1. Case on a Controllee It has been claimed in the literature that the null subject, PRO, in control complements inherits the Case along with phi-features of the antecedent when it is 163 obligatorily controlled, whereas it bears some ‘independent’ Case when it occurs in a structural environment which blocks OC. Contrary to this, the null subject in OC complements in Korean can bear the nominative Case that is checked in an independent locality domain. As discussed in the previous section, the null subject in finite control complements in Korean can be replaced with an overt pronoun which is marked for nominative Case. This tells us that the null subject can also take nominative Casemarking. In this section, I will provide additional evidence showing that the null subject in OC complements in Korean must have nominative Case. 4.2.1.1 Quantifier agreement It is well known that in Korean, post-nominal quantifiers must agree with the head noun, as follows: (13) a. Haksayng-(tul)-i seys-(i)/*ul ku chayk-ul ilk-ess-ta. student-PL-NOM three-NOM/ACC the book-ACC read-PST-DC ‘Three students read the book.’ b. Haksayng-(tul) seys-i student-PL ku chayk-ul ilk-ess-ta. three-NOM the book-ACC read-PST-DC c. Haksayng-(tul)-i student-PL-NOM ku chayk-ul seys-*(i) ilk-ess-ta. the book-ACC three-NOM read-PST-DC (14) a. Nay-ka chinkwu-(tul)-hanthey seys-(hanthey)/*ul pheynci-lul I-NOM friend-PL-DAT ‘I sent a letter to three friends.’ three-DAT/ACC letter-ACC ponay-ss-ta. send-PST-DC 164 b. Nay-ka chinkwu-(tul) seys-hanthey pheynci-lul I-NOM friend-PL three-DAT letter-ACC ponay-ss-ta. send-PST-DC c. Nay-ka chinkwu-(tul)-hanthey pheynci-lul seys-*(hanthey) ponay-ss-ta. I-NOM friend-PL-DAT letter-ACC three-DAT send-PST-DC In (13a) and (14a), the quantifiers have nominative and dative Case, which are the same Case as the nominals they modify, respectively. Accusative Case on the quantifier is illicit. This is because of the restriction that in simplex sentences the Case marker of a quantifier must agree with that of the noun it quantifies. The examples above show that either the Case marker of the head noun or the quantifier can be dropped. However, when the quantifier is floated, the omission of the Case marker causes the sentence to be ungrammatical, as shown in (13c) and (14c). This clearly indicates that post-nominal quantifiers in Korean must agree with the head noun. Based on this generalization, one can deduce that the null subjects in the following examples must be marked by nominative Case because their quantifiers are nominative. 71 (15) a. Sensayng-nim-kkeyse haksayng-tuli-ul [eci motwu-(ka) yelsimhi teacher-HON-NOM.HON student-PL-ACC kongpwuha-tolok seltukha-si-ess-ta. study-C persuade-HON-PST-DC all-NOM hard ‘The teacher persuaded all the students to study hard.’ 71 See Sigurðsson (1991:331f) for Icelandic where floated quantifiers in infinitive constructions bear the case of the null subject. 165 b. Kim sacang-i cikwen-tuli-eykey [eci ku Kim president-NOM employee-PL-DAT chamsekha]-l kes-ul hoyui-ey motwu-(ka) the meeting-LOC all-NOM myenglyengha-yess-ta. participate-ADN thing-ACC order-PST-DC ‘President Kim ordered all his employees to participate in the meeting.’ c. Haksayng-(tul)i -i [eci cip-ey student-PL-NOM seys-i ka-keyss-ta]-ko yaksokha- home-LOC three-NOM go-VOL-DC-C promise- yess-ta. PST-DC ‘The students promised to, the three of them, go home.’ (15a,b) and (15c) show obligatory object and subject control, respectively. Notice that the null subjects in (15a,b) can bear a distinct Case from the controller. That is, unlike the object controller in (15a) and (15b) which bears Accusative and Dative Case, respectively, the null subject must bear Nominative Case, which is evidenced by the presence of a nominative Case-marked quantifier. This indicates that an analysis in which the null subject in OC contexts inherits Case from its controller 72 cannot be applied to Korean and that the null subject must be able to check Case in its locality domain. 4.2.1.2. Honorific suffix agreement In Korean, where subjects denote a person who is honorified, a verbal suffix -(u)siis added to the predicate. 72 Cecchetto & Oniga (2004) propose such a theory of inheritance based on the fact that PRO always shares Case with its controller in Latin (and Italian). 166 (16) a. Sensayng-nim-kkeyse ku ai-lul manna-si-ess-ta. teacher-HON-NOM.HON the child-ACC meet-HON-PST-DC ‘The teacher met the child.’ b. Nay-ka I-NOM ecey ku sensayng-nim-ul manna-ss-ta. yesterday the teacher-HON-ACC meet-PST-DC ‘I met the teacher yesterday.’ c. Sensayng-nim-uy atul-i cip-ey ka-n-ta. teacher-HON-GEN son-NOM home-LOC go-PRS-DC ‘The teacher’s son goes home.’ As shown in (16), -(u)si- can be attached to verbs when the subject is honored by the speaker. As Hong (1994) among many others has pointed out, grammatical subjects honored by the speaker are responsible for -si- marking, although semantic and pragmatic factors play a significant role. That is, while a nominative subject determines -(u)si- marking, as shown in (16a), other NPs with Case other than nominative do not trigger the marking, as shown in (16b) and (16c). If we assume that only nominative NPs can trigger honorific agreement, the following data provide evidence that the embedded null subjects are marked for nominative Case. (17) a. Halapecii-kkeyse [eci chayk-ul ilk-(usi)]-ki sicakha-(si)-ess-ta. grandfather-NOM.HON book-ACC read-HON-NOML begin-HON-PST-DC ‘Grandfather began to read a book.’ 167 b. Sensyangi-nim-kkeyse [eci cha-lul teacher-HON-NOM.HON kochi-(si)]-lyeko nolyekha-(si)-ess-ta. car-ACC repair-HON-C try-HON-PST-DC ‘The teacher tried to repair the car.’ c. Minai-ka emenij-lul [ecj hakkyo-ey M.-NOM mather-ACC ka-(si)]-tolok seltukha-yess-ta. school-LOC go-HON-C persuade-PST-DC ‘Mina persuaded (her) mother to go to school.’ In (17), -(u)si- can be added to the embedded predicates, providing evidence that the null subjects in the control constructions are marked for nominative Case. 4.2.1.3. Occurrence of an Emphatic Subject Pronoun in Control Constructions Further evidence that embedded null subjects in Korean receive nominative Case comes from the fact that the complement clause of control constructions can have an emphatic subject pronoun which is obligatorilymarked with nominative Case. (18) a. Minai-ka [nayil eci/*j/kunyecasini/*j/cakicasini/*j-i/*ul ttena]-lyeko M.-NOM tomorrow herself self-NOM/ACC leave-C nolyekha-yess-ta. try-PST-DC ‘Mina herself tried to leave tomorrow.’ b. Minai-ka Pataj-eykey [nayil M.-NOM P.-DAT cenyek-ul sa]-ki-lo eci/*j/*k/kunyecasini/*j/*k/cakicasini/*j/*k-i/*ul tomorrow herself yaksokha-yess-ta. dinner-ACC buy-NOML-DR promise-PST-DC self-NOM/ACC 168 ‘Mina promised Pata that she herself would buy dinner tomorrow.’ c. Minai-ka Pataj-lul [nayil M.-NOM P.-ACC eci/*j/*k/kunyecasin*i/j/*k/*cakicasin-i/*ul tomorrow herself ttena]- self-NOM/ACC leave- tolok seltukha-yess-ta. C persuade-PST-DC ‘Mina persuaded Pata that she herself would leave tomorrow’ The placement of the temporal adverb nayil ‘tomorrow’ in (18) indicates that the emphatic pronouns kunyecasin and cakicasin occur in the embedded clauses. The emphatic pronouns are complex reflexives which can be morphologically divided into two lexical items: ‘Pronoun + casin’ and ‘caki + casin. These examples demonstrate that emphatic pronouns can alternate with the null subjects in OC constructions, suggesting that even when the embedded subject is null, it must be assigned nominative Case. Note that the emphatic pronoun shows the same referential property as the null subject, although cakicasin is not permitted in the case of object control due to the subject orientation of Korean reflexives. This suggests that the null subjects in OC constructions are more compatible with a pronominal nature. 4.2.1.4 Availability of a Controllee Conjoined with a Lexical DP In Korean, when two lexical NPs are conjoined, Case is marked only on the second conjunct and the first conjunct carries the nominal conjunctive suffix -(k)wa, 73 as shown in the following examples. 73 There is another way to make an NP coordination in Korean. In this case, Case is marked on both conjuncts and the conjunction kuliko occurs between the conjuncts, as follows: 169 (19) a. Mina-ka Pata-wa hakkyo-ey ka-ss-ta. M.-NOM P.-and school-LOC go-PST-DC ‘Mina went to school with Pata.’ b. Mina-wa Pata-ka M.-and hakkyo-ey ka-ss-ta. P.-NOM school-LOC go-PST-DC ‘Mina and Pata went to school.’ Unlike (19a) in which only Mina is the subject of the predicate ka- ‘go’ since the two NPs Mina and Pata are not conjoined, Mina and Pata in (19b) are conjoined and function as the subject of the predicate ka- ‘go’. However, nominative Case is marked only the second conjunct. The following data show the same pattern. (20) a.*Mina-ka Pata-wa twul-man M.-NOM P.-and hakkyo-ey ka-ss-ta. two-only school-LOC go-PST-DC ‘Only two, Mina and Pata, went to school.’ b. Mina-wa Pata-(ka) twul-man M.-and P.-NOM hakkyo-ey ka-ss-ta. two-only school-LOC go-PST-DC ‘Only two, Mina and Pata, went to school.’ As shown in (20), the quantifier ‘twul-man’ can follow a head noun which involves an NP coordination, but cannot follow a head noun which is a semantically and Mina-ka kuliko Pataj-ka hakkyo-ey ka-ss-ta. M.-NOM and P.-NOM school-LOC go-PST-DC ‘Mina and Pata went to school.’ 170 syntactically singular subject. Based on this property, the following examples show that the null subject is conjoined with an overt lexical NP. (21) a. Minai-ka [nayil Wucin-kwa eci/kunyei/*j/cakii/*j twul-man M.-NOM tomorrow W.-and she self two-only hakkyo-ey school-LOC ka]-leyko nolyekha-yess-ta. go-C try-PST-DC ‘Mina tried (only two, Wucin and ec/she/herself,) to go to school, tomorrow.’ b. Minai-ka Pataj-eykey [nayil M.-NOM P.-DAT hakkyo-ey Wucin-kwa eci/*j/kunyei/*j/cakii/*j twul-man tomorrow W.-and she self two-only ka-keyss-ta]-ko yaksokha-yess-ta. school-LOC go-VOL-DC-C promise-PST-DC ‘Mina promised Pata that only two, Wucin and ec/she/herself, would go to school tomorrow.’ c. Minai-ka Pataj-lul [nayil Wucin-kwa ecj/kunye*i/j /*caki twul-man M.-NOM P.-ACC tomorrow W.-and she self two-only hakkyo-ey ka]-tolok seltukha-yess-ta. school-LOC go-C persuade-PST-DC ‘Mina persuaded Pata that only two, Wucin and ec/Pata, go to school, tomorrow.’ In the examples above, the temporal adverbs occur in the embedded clause, which, in turn, shows that the overt NP Wucin belongs to the embedded clauses. The distribution of the quantifier twul-man shows that the subject of the embedded predicat must be 171 plural. This indicates that the NP Wucin is conjoined with a null subject. In fact, the occurrence of the lexical pronoun kunye ‘she’ or the reflexive caki in the null subject position shows that the null subject in the complement clauses is conjoined with the NP Wucin, which carries the nominal conjunctive suffix -(k)wa. Therefore, it is analyzed that the null subject in the NP coordination carries nominative Case. 4.2.2. Agree Features of Controllee It has been claimed that PRO is anaphoric and thus, has no inherent phi features, 74 while pro is pronominal and possesses inherent phi features, as lexical NPs do. According to Landau (2000, 2004) who argues that PRO is ‘active’ for Agree due to its anaphoric nature, PRO in tensed complements inherits all phi-features from the controller, including semantic plurality. However, Landau argues that PRO does not necessarily inherit semantic singularity. This helps explain instances of PC in English, as follows: (22) a. John1 wanted [PRO1+ to leave together]. b. The committee1 was glad that the chair had agreed [PRO1 to gather before the elections]. Sem. SG. Sem. PL. c.*The chair1 was glad that the committee had agreed [PRO1 to wear a tie]. Sem. PL. Sem. SG (22a) shows that controllees can differ from their controllers in semantic plurality because the controlees refer to a set including the matrix controllers. In contrast to this, 74 Terzi (1997:343) claims that PRO in Greek subjunctive complements bears Agr features and thus, it has to be raised to the Spec, AgrSP to check the Agr feature. 172 (22b) and (22c) 75 show that a semantically plural controller cannot control a semantically singular PRO. Based on this evidence, Landau claims that the controller is semantically and syntactically singular, but the controllee is semantically plural and syntactically singular. However, this analysis is problematic when we look at the following data from Romanian. (23) a. *Eu vreau I want b. Eu vreau I want [să plec ȋmpreună] SBUJ leave.1SG together [să plecăm ȋmpreună] SBUJ leave.1PL together ‘I want to leave together.’ (Alboiu 2004:60) (23a) shows that a syntactically singular subject cannot license a semantically plural predicate (i.e., a collective predicate) as in English (22a). The subject has to be plural, as in (23b), in which case the OC constructions trigger exhaustive control, not partial control. These data indicate that the null subject in OC constructions does not necessarily inherit its phi-features from its controller. Also, as Madigan (2006) points out, this analysis becomes problematic for Korean PC. (24) a.*Minai-ka moyess-ta. M.-NOM gathered-DC ‘Mina gathered.’ 75 In Landau (2000:53), the coindexation of the PRO and the matrix argument shows that they share the same semantic number. 173 b. Haksayng*(tul)-i ppali-tul moyess-ta student-(PL)-NOM quickly-PL gathered-DC ‘Students gathered quickly.’ c. Wiwonhoy-ka ppali-(*tul) moyess-ta. committee-NOM quickly-PL gathered-DC ‘The committee gathered.’ In Korean, there is no agreement marking on a verb. However, number agreement for subjects is optionally realized with the plural marker -tul on an adverb, a noun, a postposition, or a complementizer. As shown in (24), the predicate moi- ‘gather’, which is a semantically plural predicate, requires a semantically plural subject. Namely, if the subject is not semantically plural, the insertion of a plural marker -tul is necessary, as in (24b), whereas syntactically singular subjects can license the predicate if the subjects are semantically plural, as in (24c). Given this, the empty categories in (25) must be syntactically plural. This is proved by the plural marker -tul in the adverb phrase which can occur only with a plural subject. (25) a. Elun-tuli-un [eci ilccik-tul ttena]-ki-lo the.elder-PL-TOP kyelcengha-yess-ta. early-PL leave-NOML-DR decide-PST-DC ‘The elders decided to leave early.’ b. Minai-ka [eci+ tosekwan-eyse-(tul) moi]-ki-lo M.-NOM library-LOC-PL yaksokha-yess-ta. gather-NOML-DR promise-PST-DC ‘Mina promised to gather in the library.’ 174 (26) a. Minai-ka Pataj-eykey [eci+ motwu/twul-man/selo M.-NOM P.-DAT moi-la]-ko malha-yess- all/two-only/each other gather-IMP-C tell-PST- ta. DC ‘Mina told Pata, all/two of them only/each other, to gather.’ b. Minai-ka [caki-*(tul)i+ -i tosekwan-eyse-(tul) moi-keyss-ta]-ko M.-NOM self-(PL)-NOM library-LOC-PL yaksokha- gather-VOL-DC-C promise- yess-ta. PST-DC ‘Minai promised that theyi+ , themselves, would gather in the library.’ (25a) is an instance of EC in which the null subject is controlled by the matrix plural subject and carries a [+pl] feature, while (25b) is an instance of PC in which the null subject is not exhaustively controlled by the matrix singular subject and carries a [+pl] feature which licenses the plural affix on the locative NP tosekwan ‘library’. In Landau’s analysis, the controllee in EC, by definition, must match all phi-features of its controller, while that in PC must inherit all phi-features from the controller, except semantic singularity which is not necessarily inherited. However, (25b) shows that the controllee in Korean PC does not inherit the syntactic singularity of its controller. Similarly, (26a) shows that the empty subject can combine with selo or any quantifier which is compatible with a plural subject. 76 In addition, the occurrence of an overt 76 Madigan (2006) considers (1a) an example of PC under the analysis that hamkkey ‘together’ behaves in the same manner as selo and quantifiers, such as ‘all’. However, as shown in (1b,c), hamkkey can be compatible with a singular subject. That is, hamkkey does not necessarily require a syntactically plural subject, unlike selo. 175 plural subject in (26b) clearly demonstrates that the controllee in this context must be plural. Consequently, this evidence indicates that null subjects in Korean PC contexts do not need to match the syntactic plurality of the controllers, contrary to what Landau (2004) suggests. Since the control complements are CPs which are fully specified for Tense, Agreement, and Mood, the null subjects can be licensed within their local domain and thus, can carry inherent phi-features, just as lexical NPs can. Therefore, these data support the claim that null subjects in Korean OC contexts are pro, not PRO. 4.2.3. Pro Different From the LD reflexive Caki and a Pronoun Madigan (2006) argues that the long-distance (LD) reflexive caki is an overt form of the null subject, PRO, in OC constructions. 77 This is reminiscent of the analysis maintaining that control is reducible to a highly constrained anaphoric relation. 78 Following Landau’s (2000) diagnostics, 79 Madigan (2006) shows some evidence that LD reflexive caki and the null subject of embedded clauses in Korean control constructions behave identically under all conditions. Contrary to this proposal, I will (1) a. Jwuhi1-ka [PRO1+ /caki-*(tul)-i1+ hamkkey ka-keyss-ta]-ko yaksokha-yess-ta. J.-NOM self-PL-NOM together go-VOL-DC-C promise-PST-DC ‘Jwuhi promised/themselves to go together.’ b. Jwuhi1-ka [ec1/caki1-ka hamkkey/*selo ka-keyss-ta]-ko yaksokha-yess-ta. J.-NOM self-NOM together/each.other go-VOL-DC-C promise-PST-DC ‘Jwuhi promised that (she/self) would go together.’ c. Jwuhi1-ka [caki1-ka Mina-wa hamkkey ka-keyss-ta]-ko yaksokha-yess-ta. J.-NOM self-NOM M.-with together go-VOL-DC-C promise-PST-DC ‘Jwuhi promised to go with Mina.’ 77 Kayne (1994:70) proposes a similar claim that the PRO subject of indicative-like infinitives is a subject-oriented anaphor. 78 It has been proposed that many of the properties of OC PRO are shared by reflexive anaphors (Manzini, 1983; Bouchard, 1984; Martin, 1996, among others). 79 Madigan adds one more property- a referential lexical DP may not replace PRO- to Landau’s (2000) diagnostics as properties of OC in Korean. 176 argue that the LD reflexive caki cannot be an overt form of the null subject in OC contexts and provide some evidence supporting this position. This analysis does not simply argue against the analysis of the LD reflexive as an overt form of PRO; it also contends that the controlled null subject in Korean OC is not PRO, but pro, which is different from an overt pronoun and the LD reflexive. 4.2.3.1. Different Distributions of the LD reflexive Caki and Pro Madigan (2006) argues that wherever OC PRO occurs, the LD reflexive caki can also occur and that the LD reflexive caki is an overt form of PRO in OC situations. Indeed, if this is the case, the LD reflexive caki should show the same distribution as that of PRO in OC constructions. The examples in (27) appear to support the analysis that OC caki is an overt form of OC PRO. (27) a. Jwuhii-ka [eci/*arb/cakii/*j/*arb/kunyei/*j/*arb-ka ttena-keyss-ta]-ko yaksokha-yess-ta. J.-NOM self she-NOM leave-VOL-DC-C promise-PST-DC ‘Jwuhi promised that self/she would leave.’ b. Minai-ka Pataj-eykey [ec*i/j/caki*i/j/kunye*i/j-ka cip-ey M.-NOM P.-DAT malha-yess-ta. tell-PST-DC ‘Mina told Pata to go home.’ self she-NOM ka-la]-ko home-LOC go-IMP-C 177 Both the empty category and caki show the same control readings: subject control in (27a) and object control in (27b) 80. In particular, (27b) is interesting in that caki refers to the matrix object Pata, despite the fact that caki, in general, is a subject-oriented reflexive. However, notice that the 3rd person pronoun kunye ‘she’ in OC contexts also shows the same distribution as the empty category, just as the controlled caki constructions do. Moreover, the following examples in (28) show that the LD reflexive and the null subject can have different distributions in OC constructions and that a personal pronoun shows the same distribution as the null subject, as briefly discussed in section 4.1. Therefore, these data provide evidence against the analysis that caki is an OC PRO. (28) a. Minai-ka Pataj-eykey [ec*i/j/*caki/kunye*i/j -ka cip-ey M.-NOM P.-DAT self/she-NOM ka-la]-ko home-LOC go-IMP-C seltukha-yess-ta. persuade-PST-DC ‘Mina persuaded Pata to go home.’ b. Nayi-ka [eci/*cakii/nayi-ka ku il-ul I-NOM kkuthnay-keyss-ta]-ko yaksokha- self/I-NOM the work-ACC finish-VOL-DC-C promise- yess-ta. PST-DC ‘I promised to finish the work.’ 80 Some Korean speakers, including myself, consider caki unacceptable in (27b). The form caki can be used as a discourse pronoun indicating ‘you’, when addressing a person who is close to the speaker. Therefore, for the Korean who accepts caki in (27b), caki in that case is not a pure anaphor. 178 c. Patai-ka Minaj-lopwuthe [ec*i/j/*caki/kunye*i/j-ka ttena]-ki-lo P.-NOM M.-from self/she-NOM yaksok- leave-NOML-DR promise- pat-ass-ta. receive-PST-DC ‘Pata got a promise from Mina to leave.’ d. Patai-ka Minaj-lopwuthe [eci/*j/*caki/kunyei/*j -ka ttena]-tolok seltukP.-NOM M.-from self/she-NOM leave-C persuasion- pat-ass-ta. receive-PST-DC ‘Pata got persuaded by Mina to leave.’ Since (28a) has the same embedded clause containing the imperative Mood maker ‘-la-’ as (27b), it should allow the same occurrence of caki as (27b). However, unlike (27b), caki cannot occur in (28a). In contrast, the empty category appears in both constructions and shows object control induced by the Mood marker. The examples in (28b-d) also illustrate that caki cannot always occur in OC contexts. (28b) shows that the reflexive caki cannot have an antecedent referring to a 1st person singular pronoun, while the null subject can. The examples in (28c, d) show that in passivized OC constructions, the reflexive caki cannot replace the null subject. Given the fact that caki is subject-oriented, the impossibility of caki replacement in (28c) is natural because the controller Mina is not a subject of the matrix clause. On the other hand, in (28d) although the controller Pata is the subject of the matrix clause, caki cannot replace the null subject. In these constructions, the controllers are non-agents which do not match with the semantic role 179 of caki; namely, ‘Agent’ of the embedded predicate. That is to say, while a null subject can have an antecedent which is an agent or non-agent controller, caki can only have an antecedent which must be syntactically a subject and semantically an agent. This suggests that even though the reflexive caki can replace the empty category in many cases of OC constructions, the controlled empty category should be treated differently from the reflexive. This also holds true for an overt pronoun, as will be addressed in detail in the following sections. 4.2.3.2. Different References and Capability of LD Control Another piece of evidence against the analysis of the LD reflexive as an overt form of a controllee is that the LD reflexive and a controllee can have different referents, as shown in the following examples. (29) a. Minai-ka Pataj-eykey [ec*i/j/i+j ttena]-l kes-ul M.-NOM P.-DAT yokwuha-yess-ta. leave-C thing-ACC ask-PST-DC ‘Minai asked Pataj that shej/theyi+j would leave.’ b. Minai-ka Pataj-eykey [cakii/*j-ka/caki-tul?i+k-i M.-NOM P.-DAT ttena]-l kes-ul yokwuha- self-NOM /self-PL-NOM leave-C thing-ACC ask- yess-ta. PST-DC ‘Minai asked Pataj that shei /they?i+k would leave.’ c. Minai-ka Pataj-eykey [kunyei/j-/ku-tul?i+j-i M.-NOM P.-DAT ttena]-l kes-ul yokwuha- she-NOM/he-PL-NOM leave-C thing-ACC ask- 180 yess-ta. PST-DC ‘Mina asked Pata that shei/j would leave/that they (Mina and Pata) would leave.’ (30) a. Patai-ka Minaj-eykey [eci/i+j ttena]-ki-lo P.-NOM M.-DAT yaksokha-yess-ta. leave-NOML-DR promise-PST-DC ‘Patai promised Minaj to leave/Patai promised Minaj that theyi+j would leave.’ b. Patai-ka Minaj-eykey [cakii/*j-ka/caki-tuli+k-i P.-NOM M.-DAT ttena]-ki-lo yaksokha- self-NOM/self-PL-NOM leave-NOML-DR promise- yess-ta. PST-DC ‘Patai promised Minaj that she herselfi/they themselvesi+k would leave.’ c. Patai-ka Minaj-eykey [kunyei/*j-ka/ku-tul?i+k-i ttena]-ki-lo P.-NOM M.-DAT yaksokha- she-NOM/he-PL-NOM leave-NOML-DR promise- yess-ta. PST-DC ‘Patai promised Minaj that shei/they?i+k would leave.’ In (29a), the null subject can refer to the indirect object or both the matrix subject and the indirect object. However, in (29b), the reflexive caki can only refer to the matrix subject and when it takes the plural suffix -tul, it cannot include the matrix object as its antecedent. Compared to the reflexive caki, the overt singular pronoun in (29c) can refer to either the matrix subject or the matrix object and the overt plural pronoun can have a split antecedent. The similar pattern is found in the examples in (30). That is, 181 compared to the null subject in (30a) which can have split control as well as subject control, the caki in (30b) and the pronoun in (30c) can only refer to the matrix subject, when they are singular and have referents different from the null subject, even when they are plural. These different referential properties provide evidence against the analysis that the controlled caki is an overt form of a controllee. The following examples further support the aforementioned claim, as they exhibit different referential properties. (31) a. Jwuhii-ka [ec yeses-si-ey manna-keyss-ta]-ko yaksokha-yess-ta. J.-NOM six-time-at meet-VOL-DC-C promise-PST-DC i) ‘Jwuhii promised that (shei) would meet (someone) at 6.’ ii) ‘Jwuhii promised that (shei and someonej) would meet at 6.’ b. Jwuhii-ka [cakii/kunyei-ka yeses-si-ey manna-keyss-ta]-ko yaksokha-yess-ta. J.-NOM self/she-NOM six-time-at meet-VOL-DC-C promise-PST-DC i) ‘Jwuhii promised that shei, herself, would meet (someone) at 6.’ ii)*‘Jwuhii promised that shei (and someonej) would meet at 6.’ Madigan (2006) argues that (31a) is an example of PC in Korean, and thus, the singular form of caki is ungrammatical. However, almost all arguments can be deleted in Korean dialogue when they are clear in the discourse, and the predicate ‘meet’ does not always select a plural subject. Therefore, the embedded sentence can also be interpreted with a singular null subject, which is coindexed with the matrix subject, and a missing object, as shown in interpretation i). That is, (31a) is actually ambiguous in that it can be EC in 182 which the null subject can refer to Jwuhi only, or PC in which the null subject can refer to a plural entity including the matrix subject. Unlike the empty category in (31a), the reflexive caki and the personal pronoun kunye in (31b) can only refer to Jwuhi, which forces the EC interpretation. This shows that the LD reflexive and a controllee can have different referents With regard to LD control, Madigan (2006) claims that LD control is impossible with OC PRO and caki, as follows: (32) a. Minai-ka [Pataj-ka [ec*i/j ttena-keyss-ta]-ko yaksokha-yess-ta]-ko mitM.-NOM P.-NOM leave-VOL-DC-C promise-PST-DC-C believe- nun-ta. PRS-DC ‘Mina believes that Pata promised her to leave.’ b. Minai-ka [Pataj-ka [caki*i/j-ka ttena-keyss-ta]-ko yaksokha-yess-ta]-ko M.-NOM P.-NOM self-NOM leave-VOL-DC-C promise-PST-DC-C mit-nun-ta. believe-PRS-DC ‘Mina believes that Pata promised her to leave.’ c. Minai-ka [Pataj-ka [kunye*i/j-ka ttena-keyss-ta]-ko M.-NOM P.-NOM she-NOM leave-VOL-DC-C yaksokha-yess-ta-ko promise-PST-DC-C mit-nun-ta. believe-PRS-DC ‘Mina believes that Pata promised that she would leave.’ 183 In (32a, b), the empty category and caki both show the impossibility of LD control supporting Madigan’s analysis. That is, they both are not able to take the long distant antecedent Mina in the examples. However, (32c) is problematic for the analysis because the overt pronoun also disallows LD control. The more critical argument against the analysis derives from the following examples which show that unlike the null subject in an OC context, an overt pronoun and an LD reflexive can permit LD control. (33) a. Minai-ka [Pataj-ka [ec*i/j cenyek-ul M.-NOM P.-NOM ul ic-ess-ta-ko sa]-ki-lo yaksokha]-n kes- dinner-ACC buy-NOML-DR promise-ADN thingsayngkakha-n-ta. ACC forget-PST-DC-C thingk-PRS-DC ‘Minai thinks that Pataj forgot that shej promised heri to buy dinner.’ b. Minai-ka Pataj-ka [cakii/j-ka cenyek-ul sa]-ki-lo yaksokha-n M.-NOM P.-NOM self-NOM dinner-ACC buy-NOML-DR promise-ADN kes-ul ic-ess-ta-ko sayngkakha-n-ta. thing-ACC forget-PST-DC-C thingk-PRS-DC i) ‘Minai thinks that Pataj forgot that shej promised heri to buy dinner.’ ii) ‘Minai thinks that Pataj forgot that shei promised herjto buy dinner.’ c. Minai-nun Pataj-ka [kunyei/j-ka cenyek-ul M.-TOP sa]-ki-lo yaksokha-n P.-NOM she-NOM dinner-ACC buy-NOML-DR promise-ADN 184 kes-ul ic-ess-ta-ko sayngkakha-n-ta. thing-ACC forget-PST-DC-C think-PRS-DC i) ‘Minai thinks that Pataj forgot that shej promised heri to buy dinner.’ ii) ‘Minai thinks that Pataj forgot that shei promised herj to buy dinner The examples in (33) show that the LD reflexive caki and the overt pronoun are capable of LD control, whereas the null subject is incapable of it. That is, unlike the null subject in (33a) which cannot refer to the long distance antecedent Mina, the LD reflexive caki in (33b) and and the overt pronoun in (33c) can refer to either the local antencedent Pata or the long distance antecedent Mina. This is decisive evidence against the analysis of controlled caki as an overt form of PRO and clearly shows that null subjects in OC contexts behave differently from the reflexive caki and normal pronouns. One reason that the controlled caki and pronoun constructions have the same property as the control constructions with a null subject might be because caki and a pronoun can optionally alternate with the null subject in control constructions. However, this does not mean that caki and a pronoun are overt forms of the null subject in OC constructions, unless they always show the same properties as the null subject in that position. 4.2.3.3. Sloppy vs. Strict reading and de re vs. de se interpretation It is generally claimed that PRO only supports a sloppy reading under VP ellipsis, while lexical pronouns support a strict reading under ellipsis. In other words, PRO behaves like an anaphor in OC contexts and like a pronoun in NOC contexts (Bouchard 1984). This is the case with Korean OC constructions, as shown in (34a, b). Interestingly enough, however, when the null subject position of an embedded clause in 185 OC contexts is filled with a pronoun, it also allows only a sloppy reading, as shown in (34c). (34) a. Patai-ka Minaj-eykey [eci ttena-keyss-ta]-ko P.-NOM M.-DAT leave-VOL-DC-C yaksokha-yess-ta. promise-PST-DC Enju-to kuriha-yess-ta. E.-also same-PST-DC i) Pata promised Mina that she would leave and Enju promised Mina that she (Enju) would leave. ii) *Pata promised Mina that she would leave and Enju promised Mina that she (Pata) would leave.’ b. Patai-ka Minaj-eykey [cakii-ka P.-NOM M.-DAT ttena-keyss-ta]-ko yaksokha-yess-ta. self-NOM leave-VOL-DC-C promise-PST-DC Enju-to kuriha-yess-ta. E.-also same-PST-DC i) Pata promised Mina that she would leave and Enju promised Mina that she (Enju) would leave. ii) *Pata promised Mina that she would leave and Enju promised Mina that she (Pata) would leave.’ c. Patai-ka Minaj-eykey [kunyei-ka ttena-keyss-ta]-ko P.-NOM M.-DAT Enju-to kuriha-yess-ta. E.-also same-PST-DC yaksokha-yess-ta. she-NOM leave-VOL-DC-C promise-PST-DC 186 i) Pata promised Mina that she would leave and Enju promised Mina that she (Enju) would leave. ii) *Pata promised Mina that she would leave and Enju promised Mina that she (Pata) would leave.’ The above examples clearly show the possibility of a sloppy reading, but not a strict reading with a remote controller, regardless of the type of the embedded subject. That is, in all three constructins above, what Enju promised is that she (Enju) would leave, a sloppy reading, but not that Pata would leave, a strict reading. This fact seems to be directly related to properties of OC. Indeed, a pronoun and a null subject in NOC constructions are free to get either a sloppy or a strict reading. (35) a. Minai-ka [eci hakkyo-ey M.-NOM ka]-ki-lul wenha-yess-ta. Pata-to kuriha- school-LOC go-NOML-ACC want-PST-DC P.-also same- yess-ta. PST-DC i) Mina wanted to go to school and Pata wanted to go to school. ii) Mina wanted to go to school and Pata wanted Mina to go to school.’ b. Minai-ka [kunyei-ka hakkyo-ey ka]-ki-lul wenha-yess-ta. Pata-to M.-NOM she-NOM school-LOC go-NOML-ACC want-PST-DC P.-also kuliha-yess-ta. same-PST-DC i) Mina wanted to go to school and Pata wanted to go to school. 187 ii) Mina wanted to go to school and Pata wanted Mina to go to school. c. Minai-ka [cakii-ka hakkyo-ey ka]-ki-lul wenha-yess-ta. Pata-to M.-NOM self-NOM school-LOC go-NOML-ACC want-PST-DC P.-also kuliha-yess-ta. same-PST-DC i) Mina wanted to go to school and Pata wanted to go to school. ii)*Mina wanted to go to school and Pata wanted Mina to go to school The structure in (35a) is an example of NOC in Korean in that it is possible to insert a referential DP to appear in the subject position of the embedded clause. While (35a) with the null subject and (35b) with the overt pronominal subject kunye ‘she’ admit either a sloppy reading or strict reading, (35c), in which caki is the embedded subject, can only have the sloppy reading. That is, unlike the reflexive caki which always exhibits a sloppy reading 81, a null subject and pronoun allow either a sloppy or a strict reading in NOC contexts. In this respect, a null subject and pronoun behave the same. Finally, Madigan (2006) argues for the LD reflexive caki as an overt form of PRO in OC contexts, on the basis of the fact that when the null subject position in OC contexts is filled with a pronoun, it allows only a de re interpretation, as follows: (36) “(There is) a war hero who suffers from amnesia and remembers nothing of the war. Suppose this person (hereafter, “the soldier”) sees a TV program describing his own exploits, and is impressed with the courage exhibited by that person, who he 81 In some languages, it has been noted that reflexives can only be interpreted as bound pronouns. (Zec 1987). 188 does not know is himself.” (37) a. Ku pyengsa-nun [ec/caki-ka the soldier-TOP (Madigan, 2006) sang-ul patu]-lyeko nolykha-yess-ta self-NOM prize-ACC receive-C try-PST-DC ‘That soldier tried to get the medal.’ b. Ku pyengsa-nun the soldier-TOP [ku-ka sang-ul patu]-lyeko nolykha-yess-ta he-NOM prize-ACC receive-C try-PST-DC ‘That soldier tried to get the medal.’ Under the scenario in (36), Madigan claims that (37a) can only be interpreted in a de se manner and is therefore false. However, this does not provide evidence that caki is PRO since an overt pronoun in the same context displays the same property. That is, (37b) is false since the pronoun can only refer to the matrix subject ku pyengsa and thus, can only be satisfied by de se beliefs. This shows that only a de se interpretation is possible in Korean OC constructions, regardless of the type of the embedded subject: null, reflexive, or pronoun. This evidence suggests that we cannot necessarily claim that the controlled caki is an overt form of PRO. Although some controlled caki constructions show the same properties as OC constructions, this does not necessarily mean that the controlled caki is an overt form of the null controllee in OC constructions. Rather, controlled caki constructions seem to show the same properties as OC constructions with null subjects just because OC constructions can allow caki to appear in the null subject position of embedded clauses under appropriate circumstances. That is, it is simply one thing that can occur in the null subject position and be controlled. This is also the case for controlled pronouns in OC contexts. 189 Thus far, it has been claimed that the finite OC in Korean can be established with pro whose Case and formal features are checked in an independent locality domain. In this respect, pro in finite OC is parallel to lexical subjects. However, a controlled pro differs from a lexical subject in that it is referentially dependent, although it is thematically an independent NP. The more thorough investigation of finite OC contexts in Korean further reveals that null subjects in OC contexts can have different referential properties from the LD reflexive caki and an overt pronoun and thus, the analysis of the LD reflexive caki as an overt form of PRO in Korean finite control cannot be maintained. 4.3. The Status of Serial Verb Constructions As Non-Finite Control In the literature on control in Korean, some of the sentential complements, such as -ki nominalization and -leyko and -tolok complements, have been analyzed as non-finite control. However, as discussed in the previous section, those complements in fact turn out to be finite in that they can specify dependent Tense, Agreement, and Mood and allow an overt lexical NP in nominative Case to occur in the subject position. Therefore, it is assumed that there is no identifiable infinitival embedded clause in Korean. According to Stiebels (2007), cross-linguistically languages that lack a finite-infinite distinction on the morphological level typically lack control-inducing structures such as infinitival complements, but may exhibit verb incorporation or verbal compounding. In fact, besides clausal complementation, Korean displays serial verb constructions (SVCs) 82 82 that have been analyzed as control constructions in the literature The constructions have often been called compound, complex, or auxiliary verb constructions in the literature (H-M. Sohn 1976, S-H. Choi 1988, and Sells 1998 among others). In fact, two or more verbs connected with a suffix can occur in SVCs, but this 190 (Gamerschlag 2007). Studies on Korean control generally have focused on a few types of complements that trigger Obligatory Control and have not given a full account of the status of SVCs as control constructions. However, some of the SVCs have commonly been considered control constructions (Sells 1998, Choi 2003) since they contain a typical control verb, such as ‘try’. Moreover, sentences corresponding to the Korean SVCs in other languages, such as Japanese, have been analyzed as control structures. Therefore, it is worthwhile to examine whether the SVCs in Korean involve control or not. In this section, the structural property of the SVCs in Korean which take the socalled control verb, such as ‘try’ and ‘want’ will be first investigated and then it will be proposed that the SVCs do not actually involve control, but simply argument sharing. 4.3.1. Monoclausal Status of Serial Verb Constructions In SVCs, two verbs are combined by attaching a suffix -e/a or -ko 83 to the verb stem in order to form a compound lexical item. Unlike the complementizers and the complementizer-like elements, such as a nominalizer and an adnominalizer, in finite control constructions, the suffixes -e/a and -ko do not carry any inherent Mood/Modal or Tense/Aspect meaning. They are simply attached to the verb to combine with another verb, as follows. (38) a. Kyoswu-nim-kkeyse [ku nonmwun-ul ilk-(*usi)-(*ess)-e] po-si- professor-HON-NOM.HON the paper-ACC read-HON-PST-E try-HONthesis concerns only SVCs in which the second verb in a sequence of two verbs is a typical control verb. 83 It is not clear how best to label these two suffixes. Some researchers call the suffixes complementizers (Sells 1998) or connectives (H.S. Lee 1991), and some other researchers (Martin 1992; Sohn 1994 among others) call the suffixes infinitive and gerundive, respectively. 191 ess-ta. PST-DC ‘The professor tried to read the paper.’ b. Emeni-nim-kkeyse [cip-ey ka-(*si)-(*ss)-ko] siph-(usi)-ess-ta. mother-HON-NOM.HON home-LOC go-HON-PST-KO want-HON-PST-DC ‘Mother wanted to go home.’ As shown in the above examples, the SVCs employ a sequence of verbs in which verbal inflections such a Tense, Agreement, and Mood are marked only on the second verb. Just looking at the English gloss, the second verbs of the SVCs above seem to be the same predicates found in finite control constructions. However, the second verbs in (38) are not attested as independent verbs. That is, pota ‘try’ in (38a) denotes ‘see’ when it is used as a main verb and siphta ‘want’ in (38b) cannot occur as an independent verb. The two verbs in the SVCs function as a single predicate denoting a single overall event. The following examples further illustrate the absence of Tense in the verbal complements. (39) a. *Ecey Con-i [nayil ttena-a] po-ass-ta. yesterday John-NOM tomorrow leave-A try-PST-DC ‘Yesterday, John tried to leave tomorrow.’ b. *Ecey Nay-ka [nayil yesterday I-NOM cip-ey ka-ko] siph-ess-ta. tomorrow home-LOC go-KO want-DC ‘Yesterday, I wanted to go home tomorrow.’ 192 The examples in (39) show that the embedded clauses cannot take a temporal adverb different from the matrix event. That is, the verbal complements, unlike sentential complements, cannot have even dependent Tense. Therefore, it is concluded that the verbal complements cannot be specified for Tense, Agreement, and Mood. Since they lack those features they cannot contain any functional head associated with the features. Hence, there can be no syntactic structure (i.e., Spec, FP) to receive nominative Case. Indeed, the following examples show that the SVCs at issue have only one overtly expressed syntactic subject. (40) a. Coni-i [*casini-i/*kui-ka ku chayk-ul ilk-e] po-ass-ta. John-NOM self-NOM/he-NOM the book-ACC read-E try-PST-DC ‘John tried to read the book.’ b. Nayi-ka [*casini-i/*nayi-ka I-NOM cip-ey ka-ko] siph-ess-ta. self-NOM/I-NOM home-LOC go-KO want-DC ‘I wanted to go hom.’ Unlike the sentential complements discussed in chapter 3, the SVCs in (40) do not allow an overt lexical subject in the complement. Gamerschlag (2007) argues that these types of constructions are instances of syntactically induced control and that this type of complement always involves obligatory subject control. Generally, SVCs require argument sharing between the serialized verbs due to simultaneous or immediately consecutive events denoted by the verbs of the SVCs. Therefore, although Gamerschlag 193 (2007) does not address this issue, the syntactic control constructions can be analyzed as mono-clausal in the sense of Wurmbrand (2001), which treats control complements as VP. In fact, there is some evidence that the complements in the SVCs in Korean are simply verb phrases, lacking an embedded subject position. Besides the unavailability of overt NP in the controllee position and lack of any Tense, Agreement, and Mood morphology, the SVCs differ from finite OC constructions in Korean in a number of ways. The next three subsections examine several syntactic phenomena, all of which point to a structural difference between the two constructions. 4.3.1.1. Clausemate Condition on NPIs One piece of evidence for the monoclausal status of the SVCs can be found by the clausemate condition on Negative Polarity Items (NPIs). NPIs in Korean, such as amwuto ‘nobody’, are construed with negation in a same clause, as shown in (41). (41) a. Amwuto ku salam-ul nobody cohaha-*(ci anh)-nun-ta. the man-ACC like-NEG-PRS-DC ‘Nobody likes the man.’ b. Minki-nun amwuto *(an) manna-ss-ta. M.-TOP nobody not meet-PST-DC ‘Minki met nobody.’ c. Minki-ka [amwuto hakkyo-ey o-ci anh-ass-ta]-ko malha-yess-ta. M.-NOM nobody school-LOC come-NEG-PST-DC-C tell-PST-DC ‘Minki told that nobody came to school.’ 194 d. *Minki-ka [amwuto hakkyo-ey wa-ss-ta]-ko malha-ci anh-ass-ta. M.-NOM nobody school-LOC come-PST-DC-C tell-NEG-PST-DC The NPIs amwuto ‘nobody’ in (41a,b) occur in a simple clause and are licensed by the long form and the short form negation attached to the verbs, respectively. In contrast, the NPI in (41c) occurs in the embedded clause with the long form negation and thus, the scope of the negation ranges over only the embedded clause. Notice that when the NPI is separated from a negative expression by a clause boundary as in (41d), the sentence is ill-formed. This demonstrates that the NPIs and negative expressions cooccur in the same clause. This clausemate condition on NPIs supports the mono-clausal status of serial verb constructions and the biclausal status of sentential complements. The following data show the behavior of NPIs in SVCs. (42) a. Amwuto ku nonmwun-ul ilk-e nobody the paper-ACC po-ci anh-ass-ta. read-E try-NEG-PST-DC ‘Nobody tried to read the paper.’ b. Amwuto ku nobody nonmwun-ul an ilk-e the paper-ACC po-ass-ta. not read-E try-PST-DC ‘Nobody tried to read the paper.’ c. Na-nun amwu chayk-to ilk-e I-TOP any po-ci anh-ass-ta. book-too read-E try-NEG-PST-DC ‘I didn’t try to read any book.’ 195 (43) a. Amwuto cip-ey ka-ko siph-ci anh-ass-ta. nobody home-LOC go-KO want-NEG-PST-DC ‘Nobody wanted to go home.’ b. Amwuto cip-ey an ka-ko siph-ess-ta. nobody home-LOC not go-KO want-PST-DC ‘Nobody wanted to go home.’ c. Na-nun amwuto po-ko I-TOP siph-ci anh-ass-ta. nobody see-KO want-NEG-PST-DC ‘I didn’t want to see anyone.’ The NPIs in the examples above are properly licensed by the long form or the short form negation, regardless of which verb the negation is attached to. In other words, the negation can be attached to either the first verb or the second verb and the scope of the negation is not changed. For instance, in (42a) the second verb po- ‘try’ is negated and in (42b) the first verb ilk- ‘read’ is negated; regardless, the subject NPI amwuto ‘nobody’ is licensed in each sentence and the sentences are completely synonymous. This provides evidence that the whole constructions is monoclausal just as the simplex clauses in (41a, b). In contrast, finite control complements behavior differently in terms of the scope of negation, as follows. (44) a. Amwuto [ec ku sang-ul nobody pat]-uleyko noleykha-ci anh-ass-ta. the prize-ACC receive-C ‘Nobody endeavored to receive the prize.’ endeavor-NEG-PST-DC 196 b. *Amwuto [ec ku sang-ul nobody pat-ci anh]-uleyko noleykha-yess-ta. the prize-ACC receive-NEG-C endeavor-PST-DC c. Con-i [ec amwuto an manna]-leyko noleykha-yess-ta. John-NOM nobody not meet-C endeavor-PST-DC ‘John endeavored not to meet anyone.’ (45) a. Amwuto Mina-eykey [ec hakkyo-ey ka]-tolok seltukha-ci anh-ass-ta. nobody M.-DAT schoo-LOC go-C persuade-NEG-PST-DC ‘Nobody persuaded Mina to go to school.’ b. *Amwuto Mina-eykey [ec hakkyo-ey nobody c. Con-i M.-DAT ka-ci anh]-tolok seltukha-yess-ta. schoo-LOC go-NEG-C persuade-PST-DC Mina-eykey [ec amwuto manna-ci anh]-tolok seltukha-yess-ta. John-NOM M.-DAT nobody meet-NEG-C persuade-PST-DC ‘John persuaded Mina not to meet anyone.’ The examples in (44) and (45) employ the so-called non-finite control constructions. The NPIs in (44a) and (45a) functioning as the matrix subjects are construed with the matrix verbs. As a result, the sentences meet the locality condition on NPIs. On the other hand, the sentences in (44b) and (45b) violate the locality condition and thus, are ruled out. Notice that the scope of the negation in (44a) and (45a) encompasses the matrix sentence, while that in (44c) and (45c) encompasses only the embedded clause. The same pattern occurs in ko-complement (46) and adnominalized complement (47) constructions. 197 (46) a. Amwuto Con-eykey [ec cenyek-ul nobody John-DAT sa-keyss-ta]-ko yaksokha-ci anh-ass-ta. dinner-ACC buy-VOL-DC-C promise-NEG-PST-DC ‘Nobody promised John to buy dinner.’ b. *Amwuto Con-eykey [ec cenyek-ul nobody John-DAT c. Con-i sa-ci anh-keyss-ta]-ko yaksokha-yess-ta. dinner-ACC buy-NEG-VOL-DC-C promise-PST-DC Mina-eykey [ec amwuto an manna-keyss-ta]-ko yaksokha-yess-ta. John-NOM M.-DAT nobody not meet-VOL-DC-C promise-PST-DC ‘John promised Mina not to meet anyone.’ (47) a. Amwuto Enjwu-eykey [ec ttena]-l nobody E.-DAT kes-ul yokwuha-ci anh-ass-ta. leave-ADN thing-ACC ask-NEG-PST-DC ‘Nobody asked Enjwu to leave.’ b. *Amwuto Enjwu-eykey [ec ttena-ci anh]-ul nobody E.-DAT kes-ul yokwuha-yess-ta. leave-NEG-ADN thing-ACC ask-PST-DC c. Minki-ka Enjwu-eykey [ec amwuto manna-ci anh]-ul kes-ul M.-NOM E.-DAT yokwuha- nobody meet-NEG-ADN thing-ACC ask- yess-ta. PST-DC ‘Minki asked Enjwu not to meet anyone.’ The locality condition on NPIs shows that unlike the SVCs which do not involve any embedding structures, the finite control constructions are bi-clausal. 198 4.3.1.2. Placement of Sentential Adverbs The second motivation for assuming that finite control complements are CPs while the SVCs are mono-clausal comes from placement of sentential adverbs. As shown in (48), sentential adverbs in Korean, such as cengmallo ‘really’, show free word-order. (48) a. Cengmallo Mina-ka ssawum-ul sicakha-yess-ta. really M.-NOM fight-ACC start-PST-DC ‘Mina really started a fight.’ b. Mina-ka cengmallo ssawum-ul sicakha-yess-ta. M.-NOM really c. Mina-ka fight-ACC start-PST-DC ssawum-ul cengmallo sicakha-yess-ta. M.-NOM fight-ACC really start-PST-DC Compared to (48a) and (48b), in which the sentential adverb cengmallo ‘really’ precedes the subject and the object, respectively, (48c) shows that the sentential adverb can occur between the object and the verb, just like VP adverbs that modify only verbs. However, (48c) does not trigger a difference in meaning and the adverb expresses the speaker’s certainty as to the truth of the whole sentence, just as it does in (48a) and (48b). The lack of meaning difference between these examples clearly shows that sentential adverbs distribute freely within a single clause. The same pattern occurs in the SVCs, as follows. 199 (49) a. Cengmallo Mina-ka really ku maswul-ul sihemha-ye po-ass-ta. M.-NOM the magic-ACC test-YE try-PST-DC ‘Mina really tried to test the magic.’ b. Mina-ka cengmallo ku maswul-ul M.-NOM really c. Mina-ka sihemha-ye po-ass-ta. the magic-ACC test-YE ku maswul-ul cengmallo sihemha-ye po-ass-ta. M.-NOM the magic-ACC really test-YE (50) a. Cengmallo Minki-ka Sanghwun-ul iki-ko Really try-PST-DC M.-NOM S.-ACC try-PST-DC siph-ess-ta. beat-KO want-PST-DC ‘Minki really wanted to beat Sanghwun.’ b. Minki-ka cengmallo Sanghwun-ul iki-ko M.-NOM really S.-ACC beat-KO want-PST-DC c. Minki-ka Sanghwun-ul cengmallo iki-ko M.-NOM S.-ACC really siph-ess-ta. siph-ess-ta. beat-KO want-PST-DC The examples in (49) and (50) show free word-order of the sentential adverb without changing the meaning of the sentence. That is, the sentential adverbs in (49c) and (50c) have scope over the whole sentence, even though they precede the first verb of the serial verbs and follow the object. If the SVCs were bi-clausal, it would be expected that the sentential adverb could only have embedded scope. This indicates that the SVCs are mono-clausal. Finite control constructions which have been treated as non-finite control constructions do not show the same behavior in terms of the scope of sentential adverbs, as follows. 200 (51) a. Cengmallo Mina-ka [ec ku sang-ul really M.-NOM pat]-uleyko noleykha-yess-ta. the prize-ACC receive-C endeavor-PST-DC ‘Mina really endeavored to receive the prize.’ b. Mina-ka cengmallo [ec ku sang-ul M.-NOM really pat]-uleyko noleykha-yess-ta. the prize-ACC receive-C endeavor-PST-DC ‘Mina really endeavored to receive the prize.’ c. Mina-ka [ec ku sang-ul M.-NOM cengmallo pat]-uleyko noleykha-yess-ta. the prize-ACC really receive-C endeavor-PST-DC ‘Mina endeavored to really receive the prize.’ (52) a. Cengmallo Minki-ka Sanghwun-eykey [ec ku sihap-eyse really M.-NOM S.-ACC ppaci]-tolok the match-from drop.out-C kangyoha-yess-ta. force-PST-DC ‘Minki really forced Sanghwun to drop out of the match.’ b. Minki-ka cengmallo Sanghwun-eykey [ec ku sihap-eyse M.-NOM really S.-DAT ppaci]-tolok the match-from drop.out-C kangyoha-yess-ta. force-PST-DC ‘Minki really forced Sanghwun to drop out of the match.’ c. Minki-ka Sanghwun-eykey [ec ku sihap-eyse cengmallo ppaci]-tolok M.-NOM S.-ACC the match-from really drop.out-C 201 kangyoha-yess-ta. force-PST-DC ‘Minki forced Sanghwun to really drop out of the match.’ The sentential adverbs in (51a, b) and (52a, b) occur in the matrix clauses and thus, scope over the matrix verbs. As opposed to this, the sentential adverbs in (51c) and (52c) occur in the embedded clauses and scope over only the embedded clauses. These different scope relations cannot be explained if the constructions are not bi-clausal. The following examples further show that unlike serial verb constructions, but like the so-called non-finite constructions in (51) and (52), finite control complements headed by the complementizer -ko and the adnominalizer -l can take sentential adverbs having embedded scope. (53) a. Hwaksilhi Minki-ka Con-eykey [ec ku sihap-eyse surely M.-NOM John-DAT iltung-ul ha-keyss-ta]- the match-LOC best-ACC do-VOL-DC- ko yaksokha-yess-ta. C promise-PST-DC ‘Minki surely promised John to win first place at the match.’ b. Minki-ka Con-eykey [ec ku sihap-eyse M.-NOM John-DAT hwaksilhi iltung-ul the match-LOC surely ha-keyss-ta]- best-ACC do-VOL-DC- ko yaksokha-yess-ta. C promise-PST-DC ‘Minki promised John to surely win first place at the match.’ 202 (54) a. Elisekkeyto Mina-ka [ec cip-ul stupidly M.-NOM ttena]-l kes-ul kyelsimha- home-ACC leave-ADN thing-ACC determine- yess-ta. PST-DC ‘Stupidly, Mina determined to leave home.’ b. Mina-ka [ec elisekkeyto cip-ul M.-NOM stupidly ttena]-l kes-ul kyelsimha- home-ACC leave-ADN thing-ACC determine- yess-ta. 84 PST-DC ‘Mina determined to leave home stupidly.’ While the sentential adverb hwaksilhi ‘surely’ in (53a) and (54a) scopes over the matrix predicate, that in (53b) and (54b) scopes over only the embedded clause. This indicates that the constructions are biclausal. All the examples above show that the SVCs behave differently from finite OC constructions in terms of the scope of sentential adverbs. Unlike finite OC constructions, the sentential adverbs intervening between the embedded object and the embedded predicate have scope over the matrix verbs. This clearly indicates that the SVCs are monoclausal. 84 As shown below, the sentential adverb can also occur between the embedded object and verb, possibly resulting in ambiguity of interpretation of the sentence. That is, the adverb can modify either the whole embedded clause, or just the embedded verb. In the latter case, the adverb is considered to have a manner reading and thus, the sentence would be semantically awkward, leading the sentence to be ungrammatical. Minai-ka [eci cip-ul elisekkeyto ttena-l] kes-ul kyelsimha-yess-ta. ‘Mina determined to leave home stupidly.’ 203 4.3.1.3 Scope of Wh-Adverbs The third motivation for the claim that unlike finite control complements the SVCs are mono-clausal is derived from different scope reading of wh-adverbs. (55) a. Mina-ka M.-NOM hakkyo-ey encey ka-ss-ni? school-LOC when go-PST-Q ‘When did Mina go to school?’ b. Mina-ka encey hakkyo-ey ka-ss-ni? M.-NOM when school-LOC go-PST-Q ‘When did Mina go to school?’ c. Encey Mina-ka hakkyo-ey ka-ss-ni? when M.-NOM school-LOC go-PST-Q ‘When did Mina go to school?’ (56) a. Mina-ka [ku chayk-ul encey ilk]-e po-ass-ni? M.-NOM the book-ACC when read-E try-PST-Q ‘When did Mina try to read the book?’ b. Encey Mina-ka [ku chayk-ul ilk]-e po-ass-ni? when M.-NOM the book-ACC read-E try-PST-Q ‘When did Mina endeavor to go to Paris?’ (57) a. Mina-ka [ppali-ey encey ka]-ko siph-ess-ni? M.-NOM Paris-LOC when go-KO want-PST-Q ‘When did Mina want to go to Paris?’ 204 b. Encey Mina-ka [ppali-ey ka]-ko siph-ess-ni? when M.-NOM Paris-LOC go-KO want-PST-Q ‘When did Mina want to go to Paris?’ As shown in (55), wh-adverbs can freely occur in a single clause with no difference in meaning. In (55a) encey ‘when’ occurs before the verb, in (55b) before the locative, and in (55c) clause initially. Similarly, the different placements of wh-adverbs in (56) and (57) do not change the meaning of the sentence; thus the placement induces no difference in scope relations. This suggests that the SVCs describe a single event, not a sequence of separate events. Tense marked only after the second verb has scope over the whole sentence and thus, the two verbs are interpreted as having the same Tense and as sharing the same subject. Therefore, wh-adverbs can freely occur in any position, just as in a single clause. Contrary to this, different placements of wh-adverbs in finite OC contexts establish different scope relations. (58) a. Mina-ka [ec encey ppali-ey M.-NOM ka]-leyko noleykha-yess-ni? when Paris-LOC go-C endeavor-PST-DC ‘Did Mina endeavor when to go to Paris?’ b. Encey Mina-ka [ec ppali-ey when M.-NOM ka]-leyko noleykha-yess-ni? Paris-LOC go-C endeavor-PST-DC ‘When did Mina endeavor to go to Paris?’ (59) a. Mina-ka Con-eykey [ec encey hakkyo-ey M.-NOM John-DAT ka]-tolok kanyoha-yess-ni? when school-LOC go-C force-PST-Q 205 ‘Did Mina force John when to go to school?’ b. Encey Mina-ka Con-eykey [ec hakkyo-ey ka]-tolok kanyoha-yess-ni? when M.-NOM John-DAT school-LOC go-C force-PST-Q ‘When did Mina force John to go to school?’ (58) and (59) are examples of the so-called non-finite control contructions. In (58a) and (59a), encey ‘when’ occurs in the complement clauses, and thus it has the embedded scope reading. In contrast, in (58b) and (59b), encey ‘when’ occurs before the matrix subject Mina and thus it has the matrix scope reading. That is, the different placement of wh-adverbs in the constructions induces differences in scope relations. The same pattern can be found in finite complements selected by the complementizer -ko and -l. (60) a. Mina-ka Con-eykey [ec encey ceneyk-ul sa-keyss-ta]-ko M.-NOM John-DAT when dinner yaksokha-yess-ni? buy-VOL-DC-C promise-PST-Q ‘Did Mina promise John when she would buy dinner?’ b. Encey Mina-ka Con-eykey [ec ceneyk-ul sa-keyss-ta]-ko when M.-NOM John-DAT dinner yaksokha-yess-ni? buy-VOL-DC-C promise-PST-Q ‘When did Mina promise John to buy dinner?’ (61) a. Mina-ka Con-eykey [ec encey ttena]-l M.-NOM John-DAT kes-ul yokwuha-yess-ni? when leave-ADN thing-ACC ask-PST-Q ‘Did Mina ask John when to go to school?’ 206 b. Encey Mina-ka Con-eykey [ec ttena]-l when M.-NOM John-DAT kes-ul yokwuha-yess-ni? leave-ADN thing-ACC ask-PST-Q ‘When did Mina ask John to leave?’ The examples in (60)-(61) show that wh-adverbs in finite OC constructions can only have either an embedded or a matrix scope reading, depending on their placement. This clearly demonstrates that those constructions are biclausal. To sum up, the SVCs at issue do not exhibit any of the syntactic behaviors that are shown in finite OC constructions. That is, the SVCs do not consist of two different TP clauses where an OC PRO can be involved, but are monoclausal. However, the size of the embedding structure of SVCs is not straightforward in terms of whether it is vP or VP. In fact, there have been competing proposals regarding the structure of OC which is considered to be monoclausal. According to Wurmbrand (1998, 2001, 2002) and Baltin (1995), OC infinitives are VP, lacking a small v projection. For Wurmbrand, OC is compatible with a subjectless, i.e., PRO-less infinitive, while Baltin (1995) argues that PRO is Caseless DP in Spec VP. On the other hand, Aoshima (2000, 2001) proposes that OC in Japanese cannot be smaller than vP, convincingly arguing that VP is not a size appropriate for OC complements. In the next section, I will provide some evidence that the embedded structure of the SVCs in Korean is VP, and so they can be analyzed as argument sharing, not as control constructions. 207 4.3.2. VP Analysis of Serial Verb Constructions Wurmbrand (1998, 2001, 2002, 2004) argues that restructuring infinitives, which crucially involve OC, do not contain any projection above the VP. That is, restructuring infinitives do not only lack CP and TP, but also vP in which an embedded subject is merged. This analysis is based mostly on the lack of clause-boundedness effects on restructuring infinitives in German, such as long-passive movement. As shown in (62b), she argues that matrix passivization in restructuring infinitives triggers an embedded object to undergo long-passive movement, which is unexpected under a clausal view of the infinitives. (62) a. weil Hans den Traktor since John zu reparieren versuchte. the tractor-ACC to repair tried ‘since John tried to repair the tractor.’ b. dass der Traktor wurde zu reparieren versucht. that the tractor-NOM was to repair tried ‘that they tried to repair the tractor.’ Wurmbrand argues that “long” object movement would not be possible if there were a subject position in the infinitives since the presence of a syntactic subject would block the A-movement operation. Thus, Wurmbrand proposes the syntactic structure for restructuring infinitives given in (63). 208 (63) TP Johni T' vP T the tractorj v' ti v' VP v VP tj V V tried to repair However, Aoshima (2000, 2001) challenges Wurmbrand’s VP analysis of restructuring infinitives, arguing that unlike the German example in (62b), a long passive cannot be formed in Japanese or English, as shown in (64). (64) *sono zitensya-ga naosi te mi-rare-ta. the bike-NOM repair to try-PASS-PST ‘*The bike was tried to repair.’ (Aoshima 2001:41) According to Aoshima (2001:42), the derivation of the Japanese and English long passive crashes simply because one argument is given incompatible external and internal θ- features by one transitive verb. With regard to the thematic relation between the embedded transitive verb repair and the argument the bike, the NP moved from the VP complement position to the Spec vP would get both Theme and Agent roles. Thus, 209 the control clause in (64) would derive an inappropriate interpretation such that the bike repairs itself. Therefore, Aoshima argues that the long passive in try type of control constructions in Japanese and English can be excluded if the control complement is assumed to involve vP, not VP. However, unlike the Japanese example in (64), but similar to the German in (62b), the SVCs in Korean in question allow the long passive, as shown in (65) and (66). (65) a. Sensayng-nim-kkeyse Bil-ul teacher-HON-NOM mopem-ulo sam-a po-si-ess-ta. Bill-ACC example-as make-A try-HON-PST-DC ‘The teacher tried to make Bill as an example.’ b. Bil-i mopem-ulo sam-a po-a-cyess-ta. Bill-NOM example-as make-A try-A-PASS.PST-DC ‘Bill was tried to make an example.’ (66) a. Mali-ka emeni-lul posalphi-ko siph-ess-ta. Mary-NOM mother-ACC take.care-KO want-PST-DC ‘Mary wanted to take care of the mother.’ b. Emeni-kkeyse (Mali-hanthey) posalphi-ko mother-HON.NOM Mary-DAT siph-e-ci-syess-ta. take.care-KO want-E-PASS-HON.PST-DC ‘The mother was wanted to take care of by Mary.’ (65b) and (66b) show that the embedded objects in (65a) and (66a) are assigned nominative Case and become the matrix subject via the long-passive movement. The loss of and the occurrence of the honorific agreement marker -si- on the second 210 predicate further indicates the subject status of the embedded object. Unlike the SVCs, long-passive movement is not possible in finite control constructions, given in (67) and (68). (67) a. Bil-i [pro Con-ul Bill-NOM ttalacap]-uleyko sitoha-yess-ta. John-ACC catch.up.with-C try-PST-DC ‘Bill tried to catch up with John.’ b. *Con-i ttalacap-uleyko sito-toy-ess-ta. John-NOM catch.up.with-C trial.PASS-PST-DC ‘John was tried to catch up with.’ (68) a. Mali-ka [pro Bil-ul Mart-NOM manna]-ki-lul wenha-yess-ta. Bill-ACC meet-NOML-ACC want-PST-DC ‘Mary wanted to meet Bill.’ b.*Bil-i manna-ki-lul wenhay-cyess-ta. Bill-NOM meet-NOML-ACC want-PASS.PST-DC ‘Bill was wanted to meet.’ The impossibility of long-passive movement in finite control constructions is predictable since finite control constructions, unlike the SVCs, involve an embedded subject. If the ability of long-passive movement explains the absence of an embedded subject, as Wurmbrand argues, the SVCs in Korean should be analyzed as a VP structure. 211 Wurmbrand (1998) gives further evidence that restructuring infinitives involving OC are always syntactically reduced to VP, based on the impossibility of a partial control reading. (69) a. *Der Bürgermeister versuchte [PROi+k sich the mayor tried in Schloβ zu versammeln]. SELF in the castle to gather ‘The mayor tried to gather in the castle.’ b. Eri beschhloβ [PROi+k sich im Schloβ He decided zu versammeln]. SELF in the castle to gather He decided to gather in the castle.’ (Wurmbrand 2002:5-6) Wurmbrand(1998) claims that a restructuring infinitive cannot take a partial control reading since it lacks a vP projection. Under her analysis, this explains the difference in grammaticality between (69a) and (69b). That is, given that partial control involves a matrix controller as well as anyone else salient from the discourse context, there must be a syntactic position for an embedded subject. Thus, partial control is not compatible with a restructuring infinitive which lacks a syntactic subject position. On the other hand, a non-restructuring infinitive, such as (69b), contains a vP projection, and thus allows partial control. Aoshima takes up this argument to claim that Wurmbrand’s VP analysis cannot account for OC complements in Japanese, based on data such as those in (70). 212 (70) a. *Sichyooi-ga PROi+k siyakusyo-ni atumat te mita. mayor-NOM city.hall-at gather to tried ‘*The mayor tried to gather in the city hall.' b. Emii-ga PROi+k issyoni Emi-NOM ringo-ga tabe-rare-ru. together apples-NOM eat-can-PRS ‘Emi can eat apples together.’ (Aoshima 2001:38) Under Wurmbrand’s (1998) analysis, both (70a) and (70b) take restructuring verbs, which trigger OC, and thus a partial control reading should not be permitted. However, Aoshima claims that the nominative object construction (70b), unlike the try type of verb (70a), takes a partial control reading, 85 and thus OC complements in Japanese should contain a vP projection which provides a subject position. If the possibility of a partial control reading can be considered evidence for the vP analysis of control complement, the following examples show that the SVCs in Korean at issue do not contain a vP. (71) a.*Bili-i PROi+j tosekwan-eyse-(tul) moi-e Bill-NOM library-LOC-PL po-ass-ta. gather-E try-PST-DC ‘*Bill tried to gather in the library.’ 85 In fact, the PC reading of (70b) is not so obvious since the sentence can be interpreted as follow. (i) Emii-ga (Kim-mo) issyoni ringo-ga tabe-rare-ru. Emi-NOM K.-and together apples-NOM eat-can-PRS ‘Emi can eat apples together (with Kim).’ The construction in (i) contains an omitted NP with comitative case, such as Kim-mo ‘with Kim’, and thus takes an exhaustive reading rather than a PC reading. 213 b.*Bili-i PROi+j tosekwan-eyse-(tul) moi-ko Bill-NOM library-LOC-PL siph-ess-ta. gather-KO want-PST-DC ‘Bill wanted to gather in the library.’ c. Minai-ka [eci+j tosekwan-eyse M.-NOM moi-keyss-ta]-ko yaksokha-yess-ta. library-LOC-PL gather-VOL-DC-C promise-PST-DC ‘Mina promised to gather in the library.’ Unlike finite OC in Korean, in which partial control can be allowed, as discussed in 4.2.2 and shown in (71c), the SVCs in (71a) and (71b) do not permit partial control. In this respect, SVCs in Korean pattern together with the German example in (69a) and the Japanese te construction in (70a), but not with the nominative object construction in (70b). As a result, the SVCs in Korean cannot be accounted for under the vP analysis, but can under the VP analysis. The following examples further show that there is no embedded subject position in the SVCs. (72) a. Kyoswu-nim-kkeyse chayk-ul ilk-usi-(ess)-na po-(*si)-ta. Professor-HON-NOM.HON book-ACC read-HON-PST-NA seem-HON-DC ‘It seems that the professor is reading/read a book.’ b. Kyoswu-nim-kkeyse chayk-ul ilk-(*usi)-e po-si-ess-ta. Professor-HON-NOM.HON book-ACC read-HON-E try-HON-PST-DC ‘The professor tried to read a book.’ 214 c. Kyoswu-nim-kkeyse chayk-ul ilk-(*usi)-ko siph-usi-ess-ta. Professor-HON-NOM.HON book-ACC read-HON-KO want-HON-PST-DC ‘The professor wanted to read a book The example in (72a) is equivalent to a Subject-to-Subject Raising construction in English. Note that the raising verb in (72a) cannot be followed by the subject agreement marker -si. This is due to the property of a raising predicate that cannot assign a thetarole to its subject. Instead, the agreement marker follows the embedded verb. The availability of the past Tense marker -ess on the embedded verb in (72a) indicates that there is a subject position in the clause. On the other hand, as briefly discussed in 4.3.1, embedded verbs in the SVCs cannot take the subject agreement marker -si. The examples in (72b) and (72c) show that when the subject is someone honorable, the agreement marker follows the second verb, not the first verb. This tells us that there is no subject postion for the embedded verb itself. Another piece of evidence that the SVCs in Korean should be analyzed as VP is derived from a function of the suffix between serial verbs. As discussed in 4.3.1, the suffixes -a/e and -ko do not carry any inherent Mood/Modal, Tense, or aspectual meaning, and they are simply attached to the stem of a main verb to link to an auxiliary verb. Moreover, recall that the embedded verb in the SVCs cannot take any Tense, Agreement, or Mood marker. In these respects, the SVCs in Korean are different from te and yooni constructions in Japanese which are considered to be bigger than vP in the literature (Aoshima 2000, 2001; Dubinsky 2007). 215 (73) a. John-ga [sono ronbun-o J.-NOM that yomi] sobireta. paper-ACC read failed ‘John failed to read that paper.’ b. John-ga [sono ringo-o J.-NOM that tabe-te] mita. apple-ACC eat-to tried ‘John tried to eat that apple.’ c. John-ga Mary-ni [ sono ringo-o tabe-ru]-yooni itta. J.-NOM Mary-DAT that apple-ACC eat-IMP-to told ‘John told Mary to eat that apple.’ According to Aoshima (2001), 86 te and yooni constructions as well as the compound verb constructions above are analyzed as monoclausal, lacking a TP projection and taking an OC reading. Consequently, this leads her to support Hornstein’s movement approach to OC constructions. However, Aoshima (2001:42-43) proposes a different size of structure for each construction, given below: (74) a. [TP NPi [vP ti [VP [PartP [vP ti [VP … V] v] te] V] v] T] 86 In fact, Aoshima (2000, 2001) divides Japanese control verbs into five types with respect to their selections of complements, as follows: (i) V-V compound verbs. (ii) Verbs followed by a particle te/de. (iii) Verbs followed by a subjunctive yooni. (iv) Verbs followed by a complementizer to, which heads a Modal verb. (v) Verbs followed by a koto complementizer. According to Aoshima, the cases in (i)-(iii) are OC constructions that invlove a monoclausal structure. 216 b. [TP NPi [vP ti [VP NPj [VP [PartP [vP tj [VP … V] v] te] V] v] T] c. [TP NPi [vP ti [VP NPj [MODP [vP tj [VP … V] v] yooni] V] v] T] According to Aoshima, te constructions trigger subject control (74a) or object control (74b) and te is analyzed as a gerund form of participle that heads a Participle projection (PartP) above an embedding vP. On the other hand, yooni in (74c) always triggers object control and is considered a subjunctive auxiliary verb that heads a MoodP. The structural difference between te and yooni constructions is based on the fact that unlike te constructions, yooni constructions can bear an imperfect aspect marker, as shown in (73c). Aoshima takes this as evidence supporting the claim that the VP analysis cannot account for the structure of OC constructions in Japanese since VP is too small to account for the aspect interpretation of the embedded verb in yooni constructions. However, the monoclausal property of yooni constructions is questionable, given the following examples from Dubinsky (2007:18-19). (75) a. Mary1 ni, Taroo zyanakute, kanozyo1-ga sono kaisya-o NI yoo-ni is.not she-NOM tyoosa-suru that company-ACC investigate-do meizita. manner-NI ordered ‘I ordered Mary that she, and not Taroo, should investigate that company.’ b. Mary1 ni, kanozyo1 zyanakute, Taroo-ga NI she is.not sono kaisya-o tyoosa-suru -NOM that company-ACC investigate-do 217 yoo-ni meizita. manner-NI ordered ‘I ordered Mary that Taroo, and not she, should investigate that company.’ The above examples show that under the right contrastive conditions, a resumptive pronoun can occur in yooni constructions (75a), and the subject of the constructions can be other than the matrix object (75b). Given this, yooni constructions can be analyzed as biclausal, not monoclausal. Furthermore, Dubinsky (2007) argues that -te is a “tenselike” affix, and thus -te heads an event-denoting TP or Event projection (EvP), which blocks movement of the subject into the matrix subject position. 87 Consider the following examples. 87 Dubinsky provides two types of evidence for the analysis, examining locative/manner adverbial clause headed by ni and ni-site. (i) a. Mari-wa [tue-o yoko ni] tatiagatta. TOP cane-ACC side at stood.up ‘Mari stood up, with the cane at [her] side.’ b. Mari1-wa neko-o (*kanozyo1-no) aite ni syokozi-o siteiru. TOP cat-ACC she-GEN companion at meal-ACC is.doing ‘Mari is eating a meal with a cat as her companion.’ (ii) a. Mari-wa [tue-o yoko ni site] tatiagatta. TOP cane-ACC side at do.TE stood.up ‘Mari stood up, with the cane at [her] side.’ Or ‘Mari stood up, having laid the cane flat [i.e. on its side].’ b. Mari1-wa neko-o (kanozyo1-no) aite ni site syokozi-o siteiru. TOP cat-ACC she-GEN companion at do.TE meal-ACC is.doing ‘Mari is eating a meal with a cat as her companion.’ (Dubinsky 2007:23-25) According to Dubinsky, contrary to bare ni adjuncts in (i), ni-site adjuncts in (ii) show that the ni-marked NP can be co-referential with other than the matrix argument and the possessor of the ni-marked NP can be overt. 218 (76) a. Tabete miru b. Tabete miseru eat.TE see eat.TE ‘eat and see’/ ‘try eating’ ‘eat and show’/‘demonstrate eating’ c. Iki-masyoo or go-TENTATIVE show Ik-oo go-TENTATIVE ‘Let’s go.’ (77) a. Tabe-tai. b. Tabete hosii. eat-want eat.TE want ‘I want to eat’ ‘I want him to eat.’ (Dubinsky 2007:21-22) According to Dubinsky, unlike bare verb phrases such as (76c), which are interpreted as part of the event/state denotation of the main clause, V-te phrases often denote ‘temporal sequence’ rather than any (in)dependent event/state, as shown in (76a) and (76b). Also, note that while the subject of tabe ‘eat’ in (77a) is syntactically controlled by the subject of tai ‘want’, the subject of the te phrase in (77b) is different from the subject of hosii ‘want’. This suggests that te constructions can be analyzed as biclausal with te heading a TP projection. Now, consider the following Korean examples in which two verbs are linked by the suffix -a/e and -ko, just as two verbs in Japanese te constructions are linked by -te. (78) a. Con-i koki-lul kwu-e/*ese po-ass-ta. John-NOM meat-ACC broil-E/ESE try-PST-DC ‘John tried to broil the meat.’ 219 b. Con-i koki-lul kwu-e/*ese tay-ess-ta. John-NOM meat-ACC broil-E/ESE supply-PST-DC ‘John kept broiling the meat.’ c. Con-i koki-lul kwu-e-/ese mek-ess-ta. John-NOM meat-ACC broil-E/ESE eat-PST-DC ‘John broiled the meat and then ate it.’ (79) a. Con-i thayksi-lul tha-ko/*ese John-NOM taxi-ACC siph-ess-ta. take-KO/ESE want-PST-DC ‘John wanted to take a taxi.’ b. Con-i thayksi-lul tha-ko/*ese John-NOM taxi-ACC iss-ess-ta. take-KO/ESE exist-PST-DC ‘John was taking/getting in a taxi’ c. Con-i thayksi-lul tha-ko/se ka-ss-ta. John-NOM taxi-ACC take-KO/SE go-PST-DC ‘John went by taxi/took a taxi and then went.’ The verbal complements in (78) bear the same suffix -e, but they express different eventualities. The event of the verbal complements in (78a) and (78b) cannot be separated from the event denoted by the second verb, and thus the constructions are perceived as denoting a single event. Contrary to this, (78c) denotes a sequence of events. In other words, the event of the verbal complement in (78c) denotes a completed event followed by an immediate action denoted by the second verb. In this respect, the former is equivalent to the compound verb construction in Japanese, while the latter is 220 equivalent to Japanese te constructions. This difference is validated by the (un)availability of the suffix -(e)se ‘and then’ which is attached to a subordinate clause. 88 Similarly, (79a) and (79b) denote a single event and thus, cannot take the subordinating suffix, whereas (79c) denotes two separate events and thus, the subordinating suffix is available. 89 If the suffixes -a/e and -ko 90 carry any inherent Tense or aspectual meaning, the difference in event property between the examples above cannot be explained. In fact, it is assumed that the suffixes are simply connectives which link two verbs and that the difference in event property of SVCs is derived from a distinct structure formed by the serial verbs. Note that the examples in (78c) and (79c) are composed of fully autonomous verbs, while the examples in (78a, b) 88 This is illustrated in the following example: (i) Con-i cip-ey ka-se swukcey-lul ha-yess-ta. John-NOM home-LOC go-and.then homework-ACC do-PST-DC ‘John went home and then did the homework.’ 89 The following examples further demonstrate that the constructions in (78c) and (79c) can denote two separate events. (i) a. Con-i koki-lul kwu-e manul-kwa hamkkey mek-ess-ta. John-NOM meat-ACC broil-E galic-and together eat-PST-DC ‘John broiled the meat and then ate it with a galic.’ b. Con-i thayksi-lul tha-ko cip-ulo ka-ss-ta. John-NOM taxi-ACC take-KO home-DR go-PST-DC ‘John went home by taxi.’ In addition, the serial verbs in the constructions can be freely separated by many elements, such as the emphatic particles -man/nun ‘only’ and a variety of the subordinating markers: -(u)myense ‘while’, -kose ‘after’, and -camaca ‘as soon as’, while the serial verbs in (78a, b) and (79a, b) can be separated only by the emphatic particles. 90 Sells (1998:2) points out that the suffix in complex predicates in Korean is specific to the particular complex predicate. Following Sells, I assume that the second verb in SVCs selects the suffix in the lexicon. That is, po- ‘try’ always selects -a/e and siph‘want’ always selects -ko. 221 and (79a, b) are composed of a main verb and an auxiliary verb. That is, unlike the second verb in the former examples which functions as an independent predicate, the second verb in the latter examples functions as an auxiliary, modifing the meaning of the first verb with a meaning such as ‘try’ and aspectual meaning. In fact, SVCs in Korean, which are seemingly identical to each other in terms of the form of [V1-e/ko V2], can be divided into three types, depending on the syntactic and semantic functions of the second verb. The first type of SVC consists of two independent lexical verbs, as in (78c) and (79c). In this case, the phrases with the suffixes usually denote a completed event or a means for achieving an event denoted by the second verb. That is, this type of SVC is perceived of as comprising separate actions. Thus, each verb selects its own internal argument structure, although it involves an external argument sharing. Simplified structures for (78c) and (79c) are given in (80). (80) a. vP tSubj b. v' VP vP tSubj v v' VP VP VP VP obj V1 pro V2 obj V1 v VP V2 The second type of SVC consists of a main verb and an aspectual verb, as in (78b) and (79b). The second verb in this type of SVC is unable to function as a predicate, but adds an aspectual meaning to the first verb. 91 That is, the second verb does not bear a 91 The second verbs in this type of SVC can be used as main verbs as in (i): 222 theta-role and simply heads an Aspect Phrase (AspP) above vP. 92 As a result, the first verb in this type of SVC functions as a predicate, and thus determines the argument structure of the serial verbs. The third type of SVC consists of a main verb and an auxiliary verb, as in (78a) and (79a). Unlike the aspectual verbs in the second type of SVC, the auxiliary verb in this type establishes a thematic relation with an external argument. 93 Thus, it can select its own external argument that is shared with the first verb. Simplified structures for these two types of SVCs are represented in (81). (i) a. Mina-ka ku chayk-ul peli-ess-ta. M.-NOM the book-ACC throw.away-PST-DC ‘Mina threw away the book.’ b. Ku-ka ku kakey-ey mulken-ul tay-n-ta. he-NOM the store-LOC stuff-ACC supply-PRS-DC ‘He supplies stuff to the store.’ c. Minswu-nun cip-ey iss-ta. M.-TOP house-LOC exist-DC ‘Minswu is at home.’ 92 A functional head analysis for aspectual verbs has been proposed in the literature. See Wurmbrand (2001) for German, Cinque (2003) for Romance languages, and Fukuda (2006) for Japanese. 93 Unlike raising verbs and aspectual verbs of SVCs, the auxiliary verbs impose selectional restrictions on their subjects, as shown below: (i) a. Pi-ka wa-ss-na po-ta. rain-NOM come-PST-NA seem-DC ‘It seems raining.’ b. Pi-ka nayli-e peli-ess-ta. rain-NOM come-E throw.away-PST-DC ‘It rained/has rained.’ (ii) a. Mina/*pi-ka wa-a po-ass-ta. M./rain-NOM come-A try-PST-DC ‘Mina/*rain tried to come.’ b. Mina/*pi-ka o-ko siph-ess-ta. M./rain-NOM come-KO want-PST-DC ‘Mina/*rain wanted to come.’ 223 (81) a. AspP b. Asp' vP tSubj tSubj Asp v' VP vP v' VP v V2 V' v VP V1 V2 V' NP V1 Note that the aspectual verb in (81a) heads vP, while the second verbs in (81b) heads VP. This structural difference explains the impossibility of long-passive movement in the former construction, as shown below. (82) a. Koyangi-ka cwui-lul cat-NOM cap-a peli-ess-ta. mouse-ACC catch-A throw.away-PST-DC ‘The cat has caught/caught the mouse.’ b. *Cwui-ka koyangi-eykey cap-a mouse-NOM cat-DAT peli-e ci-ess-ta. catch-A throw.away-PASS-PST-DC ‘The mouse was caught by the cat.’ c. Cwui-ka koyangi-eykey cap-hi-e mouse-NOM cat-DAT peli-ess-ta. catch-PASS-E throw.away-PST-DC ‘The mouse was caught by the cat.’ 224 Unlike the SVCs in which the second verb is the so-called control verb, long-passive movement is blocked in the SVCs in which the second verbs are aspectual verbs, as shown in (82b). Passivization of an embedded object is possible only when the passive morpheme occurs on the first verb, as in (82c). So far, it has been argued that the Korean SVCs at issue differ from the Japanese OC constructions which have been considered to be monoclausal, but behave the same as the restructuring OC infinitives in German in terms of the availability of long-passive movement and the unavailability of a PC reading. Therefore, the Korean SVCs are analyzed as having a VP structure. However, they are not considered control constructions. According to Wumbrand (2001, 2002), at least some instances of OC involve structure sharing, resulting in subjectless infinitives. Under the Wumbrand (2001, 2002)’s analysis, the Korean SVCs should be analyzed as OC. However, note that the SVCs in Korean and the restructuring infinitives in German do not display a ‘control’ relation between a controller and a controllee since the verbal complements lack a subject position. In those constructions, the subjects are simply the same arguments of two different predicates. Therefore, given the very definition of control – the coreferent relation between the missing subject of the embedded structure and a matrix argument, there is no motivation for the control analysis of the SVCs. In other words, the control analysis of the SVCs in Korean and the restructuring infinitives in German are not theoretically consistent with control constructions since the external argument sharing in the constructions is not mediated by an empty category. In this chapter, it has been claimed that OC with a pro subject in a finite context can be attested in Korean and that serial verb constructions do not exhibit any of the 225 syntactic behaviors that are shown in finite OC constructions and thus, are analyzed as monoclausal. Given the control facts, the standard theory of control establishing only a PRO controllee needs to be refined. One implication of the pro analysis of finite control in Korean is that control cannot be adequately accounted for based purely on syntax, which is assumed to be the case in many languages which in general depend on finiteness of control complements: i.e., a PRO occurs in a nonfinite context, but not in a finite context. Furthermore, if it is the case that OC PRO can appear in a finite context, as claimed by Landau (2000, 2004) for Hebrew and Balkan languages, Case and finiteness cannot distinguish the distribution of a PRO and a pro/lexical DP. This can also be invoked as evidence for the absence of syntactic control. In the next chapter, I will address the issue of what determines controllers and argue for the role of semantics and pragmatics in determining the controller in Korean OC constructions. 226 CHAPTER 5 CONTROLLER CHOICE IN OBLIGATORY CONTROL IN KOREAN One of most important issues in accounting for control phenomena is identifying the controller and explaining its relation to the controllee. Regarding this issue, there have been purely syntactic approaches to control (Hornstein 1999; Boeckx & Hornstein 2003) and purely semantic approaches (Culicover & Jackendoff 2001, 2006). Still other researchers have proposed both syntactically and semantically based accounts of control. For example, Landau (2000) argues that the Obligatory vs. Non-Obligatory Control distinction is syntactic rather than semantic: i.e., VP-internal infinitives trigger OC and VP-external infinitives allow NOC, while the exhaustive vs. partial control distinction is based on semantic differences. In a similar vein, Wurmbrand (2002) argues that NOC constructions are clauses syntactically and propositions semantically, while OC constructions are VP, and are thus determined semantically. However, in the case of Korean, the OC vs. NOC distinction as well as exhaustive vs. Obligatory Control distinction is more attributable to a semantic rather than a syntactic distinction in that OC constructions in Korean are analyzed as CP constructions just as NOC constructions are, as already discussed in the previous chapters. In this chapter, I will discuss some issues regarding OC constructions in Korean which can be problematic for the purely syntactic based analyses of control in the literature and argue for the key role of semantics in determining the controller in OC constructions in Korean. However, unlike the previous studies (Chierchia 1983, 1984; Farkas 1988; Pollard & Sag 1991; Culicover & Jackendoff 2001, 2006, among others) proposing that OC is determined by the lexical semantics of the matrix predicates 227 involved, I argue that OC in Korean cannot be completely explained by the lexical properties of a selecting predicate alone. Rather, it is derived through a more complicated array of factors. In the literature on Greek control, there has been a proposal that control may be linked not only to the lexical properties of the selecting predicates, but also to some other factors, such as the combination of the lexical or inflectional properties of the selecting predicate along with a complement clause (i.e., Roussou 2007). Similarly, OC in Korean can be enforced by the lexical properties of the selecting predicates/complementizers, the combination of a matrix predicate and complement type, or syntactic/semantic properties of complement clauses. 5.1. Problems for Syntactic Analyses of Obligatory Control In this section, I will examine some syntactic issues concerning OC constructions in Korean that raise problems for syntactic analyses of control in the literature and argue that syntactic properties are not sufficient to determine controller choice in Korean. 5.1.1. Locality and C-Command Syntactic analyses of control attempt to explain control relations based on properties of structural configurations. That is, traditional syntactic approaches to control within GB frameworks, such as Williams (1980), argue that an empty subject in OC complements must be controlled by the closest c-commanding matrix argument, using the Minimal Distance Principle (MDP) proposed by Rosenbaum (1967). This strictly local configuration between a controller and a controllee has also been accounted for in the movement analyses of control by the Minimal Link Condition (MLC), which prohibits an element from moving across another element with a similar feature specification. In short, these syntactic constraints are taken to be the crucial 228 factor determining controller choice in the syntactic approaches to control. In most cases, this is true of OC in Korean, as shown in (1). (1) a. Minai-ka [proi/*j ttena]-leyko noleykha-yess-ta. M.-NOM leave-C try-PST-DC ‘Mina tried to leave.’ b. [Minai-uy emma]j-ka [pro*i/j kunyecasin*i/j-ul salangha]-leyko noleykha-yess-ta. M.-GEN mom-NOM herself-ACC love-INT-C try-PST-DC ‘Mina’s mom tried to love herself.’ c. Minai-ka Pataj-lul [pro*i/j ttena]-tolok seltukha-yess-ta. M.-NOM P.-ACC leave-C persuade-PST-DC ‘Mina persuaded Pata to leave.’ d. [Mina-uy emma]i-ka [Pataj-uy emma]k-eykey [pro*i/*j/k cakicasini/*j/k-ul M.-GEN mom-NOM P.-GEN mom-DAT self-ACC chingchanha-la]-ko malha-yess-ta praise-IMP-C tell-PST-DC ‘Mina’s mom told Pata’s mom to praise herself.’ The matrix subject DP in (1a, b) and the matrix object DP in (1c, d) are the closest ccommanding matrix arguments, so they control the controllees. However, Landau (1999, 2000) points out a serious shortcoming of the MDP and the MLC, based on wellknown empirical evidence, as found in (2). 229 (2) a. Johni promised Maryj [PROi to leave]. b. Johni asked Maryj [PROi to be allowed to leave]. In the examples, John and Mary both c-command the empty category, but the controller is the matrix subject John, not the matrix object Mary which is the closest ccommanding argument. Landau (2000:31) even claims that c-command is not a necessary condition for OC, providing the following examples in (3). (3) a. Yesterday, it spoiled Mary’si mood [PROi/*arb to listen to the news]. b. [PROi/*Bill’s/hisi/*j making that comment] was very rude of Johni. Similarly, OC constructions in Korean show that controllers do not always need to be in c-command relationships with the controllees. Consider the following examples. (4) a. Minai-ka Pataj-wa [proi+j/*??i/*j twu-si-ey M.-NOM P.-with Wujin-ul manna]-l kes-ul two-time-at W.-ACC meet-ADN thing-ACC yaksokha-yess-ta. promise-PST-DC ‘Minai promised to Pataj that shei would meet Wujin at two with herj.’ b. Minai-ka [proi/i+j twu-si-ey M.-NOM yess-ta. PST-DC Wujin-ul manna]-ki-lo Pataj-wa yaksokha- two-time-at W.-ACC meet-NOML-DR P.-with promise- 230 i) ‘Mina promised to Pata to meet Wujin at two.’ ii) ‘Minai promised to Pataj that shei would meet Wujin at two with herj.’ c. Minai-ka [proj hakkyo-ey M.-NOM ka]-tolok Pataj-eykey seltukha-yess-ta. school-LOC go-C P.-DAT persuade-PST-DC ‘Mina persuaded Pata to go to school.’ (4a) shows split control in which two matrix arguments, the subject Mina and the oblique Pata, jointly control the controllee. Neither the oblique DP nor the matrix subject can exclusively control the controllee. 94 Compared to (4a), (4b) can take subject control as well as split control. The possibility of subject control in (4b) seems to be explained simply by the c-command and precedence requirement. 95 However, in the case of split control in the examples, the controllee includes the matrix subject as well as the oblique DP which is not in a c-commanding position. Similarly, the embedded null subject in (4c) must refer to the matrix object, Pata, which fails to c-command the embedded clause. A simplified structure for (4a) and (4c) are represented in (5). 94 In fact, (4a) is ambiguous in that the oblique DP can be either in the matrix clause or in the embedded clause. However, (4b) gives evidence supporting the oblique DP as an element of the matrix clause. When the oblique DP belongs to the embedded clause, the sentence means ‘Mina promised (someone) to meet Wujin at two with Pata’. 95 As discussed in 1.2.1.1, Williams (1980) proposes that the antecedent precedes the controlled PRO. 231 (5) a. vP b. NP vP NP Minai-ka VP v Minai-ka VP V' PP NP P V' V' NP Pataj -wa CP v CP V N yaksokha- proi+j ... mannal kes-ul V' proj ... ka-tolok NP Pataj PP V P seltukha- eykey The structural representations in (5) clearly show that OC in Korean does not obey strict c-command. Specifically, the oblique argument Pata in (5a) does not c-command the null subject of the complement clause. Thus, it is obvious that a solely syntactic account of control based on the strict locality conditions cannot account for the Korean data. Hashemipour (1989) proposes that the antecedent of the controlled NP must be in the superordinate clause immediately dominating the controlled clause in Persian; thus, long-distance control is not possible. This also applies to OC in Korean. In determining the controller in OC, syntax simply plays the role of delimiting a potential domain for control; namely, the domain of an immediately higher clause of a controlled clause. However, within that domain, the licensing mechanism for controllees is not syntactic. Rather, the choice between subject and object control is sensitive to semantic or pragmatic factors, as it will be discussed in 5.2. 232 In addition, recall that movement analyses of control necessitate a non-CP (i.e., TP/IP) status of control complements since the phasal status of CP blocks A-movement of a DP out of the phase. Therefore, any syntactic analysis based on movement cannot account for the CP status of the embedded clause in finite OC in Korean. 96 5.1.2. Tense, and Agreement Although Landau (2000) argues that the exhaustive vs. partial control distinction is based on semantic differences, he proposes a syntactic analysis of control (i.e., Agree-based analysis), employing two crucial syntactic features, Tense and Agreement. Under his analysis, infinitives combining with implicative, aspectual and Modal predicates yield EC, whereas infinitives with, for instance, desiderative and interrogative predicates allow PC. However, he proposes that Tense is the crucial factor distinguishing PC from EC: PC complements involve their own semantic Tense, whereas EC complements contain no such independent Tense, as shown in (6). (6) a. *Yesterday, Johni began to PROi solve the problem, tomorrow. (EC) b. Yesterday, Johni wondered how to PROi+ solve the problem, tomorrow. (PC) Unlike (6a), (6b) shows the possibility of conflicting temporal adverbs between the matrix clause and the infinitive clause. Landau reflects on this contrast in the derivation of EC and PC to explain their interpretational differences, as illustrated in (7). 96 Madigan (2008) also points out some shortcomings of the movement theory of control (Hornstein 1999 and subsequent) in explaining partial and split control in Korean. 233 (7) a. FP F' F VP DP Agree2 V' V CP C TP CT PRO T-Agr Agree3 T tT-Agr VP Move tPRO V' Agree1 b. FP F' F VP DP Agree2 V' V CP C TP PRO Agree3 T T-Agr VP tPRO V' Agree1 For the derivation, Landau (2000:66) stipulates that DP and PRO are inherently [+SP] ‘Semantically plural’ or [-SP], F(unctional) head and Agr are initially unspecified for 234 semantic plurality, [ØSP], and acquire an SP value under agreement with DP, though not with PRO. Following Pesetsky & Torrego (2000), Landau also assumes that C contains an uninterpretable T(ense)-feature in a tensed clause, which triggers an overt T-to-C movement. This is the crucial factor distinguishing PC from EC. That is, the derivations explain how PRO can have a different semantic number than its controller in a PC context, but not in an EC context. In the case of PC in (6b) shown in (7a), first, the [+SP] PRO agrees with [ØSP] T-Agr. Next, the T-Agr which still remains unspecified for [SP] since PRO cannot value it moves to C to check off uninterpretable Tense feature on C. Then, the [ØSP] F agrees with the [-SP] DP and acquires [-SP]. Finally, the [-SP] F agrees with the [ØSP] T-Agr under the assumption that [-SP] and [ØSP] are non-distinct on functional heads. Therefore, this derivation allows a [+SP] controllee to have a [-SP] controller. Contrary to the PC derivation, the derivation of EC in (7b) does not include a T-to-C movement since EC complements are not tensed. Thus, PRO and F directly fall in an Agree relation. If PRO matches with F in SP, then the derivation will converge, but if PRO and F are mismatched in SP, then the derivation will crash. However, there are good reasons to claim that the Agree-based analysis cannot account for OC in Korean. First and foremost, Tense cannot distinguish EC from PC in Korean since they both occur in [+Tense] contexts in the sense of Landau (2000). (8) a. Ecey Minai-ka Pataj-eykey [proi/*j nayil yesterday M.-NOM P.-DAT cenyek-ul sa-keyss-ta]- tomorrow dinner-ACC buy-VOL-DC- 235 ko yaksokha-yess-ta. C promise-PST-DC ‘Yesterday, Mina promised Pata to buy dinner tomorrow.’ b. Ecey Minai-ka Pataj-lul [pro*i/j nayil yesterday M.-NOM P.-ACC ttena]-tolok seltukha-yess-ta. tomorrow leave-C persuade-PST-DC ‘Yesterday, Mina persuaded Pata to leave tomorrow.’ (9) a. Ecey Minai-ka [nayil proi+ tosekwan-eyse moi-keyss-ta]-ko yesterday M.-NOM tomorrow library-LOC gather-VOL-DC-C yaksokha-yess-ta. promise-PST-DC ‘Yesterday, Mina promised that ec i+ would gather at the library tomorrow.’ b. Ecey Minai-ka Pataj-eykey [proj+ nayil yesterday M.-NOM P.-DAT tosekwan-eyse-tul moi]- tomorrow library-LOC-PL gather- tolok seltukha-yess-ta. C persuade-PST-DC ‘Yesteday, Mina persuaded Pata to gather at the library tomorrow.’ The examples in (8) and (9) are instances of EC and PC, respectively. According to Landau’s (2000) analysis, only the examples in (9) should be tensed since they are PC, not EC. However, contrary to English, EC in Korean does not differ from PC from the standpoint that its complements have dependent Tense just as PC complements do. Another piece of evidence supporting the claim that the Agree-based analysis cannot account for OC in Korean derives from the fact that the null subjects of the 236 embedded clauses in Korean OC contexts do not need to match the controllers in Agree features, as already discussed in 4.2.2. The relevant examples are repeated below. (10) a. Minai-ka [proi+ tosekwan-eyse-(tul) moi]-ki-lo M.-NOM library-LOC-PL yaksokha-yess-ta. gather-NOML-DR promise-PST-DC ‘Mina promised to gather in the library.’ b. Minai-ka [caki-*(tul)i+-i tosekwan-eyse-(tul) moi-keyss-ta]-ko M.-NOM self-(PL)-NOM library-LOC-PL yaksokha- gather-VOL-DC-C promise- yess-ta. PST-DC ‘Minai promised that theyi+ , themselves, would gather in the library.’ According to Landau (2000, 2004), PRO in OC structures is ‘active’ for Agree due to its anaphoric nature, and thus PRO in tensed complements inherits all phi-features from the controller, although not necessarily semantic singularity. However, the examples in (10) show that a controllee in the Korean PC context does not inherit the syntactic singularity of its controller. That is, the null subject in (10a) cannot be exhaustively controlled by the matrix singular subject Mary since the predicate moi- ‘gather’, which is a semantically plural predicate, requires a semantically or syntactically plural subject. The availability of a plural marker -tul on the postpositional phrase tosekwan-eyse shows the plural properties of the null subject. The ungrammaticality of the omission of the plural marker in (10b) clearly demonstrates that the controllee in this context must be plural. Consequently, this evidence indicates that null subjects in Korean PC contexts 237 do not need to match the syntactic plurality of the controllers, contrary to what Landau (2004) suggests for English PC. The examples in (11) further indicate that even in EC contexts, the null subject does not need to inherit the [+honorific] feature from the controller. (11) a. Apecii-kkeyse halapenimj-kkey [proi/*j cip-ey ka-(*si)-keyss- father-NOM.HON grandfather.HON-DAT.HON home-LOC go-HON-VOLta]-ko yaksokha-si-ess-ta. DC-C propmise-HON-PST-DC ‘Father promised the grandfather to go home.’ b. Halapecii-kkeyse apecij-kkey [pro*i/j ku il-ul grandfather-NOM.HON father-DAT.HON kumantwu-(si)-la]- the job-ACC quit-HON-IMP- ko seltukha-si-ess-ta. C persuade-HON-PST-DC ‘Grandfather persuaded father to quit the job.’ The controllers of the embedded subjects in the above examples are individuals of honorific status relative to the speaker. However, the honorific marking on the embedded verb in (11a) actually leads the sentence to be ungrammatical and that in (11b) is not obligatory. These empirical data are also problematic for Madigan (2008)’s analysis in which controller choice is determined by a syntactic agreement relationship between PRO and the Speaker/Addressee head that carries the presuppositional features speaker and addressee. Madigan argues that the Modal and Mood markers -keyss and - 238 la, which have reference to a speech act, trigger an OC reading and posits that there is a functional category Speaker/Addressee whose head checks syntactic features Speaker and Addressee that can be further divided into specific Mood-type features, such as [±volitional] and [±required]. Under his mechanism, the derivation for a subject control that involves -keyss as in (11a) is represented below. (12) Addressee/SpeakerP MP Addressee/Speaker [+speaker] IP M [+volitional] Spell out as -keyss PRO [+speaker] Agree Following Kratzer (2006) in which a bound variable or a minimal pronoun comes into a derivation with minimal feature specifications and acquires its features via local chains of agreement with its antecedent, Madigan assumes that PRO is a minimal pronoun and thus inherits its features from its controller. This assumption is apparently verified by the following data which show subject-honorific agreement. (13) a. Sensayng-nimi-i haksayng-tul-eykey [PROi yeset-si-ey ttena-si-keyss- teacher-HON-NOM student-PL-DAT ta]-ko yaksokha-si-yess-ta. DC-C promise-HON-PST-DC six-time-at leave-HON-VOL- 239 ‘The teacher promised the students to leave at 6.’ b. Kim sensayng-nimi-i K. Pak sensayng-nimj-eykey [PROi+j teacher-HON-NOM P. tungsan-ul teacher-HON-DAT ha-si-ca]-ko ceyanha-si-ess-ta. mountain.climbing- ACC do-HON-EXH-C propose-HON-PST-DC ‘Teacher Kim proposed teacher Pak to go to mountain climbing.’ c. Hwuni-i Inhoj-eykey [PROi+j kongpwu-ha-ca]-ko yaksokha-yess-ta. H.-NOM I.-DAT study-do-EXH-C promise-PST-DC ‘(lit.) Hwun promised Inho to study together.’ The data from Madigan (2008) above show that the honorific marker -si occurs on the verb when the subject is a person who deserves the speaker’s deference. That is, the honorific marking on the matrix verbs in (13a, b) agrees with the matrix honorific subject. Thus, it is reasonable to think that the honorific marking on the embedded verbs in (13a, b) must agree with the PRO subjects. This leads Madigan to conclude that the honorific marker -si in Korean must be marked on a verb whose subject is a person of honorific status relative to the speaker and that PRO inherits its honorific feature from its controller. However, his analysis cannot account for the data in (11). If his analysis were correct, the occurrence of the honorific marker -si in (11) should be obligatory since the speaker in (11a) and the addressee in (11b) are the embedded subjects who are individuals of honorific status relative to the speaker. However, the honorific marking 240 on the embedded verb in (11a) actually leads the sentence to be ungrammatical and that in (11b) is not obligatory. 97 The following data raise a further problem for his analysis. (14) a. Minai-ka Kim sensayng-nimj-kkey [proi+j cip-ey M.-NOM K. teacher-HON-DAT.HON ka-*(si)-ca]-ko home-LOC go-HON-EXH-C ceyanha-yess-ta. propose-PST-DC ‘Mina proposed teacher Kim to go home.’ 97 As a matter of fact, the occurrence of the honorific marker in (13a) and (13b) is not obligatory as Madigan argues. Subject-honorific agreement in Korean is very intriguing regarding the proper usage of the honorific marker in relation to its trigger. In general, there are two main dimensions of honorifics: referent honorifics (speaker-referent perspective) and addressee honorifics (speaker-addressee perspective). The former concerns the speaker’s regard for a referent denoted by a nominal appearing overtly or covertly in a sentence, whereas the latter is associated with the speaker’s regard for the addressee in a speech situation. In an actual speech situation, the addressee honorifics are more important and thus more finely segmented than referent honorifics in view of the subtlety of interpersonal feelings and ‘face’ involved in face-to-face interactions (Sohn 1999:414). The referent honorifics relevant to the examples above are widely understood to occur when the subject denotes an individual who deserves the speaker’s deference. However, the referent honorifics are more finely segmented than the general view of the honorifics. For instance, the referent honorifics are also triggered by the relationship between the referents: the subject ‘the speaker’ and the indirect object ‘the addressee’ in the sentence. (11a) is one of the typical examples. That is, the matrix subject apeci ‘father’ is inferior to the matrix indirect object halapenim ‘grandfather’ and thus the honorific marking on the verb is not allowed, although the matrix subject is a person of honorific status relative to the speaker. Also, there is one important restriction on referent honorifics. When a speaker describes his/her own action, the honorific marking on the verb must be dropped, although the speaker is a person of honorific status relative to the addressee. For example, in (13a) the subject sensayngnim ‘teacher’ is superior to haksayngtul ‘students’ and the speaker in status, but since the embedded subject is the speaker of the embedded sentence, the honorific marking on the embedded verb is ungrammatical (unless the speaker intends to honorify himself/herself). However, Korean people commonly misuse or overuse the honorific marker in this case due to the honorific status of the subject relative to the speaker/writer of the whole sentence. 241 b. Kim sensayng-nimi-i K. Pataj-eykey [proi+j cip-ey teacher-HON-NOM P.-DAT ka-(*si)-ca]-ko hom-LOC go-HON-EXH-C ceyanha-si-ess-ta. propose-HON-PST-DC ‘Teacher Kim proposed Pata to go home.’ c. Halapecii-kkeyse apecij-kkey [pro*i/j yehayng ka-(*si)-ca]-ko grandfatehr-NOM.HON father-DAT.HON travel go-HON-EXH-C yaksokha-si-ess-ta. promise-HON-PST-DC ‘Grandfather promised father to go travel together.’ All of the examples in (13b, c) and (14) show split control involving the exhortative marker -ca, but they show differences in the occurrence of the honorific marker. Compared to the examples in (13b, c) in which the matrix subject and indirect object are equal in honorific status, the examples in (14) include the subject and indirect object referents that differ in terms of honorific status. Contrary to (14a), in which the matrix subject is inferior to the matrix indirect object in honorific status, the matrix subjects in (14b, c) are superior to the matrix indirect objects. Note, however, that the matrix subject and indirect object in (14c) are individuals of honorific status relative to the speaker. Madigan (2008) employs Kratzer’s (2006) [sum] feature as one of pronominal features in (15) in order to account for plurality in split control. (15) a. [[ [sum] ]] = _x._y. x+y 242 b. Each of the cats chased each of the dogs before they curled up with each other. (Kratzer 2006) Based on (15), it can be easily explained why (13b), but not (13c), takes the honorific marker on the embedded verb. That is, PRO in (13b), unlike that in (13c), bears [+honorific] feature inherited from the split controllers, the matrix subject and indirect object (the Speaker and the Addressee in Madigan’s terms), which in turn triggers the honorific marking on the embedded verb. However, Madigan (2008) does not address how the [honorific] feature on PRO accounts for the differences in the honorific marking on the embedded verbs in (14). As Madigan proposes, if the exhortative marker -ca on the Speaker/Addressee head triggers split control under the Agree relation between PRO and the Speaker/Addressee head and the [sum] feature on PRO is simply a set of atomic individuals and not a collective group, it is not clear what determines the honorific marking on the embedded verbs in (14). Given his definition of the [sum] feature on PRO, it is assumed that PRO in (14a, b) also bears a [+honorific] feature inherited from one of their controllers– the matrix indirect object in (14a) and the matrix subject in (14b) and PRO in (14c) from its split controllers. However, the honorific marking on the embedded verb is obligatory in only (14a), and in the other examples it renders the sentences ungrammatical. In fact, the examples in (13b, c) and (14) show that the honorific marking on the embedded verb is permitted only when the matrix indirect object is superior to the matrix subject in status. This is different from the honorific marking on the matrix verb that is triggered by the relationship between the matrix subject ‘referent’ and the speaker of the whole sentence. In conclusion, 243 Madigan’s Agree-based analysis also falls short in explaining the occurrence of the honorific marking on the control complements. As a matter of fact, the examples above suggest that the empty subjects of control complements do not inherit their features from their controllers, but bear their own features. This also supports the claim that the null subject in finite control cannot be PRO but must be pro, which is independently licensed in the complement clause. Finally, it is worthwhile to note that OC and NOC complements in Korean have a uniform structure. The mainstream literature on control considers the OC vs. NOC distinction a syntactic distinction rather than a semantic distinction (cf. Hornestin 1999; Landau 2000). However, syntactic properties, such as structural configuration and Tense, are not crucial for the OC vs. NOC distinction in Korean. Recall that finiteness cannot distinguish OC from NOC in Korean, since Korean OC occurs in finite contexts, as discussed in chapter 3. (16) a. Minai-ka Wujinj-eykey [proi/*j/*k hakkyo-ey M.-NOM W.-DAT ka-keyss-ta]-ko malha-yess-ta. school-LOC go-VOL-DC-C say-PST-DC ‘Minai told Wujin that shei would go to school.’ b. Minai-ka [proi/*j ilccik ttena]-l M.-NOM kes-ul kyelsimha-yess-ta. early leave-ADN thing-ACC resolve-PST-DC ‘Mina resolved to leave early.’ c. Minai-ka [proi/*j cip-ey M.-NOM ka]-ki-lul kepwuha-yess-ta. home-LOC go-NOML-ACC refuse-PST-DC ‘Mina refused to go home.’ 244 (17) a. Minai-ka Wujinj-eykey [pro*i/*j/k hakkyo-ey ka-keyss-ta]-ko malha-yess-ta. M.-NOM W.-DAT school-LOC go-CONJ-DC-C tell-PST-DC ‘Minai told Wujin that someone might go to school.’ b. Minai-ka [proi/j ilccik ttena]-l M.-NOM kes-ul wenha-yess-ta. early leave-ADN thing-ACC want-PST-DC ‘Mina wanted to leave early.’ c. Minai-ka [proi/j cip-ey M.-NOM ka]-ki-lul wenha-yess-ta. home-LOC go-NOML-ACC want-PST-DC ‘Mina wanted to go home.’ As shown in above examples, the matrix verbs in OC (16) select the same types of complements that the matrix verbs in NOC (17) select. The only difference between the constructions lies in the Modal marker -keyss and the matrix verbs. This clearly shows that the basic OC vs. NOC distinction in Korean is not determined by a syntactic structure, but is triggered by lexical semantic or pragmatic factors. Similarly, lexical semantics plays an important role in controller choice in OC contexts. This issue will be discussed in the following sections. 5.2. Roles of Syntax, Semantics, and Pragmatics in Controller Choice It has been claimed that different classes of complementizers or Mood markers can make differences in controller choice, regardless of the lexical properties of the matrix predicates (Madigan 2006, 2008; Gamerschlag 2007) and control in Korean is the result of the combination of a complementizer/Mood marker and a matrix predicate capable of control (Yang 1984; Kim 1995). By these analyses, some of the previous 245 studies suggest that OC in Korean cannot be explained in terms of pure syntax, but may be semantically determined in part. However, they do not make a comprehensive investigation of the various data that may be related to control constructions, resulting in a failure to capture the fact that OC in Korean is a largely semantically based phenomenon derived by an intricate set of factors. Also, although they address the control effect of Mood or Modal markers in -ko complement clauses, they fail to point out the effect of semantic properties of complement clauses on controller choice. A farranging investigation of sentential complementation in Korean reveals that control in Korean is the result of certain semantic or pragmatic implicature imposed by the semantics of a matrix predicate, complementizer, the combination of a matrix predicate and complement type, or the content of a complement clause. In short, controller choice in Korean cannot be completely explained by a simple semantic mechanism, such as a semantic approach based on the semantics of matrix predicates. 5.2.1. Roles of Matrix Predicates and Complementizers The semantic properties of individual predicates have been considered to be the main factor in establishing Obligatory Control in the literature that argues for the semantic dimension of control (c.f. Chierchia 1983, 1984; Pollard & Sag 1991; Jackendoff & Culicover 2001, 2003; Gamerschlag 2007). Generally speaking, this is true in OC constructions in Korean. That is, in general, aspectual and implicative predicates trigger subject control, directive/manipulative predicates object control, and other types of predicates trigger Non-Obligatory Control or no control at all. 98 Also, in 98 Most of Desiderative and Factive predicates tend to trigger NOC, and when the null subject is locally controlled, the controller is usually the matrix subject. On the other hand, Utterance, Propositional, Perception, and Interrogative predicates almost always show no control. These types will be called control-neutral verbs. 246 the sense that complement clauses are subject to the lexical properties of matrix predicates, control cannot be seen independently of the semantics of the predicate. However, OC in Korean cannot be explained by a single factor, but by multifaceted factors that include the lexical properties of a complementizer, the combination of a matrix predicate and a particular type of a complement clause, and so on. In what follows, I will address the role of the matrix predicates and the effect of the combination of a matrix predicate and a complement type in determining OC in Korean. 5.2.1.1. Predicates and Complementizers Inducing Inherent Control As briefly discussed in section 3.3, there are some predicates and complementizers which always induce Inherent control. As to predicates, it has been proposed that a delimited number of verbs, such as aspectual verbs (begin/start, continue), allow only OC in all languages. This suggests that at least some properties of control should be related to the lexical characteristics of the predicates involved. This is true of some OC constructions in Korean. There are limited numbers of predicates which always trigger OC regardless of the complementation type: i.e., ‘Inherent control’. Those predicates are classified into three types: 1) aspectual predicates, 2) implicative predicates silphayha- ‘fail’, kyeuliha/soholha- ‘neglect’, and samka- ‘refrain’, and 3) the desiderative predicate keylsimha- ‘resolve/determine’. Cross-linguistically, aspectual verbs select a sentential complement whose external argument has to be coreferential with the matrix subject DP. Unsuprisingly then, Korean aspectual verbs, such as sicakha- ‘begin’, kyeysokha- ‘continue’, characteristics, as illustrated in (18) and (19). and cungtanha- ‘stop’, show OC 247 (18) a. Minswui-ka [proi kongpwuha]-ki-(lul) sicakha/kyesokha-yess-ta M.-NOM study-NOML-ACC begin/continue-PST-DC ‘Minswu began/continued to study.’ b. Minswui-ka [proi kongpwuha]-nun/*l kes-ul M.-NOM study-ADN sicakha/kyeysokha-yess-ta thing-ACC begin/continue-PST-DC ‘Minswu began/continued studying.’ (19) a. Minswui-ka M.-NOM [proi kongpwuha]-ki-lul cungtanha-yess-ta study-NOML-ACC stop-PST-DC ‘Minswu stopped to study.’ b. Minswui-ka [proi kongpwuha]-nun/*l kes-ul M.-NOM study-ADN cungtanha-yess-ta thing-ACC stop-PST-DC ‘Minswu stopped to study.’ c. Minswui-ka [proi kongpwuha-te]-n M.-NOM kes-ul cungtanha-yess-ta study-RET-ADN thing-ACC stop-PST-DC ‘*Minswu stopped that he was studying.’ As shown in the above instances, aspectual predicates usually select -ki nominalizations, or -nun adnominalizations. Due to their intrinsic meanings, these predicates are not compatible with the -l adnominalization, which implies prospective Modality or future Tense. The verb cungtanha- ‘stop’ can also take an embedded clause which involves a retrospective Modal, as shown in (19c). Since the retrospective Modal denotes the speaker’s perception of the time of the embedded events that has already taken place, the Modal marker -te- can be considered to mark Past Progressive Tense. 248 Therefore, it is compatible with the inherent meanings of the predicate cungtanhata ‘stop’, putting an end to what one is doing. Conversely, the -l adnominalization in (19b), which denotes that the event has not taken place, is not. This shows that the inherent meaning of the predicates impose restrictions on the Tense or aspectual specification of the embedded clause. Furthermore, these verbs impose selectional restrictions on their arguments in the situation denoted by the clause. That is, they cannot select an animate internal argument in a simplex sentence and in the case of a complex sentence, they select two arguments, an external argument and a complement whose subject is necessarily linked to the external argument due to the lexical meaning of the predicates. 99 Therefore, no other controller, except the matrix subject, is possible. Other predicates that trigger Inherent control also impose the same restrictions on theTense/Aspect of their complements and on what type of arguments can occur with them. (20) a. Eunmii-nun [proi kongpwuha]-ki-lul/nun/*n kes-ul E.-TOP keyuliha-yess-ta. study-NOML-ACC/ADN thing-ACC neglect-PST-DC ‘Eunmi neglected to study.’ b. Kithayi-nun [proi pam-ey kansik mek]-ki-lul/nun/*n K.-TOP kes-ul samka- night-at snack eat-NOML-ACC/ADN thing-ACC refrain- ss-ta. PST-DC 99 Farkas (1992:97) consider these types of complements dependent Subject complements and argues that languages frequently use a special complement form, such as infinitives, to indicate Subject dependency. In the case of Korean, however, there is no special complement form indicating such a Subject dependency. 249 ‘Kithay refrained from having a snack at night.’ c. Kui-nun [proi ku-uy he-TOP kes-ey atul-ul wang-ulo onglipha]-m-ey/nun/*l he-GEN son-ACC king-as enthrone-NOML-LOC/ADN silphayha-yess-ta. thing-LOC fail-PST-DC ‘He failed to enthrone his son as the King.’ As shown in (20), the implicative predicates silphayha- ‘fail’, kyeuliha/sohol(hi)ha‘neglect’, and samka- ‘refrain’ show restrictions on their complementation types. The predicates kyeuliha- ‘neglect’ and samka- ‘refrain’ in (20a, b) select only -ki nominalization and -nun adnominalization. Other complementation types are not compatible with these predicates. For instance, the adnominalizer -n, which encodes indicative Modality or past Tense, cannot be compatible with these predicates since the lexical meanings of the predicates imply that the events of the complements have not been realized. In contrast, the predicate silphayha- ‘fail’ in (20c) selects -m nominalization and -nun adnominalization, but it cannot select the adnominalizer -l encoding an unrealized future event, which is not compatible with the lexical meaning of the matrix verb. Also, none of these predicates can select an animate internal argument, and thus the null subject necessarily refers to the matrix external argument, resulting in obligatory subject control. A similar pattern is found in the desiderative predicate keylsimhata ‘resolve/determine’, as in (21). 250 (21) a. Kui-nun [proi Hankwuk-ulo tolaka]-ki-lo/*m-ul he-TOP ul Korea-DR /l kes- go.back-NOML-DR/NOML-ACC/ADN thing- keylsimha-yess-ta. ACC determine-PST-DC ‘He determined to go back to Korea.’ b. Kui-nun [proi cito-lul he-TOP mantul-keyss-ta]-ko keylsimha-yess-ta. map-ACC make-VOL-DC-C determine-PST-DC ‘He determined to make a map.’ c. Nai-nun [proi nay salm-uy yangsik-ul I-TOP pakkwu]-leyko keylsimha-yess-ta. my life-GEN style-ACC change-C determine-PST-DC ‘I determined to change my lifestyle.’ d. Hyengkii-nun [proi kunye-ka kyelhonha-nun namca-lul cikcep H.-TOP she-NOM marry-ADN po]- man-ACC in.person see- kosa keylsimha-yess-ta. C determine-PST-DC ‘Hyengki determined to see the man who she is going to marry.’ Compared to the implicative predicates, the desiderative predicate keylsimha‘resolve/determine’ can select not only -ki nominalization and -l adnominalization (21a), but also quotative (21b) and intentive complements (21c, d). However, all the complement clauses denote unrealized events, which are compatible with the lexical meaning of the predicate, and the nominallizer -m, which encodes indicative Modality or past Tense cannot be selected, as shown in (21a). Similar to the aspectual and the 251 implicative predicates discussed above, this predicate imposes a restriction on its arguments, which always triggers subject control regardless of the complement types. Therefore, these predicates are analyzed as predicates inducing Inherent control. In addition, there are some complementizers which induce Inherent control. Recall that Intentive complementizers -leyko and -koca always trigger OC irrespective of verb type, as discussed in section 3.3.5. Further empirical data in (22) and (23) show that control-neutral predicates with intentive complements induce Obligatory Control. (22) a. Hananimi-kkeyse-nun [wuli-ka coy-lul cis]-tolok uytoha-ci anh-usi- God-NOM.HON-TOP we-NOM sin-ACC commit-C intend-NEG-HONess-ta. PST-DC ‘God didn’t intend us to commit sin.’ b. Hananimi-kkeyse-nun [salam-tul-i ku pwun-uy yengwonhan God-NOM.HON-TOP man-PL-NOM the person.HON-GEN eternal sayngmeyng-ul patatuli]-ki-lul life-ACC uytoha-si-ess-ta. accept-NOML-ACC intend-HON-PST-DC ‘God intended that people would accept his eternal life.’ c. Pothong yensway salinpemi-un [proi/*j casin-uy pemhayng-ul tulenay]common serial killer-TOP self-GEN crime-ACC leyko/koca uytoha-n-ta. C intend-IND-DC ‘A common serial killer intends to reveal his/her crime.’ reveal- 252 As shown in (22a) and (22b), when the predicate uytoha- ‘intend’ selects a -tolok complement and a -ki nominalized complement, no control can occur. This indicates that the predicate uytoha- ‘intend’ is control-neutral. However, when the predicate selects an intentive complement, it always triggers OC, as shown in (22c). The examples in (23) show the same pattern. (23) a. Nai-nun [kunye-ka na-man-ul I-TOP salangha]-ki-lul/l kes-ul wenha she-NOM I-only-ACC love-NOML-ACC/ADN thing-ACC want- yess-ta. PST-DC ‘I wanted her to love only me.’ b. Salam-tuli-un [cheycwung-i cek-key naka]-nun kes-ul wenha-n-ta. man-PL-TOP weight-NOM small-ADV weigh-ADN thing-ACC want-IND-DC ‘People want their weight to weigh less.’ c. Kui-nun [proi/*j chengmallo kunye-wa salang-ul nanwu]-koca wenha-yess-ta. he-TOP really she-with love-ACC share-C want-PST-DC ‘He really wanted to fall in love with her.’ The desiderative predicate wenha- ‘want’ has been considered a NOC predicate in the literature. Indeed, the examples in (23a) and (23b), with nominalized and adnominalized complements, show no control. However, when it selects a -kosa complement, it always triggers obligatory subject control, as shown in (23c). This also applies to other 253 desiderative predicates, which are generally considered NOC predicates, such as cwunpiha- ‘prepare’, kyeyhoykha- ‘plan’, para- ‘want’, hyimangha- ‘hope’ and kalmangha/kalkwuha- ‘desire’. Therefore, we can conclude that the intentive complementizers are the key factor triggering Obligatory Control. 5.2.1.2. Control-neutral Predicates In addition to the predicates that induce Inherent control above, there are many predicates that have been treated as obligatory subject control in the literature. Some of them are given in (24). (24) Subject Control verbs sidoha- ‘try’, noleykha/ayssu- ‘endeavor’, yaksokha- ‘promise’, kyeyhoykha- ‘plan’, cwunpiha- ‘prepare’, kecelha- ‘refuse’, kepwuha- ‘decline’, hwuhoyha- ‘regret’ Similar to the verbs inducing Inherent control, these verbs cannot take an animate internal direct argument and denote that the event denoted by the embedded verb is brought about by the referent of the matrix subject, resulting in subject control. In fact, the majority of the data involving the verbs above in the corpora show subject control. Consider the following examples involving some of the predicates. (25) a. Minai-ka [proi/*j ttena]-lyeko/koca sitoha/noleykha-yess-ta. M.-NOM leave-C ‘Mina tried to leave.’ try/endeavor-PST-DC 254 b. Nai-nun [proi/*j ku ceyan-ul I-TOP patatuli]-ki-lul/nun kes-ul the proposal-ACC accept-NOML-ACC/ADN thing-ACC kecelha-yess-ta. refuse-PST-DC ‘I refused to accept the proposal.’ c. Kui-nun [proi/*j nayil he-TOP achim-ey ttena]-leyko cwunpiha-yess-ta. tomorrow morning-at leave-C prepare-PST-DC ‘He prepared for leaving at tomorrow morning.’ d. Wulii-nun [proi/*j nayneyn-ey khun aphathu-lo we-TOP ul next.year-at big isaka]-l kes- apartment-DR move.in-ADN thing- kyeyhoykha-yess-ta. ACC plan-PST-DC ‘We planned to move into a big apartment next year.’ The examples above show that the null subject is coreferent with the matrix subject and it cannot be coindexed with any other referent. However, there are some data involving the same verbs in which there is no control, although the numbers of the examples are very small, compared to the examples with the same verbs showing subject control. 100 Consider the following 100 This is also true for other predicates which are often treated as control verbs in the literature. Moreover, most instances of non-control involving the predicates take an inanimate subject which may have some close relationship with the matrix argument, as in (26a) and (28c). When the complement clause takes an animate subject, the clause is often interpreted as a passive sentence, as in (26b). A further example appears below: (i) Pwucang-un [Mikwuk-ey kwanhan kisa-ka sil-li-l ke-la]-ko chief-TOP USA-about news-NOM print-PASS-ADN thing-DC-C 255 examples involving ‘try’-type verbs that have been commonly treated as an example of typical OC in the literature. (26) a. Ku-nun [kunpwu-ka khwutheytha-lul ilukhi]-tolok sitoha-yess-ta. he-TOP the military-NOM coup d’état-ACC carry.out-C try-PST-DC ‘(lit.) He tried for the military to carry out a coup d’état.’ b. Na-nun [ku-salam-i I-TOP sekpang-toy-l swu iss]-tolok noleyha-keyss-ta. the man-NOM release-become-can-C endeavor-VOL-DC ‘(lit.) I will make an effort so that the man can be released from jail.’ Unlike the common view of ‘try’-type verbs as obligatory subject control verbs, the examples in (26) show that an overt NP different from the matrix argument can occur in the null subject position of the complements. A semantic approach to control considers that the verb ‘try’ necessarily entails that the matrix subject is identical to the null subject of the complement clause, which Chierchia (1983) treats as a meaning postulate. Given this, the example in (26a) is predicted to be ungrammatical, contrary to fact, since the subject of the complement clause differs from that of the matrix clause. However, the ‘try’ type verbs in (26a) and (26b) are different from those in (25a) in that the former is reminiscent of a causative construction. Recall that the resultative complementizer -tolok can often denote a causative meaning, as discussed in section 3.3.6. Also, it is worthwhile to note that the verb ‘try’ in Greek can select a na yasokha-yess-ta. promise-PST-DC ‘The chief promised that a piece of news about USA would be printed.’ 256 subjunctive complement with an overt lexical NP, which is more accurately translated as ‘try to make it so that’ (Terzi 1997:340), as shown in (27b). (27) a. I Mariai prospathise eci na Mary tried-3SG diavasi. PRT read-3SG ‘Mary tried to read.’ b. I Maria prospathise na Mary tried-3SG diavasoun (ta PRT read-3PL pedia). the children ‘Mary tried for the children to read.’ Compared to the Greek examples, in which the matrix verbs select the same na subjunctive complement, the ‘try’ type verbs in Korean select different types of complement clauses: the resultative complement in (26) and the intentive complement in (25a). This suggests two important aspects of OC in Korean. First, the ‘try’ type verbs in Korean are not the strict OC predicates, as is commonly accepted in the literature. Second, matrix predicates are not the sole factor in establishing OC in Korean. In fact, there are many verbs that are classified as control-neutral verbs but which trigger OC in combination with a particular complementizer; these will be discussed in detail in the next section. Consider the following desiderative predicates, which have been treated as subject control verbs in the literature, but show no control at all. 257 (28) a. Ku yeca-ka pwulsonhan thayto-lo the woman-NOM impolite ey tuleka]-nun kes-ul [wuli ilhayng-i ku pang- attitude-DR we company-NOM the roomkecelha-yess-ta. LOC enter-ADN thing-ACC refuse-PST-DC ‘(lit.) The woman refused us entering the room with an impolite attitude.’ b. Tangsin-un [nay-ka Yi hoycang-kwa ochan-ul hamkkey ha-l swu iss]- you.-TOP.HON I-NOM Y. president-with lunch-ACC together do-cantolok cwunpiha-si-o. C prepare-HON-IMP ‘(lit.) You prepare that I can have a lunch with president Yi.’ c. Haksayng-tul-un [cenkwuk-uy student-PL-TOP i-la-nun kak tayhak-i Minchenghakleyn whole.country-GEN each univserty-NOM M. ilum-ha-ey taykyumo siwi-lul peli]- COP-DC-ADN name-under-LOC large.scale demonstrateion-ACC spread.outl kes-ul kyeyhoykha-yess-ta. ADN thing-ACC plan-PST-DC ‘(lit.) The students planned that each university in the whole country hold a demonstration on a large scale under the name of Minchenghakleyn.’ Compared to the examples in (25b-d) which show subject control, the examples in (28), which employ the same predicates, show no control. This clearly indicates that contrary to common presumption, these verbs are not obligatory subject control verbs. The same conclusion applies to the verb yaksokhata ‘promise’. 258 (29) a. Minai-ka Minswuj-eykey [proi/*j ttena-keyss-ta]-ko yaksokha-yess-ta. M.-NOM M.-DAT leave-VOL-DC-C promise-PST-DC ‘Mina promised Minswu that she would leave.’ b. Kim taypyenini-un [proi/*j taypyenin-ulose chwungsilhi yekhwal-ul ha-ye K. spokesman-TOP naka-l kes-i]-m-ul spokesman-as sincerely role-ACC do-YE yaksokha-yess-ta. go.forth-ADN thing-COP-NOML-ACC promise-PST-DC ‘Spokesman Kim promised that he sincerely play his role as a spokesman.’ c. Minai-ka Minswuj-eykey [proi/*j ku chayk-ul M.-NOM M.-DAT cwu]-ki-lo yaksokha- the book-ACC give-NOML-DR promise- yess-ta. PST-DC ‘Mina promised Minswu to give the book.’ As can be seen in the examples above, the verb yaksokha- ‘promise’ can select all types of complementation, except Intentive complements. This verb has almost universally been considered an obligatory subject control verb in the literature, since the verb requires that the referent of the matrix subject intentionally executes the action referred to by the embedded verb. This is true in most cases of ‘promise’ constructions in Korean, as can be seen in (29). However, the examples in (30) show that the verb does not always trigger OC. That is, the lexical properties of the verb yaksokha- ‘promise’ do not guarantee OC. 259 (30) a. Minai-ka Minswuj-eykey [proi/i+j ttena]-ki-lo M.-NOM M.-DAT yaksokha-yess-ta. leave-NOML-DR promise-PST-DC ‘Mina promised Minswu to leave/that they (Mina and Minswu) would leave.’ b. Kim sensayng-nimi-kkeyse K. Im sensayng-nimj-kkey teacher-HON-NOM.HON I. pan haksayng-tul-i [tangsin teacher-HON-DAT.HON self.HON tosekwan-ul chengsoha]-ki-lo yaksokha-yess-ta. class student-PL-NOM library-ACC clean-NOML-DR promise-PST-DC ‘Teacher Kim promised teacher Im that his students would clean the library.’ c. Hananimi-kkeyse [milay-ey hananim-uy nala-ka God-NOM.HON future-at god-GEN l kes-ul i ttang-ey imha]- nation-NOM this land-LOC arrive- yaksokha-si-ess-ta. ADN thing-ACC promise-HON-PST-DC ‘God promised that his kingdom will come to the world in future.’ d. Nay-ka [kulen il-i I-NOM such eps]-tolok yaksokha-li-ta. incident-NOM do.not.exist-C promise-VOL-DC ‘I promise that such an instance won’t happen.’ (30a) is ambiguous in that it can have an EC or split control reading. 101 Although the EC reading is preferred, the sentence does not exclude the possibility of split control. That is, the agent of the action described in the embedded clause need not be exclusively Mina. It is indeterminate whether Mina promised that only she herself will 101 The ambiguity discussed here should not be construed as structural ambiguity, so multiple syntactic structures are not being suggested. Rather the ambiguity is one of ambiguity of reference. 260 leave, or that she and Minswu will leave together, and thus it is not semantically clear who the controller is. In other words, Mina and Minswu can both be potential controllers who are the agents of leaving. In this case, the control relation between the matrix arguments and embedded subject seems to depend on pragmatic factors. That is, the ambiguity between an EC and split control reading will be resolved by the discourse situation in which the sentence is uttered. This suggests that the semantics of the matrix verb alone cannot determine the actual control relation for the complement clause. The examples in (30b)-(30d) show further that an overt NP subject different from the matrix subject can occur in the null subject position of the complement, resulting in no control. Even though a quite limited number of the examples involving the matrix verbs discussed so far show no control in the corpora, it is very suggestive evidence of finite control in Korean in two respects. First, the lexical meaning of matrix predicates cannot be the sole factor determining OC; some other factors may play a role in determining control. Second, the possibility of an overt embedded subject distinct from the matrix subject indicates that the complement clause selected by the matrix verb is finite. Lastly, the possibility of an overt embedded subject identical to the matrix subject cannot be considered just an emphatic pronoun. Similar phenomena are observed in the directive/manipulative predicates, which have also been treated as obligatory object control in the literature. Consider the following examples, which often show object control in the corpora. (31) a. Minai-ka Minswuj-lul/eykey [pro*i/j ttena]-tolok seltukha-yess-ta. M.-NOM M.-ACC/DAT leave-C persuade-PST-DC 261 ‘Mina persuaded Minswu to leave’ b. Aii-nun emmaj-eykey [pro*i/j calmos-ul yongseha-ye cwu-si]-tolok child-TOP mom-DAT fault-ACC forgive-YE give-HON-C kanchengha-yess-ta. beg-PST-DC ‘The child begged mom to forgive his/her fault.’ The majority of directive/manipulative verbs in the corpora select a -tolok complement clause, and in that case, the construction always takes object control, just as in the English analogue of the Korean structure. Recall that the complementizer -tolok cannot trigger Inherent control, as discussed in section 3.3.6. This suggests that OC in (31) may be triggered by the lexical meaning of the matrix verbs. In fact, the directive/manipulative verbs appear to denote manipulation of the object referent to make him/her bring about the event denoted by the embedded verb, resulting in object control. However, when the directive verbs select different types of complements, object control is not guaranteed, as can be seen in the following examples. (32) a. Samchoni-un [proi/*j na-lul math-ase uncle-TOP khiwu-keyss-ta]-ko apecilj-lul I-ACC take-ASE raise-VOL-DC-C father-ACC seltukha-si-ess-ta. persuade-HON-PST-DC ‘The uncle persuaded father that he would take over and take care of me.’ 262 b. Minai-ka Minswuj-eykey [proj/i+j ttena]-l M.-NOM M.-DAT kes-ul seltukha-yess-ta. leave-ADN thing-ACC persuade-PST-DC ‘Mina persuaded Minswu to leave/that (she and Minswu) leave.’ (33) a. Kui-nun imkum-kkey [proi/*j nomo-lul he-TOP king-DAT.HON mosi-keyss-ta]-ko old.mother-ACC serve-VOL-DC-C kanchengha-yess-ta. beg-PST-DC ‘He begged the king that he would serve his old mother.’ b. Akhan yeng-tuli-un [proi/*j twayci ttey-ey evil spirit-PL-TOP pig tuleka]-ki-lul kanchengha- herd-LOC enter-NOML-ACC beg- yess-sup-ni-ta. PST-AH-IND-DC ‘The evil spirits begged to dwell in the body of the pigs.’ As shown in the examples above, some directive predicates, such as seltukha- ‘persuade’ in (32), can additionally take -ko complements and -l adnominalized complements, and others such as kanchengha- ‘beg’ in (33), can also select ko complements and -ki nominalized complements. 102 The example (32a) shows that the agent of the action described in the embedded clause is the matrix subject, not the matrix direct object, 102 The directive predicates heyongha/helakha- ‘allow’, myengleyngha- ‘command’, cwumwunha- ‘order’, and hyeppakha/wihyepha- ‘threaten’, belong to the former group, while the directive predicates kangyouha- ‘force’, pwutakha/tangpwuha- ‘ask’, colu‘coax’, yochengha- ‘request’, and kanchengha- ‘beg’, belong to the latter group. In addition, some directive predicates, such as heyongha/helakha- ‘allow’ and ceyanha‘propose/suggest’ can further take -nun adnominalized complements, yet others, such as mantul- ‘make’, sikhi- ‘let’, ceykongha- ‘offer’, and pangciha/mak- ‘prevent/stop’, impose more restrictions on their complementation types. 263 resulting in subject control. Moreover, (32b) shows that the verb cannot guarantee the sentence will be OC. That is, the null subject in (32b), similar to (30a), does not exclude split control, although the object control reading is preferred. Similarly, the verb kanchengha- ‘beg’ does not always trigger object control. The examples in (33) show subject control. This indicates that the directive/manipulative predicates are not obligatory object control verbs and that the lexical meaning of matrix predicates cannot be the sole factor determining the controller choice. The following examples show further evidence that the control relation cannot be predicted solely based on the semantics of the matrix verbs, but may be determined by some other factors, such as the lexical meaning of the complementizer. (34) a. Minai-ka [proi ttena]-lyeko ha-yess-ta. M.-NOM leave-C do-PST-DC ‘Mina tried to leave.’ b. Minai-ka Minswuj-eykey [proi cenyek-ul M.-NOM M.-DAT sa]-ki-lo ha-yess-ta. dinner-ACC buy-NOML-DR do-PST-DC ‘Mina promised Minswu to buy dinner.’ c. Minai-ka Minswuj-lul [pro*i/j ttena]-tolok ha-yess-ta. M.-NOM M.-ACC leave-C do-PST-DC ‘Mina persuaded Minswu to leave’ The constructions in (34) take the same matrix predicate ha- ‘do’ which does not have an inherent meaning other than “do”. However, the constructions display different types 264 of OC just as the English analogue of the Korean constructions shown in the English translations above. Determining the OC status of a given complement solely on the basis of the semantics of the selecting predicate will not explain how the same predicate contrasts in its control properties. In the above examples, the complementizer or the complementizer-like element appears to be key in determining the control relation. Recall that the intentive complementizers, -lyeko and -koca, always trigger obligatory subject control irrespective of verb type. However, other complementizers do not always trigger Inherent control. Thus, at least in the case of (34b) and (34c), we cannot say that the complementizer is the sole factor inducing Obligatory Control. Considering that the matrix verb ‘do’ along with the complementizer creates the same control effect as the corresponding English verbs do, it is reasonable to assume that the combination of a complement type and a matrix predicate play a role in determining control relation. This issue is discussed in the next section. 5.2.1.3. Combination of a Predicate and a complementizer There are two types of structures illustrating that control effects are the result of the combination of a particular type of a complement and a matrix predicate. One is -ki complements selected by some implicative and desiderative predicates. This type obligatorily triggers subject control, and the implicative verb sitoha- ‘try’ and some desiderative predicates, such as tonguyha- ‘agree’, kyelcengha- ‘decide’, senthayha‘choose’, and sayangha- ‘decline’ belong to this type. The other type is -tolok complements selected by all directive/manipulative predicates, which trigger obligatory object control. First, consider some examples of the first type. 265 (35) a. Salam-tuli-i [proi/*j ku kos-eye people-PL-NOM tuleka]-ki-lul sitoha-yess-ta. the place-LOC enter-NOML-ACC try-PST-DC ‘People tried to enter the place.’ b. Kihoi-ka [proi/*j Swuni-wa thonghwaha]-ki-lul K.-NOM S.-with sitoha-yess-ta. talk.over.the.phone-NOML-ACC try-PST-DC ‘Kiho tried to talk to Swuni over the phone.’ (36) a. Kui-nun tanhohi [proi/*j ton-ul he-TOP firmly pat]-ki-lul kecelha-yess-ta. money-ACC accept-NOML-ACC refuse-PST-DC ‘He firmly refused to accept the money.’ b. Noini-un [proi/*j atul-ul old.man-TOP manna]-ki-lul kecelha-yess-ta. son-ACC meet-NOML-ACC refuse-PST-DC ‘The old man refused to meet his son.’ As discussed in 5.2.1.2, the implicative verb sitoha- ‘try’ and the desiderative verb kecelha- ‘refuse’ are analyzed as control-neutral verbs. However, when these verbs select -ki complements, subject control is always triggered, as shown in (35) and (36). The following examples further show that -ki complements with some desiderative predicates always trigger subject control, although the predicates are not Innherent control predicates. (37) a. Yecai-nun namcaj-eykey [pro*i/j kulehan youngto-lo ton-ul woman-TOP man-DAT such use-DR ssu]- money-ACC spend- 266 tolok tonguyha-yess-ta. C agree-PST-DC ‘The woman agreed the man to spend the money on such things.’ b. Cengpwui-nun [Cwungkwuk kyengchal-i government-TOP Chinese ku-lul teyli-e-ka]-tolok policeman-NOM he-ACC take-E-go-C tonguyha-yess-ta. agree-PST-DC ‘The government agreed that the Chinese policeman would take him.’ c. Na-nun [ku-tul-i I-TOP itan-i-la]-nun kes-ey tonguyha-n-ta. he-PL-NOM heretic-COP-DC-ADN thing-LOC agree-PRS-DC ‘I agree that they are heretics.’ (38) a. Wuli-tuli-un [proi/*j nayil-kkaci we-PL-TOP memwul]-ki-lo tonguyha-yess-ta. tomorrow-until stay-NOML-DR agree-PST-DC ‘We agreed to stay until tomorrow.’ b. Kui-nun [proi/*j ku coken-ul he-TOP patatuli]-ki-lo tonguyha-yess-ta. the condition-ACC accept-NOML-DR agree-PST-DC ‘He agreed to accept the condition.’ As can be seen in the above examples, the predicate tonguyha- ‘agree’ can select -tolok complements and -nun adnominalized and ki nominalized complements. Unlike (37a), which shows object control, (37b) and (37c) show no control. This indicates that tonguyha- ‘agree’ is a control-neutral predicate. However, when it selects a -ki nominalized complement, it always triggers obligatory subject control, as in (38). 267 Next, consider some examples of the second type. (39) a. Ku namcai-ka naj-eykey ton-ul he man-NOM I-DAT modeyl-ul yokwuha-mye [proi/*j sacin money-ACC ask-and se-keyss-ta]-ko picture kangyoha-yess-ta. model-ACC stand-VOL-DC-C force-PST-DC ‘The man forced me to make him a model of my picture, asking me for some money.’ b. Minai-ka Pataj-eykey [proi+j swul-ul M.-NOM P.-DAT masi-ca]-ko kangyoha-yess-ta. liquor-ACC drink-PROP-C force-PST-DC ‘Mina forced Pata to have a drink together.’ c. Minai-ka Pataj-eykey [pro*i/j/i+j ttena]-l M.-NOM P.-DAT kes-ul/ki-lul leave-ADN thing-ACC/NOML-ACC kangyouha-yess-ta. force-PST-DC ‘Mina forced Pata to leave.’ As shown in (39), the predicate kangyoha- ‘force’ can select different types of complements. When it selects -ko, -l and -ki complements, it can have control relations other than object control. That is, (39a) shows subject control, (39b) split control, and (39c) shows variable control, i.e. either object control or split control. However, when it selects -tolok complements, object control always arises, as shown in (40). 268 (40) a. Sinpwui-nun kutulj-eykey [pro*i/j/*k casin-ul priest-TOP they-DAT ttalu]-tolok kangyoha-yess-ta. self-ACC follow-C force-PST-DC ‘The priest forced them to follow him.’ b. Minai-ka naj-eykey [pro*i/j/*k kecis-ul malha]-tolok kangyoha-yess-ta. M.-NOM I-DAT lie-ACC tell-C force-PST-DC ‘Mina forced me to tell a lie.’ The same pattern is observed in the following examples, which take the predicate yokwuha- ‘ask’. (41) a. Nai-nun kisaj-eykey [pro*i/j cha-lul I-TOP driver-DAT seywu]-l kes-ul /ki-lul car-ACC stop-ADN thing-ACC/NOML-ACC yokwuha-yess-ta. ask-PST-DC ‘I asked the driver to stop the car.’ b. Wuli tangi-un taythongleyngj -eykey [proi+j myentamha]-l we part -TOP president-DAT /ki-lul kes-ul meet.and.talk-ADN thing-ACC youkwuha-yess-ta. NOML-ACC ask-PST-DC ‘Our part asked the President to meet and talk.’ c. Yecai-nun namcaj-eykey [pro*i/j ttena-la]-ko woman-TOP man-DAT yokwuha-yess-ta. leave-IMP-C ask-PST-DC ‘The woman asked the man to leave.’ 269 d. Yecai-nun namcaj-eykey [proi+j ttena-ca]-ko woman-TOP man-DAT yokwuha-yess-ta. leave-EXH-C ask-PST-DC ‘The woman asked the man to leave together.’ e. Namphyeni-un anayj-eykey [pro*i/j casin-eykey pokcongha]-tolok yokwuhahusband-TOP wife-DAT self-DAT obey-C ask- yess-ta. PST-DC ‘The husband asked his wife to obey him.’ As can be seen in (41), the predicate yokwuhata ‘ask’ can take different types of complements. The examples in (41a) and (41b) show that the predicate can select -l and -ki complements and triggers object control and split control, respectively. When the predicate selects -ko complements, the result is object control or split control, as shown in (41c) and (41d). In contrast, when it selects a -tolok complement, it obligatorily triggers object control, as shown in (41e). So far, it has been argued that there are some predicates and complementizers which always trigger Inherent control. Also, it has been shown that the semantics of the main verb cannot be the sole determining factor in controller licensing in many cases. Rather, the choice of a complementizer or a particular type of a complement clause along with a matrix predicate plays an important role in determining what the controller will be. The table in (42) displays a list of the predicates or complementizers that always trigger Inherent control and of the predicates that always trigger OC in association with a particular complement type. 270 (42) OC triggered by the Matrix Predicate, Complementizer, or Their Combination Control Type Key Factors Matrix Predicates Complementizer -ki complements selected by -tolok complements selected by Inherent Control Subj C OC induced by Comp + Pred. Subj C Obj C Aspectual verbs, Implicative verbs: samka-‘refrain’, keyulliha-/soholhiha‘neglect’, silphayha‘fail’ Desiderative verbs: kyelsimha-‘resolve/ determine’ Intentive -leyko, -koca Desiderative verbs: kecelha-‘refuse’, tonguyha-‘agree’, senthaykha-‘choose’, kyelcengha-‘decide’ Implicatives verbs: sidoha-‘try’ Directive verbs 103 103 There has been a proposal that the directive predicate seltukha- ‘persuade’ can have disjoint subject referents (Monahan 2003; Cormack & Smith 2004; Gamershlag 2007), as shown in (ib). (i) a. Chelswu-nun pwumo-eykey [kakkak-uy ai-ka swukcey-lul ha-tolok C.TOP parents-DAT each-GEN child-NOM home-ACC do-C seltukha-yess-ta. persuade-PST-DC ‘Chelswu persuaded the parents to make each child do the homework.’ b. Chelswu-nun Yenghuy-lul/eykey [Swuyeng-i kakey-ey ka-yaha-n-ta]-ko C.TOP Y.-ACC/DAT S.-NOM store-DAT go-should-PRS-C seltukha-yess-ta. persuade-PST-DC ‘Chelswu persuaded Yenghi that Swuyeng should go to the store’ (Cormack & Smith 2004:68) Cormack & Smith (2004) assume that (ia) is an instance of causative coercion, i.e., it is interpreted by implicitly causativizing the embedded verb. The unexpressed causer, then, is understood to be coreferential with the matrix object, which renders the example as a special case of object control. If we considers (ia) grammatical, then the combination of 271 5.2.2. Control Effects of Mood and Modal Markers in -Ko complements It has been noted in the literature that there is a suggestive correlation between the Mood of inflected infinitives and Obligatory Control. For example, it is well known that some subjunctive clauses in Balkan languages display OC (cf. Terzi (1997), and Krapova & Petkov (1999), among others). Despite this, Landau (2004:849) argues that Mood is not directly involved in control. However, in Korean, Mood and Modality definitely play a crucial role in determining control. That is, the determination of the controller can vary based on the selection of Mood and Modal markers in embedded clauses, as already noted by some researchers. This is illustrated in the following examples. (43) a. Minai-ka Pataj-eykey [Wujin-i M.-NOM P.-DAT cip-ey ka-n-ta]-ko malha-yess-ta. W.-NOM home-LOC go-IND-DC-C tell-PST-DC ‘Mina told Pata that Wujin is going home.’ b. Minai-ka Pataj-eykey [proi/*j cip-ey M.-NOM P.-DAT ka-keyss-ta]-ko malha-yess-ta. home-LOC go-VOL-DC-C tell-PST-DC ‘Mina told Pata that she would go home.’ c. Emenii-kkeyse Pataj-eykey [proi/*j cip-ey mother-NOM.HON P.-DAT ka-ma]-ko malssumha-si- home-LOC go-PROM-C tell-HON- a -tolok complement and the predicate seltukha- ‘persuade’ does not cause object control in this case, and thus the prediate should be excluded from the table. However, the grammaticality or acceptability of (ia) is inconsistent among Korean native speakers, compared to (ib), which was accepted as a grammatical sentence by the majority of the Korean native speakers whom I consulted. Moreover, there was no token showing the same phenomenon in the copora that I investigated. Therefore, this predicate is classified with the other directive predicates which trigger obligatory object control in combination with the complementizer -tolok. 272 ess-ta. PST-DC ‘Mother told Pata that she would go home.’ d. Minai-ka Pataj-eykey [pro*i/j cip-ey M.-NOM P.-DAT ka-la]-ko malha-yess-ta. home-LOC go-IMP-C tell-PST-DC ‘Mina told Pata to go home.’ e. Minai-ka Pataj-eykey [proi+j cip-ey M.-NOM P.-DAT ka-ca]-ko malha-yess-ta. home-LOC go-EXH-C tell-PST-DC ‘Mina said to Pata that they should go home.’ The examples above take the same matrix predicate and the same complement clause headed by the -ko complementizer. In general, the verb malhata ‘tell’ is not considered an OC predicate since an R-expression can alternate with a controllee, as shown in (43a). However, if we compare (43b) with (43a), the presence of the volitional Modal marker -keyss- in the same context creates obligatory subject control. Similarly, the examples in (43c)-(43e) show that the Mood marker in an embedded clause determines whether the null subject is co-referential with the subject, the object, or both. That is, the promissive marker -ma in (43c) requires subject control, just as the volitional marker in (43b) does. In the same vein, the imperative marker -la in (43d) requires object control and the exhortative marker -ca in (43e) split control. These data clearly demonstrate that control effects can be enforced by Mood or Modal markers even for predicates which are generally considered to be NOC predicates. This is also the case for the predicates which are commonly treated as OC predicates in the literature. 273 (44) a. Emmai-ka Minaj-eykey [proj ttena-la]-ko seltukha-yess-ta. mom-NOM M.-DAT leave-IMP-C persuade-PST-DC ‘Mom persuaded Mina to leave.’ b. Emmai-ka Minaj-eykey [proi ttena-keyss-ta]-ko seltukha-yess-ta. mom-NOM M.-DAT leave-VOL-DC-C persuade-PST-DC ‘Mom persuaded Mina that she would leave.’ c. Emmai-ka Minaj-eykey [proi+j ttena-ca]-ko mom-NOM M.-DAT seltukha-yess-ta. leave-EXH-C persuade-PST-DC ‘Mom persuaded Mina that they (mom and Mina) should leave’ As discussed in section 5.2.1.2, the verb seltukhata ‘persuade’ in Korean is controlneutral, but it triggers obligatory object control when it selects the complementizer tolok. However, when the complement clauses headed by the -ko complementizer contain some Mood or Modal markers, the sentence can create a different type of OC. That is, the examples in (44) differ soley in terms of a Mood or Modal marker, yet the control relations are different in each. Unlike (44a), which involves object control induced by the imperative -la, the volitional -keyss- in (44b) and the exhortative Mood marker -ca in (44c) force subject and split control, respectively. Thus, the Mood and Modal markers determine the control relations here, and not simply the matrix verb. In other words, these examples show that the matrix verb cannot induce local control by itself. Rather, the Mood and Modal markers are the key to determining a controller in (44). 274 Importantly, however, not all Mood markers display OC. (45) a. Emmai-ka Pataj-eykey [proi/*j/k ttena-n-ta]-ko mom-NOM P.-DAT malha-yess-ta. leave-IND-DC-C tell-PST-DC ‘Momi told Pataj that shei/k would leave.’ b. Emmai-ka Minaj-eykey [pro??i/*j ttena-n-ta]-ko mom-NOM M.-DAT seltukha-yess-ta. leave-IND-DC-C persuade-PST-DC ‘Mom persuaded Mina that she would leave’ Compared to the other Mood markers in (43) and (44), the indicative Modal marker -nin (45) cannot induce OC. Considering that the event time of embedded clauses in control constructions is usually unrealized or future with respect to that of the matrix, indicative Modality, which denotes assertion and involves present Tense, is not likely to create control relations. However, the examples in (46) show that while indicative Modality itself does not cue OC, it can play some role when accompanied by an appropriate control predicate. (46) a. Emmai-ka Pataj-eykey [proi/*j ttena-n-ta]-ko mom-NOM P.-DAT yaksokha-yess-ta. leave-IND-DC-C promise-PST-DC ‘Mom promised Pata to leave.’ b. Emmai-ka Pataj-eykey [proi/i+j ttena]-l mom-NOM P.-DAT kes-ul yaksokha-yess-ta. leave-ADN thing-ACC promise-PST-DC ‘Mom promised Pata to leave/ that she (mom) and Pata would leave.’ 275 As already discussed in the previous section, in some cases ‘promise’ constructions show different referential interpretations of controllees induced by different complementizers. In contrast to (46b), which can show either subject control or split control, (46a) shows obligatory subject control, even though both exa mples contain the same constituents, other than the complement type and the presence vs. absence of the Mood marker. Considering that ‘indicatives’ refer to an actual world, while ‘nonindicatives’ refer to possible worlds (Roussou, 1999:171), the matrix subject, Mina, in (44a) is making the promise of leaving with certainty that the event of leaving will really happen as an actual event. In essence, the matrix subject can be sure about the occurrence of the leaving because she will be the agent of the event. On the other hand, the predicate ‘persuade’ involves causative meaning and usually its matrix object (undergoer) will be the agent of the event described in the complement clauses. Thus, the indicative Modality, which denotes ‘assertion’: the event described in the embedded clause is conceived of as a fact, is not compatible with the verb. Consequently, this causes the sentence (45b) to be ungrammatical. This suggests that predicates show some restrictions on the Mood or Modal marker occurring in their complement clauses, depending on their intrinsic meaning. However, the selection of a Mood or Modal marker is important for determining controllers. The following table displays a list of the Obligatory Control triggered by the intrinsic meaning of the Mood and Modal markers in Korean. 276 (47) OC triggered by Mood or Modal markers in -ko Complement Clauses Mood/Modal markers keyss ma la ca Meaning Volitional Promissive Imperative Exhortative Control type (Inherent Control) Subject Control Subject Control Object Control Split Control 5.2.3. Control Effects of Syntactic and Semantic Properties of Embedded Clauses and Pragmatics As discussed in the previous sections, OC in Korean cannot be simply explained by the semantic properties of a matrix predicate. In many cases, OC is determined by the complex interplay of a matrix predicate and complementation type, and a Mood or Modal marker in a -ko complement clause can induce control effects. This makes it extremely difficult to establish the exact formulation of OC in Korean. The following data showing that control depends on syntactic or semantic properties of embedded clauses in association with the matrix predicate make it even more challenging. (48) a. Minai-ka Minswuj-eykey [proi cenyek-ul M.-NOM M.-DAT sa]-ki-lo /-l kes-ul dinner-ACC buy-NOML-DR/ADN thing-ACC yaksokha-yess-ta. promise-PST-DC ‘Mina promised Minswu to buy dinner.’ b. Minai-ka Minswuj-eykey [proi/i+j ttena]-ki-lo M.-NOM M.-DAT /-l kes-ul leave-NOML-DR/ADN thing-ACC 277 yaksokha-yess-ta. promise-PST-DC ‘Mina promised Minswu to leave/that they (Mina and Minswu) would leave.’ The examples above demonstrate that the semantic characterization of a matrix predicate cannot guarantee invariable control, as discussed in section 5.2.1. Compared to (48a), in which subject control is the only possible interpretation, (48b) can take either subject or split control. That is, while the inherent meaning of the verb yaksokha‘promise’ appears to prevent the null subject from being exclusively controlled by the matrix indirect object, it cannot restrict the possibility of split control. Similarly, the inherent meaning of the verb seltukha- ‘persuade’ excludes subject control, but it cannot restrict the null subject to be exclusively controlled by the matrix object. This is shown in (49b). (49) a. Minai-ka Minswuj-eykey [proj casin-ul M.-NOM M.-DAT ttena]-l kes-ul seltukha- self-ACC leave-ADN thing-ACC persuade- yess-ta. PST-DC ‘Mina persuaded Minswu to leave herself.’ b. Minai-ka Minswuj-eykey [proj/i+j ttena]-l M.-NOM M.-DAT kes-ul seltukha-yess-ta. leave-ADN thing-ACC persuade-PST-DC ‘Mina persuaded Minswu to leave/that (she and Minswu) should leave.’ 278 c. Minai-ka Minswuj-eykey [proi+j hamkkey ttena]-l M.-NOM M.-DAT kes-ul seltukha- together leave-ADN thing-ACC persuade- yess-ta. PST-DC ‘Mina persuaded Minswu to leave together.’ In contrast to (49a), which only permits object control, either object control or split control is possible in (49b). The null subject in (49b) can be appropriately interpreted as having split antecedents, depending on the discourse context in which the sentence is uttered. The example in (49c), in which the adverb hamkkey ‘together’ is inserted in the embedded clause, clearly shows the possibility of split control. Note that the only difference among the examples is the embedded clauses. In particular, the examples in (49) show that different argument structures of the embedded predicates can cause different control effects. These data clearly show that control can be determined by the syntactic or semantic properties of embedded clauses. The following examples show a similar pattern. (50) a. Nai-nun sensayng-nimj-kkey [pro*i/j halapenim-ul I-TOP ki-ul teacher-HON-DAT.HON mannapoy-p]- grandfater.HON-ACC meet.HUM-HUM- yochengha-yess-ta. NOML-ACC request-PST-DC ‘I requested the teacher to meet the grandfather.’ 279 b. Nai-nun sensayng-nimj-kkey [proi/*j mannapoy-p]-ki-lul I-TOP teacher-HON-DAT.HON meet.HUM-HUM-NOML-ACC yochengha-yess-ta. request-PST-DC ‘I requested the teacher that I (could) meet him.’ c. Sinsai-nun yeinj-eykey [proi+j/j chwum-ul gentleman-TOP woman-DAT chwu]-ki-lul dance-ACC dance-NOML-ACC yochengha-yess-ta. request-PST-DC ‘The gentleman requested the woman to dance (together).’ The examples in (50) show various control for the complement clauses selected by the matrix predicates yochenghata ‘request’. Although this predicate is control-neutral just as other directive predicates are, the majority of the data involving the predicate in the corpora show that the null subject in the embedded clause is controlled by the matrix object, as shown in (50a). However, the example in (50b) shows that the complement clause selected by the same matrix verb can create subject control. The contrast between (50a) and (50b) is caused by the different argument structures of the embedded verbs. Moreover, the null subject in (50c) can be controlled either by the split antecedents or by only the matrix object antecedent. Compared to the examples in (50a) and (50b), the referential ambiguity of the construction in (50c), which can be resolved by discourse context, appears to be caused by the semantics of the embedded clauses. 280 The same pattern can be observed in (51), which involves the predicate kanchenghata ‘beg’. (51) a. Akhan yeng-tuli-un [proi/*j twayci ttey-ey evil ul spirit-PL-TOP pig tuleka]-ki-lul /-l kes- herd-LOC enter-NOML-ACC/ADN thing- kanchengha-yess-sup-ni-ta. ACC beg-PST-AH-IND-DC ‘The evil spirits begged to dwell in the body of the pigs.’ b. Kui-nun anayj-eykey [pro*i/j cip-ulo cwu]-ki-lul /-l home-DR come.back give-NOML-ACC/ADN he-TOP wife-DAT kes-ul tolawa kanchengha-yess-ta. thing-ACC beg-PST-DC ‘He begged his wife to come back home’ c. Ku-tuli-un sin-kkey [caki-tul-eykey kwuwen-i ilwu-e ci]- he-PL-TOP God-DAT.HON self-PL-DAT salvation-NOM achieve-PASSki-lul /l kes-ul kanchengha-yess-ta. NOML-ACC/ADN thing-ACC beg-PST-DC ‘They begged God that their salvation would be achieved.’ (52) a. Minai-ka sensayng-nim j-kkey [pro*i/j mence ka-si]-l M.-NOM teacher-HON-DAT.HON kanchengha-yess-ta. beg-PST-DC ‘Mina begged the teacher to go first.’ first kes-ul go-HON-ADN thing-ACC 281 b. Minai-ka sensayng-nim j-kkey [proi/*j mence ka]-l M.-NOM teacher-HON-DAT.HON first kes-ul go-ADN thing-ACC kanchengha-yess-ta. beg-PST-DC ‘(lit.) Mina begged the teacher so, she could go first.’ c. Minai-ka Pataj-eykey [proi/j mence ka]-l M.-NOM P.-DAT first kes-ul kanchengha-yess-ta. go-ADN thing-ACC beg-PST-DC ‘(lit.) Mina begged Pata to go first/begged Pata so, she could go first.’ The same matrix verb kanchenghata ‘beg’ creates different control types. The examples in (51a) and (51b) take subject and object control, respectively, and the example in (51c) shows that the matrix verb can even show non-control. The examples in (52) are particularly interesting, since the presence vs. absence of the subject honorific marker si can establish a different controller choice. In (52a) the presence of the honorific marker determines object control, while its absence can cause either subject control (52b) or variable control (52c). The controller differences between (52b) and (52c) are caused by solely pragmatic factors. That is, since the matrix indirect object, sensayngnim ‘teacher’, in (52b) is a person who should be honored by the speaker, the absence of the honorific marker indicates that the null subject is not coreferential with the matrix indirect object but with the matrix subject, Mina. In contrast, the matrix indirect object Pata in (52c) is not someone who should be honored and thus is equal to Mina in social status. Therefore, it is ambiguous who the agent of ‘going’ is. 104 These 104 Also, the predicate yochengha- ‘ask’ can show shifting between object control and subject control, based on the discourse context. 282 data clearly demonstrate that in some cases, not only are the lexical properties of the matrix subject insufficient to determine control type, but also the matrix predicate and the particular type of the complementation in combination do not always trigger the same type of control. Although the control effect of a Mood or Modal marker in -ko complement clauses has been addressed in the literature (Yang 1984; Madigan 2006, 2008; Gamerschlag 2007), the relevance of other syntactic/semantic properties of the embedded clauses and pragmatic factors for the determination of control in Korean has not been addressed. However, as discussed so far, control can depend on the syntactic/semantic properties of embedded clauses and even on pragmatic conditions. Therefore, it is argued that syntactic/semantic properties of embedded clauses and pragmatic factors should be taken into account as well for an adequate analysis of control relations in Korean. 5.2.4. Control Shift The phenomenon of control shift, which involves matrix passivization, has focused on English, and control shift in Korean has been rarely discussed in the literature. In this section, I will examine control shift in Korean, comparing it with English, and argue that control shift in Korean OC constructions is well accounted for in semantic terms, as the literature on control shift in English has commonly explained control shift in semantic terms (Bresnan 1982; Melvold 1985; Sag & Pollard 1991; and Petter 1998, among others). This analysis may be in accordance with the semantic based approach to control in that thematic relations rather than argument structures better explain the phenomenon of control shift. 283 In English, there is a restriction on passivization of matrix clauses with subject controlled complements: Visser’s generalization states that matrix clauses with subject controlled complements cannot be passivized. For example, when a ‘promise’ construction is passivized, it seems to require be allowed to complements, as shown in (53b) below. (53) a. Johni promised Annj [eci to come]. b. Anni was promised [eci to be allowed to come]. c.*Anni was promised [eci to come]. d. Johni promised Annj [ecj to be allowed to come]. (54) a. Johni persuaded Annj [ecj to leave]. b. Anni was persuaded by Johnj [eci to leave]. When the null subject in an embedded clause is an agent, the ‘promise’ construction cannot be passivized, as shown in (53c). In ‘promise’ constructions, a matrix object can be the controller of the null subject in an embedded clause, only if the embedded clause takes a be allowed to infinitive. Therefore, in ‘promise’ constructions, control shift involves insertion of be allowed to in an embedded clause or passivization of a matrix verb along with the insertion of be allowed to. Notice that in the case of passivization of matrix verbs (53b), the controller Ann is the subject as well as the agent in the active construction and that in a passivized construction (53d), the controller John remains a subject, but is semantically a non-agent. However, as shown in (54), when an object 284 controlled construction, such as a ‘persuade’ construction, is passivized, the semantic role of the controller remains the same. Korean differes from English in this regard. Passivization of ‘promise’ involves not only a shift in the controller, but also a limitation of a potential controller. (55) a. Minai-ka Pataj-eykey [proi/i+j ttena]-ki-lo M.-NOM P.-DAT yaksokha-yess-ta. leave-NOML-DR promise-PST-DC ‘Mina promised Pata to leave/that they (Mina and Pata) would leave.’ b. Patai-ka Minaj-lopwuthe [proj/*??i+j ttena]-ki-lo P.-NOM M.-from yaksokpat-ass-ta. leave-NOML-DR got.promise-PST-DC ‘Pata got a promise from Mina to leave.’ (56) a. Minai-ka Pataj-eykey [proi/i+j ttena]-l M.-NOM P.-DAT kes-ul yaksokha-yess-ta. leave-ADN thing-ACC promise-PST-DC ‘Mina promised Pata to leave.’ b. Patai-ka Minaj-lopwuthe [proj/*??i+j ttena]-l P.-NOM M.-from kes-ul yaksokpat-ass-ta. leave-ADN thing-ACC got.promise-PST-DC ‘Mina got a promise from Pata to leave.’ As shown in (55a) and (56a), some structures involving control predicates allow the constructions to have more than one controller choice. However, if the constructions are passivized, the option is not allowed, as shown in (55b) and (56b). Since the matrix subject is a non-agent undergoer, it is excluded from being a potential controller of the agent of the embedded clause. Notice that contrary to English ‘promise’ constructions 285 (53b), Korean yaksokhata ‘promise’ constructions can be passivized without involving the passivization of the complement clause. This indicates that Visser’s generalization is completely irrelevant to Korean. Moreover, note that when the yaksokhata ‘promise’ construction is passivized, it no longer involves subject control, but it does retain agent control. That is, Mina is the agent of ‘promise’ as well as that of ‘leave’. Therefore, the semantic role of the controller in ‘promise’ constructions in Korean is rather consistent. In this respect, the passivization of ‘persuade’ constructions in Korean shows a parallel with that of ‘promise’ constructions, as follows: (57) a. Minai-ka Pataj-eykey [proj ttena]-tolok seltukkha-yess-ta. M.-NOM P.-DAT leave-C persuade-PST-DC ‘Mina persuaded Pata to leave.’ b. Patai-ka Minaj-lopwuthe [proi ttena]-tolok seltukkpat-ass-ta. P.-NOM M.-from leave-C got.persuade-PST-DC ‘Pata got persuaded by Mina to leave.’ The example in (57) shows that null subject in the active construction is the matrix object Pata, who is the agent of the embedded predicate. When the matrix verb is passivized, the null subject refers to the matrix subject Pata, who is still the agent of the embedded predicate. The apparent control shift from object to subject does not actually involve control shift in semantic terms. That is, the embedded clauses in (57) remain semantically the same as non-agent control in both constructions. 286 Similar patterns occur in the following examples, in which a possibility of split control is excluded. (58) a. Minai-ka Pataj-eykey [proj/i+j ttena]-l M.-NOM P.-DAT kes-ul seltukkha-yess-ta. leave-ADN thing-ACC persuade-PST-DC ‘Mina persuaded Pata to leave.’ b. Patai-ka Minaj-lopwuthe [proi/*i+j ttena]-l P.-NOM M.-from kes-ul seltukkpat-ass-ta. leave-ADN thing-ACC got.persuade-PST-DC ‘Pata got persuaded by Mina to leave.’ As is clearly shown in (58), while either object control or split control is possible in the active construction (58a), in the passivized construction, split control is not allowed and only the undergoer, which is the matrix subject, can control the null subject (58b). Therefore, the passivized construction becomes subject control, but semantically remains non-agent control. Therefore, ‘persuade’ constructions in Korean basically show the same pattern as those in English, whereas ‘promise’ constructions in Korean differ from those in English. While English remains subject control and the semantic role of the controller changes to non-agent, Korean involves a change to ‘non-subject control’, and the agent remains the controller. 105 This is summarized in (59). 105 In fact, this phenomenon was noted by Kim (1995). In his analysis, the passivization of ‘promise’ constructions in Korean involves a shift from subject control to object control. (i) Johni-i Maryj-eykey [ecj ku il-ul ha]-ki-lo yaksokpat-ass-ta. J.-NOM M.-DAT the work-ACC do-NOML-DR got.promise-PST-DC ‘John received a promise from Mary to do the work.’ 287 (59) Control Shift in Enlgish and Korean Voice Active Predicates Language Promise English Persuade Grammatical Role Subj C Passive Semantic Role Agent Grammatical Role Subj C Semantic Role Non-agent Korean Subj C Agent Non-Subj C Agent English Obj C Non-agent Subj C Non-agent Korean Obj C Non-agent Subj C Non-agent The table in (59) shows that in Korean, the semantic rather than grammatical roles of the controllers are most important for determining the controller. In other words, control shift in Korean yields a semantically coherent interpretation. This also supports the analysis that OC in Korean is better explained in semantic terms. In addition, passivized OC constructions in Korean show a different aspect from active OC constructions with respect to replacement of reflexive caki. Specifically, caki is no longer an option for the embedded subject. (60) a. Patai-ka Minaj-lopwuthe [kunye*i/j/*caki-ka P.-NOM M.-from she ttena]-ki-lo yaksokpat- /self-NOM leave-NOML-DR got.promise- ass-ta. PST-DC However, the Korean speakers I consulted with considered the above sentence ungrammatical or unnatural because of the dative case on ‘Mary’. Instead, they preferred using postposition particles -eykeyse/lopwuthe ‘from’. 288 ‘Pata got a promise from Mina to leave.’ b. Patai-ka Minaj-lopwuthe [kunye*i/j/*caki-ka P.-NOM M.-from she ttena]-l kes-ul /self-NOM leave-ADN thing-ACC yaksokpat-ass-ta. got.promise-PST-DC ‘Pata got a promise from Mina to leave.’ c. Patai-ka Minaj-lopwuthe [kunyei/*j/*caki-ka P.-NOM M.-from she ttena]-tolok seltukkpat-ass- /self-NOM leave-C got.persuade-PST- ta. DC ‘Pata got persuaded by Mina to leave.’ d. Patai-ka Minaj-lopwuthe [kunyei/*j/*caki-ka P.-NOM M.-from she ttena]-l kes-ul /self-NOM leave-ADN thing-ACC seltukpat-ass-ta. got-persuade-PST-DC ‘Pata got persuaded by Mina to leave.’ As shown in the above examples, compared to the pronoun kunye, caki cannot replace the null subject. Given the fact that caki is subject-oriented, the impossibility of caki replacement in (60a, b) is natural because the controller Mina is not a subject of the matrix clause. On the other hand, in (60c, d), although the controller Pata is the subject of the matrix clause, caki cannot replace the null subject, either. In these constructions, 289 the controllers are non-agent, which does not match with the agent semantic role of caki. This is displayed in the following table. (61) Contrast between a Null Subject and Caki in Active and Passive Sentences Predicates Promise Persuade Voice Active Passive Active Passive Subject Agent Non-agent Agent Non-Agent Non-subject Non-agent Agent Non-agent Agent Null subj √ Agent √ Agent √ Agent √ Agent Caki subj √ Agent * Agent * Agent * Agent In the table above, the underlined elements represent the controllers. Both caki and the null subject are the agent of the embedded clauses, but unlike null subjects, caki can occur only in the active ‘promise’ construction. As shown the table in (61), the controller of the null subject in ‘promise’ constructions is agent, while that in ‘persuade’ constructions is non-agent. However, caki can occur only when the controller is a subject and agent. This leads us to conclude that there is no subject control or object control in Korean in the sense of English, or other languages with the typical type of infinitival complement. In other words, an analysis of controller choice based on semantics is more consistent and faithful to the facts than an analysis of controller choice based on syntax. In this chapter, it has been amply illustrated that OC in Korean cannot be explained by a purely syntax-based analysis. Also, it has been argued that OC in Korean cannot be simply determined by the lexical properties of matrix predicates alone, although there are some correlations with the general classification of control and the inherent meaning of matrix predicates and in some cases, such as aspectual verbs, the 290 lexical properties of the predicates are the crucial factor determining Obligatory Control. In many cases, it is also influenced by the semantics of complementizers, the complex interplay of a matrix predicate and a particular type of a complement clause, or the syntactic/semantic properties of embedded clause. Also, as touched on briefly in some cases, pragmatic considerations play a role in determining what the controller will be. Therefore, it is argued that controller choice in Korean is determined primarily by multiple semantic factors and syntactic or pragmatic factors to a lesser extent. 291 CHAPTER 6 CONCLUSION This thesis attempted to achieve two main goals. The first was to provide an empirical and theoretical analysis of finite control, which contains a pro subject in Korean, through exploring the concept of finiteness and the finite properties of control complements. The second was to provide an analysis of how controller choice is determined in Korean. One implication of the pro analysis of finite control in Korean is that control cannot be adequately accounted for based purely on syntax, which is assumed to be the case in many standard syntactic approaches to non-finite control. This thesis explores the Korean data displaying that Obligatory Control cannot be explained by a purely syntax-based or semantically based theory, but is derived by more complicated factors. In chapter 1, I extensively reviewed the previous studies of control, pointing out major problems that are directly related to the main issues of this study. The mainstream literature on control studies has consistently argued that referential dependence between an overt matrix argument and an embedded null subject is a characteristic of non-finite clauses which contain a PRO subject. Moreover, even though some evidence for finite control in structures involving pro in several languages has been presented, many researchers have proposed a PRO analysis of finite control complements. This thesis, however, argued that the currently established approach to OC, which is confined to PRO, cannot account for OC in Korean and proposed a pro analysis of finite control. In chapter 2, I briefly reviewed the formal approaches to finiteness. In the formal approaches, finiteness has been defined as the feature licensing nominative Case on 292 subjects of clauses, and Tense and Agreement features are widely accepted as the elements responsible for the assignment of nominative Case. In spite of this, crosslinguistic data in the literature reveal that languages may vary in how finiteness is determined and the features determining finiteness may not be restricted to just Tense and Agreement. Finiteness may be associated with a variety of factors, such as Mood/Modality and Aspect, which the functionalist approaches to finiteness commonly take into account as syntactic and semantic categories defining finiteness. Based on these facts, this thesis pointed out the problems of Tense and Agreement as licensing features of a nominative subject in Korean and explored the relevance of Mood and Modality as the manifestation of finiteness in Korean. This analysis basically held the conventional notion that nominative Case is correlated with finiteness. However, unlike most formal approaches, in this analysis finiteness was assumed to be closely associated with Mood or Modality as reflected in morpho-syntactic expressions. In chapter 3, this thesis provided some evidence for finiteness of control complement clauses in Korean and discussed the existence of OC in finite clauses. There have been some previous studies arguing that Korean exhibits finite control. However, the previous studies failed to consider the full range of sentential complement types displaying Obligatory Control. Moreover, non-finite control, which has been taken for granted in the previous studies, exhibited similarities to finite control in terms of syntactic properties of finiteness. The far-ranging investigation of complementation types with respect to their relation to OC vs. NOC provided a clear account for the syntactic and semantic properties of OC constructions in Korean, and revealed that there are some semantic factors, such as the lexical meaning of aspectual predicates and 293 intentive complementizers, which always trigger Inherent Control, and that there is an interaction between the meaning of matrix predicates and the complement type in determining control. In chapter 4, this thesis provided empirical evidence for the CP status of OC complements in Korean and for a pro analysis of the null subject in the constructions. That is, it was argued that the null subject in finite control complements in Korean is pro, whose case and formal features are checked in an independent locality domain, i.e. CP, which is fully specified for Tense, Agreement and Mood/Modality. This thesis further argued that pro in OC complements is different from the long distance reflexive caki and from a pronoun, providing some evidence supporting the argument. In addition, this thesis argues that some serial verb constructions in Korean, which have previously been considered non-finite control constructions since they contain a typical control predicate, are best analyzed as having a VP structure, providing empirical data showing that serial verb constructions do not exhibit any of the syntactic behaviors that are shown in finite OC constructions. However, the serial verb constructions are not grouped with OC constructions, but simply involve argument sharing since the external argument sharing in the constructions is not mediated by an empty category. In chapter 5, some evidence against purely syntax-based analyses of OC in Korean was provided and the key role of semantics in determining a controller in OC construction in Korean was argued. However, unlike the previous studies maintaining that OC is determined by the lexical semantics of the matrix predicates involved, it was argued that OC in Korean cannot be completely explained by the lexical properties of a 294 selecting predicate alone. Rather, control effects are derived from more complicated factors. That is, although there are some correlations with the general classification of control and the inherent meaning of matrix predicates and, in some cases, the lexical properties of certain predicates are the crucial factor determining Obligatory Control, in many cases, OC is derived by the semantics of complementizers, Mood and Modal markers, the complex interplay of a matrix predicate and a particular type of a complement clause, or the syntactic/semantic properties of embedded clause. Also, it was briefly mentioned that in some cases, pragmatic considerations play a role in determining what the controller will be. The major contributions of this study lie in the further extension of control theory to finite control which contains a pro subject. Considering the peculiar property of finite OC in Korean, i.e. the ability to control an overt NP in the null subject position, one might argue that there is truly no control in Korean, as found in other languages such as English. This argument assumes the concept of control to indicate the relation of coreference between the null subject of non-finite complements and one of the matrix arguments. That is, the main criterion for OC is unavailability of an overt subject in the embedded clause and where a lexical subject is possible, control is optional. With respect to the syntactic property of OC, Davison (2008) introduces a Case restriction which distinguishes OC constructions from other constructions such as raising and NOC constructions: the covert subject in an OC constructions in Hindi-Urdu cannot have dative Case, while the covert subject in a raising or NOC construction may. From the point of this analysis as well as the standard theory of Control restricting OC to only non-finite clauses and PRO, finite control in Korean may not be OC, but optional 295 control which happens to have the same co-referent relation between the null subject of finite complements and one of the matrix arguments. However, there are other important criteria for OC, which is commonly employed for non-finite OC in English and other languages, such as the unavailability of strict and de re reading of the null subject. In addition to the impossibility of an overt disjoint embedded subject, given the isomorphism of finite OC in Korean with non-finite OC in English and other languages in terms of the criteria above, it is argued that OC in a finite context is attested in Korean. This may provide further insight to the nature of language, showing that not all languages make use of the same syntactic structure to display the same linguistic phenomenon. That is, languages may vary in how a controlled subject is represented. Compared to finite control, non-finite control, by definition, may show more restrictions on the null subject. In this respect, Davison (2008)’s Case restriction which is found only in non-finite clauses where a lexical subject is not possible is very suggestive. Also, it is worthwhile to mention that Davison (2008) proposes that languages with both forward and backward control systematically lack the Case restriction. This seems to hold for Korean, which has been claimed to be a language with forward and backward control. In addition, this thesis provides a more concrete classification of control complements and predicates than has previously been discussed in the literature. However, this thesis does not address whether OC triggered by semantic factors can be explained by a uniform semantic theory and a discussion of backward control. These issues remain to be investigated in future work. 296 REFERENCES Adger, D. 2007. Three domains of finiteness: a Minimalist perspective. Finiteness: theoretical and empirical foundations, Oxford University press. ed. Nikolaeva, I. Ahn, H.-D. & Yoon, H.-J. 1989. Functional categories in Korean. In Harvard studies in Korean Linguistics 3:79-88. Department of Linguistics, Harvard University, Cambridge, Mass. Alboiu, G. 2004 Shared arguments in control. Toronoto working papers in linguistics 22. Aoshima, S. 2000. Mono-clausal control constructions in Japanese. In Yang, Youngjun and Jeong-Seok Kim (eds.): Proceedings of the 2000 International Workshop on Generative Grammar Seoul, Korean: Hanshin 33-48. Aoshima, S. 2001. Obligatory control constructions in Japanese. Report of the LACEW Project: Special Research Project of the Typological Investigation into Language & Cultures of the East & West. University of Tsukuba, Tsukuba, Japan. Asudeh, A. 2005. Control and Semantic Resource Sensitivity. Journal of Linguistics, Cambridge University Press. Aygen, N. G. 2002. Finiteness, Case and Clausal Architecture. Ph.D dissertation, Harvard University. Bhatt, R. & Yoon, J. 1991. On the composition of COMP and parameters of V2. The Proceedings of the Tenth West Coast Conference on Formal Linguistics. ed. Dawn Bates. The Center for the Study of Language and Information. Bisang, W. 2007. Categories that make finiteness: discreteness from a functional perspective and some of its repercussions. Finiteness: theoretical and empirical foundations, Oxford University press. ed. Nikolaeva, I. Boeckx, C. & Hornstein, N. 2003. Reply to ‘Control is not movement’. Linguistic Inquiry 34/2:269-280. Boeckx, C. & Hornstein, N. 2007. On (non-)obligatory control. In Davies, William D. Davies and Stanley Dubinsky (eds.): 2007, New horizons in the analysis of control and raising. (Studies in natural language and linguistic theory 71) Dordrecht, The Netherlands: Springer 251-262. Borer, H. 1989. Anaphoric AGR. In Jaeggli, Osvaldo and Kenneth J. Safir (eds.): The null subject parameter. (Studies in natural language and linguistic Theory 15) Dordrecht, The Netherlands: Kluwer 69-109. 297 Bošković, Ž. 1995. On certain violations of the Superiority Condition, AgrO and the economy of derivation. Ms.,University of Connecticut. Bošković, Ž. 1997. The syntax of nonfinite complementation. Cambridge, Mass.: MIT Press. Bouchard, D. 1984. On the content of empty categories. (Studies in generative grammar 14) Dordrecht, The Netherlands: Foris. Bresnan, J.W. 1972. Theory of complementation in English syntax. Doctoral dissertation, MIT, Cambridge, Mass. Bresnan, J. W. 1982. Control and complementation. Linguistic inquiry 13/3:343-434. Bresnan, J. W. 2001. Lexical-functional syntax. Oxford: Blackwell. Cecchetto, C. & Oniga, R. 2004. A challenge to Null Case theory. Linguistic inquiry 35/1:141-149. Chierchia, G. 1983. 'Outline of a semantic theory of (obligatory) control'. In Barlow, Michael, Daniel P. Flickinger, and Michael T. Wescoat (eds.): 1983, Proceedings of the West Coast conference on formal linguistics. Stanford, Ca.: Stanford Linguistics Association (Center for the Study of Language and Information, Stanford University) 19-31. Chierchia, G. 1984. Anaphoric properties of infinitives and gerunds. In Cobler, Mark, Susannah Mackaye, and Michael Westcoat (eds.), Proceedings of the third West Coast conference on formal linguistics. Stanford, CA., Standford Linguistics Association. Cho, E. 1998. Why, Contrastive Topic, and LF Movement, in Silva, David (ed.) Japanese and Korean Linguistics 8, Stanford:CSLI Publications, 403-415. Choe, H.-S. 2006. On (backward) object control in Korean. Harvard studies in Korean linguistics. Kyunggi, South Korea: Hanshin Publishers 373-386. Choi, S.-S. 2003. Serial verbs and the empty category. Proceedings of the workshop on Mult-Verb constructions. Trondheim Summer School. Chomsky, N. 1980. On Binding. Linguistic Inquiry 11:1-46. Chomsky, N. 1981. Lectures on government and binding. Dordrecht, The Netherlands, Foris Publications. 298 Chomsky, N. 1986. Knowledge of language: Its nature, origin and use. Westport, Connecticut: Praeger. Chomsky, N. 1993. A Minimalist program for linguistic theory. In the view from building 20: Essays in Linguistics in honor of Sylvain Bromberger, ed. Kenneth Hale and Samuel Jay Keyser, 1-52. Cambridge, MA:MIT Press. Chomsky, N. 1994. Bare Phrase Structure. In MIT Occasional Papers in Linguistics, 5. Chomsky, N. 1995. The Minimalist program. Cambridge, MA: MIT Press. Chomsky, N. 2000. Minimalist inquiries: the framework. In: R. Martin, D. Michaels, J. Uriagereka (eds.), Step by Step: Essays on Minimalist Syntax in Honor of Howard Lasnik. Cambridge, Mass:MIT Press, 89-155. Chomsky, N. 2001. Derivation by phrase. In: M. Kenstowicz (ed.), Ken Hale: Life in Language. Cambridge. Mass: MIT Press, 1-52. Chomsky, N. & Lasnik, H. 1993. The theory of principles and parameters. In Joachim Jacobs, Arnim von Stechow, Wolfgang Sternefeld, and Theo Vennemann (eds.), Syntax: An international handbook of contemporary research, Mouton de Gruyter, Berlin, PP. 506-569. Cinque, G. 2003. The interaction of passive, causative, and “restructuring” in Romance. In C. Tortora (ed.), The Syntax of Italian Dialects. Oxford: Oxford University Press. Cormack, A. and Neil, S. 2004. Backward control in Korean and Japanese.UCLWorking papers in linguistics 16:67-83. Cowper, E. 2002. Finiteness. Ms., University of Toronto. Cristofaro, S. 2007. Deconstructing categories: finiteness in a functional-typological perspective. In Finiteness:theoretical and empirical foundations, Oxford University press. ed. Nikolaeva, I. Culicover, P. W. & Jackendoff, R. 2001. Control is not movement. Linguistic inquiry 32/3:493-512. Culicover, P. W. & Jackendoff, R. 2006. Turn over control to the semantics. Syntax 9/2:131-152. Culicover, P. W. & Wilkins, W. 1986. Control, PRO, and the projection principle. Language 62:120-153. 299 Dalrymple, M. 2001. Lexical Functional Grammar. (Syntax and Semantics 34) [Chapter 12] San Diego, Ca: Academic Press. Davies, W. D. and Dubinsky, S. 2004. The grammar of raising and control. Blackwell Publishers. Davison, A. 2008. A Case restriction on control: Implications for Movement. Journal of South Asian Linguistics. Center for the Study of Language and Information. Dik, S. 1997. The Theory of Functional Grammar. Berlin: Mouton de Gruyter. Dobrovie-Sorin, C. 1994. The Syntax of Romanian: Comparative Studies in Romance. Berlin: Mouton de Gruyter. Dowty, D. R. 1985. On recent analyses of the semantics of control. Linguistics and Philosopy 8/3:291-331 Dubinsky, S. 2007. Mediating the syntax and semantics of control. Ms. submitted to Language. Farkas, D. 1988. On obligatory control. Linguistics and Philosophy 11:27-58. Farkas, D. F. 1992. On obviation. In I. A. Sag & A. Szabolcsi (eds.) Lexical Matters. Center for the Study of Language and Information. Farrell, P. 1995. 'Backward control in Brazilian Portuguese'. In Fuller, Janet M., Ho Han, and David Parkinson (eds.): ESCOL '94 (Proceedings of the eleventh Eastern States conference on linguistics), 116-127 Frajzyngier, Zygmunt. 1995. A functional theory of complementizers. In Bybee, Joan and Feischman, Suzanne (eds): Modality in grammar and discourse. Jonh Benjamins publishing company. Fuji, T. 2006. Some theoretical issues in Japanese control. Doctoral dissertation, University of Maryland, College Park. Fukuda, S. 2007. On the control/raising ambiguity with aspectual verbs: a structural account. ZAS Papers in Linguistics 47:159-195. Gamerschlag. T. 2007. Semantic and structural aspects of complement control in Korean. ZAS Papers in Linguistics 47: 81-123. George, L. & Kornfilt, J. 1981. Finiteness and boundedness in Turkish. In Binding and Filtering, ed. Frank Heny, 105-128. London: Croomhellm Ltd. 300 Ghomeshi, J. 2001. Control and thematic agreement. Canadian Journal of Linguistics 8/1:149-160. Givón, T. 1990. Syntax: A functional-typological introduction, vol. 1-2. Amsterdam: Benjamins. Grover, C. 1995. Rethinking some empty categories: Missing object and parasitic gaps in HPSG. Doctoral dissertation. University of Essex. Haegeman, L. 1985. INFL, COMP and nominative Case assignment in Flemish infinitivals. Features and Projections. Foris publications, Dordrecht–Holland/ Riverton– U.S.A. Halliday, M. A. K. 1994. Functional Grammar. London: Arnold. Han, H.-S. 1989. Nominative Case Assignment in Korean: Against the asymmetrical Case marking hypothesis. Language Research 25/3:611:644. Hashemipour, M. M. 1989, Pronominalization and control in Modern Persian. Doctoral dissertation, Univ. of California, San Diego. Hoji, H. 1985. Logical Form Constraints and Configurational Structures in Japanese, Doctoral dissertation, University of Washington. Holmberg, A., Nikanne, U., Oraviita, I., Reime, H., & Trosterud, T. 1993. The structure of INFL and the finite clause in Finnish. In A. Holmberg & U. Nikanne (eds.), Case and other functional categories in Finnish syntax. Berlin: Mouton de Gruyter. Hong, K.-S. 1994. Subjecthood tests in Korean. Languge Research 30/1:99-136. Hornstein, N. 1999. Movement and control. Linguistic inquiry 30:69-96. Hornstein, N. 2003. Reply to ‘Control is not movement’. Linguistic Inquiry 34: 269280. Hwang, K.-H.1997. Nominative and default case checking in Minimalist syntax. Doctoral Dissertation, University of Washington. Iatridou, S. 1993. On nominative Case assignment and a few related things. MIT Woring papers in linguistics. 19: 175-196. Jackendoff, R. 1972. Semantic interpretation in generative grammar. Cambridge, Mass: MIT Press. 301 Jackendoff, R. & Culicover. P. W. 2003. The semantic basis of Control in English. Language 79:517-556. Kang, E.-K. 2001. The -key constructions in Korean: Predicate Head or Complementizer. Paper presented at the 9th international symposium on Korean Linguistics, Harvard University. Kapetangianni, K. & Seely. T. D. 2007. Control in Greek: It's another good move. In Davies, William D., and Stanley Dubinsky (eds.): 2007, New horizons in the analysis of control and raising. (Studies in natural language and linguistic theory 71) Dordrecht, The Netherlands: Springer 133-157 Kaplan, R. M. & Bresnan, J. 1982. Lexical-Functional Grammar: A formal system for grammatical representation. In Bresnan (ed.) 1982. The mental representation of grammatical relations. Cambridge, MA: MIT Press. Kayne, R. 1994. The Antisymmetry of Syntax. Cambridge, MIT Press. Kim, K-A. 1995. Causativity in Korean: Syntactic causative, control and purpose. Doctoral dissertation, University of Southern California. Kim, N.-K. 1984. The grammar of Korean Complementation. University of Hawaii Center for Korean Studies Monograph 11. Kiss, T. 2005. On the empirical viability of the movement theory of control. Ms., University of Bochum. Klein, W. 1998. Assertion and finiteness. In N. Dittmar & Z. Penner (eds.), Issues in the theory of language acqisiton: Essays in honor of Jürgen Weissenborn. Bern: Lang. Kornfilt, J. 1984. Case marking, agreement, and empty categories in Turkish. Ph. D. dissertation. Harvard University. Kornfilt, J. 2007. Verbal and nominalized finite clauses in Turkish. In Finiteness: theoretical and empirical foundations, Oxford University press. ed. Nikolaeva, I. Kotzoglou, G. & Papangeli, D. 2007. Not really ECM, not exactly control: The 'QuasiECM' construction in Greek. In Davies, William D., and Stanley Dubinsky (eds.): 2007, New horizons in the analysis of control and raising. (Studies in natural language and linguistic theory 71) Dordrecht, The Netherlands: Springer 111-131. 302 Krapova, I. 2001. Subjunctives in Bulgarian and Modern Greek. In M. L. Rivero & A. Ralli (eds.), Comparative Syntax of Balkan Languages. Oxford: Oxford University Press, 105-26. Krapova, I. & Petkov, V. 1999. Subjunctive complements, null subjects and Case checking in Bulgarian. In Dziwirek Katarzyna, Herbert Coats, and Cynthia M. Vakareliyska (eds.) Proceedings of the 7th Meeting of Formal Approaches to Slavic Linguistics, Michigan Slavic Publications, Univ. of Michigan, Ann Arbor. Kratzer, A. 2006. Minimal pronouns: fake indexicals as windows into the properties of bound variable pronouns. ms. University of Massachusetts, Amherst. Kuroda, K. 1978. The distribution and ecology of Noctiluca scintillans. Osaka Bay. Bull. Jpn. Soc. Fish. Oceanogr., 32, 56–67 (in Japanese). Kwon, N-Y. & Polinsky, M. 2006. Object control in Korean: structure and processing. Japanese/Korean Linguistics 15. Landau, I. 1999. Control theory and the identification of PRO. Landau, I. 2000. Elements of control. Structure and meaning in infinitival constructions. (Studies in natural language and linguistic theory 51) Dordrecht, The Netherlands: Kluwer. Landau, I. 2003. Movement out of control. Linguistic Inquiry 34/:471-498. Landau, I. 2004. The Scale of Finitess and the Calculus of Control, NLLT 22. Landau, I. 2005. Severing the Distribution of PRO from Case. Handout of the 79th LSA annual meeting, Jan.7: Panel on Rasing and Control. Landau, I. 2007. 'Movement-resistant aspects of control'. In Davies, William D., and Stanley Dubinsky (eds.): 2007, New horizons in the analysis of control and raising. (Studies in natural language and linguistic theory 71) Dordrecht, The Netherlands: Springer 293-325. Langacker, R. W.1991. Foundation of Cognitive Grammar. Vol. 2: Descriptive application. Stanford, CA: Standford University Press. Lee, H.-S. 1991. Tense, Aspect, and Modality: a Discourse-Pragmatic Analysis of Verbal Affixes in Korean from a Typological Perspective. Unpublished Ph.D. dissertation, University of California at LA. Lee, I.-S. & Ramsey. R. 2000. The Korean Language. State University of New York Press. 303 Lee, K.-Y. 2007. Control in Korean. Unpublished ms., University of Iowa. Lyons, J. 1977. Semantics. 2 vol. Cambridge: Cambridge University Press. Madigan, S. 2006. Exhaustive and partial control in Korean: Controlled caki as an overt from of PRO. In the proceedings of Harvard Studies in Korean Linguistics 11. Madigan, S. 2008. Control Constructions in Korean. Ph.D. dissertation. Manzini, M. R. 1983. On control and control theory. Linguistic inquiry 14/3:421-446. Manzini, M. R. & Roussou, A. 2000. A minimalist theory of a-movement and control. Lingua 110:409-447. Martin, R. A. 1996. A minimalist theory of PRO and control. Doctoral dissertation, Univ. of Connecticut, Storrs, Con. Martin, R. A. 2001. Null Case and the distribution of PRO. Linguistic inquiry 32/1:141166. Martin, S. 1992. A Reference Grammar of Korean. Rutland, Vermont, Charles E. Tuttle. Melvold, J. 1985. Getting PRO under Control. Ms., MIT. Modesto, M. 2007. Null subjects in Brazilian Portuguese and Finnish: They are not derived by movement. In Davies, William D., and Stanley Dubinsky (eds.): 2007, New horizons in the analysis of control and raising. (Studies in natural language and linguistic theory 71) Dordrecht, The Netherlands: Springer 231248. Mohanan, K.P. 1983. Functional and anaphoric control. Linguistic inquiry 14/4:641674. Moulton, K. 2005. Strong reflexivity and attitudes de se: Evidence from English ECM. Presented at ConSOLVE XIV, Univ. of the Basque Country, December 20. Monahan, P. J. 2003. Backward object control in Korean. In Garding, Gina and Mimu Tsujimura (eds.): 2003, Proceedings of the 22nd West Coast conference on formallinguistics . Somerville, MA: Cascadilla Press 356-369. Motapanyane, V. 1995. Theoretical implications of complementation in Romanian. Padova: Unipress. 304 Nikolaeva, I. 2007. Finiteness:theoretical and empirical foundations, Oxford University press. ed. Nikolaeva, I. Nishigauchi, T. 1984. Control and the thematic domain. Language 60. 215-250. O’Neil, J. 1995. Out of Control. Proceedings of NELS 25. GLSA, University of Massachusetts, Amherst. Pesetsky, D. 1992. Zero syntax II: an essay on infinitives. Ms., MIT, Cambridge, Mass. Pesetsky, D. & Torrego, E. 2004. The syntax of valuation and the interpretability of features. In Phrasal and Clausal Architecture: Syntactic derivation and interpretation. 2007. John Benjamins. Petter, M. 1998. Getting PRO under Control. LOT International Series 8, HIL, Holland Academic Graphics- The Hague. Philippaki-Warburton, I. 1987. The theory of empty categories and the pro-drop parameter in Modern Greek, Journal of Linguistics 23:289-318. Philippaki-Warburton, I. 2004. Peremfatikotita ke ipotaktiki: anazitondas to xameno aparemfato se dhomes elegxou. Proceedings of the 6th International Conference of Greek Linguistics. University of Crete, CD-Rom edition. Philippaki-Warburton, I. & Catsimali, G. 1999. On control in Greek, Studies in Greek syntax, edited by Artemis Alexiadou, Geoffrey Horrocks and Melita Stavrou, 153-168. Platzack, C. & Holmberg, A. 1989. The role of AGR and finiteness in Germanic VO languages. Working papers in Scandinavian Syntax 43:51-76. Polinsky, M. 2007. Object control in Korean: Obligatory vs. non-obligatory control. ZAS Papers in Linguistics 47, 81-123. Polinsky, M, P. J. Monahan, and N.-Y. Kwon. 2007. Object control in Korean: How many constructions? Language research 43/1, 1-31. Polinsky, M. and Eric Potsdam. 2002a. Backward control: evidence from Malagasy. In Rakowski, Andrea and Norvin Richards (eds.): 2002, Proceedings of the eighth meeting of the Austronesian Formal Linguistics Association. MIT Woring papers in linguistics. 44: 257-272. Polinsky, M. and Eric Potsdam. 2002b. Backward control. Linguistic Inquiry 33/2:245282. Polinsky, M. and Eric Potsdam. 2006. Expanding the scope of control and raising. Syntax 9/2, 171-192. 305 Pollard, C. & Sag, I. A. 1987. Information-based Syntax and Semantics. Volume 1: Fundamentals. Stanford: CSLI Publications. Pollard, C. & Sag, I. A 1991. An Integrated Theory of Complement Control. Language. 67(1), 63–113. Pollard, C. & Sag, I. A. 1994. Head-driven Phrase Structure Grammar. Chicago & Stanford, CA: The University of Chicago Press & CSLI Publications. Postal, P. M. 1970. On coreferential complement subject deletion. Linguistic inquiry 1: 439-500. Potsdam, E. 2006. Backward object control. Against an empty category analysis. Paper read at The 25th West Coast Conference on Formal Linguistics (WCCFL 25), Seattle, WA April 28-30, 2006. Raposo, E. 1987. Case theory and Infl-to-Comp: the inflected infinitive in European Portuguese. Linguistic Inquiry 18:85-109. Reinhart, T. & Reuland, E. 1993. Reflexivity. Lingistic Inquiry 24:657-720. Reuland, E. & Reinhar, T. 1995. Rronouns, Anaphors and Case. In Hubert Haider, Susan Olsen, and Sten Vikner (eds.), Studies in Comparative Germanic Syntax, Kluwer Academic Publishers, Dordrecht, 241-268. Rivero, M. L. 1994. Clause structure and V-movement in the languages of the Balkans. Natural Langue and Linguistic Theory 12:63-120. Rizzi, L. 1997. The fine structure of the left periphery. Elements of Grammar: Handbook of Generative Syntax. ed. By Liliane Haegeman,, 281-337. Dordrecht: Kluwer. Roberts, I. G. 1993. Verbs and diachronic syntax: A comparative history of English and French. Studies in Natural Language and Linguistic Theory 28. Kluwer Academic Publishers. Roussou, A. 2000. On the left periphery: Modal particles and complementisers. Journal of Greek Linguistics 1:65-94. Rosenbaum, P. S. 1967. The grammar of English predicate complement constructions. Cambridge, Mass.: MIT Press. Roussou, A. 1999. Modals and the Subjunctive. Studies in Greek syntax, edited by Artemis Alexiadou, Geoffrey Horrocks and Melita Stavrou, 169-184. 306 Ryu, D.-S. 1994. Agreement in Korean. Studies in Generative grammar 4:211-247. Sag, I. A. & Pollard, C. 1991. An integrated theory of complement control. Language 67/1:63-113. Schütze, C. T.1997. INFL in child and adult language: Agreement, Case and licensing. Ph.D. dissertation, MIT, Cambridge, Mass. Sells, P. 1998. Optimality and economy of expression in Japanese and Korean. Japanese Korean Linguistics. ed. N. Akatsuka, H. Hoji, S. Iwasaki, S. Sohn, & S. Strauss, 499-514. United States: CSLI Publications. Shlonsky, Ur. 1997. Clause structure and word order in Hebrew and Arabic: An essay in comparative semitic syntax. Oxford university Press, New York. Sigurðsson, H. A. 1991. Icelandic Case-marked PRO and the licensing of lexical argument. Natural Language and Linguistic Theory. 9:327-63. Singh. J. 1994. Case and Agreement in Hindi: a GB approach. Ph. D. dissertation, University of York. Sohn, H.-M. 1976. Semantics of Compound Verbs in Korean. Linguistic Journal of Korean 1/1:142- 150. Sohn, H.-M. 1994. Korean: Descriptive Grammars. London: Routledge. Sohn, H.-M. 1999. The Korean Language. Cambridge University Press. Sohn, S-O. 1995. Tense and Aspect in Korean. Center for Korean Studies: Monograph 18, Univ. of Hawai’i. Sohn, Y.-S., Hong, Y.-Y., & Hong, K.-S. 1996. The functional category IP in Korean reconsidered. Studies in Generative grammar 6/2:351-384. Spyropoulos, V. 2007. Finiteness and control in Greek. In Davies, William D. Davies and Stanley Dubinsky (eds.): 2007, New horizons in the analysis of control and raising. (Studies in natural language and linguistic theory 71) Dordrecht, The Netherlands: Springer. 159-183. Spyropoulos, V. & Philippaki-Warburton, I. 2001. Ipotaktiki, taftoprosopia ke I ipotheoria tou elegxous Subjunctive, coreference, and the module of control in Greek. Paper presented at the 5th International Conference on Greek Linguistics. Paris. 13-15 September 2001. Stiebels, B. 2007. Towards a typology of complement control. ZAS Papers in Linguistics 47, 1-80. 307 Stowell, T. 1982. The tense of Infinitives. Linguistic inquiry 13:561-570. Suh, C.-S. 1977. Remarks on subject honorification. In C.-W. Kim (ed.) Papers in Korean Linguistics. 297:304. Suñer, M. 1984. Controlled pro. Linguistic Symposium on Romance Lnaguages 12:253273. Terzi, A. 1992. PRO in finite clauses. A study of the inflectional heads of the Balkan languages. Doctoral dissertation, New York University, City University of New York. Terzi, A. 1997. PRO and Null Case in finite clauses. Linguistic review 14/4:335-360. Uchibori. A. 2000. The syntax of subjunctive complements: The evidence from Japanese. University of Connecticut dissertation. Vincent, N. 1998. On the grammar of inflected nonfinite forms (with special reference to old Neapolitan). In I. Korzen & M. Herslund (eds.), Clause combining and text structure. Copenhagen: Samfunds literature. Watanabe, A. 1993. The notion of finite clauses in AGR-based Case Theory. MIT Working Papers in Linguistics 18:281-296. Wechsler, S. 1997. Resultative predicates and control. In Texas Linguistic Forum 38: The Syntax and Semantics of Predication, edited by R. Blight and M. Moosally, 307-321. Williams, E. 1980. Predication. Linguistic inquiry 11:203-238. Wurmbrand, S. 2001. Infinitives. Restructuring and Clause Structure. Studies in generative grammar 55. Berlin (Germany): Mouton de Gruyter. Wurmbrand, S. 2002. Syntactic versus semantic control. In Zwart, C. Jan-Wouter and Werner Abraham (eds.): Studies in Comparative Germanic Syntax Proceedings form the 15th Workshop on Comparative Germanic Syntax. Amsterdam, The Netherlands: John Benjamins 93-127. Yang, D-W. 1982. Control and binding in Korean. Ono. Linguistic journal of Korea 7/2. Yang. D-W. 1984. Extended Control Theory, Language Research. Vol 20:1. Yang, D-W. 1985. On the integrity of control theory'. In Berman, S., J.-W. Choe, and J. McDonough (eds.): Proceedings of the North East Linguistic Society. Amherst, Mass.: GLSA, University of Massachusetts 15/389-408. 308 Yang. D-W. 1996. The Korean Case structure in the Minimalist Program. Korean Journal of Linguistics, 21:139-189. Yim, Y.-J. 19985. Multiple subject constructions. Harvard studies in Korean Linguistics, ed. By S. Kuno. Seoul: Hanshin Publishing Company. Yoon, J.-M. 1989. ECM and multiple subject constructions in Korean. Harvard studies in Korean Linguistics 3. ed. By S. Kuno. Seoul: Hanshin Publishing Company. Zec, D. 1987. On Obligatory Control in Clausal Complements.Working papers in grammatical theory and discourse structure: interactions of morphology, syntax,and discourse. Standford, Ca.: Center for Study of Language and Information 139-168.