<%BANNER%>

Adjunct Control in Telugu and Assamese

Permanent Link: http://ufdc.ufl.edu/UFE0021164/00001

Material Information

Title: Adjunct Control in Telugu and Assamese
Physical Description: 1 online resource (307 p.)
Language: english
Creator: Haddad, Youssef Abdallah
Publisher: University of Florida
Place of Publication: Gainesville, Fla.
Publication Date: 2007

Subjects

Subjects / Keywords: adjunct, backward, case, copy, expletive, forward, linguistics, movement, multiple, role, sideward, syntax, theta, visibility
Linguistics -- Dissertations, Academic -- UF
Genre: Linguistics thesis, Ph.D.
bibliography   ( marcgt )
theses   ( marcgt )
government publication (state, provincial, terriorial, dependent)   ( marcgt )
born-digital   ( sobekcm )
Electronic Thesis or Dissertation

Notes

Abstract: ADJUNCT CONTROL IN TELUGU AND ASSAMESE My study explores Adjunct Control in two South Asian languages, Telugu(Dravidian) and Assamese (Indo-Aryan), within the Minimalist Program of syntactic theory. Adjunct Control is a relation of obligatory co-referentiality between two subjects, one in the matrix clause and one in an adjunct/subordinate clause of the same structure. Telugu and Assamese have non-finite Conjunctive Participle (CNP) clauses that function as adjuncts. Both languages show evidence of Adjunct Control into CNP clauses. Three types of Adjunct Control are examined. These are Forward Control, in which only the matrix subject is pronounced; Backward Control, in which only the subordinate subject is pronounced; and Copy Control, in which case both subjects are pronounced. Telugu licenses all three types of Adjunct Control, while Assamese licenses only Forward and Copy Control. Sentences (1-3) are examples from Telugu. (1) Forward Control aakali wees-i Kumar sandwic tinnaa-Du hunger fall-CNP Kumar.NOM sandwich ate-3.M.S Having felt hungry, Kumar ate a sandwich. (2) Backward Control Kumar-ki aakali wees-i sandwic tinnaa-Du Kumar-DAT hunger fall-CNP sandwich ate-3.M.S Having felt hungry, Kumar ate a sandwich. (3) Copy Control Kumar-ki aakali wees-i atanu/Kumar sandwic tinnaa-Du Kumar-DAT hunger fall-CNP he/Kumar.NOM sandwich ate-3.M.S Having felt hungry, Kumar ate a sandwich. I analyze Adjunct Control as movement, providing a detailed account of the conditions that drive and constrain each type of control. I suggest that the subject starts out in the adjunct before it moves to the matrix clause. The result is non-distinct copies of the same element in both clauses. Decisions regarding the pronunciation of copies take place on the phonological side of the computation. The pronunciation of one copy only (the matrix or adjunct copy) results in Forward or Backward Control. The pronunciation of both copies results in Copy Control.
General Note: In the series University of Florida Digital Collections.
General Note: Includes vita.
Bibliography: Includes bibliographical references.
Source of Description: Description based on online resource; title from PDF title page.
Source of Description: This bibliographic record is available under the Creative Commons CC0 public domain dedication. The University of Florida Libraries, as creator of this bibliographic record, has waived all rights to it worldwide under copyright law, including all related and neighboring rights, to the extent allowed by law.
Statement of Responsibility: by Youssef Abdallah Haddad.
Thesis: Thesis (Ph.D.)--University of Florida, 2007.
Local: Adviser: Potsdam, Eric H.
Electronic Access: RESTRICTED TO UF STUDENTS, STAFF, FACULTY, AND ON-CAMPUS USE UNTIL 2012-08-31

Record Information

Source Institution: UFRGP
Rights Management: Applicable rights reserved.
Classification: lcc - LD1780 2007
System ID: UFE0021164:00001

Permanent Link: http://ufdc.ufl.edu/UFE0021164/00001

Material Information

Title: Adjunct Control in Telugu and Assamese
Physical Description: 1 online resource (307 p.)
Language: english
Creator: Haddad, Youssef Abdallah
Publisher: University of Florida
Place of Publication: Gainesville, Fla.
Publication Date: 2007

Subjects

Subjects / Keywords: adjunct, backward, case, copy, expletive, forward, linguistics, movement, multiple, role, sideward, syntax, theta, visibility
Linguistics -- Dissertations, Academic -- UF
Genre: Linguistics thesis, Ph.D.
bibliography   ( marcgt )
theses   ( marcgt )
government publication (state, provincial, terriorial, dependent)   ( marcgt )
born-digital   ( sobekcm )
Electronic Thesis or Dissertation

Notes

Abstract: ADJUNCT CONTROL IN TELUGU AND ASSAMESE My study explores Adjunct Control in two South Asian languages, Telugu(Dravidian) and Assamese (Indo-Aryan), within the Minimalist Program of syntactic theory. Adjunct Control is a relation of obligatory co-referentiality between two subjects, one in the matrix clause and one in an adjunct/subordinate clause of the same structure. Telugu and Assamese have non-finite Conjunctive Participle (CNP) clauses that function as adjuncts. Both languages show evidence of Adjunct Control into CNP clauses. Three types of Adjunct Control are examined. These are Forward Control, in which only the matrix subject is pronounced; Backward Control, in which only the subordinate subject is pronounced; and Copy Control, in which case both subjects are pronounced. Telugu licenses all three types of Adjunct Control, while Assamese licenses only Forward and Copy Control. Sentences (1-3) are examples from Telugu. (1) Forward Control aakali wees-i Kumar sandwic tinnaa-Du hunger fall-CNP Kumar.NOM sandwich ate-3.M.S Having felt hungry, Kumar ate a sandwich. (2) Backward Control Kumar-ki aakali wees-i sandwic tinnaa-Du Kumar-DAT hunger fall-CNP sandwich ate-3.M.S Having felt hungry, Kumar ate a sandwich. (3) Copy Control Kumar-ki aakali wees-i atanu/Kumar sandwic tinnaa-Du Kumar-DAT hunger fall-CNP he/Kumar.NOM sandwich ate-3.M.S Having felt hungry, Kumar ate a sandwich. I analyze Adjunct Control as movement, providing a detailed account of the conditions that drive and constrain each type of control. I suggest that the subject starts out in the adjunct before it moves to the matrix clause. The result is non-distinct copies of the same element in both clauses. Decisions regarding the pronunciation of copies take place on the phonological side of the computation. The pronunciation of one copy only (the matrix or adjunct copy) results in Forward or Backward Control. The pronunciation of both copies results in Copy Control.
General Note: In the series University of Florida Digital Collections.
General Note: Includes vita.
Bibliography: Includes bibliographical references.
Source of Description: Description based on online resource; title from PDF title page.
Source of Description: This bibliographic record is available under the Creative Commons CC0 public domain dedication. The University of Florida Libraries, as creator of this bibliographic record, has waived all rights to it worldwide under copyright law, including all related and neighboring rights, to the extent allowed by law.
Statement of Responsibility: by Youssef Abdallah Haddad.
Thesis: Thesis (Ph.D.)--University of Florida, 2007.
Local: Adviser: Potsdam, Eric H.
Electronic Access: RESTRICTED TO UF STUDENTS, STAFF, FACULTY, AND ON-CAMPUS USE UNTIL 2012-08-31

Record Information

Source Institution: UFRGP
Rights Management: Applicable rights reserved.
Classification: lcc - LD1780 2007
System ID: UFE0021164:00001


This item has the following downloads:


Full Text

PAGE 1

1 ADJUNCT CONTROL IN TELUGU AND ASSAMESE By YOUSSEF A. HADDAD A DISSERTATION PRESENTED TO THE GRADUATE SCHOOL OF THE UNIVERSITY OF FLORIDA IN PARTIAL FULFILLMENT OF THE REQUIREMENTS FOR THE DEGREE OF DOCTOR OF PHILOSOPHY U NIVERSITY OF FLORIDA 2007

PAGE 2

2 2007 Youssef A. Haddad

PAGE 3

3 To the queen and princess of my heart To my wife Soraya and our daughter Elena Yo las adoro tanto!

PAGE 4

4 ACKNOWLEDGMENTS I thank my supervisory committee chair (Dr. Eric Potsdam) for being my mentor and for all the guidance and support. This work would not have been possible without him. First, he chose me to be his research assistant and to be part of his NSF grant which funded my research. Further, his comments and questions always urged me to dig deeper and to think more logically. I will always be grateful for having worked under his supervision. I also thank Dr. Brent H enderson who joined my committee at the right time. I owe him much gratitude for the discussions we had and for all the questions he asked, although I am not sure I have provided answers to all of them. My wholehearted thanks go to Dr. Ann Wehmeyer for all the help and encouragement, not only regarding this study but ever since I joined the Linguistics Program at University of Florida. Many thanks are also owed to Dr. Murali Rao who accepted my request to join my committee despite his busy schedule. I am ve ry grateful to my Telugu and Assamese consultants. They all were graduate students at the University of Florida T hey were born in India where they grew up and lived most of their lives. On the Telugu side, I was fortunate to work with Karthik Boinapally, Mahesh Tanniru, Santhosh Kopidaka Venicata Ramana Cheekoti, Krishna Chaitanya Nimmagadda, and Sankara Sarma V. Tatiparti. On the Assamese side, I had the chance to work with Priyankoo Sarmah, Chandan Talukdar, Randeep Pratim Khaund, and S akib R. Saikia I am especially thankful to Karthik Boinapally, Mahesh Tanniru, Santhosh Kopidaka, Randeep Pratim Khaund, and S akib R. Saikia for their willingness to sit with me week after week and answer my repetitive questions about their languages. I also thank the Col lege of Liberal Arts and Sciences at University of Florida for granting me the Russell Dissertation Fellowship. I t was an honor, and it also helped me cut down on my work load and focus on my research. Thanks to this grant, I was able to graduate in a time ly

PAGE 5

5 manner. I also owe much gratitude to Dr. Caroline Wiltshire and Dr. Virginia LoCastro for the academic and emotional support and for always being there for all of us graduate students when we needed them. To the people at the Linguistics Happy Hour that is held every Friday in downtown Gainesville I say thank you for all the drinks and laughs. you have made. I hope I have made you proud, and I hope I will not ha ve to stay away f r om you And to my wife Soraya and my daughter Elena, I dedicate this work, because I love them beyond words, and also because I think they put as much into it as I did. I want to take this opportunity to let them know how happy I am to be with them and that I want to grow old with them. Thy rod and thy they always have The material in this disserta tion is based on work supported by the Nationa l Science Foundation under Grant No. BCS 0131993. Any opinions, findings, and conclusions or recommendations expressed in this material are mine and do not necessarily reflect the views of the National Science Foundation.

PAGE 6

6 TABLE OF CONTENTS page ACKNOWLEDGMENTS ................................ ................................ ................................ ............... 4 LIST OF TABLES ................................ ................................ ................................ ......................... 10 LIST OF ABBREVIATIONS ................................ ................................ ................................ ........ 11 ABSTRACT ................................ ................................ ................................ ................................ ... 13 CHAPTER 1 INTRODUCTION ................................ ................................ ................................ .................. 15 1.1 Research Questions ................................ ................................ ................................ ......... 15 1.2 Domain of Investigation ................................ ................................ ................................ 16 1.3 Analytic Approach ................................ ................................ ................................ .......... 19 1.3.1 From Government and Binding to Minim alism: An Overview ......................... 20 1.3.1.1 The architecture of the grammar in Government and Binding ............ 21 1.3.1.2 The architect ure of the grammar in the Minimalist Program ............... 24 1.3.2 Control in Government and Binding ................................ ................................ .. 29 1.3.3 Control in the Minimalist Progr am ................................ ................................ .... 32 1.3.3.1 PRO Theory of Control ................................ ................................ ........ 32 1.3.3.2 The Movement Theory of Control ................................ ....................... 37 1.3.4 Multiple Copy Spell Out and the Realization vs. Deletion of Copies ............... 42 1.3.4.1 Deletion of copies ................................ ................................ ................. 42 1.3.4.2 Realization of multiple copies ................................ .............................. 47 1.4 Structure of the Study ................................ ................................ ................................ ...... 49 2 ADJUNCT CONTROL IN TELUGU AND ASSAMESE: A DESCRIPTIVE OVERVIEW ................................ ................................ ................................ ........................... 52 2.1 Introduction ................................ ................................ ................................ ...................... 52 2.2 Linguistic Overview ................................ ................................ ................................ ........ 52 2.3 Case in Telugu and Assamese: A Descriptive Overview ................................ ................ 54 2.3.1 Telugu Case Marked Subject NPs ................................ ................................ ..... 55 2. 3.2 Assamese Case Marked Subject NPs ................................ ................................ 57 2.4 Finite Clauses in Telugu and Assamese ................................ ................................ .......... 61 2.4.1 Finite Clauses in Telugu ................................ ................................ .................... 61 2.4.2 Finite Clauses in Assamese ................................ ................................ ................ 62 2.5 Non Finite Clauses in Telugu and Assamese ................................ ................................ .. 64 2.5.1 Infinitive Clauses in Telugu and Assamese ................................ ....................... 64 2.5.1.1 Infinitive clauses in Telugu ................................ ................................ .. 64 2.5.1.2 Infinitive clauses in Assamese ................................ ............................. 66

PAGE 7

7 2.5.2 Conjunctive Participle Clauses in Telugu and Assamese ................................ .. 67 2.5.2.1 Conjunctive Participle clauses in Telugu ................................ ............. 68 2.5.2.2 Conjunctive Participle clauses in Assamese ................................ ........ 68 2.5.2.3 The subordinate na ture of CNP clauses ................................ ............... 69 2.6 CNP Clauses and Adjunct Control ................................ ................................ .................. 73 2.6.1 Forward Control in Telugu and Assamese ................................ ......................... 75 2.6.1.1 Forward Control in Telugu ................................ ................................ ... 75 2.6.1.2 Forward Control in Assamese ................................ .............................. 78 2.6.2 Backward Control in Telugu and Assamese ................................ ...................... 79 2.6.2.1 Backward Control in Telugu ................................ ................................ 79 2.6.2.2 Backwa rd Control in Assamese ................................ ........................... 80 2.6.3 Copy Control ................................ ................................ ................................ ...... 82 2.6.3.1 Copy Control in Telugu ................................ ................................ ........ 83 2.6.3.2 Copy Control in Assamese ................................ ................................ ... 88 2.6.4 Exceptions ................................ ................................ ................................ .......... 93 2.7 Conclusion ................................ ................................ ................................ ....................... 95 3. FORWARD/BACKWARD ADJUNCT CONTROL IN TELUGU AND ASSAMESE: THE ANALYSIS ................................ ................................ ................................ .................... 97 3.1 Introduction ................................ ................................ ................................ ...................... 97 3.2 Forward/Backward Control: The Facts ................................ ................................ ........... 98 3.2.1 Forward Control ................................ ................................ ................................ 98 3.2.2 Backward Control ................................ ................................ ............................ 103 3.3 Telugu and Assamese Adjunct Control as Obligatory Control ................................ ..... 107 3.4 Telugu and Assamese Adjunct Control as Sideward Movement ................................ .. 113 3.4.1 Forward Control ................................ ................................ ............................... 116 3.4.1.1 Forward Control as sideward movement ................................ ........... 116 3.4.1.2 Forward Control as sideward plus remnant movement ...................... 119 3.4.2 Backward Control ................................ ................................ ............................ 123 3.4.2.1 Backwar d Control in Telugu ................................ .............................. 123 3.4.2.2 Backward Control in Assamese ................................ ......................... 126 3.5 Multiple Case Checking and Copy Control ................................ ................................ ... 133 3.5.1 Multiple Case Checking: Inherent Structural ................................ .................. 134 3.5.2 Multiple Case Checking: Structural Structural ................................ ................ 135 3.6 Case in Raising vs. Control ................................ ................................ ........................... 136 3.6.1 ................................ ................................ ............................ 137 3.6.2 Raising vs. Co ntrol in Telugu and Assamese ................................ .................. 139 3.6.3 Case in Raising vs. Control: the Counterargument ................................ .......... 141 3.6.3.1 Boeckx and Hornstein ................................ ....................... 141 3.6.3.2 Case and Theta Role Visibility ................................ .......................... 142 3.7 Conclusion ................................ ................................ ................................ ..................... 147 4. COPY ADJUNCT CONTROL IN TELUGU AND ASSAMESE: THE ANALYSIS ....... 148 4.1 Introduction ................................ ................................ ................................ .................... 148 4.2 Copy Control as Movement ................................ ................................ ........................... 151

PAGE 8

8 4.3 Copy Control: The Derivational History ................................ ................................ ....... 155 4.4 Copy Control and Linearization ................................ ................................ .................... 158 4.4.1 Multiple Copy Spell Out ................................ ................................ .................. 160 4.4.1.1 ................................ ................................ ................. 161 4.4.1.2 ................................ ................................ .................... 162 4.4.2 Multiple Copy Spell Out and Multiple Spell Out ................................ ........... 165 4.4.2.1 Multiple Spell Out and copy raisin g ................................ .................. 167 4.4.2.2 Multiple Spell Out and Copy Control ................................ ................ 170 4.5 Adjunction to CP and Unwanted Instances of Sideward Movement ............................ 177 4.6 Phonological Realization of Copies ................................ ................................ ............... 182 4.6.1 Movement and the PF Realization of Copies ................................ .................. 183 4.6.2 Lack of Cataphoricity and the Nature of the CNP Subject .............................. 189 4.7 Conclusion ................................ ................................ ................................ ..................... 194 5. TRI GGER: WHY MOVEMENT IN CONTROL? ................................ .............................. 196 5.1 Introduction ................................ ................................ ................................ .................... 196 5.2 Enlightened Self Interest and Control ................................ ................................ ............ 196 5.2.1 ................................ .............................. 197 5.2.2 ................................ .............................. 199 5.3 Event and Control ................................ ................................ ................................ .......... 204 5.4 CP vs. IP and Control ................................ ................................ ................................ .... 208 5.4.1 IP as Defective for [Person] ................................ ................................ ............. 212 5.4 .2 IP as Defective for [Tense] ................................ ................................ .............. 213 5.5 Probe Goal Relationships ................................ ................................ .............................. 215 5.5.1 The Vehicle Requirement on Merge Revisited ................................ ................ 219 5.5.2 Telugu and Assamese Adjunct Control and the Vehicle Requirement on Set and Pair Merge ................................ ................................ ................................ ..... 226 5.6 Adjunct Control in English and the Vehicle Requirement on Set and Pair Merge ....... 233 5.7 Conclusion ................................ ................................ ................................ ..................... 237 6. EXCEPTIONS TO ADJUNCT CONTROL AS NON EXCEPTIONS ............................... 239 6.1 Introduction ................................ ................................ ................................ .................... 239 6.2 Non Volitional as Unaccusative ................................ ................................ .................... 243 6.3 Unaccusative Predicates and Expletive Control ................................ ............................ 246 6.3.1 Adjunct Control and the Target of Sideward Movement ................................ 255 6.3.2 Expletive Control and Late Merge ................................ ................................ ... 257 6.3.2.1 Late merge: empirical evidence ................................ ......................... 262 6.3.2.2 Parasitic gaps minus cyclic merge ................................ ..................... 265 6.4 Expletive Control in English vs. Telugu and Assamese ................................ ................ 266 6.4.1 The Movement Approach to Expletive Control ................................ ............... 269 6.4.2 Expletives Are Inserted as Needed ................................ ................................ .. 274 6.4.3 It vs. There Expletive Control in English ................................ ....................... 281 6.5 Conclusion ................................ ................................ ................................ ..................... 283

PAGE 9

9 7. SUMMARY AND CONCLUSION ................................ ................................ ..................... 286 7.1 Summary ................................ ................................ ................................ ........................ 286 7.2 Theoretical Implications ................................ ................................ ................................ 289 7.2.1 Multiple Case Checking ................................ ................................ ................... 289 7.2.2 R Expressions vs. Pronomina ls in Copy Control ................................ ............ 290 7.2.3 Why Movement ................................ ................................ ............................... 291 7.3 Concluding Remarks ................................ ................................ ................................ ..... 292 APPENDIX ADJUNCT CONTROL AS EXHAUSTIVE CONTROL ................................ 293 LIST OF REFERENCES ................................ ................................ ................................ ............. 295 BIOGRAPHICAL SKETCH ................................ ................................ ................................ ....... 307

PAGE 10

10 LIST OF TABLES Table page 2 1 Some types of case in Telugu ................................ ................................ ................................ .. 55 2 2 Some types of case in Assamese ................................ ................................ ............................. 57 2 3 Subjects licensed in Telugu ................................ ................................ ................................ ..... 95 2 4 Subjects licensed in Assamese ................................ ................................ ................................ 95 2 5 Types of Adjunct Control allowed in Telugu and Assamese ................................ .................. 96

PAGE 11

11 LIST OF ABBREVIATIONS unacceptable/ungrammatical ?? degraded acceptable / grammatical (only used in contrast with or ?? ) 1 1 st person 3 3 rd person ABS absolutive ACC accusative CL classifier CNP conjunctive participle DAT dative EMPH emphatic EXP NOM experiential nominative F Feminine GEN genitive GRND gerund HON honorific INF infinitive LOC locative M Masculine N neutral NEG negative N OM nominative

PAGE 12

12 P plural pro EXP null expletive S singular

PAGE 13

13 Abstract of Dissertation P resented to the Graduate School of the University of Florida in Partial Fulfillment of the Requirements for the Degree of Doctor of Philosophy ADJUNCT CONTROL IN TELUGU AND ASSAMESE By Youssef A. Haddad August 2007 Chair: Eric Potsdam Maj or: Linguistics My study explores Adjunct Control in two S outh Asian languages, Telugu (Dravidian ) and A ssamese ( Indo Aryan), within the Minimalist Program of syntactic theory. Adjunct Control is a relation of obligatory co referentiality between two subj ects, one in the matrix clause and one in an adjunct/subordinate clause of the same structure. Telugu and Assamese have non finite Conjunctive Participle (CNP) clauses that function as adjuncts Both languages show evidence of Adjunct Control into CNP clau ses. Three types of Adjunct Cont rol are examined. These are Forward Control, in which only the matrix subjec t is pronounced; Backward Control, in which only the subordinate s ubject is pronounced; and Copy Control, in which case both subjects are pronounced Telugu licenses all three types of Adjunct Control, while Assamese licenses only Forw ard and Copy Control. Sentences (1 3) are examples from Telugu. (1) Forward Control [aakali wees i] Kumar sandwic tinnaa Du [hunger fall CNP] Kumar.NOM sandwich ate 3.M. S Having felt hungry, Kumar ate a sandwich (2) Backward Control [Kumar ki aakali wees i] sandwic tinnaa Du [Kumar DAT hunger fall CNP] sandwich ate 3.M.S

PAGE 14

14 (3) Copy Control [Kumar ki aakali wees i] atanu/Kum ar sandwic tinnaa Du [Kumar DAT hunger fall CNP] he/Kumar.NOM sandwich ate 3.M.S Having felt hungry, Kumar ate a sandwich I analyze Adjunct Control as movement providing a detailed account of the conditions that driv e and constrain each type o f control I suggest that t he subject starts out in the adjunct before it moves to the matrix clause. The result is non distinct copies of the same element in both clauses Decisions regarding the pronunciation of copies take place on the phonological side of the computation. The p ronunciation of one copy only ( the matrix or adjunct copy) results in Forward or Backward Control. The pronunciation of both copies results in Copy Control.

PAGE 15

15 CHAPTER 1 INTRODUCTION 1. 1 Research Questions This study is based in the Minimalist Program of the Principles and Parameters approach to syntactic theory (Chomsky 1981, 1995, 2000, and Chomsky and Lasnik 1995). It explores a phenomenon of Adjunct Control in two South Asian languages: Telugu and Assamese. Adjunct Control is a relation of obligatory co referentiality between the subject in the matrix clause and the subject in the adjunct. Control has been a controversial issue in the Chomskyan Generative Grammar for a long time. Until recently, the assumption in the literatu re has been that control is a relation of co referentiality between an overt NP in a higher (matrix) clause and a silent NP in a lower ( subordinate ) clause, as sentences (1 2) illustrate The silent NP is symbolized by (1) [ MATRIX Tom i hopes [ SUBO RDINATE COMPLEMENT i to win]] (2) [[ MATRIX Tom i won] [ SUBORDINATE ADJUNCT without i knowing it]] These patterns are not the only attested ones, however. Other patterns do exist, leading to the following typology of control ( Polinsky and Potsdam 2006: 3 4). (3) Typology of Control a. Forward Control only the matrix NP is pronounced: [ M atrix NP i [ Subordinate i ] ] b. Backward Control only the subordinate NP is pronounced: [ M atrix i [ Subordinate NP i ] ] c. Copy Control both the matrix and subord inate NPs are pronounced: [ M atrix NP i [ Subordinate NP i ] ] Forward Control is the most researched. Its history goes back to the 6 (Chomsky 1965, Rosenbaum 1967) Backward Control is a less s tudied phenomenon. It has been investigated in a

PAGE 16

16 number of languages, including Japanese (Kuroda 1965, 1978), Tsez (Polinsky and Potsdam 2002), Malagasy (Polinsky and Potsdam 2003), and Korean (Monahan 2003). Copy Control is the least studied phenomenon. I t has been explored in Tongan (Chung 1978), and San Lucas Quiavini Zapotec (Lee 2003; Boeckx, Hornstein, and Nunes 200 7 ) Polinsky and Potsdam ( 2006 ) provide a survey Interestingly, all three types of control are attested in Telugu. Assamese, on the other hand, allows Forward and Copy Control, while the status o f Backward Control is less certain. The main questions that the study means to answer are the following: Research question 1: What are the syntactic characteristics of Adjunct Control in Telugu and Assamese? Research question 2: What ar e the mechanics invo lved in the derivation of the different types of control (Forward, Backward, and Copy) ? Research question 3: How does Adjunct Control contri bute to the analysis of Control in general? The rest of the chapter is organized as follows. Section 1. 2 presents th e domain of investigation of the study. Section 1. 3 lays out the theoretical background upon which the following chapters are built. Section 1. 4 provides a brief overview of the dissertation. 1.2 Domain of Investigation This study is mainly concerned with two South Asian languages: Modern Telugu 1 (hereafter Telugu), a Dravidian language and Assam ese, an Indo Aryan language The Indo Aryan and Dravidian language families are two major language families that share the South As ian subcontinen t They are also two of the top five largest language families in the world: the Indo Aryan languages have more than 640 million speakers (est. 1981) (Masica 1991: 1 Modern Telugu refers to the Telugu spoken from the 17 th Century to the present time. The Telugu spoken prior to 1100 AD) and Middle Telugu (1100 1600 AD) (Steever 1997: 8).

PAGE 17

17 8), and the Dravidian languages have more than 220 million speakers (est. 1991) (Steever 199 7: 1). 2 The Indo Aryan and Dravidian languages share a number of linguistic features F or example, they all have adverbial clauses known as Conjunctive Participle clauses ( Masica 2005 and Klaiman 1981, Chapter 4 ) These common features, however, are due t o cultural fusion rather than to a common ancestry. To elaborate, the Indo Aryan language family has Indo European origins. The origin of the Dravidian language family, on the other hand, has not Dravid ian describe the languages of South India (Krishnamurti 2003: Chapter 1). The fact that the two language families do not share a common ancestry is also confirmed by Masica (2005: 8) who describes the four major l anguage families of India Indo Aryan, Dravidian, Munda and Sino Tibetan Concerning Telugu it is a South Central Dravidian language. Of the twenty three 3 reported Dravidian languages Telugu has the l argest number of native speakers (more than 60 million). It is the official language of the State of Andhra Pradesh and one of the four officia l languages of the Indian Union. The three other languages are Kannada, Malayalam, and Tamil. In addition to bein g official, these four languages unlike the other Dravidian languages have long an d extensive literary traditions. For example, the first Telugu (Old Telugu) inscription dates to the late 6 th Century AD (Steever 1997: 1 8; Krishnamurti (1997: 202; 200 3: 19 23)). Assamese, also known as Asamiya, is the major language of the state of Assam in the far northeastern part of India. More than half of the population of Assam (c. 13 out of c. 22 million) 2 The I ndian subcontinent comprises at least four major language families the other two being Munda and Sino Tibetan 3

PAGE 18

18 speak Assamese as a native language and many others, bo th in Assam and in the neighboring states of Meghalaya, Arunachal Pradesh, and Nagaland, speak it as a second language (Goswami and Tamuli 2003: 393 394; Masica 1991). Assamese has a long liter ary tradition that arguably goes back to the 6 th or 7 th Centur y AD. However, the earliest literary work that is unmistakably Assamese date s to the 13 th Century AD (Goswami and Tamuli 2003: 397) The study focuses on one aspect of Telugu and Assamese, namely, Obligatory Control into a special type of non finite, parti cipial clauses known as Conjunctive Participle clauses. Obligatory Control is a relation of obligatory co referentiality between two arguments in a structure. One of the arguments occupies the matrix clause and is usually pronounced, while the other argume nt occupies a subordinate clause and is usually implied. Adjunct Control is control in which the two arguments are subjects, one of which occupies the matrix clause of a given structure and the second occupies an adjunct. Three types of Telugu and Assames e Adjunct Con trol are examined. These are Forward Control, in which only t he matrix subject is pronounced ( e.g., (4a) and (5a) ) ; Backward Control, in which only the subordinat e/adjunct subject is pronounced ( e.g., (4b) and (5b) ) ; and Copy Control in whi ch both subjects are pronounced ( e.g. (4c) and (5c) ) (4) Telugu a. Kumar i [ i/*k jwaram wacc i] hospital weLLaa Du Kumar.NOM [ fever come CNP ] hospital went 3.M.S b. i/*k [Kumar ki i jwaram wacc i] hospit al weLLaa Du [Kumar DAT fever come CNP ] hospital went 3.M.S c [Kumar ki jwaram wacc i] Kumar hospital weLLaa Du [Kumar DAT fever come CNP ] Kumar.NOM hospital went 3.M.S

PAGE 19

19 (5) Assa mese a. Ram e i [ i/*k xomoi no thak i] bhat na khal e Ram NOM [ time NEG keep CNP] rice NEG ate 3 b. ?? i/*k [Ram Or i xomoi no thak i] bhat na khal e [Ram GEN time NEG keep CNP] rice NEG ate 3 c [Ram Or xomoi no thak i] Ram e bhat na khal e [Ram GEN time NEG keep CNP] Ram NOM rice NEG ate 3 Altho ugh structures that involve a CNP clause are generally Obligatory Control structures, a few exceptions exist. For example, sentences (6) and (7) each involve a CNP clause. Yet, disjoint subjects are allowed. (6) Telugu [ war am paD i] janaalu taDici pooyaa ru [rain fall CNP] people.NOM wet became 3.M.P (7) Assamese [dhumuha ah i] bohut gos bhangil [storm ABS come CNP] many trees ABS broke The following chapters account for structures like (4) through (7) within syntactic theory. S ection 1.3 highlights relevant aspects of this theory. 1.3 Analytic Approach Following Hornstein (1999, 2003), I analyze Adjunct Control as an instance of movement whereby the subject is base generated in the adjunct before it moves to the matrix clause. The analysis of Adjunct Control requires answering two questions. First, w hat are the mechanics involved in the derivation of Telugu and Assamese Adjunct Control s tructures? Second, w hat are

PAGE 20

20 the mechanics involved in the pronunciation of either or both subjects in the different types of Adjunct Control structures that the Telugu and Assamese allow? The answer to the first question requires familiarity with the synta ctic theory related to control in general. Assuming the Movement Theory of Control (Hornstein 1999) and that the two subjects in an Adjunct Control structure are related via movement the answer to the second question is based in the broader phenomenon of multiple copy spell out, whereby more than one copy of the same token are pronounced in a single structure. The main task is to determine the factors that are decisive in the realization of copies, resulting in variation in Adjunct Control. I address thes e questions in Section 1. 3.2 through 1.3.4 In Section 1. 3.2 and 1. 3.3, I present a review of two opposing approaches to control theory: the PRO Theory of Control and the Movement Theory of Control. I show that the movement approach is more compatible with the Telugu and Assamese data. Section 1. 3.4 deals with the issue of multiple copy spell out. It brings to the fore the factors that may be decisive in the pronounciation of either or both subjects in the different types of Telugu and Assamese Adjunct Cont rol structures. First, however, an overview of the framework within which this study is based is appropriate. The study adopts the movement approach to control which has been made possible by c hanges in s yntactic theory during the 199 Section 1. 3.1 hi ghlights some major aspects of this theory and e xplain s how the changes came about 1. 3. 1 From Government and B inding to Minimalism: A n Overview This section is divided into two parts. Section 1. 3.1.1 offers an outline of the grammar within the Government and Binding framework as presented in (Chomsky 1981, 1986a, 1986b and Chomsky and Lasnik 1995). Section 1. 3.1.2 presents a summary of the grammar from the perspective of the Minimalist Program (Chomsky 1995, 2000, 2001, 2004), underlining the changes in th e theory along the way. In addition to the cited works, the discussion in both

PAGE 21

21 subsections has benefited extensively from Marantz 1995, Ouhalla 1999, Hornstein 2001, and Hornstein, Nunes, and Grohmann 2005. 1. 3. 1.1 The architecture of the gra mmar i n G overn ment and Binding G overnment and Binding assumes that all human beings are equipped with a Language Faculty, or a cognitive ability to acquire language. This Language Faculty comprises a computational system and a lexicon. The computational system selects i tems from the lexicon and forms a derivation in accordance with X bar Theory. Another assumption is that the grammar has four levels of representation: Deep Structure, Surface Structure, Logical Form (LF), and Phonological Form (PF). Deep Structure is an internal interface level that relates the computational system to the lexicon. At this level, lexical items are inserted into a phrase marker in accordance with the Projection Principle and Theta Theory The Projection Principle as stated in (8) ensures th at the Deep Structure thematic information is preserved at all four levels of representation. Theta Theory dictates that all thematic positions are filled. S ubsequent movement into a thematic position is disallowed as it violates the Theta Criterion (9). (8) Projection Principle Representations at each syntactic level (i.e. LF, and Deep and Surface Structure) are projected from the lexicon, in that they observe the Subcategorization properties of lexical items. (Chomsky 1981:29) (9) Theta Criterion Each argume nt bears one and only one theta role, and each theta role is assigned to one and only one argument. (Chomsky 1981:36) To illustrate, in order to derive a sentence like (10), the computational system selects the lexical items in (11) and inserts them in th e phrase marker in (12). (10) Sue arrived. (11) {Sue, arrived}

PAGE 22

22 (12) Deep Structure = [ CP C 0 [ IP [ I' I 0 [ VP arrived Sue]]]] Notice that arrive is an unaccusative verb that has one thematic position. This means that it requires only one argument. By leaving no thematic pos ition unoccupied, the derivation satisfies the Projection Principle. If a thematic position is left empty, the derivation crashes. Overt movement applies between Deep and Surface Structure. In this sense, Surface S tructure reflects the final word order of a structure. For example, the Surface Structure of (12) above is (13). By Surface Structure, Sue moves to Spec, IP in order to be assigned Case and to satisfy the EPP 4 leaving a trace behind. (13) Surface Structure = [ C P C 0 [ IP [ I' Sue i I 0 [ VP arrived trace i ] ]]] Notice that the movement of Sue obeys the Projection Principle which holds that thematic information has to be preserved at all levels of representation. Sue moves into a Case position but not into a new theta position. And by leaving a trace behind, t he thematic information encoded at Deep Structure is preserved. Further, Surface Structure is the level responsible for sending the derivation to the two external interface levels: PF and LF. These two levels are needed for pronunciation (form) and interpr etation (meaning) respectively. PF is interpreted by the sensorimotor system, providing the information needed for the phonetic interpretation/realization of a structure. LF is interpreted by the system of thought, providing the information needed for the semantic interpretation of a linguistic expression. A structure must satisfy Full Interpretation at PF and LF. Full Interpretation means that an appr o (Chomsky 1986: 98). A t PF, if the derivation has phonological information (e.g., a stress pattern) 4 EPP or the Extended Projection Princ iple dictates that all clauses must have subjects.

PAGE 23

23 that cannot be realized or interpreted by the sensorimotor system, the derivatio n crashes. Full Interpretation at LF is a little more complicated. This is the level at which cert ain syntactic conditions apply. For example, an argument is not allowed to move into a new thematic position because this means that the Deep Structure thematic information is not preserved at LF, which is a violation of the Projection Principle in (8 ) An other requirement is the Case Filter which dictates that an NP be C ase marked ( with abstract or morphological C ase) in order to be visible. Visibility has two facets (14 15). Notice that (14) is also a PF requirement. (14) An NP must be C ase marked in order to be pronounced (C homsky 1981: 49, Vergnaud 1982) (15) An argument, [or more appropriately, an argument chain ], must be C ase marked to be visible for theta role assignment (Chomsky and Lasnik 1995:46, following Joseph Aoun) Another property of Government and Binding is that it is a modular grammatical theory which holds that the grammar is made up of several modules: Case Theory, Binding Theory, Bounding Theory, Phrase Structure or X Bar Theory, Movement Theory, Control Theory, Theta Theory, and Trace Theory. Each module is distinct and subject to constraints and well formedness requirements. What is common to all of them is that they are all relational. T hey require interaction between two elements. For e xample, Case Theory requires a Case assigner and a C in (16) (based on Chomsky 1986a : 10 16 and Chomsky and Lasnik 1 995: 79) For example, the nominative C ase on the subject in (13) is assigned under Government by I 0 (16) only if a. is a head b. c. there is no barrier (mainly, a CP) that intervenes between and

PAGE 24

24 The purpose behind the different modules is to capture the more specific, more abundant, and seemingly unrelated grammatic al rules that describe individual syntactic s tructures and to capture them with more general grammatical principles. For example, grammatical rules that describe anaphoric relations among nominal expressions are realized as Conditions A, B, and C within Bi nding Theory. This trend to move from specific rules to general grammatical principles main focus of syntactic theory, even more so within the framework of the Mini malist Program as presented in Chomsky 1995 and further developed by Chomsky (2000, 2001, 2004) and by other researchers. 1. 3. 1.2 The architecture o f the grammar i n the Minimalist Program The Minimalist Program, as the name indicates, is an ongoing reducti onist project the grammar, preserving grammatical notions based on naturalness, simplicity, and e conomy, as defined in (17) ( Hornstein, Nunes, and Grohmann 2005: Chapter 1). (17) a. Naturalness implies that only notions that correspond to self evident facts about language should be preserved. b. Simplicity follows from naturalness. If only natural notions are preserved and all other theory internal notions are remo ved, the grammar becomes simpler. Further, given two theories A and B that are equal in every way except that A has fewer rules than B, A is considered superior. c. Economy is pertinent to derivations and derivational rules. Everything else being equal, a derivational step that requires the least effort (e.g., fewer steps) and that happens only when necessary (i.e., as a Last Resort) is optimal. Let us begin by examining the four Government and Binding levels of representation in accordance with (17). It is a fact about language that linguistic expressions are a combination of form and meaning. This fact justifies the preservation of the two external interface levels, LF and

PAGE 25

25 Deep Structure and Surfa removed fr originally made by Chomsky (1995), are accompanied by analyses which show that the grammar can not only do without these theory internal levels but also be better off without them. To illustrate, the idea that at Deep Structure the whole phrase marker of a linguistic expression is available all at once and that the thematic positions of the phrase marker are all filled before any movement takes place at Surface Structure is a purely theory internal idea. What is certain is that words are combined into phrases and that nominal expressions do receive a thematic interpretation at LF. Therefore, Deep Structure could be dispensed with in favor of a simple operation that brings lexi cal items together; call this operation Merge (Chomsky 2000: 101). Since it is evident that nominal expressions re ceive a thematic meaning ( agent, patient, roles can only The discussion suggests that the theory internal levels are superfluous. The Minimalist Program recognizes the problem, marking the end of the Deep and Surface Structure era and reducing the level s of representation from four to two. T his reduction is also a step towards simplicity. Everything else being equal, a grammar with two levels of representation is Minimalistically more desirable than a grammar w ith four. Now the questio n is: H ow does a derivation take place without Deep Structure and Surface Structure? Like Government and Binding, the Minimalist Program considers language to comprise a lexicon and a computational system. Preserving these two notions is also in line with natural ness.

PAGE 26

26 It is a fact that linguistic expressions are made of lexical items combined together to form phrases. These observations are a reason to believe that the computational system comp rises two operations: form a numeration and merge. Forming a nu meration means copying from the lexicon all and only the syntactic objects needed for the derivation. Thus the numeration for sentence (18) is (19): (18) Sue arrived (19) Numeration = {Sue 1 arrived 1 I 0 1 C 0 1 ) The indices in (19) show how many tokens of an item ar e copied from the lexicon. At the end of the derivation, all the items in the numeration must be exhausted. Further, no new features or items other than those in the numeration may be introduced during the derivation. This requirement is called the Inclusi veness Condition (Chomsky 2000: 113). Merge combines two objects together to form a new syntactic object. For example, Merge applies to the NP Sue and V 0 arrived in (19 ) above. V 0 projects, yielding (20). As indicated by the superscript, Sue receives a th eta role that is licensed by V 0 Subsequently, I 0 merges with VP, yielding (21). (20) [ VP arrived Sue ] (21) [I 0 [ VP arrived Sue ]] Another fact about language is that elements within a linguistic expression may be pronounced in one position and interpreted in anot her. Therefore, it seems that the computational system does not only select lexical items and combine them; it also moves them around. One straightforward example is the case of wh questions in English. For instance, what in (22) is pronounced sentence ini tially although it is interpreted as a complement of eat (22) What did you eat?

PAGE 27

27 This fact about language has led to the intuition that when an element moves, it does not really evacuate its site. More likely, it copies and merges leaving behind a copy that is available for interpretation at LF but that is usually deleted at PF. Therefore, it is more accurate to describe movement as a dual operation of copy plus merge (Chomsky 1995). Economy considerations (17c) constrain the applications of copy plus merge. Th e operation takes place so that a structure may be interpreted at PF and LF. In Minimalist term inology, lexical items (e.g., noun s) enter the derivation with features, some of which are interpretable (e.g., phi features) and some uninterpretable (e.g., Cas e). The latter cannot be interpreted at the interfaces and must be checked by an appropriate head before the derivation reaches LF and PF. Movement happens for the purpose of feature checking which renders uninterpretable features invisible at PF/LF 5 For example, in (23) arrived is an unaccusative verb that cannot check the Case feature of its complement. Sue moves (copy plus merge) to Spec, IP in order to check its Case feature, and I 0 projects. Finally, assuming that all complete sentences are CPs, a de clarative C 0 merges with IP, resulting in the structure in (24). (23) [ IP Sue Case [ I' I 0 [ VP arrived Sue ]]] (24) [ CP C 0 [ IP Sue Case [ I' I 0 [ VP [ VP arrived Sue ] ]] ] ] In Government and Binding, a syntactic object is shipped to LF and PF via Surface Structure. In th e Minimalist Program, an operation called Spell Out (or Transfer) does the job (Chomsky 2000: 118 119, Chomsky 2004: 115 116). Un like Surface Structure, Spell O ut is not a level of representation. It applies to (24) above, and the derivation converges at LF. At PF, the lower copy of Sue is deleted, resulting in (25). D eletion of copies is the topic of Section 1.3. 4. 5 Chomsky (2004) holds that feature checking can take place via Agree (a c command relationship between a Probe and a Goal) and that all movement happens for the purpose of the EPP.

PAGE 28

28 (25) [ CP C 0 [ IP Sue Case [ I' I 0 [ VP [ VP arrived Sue ]]]]] To summarize the Minimalist Program assumes that the gramm ar comprises a basic operation M erge and two interface levels LF and PF responsible for form and meaning Displacement is considered a fact about language. It takes place via a dual operatio n, copy plus merge, also kno wn as movement. Movement happens for a purpose, namely, feature checking. The reductionist project of Minimalism does not stop here. As pointed out in the previous section, Government and Binding is a modular grammar, made up of a number of modules as well as certain notions, such as Government. In the Minimalist Program, there is a serious 2000: 113). Therefore, optimally, government and the different modules must be eliminated. One radic al attempt along these lines is carried out by Hornstein (2001) who reviews the role of the modules, deeming them unnecessary and attributing all construal to movement. One module that This sec tion has presented an overview of the changes that took place in syntactic theory during the 199 The most relevant points are the following: (26) a. Deep Structure and Surface Structure are no longer part of the theory. Only the interface levels are preser ved. These are LF and PF. b. Merge is the basic structure building operation. It can apply to items selected in the numeration as well as to phrasal structures When combined with the operation copy, merge can also apply to an item already in the deriva tion. c. No new element can be introduced to the derivation if it is not originally available in the numeration (the Inclusiveness Condition).

PAGE 29

29 d. Only grammatical relations that are necessary for intepretation at t he interfaces must be preserved ( e.g., scope ) Grammatical relations that are made available only for theory internal reasons but are not necessary for inte rpretation should be eliminated ( e.g., Government ) The following section offers a presentation of how control structures are analyzed i n a Government and Binding framework and how the innovations of the Minimalist Program have made an alternative analysis possible. 1.3.2 Control in Government and Binding In Government and Binding, c ontrol constructions similar to (27 ) have generally been considered to comprise two base generated subjects, one upstairs and one downstairs. Both are available for interpretation at LF. The former is a lexical ized subject, whereas the latter is a silent PRO. The arguments are coreferential, a relation that is d etermined through co indexation. In other words, sentence (27) has the structure in (28). This approach is the PRO Theory of Control (27) [Sue tried [ to impress Tom ] ] (28) [ IP Sue i [ vP trace Sue tried [ IP PRO i to [ vP trace PRO impress Tom ] 6 In the early frame w ork of Government and Binding (Chomsky 1981, 1986a, 1986b ), PRO is presented as a Case less, phonetically null, and base generated NP that occupies the subject position of non finite clauses Later, based on the observation that a Case less PRO cannot be visible for theta role assignment and is thus a violation of the Case Filter as defined in (15) above, Chomsky and Lasnik ( 1995) hold that PRO is necessarily Case marked and that it is assigned a special type of Case they call Null C ase. P th which is licensed in a spec head relation between a non finite I 0 and PRO According to the authors, it is only logical that a minimal I 0 [ Tense, Finite] assigns minimal or Null Case 6 The structure anachr onistically contains vP which was not available in the earlier versions of Government and Binding.

PAGE 30

30 ( Chomsky and Lasnik 1995: 119 120). Null Cas e serves two jobs. B y virtue of being null, it dictates that the subject it marks be obligatorily unpronounced (Unlike the nominative Case of the Lexical DP Sue which makes it visible or pronounced) At the same time, b y virtue of being a C ase, Null Case q ualifies an argument chain for theta marking at LF. A movement approach to control as illustrated in (29) is simply not possible within the framework of Government and Binding. The traces in (29 ) ind icate that the matrix subject Sue starts out in one thema tic position in the embedded clause before moving to another thematic position in the matrix clause. This means that the same argument is assigned two theta roles, which is a violation of the Theta Criterion as formulated in (9), repeated here as (30). (29) [ I P Sue Case/EPP [ vP trace Sue 1/ 2 tried [ IP trace Sue EPP to [ vP trace Sue 1 impress Tom ] (30) Theta Criterion Each argument bears one and only one theta role, and each theta role is assigned to one and only one argument. (Chomsky 1981:36) A non movement/PRO approach as exemplified in (31 ) does not violate the Theta Criterion. The sentence assumes two external theta roles of the embedded clause before movi ng to Spec,IP in order to check Null Case and the EPP. The Sue in thematic Spec,vP of the matrix clause before moving to Spec,IP w here it checks nominative Case and the EPP. (31) [ IP Sue i Case/EPP [ vP trace Sue tried [ IP PRO i Null Case/EPP to [ vP trace PRO impress Tom ] The above discussion lays out the details about the distribution of PRO. With regard to its interpretation PRO l ike any NP trace is considered anaphoric. For example, in ( 31 ), PRO refers bac k to its antecedent Sue In other words, it satisfies Condition A which holds that an anaphor is bound in its governing category. However, structures like ( 32a b ) in which PRO is

PAGE 31

31 free are a violation of Condition A. In such sentences, PRO behaves like a no n anaphoric pronominal I t obeys Condition B according to which a pronoun is free in its governing category. (32) a. [PRO to escape when everybody is watching] is not a good idea. b. John wondered [how PRO to behave oneself in public] This bi polar qualit y of PRO led to the PRO Theorem in (33) which states that PRO is ungoverned simply because it occupies the subj ect position of a non finite CP. A non finite I 0 and presumably a non finite C 0 is too weak to govern PRO. Further, a CP, a ccordin g to Chomsky (1986a : 10 16), is a barrier. I n other words, elements inside CP or, more precisely, elements in IP complement of C 0 cannot be governed by a head outside CP and thus, cannot be bound, by a node higher than CP. Since PRO is ungoverned, it vacuously sati sfies both Conditions A and B. This leads to the dual nature of PRO. U nlike reflexives which are [+anaphoric, pronominal] or pronouns which are [ anaphoric, +pronominal], PRO can be both: [+anaphoric, +pronominal]. (33) PRO is ungoverned. The PRO Theorem was a cceptable when PRO was considered Case less in the 1981 1986 Government and Binding version of Control Theory. In the latest version of Government and Binding, however, PRO is Null Case marked as we saw above. Case assignment requires government by a Case assigning head. Therefore, to assume that PRO is Case marked and ungoverned at the same time is contradictory. This is not to mention that it is only stipulative to assume that I 0 and C 0 can govern the Specifier of a finite IP but not the Specifier of a no n finite IP, as Martin (2001: 142; fn. 3) and Watanabe (1996) point out. With the arrival of the Minimalist Program in the mid 199 the aforementioned challenges became even more problematic, and many of the Government and Binding assumptions that lead to the derivation in (31) became eithe r orthogonal or unavailable

PAGE 32

32 To elaborate, government has no place in Minimalism and can no longer be used as a tool for the interpretation of PRO. Concerning the distribution of PRO, the Theta Criterion that justifie s ) in favor of (31), is a Deep Structure requirement. The Minimalist Program recognizes two level s of representatin that are con s i meanin g PF and LF respectively. Accordingly, Deep Structure is eliminated from the theory. With the elimination of Deep Structure, the Theta Criterion as stated (30) is also done away with Whereas the idea that every argument must be assigned a theta role stil l holds, the restriction that an argument can be assigned one and only one t heta role no longer holds ( Polinsky and Potsdam 2002: 264 265 and works within). Based on the above observations, several researchers reth ought the distribution and interpretatio n of PRO within the framework of the Minimalist Program (e.g., Martin 1996, San Martin 2004, Landau 2000, 2004) At the same time, other researchers have seen it Minimali stically viable and desirable to eliminate PRO completely from the theory and resort t o movement instead. One such approach is known as the Movement Theory of Con trol 1995, Hornstein 1999, 2001, 2003) The following section spells out the details. 1.3.3 Control in the Minimalist Program 1.3.3.1 PRO Theory of Control The PRO Theory o f Control within the Minimalist Program has different incarnations all of which depart from the Government and Binding approach. Researchers have taken into account work on the distribution of PRO in other languages, such as Icelandic (Sigur sson 1991), R om an i a n, and Arabic, among others ( San Martin 2004 ). These lang uages show that PRO occupies a C ase position just like lexical DP s and that finiteness is not always d ecisive in the

PAGE 33

33 licensing of PRO. For example, control obtains even if an embedded clause is subjunctive rather than infinitival. To illustrate, floating quantifiers in Icelandic show agreement with the null subject in the embedded clauses of control structures. That is, the d ative floating quantifier llum in ( 34a ) i ndicate s that PRO is i tself d ative, whereas the g e nitive floating quantifier allra in (34b ) i ndicate s that PRO is g enitive (Sigur sson 1991: 331 332 (8c d)). (34) Icelandic a. vonast til the boys.NOM hope for [a PRO lei ast ekki llum [to PRO.DAT to be bored not all.DAT in school] b. vonast til the boys.NOM hope for [a PRO ver a allra geti r nnie]. [t o PRO.GEN be all.GEN mentioned in the speech] The boys hope to be all mentioned in the speech. In addition evidence from languages like Greek, Romanian, and Arabic shows that c ontrol into finite cl auses is possible (San Martin 2004: Chapter 4). The following sentence is an example from Standard Arabic. Notice that the embedded verb is subjunctive. (35) Standard Arabic l walad u [PRO an yan ] tried.3.M.S the child NOM [PRO SUBJ UNCTIVE succe ed.3.M.S ] The child tried to succeed. Accordingly, it has been suggested that the distribution of PRO is determined by factors other than Case and finiteness. These factors also determine the interpretation of PRO Different PRO theories have been propos ed in support of this view. In the rest of this section, I provide a synopsis of two theories, one proposed by Landau (2000, 2004 ) and one by San Martin ( 2004).

PAGE 34

34 Landau (2004, 2006) holds that lexical DPs and PRO are in complementary distribution The form er is R]. Both [+R] and [ R] are interpretable features on lexical DPs and PRO respective ly. The distribution of lexical subjects and PRO is distinguished by the Tense and Agreement features [T, Agr] on I 0 and C 0 automatically come to bear [+R] le feature on I 0 a feature that can be checked by lexical DP as it bears an interpretable [+R]. ny other feature constitution that is, [+T, Agr], [ T, +Agr], or [ T, Agr] is associated with [ that is assigned as an uninterpretable feature on I 0 and can only be checked by PRO which has an interpretable [ R]. In this sense, PRO cannot be substituted with a lexical DP, not because PRO is Case less or Null Case marked or even ungoverned, but because it is a null anaphor that can delete an uninterp retable [ R] feature that a lexical DP cannot (Landau 2006: 163 ). framework and much earlier work, it is a n I San Martin uses examples from Romanian Macedonain, Hungarian, Spa nish, Arabic, and Basque to argue that regardless of Case and Finiteness, it is the size of the complement that determines the type of control. If the matrix verb selects for an IP complement, the result is an Obligatory Control structure that obeys the c ross linguistic generalizations in (36) ( San Martin 2004: 48). (36) T he embedded subject of an Obligatory Control structure must be a strictly coreferential with the matrix controller and b phonetically null.

PAGE 35

35 Therefore, like La ndau, San Martin shows that the licensing of lexical DPs and PRO is divorced from finiteness and Case. Unlike Landau, however, San Martin (2004: 169 (85)) argues that it is the size of the complement clause that is decisive in the licensing process: (37) In com plement clauses, lexical subjects arise in CPs, whereas PRO is licensed in bare IPs. her IP complements have incomplete I 0 with [+T ense ] feature but no [Person] feature T his I 0 [+T ense Person] is able to check Case on the embedded subject, which normally shows on elements such as the floating quantifiers in Icelandic, but it is not able to license a lexical/overt DP. An indicati ve CP, however, licenses a lexical subject because it has a complete I 0 [+T ense +Person], with the [Person] feature being provided by C 0 Now the question is: How does PRO receive its interpretation? San Martin (2004: 207) building on Martin (1996), of fers the following explanation: The interpretation of PRO is derived as follows: PRO is a featureless element that is inserted off line in to the derivation as Last Resort (only when there is no DP left in the Numeration to sa turate the existing theta r oles ). Although it appears in a local relation to a Case assigning Probe [+T ense ], its defective nature makes it unable to host the Case Value. In order to prevent a FI [Full Interpretation] violation, the chain of PRO collapses to the most local chain that bi nds it, the subject or the object chain in Subject and Object Control respectively. This de rives the Control effect. What is it that accounts for the complementary distribution between PRO and lexical complements? Here San Martin follows evaluate d for interpretation once completed I f PRO occurs inside a CP phase, it is sent for interpretation before the chain it occ urs in gets the chance to collapse with a matrix NP chain.

PAGE 36

36 chain of PRO simply does not have a local well formed and interpretable chain with which it can collapse and the chain of PRO will violate F ull I nterpretability re different in several ways but they both sha re the standard view of control. T hey consider control structures to involve two coreferential argument chains, one of which is PRO. Crucially, PRO and lexical DPs are in complement ary distribution. Any theory of control that is built on this assumption is incompatible with the Telugu and Assamese data. Both languages h ave Copy Control structures (38a, 39a). In t hese structures both subject are obligatorily coreferential and, most i mportantly, pronounced. A silent subordinate subject, or what PRO theories consider as PRO, is also possible (38b, 39b) The examples in (38 39) violate the essence of PRO theory, namely, that PRO and lexical DP are in complementary distribution. (38) Telugu a [Kumar ki jwaram wacc i] Kumar hospital weLLaa Du [Kumar DAT fever come CNP ] Kumar.NOM hospital went 3.M.S b. Kumar i [PRO i/*k jwaram wacc i] hospital weLLaa Du Kumar.NOM [PRO fever come CNP ] hospi tal went 3.M.S (39) Assamese a. [Ram Or xomoi no thak i] Ram e bhat na khal e [Ram GEN time NEG keep CNP] Ram NOM rice NEG ate 3 b. Ram e i [PRO i /*k xomoi no thak i] bhat na khal e Ram NOM [PRO time NEG keep CNP] rice NEG ate 3 Further, Tel ugu allows Backward Control as (40) illustrates Note that the dative subject is licensed by the subor dinate clause. Compare with the Forward Control structure (39b) in which the nominative pronounced subject is licensed by the matrix clause. A PRO theory of control

PAGE 37

37 would hold that PRO appears in the matrix clause and the lexical DP appears in the subordin ate clause. In this case, the structure would involve an anaphoric PRO c commanding a correferential lexical DP (R expression), which should induce a Condition C violation, contrary to fact (Potsdam 2006, Polinsky and Potsdam 2002). (40) Telugu PRO i/*k [Kumar ki i jwaram wacc i] hospital weLLaa Du PRO [Kumar DAT fever come CNP ] hospital went 3.M.S The following section presents the relevant details of an alternative approach and shows that it is superior to the PRO theory, at least insofar as the Telugu and Assamese data are concerned. 1.3.3.2 The Movement Theory of Control As Section 1. 3.1 pointed out the Theta Criterion and the Projection Principle are the main reason s why G overnment and B inding rejects a movement approach to Cont rol. With the elimination of Deep Structure, the Theta Criterion is also eliminated. A n argument may be assigned more than one theta role without inducing any violation. Note, however, that every argument must still be assigned a theta role, and every theta role must be assigned to an argument. Nonetheless, these restrictions now follow naturally from Full Interpretation rather than from the Projection Principle and Deep Structure ( Brody 1993, 1995, and Polinsky and Potsdam 2002: 264). Criterion holds at LF only to the extent required for meaningful interpretation Now that multiple theta role assignment is possible, a movement app roach to c ontrol has become possible, and with it the derivation in (41). Note that copies replace traces, in accordance with the Copy Theory of Movement (Chomsky 1995) (41) [ IP Sue Case/EPP [ vP Sue 2 tried [ IP Sue EPP to [ vP Sue 1 impress Tom]

PAGE 38

38 The major departu re from the PRO tradition in the above derivation is the list of assumptions in (42) below (especially ( 42 a b) ; (42c d) were introduced earlier in the Minimalist Program ). These are the main grounds on which the PRO Theory of Control is abolished (Hornstei n 2003: 22 (40)) (42) a. Theta roles are features and can thus trigger movement b. There is no upper bound on the number of theta features that a DP can have. c. Movement is Greedy. d. Greed is understoo whereby an element moves to check a feature of its own or a feature of the target ( Lasnik 1995). Based on (42), sentence (41) is derived in this manner: Sue merges in Spec,vP of the embedded clause where it checks a theta role feature. It moves to Spec,IP to check the EPP feature of th e target. This is followed by movement to Spec,vP of the matrix clause where another theta role feature is checked. The last move is to Spec, IP. This is where the nominative Case feature is checked. Finally, the derivation is shipped to the interfaces via spell out. At PF, the lower copies of Sue are deleted for reasons to be specified. Theoretically, the movement approach to control is in line with the grammatical downsizing project of the Minimalist Program. A movement analysis does away w ith all unnecessary construal processes in Control T heory All construal is now attributed to movement Minimalistically, the Movement Theory of Control does not assume any levels of representation apart from LF and PF. No features or elements (e.g., indi ces) other than the ones in the numeration are inserted during the derivation, which satisfies the Inclusiveness Condition. The two arguments in a control structure are interpreted as coreferential for the mere reason that they are copies of the same token An argument starts out in the subordinate clause of a given structure. It copies out of the subordinate clause and merges in the matrix clause. The result is

PAGE 39

39 non distinct copies of the same argument in both clauses. Being non distinct, the two copies are co referential. Empirically, the theory is able to account for cases of control which are problematic for the PRO Theory. One such case is Backward Control. According to the movement approach, the Forward and Backward Control structures in (39b) and (40) look like (43a b). Both structures are alike. The only difference is that the higher copy of the subject is pronounced in the Forward Control structure (43a), while the lower copy is pronounced in the Backward Control structure (43b). No anaphoric elements are involved. (43) Telugu a Kumar [ Kumar ki jwaram wacc i] hospital weLLaa Du Kumar.NOM [ Kumar DAT fever come CNP ] hospital went 3.M.S b Kumar [Kumar ki jwaram wacc i] hospital weLLaa Du Kumar.NOM [Kumar DAT fever come CNP ] hospital went 3.M.S Further, the movement approach is potentially capable of accounting for Copy Control structures like (38a) and (39a), repeated as (44 45). From a movement perspective, (44 45) are derivationally similar to (43a b) in that they involve multiple copies of the same argument. The difference is that one copy survives deletion in the Forward and Backward Control structures (43a b), while two copies survive deletion in each of (44) and (45). Realization of multiple copies of the same token is not an isolated phenomenon that is unique to Copy Control. It is attested in other types of structures as we will see shortly. Compare to a PRO theoretic attempt at accounting for the fact that a necessarily silent PRO is realized as a lexical DP. My impression is that such an attempt, even if successful, will be unique to control, just as PRO itself as a syntactic object is unique to control.

PAGE 40

40 (44) Telugu [Kumar ki jwaram wacc i] Kumar hospit al weLLaa Du [Kumar DAT fever come CNP ] Kumar.NOM hospital went 3.M.S (45) Assamese [Ram Or xomoi no thak i] Ram e bhat na khal e [Ram GEN time NEG keep CNP] Ram NOM rice NEG ate 3 Having no The discussion in this section does not mean to imply that the Movement Theory of Control is without problems. On the theoretical side, a major premise in the movement approach, as we have seen, is that theta roles are features and can trigger movement. According to Chomsky (1995) theta roles are configurational in the sense that they are the result of relations between a head and its specifier or complement. Such relations, as maintained by Landau (2003), are traditionally acce ssed only at LF (which is reasonable, since LF is the level of interpretation) and not during the derivation. If this is correct, then theta roles cannot trigger movement. This means that movement into a theta position must still be disallowed. Despite the argu ments and counterarguments ( Boeckx a nd Hornstein 2004 ), the issue is theory internal and remains an open debate. On the empirical side, another problem with the Movement Theory of Control is that it attributes all types of control interpretation and c hoice of controller to the narrow syntax, dismissing those instances in which the choice of controller is determined by semantic and pragmatic factors. For example, from a movement perspective, sentence (46) below can only mean that the teacher will eventu ally go the restroom. It cannot mean that the student needs to go the restroom, which is also a possible meaning. (46) The student asked the teacher to go to the restroom

PAGE 41

41 The reason for incorrectly ruling out the latter interpretation is that movement can onl y target the closest possible site, a restriction that is known as the Minimal Link Condition (Hornstein 1999) In this sense, the embedded subject in (46) can only move to the matrix object position, as (47) shows. Movement to the farther matrix subject p osition, as illustrated in (48), induces a violation of the Minimal Link Condition. Landau ( 2003 ) and Cullicover and Jackendoff ( 2001 ) provide more arguments along these line s (47) Satisfying MLC The student asked the teacher [the teacher to go to the res troom]. (48) Violating MLC The student asked the teacher [the student to go to the restroom]. The conclusion is that the Movement Theory of Control seems to incorrectly try to promote a theory of control that is totally free from any semantic or pragmatic interference. This observation is an important caveat for any analysis that adopts the movement approach. For example, a movement account of Copy Control in Telugu and Assamese must leave some room for semantic and pragmatic interference in order to be abl e to account for structures like (49 50), repetion of (6 7) above. Although a syntactic account is viable, as chapter 6 show s semantic factors still play a role. (49) Telugu [war am paD i] janaalu taDici pooyaa ru [rain fall CNP] people.NOM wet became 3.M. P (50) Assamese [dhumuha ah i] bohut gos bhangil [storm ABS come CNP] many trees ABS broke Minimal Link Minimal Link Movement Movement

PAGE 42

42 H owever, if Backward Control and Copy Control are facts about natural languages, w hich (Chomsky 2000: 92), then the movement approach seems to have more chances of survival a s a theory of control. Problems are only normal in an on going project, and only time and more research can tell whether they are really a challenge to the movement approach or whether there are ways around them that still need to be worked out. One issue still needs to be addressed, namely, the pronunciation vs. deletion of copies in Adjunct Control. The different types of Telugu and Assamese Adjunct Control structures stand out as different due to one salient pr operty: the phonological nature of the subje cts. Forward Control involves a pronounced subject in the matrix clause and a silent subject in the subordinate clause. Backward Control is the mirror image of Forward Control; the matrix subject is silent and the subordinate subject is pronounced. Copy Co ntrol, on the other hand, includes two pronounced subjects. What factors are involved in the phonological (non )realization of the copies? This is the topic of the following section. 1.3.4 Multiple Copy Spell Out and the Realization vs. Deletion of Copies 1.3.4.1 Deletion of c opies In pre Minimalis m when an element moves, it leaves behind a phonetically null trace. The Minimalist Program does away with traces and adopts the Copy Theory of Movement (Chomsky 1995). Under this theory, when an element moves, it copies out of one position and merges in another. The outcome is two copies of the same token. For example, the man in (51) copies out of its theme position downstairs and merges in Spec,IP upstairs. The less marked situation is that only one copy, usua lly the higher/highest copy is pronounced. The lower copies are normally deleted.

PAGE 43

43 (51) [ CP [ IP [ DP the man] [ I' I 0 [ VP arrived [ DP the man ] ]]]] Why do the lower copies get deleted? Several answers have been pr oposed in the literature ( Brody 1995, Peset sky 1998, among several others). One answer that has received considerable support in the literature is provided by Nunes (1995, 2004) Building on work by Kayne (1994), Nunes provides a systematic analysis to account for the deletion of copies According to Kayne, linear order in a structure is a precedence relation that is regulated by heirarchical structure. If a non terminal X c commands a non terminal Y, but Y does not c command X, this means that X precedes Y. By the same token, every terminal that is dominated by X precedes the terminals that are dominated by Y. Kayne formulates this idea as the Linear Corre spondence Axiom (52). (52) The Linear Correspondence Axiom Let X, Y be nonterminals and x, y terminals such that X dominates x and Y dominates y. Then if X asymmetrically c commands Y, x precedes y. (Kayne 1994: 33) Nunes adopts the formulation in (52) and holds that the deletion of the lower copy in (51) takes place in order for the structure to be linearized in accordance with the Linear Corresponde nce Axiom. As the dotted arrows in (53) illustrate, DP in Spec,IP c commands the lower DP and thus precedes it. The fact that the lower copy is non distinct from the copy in Spec,CP means that the man precedes and follows itself, which induces a violation of irreflexivity as formulated in (54). At the same time, the verb arrived c commands and is c commanded by the man This means that it precedes and is preceded by the same element, which is a viola tion of asymmetry in (55) (Nunes 2004: 24). The se violations do not allow the structure to be linearized in accordance with the Linear Correspondence Axiom in (52). The reason is that no linear order can be established between the two copies of the man or between arrived and the man and consequently t he structure does not converge at PF.

PAGE 44

44 (53) IP 3 DP I' 6 3 the man I VP 3 V DP arrived 6 th e man (54) Irreflexivity If x precedes y, then x and y are distinct copies. (55) Asymmetry If x precedes y, y necessarily does not precede x. In order for the structure in (53) to be linearized, only one copy may be phonologically realized and the other copy must the violation, linear order can then be established, yielding an appropriate PF object (Nunes 2004: 25). Deletion of copies does not just happen, however. According to Nunes (2004), it is a PF o peration that is arranged for in the syntax. To elaborate, Nunes considers deletion of copies as a PF operation that applies to chain s. Chains, on the other hand, are formed in the syntax in accordance with the Conditions on Form Chain in (56) (Nunes 2004: 91 (4)). (56) Conditions on Form Chain m the non a. b. c. sublabel of the head of the projection with whic system; and d. n

PAGE 45

45 Conditions ( 56 a) and ( 56 b) are straightforward. In (53), for example, the two DPs are non distinct by virtue of being copies of the same token, the man brought about by movement as copy plus merge. Further, the copy in Spec,IP c commands the copy in VP. According to (56c d), the movement of the man has to take place as a last resort and in accordance with the Minimal Link Condition (Chomsky 1995). In other words, movement must result in feature checking. Also, there should be no other DP that may check the same feature and that c commands the man and is c commanded by Spec,IP. The two DPs in (53) satisfy the conditons in (56) and, accordingly, they form a chain At PF, deletion applies to the chain via the operation Chain R eduction in (57) (Nunes 2004: 27 (44)) prevents cases where more copies are deleted than is necessary in order for the structure to converge (Nunes 2004: 27). (57) Chain Reduction Delete the minimal number of constituents of a nontrivial chain CH that suffices for CH to be mapped into a linear order in accordance with the Linear Correspondence Ax iom. Therefore, applying (57) to the chain {[ DP the man], [ DP the man]} in (53) results in the deletion of one copy only, which usually is the lower copy, as (58) shows Nunes (2004: 30 38) holds that uninterpretable features are not legible at PF. If a copy survives deletion at PF, but it still has formal features, they have to be eliminated by an operation he calls Formal Feature Elimination. Normally, by the time the derivation is delivered to the phonological component, the higher copy has checked mor e formal/uninterpretable features than the lower copy. This means that the phonological realization of the higher copy requires less Formal Feat ure Elimination and is thus more economical. Consequently, the lower copy is marked for deletion.

PAGE 46

46 (58) IP 3 DP I' 6 3 the man I VP 3 V DP arrived 6 the man B efore proceeding to the following subsection, it is worth noting that according to Nunes (2004: 50 55) if two non distinct copies do not form a chain neither of them can be deleted. As I mentioned above, the reason is that the PF operation Chain Reduction is parasitic on the narrow syntax operation Form Chain. Consider the example in (59). The sentence involves coordination of vPs, as (59b) shows Assume that DP the pizza starts out in the lower conjunct before it moves to the higher one. The two copies are non distinct, but they do not enter a c co mmand relationship. Therefore, they cannot form a chain At PF, Chain Reduction can target neither copy. Accordingly, neither is deleted. The structure may not be linearized, however. According to the Linear Correspondence Axiom in (52), the two copies of the pizza are still in a precedence relationship. The reason is that the higher vP c command s the lower vP and thus all the terminals dominated by the higher vP precede all the terminals dominated by the lower vP. This means that if both copies are pronoun ced, the same element follows and precedes itself, which is a violation of the irreflexivity condition in (54). As a result, the structure must not converge, contrary to facts. To avoid this problem, Nunes maintains that DP the pizza and the pronominal it are base generated. That is, they are not related by movement and thus they are distinct copies. (59) a. Tom ate the pizza and paid for it. b. Tom [ CONJ P [ vP ate [ DP the pizza]] [ CONJ P' and [ vP paid [ PP for [ DP the pizza/it]]]]]

PAGE 47

47 Nevertheless, there exist struc tures in which two elements are related by movement that is, they are non distinct and yet they are pronounced without inducing a violation of the Linear Correspondence Axiom. The following section spells out the details. 1.3.4.2 Realization of multipl e c opies Although the unmarked situation is that only one copy in a given chain survives deletion upon linearization, structures in which more than one copy in a chain are pronounced are attested in several languages. Such structures are classified as inst ances of multiple copy spell out. Here are three examples from Romani ( McDaniel 1986, in Nunes 2004: 38 (72)) Frisian ( Hiemstra 1986, in Nunes 2004: 38 (73)) and San Lucas Quiavin Zapotec (Lee 2003: 102 (83)) Each example includes two non distinct copi es in a c command relationship. Yet, they both escape deletion without violating the Linear Correspondence Axiom. (60) Romani Kas misline kas o Demri dikhl kas ? whom you.think whom Demir saw kas ? (61) Frisian wr tinke jo w Jan wennet wr ? Where think you where that Jan lives Where ? (62) San Lucas Quiavin Zapotec R g auh bxaady HABITUAL want Mike IRREALIS eat (Mike) grasshopper According to Nunes (2004: 40), structures like (60 62) are possible only if one of the copies is a head and it adjoins to another head. By PF, the two heads undergo morphological re analysis, or morphological fusion in the sense of Hall e and Marantz (1993), and the copy becomes a part of the node that dominates it. At PF, the structure is linearized. Following Chomsky (1995: 337), Nunes assumes that the Linear Correspondence Axiom does not apply

PAGE 48

48 word internally. Therefore, the fused copy cannot be detected as non distinct from the other copies in the chain so it escapes deletion. To illustrate, Nunes holds that the intermediate copy of the wh element in each of (60) and (61) adjoins to C 0 of the subordinate clause on its way to Spec, CP of the matrix clause. The wh element and C 0 are morphologically re analyzed as a single terminal node of the form [ C 0 WH [ C 0 C 0 ]] and therefore the wh (Nunes 2004: 40) or, as Kandybowicz (2 006) argues, it becomes distinct adjoined cop ies. Chain Reduction is s ubject to economy restrictions; that is it applies only if the phonological realization of a copy prevents the structure from being linearized. In (60) and (61), the only wh copy that leads to such consequences is the lowest copy and is therefore deleted. The intermediate copy, on the other hand, no longer causes the derivation to crash, which is why it is not deleted. Note that the morphological fusion of the intermediate wh copy and C 0 into [ C 0 WH [ C 0 C 0 ]] can be morphophonologically detected in (61). Whereas the highest and lowest copies are wr case, however, as (60) and (62) imply. According to Nunes, sentence (60) is similar to (61), except for one difference: the wh element in (60) adjoins to a null head. Sentence (62) is an example of Copy Control. Boeckx Hornstein, and Nunes (2007 ) analyze it as movement. The lower copy copies out of the subordinate clause and merges in the matrix clause. At the end of the derivation, the two non distinct copies form a chain At PF, the lower copy adjoin s to a covert head, a reflexive s elf affix 7 The copy and the head form a new word. Linearization cannot see into words. This is why the copy escapes deletion. 7 Evidence for this affi x comes from structures like (i pronoun. This phenomenon is attested in C hinese, as Boeckx, Hornstein, and Nunes (2007 ) point out. For example,

PAGE 49

49 In this section, I set the stage for the rest of the study. Most crucially, I highlighted the major premises of the Movement Theory of Control and explain ed why it is superior to other alternatives with respect to Adjunct Control in Telugu and Assamese. In addition to the determines the deletion or pronunciation of copies in Adjunct Control. This section delineated deletion and pronunciation of copies do not happen randomly but are restricted by certain conditions at the syntax phonology interface. Other theoretical assumptions will be discussed as needed in the body of the dissertation. S ection 1. 4 provides a brief overview of the following chapters. 1.4 Structure of the Study The following chapters are organized as follows. Ch apter 2 is a descriptive o verview It highlights the Telugu and Assamese morphosyntactic characteristics that are most relevant to the topic of Adjunct Control. These include word order, Case, and types of clauses. It also lays out the Adjunct Control data that are analyzed in subsequent chapters. A major contribution of this chapter is that it presents evidence of Copy Control in Telugu and Assamese, a phenomenon that to my knowledge, has not been documented in grammar books (e.g., Goswami 1982; Krishnamu rti and Gwynn 1985). sentence (i) (from Lee 2003: 84 (1)) is similar to the Chinese example (ii). T Chinese is ov ert, while in San Lucas Quiavin Zapotec it is covert. (i) S an Lu cas Quiavin Zapotec R HABITUAL like Mike Mike (ii) Chinese Mama hen xihuan mama ziji Mother very like mother self likes herself.

PAGE 50

50 Chapter s 3 and 4 provide a detailed analysis of the different types of Adjunct Control. I follow Hornstein (1999, 2003) and Nunes (1995, 2004) and consider Adjunct Control as being derived via sideward movement. This is a special type of movement that allows an element to copy out of one phrasal structure (e.g., an adjunct/subordinate clause) and merge into another unconnected and independently formed phrasal structure (e.g., a matrix clause). Chapter 3 focuses on Forward and Backward C ontrol. Both types share two characteristic s First, the adjunct merges with the matrix clause at vP Second, only one subject is phonologically realized Chapter 4 deals with Copy Control. In this case, the adjunct merges with the matrix clause at CP, an d both subjects are pronounced. The chapter places Copy Control within the bigger picture of multiple copy spell out. I argue that two copies in a Copy Control structure survive deletion because one of them becomes part of a bigger phonological word a s pelled out domain that is opaque to linearization and is inaccessible to Form Chain and Chain Reduction. In order for this to be possible, spell out and linearization have to apply cyclically (Uriagereka 1999). Chapters 3 and 4 remain silent about why movement takes place in Telugu and Assamese Adjunct Control. Given that movement in Minima lism has to happen for a reason, Chapter 5 addresses this issue and shows that the Minimalist view of movement as being triggered by feature checking is not sufficien t to account for the data under examination. Building on work by Pesetsky and Torrego (2006), I suggest that the subject moves in order to make up for a defect in the adjunct. I show that the adjunct in Adjunct Control structures is a non goal and suggest that its merge is disallowed unless it is licensed by the movement of its subject. Chapter 6 deals with the apparent exceptions to Adjunct Control in (6 7) above. These are structures that pattern the same as Adjunct Control structures, but they allow dis joint subjects.

PAGE 51

51 The chapter argues that these exceptions are instances of Expletive Control that have the same derivational history as the other more common instances of Adjunct Control The major difference is that the subjects in Expletive Control struct ures are null expletives rather than arguments. Chapter 7 is a summary and a conclusion. It summarizes the findings of the dissertation and highlights the major theoretical implications.

PAGE 52

52 CHAPTER 2 ADJUNCT CONTROL IN T ELUGU AND ASSAMESE : A DESCRIPTIVE OVE RVIEW 2.1 Introduction T his chapter present s a detailed de scription of the phenomenon of Adjunct C ontrol in Telugu and Assamese. The focus is on subject control into non finite clause known as conjunctive participle or adverbial clause To set the scene, the chapter also outlines the aspects of Telugu and Ass amese morphosyntax that are relevant to the phenomenon in question. The following sections are organized in this manner. Section 2 .2 offers a general linguistic overview of Telugu and Assamese. Section 2. 3 presents a descriptive survey of Case, especially as related to Subject NPs. Section 2. 4 is a brief description of finite clauses in both languages, with a special focus on agreement. Section 2. 5 delineates the characteristics of non finite clauses, d rawing a distinction between non finite subordinate clauses that do not enforce a control interpretation and conjunctive participle clauses that do. Section 2. 6 highlights the different types of adjunct control that are allowed in each language. These are Forward Control (Section 2. 6.1), Backward Control (Section 2. 6.2), and Copy Control (Section 2. 6.3). Exceptions to the phenomenon are presented in Section 2. 6.4. Section 2. 7 con c ludes the chapter with a summary. 2. 2 Linguistic Overview Telugu and Assamese are subject pro drop head final, SOV language s ( Krishnamurti 1997, 2003; Goswami and Tamuli 2003) That is both ( a) and (b ) of (1 2) are grammatical: (1) Telugu a. atanu Sarita ki ninna aa pustakamu iccaa Du he.NOM Sarita DAT yesterday that book gave 3.M .S b. pro Sarita ki ninna aa pustakamu iccaa Du pro Sarita DAT yesterday that book gave 3.M.S

PAGE 53

53 (2) Assamese a. xi azi ratipuwa PrOxad ok e khon kitap dil e he NOM today morning Pr oxad ACC one CL book gave 3 b pro azi ratipuwa PrOxad ok e khon kitap dil e pro today morning Proxad ACC one CL book gave 3 He gave Proxad a book this morning. Although the canonical word order in both languages is SOV, OSV is also possible. In fact, apart from the position of the verb which is usually fixed, any constituent can be sentence initial in a topic position as the sentences in (3 4) demonstrate (3) Telugu a. Kumar Sarita ki ninna aa pu stakamu iccaa Du Kumar.NOM Sarita DAT yesterday that book gave 3.M.S b. Sarita ki Kumar ninna aa pustakamu iccaa Du c. ninna Kumar Sarita ki aa pustakamu iccaa Du d. aa pustakamu Kumar Sarita ki ninna iccaa Du (4) Assamese a. Ram e azi ratipuwa PrOxad ok e khon kitap dil e Ram NOM today morning Proxad ACC one CL book gave 3 b. azi ratipuwa Ram e PrOxad ok e khon kitap dil e c. PrOxad ok Ram e azi ratip uwa e khon kitap dil e d e khon kitap Ram e azi ratipuwa PrOxad ok dil e The immediate pre verbal position in both languages is a focus position. For example, the subject may occupy a preverbal position for emphatic purposes (e.g., (5 b) and ( 6 b) ) At the same time question words, which are focal elements, occupy a preverbal position (e.g., (5c ) and ( 6c ) ) Note, however, that question words may also be pronounced in situ (e.g., (5d ) and ( 6d ) ).

PAGE 54

54 (5) Telugu a. Kumar Sarita ki phone ceesaa Du Kumar.NOM Sa rita DAT phone did 3.M.S b Sarita ki Kumar phone ceesaa Du Sarita DAT Kumar.NOM phone did 3.M.S c Sarita ki ewaru phone ceesaa Du Sarita DAT who.NOM phone did 3.M.S d ewaru Sarita ki phone ceesaa Du who.NOM Sarita DAT phone did 3.M.S (6) Assamese a. Ram e mor ghorto bhangil e Ram.NOM my house destroy 3 b. mor ghorto Ram e bhangil e my house Ram NOM destroyed 3 c mor ghorto kone bhangil e my house who.NOM destroyed 3 d. kone mor ghorto bhangil e who.NOM my house destroyed 3 The following section delineates the ma in characteristics of Case in Telugu and Assamese, focusing mainly on Case marked subject NPs. 2.3 Case in Telugu and Assamese: A Descriptive Overview Case in Telugu and Assamese is a morphological a nd syntactic category. Morphologically, C ase marking in both languages is agglutinative in nature. Syntactically, a n NP must inflect for C ase in order to be used in a sentence ; its inflection determines its function ( Krishnamurti 2003; Goswami and Tamuli 2003: 319; Masica 1991: 230 236).

PAGE 55

55 Since this study is co ncerned with subject control into adjuncts, the focus in the following subsections is mainly on the Case of Telugu and Assamese subject NPs. These can be Structural Case marked (e.g., nominative) or Inherent Case marked (e.g., dative). Structural Case is a ssociated with grammatical relationships. F or example, although the subject of passive constructions in English is a theme, it is Structural Case marked nominative because it is a subject. Inherent Case, on the other hand, is associated with a theta role. F or instance, an experiencer subject NP that is, an NP whose physical or emotional state the predicate describes is Inherent Case marked dative in Telugu and genitive in Assamese. 2. 3 .1 Telugu Case Marked Subject NPs Telugu is a nominative accusativ e language. The subject can be Case marked nominative or dative depending on the predicate. Nominative is th e unmarked c ase. I n other words, a nominative NP is also the basic stem. Dative on the other hand, is formed by adding a l ayer II suffix to the ob lique stem. According to Krishnamurti (2003: 218) and Masica (1991) an oblique suffix in Dravidian languages is a l ayer I morpheme that makes an NP e ligible to receive layer II c ase markers. Take the Telugu noun s (Table 2 1) T he nom inative form and t he basic stem coincide. The dative c ase marker as well as the other postpositions (e.g. locative) are added to the oblique stem which i s also the genitive form. Krishnamurti and Gwynn ( 1985, Ch apters 9 and 26 ) provide a more detailed d escription of morphological case in Telugu Table 2 1 Some types of c ase in Telugu Case Form Basic Stem/Nominative illu kukka Oblique Stem/genitive Varies iNTi kukka Dative ku/ ki iNTi ki kukka ku/ki Locative loo iNTi loo kukka loo

PAGE 56

56 In Telugu, unless the predicate is an experiential predicate (a predicate that expresses a physical or emotional state e.g., hunger or anger or possession), the subject is Structural Case marked nominative in which case the v erb agrees with it (e. g., (7a) and (7 b) ) (7) a. Kumar Naatyamu ceesaa Du Kumar.NOM dance did 3.M.S b. caalaa mandi Naatyamu ceesaa ru many people.NOM dance did 3.M.P In sentences with ex periential predicates, a dative NP funct io ns as the subject. For example, it fun ctions as an antecedent to an anaphor or as the unpronounced argument of the subordinate clause in a control construction. 1 In ( 8 a b ), Kumar is dative Case marked because koopam waccindi and aakali westondi (8c ), Kumar functions as the antecedent to the anaphor tana ku tana meedhe In (8d), the dative NP is the unpronounced controllee (presented as ), which can only be the subject. (8) a. Kumar ki koopam waccin di Kumar DAT anger came 3.N.S b. Kumar ki aakali weston di Kumar DAT hunger feel 3.N.S c. Kumar ki tana ku tana meede koopam waccin di Kumar DAT him DAT him upon anger came 3.N.S d. Kumar i i koopam raawaTam 2 ] ishTapaDa Du Kumar.NOM anger coming ] 3.M.S 1 The unpronounced argument of a subordinate clause in a con trol structure can only be a subject. However, Dubinsky and Hamano (2006) analyze a case of control in which the unpronounced argument is a possessor NP. 2 RaawaTam waccin di

PAGE 57

57 Unlike nominative NPs, however, dative NPs do not trigger agreement on the verb. In the above sentence, the verb agrees with koopam aakali Kumar ki For further illustration, the sentences in (9) below also have dative subjects. T he verb in (9a ) agrees with the singular NP paniwaaDu agrees with the plural NP paniwaallu (9) a atani ki paniwaaDu unna Du he DAT servant is 3.M.S b. atani ki paniwaallu unna ru he DAT servants are 3.M.P 2. 3 .2 Assamese Case Marked Subject NPs Like Telugu, Assamese is a nominative accusative language (contra Amritavalli and Sarma 2002 ). Whereas the subject in Telugu m ay be Case marked nominative or dative, depending on the predicate, the s ubject in Assamese may be case marked nominative absolutive, genitive, or accusative ( Goswami 1982, Nath 2001, and Goswami and Tamuli 2003 ) And as we will see shortly nominative is further split into two categories: nominative and experiential nominativ e Assamese Case marked NPs experience minimum morphophonemic variation, although pronouns seem to be more susc eptible to such variation ( Table 2 2 ). Table 2 2 Some types of c ase in Assamese Case Form the Nominative e/ manuh zan e xi A bsolutive manuh zan xi Accusative (a)k manuh zan ak ta k Genitive Or manuh zan or ta r Nominative subjects occur with transitive predicates (e.g., (10) ) and unergative predicates (e.g., (11 ) ) Absolutive subjects occur with unaccusative predicat es (e.g., (12 ) ).

PAGE 58

58 (10) a. Ram e khotha tu xunil e Ram NOM news CL heard 3 b. Ram e bhat khal e Ram NOM rice ate 3 (11) a. Ram e nasil e Ram NOM danced 3 b. Ram e aahil e Ram NOM came 3 (12) a PrOxad xu i goil Proxad ABS sleep INF went b. bohut manuh moril many people ABS died Accusative subjects on the ot her hand, are a rare phenomenon. T hey verb l ag Goswami and Tamuli 2003: 432). Sentence (13a) is an example. Note, however, that (13b) is an alternative with a nominative subject. Accusative subjects will not receive further attention in this stu dy. (13) a. Ram ok toka lag e Ram ACC money want 3 b. Ram e pani bisaril e Ram NOM water wanted 3 Ram wanted water. Genitive subjects occ ur with experiential predicates. They are experiencers whose emotional or physical s ta te the predicate describes (e.g., (14a b ) )

PAGE 59

59 (14) a Ram Or khong uthil Ram GEN anger raised b Ram Or thanda lagil Ram GEN cold felt In addition, g entive subjects an d possessions (e.g., (15a b) ) (Nath 2001: p. 21 (20 21)). U nlike Telugu experiential predicates, Assamese predicat es do not show agreement in non nominative subject constructions. (15) a. Ram Or du khan haat ase Ram GEN two CL hands has b. Ram Or du ta laguwa asil Ram GEN two CL servant had Ram had two servants. Li ke dative subjects in Telugu, genitive subject s in Assamese fun ction as antecedent s to anaphors (e.g., (16a ) ) and are the unpronounced argument s in control structures (e.g., (16b ) ) (16) a. Ram Or niz Or uporot khong uthil Ram GEN self GEN above/on anger raised b. Ram e i i thanda lagabo] ni bisar e Ram NOM cold feeling] NEG want 3 A further note on experiencer subjects is in order for the purpose of this study. Compare the sent ences in (17). While both (17a) and (17b ) are somewhat synonymous, s entence ( 17b ) implies a more conscious effort on the part of the subject. Using kor more volitional. The same observation applies to (18a b): (17) a. Ram Or e ta buddhi khelal Ram GEN one CL idea played

PAGE 60

60 b. Ram e e ta buddhi koril e Ram NOM one CL idea did 3 (18) a. Ram Or phurti lagil Ram GEN exhilaration felt b. Ram e phurti koril e Ram N OM exhilaration did 3 To elaborate, e xperiential predicates with kor (e.g., (19 a) and ( 20 a ) ) The same expressions make non volitional experiential predicates unacceptable (e.g., (19b ) and ( 20b) ) (19) a. Ram e jani buji e ta buddhi koril e Ram NOM knowingly one CL idea did 3 b. *Ram Or jani buji e ta buddhi khelal Ram GEN knowingly one C L idea played ea occurred to Ram on purpose (20) a. Ram e jani buji khong koril e Ram NOM knowingly anger did 3 b. *Ram Or jani buji khong uthil Ram NOM knowingly anger did 3 Nevertheless, this observation does not deprive the nominati ve subjects in (17b) and (18b), as well as in (19a) and (20a), of being experiencers on a par with their genitive counterpart s According to Abbi (1991), experiential predicates can be divided into at least three categorie s: State Experiential Process Experiential and Stative Action Process Experiential The first and second types describe a physi cal, mental, or emotional state ( )

PAGE 61

61 Th xperiencer is in a certain state or condition with respect to an 256). khong uth i l anger raised and phurti lagil exhilaration be classified as state or process predicate s which Case mark their subject s genitive. The predicates khong kor i le phurti korle exhilaration Stative Action Proces s Experiential pre dicates which C ase mark their subject experiential nominative ( EXP NOM) The two types of predicates have a difference in meaning: khong uth i l anger raised and phurti lagil exhilaration respectively, w hile khong kor i le phurti korle exhilaration The Case assigned by either type of predicat e is related to the theta role e xperiencer regardless of the morphological materializati on. The reason why experiential nominative subjects are not considered simply nominative is based on empirical grounds. The two types of nominative subjects exhibit different behaviors in Adjunct Control structures as we will see in Section 2. 6.3 below. 3 The following section provides a brief description of finite clauses in Telugu and Assamese. The focus is mainly on the agreement behavior of finite predicates in each language. 2.4 Finite Clauses i n Telugu a nd Assamese 2.4.1 Finite Clauses i n Telugu Fini te declarative clauses in Telugu take verbs that are inflected for tense and agreement. The verb may belong to one of the following morphological paradigms: past, non past (future or habitual), or negative (future or habitual). Concerning agreement, verbs inflect for person, gender 3 Note that the distincti on between nominative and experiential nominative subjects may very well be exte nded to Telugu. However, the two types of subjects do not behave differently with respect to Adjunct Control. This is why the distinction, if it exists, is not discussed.

PAGE 62

62 (only with 3 rd person), and number. For example, the verb ammu in (21a c) if used in a declarative finite clause with a 3 rd person feminine subject. Compare to (21d f) in which the verb agrees wit h a 3 rd person masculine subject. The variation in suffix forms is morphophonological ( Kirshnamutri 1997 : 216 221) 4 (21) Telugu a. amm in du sell PAST 3.N.S b. ammu tun di sell NONPAST 3.N.S c. amm a du sell NEG 3.N. S d. amm aa Du sell PAST 3.M.S e. ammu taa Du sell NONPAST 3.M.S f. amm a Du sell NEG 3.M.S 2.4.2 Finite Clauses i n Assamese Finite clauses in Assamese have verbs that are inflected for a spect, t ense, and a greement, in this order. There are three types of aspect in Assamese: imperfective is habitual and perfective which collapses with the simple past into one portmanteau morpheme il Tense is also divided 4 Neg ative past verbs are compounds made of the stem + lee (i) ammu lee du sell was not 3.N.S

PAGE 63

63 in to three categories: p ast il p resent which is associated with the stem itself and f uture ib Regarding agreement verbs inflect for person (1 st 2 nd and 3 rd ) and honorificity ( only with 2 nd Person as [ +, or honorific ] ) Assamese verbs do not infl ect for gender or number. For example, any of the forms of the verb likh in (22) below can be used in a finite clause to agree with a 3 rd person, singular or plural, feminine or masculine subject. The variation in ( 22 c) and ( 22 e) is morphophonol ogical (Goswami and Tamuli 2003:422 423). 5 (22) Assamese a. likh e write HABITUAL 3 b. likh ib a write FUTURE 3 c. likh is e write IMPERFECTIVE 3 d. likh il e write PAST 3 e. likh is il write IMPERFECTIVE PAST 3 Variation in tense and/or aspect in finite clauses does not have an effect on Adjunct C ontrol. This is why most of the examples from both languages will depict one tense form: the past. The following section provides a descriptive overview of non finite subordinate clauses. The focus is on adjuncts. 5 Aspect in both Telugu and Assamese can also be expressed by forming compound stems. An exa mple from Assamese is bhangi thak Non lies in the first stem. The second stem is only aspectual, which is expected in a head final language (Krishnamurti and Gwynn 1985: C h. 16; Goswami and Tamuli 2003: 425 430; and Masica 1991: 270 271).

PAGE 64

64 2.5 Non Finite Clauses in Telugu and Assamese Telugu and Assamese have two types of non finite subordinate clauses that func tion as adjuncts. The first type is what I will refer to as infinitive clauses (INF clauses) The second type is known as adverbial clauses or conjunctive participle clauses (CNP clause s) ( Linholm 1975 and Klaiman 1981). Section 2. 5 .1 d eals with INF clause s. Section 2. 5 .2 delineates the characteristics of CNP clauses, which are the main domain of investigation of the current study. 2. 5 .1 Infinitive Clauses in Telugu and Assamese An INF clause in Telugu and Assamese comprises a non finite verb t hat Case mar ks its subject like any finite verb. The subject of an INF clause does not have to be coreferential with the subject of the matrix clause. The following subsections spell out the details. 2. 5 .1.1 Infinitive c lauses in Telugu In Telugu, the verb of an INF clause may take one of the forms in (23 27). As the examples illustrate, the first three forms (23 25) are participial adjectives that need an overt complementizer in order to function in subordinate clauses. The forms in (26 27), on the other hand, do no t take an overt complementizer. Nevertheless, since their behavior with regard to control patterns with the behavior of the forms in (23 25), I group all five forms together. (23) Past Participle: Verb stem + ina ; e.g., a. cepp ina maaTa => b. cees ina tarwaata => c. [Kumar bhojanamu tayaru cees ina tarwaata] Sarita tinnaa di [Kumar.NOM dinner prepare do INF after] Sarita.NOM ate 3.N.S d. [Kumar ki jwaram wa cc ina anduku] [Kumar DAT fever come INF because] Sarita (atani ki) mandulu iccin di Sarita.NOM (him DAT) medicines gave 3.N.S

PAGE 65

65 (24) Future or Habitual: Verb stem + ee ; e.g., a. wacc ee eeDu => b. cees ee mundu => c. [Kumar pani ki well ee mundu] Sarita ki fon ceesaa Du [Kumar.NOM work go INF before] Sarita DAT phone did 3.M.S d. [raatri ayy ee sariki] baaluD u nidra pooyaa Du [night come INF by the time] child sleep happen 3.M.S (25) Durative: Verb stem + tunna / Tunna ; e.g., a. was tunna waaDu => b. cees tunna appudu => c. [Kumar sinima cuus tunna appuDu] Sarita popkorn tinna a di [Kumar.NOM movie watch INF while] Sarita.NOM popcorn ate 3.N.S d. [tana bharta too naaTyam cees tunna appuDu] [her husband with dance did INF while] Sarita (atani ki) kathalu ceppin di Sarita (him DAT) stories told 3.N.S (26) Conditional: Verb stem + tee ; a. amm itee => b. [Kumar kofi kalip i tee] Sarita taagutun di [Kumar.NOM coffee mix INF] Sarita NOM will drink 3.N.S c. [Kumar iNTiki tondaraga wos tee] [Kumar.NOM home early come INF] Sarita aanand istun di Sarita.NOM happy will be 3.N.S (27) Concessive: Verb stem + inaa a. amm inaa =>

PAGE 66

66 b [Kumar iNTiki tondaraga wacc ina] Kumar.NOM home early come INF Sarita aananda paDalee du Sarita.NOM happy 3.N.S c [Kumar manci kofi kalip inaa] Sarita taaga lee du [Kumar.NOM good coffee make INF] Sarita.NOM 3.N.S 2. 5 1.2 I nfinitive c lauses in Assamese In Assamese, the subordinate non finite verb may take several forms, depending on the intended meaning. Following are three examples. The first form in (28) is a nominal or gerundive that is normally Case marked. It is followed by an overt complementizer when used in an INF clause. The forms in (29 30) do not take an overt complementizer. All three forms have the same characteristics with respect to control: no cont rol interpretation is required. (28) Nominal: Verb stem + a a. thak a => b. thak a r karone => c. [Ram Or tini ta loguwa thak a r karone] [Ram GEN three CL servant keep INF GEN because] xi / tar ghoiniyek e ghoro r kam no kor e he / his wife NOM house GEN work NEG do 3 b. [loratu e bhalkoi nas a r karone] [boy NOM well dance INF GEN because] tar mak Or bhal lagil his mother GEN good felt boy danced well, the mother fel (29) Contingent: Verb stem + te a. kha te => b. [Ram e bhat kha te ] [Ram NOM rice eat INF]

PAGE 67

67 xi Proxad ok gai thaka xunil e he.NOM Proxad ACC sing keep heard 3 c. [Ram e ga te] Proxad e nasil e [Ram NOM sing INF] Proxad NOM danced 3 (30) Future conditional: Verb stem + (i)le a. kor ile => b. [Ram e ga ile] Proxad e nasib a [Ram NOM sing INF] Proxad NOM will dance 3 c. [PrOxad Or bhok lag ile] xi bhat khawib a [Praxad GEN hunger strike/feel INF] he.NOM rice will eat 3 The following section introduces another type of T elugu and Assamese subordinate clauses, namely, conjunctive participle or CNP clauses. 2. 5 .2 Conjunctive Participle Clauses in Telugu and Assamese C onjunctive Participle clauses in the Indian Subcontinent are a defining characteristic which South Asian la nguages inherited from Sanskrit (Dwarikesh 1971). In Telugu and Assamese, like in most South Asian languages, CNP clauses are non finite clauses with no (overt) complementizer T hey express an action that is anterior to or simultaneous with that of the fin ite clause. Further, the CNP verb shows no inflection for tense or agreement. Although the CNP clause and the matrix clause might have a cause effect relation, they can be fairly translated into English as two clauses joined by the conjunction and Despit e this conjunctive nature, however, they behave like adverbial subordinate clauses for example, unlike conjuncts, they may be embedded within another clause whose predicate they functionally modify which is why they are considered adverbial participl e clauses or adjuncts ( Masica

PAGE 68

68 2005: 110, Haspelmath 1995 and Jayaseelan 2004, among several others). The following sections spell out the details. 2. 5 .2.1 C onjunctive Participle c lauses in Telugu In Telugu, a CNP clause contains a verb that takes one of the forms in (31) ( Kr ishnamurti and Gwynn 1985, ch.18 ) (31) Participial: Verb stem + i ; e.g., a. wacc i => b. Kumar [jwaram wacc i] hospital weLLaa Du Kumar.NOM [fever come CNP] hospital went 3.M.S c. Sarita [niiLLu kaac i] tea tayaru ceesin di Sarita.NOM [water boil CNP] tea prepare did 3.N.S d Sarita ki [aa maaTa win i] koopamu waccin di Sarita.D [that matter hear CNP] an ger came 3.N.S (32) Durative: Verb stem + tuu/ T uu ; e.g., a. cees tuu => b. Kumar [Sarita too naTyamu cees tuu] Kumar.NOM [Sarita with dance do CNP] aami ki kata ceppaa Du h er DAT story told 3.M.S c. Kumar [atani bhaarya to diner cees tuu] Kumar.NOM [his wife with dinner take CNP] Arun ki fon ceeddaam anukunnaa Du Arun DAT phone making decided 3.M.S 2. 5 .2.2 Conjunctive Participle c lauses in Assamese Assamese CNP verb s, unlike their Telugu counterparts, have a single form, presented in (33).

PAGE 69

69 (33) Verb Stem + i ; e.g., a. thak i => b. Ram e [xomoi no thak i] bhat na khal e Ram NOM [time NEG keep CNP] rice NEG ate 3 c. Ram [bhagar lag i] xui thakil Ram ABS [exhaustion feel CNP] sleep kept Ram d. Ram Or [train dhoribo no ar i] khong uthil Ram GEN [train catch NEG able CNP] anger raised Not being able to catch the train, Ram got angry. The following section show s that CNP clauses are subordinate clauses despite their c onjunctive meaning. 2. 5 .2.3 The subordinate nature of CNP c lauses As I mentioned earlier, semantically CNP clauses may denote a conjunctive meaning. Syntactically, however, they behave like adverbial clauses. For one thing they do not obey the Coordinat e Structure Constraint T his constraint disallows extraction of an element out of a conjunct (Ross 1967, cited in Kehler 1996). To illustrat e, whereas (34a) is grammatical, (34b ) is unacceptable because an N P is extracted out of a conjunct. (34) a. Tom ate a sa ndwich and drank a soda. b. What did Tom eat a sandwich and dr i nk _______? In order to prove that CNP clauses are not conjuncts, we need to show that they do not obey the Coordinate Struct ure Constraint. Before doing s o however, we have to make sure th at conventional conjuncts in Telugu and Assamese actually obey the Coordinate Structure Constraint Examples (35 36) indicate that they do. The (a) sentences in ( 35 36 ) are grammatical just like ( 34a ) above ; the (b) sentences are ungrammatical for the same reason ( 34b) is (35) Telugu a. Kumar oka pustakam konnaa Du Kumar.NOM one book bought 3.M.S

PAGE 70

70 mariyu oka magazine cadiwaa Du and one magazine read 3.M.S b aa magazine Kumar oka pustakam konnaa Du that magazine Kumar.NOM one book bought 3.M.S mariyu _______ cadiwaa Du and _______ read 3.M.S (36) Assamese a ram e kitap e khon kinil e Ram NOM book one CL bought 3 aru alosani e khon porhil e and magaz ine one CL read 3 b alosani e khon ram e kitap e khon magazine one CL Ram NOM book one CL kinil e aru _______ porh il e bought 3 and _______ read 3 T he Coordinate Structure Constraint can be violated without affecting grammaticality if there is a cause effect relation b etween the conjuncts (e.g., (37a b) ) (Kehler 1996: 2 (5), from Lakoff (1986)) This point is important because many of the Telugu a nd Assamese constructions we are dealing with may imply a cause effect relation and might turn out to be grammatical for the wrong reasons (37) a The guys in the Caucasus drink this stuff and live to be a hundred. b asus drink and live to be a hundred. In terestingly for Telugu and Assamese conventional conjuncts. I n other words, even if the relation between the conjuncts is that of cause and effect, extraction still in duces ungrammaticality To illustrate, the (a) s entence s in (38 39 ) are coordinate structure s. The two conjuncts in each sentence can be considered as a sequence of

PAGE 71

71 ts results in ungrammaticality, as (38b) and (39b) indicate. (38) Telugu a. Kumar ki koopamu waccin di Kumar DAT anger got 3.N.S mariyu Naa illu kuulcaa Du and my house destroyed 3.M.S ed b Naa illu ni Kumar ki koopamu waccin di My house EMPH Kumar DAT anger got 3.N.S mariyu ________ kuulcaa Du and ________ destroyed 3.M.S (39) Assamese a. ram Or khong uthil Ram GEN anger raised aru mor ghorto bhangil e and my house destroyed 3 b. mor ghorto ram Or khong uthil my house Ram GEN anger raised aru ______ bhangil e and ______ destroyed 3 N ow we tu rn to structures with CNP clauses to see if they violate the Coordinate Structure Constraint. I both semantically and syntactically. Otherwise we can fairly assume that they are subordinat e clauses, as t he data s eems to indicate. Sentences ( 40 a ) and (41 a) contain a CNP clause each. They can read as ( 40 b ) and (41 b) respectively and still be grammatical. That is, they are acceptable despite the

PAGE 72

72 (40) Telugu a Kumar [koopamu wacc i] Naa illu kuulcaa Du Kumar.NOM [anger get CNP] my house destroyed 3.M.S b Naa illu ni Kumar [koopamu wacc i] _____ kuulcaa Du My house EMPH Kumar.NOM [anger get CNP] __ ___ destroyed 3.M.S (41) Assamese a Ram e [khong uth i] mor ghorto bhangil e Ram NOM [anger get CNP] my house destroyed 3 b mor ghorto Ram e [khong uth i] _____ bhangil e my house Ram NOM [anger get CNP] _____destroyed 3 Violating the Coordi n a te Structure Constraint is one way to prove that CNP clauses are in Coordinate clauses do not normally overlap. In other words, one conjunct cannot break the continuity of another conjunct. A subordinate clause, on the other hand, may be embedded in the matrix clause, breaking its con tinuity. The sentences in (42 43 ) indicate that the CNP clause ma y be realized either outside (42a 43a) or inside (42b 43 b) the matrix clause. Notice that the pronounced subject in each of the sentences is Case marked nominative by the matrix predicate Th e CNP predicate in (42 ) would Case mark its subject dative while the CNP predicate in (43 ) would Case mark its subject genitive. (42) Telugu a [aakali wees i] Kumar bhojanamu tayaru ceesikunaa Du [hunger fall CNP] Kumar.NOM dinner prepare did f or self 3.M.S b Kumar [aakali wees i] bhojanamu tayaru ceesikunaa Du Kumar.NOM [hunger fall CNP] dinner prepare did for self 3.M.S

PAGE 73

73 (43) Assamese a. [ananda lag i] Ram e pagolor nisine nasil e [happiness feel CNP] Ram NOM like a crazy person danced 3 b Ram e [ananda lag i] pagolor nisine nasil e Ram NOM [happiness feel CNP] like a crazy person danced 3 Based on the above discussion, I consider CNP clauses as subordinate clauses. More s pecifically, they are adjuncts, or adverbial subordinate clauses, whose function is to modify the matrix predicate (Haspelmath 1995: 3, Masica 2005: 110 ). The following section presents Adjunct Control data from Telugu and Assamese. These data will be the subject of analysis in the following chapters. 2.6 CNP Clauses and Adjunct Control One relevant feature of CNP clauses is that they obey what is called the Same Subject Condition (Klaiman 1981: 88) or the Common Subject Requirement (Linholm 1975: 30). This means that the unpronounced subject of the CNP clause and the subject of the matrix clause are obligatorily coreferential, and that a sentence with a CNP clause is an instance of Obligatory Control. In other words, the (b) sentences in (44 47 ) are infelicitous under the designated reading, even though the (a ) sentences are provided as context or prior knowledge. (44) Telugu a Sarita i niillu kaacin di Sarita.NOM water boiled 3.N.S b Kumar i niillu kaac i] tii tayaru ceesaa Du water boil CNP] tea prepare did 3.M .S (45) Telugu a. Sarita i sinima cuusin di Sarita.NOM a movie watched 3.N.S

PAGE 74

74 b. Kumar i sinima cuus tuu] popkorn tinnaa Du Kumar.NOM movie watch CNP] popcorn ate 3. M.S he (46) Assamese a. Proxad i bhalkoi gal e Proxad well sang b Ram Or i bhalkoi ga i] bhal lagil Ram GEN well sing CNP] good felt (47) Assamese a Proxad i xomoi no thakil Proxad ABS time NEG kept b Ram e i xomoi no thak i] bhat na khal e Ram NOM time NEG keep CNP] rice NEG ate 3 Th is obligatory coreferentiality qualifies Telugu and Assamese sentences with CNP clauses as control constructions. Typologically, there are at least three types of control: For ward Control (48a) Backward Control (48b) and Cop y Control (48 c) (Polinsky and Potsdam 2006). In Forward Control constructions, the matrix subject is pronounced, while the subordinate subject is implied. In Backward Control constructions, the opposite is true. In Copy Control constructions, both subjects are pronounced. (48) a. Forward [ Matrix [ Subordinate Subject ] [ Matrix Subject ]] b Backward [ Matrix [ Subordinate Subject ] [ Matrix Subject ]] c. Copy [ Matrix [ Subordinate Subject ] [ Matrix Subject ]] Telugu shows evidence for all three ty pes of control delineated in (48 ). Assame se shows evidence for Forward and Copy Control; Backward Control constructions are either dispreferred or unacceptable. I begin with Forward Control.

PAGE 75

75 2. 6 .1 Forward Control in Telugu and Assamese Both Telugu and Assamese show evidence for Forward Control in to CNP clauses. The following subsections provide multiple examples and point out the pecularities of each language. 2. 6 .1.1 Forward Control in Telugu Observe sentences (49 5 3 ) below These are instance s of Forward Control in which the matrix subject is pronounced, determining the identity of the unpronounced CNP subject. (49) Kumar [ Kumar ki aakali wees i] Kumar.NOM [ Kumar DAT hunger fall CNP] bhojanamu tayaru ceesikunaa Du dinner prepare did for self 3.M.S (50) Kumar [ Kumar ki jwaram wacc i ] hospital weLLaa Du Kumar.NOM [ Kumar DAT fever come CNP ] hospital went 3.M.S (51) Kumar [ Kumar ki koopamu wacc i] Naa illu kuulcaa Du Kumar.NOM [ Kumar DA T anger get CNP] my house destroyed 3.M.S (52) ??Sarita ki [ Sarita aa maaTa win i] koopamu waccin di Sarita DAT [ Sarita.NOM that matter hear CNP] anger came 3.N.S (53) ??Kumar ki [ Kumar bazaaru loo Sarita ni cuus i] Kumar DAT [ Kumar.NOM market in Sarita ACC see CNP] Sarita miida preema kaligin di Sarita on love occurred 3.N.S h e market, Kumar fell in love with h Notice that in all of the above sentences the Case marking of the CNP subject would be different from the Case marking of the matrix subject. It is worth noting, however, that different Case marking is not a requirement. The above examples have been selected in order to make obvious that the structures are instances of Forward Control. To illustrate, the matrix subject and

PAGE 76

76 the CNP subject in (54) below take on the same Case; they are both nominative. The same applies to (59), except that bo th subjects are dative. In both cases, Forward Control still obtains. (54) [ Kumar unnu Sarita diner cees tuu] [ Kumar.NOM and Sarita.NOM dinner take CNP] Kumar unnu Sarita okari ki okaru kathalu ceppukumaa ru Kumar.NOM and Sarita.NOM one to one stories te ll 3.M.P (55) ?? [ Kumar ki juttu uuDipooy i] Kumar ki picci paTTin di [ Kumar DAT hair lose CNP] Kumar DAT craziness caught A point about the d egraded sentences (52), (53), and (55) above is in order. Notice that in these sentences the matrix subject is dative unlike the matrix subjects in (49), (50), (51), and (54) which are nominative. For reasons I have not been able to determine, Adjunct Co ntrol constructions that involve a dative subject in the matrix clause were judged as awkward by my Telugu consultants. Structures (56) through (59) are additional examples that some speakers found degraded, while others found them totally unacceptable. 6 (56) ??/*[ Kumar atani bhaarya to diner cees tuu] [ Kumar.NOM his wife with dinner take CNP] Kumar ki Nidra waccin di Kumar DAT sleep came 3.N.S (57) ??/*[ Kumar waana loo taDis i] [ Kumar.NOM rain in get wet CNP] Kumar ki daggu u jalubu u waccin yi Kumar DAT cough and cold and came 3.N.P 6 Everything else being equal, sentences which involve the matrix predicate koopamu waccin di (52), were not judged as degraded as the other structures that involve a da tive subject in the matrix clause (e.g., (56 59).

PAGE 77

77 (58) ??/* Kumar ki [ Kumar illu pooy i] picci paTTin di Kumar DAT [ Kumar. NOM house lose CNP] craziness caught (59) Sarita ki [ Sarita France loo perig i] Sarita DAT [ Sarita.NOM France in grow up CNP] French baagaa waccin di French well came 3.N.S When faced with constructions like (56 59), the consultants preferred to use an INF clause instead of a CNP clause. F or example, (60) was considered more acceptable than (57). Another alternative was to express the same id ea using a matrix predicate that licenses a nominative subject; compare (61) to (58). (60) [waana loo taDis ee sariki] Kumar ki daggu u jalubu u waccin yi [rain in get wet INF since] Kumar DAT cough and cold and came 3.N.P (61) [ Kumar illu pooy i] Kumar picci waaDu ayyaa Du [ Kumar.NOM house lose CNP] Kumar.NOM a crazy man became 3.M.S It is hard to explain the reasons behind the restriction a nd the var iation among speakers. I t is even harder to explain why some sentences that involve a CNP clause and a dative subject in the matrix clause are more acceptable than others. At the same time, it is worth mentioning that the phenomenon is not exactly unique. Whereas Telugu seems to disprefer dative subjects in the matrix clauses of control structures, Hindi d isallows dative subjects in subordinate clauses. This fact is illustrated in (62). Sentence (62a) has an experiential predicate and a dative subject. The same structure is used as a non finite clause in (62b). T he sentence is ungrammatical (Davison 1993: 47 49; (3) and (6a)). I will make no attempt at explaining this restriction in this study. I present the relevant data for completeness.

PAGE 78

78 (62) Hindi a. pitaa ko o apnee bhaaii par kroodh aataa hai father DAT self brother on anger come is b. pitaa [bhaaii par kroodh aa naa] nahii caahtaa father [brother on anger come (non finite)] not want 2. 6 .1.2 Forward Control in Assamese Assamese also shows evidence for Forward Control However, i t does not display restriction s concerning the type of predicate that is allowed in the matrix clause. Sentences (63 69) below are examples of Forward Control in which only the matrix subject is pronounced (63) Ram e [ Ram Or xomoi no thak i] bhat na khal e Ram NOM [ Ram GEN time NEG keep CNP] rice NEG ate 3 (64) [ Ram Or e ta bhal buddhi khela i] Ram e phurti koril e [ Ram GEN one CL good idea play CNP] Ram NOM party did 3 (65) Ram [ Ram Or bhagar lag i] xui thakil Ram ABS [ Ram GEN exhaustion feel CNP] sleep k ept (66) [ Ram e kam tu kor i] Ram gusi gol [ Ram NOM work do CNP] Ram ABS away went (67) [ Ram e kukur tu heru i] Ram Or dukh lagil [ Ram NOM dog CL lose CNP] Ram GEN sad felt (68) [ Ram e loteri jik i] Ram Or phurti lagil [ Ram NOM lottery win CNP] Ram GEN exhilaration felt (69) Ram Or [ Ram e phurti kor i] bohk lagil Ram GEN [ Ram NOM exhilaration do CNP] hunger felt

PAGE 79

79 T he above examples are selected with a CNP subject and a matrix subject that are Case marked differently. Neve rtheless, this is not mandatory .The following examples sho w that the subjects can both be nominative (e.g., (70) ) or both genitive (e.g., (71 ) ). (70) [ Ram e loteri jik i] Ram e notun ghor kinil e [ Ram NOM lottery win CNP] Ram NOM new hous bought 3 (71) [ Ram Or e ta bhal buddhi khel i] Ram Or bhal lagil [ Ram GEN one CL good idea play CNP] Ram GEN good felt 2. 6 .2 Backward Control in Telugu and Assamese As indicated in (49b) above, repeated here as (72), Backward Control is the case when the subordinate/CNP subject is pronounced and the matrix subject is implied. (72) Backward Control [ Matrix [ Subordinate Subject ] [ Matrix Subject ]] Telugu Backward Control structures are grammatical as originally observed by Subbarao ( 20 04). Assamese Backward Control structures, however, are not as acceptable for my speakers contra Subbarao ( 2004) 2. 6 .2.1 Backward Control in Telugu Obs erve the Telugu sentences in (73 76). These are the same as the Forward Control c onstructions in (49 53) above, except for one difference: the pronounced subject in each sentence belongs to the CNP clause, making it an instance of Backward Control. (73) Kumar [Kumar ki aakali wees i] Kumar.NOM [Kumar DAT hunger fall CNP] bhojanamu tayaru ceesikunaa Du dinner prepare did for self 3.M.S (74) Kumar [Kumar ki jwaram wacc i ] hospital weLLaa Du Kumar.NOM [Kumar DAT fever come CNP ] hospital went 3.M.S

PAGE 80

80 (75) ?? Sarita ki [Sarita aa maaTa win i] koopamu waccin di Sarita DAT [Sarita.NOM that matter hear CNP] anger came 3.N.S (76) ?? Kumar ki [Kumar bazaaru loo Sarita ni cuus i] Kumar DAT [ Kumar.NOM market in Sarita ACC see CNP] Sarita miida preema kaligin di Sarita on love occurred 3.N.S I mentioned in Section 2. 6.1.1 that Adjunct Control constuctions with a dative su bject in the matrix clause are uncommon in Telugu. Such constructions are mostly considered degraded or unacceptable. Some are considered even more unacceptable when realized as instances of Backward Control. For example, the Backward Control structures (7 7 78) are judged as more degraded than their Forward Control counterparts (56 57) above. (77) Kumar ki [Kumar atani bhaarya to diner cees tuu] Kumar DAT [Kumar NOM his wife with dinner take CNP] Nidra waccin di sleep came 3.N.S (78) Kumar ki [ Kumar waana loo taDis i] Kumar DAT [ Kumar NOM rain in get wet CNP] daggu u jalubu u waccin yi cough and cold and came 3.N.P 2. 6 .2.2 Backward Control in Assamese Whereas Backward Control structures are acceptable in Telugu, they are not as easy to obtain in Assamese. If the Forward Control structures in (63 69) above are presented as instances of Backward Control, their ac ceptability drops dramatically, as sentences (79 84) below indicate. Notice that the pronouced subject is licensed by the CNP clause. The sentences are considered dispreferred to unacceptable for exactly this reason.

PAGE 81

81 (79) ??/* Ram e [Ram Or xomoi no thak i] bhat na khal e Ram NOM [Ram GEN time NEG keep CNP] rice NEG ate 3 (80) ??/* [Ram Or e ta bhal buddhi khel i] [Ram GEN one CL good idea play CNP] Ram e phurti koril e Ram N OM party did 3 (81) ??/* Ram [Ram Or bhagar lag i] xui thakil Ram ABS [Ram GEN exhaustion feel CNP] sleep kept (82) ??/* [Ram e kukur tu heru i] Ram Or dukh la gil [Ram NOM dog CL lose CNP] Ram GEN sad felt (83) ??/* [Ram e loteri jik i] Ram Or phurti lagil [Ram NOM lottery win CNP] Ram GEN exhilaration felt (84) ??/* Ram Or [Ram e phurti kor i] bohk lagil Ram GEN [Ram NOM exhilaration do CNP] hunger felt Now consider the acceptable Backward Control structure in (85) below. Compared to the structures (79 84) ab ove, (85) seems to stand out as an exception. A closer examination, however, shows that Assamese native speakers are likely to process the sentence as an instance of Forward Control. Here is why: the CNP subject is nominative, while the matrix subject is a bsolutive. The demarcation between these two types of Case, nominative and absolutive, is not as underscored as, say, between nominative and genitive. As a matter of fact, nominative does substitute absolutive in some instances, as already pointed out by E dwards (2003). According to Edwards, nominative Case is indicative of more responsibility on the part of the subject. To illustrate, compare (86a b) (from Edwards 2003: 53; (71a b)). Ram is more responsible for his

PAGE 82

82 death in (86b) than he is in (86a). Note that the verb in (86b) does not show agreement with the nominative subject, although in some instances it may. (85) [Ram e kam tu kor i] Ram gusi gol [Ram NOM work do CNP] Ram ABS away went (86) a. Ram aksident ot mori l Ram ABS accident in died b Ram e bhiri khua r karone moril Ram NOM cigarette smoking GEN because died All this is to indicate that (85), repeated below as (87a), is more likel y to be interpreted by Assamese native speakers as (87b), whereby the subject is licensed by the matrix clause. This may explain why it is not considered unacceptable on a par with the other instances of Backward Control. (87) a. [Ram e kam tu kor i] gusi g ol [Ram NOM work do CNP] away went b. Ram e [ kam tu kor i] gusi gol Ram NOM [ work do CNP] away went The following section presen ts evidence for the less studied phenomenon of Co py Control. 2. 6 .3 Copy Control Both Telugu and Assamese show evidence of a cross linguistically rare phenomenon of control: Copy Control. Copy Control constructions involve a matrix subject and a CNP subject that are, not only obligatorily co referential, but also both pronounced. Such structures are possible under the following conditions: Condition 1 : The CNP clause is sentence initial. Condition 2 : The CNP subject is an R expression (non pronominal).

PAGE 83

83 2. 6 .3.1 Copy Control in Telugu The sentences in (88 93) are the Copy Control equivalent of the Telugu Forward and Backward constructions in the previous subsections. In all of the sentences, the CNP clause is sentence initial, and the CNP subject is an R expression. Note that the CNP and matrix subjects may be Case marked differently (e.g., (88 89) and ( 92 93) ) or the same (e.g., (90 91 ) ). Sentences (91 93) are degraded because they involve a dative subject in the matrix clause. (88) [Kumar ki i aakali wees i] [Kumar.DAT hunger fall CNP] atanu i /Kumar bhoj anamu tayaru ceesikunaa Du he.NOM /Kumar.NOM dinner prepare did for self 3.M.S (89) [Kumar ki i jwaram wacc i ] [Kumar.DAT fever come CNP ] atanu i /Kumar hospital weLLaa Du he.NOM /Kumar.N OM hospital went 3.M.S (90) [Kumar i sinima cuus tuu] [Kumar.NOM movie watch CNP] atanu i /Kumar popkorn tinnaa Du he.NOM /Kumar.NOM popcorn ate 3.M.S (91) ??[Kumar ki i juttu uuDipooy i] [Kumar DAT hair lose CNP] atanu ki i /Kumar ki picci paTTin di he DAT /Kumar DAT craziness caught (92) ??[Sarita i aa maaTa win i] [Sarita NOM that matter hear CNP] Sarita ki k oopamu waccin di Sarita DAT anger came 3.N.S

PAGE 84

84 (93) ??[Kumar i bazaaru loo Sarita ni cuus i] [Kumar.NOM market in Sarita ACC see CNP] atanu ki i /Kumar ki Sarita miida preema kaligin di he DA T /Kumar DAT Sarita on love occurred 3.N.S The two subjects in the above examples are obligatorily co referential. Disjoint subjects result in ungrammaticality, as sentences (94 96) show (94) *[Kumar ki i aakali wees i] [Kumar.DAT hunger fall CNP] atanu k /Rao bhojanamu tayaru ceesaa Du he.NOM /Rao.NOM dinner prepare did 3.M.S he/ (95) *[Kumar ki jwaram wacc i ] [Kumar.DAT fever c ome CNP ] aame/Sarita (atani ki) mandulu iccin di she.NOM/Sarita.NOM (him DAT) medicines gave 3.N.S she/ (96) *[Kumar i sinima cuus tuu] [Kumar.NOM movie watch CNP] atanu k /Rao popkorn tinnaa Du h e.NOM /Rao.NOM popcorn ate 3.M.S he/ It is important to note that the grammatical sentences (88 93) are judged by the Telugu consultants as redundant, but not unacceptable. According to them, pronouncin g one of the subjects is sufficient. In other words, Forward or Backward Control can do the job. Redundancy is eliminated if the matrix subject is pronounced as an epithet and/or if the sentence is made longer. To illustrate, sentences (97 98) make use of an epithet in the matrix clause. Sentence (99) is longer compared to (88 93) above. None of these sentences is judged as redundant.

PAGE 85

85 (97) [Kumar ki koopamu wacc i] aa pichooDu akkadi nunci wellipoyinaa Du [Kumar DAT anger come CNP] that idiot.NOM there f rom left 3.M.S (98) [Kumar illu pooy i] aa pichooDu picci waaDu ayyaa Du [Kumar.NOM house lose CNP] that idiot.NOM a crazy man became 3.M.S (99) [Sarita enimid inTik i bhojanamu tayaru ceesikun i] [Sarita.NOM eight time dinner prepare do for self CNP] aame/Sarita tommid inTiki tinnaa di she.NOM/Sarita.NOM nine time ate 3.N.S I m entioned at t he beginning of this sub section that Copy Control ob tains if two conditions are met. The CNP clause has to be sentence initial, and t he CNP subject has to be an R expression (non pronominal). Concerning the first condition if the CN P clause is realized sentence internally, a Copy Control construction becomes ungrammatical, as illustrated in (100). This observation holds whether an R expression is used in the CNP clause or in the matrix clause (e.g., (100a) and ( 100 b) ) It also holds if an epithet is used (e.g., (100c) ) (100) a. atanu/Kumar [atanu ki/ Kumar ki aakali wees i] he.NOM/Kumar.NOM [he DAT/Kumar.DAT hunger fall CNP] bhojanamu tayaru ceesikunaa Du dinner prepare did for self 3.M.S b. atanu/Kumar [atanu/Kumar Sarita too naTyamu cees Tuu] he/Kumar.NOM [he NOM/Kumar.NOM Sarita with dance do CNP] aami ki kata ceppaa Du her to story told 3.M.S c. *Kumar/ aa pichooDu [aa pichooDu ki/Kumar ki koopamu wacc i] Kumar.NOM/that idiot.NOM [that idiot DAT/Kumar DAT anger come CNP]

PAGE 86

86 Naa illu kuulcaa Du my house destroyed 3.M.S Concerning the condition that the CNP subject has to be an R expression it is worth mentioning that an R expression may be any NP that is not pronominal. T he data introduced thus far involve proper nouns. This is not a requirement, however, a s (101a b) illustrate. (101) a. [yeenugu waac i] adi janaalu nu bayapeTTin di [elephant come CNP] it people ACC scared 3.N.S b [Naa boss ki pooyina waaram jwaramu wacc i] [my boss DAT last week fever come CNP] atanu muuDu roojulu mandulu waaDaa Du he.NOM three days medicine used If a pronominal is used as a CNP subject, a Copy Control construction becomes ungrammatical as (102a b) indic ate. (102) a. *[atani ki aakali wees i] [he DAT hunger fall CNP] atanu/Kumar bhojanamu tayaru ceesikunaa Du he.NOM/Kumar.NOM dinner prepare did for self 3.M.S b. [atani ki jwaram wacc i] atanu/Kumar hospital weLLaa Du [he.DAT fever come CNP ] he.NOM/Kumar.NOM hospital went 3.M.S T he pronominal in (102a b) cannot refer to someone other than Kumar. In other words, although both subjects are pronounced, they cannot have disjoint referents. This observation is true even if the context allows it. Take sentence (103a), for example. The idea can only be expressed with disjoint subjects. Yet, the sentence is ungrammatical because a CNP clause enforces a control reading. To express the same idea, an INF clause must be used (e.g., (103b) )

PAGE 87

87 (103) a. *[Kumar ki aakali wees i] [Kumar DAT hunger fall CNP] amma atani ki annam peTTin di mother him DAT food kept b [Kumar ki aakali wees ina anduku] [Kumar DAT hunger fall INF because] amma atani ki annam peTTin di mother him DAT food kept Before proceeding to the following section, a word about the copies in Copy Control is appropriate. T he matrix subject in the Copy Control examples (88 93) may be realized as an exact copy of the CNP subject. This applies only if the CNP subject does not exceed one or two words, a s (104a b) illustrate. Compare to (105a b) in which the CNP subject is a conjunct. In this case, only a pronoun or an epithet may be used as a subject in the matrix clause. (104) a. [yeenugu waac i] yeenugu janaalu nu bayapeTTin di [elephant come CNP] elephan t people ACC scared 3.N.S b [Naa boss ki pooyina waaram jwaramu wacc i] [my boss DAT last week fever come CNP] Naa boss muuDu roojulu mandulu waaDaa Du my boss.NOM three days medicine used (105) a. [Kumar maryu Sarita sinima cuus tuu] [Kumar.NOM and Sarita.NOM movie watch CNP] waLLu / aa pichooLLu/ *Kumar maryu Sarita popkorn tinna ru they/those idiots.NOM/*Kumar.NOM and Sarita.N OM popcorn ate 3.M.P b [Kumar unnu Sarita muddu peTTukon i] Kumar.NOM and Sarita.N kiss put to each other CNP]

PAGE 88

88 waLLu / aa pichooLLu/ *Kumar maryu Sarita wellipooyaa ru they/those id iots.NOM/*Kumar.NOM and Sarita.NOM left 3.M.P 2. 6 .3.2 Copy Control in Assamese Similar to Telugu, Ass amese allows Copy Control if conditions 1 and 2 below hold. Unlike Telugu, Assamese has an extra restricti on on the type of CNP subject that is allowed to be pronounced. Only subjects of psychological or e xperiential CNP predicates are allowed to be overt This restriction is expressed as condition 3 Condition 1: The CNP clause is sentence initial. Condition 2: The CNP subject is an R expression (non pronominal). Condition 3: The CNP clause contains an experiential predicate. Condition 3 is based on the fact that Copy Control structures that involve non experiential CNP predicates are considered unacceptable, at least by some of the speakers I consulted. By comparison, Copy Control structures that contain experiential predicates are judged as acceptable by all the consultants. The subjects of experiential predicates in Assamese are usually g enitive. Sentences (106 109) are some examples. (106) [Ram Or i khong uth i] [Ram GEN anger raise CNP] xi i /Ram e mor ghorto bhangil e he.NOM /Ram NOM my house destroyed 3 (107) [Ram Or i phurti lag i] [Ram GEN exhilaration do CNP] xi i /Ram e pagolor nisena nasil e he.NOM /Ram NOM crazy person like danced 3 (108) [Ram Or i bhagar lag i] etiya xi i /Ram xui thakil [Ram GEN exhaustion feel CNP] now he/Ram ABS sleep kept

PAGE 89

89 (109) [Ram Or i xomoi no thak i] [Ram GEN time NEG keep CNP] xi i /Ram e bhat o na khal e he/Ram NOM rice even NEG ate 3 As I mentioned in Section 2.3 .2, the re are two types of experiential predicates in Assamese, those that license a genitive subject and those that license an experiential nominative subject. The difference between the two types is illustrated in examples (19 20) above, repeated as (110 111). Sentences (110a) and (111a) comprise what Abbi (1991) calls State Experiential and Process Experiential predicates. These license genitive subjects, and they do not allow the d ( 111b), on the other hand, contain Stative Action Process Experiential predicates In addition to being experiential, these predicates are volitional, which is why they license experiential nominative (110) a. Ram Or (*jani buji) e ta buddhi khelal Ram GEN ( knowingly) one CL idea played b. Ram e jani buji e ta buddhi koril e Ram NOM knowingly one CL idea did 3 (111) a. Ram Or (*jani buji) khong uthil Ram NOM ( knowingly) anger did b Ram e jani buji khong koril e Ram NOM knowingly anger did 3 What is pertinent to this section is that Assamese allows Copy Control, not only if the subject is an experiential genitive NP, but also if it is an experiential nominative NP, as (112 114) below sh ow. As condition 3 above points out, Copy Control structures are judged

PAGE 90

90 as acceptable as long as the CNP clause contains an experiential predicate. This restriction holds regardless of the morphological Case of the CNP subject. (112) [Ram e i khong kor i] [Ra m EXP NOM anger raise CNP] xi i/ /Ram e mor ghorto bhangil e he.NOM/Ram NOM my house destroyed 3 (113) [Ram e i phurti kor i] [Ram EXP NOM exhilaration do CNP] etiya tar i /Ra m Or bohk lagil now he.GEN/Ram GEN hunger felt (114) [Ram e i dukh kor i] [Ram EXP NOM sadness do CNP] xi i /Ram e bhat o na khal e he/Ram NOM rice even NEG ate 3 On the other hand, if the CNP subject is not genitive or experiential nominative, judgment pertaining to Copy Control becomes inconsistent. Of the four native speakers I have consulted, two considered instances of Copy Control like (115 117) a s acceptable, while two considered them as unacceptable. Notice that the CNP clause is sentence initial and the CNP subject is an R expression. Apparently, the only reason why the sentences are considered unacceptable by two of the consultants is because the CNP predicate is n ot an Experiential predicate. (115) /*[Ram e kam tu kor i] xi gusi gol [Ram NOM work do CNP] he.ABS away went (116) /* [Ram e kukur tu heru i] tar dukh lagil [Ram NOM dog CL lose CNP] he.GEN sad felt

PAGE 91

91 (117) /* [Ram e loteri jik i] tar phurti lagil [Ram NOM lottery win CNP] he.GEN exhilaration felt The acceptable examples of Copy Control in this section comprise an R expression or a pronoun in the matrix clause. Another alternative is an epithet, as the sentences in (118) show (118) a [Ram Or khub bhok lag i] [Ram GEN very hunger feel CNP] beseratu e posa bhat khal e the poor guy NOM stale rice ate 3 b [Ram e khong kor i] [Ram EXP NOM anger raise CNP] gadhatu e mor ghorto bhangil e the donkey NOM my house destroyed 3 In the discussion about Telugu Copy Control, I mentioned that epithets make Copy Control structures sound less redundant. In Assamese, on the other hand, even if no epithet is involved, Copy Control is enforced if the sentence begins with a pronounced CNP subject that is Case marked di fferently from the matrix subject. In other words, say the speaker begins a sentence with the CNP clause in (118a) above. The clause contains an experiential predicate and an overt genitive subject. In order to finish the sentence with a matrix clause that licenses a nominative subject, the speaker automatically inserts an overt pronoun; otherwise, the sentence becomes degraded. Descriptively, not pronouncing the matrix subject means that the structure qualifies as a Backward Control construction. The tende ncy to insert a pronominal is consistent with the data in Section 2. 6.2.2 which shows that Assamese disprefers Backward Control constructions. Nevertheles s, if both the CNP and the matr ix predicates license the same Case, Copy Control becomes redundant, although not unacceptable. In this case, an epithet makes the

PAGE 92

92 sentence sounds less redundant. Otherwise, Forward Control is preferred. For example, when presented with (119a b), speakers automatically choose the latter, considering the former acceptable but redundant. The following chapter provides a detailed explanation for this preference. For now, it suffices to say that, when the predicates check the same Case (e.g., nominative) on their subject s, s peakers assign the pronounced subject to the matrix clause, leaving the CNP subject silent, as ( 119 c) indicates. (119) a. Ram e loteri jik i xi notun ghor kinil e Ram NOM lottery win CNP he.NOM new house bought 3 b. Ram e loteri jik i notun ghor kinil e Ram NOM lottery win CNP new house bought 3 c Ram e [loteri jik i] notun ghor kinil e Ram NOM [lottery win CNP] new house bough t 3 Most importantly, the two pronounced subjects in Assamese Copy Control have to be co referential. Disjoint subjects result in ungrammaticality, as the sentences in (120) illustrate. (120) a. [ Ram Or i khong uth i] xi k /Proxad gusi gol [Ram GEN anger raise CNP] he.ABS/Proxad ABS away went b *[Ram Or i xomoi na thak i] [Ram GEN time NEG keep CNP] xi k / PrOxad e bhat o na khal e he.NOM/Proxad NOM r ice even NEG ate 3 c *[Ram e loteri jik i] tar ghoiniyek Or phurti lagil [Ram NOM lottery win CNP] his wife GEN exhilaration felt In ad dition, just like in Telugu, Copy Control in Assamese is unacceptable if the CNP clause is not sentence initial (e.g., ( 121) ) and/or if the CNP subject is pronominal (e.g., ( 122 ) ).

PAGE 93

93 (121) Ram e [tar/Ram Or xomoi no thak i] Ram NOM [he/Ram GEN time NEG keep CNP] bhat na khal e rice NEG ate 3 (122) [tar bhagar lag i] [he GEN exhaustion feel CNP] etiya xi/Ram xui thakil now he/Ram ABS sleep kept T h e Telugu and Assamese data presented in Sections 2. 6.1 through 2. 6.3 indicate that structures with CNP clauses require a control interpretation. Exceptions do exist, however. These are discussed in the following section. 2. 6 .4 Exceptions Although the Same Subject Condition is usually obeyed, and thus control is normally enforced, violations do occur. Observe the structures in (123 124), for example. Notice that, contrary to expectation, disjoint subjects are allowed in the environment of a CNP clause. (123) Telu gu a. [bombu pel i] caala mandi canipoyaa ru [bomb.NOM explode CNP] many people.NOM died 3.M.P b [war am paD i] cetlu/mokkalu peri ga yi [rain.NOM fall CNP] trees/plants.NOM grew 3.N.P (124) Assamese a. [e ta ghor ot zui lag i] bohut manuh moril [one CL house LOC fire ABS happen CNP] many people ABS died b [dhumuha ah i] bohut gos bhangil [storm ABS come CNP] many trees ABS broke

PAGE 94

94 Just as Adjunct Control into CNP clauses is not unique to Assamese and Telugu but commo n to most South Asian languages, violations of adjunct control also occur in many of these languages; for example Bengali (Klaiman 1981) Marat hi ( Pandharipande 1997) Hindi (Davison 1981), systematic study on exac tly this issue. The author examines Bengali CNP clauses and arrives at the following conclusion: the Same Subject Condition applies when either the matrix clause or the CNP clause expresses a I f the activities in both clauses are non volitional, the condition can be violated (Klaiman 1981: 120). This generalization applies to Telugu and Assamese. If either of the activities in (123a b) or (124a b) above is volitional, the sentence becomes unacceptable. This idea is illustrated in (125 126) below. In the (a) sentences the CNP predicate is volitional, and in the (b) sentences the matrix clause is volitional. A ll four sentences are ungrammatical. (125) T elugu a. [Kumar bombu ni pelc i] [Kumar.NOM bomb ACC explode CNP] caala mandi canip oya ru many people.NOM died 3.M.P b [bombu pel i] ambulens waccin di [bomb.NOM explode CNP] ambulance.NOM came 3.N.S (126) Assamese a [Ram e ghortot zui laga i] [Ram NOM house fire happen CNP] bohut manuh moril many people ABS died b [e ta ghorot zui lag i] [one CL house LOC fire happen CNP]

PAGE 95

95 bohut manuh e police aloi phone koril e many people.NOM police DA T phone did 3 Commenting on a similar case in Bengali, Klaiman adds: I hope I have shown that the conditioning is to a very large extent semantic, and that it is impossible to adequately describe any o f these processes without reference to the underlying semantic opposition VOLITIONAL / NONVOLITIO The one possibility I would confidently rule out is that any existing theoretical model can handle the facts. The material presented in this study calls f or a new approach to meaning in grammar. (125 126) Chapter 6 suggests that this semantic restriction is also a conspiracy in the syntax and that the examples that violate the Same Subject Condition are in fact instances of Obligatory Control. 2. 7 Conclusi on This chapter presented a linguistic overview of Telugu and Assamese morphosyntax, highlighting aspects that are relevant to the topic of Adjunct Control. One aspect that is most pertinent for our purposes is the licensing of Case marked subjects in the different types of clauses in each language. Both language s have Inherent and Structural Case marked subjects. In Telugu, the two types are licensed in finite, as wel l as in INF and CNP clauses (Table 2 3 ) In Assamese, the status of Structural Case marke d subjects in CNP clauses is uncertain (Table 2 4 ) In Backward Control structures, such subjects are judged as degraded or unacceptable. In Copy Control structures, some speakers consider them acceptable. Table 2 3 Subjects licensed in Telugu Type and form finite clauses INF cl a u ses CNP clauses Inherent Case DAT ki Structural Case NOM Table 2 4 Subjects licensed in Assamese Type and form finite clauses INF cl a u ses CNP clauses Inherent Case GEN EXP NOM Or e/ Structural Case NOM ABS e/ /??/

PAGE 96

96 In addition, Telugu and Assamese have non finite Conjunctive Participle (CNP) clauses that function as adjuncts Both languages show evidence for Adjunct Cont rol into CNP clauses (Table 2 5). Telugu licenses Forward, Backward, and Copy Control. Assamese only licenses Forward and Copy Control; Backward Control structures are considered degraded or unacceptable. Table 2 5 Types of Adjunct Control allowed in Telugu and Assamese Forward Control Ba ckward Control Copy Control Telugu Assamese ??/ The following chapter presents a detailed analysis of Forward and Backward Adjunct Control in Telugu and Assamese It provides an account of the conditions that drive and constrain their occurrence. It also deals with the problems that the analysis brings about, especially as related to Case theory.

PAGE 97

97 CHAPTER 3 FO RWARD/BACKWARD ADJUN CT CONTROL IN TELUGU AND ASSAME SE: TH E ANALYSIS 3.1 Introduction This chapter provides an analysis of Forward and Backward Adjunct Control in Telugu and Assamese. In both types of control, only one of the two co referential subjects in a given structure is pronounced. Forward Control is the c ase when the pronounced subject is licensed by the predicate in the matrix clause (1a). Backward Control, on the other hand, means that the pronounced subject is licensed by the subordinate/CNP predicate (1b). In both cases, the overt subject determines th e identity of the unpronounced subject. (1) a. Forward Control [ Matrix [ CNP Subject [ Matrix b. Backward Control [ Matrix [ CNP [ Matrix Subject The rest of the chapter is organized as follows. Section 3. 2 puts forth the structures that this chapter means to analyze. Although these structures are documented in Chapter 2, this section goes beyond listing to show that Case differences or similarities between the CNP subject and the matrix subject have no say in the type of Adjunct Control t hat is licensed. Section 3 .3 shows that the Adjunct Control structures under examination are instances of Obligatory Control. This is important because within the framework of the Movement Theory of Control, Non Obligatory Control structures are not deriv ed via movement. Rather, they are analyzed as involving pro T his is the same pro that is believed to exist in pro drop languages. If the Telugu and Assamese c ontrol structures that this study is concerned with are occurrences of Non Obligatory C ontrol, th e discussion in the rest of the chapter becomes irrelevant. Section 3. 4 delineates the steps involved in the derivation of Forward and Backward Control structures in Telu gu and Assamese. Both types of c ontrol are analyzed as instances of sideward m ovement (Nunes 2004)

PAGE 98

98 If the analysis of c ontrol as movement is on the right track, this means that the subject in Telugu and Assamese Adjunct Control moves out of one Case po sition into a new Case position. T hat is, it undergoes multiple Case checking. This ide a is problematic on both theoretical and empirical grounds. Section s 3. 5 and 3. 6 address exactly this issue and propose a solution Section 3. 7 is a summary and a conclusion. 3.2 Forward/Backward Control: The Facts 3.2.1 Forward Control The sentences in (2 ) through (5) below are examples of Telugu and Assamese Forward Control In (2) and (4), the Forward Control structures involve CNP and matrix subjects that are Case marked differently. In (3) and (5), the CNP and matrix subjects are Case marked the same. Under the Movement Theory of Control, t he subject starts out in the CNP clause and moves to the matrix clause. The result is two identical copies of the same NP. At PF, the CNP copy is deleted for reasons to be specified in Section 3. 4. Note that the Telug u sentences (2c), (2d), and (3b) are marginal for an indepe ndent reason: Telugu seems to disprefer a dative subject in the matrix clause of Adjunct Control structures. I do not have an explanation why this is so. All that I can say at this point is that it is a semantic/pragmatic preference. (2) Telugu (CNP and matrix subject Case marked differently) a. Kumar [ Kumar ki jwaram wacc i] hospital weLLaa Du Kumar.NOM [ Kumar DAT fever come CNP] hospital went 3.M.S b. Kumar [ Kumar ki aakali wees i] sandwic tinnaa Du Kumar.NOM [ Kumar DAT hunger fall CNP] sandwich ate 3.M.S c ??Kumar ki [ Kumar bazaaru loo Sarita ni cuus i] Kumar DAT [ Kumar.NOM mark et in Sarita ACC see CNP] Sarita miida preema kaligin di Sarita on love occurred 3.N.S

PAGE 99

99 d. ??Sarita ki [ Sarita aa maaTa win i] koopamu waccin di Sarita DAT [ Sarita.NOM that mat ter hear CNP] anger came 3.N.S (3) T elugu (CNP and matrix subject Case marked the same) a Kumar [ Kumar Sarita too naTyamu cees tuu] Kumar.NOM [ Kumar.NOM Sarita with dance do CNP] aami ki kata ceppaa Du her DAT story told 3.M.S b ??Kumar ki [ Kumar ki juttu uuDipooy i] picci paTTin di Kumar DAT [ Kumar DAT hair lose CNP] craziness caught 3.N.S (4) Assamese (CNP and matrix subject Case marked differently) a. Ram e [ Ram Or phurti lag i] Ram NOM [ Ram GEN exhilaration feel CNP] pagolor nisena nasil e crazy person like danced 3 b Ram [ Ram Or bhagar lag i] xui thakil Ram ABS [ Ram GEN exhaustion feel CNP] sleep kept c Ram Or [ Ram e kukur tu heru i] dukh lagil Ram GEN [ Ram NOM dog CL lose CNP] sad felt d Ram e [ Ram e khong kor i] mor ghorto bhangil e Ram NOM [ Ram EXP NOM anger did CNP] my house destroyed 3 (5) Assamese (CNP and matrix subject Case marked the same) a. Ram e [ Ram e loteri jik i] notun ghor kinil e Ram NOM [ Ram NOM lottery win CNP] new house bought 3 b. Ram Or [ Ram Or e ta bhal buddhi khela i] bhal lagil Ram GEN [ Ram GEN one CL good idea play CNP] good felt

PAGE 100

100 Forward Control is obvious in sentences ( 2 a d) and ( 4 a d ). T he Case marking of the pronounced subjects shows that they are licensed by the matrix predicate and that the structures ar e instances of Forward Control. In ( 3 a b) and ( 5 a b), on the other hand, the matrix and CNP subjects in each construction are Case marked the same T his means that the structures can be instances of Forward Control. At the same time they may be analyzed a s instances of Backward Control. In the latter case, the CNP subject is overt and the matrix subject is implied as illustrated in (6 7) below. B oth types of control result in the same word order, which is why they cannot be teased apart. (6) Telugu (CNP and matrix subject Case marked the same) a. Kumar [Kumar Sarita too naTyamu cees tuu] Kumar.NOM [Kumar.NOM Sarita with dance do CNP] aami ki kata ceppaa Du her DAT story told 3.M.S b. ?? Kumar ki [K umar ki juttu uuDipooy i] picci paTTin di Kumar DAT [Kumar DAT hair lose CNP] craziness caught 3.N.S (7) Assamese (CNP and matrix subject Case marked the same) a. Ram e [Ram e loteri jik i] notun ghor kinil e Ram NOM [Ram NOM lottery win CNP] new house bought 3 b. Ram Or [Ram Or e ta bhal buddhi khela i] bhal lagil Ram GEN [Ram GEN one CL good idea play CNP] good felt In order to make sure that Forward Control is allowed, the four sentences are repeated in (8) and (9), only this time the CNP clauses are realized sentence initially. Notice that the overt subjects are pronounced in the vicinity of the m atrix clause, which is an indication that (3a b) and (5a b) above qualify as Forward Control constructions.

PAGE 101

101 (8) Telugu (CNP and matrix subject Case marked the same) a. [ Kumar Sarita too naTyamu cees tuu] [ Kumar.NOM Sarita with dance do CNP] Kumar aami ki kata ceppaa Du Kumar.NOM her DAT story told 3.M.S b. [ Kumar ki juttu uuDipooy i] Kumar ki picci paTTin di [ Kumar DAT hair lose CNP] Kumar DAT craziness caught (9) Assamese (CNP and matrix subject Case marked the same) a [ Ram e loteri jik i] Ram e notun ghor kinil e [ Ram NOM lottery win CNP] Ram NOM new hous bought 3 b [ Ram Or e ta bhal buddhi khe la i] Ram Or bhal lagil [ Ram GEN one CL good idea play CNP] Ram GEN good felt It could be argued that the pronounced subject in each sentence belongs to the CNP clause and that it is extraposed to a clause final position. Extraposition of this type leads to ungrammaticality, as illustrated in (10 11). In all four sentences, the pronounced subject is licensed by the CNP clause, as Case shows. The sentences are ungrammatical, not because they are instances of Back wa rd Control, but because the CNP subject is extraposed to a post verbal position. (10) Telugu (CNP and matrix subject Case marked differently) a. *[jwaram wacc i Kumar ki] Kumar hospital weLLaa Du [fever come CNP Kumar DAT] Kumar.NOM hospital went 3.M. S b *[aakali wees i Kumar ki ] Kumar sandwic tinnaa Du [hunger fall CNP Kumar DAT] Kumar.NOM sandwich ate 3.M.S (11) Assamese (CNP and matrix subject Case marked differently) a *[phurti lag i Ram Or] [ exhilaration feel CNP Ram GEN]

PAGE 102

102 Ram e pagolor nisena nasil e Ram NOM crazy person like danced 3 b *[bhagar lag i Ram Or] Ram xui thakil [exhaustion feel CNP Ram GEN] Ram ABS sleep kept It is worth noting that the ungrammaticality of the sentences in (10) and (11) follows from the fact that the v erb in both languages canonically oc cur s clause finally. Telugu seems to categorically disallow sentences of the word order OVS, as the sentences in (12) illustrate. Assamese is more permissive than Telugu, allowing OVS structures in which case the post verbal NP is interpret ed as an afterthought S uch structures, however, are only considered acceptable in finite clauses like the ones in (13 ). Extraposition in non finite clauses results in ungrammaticality. The bottom line is that the pronounced subjects in (8 a b ) and (9 a b ) a bove belong to the matrix clauses. In other words, the structures are instances of Forward Control. Whether the CNP subject and the matrix subject are Case marked differently or the same has no influence on the type of c ontrol that is allowed. (12) Telugu OVS a. hospital weLLaa Du Kumar hospital went 3.M.S Kumar b. *jwaram waccin di Kumar ki fever came 3.N.S Kumar DAT (13) Assamese OVS a pagolor nisena nasil e, Ram e crazy person like danced 3, Ram NOM b xui thakil, Ram sleep kept, Ram ABS

PAGE 103

103 3.2.2 Backward Control Unlike Forward Control, which is acceptable in both Telugu and Assamese, Backward Control is only acceptable in Telugu. The sentences in (14 ) and (1 5 ) below are examples from Telugu The sentences in (14) contain subjects that are Case marked differently, whereas those in (15) contain subjects that are Case marked the same. In both cases, t he subject starts out in the CNP c lause and moves to the matrix clause. At PF, the CNP copy is preserved and the matrix copy is deleted. (14) T elugu (CNP and matrix subject Case marked differently) a. Kumar [Kumar ki jwaram wacc i] hospital weLLaa Du Kumar.NOM [Kumar DAT fever come CNP] hospital went 3.M.S b Kumar [Kumar ki aakali wees i] sandwic tinnaa Du Kumar.NOM [Kumar DAT hunger fall CNP] sandwich ate 3.M.S c ?? Kumar ki [Kuma r bazaaru loo Sarita ni cuus i] Kumar DAT [Kumar.NOM market in Sarita ACC see CNP] Sarita miida preema kaligin di Sarita on love occurred 3.N.S d ?? Sarita ki [Sarita aa maa Ta win i] koopamu waccin di Sarita DAT [Sarita.NOM that matter hear CNP] anger came 3.N.S (15) Telugu (CNP and matrix subject Case marked the same) a Kumar [Kumar Sarita too naTyamu cees tuu] Kumar .NOM [Kumar.NOM Sarita with dance do CNP] aami ki kata ceppaa Du her DAT story told 3.M.S b ?? Kumar ki [Kumar ki juttu uuDipooy i] picci paTTin di Kumar DAT [Kumar DAT hair lose CN P] craziness caught 3.N.S

PAGE 104

104 We saw in the previous se ction that the structures in (15 ) qualify as examples of Forward Control. Now we need to make sure that they may also qualify as examples of Backward Control. Observe the sentences in (16 ). Sentence (1 6 a) is an example of Forward Control. Sentence (16 b) is ungrammatical because it involves extraction of the time expression/adverb out of the CNP clause, which is an adjunct 1 Scrambling within the adjunct boundari es, on the other hand, is acceptable as (16 c) shows I n this case, the time expression is scrambled past the CNP subject but it is still within the boundaries of the CNP clause. (16) Telugu a Kumar [ Kumar ki enimidiki aakali wees i /wees ina anduku] Kumar.NO M [ Kumar DAT at 8:00 hunger fall CNP / fall INF because] tommidiki bhojanamu tayaru ceesukunaa Du? at 9:00 dinner prepare did for self 3.M.S? b *enimidiki Kumar [ Kumar ki aakali wees i /wee s ina anduku] at 8:00 Kumar.NOM [ Kumar DAT hunger fall CNP / fall INF because] tommidiki bhojanamu tayaru ceesukunaa Du? at 9:00 dinner prepare did for self 3.M.S? c [enimidiki Kumar ki aak ali wees i /wees ina anduku] [at 8:00 Kumar DAT hunger fall CNP / fall INF because] Kumar tommidiki bhojanamu tayaru ceesukunaa Du? Kumar .NOM at 9:00 dinner prepare did for self 3.M.S? Now con sider the grammatical structure in (1 7 a). Based on the discussion about (1 6 ), we ca n conclude that the only way (17 a) can be acceptable is if it reads as (17 b) where the scrambling of the time expression aaruinTiki es of the CNP clause. If the sentence were analyzed as (17 c), it should be considered ungrammatical since 1 T hat the matrix clause has a distinct time expression, tommidiki enimidiki

PAGE 105

105 it would involve extraction out of an island. This indi cates that (17a) qualifies as an example of Backward Control. By the same token, the structures in (15 ) above may qualify as instances of Backward Control. (17) Telugu a. aaruinTiki Kumar unnu Sarita kalis i eeDuinTiki wellaa ru at 6:00 Kumar and Sarita.NOM meet CNP at 7:00 left 3.M.P b. Ku mar unnu Sarita [aaruinTiki Kumar unnu Sarita kals i] Kumar and Sarita.NOM [at 6:00 Kumar and Sarita.NOM meet CNP] eeDuinTiki wellaa ru at 7:00 left 3.M.P c. *aaruinTiki Kumar unnu Sarita [ Kumar unnu Sarita kals i] at 6:00 Kumar and Sarita.NOM [ Kumar and Sarita.NOM meet CNP] eeDuinTiki wellaa ru at 7:00 left 3.M.P Unlike Telugu, Assamese Adjunct Control structures may not be realized a s instances of Backward Control. T he result ranges from degradation to unacceptability as (18 ) shows Section 3. 4.2.2 offers a possible explanation as to why this is the case (18) Assamese (CNP and matrix subject Case marked differently) a. ?? Ram e [Ram Or phurti lag i] Ram NOM [Ram GEN exhilaration feel CNP] pagolor nisena nasil e crazy person like danced 3 b. ?? Ram [Ram Or bhagar lag i] xui thakil Ram ABS [Ram GEN exhaustion feel CNP] sleep kept c. ?? Ram e [Ram e khong kor i] mor ghorto bhangil e Ram NOM [Ram EXP NOM anger did CNP] my house destroyed 3

PAGE 106

106 d. ??/* Ram Or [Ram e kukur tu heru i] dukh lagil Ram GEN [Ram NOM dog CL lose CNP] sad felt Again, this phenomenon is independent of whether the CNP subject and the matrix subject are Ca se marked the same or differently. The sentences in (19 ) must be considered degraded to unacceptable under the proposed parsing. Of course when presented to native speakers as a string of words, they are parsed as Forward Control structures. (19) Assamese (CNP and matrix subject Case marked the same) a. ??/* Ram e [Ram e loteri jik i] notun ghor kinil e Ram NOM [Ram NOM lottery win CNP] new house bought 3 b ?? Ram Or [Ram Or e ta bhal buddhi khe la i] bhal lagil Ram GEN [Ram GEN one CL good idea play CNP] good felt When extraction is involved, however, things become clearer. For example, a structure like ( 20 a) is directly ruled out by native speakers and thus the sentences are semantically awkward We know that the structure in ( 20 b) is not allowed because it involves extraction out of an island. But ( 20 c) is also ruled out because it is an instance of Backward Control. This proves that Backward Contro l is not allowed in Assamese. Compare to (17 b) from Telugu. (20) A ssamese a *January t Ram e loteri jik i January in Ram NOM lottery win CNP March ot notun ghor kinil e March in new house bought 3 b *January t Ram e [ Ram e loteri jik i] January in Ram NOM [ Ram NOM lottery win CNP]

PAGE 107

107 March ot notun ghor kinil e March i n new house bought 3 c Ram e [January t Ram e loteri jik i] Ram NOM [January in Ram NOM lottery win CNP] March ot notun ghor kinil e March in new house bought 3 This section presented evidence that Forward Control in Telugu and Assamese and Backward Control in Telugu are licensed independently from the Case similarities or differences between the CNP an d matrix subject s For the purpose of clarity, most of the Adjunct Control examples used in the following sections comprise two differentl y Case marked subjects. Before proceeding to the analysis, S ection 3.3 shows that the structures under examination are occurren ces of Obligatory Control. 3. 3 Telugu and Assamese Adjunct Control as Obligatory Control Traditionally, Obligatory Control structures share a number of characteristics (Williams 1980, Hornstein 1999 Jackendoff and Culicover 2003, Polinsky and Pot sdam 2004, among others ). Four major characteristics are listed in (21). (21) Properties of Obligatory Control The subordinate subject a. Requires a c commanding antecedent b Requires a local antecedent c Disallows split antecedents d Requires a sloppy read ing under ellipsis To illustrate, observe the examples in (22 25) (based on Hornstein 2001: 46 (56 59)). In structure (22a), the unpronounced controllee the subordinate subject and the overt controller Tom in the matrix clause are in a c command relationship. The sentence is grammatical. Compare to (22b). The sentence is ungrammatical under the designated reading

PAGE 108

108 because Tom in the matrix clause does not enter a c command relationship with the unpronounced subordinate subject, and thus does not me et the requirement in (21a). 2 (22) a Tom left [after Tom washing himself]. b Tom washing himself]. Further, the requirement in (21b) above dictates that the controller be local, possibly in the next higher clause. Sentence (23a) m eets this requirement. In (23b), on the other hand, the controller is too remote, resulting in unacceptability. (23) a. Tom left [after Tom washing himself]. b *Tom said that Sue left [after Tom washing himself]. The prohibition against split antecedents in (21c) follows from the requirements in (21a) and (21b). If Obligatory Control requires a local c commanding antecedent, it follows that two non conjoined NPs cannot both be local, c commanding controllers. This is the same as saying that two non conjoined NPs cannot both be antecedents to the same trace (Hornstein 1999: 80). The prohibition explains why (24a) and (24b) below are ungrammatical. In (24a), Tom in the highest clause is too remote to control Tom in the adjunct. In (24b), assuming that the subord inate clause adjoins to the matrix clause at vP (or VP), Sue does not enter a c command relationship with the subordinate subject and, thus, cannot be a controller. (24) a *Tom said that Sue left [after Tom and Sue washing themselves] b *Tom kissed Sue [af ter Tom and Sue washing themselves]. The restriction in (21d) dictates that structures like (25) should mean that Sue left after she not Tom, ate. That is, the sentence must have the sloppy reading in (25a) rather than the strict reading in (25b). 2 Here and below, I use the terms controller an d controllee to refer to the two co referential elements in a control structure. The controller is the element in the matrix clause, and the controllee is the element in the subordinate clause. The usage is only a matter of convenience; it is independent o f the type of control (Forward vs. Backward), or whether the element referred to is pronounced or unpronounced.

PAGE 109

109 (25) Tom l eft [after Tom eating] and Sue did too. a Sue left after she ate. b *Sue left after Tom ate. If tested against the properties in (21), Telugu and Assamese Adjunct Control qualifies as Obligatory Control. To begin with, Adjunct Control requires a c co mmanding antecedent This is true of Forward and Backward Control in Telugu and of Forward Control in Assamese. 3 For example, the Telugu Forward and Backward Control structures (2 6a ) and (26b) are ungrammatical under the designated reading because the cont roller and the controllee do not enter a c commanding relationship The same observation applies to the Assamese Forward Control structure (27a). Compare to (26c d) and (27b) in which the controller and the controllee are in a c command relationship. (26) Telug u a Forward Control *atani i amma [ atani ki i aakali wees i] atani ki annam peTTin di his mother [ he DAT hunger fall CNP] him DAT food kept 3.N.S Intended m eaning: b Backward Control atani i amma [atani ki i aakali wees i] atani ki annam peTTin di his mother [he DAT hunger fall CNP] him DAT food kept 3.N.S Intended m eaning : c Forward Control atanu i [ atani ki i aakali wees i] sandwic tinnaa Du he.NOM [ he DAT hunger fall CNP] sandwich ate 3.M.S d Backward Control atanu i [atani ki i aakali wees i] s andwic tinnaa Du he.NOM [he DAT hunger fall CNP] sandwich ate 3.M.S 3 I will not examine Assamese Backward Control structures because they are already degraded.

PAGE 110

110 (27) Assamese a *tar i mak Or [ xi i bhalkoi nas i] bhal lagil his mother GEN [ he NOM well dance CNP] good felt Intended m eaning He b tar i [ xi i bhalkoi nas i] bhal lagil he.GEN [ he NOM well dance CNP] good felt e Further, the antecedent has to be local. The Telugu Forward and Backward Control struct ures (28a b) and the Assamese Forward Control structure (29a) each involve a non local antecedent under the intended meaning. The sentences are ungrammatical. Sentences (28c d) and (29b), on the other hand, involve a local antecedent. The sentences are acc eptable. (28) Telugu a Forward Control *[Kumar [ Sarita sinima cuus Tuu] [Kumar.NOM [ Sarita.NOM movie watch CNP] popcorn tinna Du aNi] Sarita ceppin di popcorn ate 3.M.S so/that] Sarita.NOM said 3.N.S Intended m eaning : e popcorn while Sarita was watching b Backward Control *[ Kumar [Sarita sinima cuus Tuu] [ Kumar.NOM [Sarita.NOM movie watch CNP] popcorn tinna Du aNi] Sarita ceppin di popcorn ate 3.M.S so/that] Sarita.NOM said 3.N.S Intended m eaning : c. Forward Control [Kumar [ Kumar sinima cuus Tuu] [Kumar.NOM [ Kumar.NOM movie watch CNP] popcorn tinna Du aNi] Sarita ceppin di popcorn ate 3.M.S so/that] Sar ita.NOM said 3.N.S

PAGE 111

111 d Backward Control [ Kumar [Kumar sinima cuus Tuu] [ Kumar.NOM [Kumar.NOM movie watch CNP] popcorn tinna Du aNi] Sarita ceppin di popcorn ate 3.M.S so/that] Sarita.NOM said 3.N.S (29) Assamese a *PrOxad e kol e [ze Ram e Proxad NOM said 3 [that Ram NOM [ PrOxad Or xomoi na thak i] bhat na khal e] [ Proxad GEN time NEG k eep CNP] rice NEG ate 3] Intended m eaning : b PrOxad e kol e [ze Ram e Proxad NOM said 3 [that Ram NOM [ Ram Or xomoi na thak i] bhat na khal e] [ Ram GEN time NEG keep CNP ] rice NEG ate 3] Further the unpronounced subject of Obligatory Control constructions does not allow split antecedent s As I mentioned earlier, t his requirement follows from the two previ ous properties, namely, that Obligatory Control requires a local and c commanding antecedent. In ( 30a b ), for example, the CNP subject has two c commanding et, it can only take as an antecedent The same applies to the Assamese example ( 31 (30) Telugu a Forward Control *Kumar [ Kumar unnu Sarita sinima cuus Tuu] Kumar.NOM [ Kumar unnu Sarita.N movie watch CNP] popcorn tinna Du aNi Sarita ceppindi popcorn ate 3.M.S so/that Sarita said Intended m eaning : a i d that while she and Kumar were watching a movie,

PAGE 112

112 b Backward Control Kumar [Kumar unnu Sarita sinima cuus Tuu] Kumar.NOM [Kumar unnu Sarita.N movie watch CNP] popcorn tinna Du aNi Sarita ceppindi popcorn ate 3.M.S so/that Sarita said Intended m eaning : a i d that while she and Kumar were watching a movie, (31) Assamese *PrOxad e kol e ze Ram e Proxad NOM said 3 that Ram NOM [ Ram aru PrOxad Or khong uth i] ghorto banghil e [ Ram and Proxad GEN anger arise CNP] house destroyed 3 Intended m eaning : F inally Obliga tory Control requires a sloppy reading under ellipsis. Adjunct control structures in Telugu and Assamese share this property as well. To illustrate, the elided parts of Telugu Forward and Backward Control structures ( 3 2 a b) mean that Arun left because/afte r he himself got angry They do not mean that Arun left because/after Kumar got angry. The same applies to the elided part of the Assamese Forward Control structure in ( 3 3 ). Proxad destroyed the house because/after he himself got angry ; not because/after R am got angry. (32) Telugu a Forward Control Kumar [ Kumar ki koopamu wacc i] Kumar.NOM [ Kumar DAT anger come CNP] akkadi nunci wellipoyinaa Du Arun kuDa there from left 3.M.S Arun.NOM also Meaning : Not: b Backward Control Kumar [Kumar ki koopamu wacc i] Kumar.NOM [Kumar DAT anger come CNP]

PAGE 113

113 akkadi nunci wellipoyinaa Du Arun kuDa there from left 3.M.S Arun.NOM also Meaning: Not: (33) Assamese Ram e [ Ram Or khong uth i] Ram NOM [ Ram GEN anger come CNP] ghorto bhangil e, aru PrOxad e O take koril e house destroyed 3, and Proxad NOM also same did 3 Meaning: Not: destroyed the house Based on the evidence presented in t his section, we can conclude that Telugu an d Assamese Adjunct C ontrol structures are instances of Obligatory Control. 4 Now we turn to the details of the movement analysis. 3.4 Telugu and Assamese Adjunct Control as Sideward Movement It is well known that adjuncts are islands to movement, except in such cases as parasitic gap constructions which, according to Nunes (1995, 2004), in volve sideward movement. This is an operation that allows movement between two unconnected syntactic objects. Hornstein (1999, 2 003) follows Nunes and considers Adjunct Control structures as instance s of sideward m ovement. Under the Copy Theory of Movement as proposed by Chomsky (1995), all movement takes place between two posi tions that are in a c command relationship. A constitu ent copy plus merges into a c commanding position and the two copies form a chain system, the Copy Theory of Movement is reformulated as the Copy plus Merge Theory of Movement. According to this theory, movement comprises f our independent operations: Copy, 4 The appendix at the end of this dissertation prov ides evidence that Telugu and Assamese Adjunct Control also qualifies as Exhaustive Control as analyzed in Landau 2000, 2004.

PAGE 114

114 Merge, Form Chain, and Chain Reduction. The two operations Copy and Merge are in principle except that chain formation does not follow naturally from them. C hain formation is an independent operation which Nunes formulates as follows: (34) Form Chain Two constituents X and Y form a chain iff a. X and Y are non distinct; b. X c commands Y. With Form Chain as an independent step, movement does not have to target a c commanding position. In other words, movement between two unconnected syntactic objects is now possible. For example, X in (35) may copy out of the syntactic object L and merge in the unconnected syntactic object M, as (35a) illustrates. Subsequently, L and M undergo merge in (35 b). This type of movement is called sideward movement (Nunes 2004). Note that if L is an adjunct, it becomes an island after not before merging with M. 5 (35) a. [ L X [ M b. [ M [ L M Whereas the three operations Copy, Merge, and Form Chain take place in the syntax, Chain Reduction takes place in the phonological component. According to Nunes, if two elements form a chain one of them has to be deleted. Both operations, Form Chain and Chain Reduction, take place f or the purpose of linearization T hey satisfy the Linear Correspondence 5 It is worth mentioning that the idea of a c commanding antecedent in relation to control interpretation as used in Section 3.3 becomes anachronistic when looked at from the perspective of sideward movement (or the Copy plus Merge Theory of Movement (Nunes 2004)). As the discussion here shows, sideward movement does not target a c commanding position. Therefore, it is possible for two copies of the same element to end up in a non command relationship, which is actually the case of the Copy Control structure analyzed in Chapter 4. Such cases raise the question: if Adjunct Control is derived via sideward movement, which is divorced from c command, how is the control interpretation determined? In other words, what explains why the two subjects in an Adjunct Con trol structure have the same referent? According to Hornstein (2001), movement does. The fact that the two subjects in an Adjunct Control structure are derived by movement means that they are non distinct copies of the same token, and thus they are necessa rily co referential. In this sense, movement becomes the major theory of construal, and restrictions on control interpretation are mainly restrictions on movement. COPY MERGE

PAGE 115

115 Axiom in (3 6 ) which dictates that an element cannot asymmetrically c command and be asymmetrically c commanded by the same element in a structure. By the same token, an element cannot follow and precede itself as this induces a violation of irreflexivity Both asymmetry and irreflexivity are defined in (37) and (38) respectively (Nunes 2004: 24). To satisfy the Linear Correspondence Axiom, Chain Reduction applies at PF T his PF operati on reads as ( 39 ). (36) Linear Correspondence Axiom Let X, Y be nonterminals and x, y terminals such that X dominates x and Y dominates y. Then if X asymmetrically c commands Y, x precedes y. (Kayne 1994: 33) (37) Asymmetry If x precedes y, y necessarily does not p recede x. (38) Irreflexivity If x precedes y, then x and y are distinct copies. (39) Chain Reduction Delete the minimal number of constituents of a nontrivial chain CH that suffices for CH to be mapped into a linear order in accordance with the LCA (Nunes 2004: 27 (44)). To illustrate, sentence (40) below contains two non distinct copies of Tim in a c command relationship; thus, they form a chain The pronunciation of both copies induces a violation of the Linear Correspondence Axiom. As the arrows show, the verb called ends up c commanding and being c commanded by the same element T his is a violation of asymmetry as stated in (37) At the same time, the two copies of Tim being non distinct, Tim ends up preceding and following itself, which is a violation of the i rreflexivity condition in (37). In order for the structure to be linearized in accordance with the Linear Correspondence Axiom, one of the copies has to be deleted. Chain reduction applies and marks one of the copies for deletion. The lower copy undergoes deletion because it has less checked features (40) Tim was called Tim by Sue. c comma nding c commanded

PAGE 116

116 in Telugu and Assamese. I start with Forward Control. 3.4.1 Forward Control This sec tion analyzes Forward Control s truct u r es in Telugu and Assamese as instances of sideward movement. Section 3. 4 .1.1 deals with Forward Control structures in which the CNP clause is realized sentence internally ( e.g. (41a b ) ) Section 3. 4.1.2 provides an an alysis of Forwa r d Control structures in which the CNP clause is realized sentence initial ly ( e.g., (42a b) ) (41) a Telugu Kumar [ Kumar ki daggu u jalubu u wacc i ] Kumar.NOM [ Kumar DAT cough and cold and come CNP ] mandulu waaDaa Du medicines used 3.M. S b Assamese Ram e [ Ram Or xomoi no thak i] bhat na khal e Ram NOM [ Ram GEN time NEG keep CNP] rice NEG ate 3 (42) a Telugu [ Kumar ki daggu u jalubu u wacc i ] [ Kumar DAT cough and cold and r come CNP ] Kumar mandulu waaDaa Du Kumar.NOM medicines used 3.M.S b Assamese [ Ram Or xomoi no thak i] Ram e bhat na khal e [ Ram GEN time NEG keep CNP] Ram NOM rice NEG ate 3 3.4.1.1 Forward Control as sideward m ovement Observe the Telugu sentence in ( 43 ). Following Hornstein (1999, 2003) and Nunes (1995, 2001, 2004), we can pr opose that the sentence has the derivation in ( 44). In (44a), t he CNP clause and the matrix clause form ind In

PAGE 117

117 COPY ( 44b ), The copy plus merge oper ation between the two un connected syntactic objects, the CNP and the matrix clauses, is an instance of sideward movement. 6 Following, the CNP clause adjoins to matrix vP as shown in (44c ) Upon adjunction, the CNP clause becomes an island In (44d), t ves from Spec,vP to Spec, IP to check the EPP feature. As the dotted arrows in (44e) show, t he copy of commands both the copy in the CNP clause and the copy in Spec,vP, forming a chain with each thus, Form Chain. Step ( 44f ) takes pl ace at PF; this is when Chain Reduction applies, and the lower copy in each chain is deleted in order for the structure to be linearized in accordance with the Linear Correspondence Axiom (43) Telugu Kumar [ Kumar ki jwaram wacc i ] hospital weLLaa Du Kumar .NOM [ Kumar DAT fever come CNP ] hospital went 3.M.S to the hospital (44) a. i. [ CNP [ NP Kumar ki] jwaram wacc i] [ NP Kumar] [ CNP [ NP Kumar DAT] fever come CNP] ii [ Matrix vP hospital weLLaa Du] [ Matrix vP hospital went 3.M.S] b [ Matrix vP [ NP Kumar] hospital weLLaa Du] c [ Matrix IP [ vP [ CNP [ NP Kumar ki] jwaram wacc i] [ vP [ NP Kumar] hospital weLLaa Du]]] d [ Matrix IP [ NP Kumar] [ vP [ CNP [ NP Kumar ki] jwaram wacc i] [ vP [ NP Kumar] hospital weLLaa Du]]] 6 n. I deal with this problem in Section 3.5 below.

PAGE 118

118 COPY e. CP 3 IP C qp SUBJ I' Kumar qp vP I qp CNP P 7 vP 3 3 SUBJ 6 SUBJ 6 Kumar ki jwaram wacc i Kumar hospital weLLaa Du f. At PF: [ Matrix IP [ NP Kumar] [ vP [ CNP [ NP Kumar ki ] jwaram wacc i] [ vP [ NP Kumar ] hospital weLLaa Du]]] The exact same derivation applies to the Assamese sentence ( 45 ), as illustrated in (4 6 ). (45) Assamese Ram e [ Ram e loteri jik i] notun ghor kinil e Ram NOM [ Ram NOM lottery win CNP] new house bought 3 (46) a. i [ CNP [ NP Ram e] loteri jik i] [ NP Ram] [ CNP [ NP Ram NOM] lottery win CNP] ii. [ Matrix vP notun ghor kinil e] [ Matrix vP new house bought 3] b [ Matrix vP [ NP Ram e] notun ghor kinil e] c [ Matrix IP [ vP [ CNP [ NP Ram e] loteri jik i] [ vP [ NP Ram e] notun ghor kinil e]]] d [ Matrix IP [ NP Ram e] [ vP [ CNP [ NP Ram e loteri jik i] [ vP [ NP Ram e] notun ghor kinil e]]] 7 The size of CNP clauses is not especially relevant to the discussion in this chapter or in Chapter 4 (it becomes relevant in Chapter 5). This is why I continue to label them as CNP P. A brief note is in or der, however. I follow Jayaseelan (2004) by assuming that CNP clauses are IPs. This is a reasonable assumption. A CNP clause allows negation and other adverbs (e.g., time expressions), whose locus is generally believed to be higher than vP (Cinque 1999). A t the same time, a CNP clause does not allow an overt complementizer, which may be an indication that they do not project as high as CP.

PAGE 119

119 e CP 3 IP C qp SUBJ I' Ram e qp vP I qp CNP P vP 3 3 SUBJ 6 SUBJ 6 Ram e loteri jiki Ram e notun ghor kinil e f At PF: [ Matrix IP [ NP Ram e] [ vP [ CNP [ NP Ram e ] loteri jik i] [ vP [ NP Ram e ] notun ghor kinil e]]] The following secti on deals with structures that involve a sentence initial CNP clause 3. 4.1.2 Forward Control as s ideward plus remnant m ovement Forward Control may also obtain in constructions where the CNP clause is sentence initial, such as ( 47 a) and ( 48a). As ( 47 b) and ( 48 b) show, the CNP clause in such constructions is pronounced at CP of the matrix clause. (47) Telugu a. [ Kumar ki aakali wees i] Kumar sandwic tinnaa Du [ Kumar DAT hunger fall CNP] Kumar.NOM sandwich ate 3.M.S b CP qp CNP P CP 3 3 SUBJ 6 IP C Kumar ki aakali wees i 3 SUBJ 6 Kumar sandwic tinnaa Du (48) A ssamese a. [ Ram Or phurti lag i] Ram e nasil e [ Ram GEN exhilaration feel CNP] Ram NOM danced 3

PAGE 120

120 b CP qp CNP P CP 3 3 SUBJ 6 IP C Ram Or phurti lag i 3 SUBJ 6 Ram e nasil e Let us assume that the CNP clause in (47 48) is not only pronounced at CP of the matrix clause, but also base generated there. This means that at no point in the derivation can the CNP and matrix s ubjects enter a c command relation ship and thus the two non distinct copies cannot form a chain Accordingly, Chain Reduction, which is dependent on Form Chain, must fail to apply, and no deletion must take place, contrary to fact. An alternative approach is to assume that the CNP clause is base generated at vP of the matrix clause before it moves to the position where it is pronounced. In other words, (47a) and (48a) have the structure in (49) and (50) respectively. (49) T elugu CP qp CNP P 2 CP 3 3 SUBJ 6 IP C Kumar ki aakali wees i qp SUBJ I' Kumar qp vP I qp CNP P 1 vP 3 3 SUBJ 6 SUBJ 6 Kumar ki aakali wees i Kumar sandwic tinnaa Du

PAGE 121

121 (50) Assamese CP qp CNP P 2 CP 3 3 SUBJ 6 IP C Ram Or phurti lag i qp SUBJ I' Ram e qp vP I qp CNP P 1 vP 3 3 SUBJ 6 SUBJ 6 Ram Or phurti lag i Ram e nasil e T he copy of the subject in Spec,IP of the matrix clause c commands both th e copy in Spec, vP and the copy in the lower CNP clause (CNP 1 ) I t f orms a chain with each of them. At PF, Chain Reduction applies, and the lower copy in each chain is deleted. Further, t he two copies of the CNP clause, CNP 2 and CNP 1 also form a chain ; at PF, the lower copy is deleted. The movement of the CNP clause in (49 50) is commonly referred to as r emnant movement It involves movement of a constituent out of which extraction has taken place (Mller 2000). Assuming that control is movement, the CNP c lause in each of (49 ) and ( 50) moves to matrix CP after the CNP subject has moved to the matrix clause. Now the question is: H ow does the subject in CNP 2 get deleted? In order to account for a similar case of remnant movement, Nunes (2004: 50 55) adopts a more elaborate definition of a chain and chain links. F ollowing Chomsky (1995: 300), he holds that chain identified not only in terms of their content, but also in terms of their local

PAGE 122

122 To il lustrate, consider the chain {Kumar, Kumar ki} in (49). I t is made up of the copy of the subject in Spec,IP of the matrix clause and the copy of the subject in the CNP clause. Nunes holds that the two copies must be identified, not only in terms of their c also in terms of their local structural configuration. That is, the chain {Kumar, Kumar ki} must be identified as (51) in which one link is identified as the sister of matrix I' and the other link as the sister of CNP' of the CNP cla use. At PF, Chain Reduction instructs the phonological component to delete the occurrence of Kumar ki that has the structural configuration ( Kumar ki [ CNP' aakali wees i ]). Two such copies exist in (49), one in CNP 1 and one in CNP 2 As Nunes (2004: 54) ma ( Kumar ki [ CN P aakali wees i ]) as ( 52 ) shows (51) (Kumar ki, [ CN P' aakali wees i]) } (52) Telugu [ Kumar ki aakali wees i] Kumar [ Kumar ki aakali wees i] [ Kumar DAT hunger fall CNP] Kumar.NOM [ Kumar DAT hunger fall CNP] sandwic tinnaa Du sandwich ate 3.M.S The analysi s can be extended to the Assamese example in ( 50 ) I n this case, the copy of the subject in Spec,IP of the matrix clause and the copy in the CNP clause form the chain in (53). When Chain Reduction applies, it scans the structure, not for the lower copy per se, but rather for the configuration ( Ram Or [ CNP' phurti lag i ]) T wo copies of this configuration are detected in the structure and, consequently, both are deleted as (54) illustrates (53) {(Ram ( Ram Or, [ CNP' phurti lag i] ) }

PAGE 123

123 (54) Assamese [ Ra m Or phurti lag i] Ram e [ Ram Or [ Ram GEN exhilaration feel CNP] Ram NOM [ Ram GEN phurti lag i] nasil e exhilaration feel CNP] danced 3 The analysis thus far accounts for one side of the coin: Forward C ontrol. We are left with Backward Control, which is the topic of the following section. 3. 4 .2 Backward Control This section provides an analysis of Backward Control constructions in Telugu and Assamese. These are the mirror image of the Forward Control s tructure s presented in Section 3. 4.1. T hey are construc tions in which the subordinate/ CNP subject is pronounced, determining the identity of the implied subject of the matrix clause The section is organized as follows: Section 3. 4 .2.1 presents an analysi s of Telugu Backward constructions as instances of sideward m ovement. Concerning Backward Contro l constructions in Assamese, w e saw in Section 2. 2 that they are judged as unacceptable or degraded. Section 3. 4 .2.2 provides a possible explanation and highlig hts the implications for Case theory. 3. 4.2.1 Backward Control in Telugu Observe the structure in ( 55 ). In both sentences, the CNP subject is pronounced and the matrix subject is implied. (55) Telugu Kumar [Kumar ki aakali wees i] sandwic tinnaa Du Kumar.N OM [Kumar DAT hunger fall CNP] sandwich ate 3.M.S The same mechanism involved in Forward Control and the pronunciation of the matrix subject is involved in Backward Control and the pronounciation of the CNP subject. In other

PAGE 124

124 words, the derivational history of the sentenc e in (55) is almost identical to that of ( 43 ) and ( 45 ) above. More specifically, sentence ( 55 ) has the derivational history delineated in (5 6 ). What makes this derivation different from the derivation of a Forward Control structure is the outcome of the PF operation Chain Reduction. In Forward Control constructions, Chain Reduction deletes the CNP copy of the subject NP, as illustrated in (44 f ) and ( 46f ) above. In ( 56 ), however, Chain Reducti on deletes the matrix copy of the subject NP, leading to Backward Control, as ( 56f ') illustrate s (56) a. i [ CNP [ NP Kumar ki] aakali wees i] [ NP Kumar] [ CNP [ NP Kumar DAT] hunger fall CNP] ii [ Matrix vP sandwic tinnaa Du] [ Matrix vP sandwich ate 3.M.S] b [ Matrix vP [ NP Kumar] sandwic tinnaa Du] c [ Matrix IP [ vP [ CNP [ NP Kumar ki] aakali wees i] [ vP [ NP Kumar] sandwic tinnaa Du]]] d [ Matrix IP [ NP Kumar] [ vP [ CNP [ NP Kumar ki] aakali wees i] [ vP [ N P Kumar] sandwic tinnaa Du]]] e CP 3 IP C qp SUBJ I' Kumar qp vP I qp CNP P vP 3 3 SUBJ 6 SUBJ 6 Kumar ki aakali wees i Kumar sandwic tinnaa Du f At PF: [ Matrix IP [ NP Kuma r ] [ vP [ CNP [ NP Kumar ki] aakali wees i] [ vP [ NP Kumar ] sandwic tinnaa Du]]]

PAGE 125

125 The outcome in (5 6 ) is a little suprising The unmarked situation is for the higher/matrix copy to be pronounced a nd the lower/subordinate copy to be deleted. Why is it possible to delete usually deleted b ecause in most cases it has fewer checked features than the higher copy. This puts the higher copy at an advantage. W hen Chain Re duction applies, it picks the co py with more unchecked features (i.e., the lower copy) and the higher copy escapes deletion. Let us extend this idea to the Telugu structure (55 ) and its derivation in (56 ). As the dotted arrows in (56e) indicate at least two chain s of the subject Kumar are formed. The first chain is {(SUBJ, [ Matrix I']), (SUBJ, [ Matrix v'])}. Out of these two copies, the higher copy in Spec,IP has an advantage of checking more features ( mainly Case), which is why the lower copy is deleted The second chain is {(SUBJ, [ Matrix I']), (SUBJ, [ CNP as far a s feature checking is concerned. Both copies have checked Case, and n either copy has an uninterpretable feature that needs to be checked. When Chain Reduction applies, the operation is free to select either copy for deletion. If Chain Reduction chooses the lower copy, Forward Control obtains I f Chain Reduction chooses the higher copy, Backward Control obtains. The above analysis raises the question: Why does the CNP subject move if it d oes not have a feature to check? For now, I will assume that it moves, not to check a feature of its own, but to check a feature on the target T his kind of movement is triggered by Enlightened Self Interest (L asnik 19 95), rather than Greed. Chapter 5 offers a different and detailed analysis of this issue. It is interesting to note that this optionality is not unique to Telugu. It is also attested in Malagasy Object Control (Potsdam 2006). To illustrate, Potsdam offers evidence from Malagasy that shows an alternation between Forward and Backward Control within the same structure, as exemplified in ( 57 ) (in original (3a b) ) Potsdam adopts the movement approach to control in

PAGE 126

126 order to analyze the relevant structures T he e mbedded su bject c opy plus m erges into the matrix object position B oth copies have all their features checked A t PF, Linearization detects the two copies as non distinct elem ents in a c command relation ship At PF, Chain Reduction applies. S ince both copi es have all their features check, either one of them can be deleted. In the case of Forward Control, the copy in the mat rix clause is pronounced ( 57a). I n the case of Backward Control, the embedded copy is pronounced ( 57b) (57) Malagasy a. Mery ny zaza [hofafana ny zaza ny trano forced Mary the child [sweep the child the house b. Mery ny zaza ny zaza ny trano forced Mary the child [sweep the child the house Now we turn to Backward Control (or the lack of it) in Assamese. 3. 4.2.2 Backward Control in Assamese Unlike in Telugu where Backward Control structures are grammatical the acceptability of Backward Control in Ass amese ranges from degraded to totally unacceptable. Here are the details. All four speakers I have consulted consider Backward Control constructions like the ones in (5 8 ) and ( 59 ) as degraded. These are structures with Inherent Case marked CNP subjects I n ( 58 a b) the CNP subject is genitive, while in ( 59 a b) the subject is experiential nominative (58) Genitve CNP subject a. ?? Ram e [Ram Or dukh log i] bhat na khal e Ram NOM [Ram GEN sad feel CNP] rice NEG ate 3 b ?? Ram [Ram Or bhagar lag i] xui thakil Ram ABS [Ram GEN exhaustion feel CNP] sleep kept

PAGE 127

127 (59) Experiential Nominative CNP subject a. ?? Ram Or [Ram e phurti kor i] bohk lagil Ram GEN [Ram EXP NOM exhilaration do CNP] hunger felt b. ?? Ram Or [Ram e khong kor i] kosto hoise Ram GEN [Ram EXP NOM anger did CNP] trouble was When the CNP subject is Structural Case marked like in ( 60 a b), two consultants fou nd Backward Control degraded, while the two other consultants fou nd them completely unacceptable. 8 (60) Nominative CNP subject a. ??/* Ram Or [Ram e kukur tu heru i] dukh lagil Ram GEN [Ram NOM dog CL lose CNP] sad felt b. ??/* Ram Or [Ram e loteri jik i] phurti lagil Ram GEN [Ram NOM lottery win CNP] exhilaration felt Regardless of the Case marking of the CNP subject, when presented with Backward Control constructions similar to those in ( 58 60 the subject in the matrix clause is missing. To them, the s tructures are salvaged if they read like (6 1 a f), which are instances of Copy Control. Of course, another option is to realize them as Forward Control structures like the ones in (4) above, repeated here as (62) (61) Copy Control a. [Ram Or dukh log i ] xi/Ram e bhat na khal e [Ram GEN sad feel CNP] he.NOM/Ram NOM rice NEG ate 3 8 Regional variation does not seem to have a say in these judgments. Two of the Assamese consultants come from Upper A ssam where a more standard dialect of Assamese is spoken, and two come from Lower Assam where a more Urban dialect is spoken. The sentences in (60) are judged differently by speakers of the same region. In other words, one Upper Assam speaker and one Lower Assam speaker find the sentences marginal, and one Upper Assam speaker and one Lower Assam speaker find them totally unacceptable. Obviously, more field research needs to be done concerning Assamese Backward Control structures.

PAGE 128

128 b. [Ram Or bhagar lag i] xi/Ram xui thakil [Ram GEN exhaustion feel CNP] he.ABS/Ram ABS sleep kept c. [Ram e phurti kor i] tar/Ram Or bohk lagil [Ram EXP NOM exhilaration do CNP] he.GEN/Ram GEN hunger felt d. [Ram e khong kor i] tar/Ram Or kosto hoise [Ra m EXP NOM anger did CNP] he.GEN/Ram GEN trouble was e. /* [Ram e kukur tu heru i] tar/Ram Or dukh lagil [Ram NOM dog CL lose CNP] he.GEN/Ram GEN sad felt f. /* [Ram e loteri jik i] tar/Ram Or phurti lagil [Ram NOM lottery win CNP] he.GEN/Ram GEN exhilaration felt (62) Assamese (CNP and matrix subject Case marked differently) a. Ram e [ Ram Or phurti lag i] Ram NOM [ Ram GEN exhilaration feel CNP] pagolor nisena nasil e crazy person like danced 3 b. Ram [ Ram Or bhagar lag i] xui thakil Ram ABS [ Ram GEN exh austion feel CNP] sleep kept c. Ram Or [ Ram e kukur tu heru i] dukh lagil Ram GEN [ Ram NOM dog CL lose CNP] sad felt d. Ram e [ Ram e khong kor i] mor gh orto bhangil e Ram NOM [ Ram EXP NOM anger did CNP] my house destroyed 3 Notice that not all speakers accept ( 61 e f). The speakers who find the Backward Control constructions in ( 60 ) in which the CNP subject is nominative, degraded find their Copy Control counterparts ( 61 e f) acceptable. Those who find ( 60 a b) unacceptable also rule out ( 61 e f).

PAGE 129

129 That a Backward Control derivation is problematic in Assamese is somewhat surprising. Given wh at we know about Backward Control derivation in Telugu, Assamese should behave the same. The answer to this puzzle resides in the Case characteristics of the CNP subject. Let us assume that every argument enters the computation with an uninterpretable Str uctural Case feature that needs to be checked, even if the argument takes on Inherent Case. Let us also assume that Assamese CNP clauses do not check Structural Case while Telugu CNP clauses do This means that when an Assamese Adjunct Control structure re aches PF, the CNP subject will still have an uninterpretable Structural Case feature that needs to be checked. Take sentence ( 63), for example. T he derivational history of ( 63 ), as delineated in ( 64 ), is almost identical to the derivational history of the Telugu Backward structure in ( 56 ) above B oth derivations involve steps (a e ). What makes ( 64 ) different is step ( f ') where Chain Reduction applies. In ( 56f ) above, the matrix subject and the CNP subject are on equal footing with regard to feature chec king T his is why the deletion of the matrix subject yields an acceptable outcome: Backward Control. In ( 64 ) the CNP copy has an unchecked Stru ctural Case feature This puts it at a disadvantage. W hen Chain Reduction applies, the operation prefers to dele te the CNP subject, as ( 64f ') indicate s. This is why the matrix copy is normally the one that escapes deletion. (63) Ram e [Ram Or khub bhok lag i] posa bhat khal e Ram NOM [Ram GEN much hunger felt CNP] stale rice ate 3 (64) a. i. [ CNP [ NP Ram Or] khub bhok lag i] [ NP Ram] [ CNP [ NP Ram GEN] much hunger felt CNP] ii. [ Matrix vP posa bhat khal e] [ Matrix vP stale rice ate 3] b. [ Matrix vP [ NP Ram e] posa bhat khal e] COPY

PAGE 130

130 c. [ Matrix IP [ vP [ CNP [ NP Ram Or] khub bhok lag i] [ vP [ NP Ram e] posa bhat khal e]]] d. [ Matrix IP [ NP Ram e] [ vP [ CNP [ NP Ram Or khub bhok lag i] [ vP [ NP Ram e] posa bhat khal e]]] e. CP 3 IP C qp SUBJ I' Ram e qp vP I qp CNP P vP 3 3 SUBJ 6 SUBJ 6 Ram Or khub bhok lag i Ram e posa bh at khal e (f ') At PF: [ Matrix IP [ NP Ram e] [ vP [ CNP [ NP Ram Or ] khub bhok lag i] [ vP [ NP Ram e ] posa bhat khal e]]] What about the speakers who fou nd the Backward Control structures in ( 60 a b) marginal but not totally unacceptable? These structur es involve Structural Case marked CNP subjects. Two explanations o r, more likely, stipulations are possible. It might be speculated that nominative Case on CNP subjects is a default Case that is realized in the absence of a licensing head. Although a D efault Case marked subject makes Backward Control tolerable, sentences ( 60 a b) are still considered marginal because the matrix subject, whose Case is licensed by a functional head, has an advantage over the Default Case marked CNP subject T hus, Chain Red uction favors the former over the latter. Another possibility is that th e phenomenon is a change in progr ess and that Assamese CNP clau ses are becoming like Telugu CNP clauses that license Structural Case marked subjects. Before moving to the following sec tion, it is worth noting that Subbar ao (2004) presents data that is in line with the data presented in this section. His conclusion, however, is different.

PAGE 131

131 According to Subbarao, Assamese does allow Backward Control only if the matrix subject is not Case m arked genitive The author provides the example s in ( 65 66 ) (Subbarao 2004: 20 22; (11), (13), (18) and (19)). In ( 65 ), the matrix subject is absolutive. Both Forward Control (6 5 a) and Backward Control (65b) are allowed. 9 In (66 ), however, the matrix subje ct is genitive. Only Forward Control is allowed (66a). Backward Control is unacceptable (66b). (65) a. Forward Control [ CP [ IP Xita [ vP [ CNP Xita Or xahOx thak i u] [ CP [ IP Xita ABS [ vP [ CNP Xita GEN courage keep CNP also] [ vP Xita pOlay gol]]]] [ vP Xita ABS ran away]]]] b. Backward Control [ CP [ IP Xita [ vP [ CNP Xita Or xahOx thak i u] [ CP [ IP Xita ABS [ vP [ CNP Xita GEN courage keep CNP also] [ vP Xita p Olay gol]]]] [ vP Xita ABS ran away]]]] (66) a. Forward Control [ CP [ IP Ram Or [ vP [ CNP Ram ei kotha tu xun i] [ CP [ IP Ram GEN [ vP [ CNP Ram ABS this news CL heard CNP] [ vP Ram Or khong uthil ]]]] [ vP Ram GEN anger raised ]]]] b. Backward Control [ CP [ IP Ram Or [ vP [ CNP Ram ei kotha tu xun i] [ CP [ IP Ram GEN [ vP [ CNP Ram ABS this news CL heard CNP] [ v P Ram Or khong uthil ]]]] [ vP Ram GEN anger raised ]]]] 9 My consultants considere d structures like (65b), with an Inherent Case marked CNP subject, degraded. Nevertheless, (65b) was judged as more acceptable than control structures with a tructural Case marked CNP subject (e.g., (66b)).

PAGE 132

132 [genitive] subject in a Forward Control structure the correspondin g Backward Control structure 22). If the explanation presented in this section is on the right track, then the reason why the Backward Control structure ( 66 b) is unacceptable is due to the typ e of the CNP predicate rather than the Case marking of the matrix subject. As the sentences in ( 67 ), repetition of (59) above s how, even if the matrix predicate licenses a genitive subject, Backward Control constructions are somehow tolerated as long as the CNP clause has an experien tial predicate. In addition, sentence (19c) above, repeated here as (6 8 ) shows that Backward Control may be unacceptable even if the matrix subject is nominative Case marked ( Section 3. 2.2 provides more details ). The reason w hy (68) is totally unacceptable at least to some speakers is because the CNP predicate is a non experiential predicate. (67) Assamese a. ?? Ram Or [Ram e phurti kor i] bohk lagil Ram GEN [Ram EXP NOM exhilaration do CNP] hunger felt b. ?? Ram Or [Ram e khong kor i] kosto hoise Ram GEN [Ram EXP NOM anger did CNP] trouble was (68) Assamese Ram e [January t Ram e loteri jik i] Ram N OM [January in Ram e lottery win CNP] March ot notun ghor kinil e March in new house bought 3 This section has presented an analysis of Telugu a nd Assamese Adjunct Control as s i dewa rd m ovement. The section has also concluded that the CNP subject in both languages is Case marked I n Telugu, the CNP subject is Inherent and Structural Case marked, which is why

PAGE 133

133 Forward and Backward Cont rol can be used interchangeably. I n Assamese, on the other hand, the CNP subject takes on Inherent Case and maybe default Case but not Structural Case, which is why Backward Control is not favored. Knowing that the subject also checks Case in the matrix clause, we are now face to face with a theoretical problem of multiple Case checking. The theoretical issues are built on empirical grounds. The Movement Theory of Control claims that control is just like raising in that they both are derived via movement. The moving element in raising structures does not undergo multiple Case checking. If Control is derived in the same manner then the moving element in c ontrol structres must not undergo multiple Case checking either (Landau 2003). Sections 3. 5 and 3. 6 address these problems and show that they are not ne cessarily a challenge to the approach adopted here. 3.5 Multiple Case Checking and Copy Control Case Theory and the Case Filter as delineated in (Chomsky 1981) and (Chomsky and Lasnik 1995) require that every argument be Case marked in order to be visibl e Visibility can be understood as a PF requirement, as the original formulation of the Case Filter in (69a) indicates; if an NP is not Case marked, it cannot be phonologically realized. It can also be understood as an LF requirement, as the later formulat ion of the Case Filter in (69b) shows; at LF, a link in an argument chain must have Case in order for the chain to be visible for a theta role. Th e Case Filter can be satisfied by either Structural or Inherent Case. Once an argument is Case marked, however it freezes for all further A movement (Chomsky and Lasnik 1995: 111 119; Chomsky 2000: 127) (69) Case Filter a. An NP must be C ase marked in order to be pronounced (Chomsky 1981: 49, Vergnaud 1982).

PAGE 134

134 b. An argument chain must be C ase marked to be visible fo r theta role assignment (Chomsky and Lasnik 1995:46, 119; following Joseph Aoun). Evidence from Backward Control (and Copy Control) indicates that the CNP subject in Adjunct Control structures is Case marked prior to sideward movement, which makes the m ovement approach to control suspect from a traditional Case Theory perspective. The reason is that the movement approach seems to suggest that multiple Case checking is possible, something that Case Theory does not allow. Fortunately, there is strong evide nce that multiple Case checking is a fact about natural languages, rendering the idea that an NP can only check Case once a stipulation. 3.5.1 Multiple Case Checking: Inherent Structural The literature is replete with evidence that multiple Case checking is possible. Belletti (1988), Mohanan (1994), Sigur sson (2004), Woolford (2006), and Yoon (2004) among others provide evidence from several languages (e.g., Hindi, Finish, Icelandic, Korean) to argue that an Inherent Case marked NP may also check Structur al Case. In some languages, both Case markers are phonologically realized. For example, Yoon (2004 : 268 (12a) ) shows that Korean subject NPs check nominative Case on top of the dative Case marker and that the two Case markers are pronounced (e.g., ( 70 ) ) E van s ( 2005 ) offers similar examples from Ka yardild, an Australian language (70) Korean Cheli eykey ka ton i manh ta Cheli DAT NOM money NOM a lot DECLARATIVE T he Korean case is exceptional however. Although an Inherent Case marked argument may also check Structural Case, often morphological restrictions in the language allow only one Case to be realized. The tendency is that Inherent Case take s precedence morphologically. The sentences in ( 71 ) are an example. Sentence ( 71 b) is the passive equivalent of ( 71 a). In ( 71 b), the

PAGE 135

135 genitive Case marked NP Sjklinganna checks nominative Case on a par with vi 71 a). Whereas the nominative Case marker is realized on vi only the genitive is pronounced on Sjklinganna (Bejar and Massam 1999: 68: (6), from Andrews 1990 as cited in Harley 1995). (71) Icelandic a Vi vitjuum sjklinganna We.NOM visited the patients.GEN b. Sjklinganna var vitja sjklinganna The patients.GEN was visited the patients.GEN The reason behind the realization of Inherent Case Woolford (2006: 117; fn. 4) suggests, is that it nonstructural Cases is more import ant than using the less marked S Therefore, the idea that in Assa mese an Inherent Case marked CNP subject moves to the matrix clause where it checks Structural Case is not unheard of One question remains: What about an Assamese CNP subject that checks Inherent Case in the matrix clause as well? Is m ultiple In herent Cas e c hecking possible? Since Inherent Case is interwined with theta role, the answer depends on the possibility of multiple theta role checking. Ac cording to Hornstein (1999 2003), the latter is possible I t follows that Multiple Inherent Case checking is also possible ( Boeckx and Hornstein 2006). 3.5.2 Multiple Case Checking: Structural Structural Unlike Assamese, Telugu CNP subject NPs check not only Inherent Case but also Structural Case before moving to a new Structural Case position This requires mul tiple

PAGE 136

136 Structural Case checking, a phenomenon that is also attested in several languages (Bejar and Massam 1999, Merchant 2006, among others). One example comes from Hungarian (e.g., ( 72 ) ) (adapted from Bejar and Massam 1999: 74 (2), from Kiss 1985 as cited in Massam 1985). Kiket ki in the embedded clause where it is Case marked nominative. On its way to Spec, CP, it checks accusative case against szeretnl that is checked last is pronounced. (72) Hungari an kiket mondtad hogy szeretnl ha ki eljnnnek who.ACC you.said that you.would.like if who.NOM came A ccording to Bejar and Massam (1999: 74), the reason the last Case is pronounced is t hat functional head. In other words, Case is interpreted compositionally. Effectively, this means that the Case subscript is left behind when DP moves out o f one Case checking configuration into a r aising constructions. Accordingly, an alternative explanation will be provided. What is important for our purposes, however, is tha t multiple Case checking is needed on independent grounds and that it is not an ad hoc stipulation that is used to account for Telugu and Assamese Adjunct Control The following section addresses the issue of discrepancy in the C ase behavior of subject NPs in control vs. r aising. 3.6 Case in Raising vs. Control One type of evidence used to argue against the movement approach comes from the diffe rence in Case behavior between r aising subject NPs and their c ontrol counterparts. This section delineates the pr oblem as presented in Landau (2003), highlights its relevance to Telugu and Assamese, and suggests a solution

PAGE 137

137 3.6.1 Landau Analysis One proponent of the PRO Theory of Control is Landau (2000, 2003, 2004, 2006) who holds that control is different from r a ising and that the two constructions have different derivational histories. Landau holds that Control structures consist of two distinct argument chains, one comprising PRO and the other the matrix subject. R aising structures, on the contrary, involve only one argument chain. To illustrate, the c ontrol sentence ( 73 a) is assumed to have the structure in (73 b). PRO is base generated in the vP of the embedded CP where it is assigned a theta role (unlike Hornstein, Landau and PRO Theory in general does not consider theta roles to be features). PRO then moves to Spec, IP where it checks Case and the EPP feature of the embedded I 0 The two copies of PRO form a chain Tom is base generated in vP of the matrix clause where it is assigned a theta role. It moves to Spec, IP where it checks Case and the EPP feature of matrix I 0 The two copies of Tom form a chain Details aside, the identity of PRO is determined by the identity of Tom through Agree (73) a. Tom hopes to know the answer. b. [ CP [ IP Tom i Case, EPP [ vP Tom (b) hopes [ CP [ IP PRO i Case,EPP to [ vP PRO know the answer] The r aising sentence ( 74 a), on the other hand, is assumed to have the structure in ( 74 b). Tom starts out in vP of the embedded IP (according to PRO Theory, with the exception of San Martin 200 4 only c ontrol struc tures have CP emb edded clauses ). Tom makes the trip to matrix Spec, IP, passing through subordinate Spec, IP. Notice that seems occupies a VP rather than a vP because it is unaccusative. At the end of the derivation, all the copies of Tom form a chain (74) a. Tom seems to know the answer. b. [ CP [ IP Tom Case,EPP [ VP seems [ IP Tom EPP to [vP Tom know the answer]

PAGE 138

138 Landau, building on work by Sigurisson (1991), uses evidence from Icelandic to prove his point. Compare the c ontrol structure (7 5 ) (Land au 2003: 492 (40b), from Sigur sson 1991) with the r aising Structure (7 6 ) (Landau 2003: 492 (41b) H skuldur Thra insson) Notice that the dative floating quantifier llum agree s with the unpronounced subject. I t agrees with PRO in (7 5 ) and with a deleted copy or trace in ( 76 ). (75) Icelandic nir vonast til the boys .NOM hope for [a PRO lei ast ekki llum to PRO.DAT to be bored not all.DAT in school (76) Icelandic vir ast the boys .DAT seem [ 10 lei ast ekki llum the boys .DAT to be bored not all.DAT in school What is crucial for our purposes is the difference in the Case marking of the matrix subjects. The matrix subject in the c ontrol structure (7 5 ) is Case marked nominative, which is different from the Case marking of the embedded subject which is dative. The matrix subject of the r aising structure ( 76 ), however, takes on the same Case as the embedded subject: dative. Landau maintains that if control is movement, just like r aising, then Case in (7 5 ) s hould pattern the same as in (76 ). He uses this as an argument that c o ntrol is different from r aising. Only r aising is derived by movement, whereby the embedded subject an d the matrix subject form one A chain which is Case marked once Con trol is not derived by movement. PRO does 10 In Landau (2003), the embedded subject is prese nted as a t trace.

PAGE 139

139 exist, and PRO and the matrix subject form two distinct A chains that are each Case marked on ce. I n this case, the PRO chain is dative, and the ma trix subject is nominative. 3. 6 .2 Raising vs. Control in Telugu and Assamese in ( 77 ). In ( 77 a), the subject is dative. I n ( 77 b), it is nominative. (77) Telugu a. Sarita ki Nidra waccin di Sarita DAT sleep came 3.N.S b. Sarita aadiwaaraal lo pani ceestun di Sarita.NOM Sundays on work do 3.N.S W hen each of the sentences in ( 77) is used in a r aising construction with the the subject is realized with the Case it checks in the embedded clause. Sentence ( 78 a) is the r aising counterpart of (7 7 a). Sarita is dative in both. Sentence (78b) is the r aising counterpart of (78 b). Sarita is nominative in both. The r aisi ng verb has no effect on the subject as far as Case is concerned. (78) Telugu a. Sarita ki Nidra raaw aTam modalu ayyin di Sarita DAT sleep come GRND begin happened 3.N.S b. Sarita aadiwaaraal lo pani ceey aTam modalu p eTTin di Sarita.NOM Sundays on work do GRND begin did 3.N.S W hen the sentences in ( 77) are used in c ontrol constructions, however, the subjects take o n the Case associated with the c ontrol verb. For example, in ( 79 ) the verb ishTam leeDu ontrol verb that Case marks its subject dative. This is why the subject is dative in both ( 79a b). According to Landau, if c ontrol is movement, ( 79a b ) should pattern with ( 78a b),

PAGE 140

140 (79) Telugu a. S arita ki Nidra raaw aTam ish Tam lee Du Sarita DAT sleepy come GRND like GRND is not 3.M.S b. Sarita ki aadiwaaraal lo pani ceey aTam ish Tam lee du Sarita DAT Sundays on work do GRND like GRND is not 3.N.S T he same applies to Assamese. Observe the sentences in (80). I n (80 a) the subject is genitive. I n (80 b), it is nominative. When the sentences are used in r ai sing constructions, the subject preserve s the Case it has c hecked downstairs, as (81a b ) illustr ate (80) Assamese a. Ram Or khong uthil Ram GEN anger got b. Ram e khong koril e Ram NOM anger did 3 (81) Assamese a. Ram Or khub xonkale khong uth a zen lage Ram GEN very fast anger get GRND like feel b. Ram e khub xonkale khong kor a zen lage Ram N OM very fast anger do GRND like feel C ompare to ( 82). Sentence (82 a) has a dat ive subject. When used in a c on trol structure like ( 82 b), its subject takes on the case associated with the matrix verb; in this case, nominative. (82) Assamese a. Ram Or thanda lagil Ram GEN cold felt b. Ram e thanda lag a bo ni bisar e Ram NOM cold feel GRND FUT NEG want 3

PAGE 141

141 to make the movement approach to Control questionable. A counterargument, however, is available as the following section shows. 3.6.3 Case in Raising vs. Control: the Counterargument One counterargument has been proposed by Boeck x and Hornstein (2006) Their reply is summarized in Section 3. 6.3.1. The argument falls short of accounting for multiple Structural Case checking. Section 3. 3. 6.3.1 Boeckx and Hornstein nalysis Boeckx and Hornstein (2006) reply to Landau by giving a mov em ent account of the Icelandic c ontrol data. They observe that multiple Case checking in Icelandic Control structures only occurs when the case on either the embedded subject or the matrix subject or both is Inherent Case. Under the standard assumption that Inherent Case is directly associated with theta role and that an A chain in a c ontrol configuration bears two theta roles, it follows naturally that multiple Inherent Case assignment is po ssible. In other words, if the c ontroll er and the c ontrollee, which are two copies of the same argument, bear two different theta roles, there is no reason why these two copies should not bear two distinct Inherent Cases associated with the theta roles. As I understand Boeckx we en control and r aising in Icelandic can be presented schematically as in ( 83 ). In the c ontrol structure ( 83 a), the subject checks a theta role and possibly Inherent Case in the embedded clause. Then the subject moves to the matrix clause where it checks an other theta role and possibly another Inherent Case. Finally, the subject lands in Spec,IP where it checks Structural Case. If Inherent Case is checked in the matrix clause, it gets pronounced since it is the more marked situation O therwise,

PAGE 142

142 Structural Ca se is realized. In r aising constructions ( 83 b), the subject checks a theta role feature and possibly Inherent Case in the embedded clause before it moves to Spec,IP of the matrix clause where it checks Structural Case. If Inherent Case is checked downstair s, it is eventually realized fo r the same aforementioned reason, namely, markedness. (83) a. Control: [ CP [ IP Subject [ vP Subject [ IP Subject [ vP STRUCTURAL (INH ERENT (INH ERENT b. Raising: [ CP [ IP Subject [ IP Subject [ vP STR UCTURAL (INH ERENT proposes a solution 3.6.3.2 Case and Theta Role Visibility The problem with Boeckx that Structural Case cannot be checked in embedded non finite clause s and that a chain is always realized with one and only one Structural Case. These assumptions might work for Assamese in which a CNP subject most probably checks Inherent Case only. The assumptions do not work for Telugu, however, in which the CNP subject do es check Structural Case. Here we might adopt Bejar and is possible and simply add it to Boeckx According to the ir system, Case is interpreted locally, and the last Case that an argument checks is the one that is phonol ogically realized This obs ervation does not apply to the r aising constructions presented in the previous section (e.g. (80 81) ) What seems to be ha ppening in r aising constructions is that a subject NP checks Case downstairs before moving into a Case the Case associated with

PAGE 143

143 the subordinate clause is realized. In other words, altho dominated by an appropriate functional head the Case subscript that the subject takes on in the subordinate clause is [NOT] left behind when DP moves out of one Case checking In order to solve this problem, I build on both systems and suggest a Principle that I call Theta Role Visibility. (84) Theta Role Visibility 11 a. An argument is visible for one round of Case checking iff it merge s into a t hematic position. b. A round of Case comprises Inherent C ase followed by Structural Case, depending on the availability of an appropriate licenser for each. What Theta Role Visibility amounts to is the following. erge is a thematic position T his makes it visible for a round of Inherent and Structural Case checking, depending on the availability of an appropriate licenser for each I f both licensers are available the argument checks both Cases. T he result is usually the ph onological realization of Inherent Case If an argument moves into a new thematic position, the argument becomes visible for a new round of Case checking even if it has already checked Case This e xplains why the subject in the c ontrol structures in Icelan dic, Telugu, and Assamese is realized with the Case of the matrix clause. This is because the subject checks a new theta role feature in the matrix clause, which makes it visible for a new round of Case checking. Let us see how Theta Role Visibility as def ined in (84) applies to the Adjunct Control structures under examination. Take the Telugu sentences in (85) as an example Kumar checks a theta role feature ] and Case [ CASE1] in the CNP clause. [ CASE1] comprises Structural Case 11 This is just the reverse of the traditional Case Visibility which assumes that an argument must check Case in order to be visible for theta role assignment (Chomsky and Lasnik 1995:46, following Joseph Aoun)

PAGE 144

144 only (nominative) in (85a), but Inherent Case plus Structural Case (dative plus nominative, with the former phonologically realized) in (85b). Kumar undergoes Sideward Movement to a new thematic position [ ] in the matrix clause T his makes it visible for a new round of Case checking [ CASE2 ] which happens to comprise Structural Case only ( nominative ) in both sentences At PF, Chain Reduction chooses which copy to delete and which t o pronounce.The result can be either Forward or Backward Control. (85) a. [ CP [ IP Kumar CASE2 [ vP [ CNP Kumar sinema cees tuu] [ CP [ IP Kumar.NOM [ vP [ CNP Kumar.NOM movie watch CNP] [ vP Kumar popkorn tinna Du]]] [ vP Kumar popcorn ate 3.M .S]]] b. [ CP [ IP Kumar CASE2 [ vP [ CNP Kumar ki jwaram wacc i] [ CP [ IP Kumar.NOM [ vP [ CNP Kumar.DAT fever come CNP] [ vP Kumar hospital weLLaa Du]]] [ vP Kumar hospital went 3.M.S]]] What happens if the argument moves into a new Case position without moving into a new thematic position? This is exactly the case of the Icelandic passive construction in (71b) and the I celandic, Telugu, and Assamese r aisin g constructions presented above. T he argument moves into a Case position without landing in a new/second theta role position first. According to Theta Role Visibility if the argument has already completed a round of Case checking that i s, if it has checked Structural Case, possibly on top of Inherent Case no further Case checking i s possible. In other words, movement does not make the argument visible for a new round of Case checking. This is why the subject is realized with the Case it checks in the subordinate clause Theta Role Visibility seems to work for Telugu and Assamese, as well as Icelandic. As a matter of fact, all the cases of multiple Structural Case checking that I know of involve

PAGE 145

145 movement into a new theta role position. Nevertheless, there is one instance of multiple Case checking that involves movement into a non theta role position: copy raising structures of the type illustrated in (86) below. Fujii (2005: 13 15) argues that structures similar to (86) involve movement ( contra Potsdam and Runner 2001), and that this movement involves multiple Case checking. (86) a. Tom seems as if he knows the answer b. [ CP [ IP Tom CASE2 [ VP seems [ CP as if [ IP Tom CASE1 he [vP Tom know s the answer] The argument for multiple Case checking is based on structures like (87a b) (in original (43a) and (45)). In both structures, John starts out in the lower CP before moving to the subject position of the infinitival clause to seem Acc ording to Fujii, the ungrammaticality of (87a) is a proof that John needs to check Case in its new position. Following Bejar and Massam (1999), he holds that John strands its Case behind before moving to the higher subject position. The infinitival clause does not check Case; neither does the nominal belief Consequently, the higher copy of John grammatical alternative (87b). The verb believe is able to check the Case feature on Jo hn Accordingly, no Case Filter violation is induced. (87) a. *the belief [John to seem [ CP like John is intelligent] b. I believe/expect [John to seem [ CP like John is intelligent]] Closer examination mmaticality of (87a) follows from the selectional requirements and the uninterpretable Case feature on belief rather than on John As originally observed by Stowell (1982), even if the nominal in (87) has a finite CP complement that licenses a Case marked subject, the structure will still be ungrammatical

PAGE 146

146 unless the complement is introduced with an overt complementizer. This observation is illustrated in (88a b) (Pesetsky and Torrego 2006: 5 (8a) and (9a)) In both structures, Mary checks its Case feature a gainst the functional layers in the CP complement. Therefore, the Case feature on Mary cannot be the reason why (88a) is ungrammatical. According to Pesetsky and Torrego (2006), (88a) is ungrammatical because the nominal proof has an uninterpretable Case f eature that may be checked by that or by a prepos ition, but not by an NP/DP. Section 5. 5 in this dissertation offers more details. (88) a. *your proof Mary could not have committed the crime b. your proof that Mary could not have committed the crime If Pesets (87a) is the result of an unchecked Case feature on belief rather on John Accordingly, it is reasonable to assume that even if copy raising is movement, no multiple case checkin g takes place. The reason is that the subject does not touch down in a thematic position when it moves to the higher clause. If this is correct, then the Theta Role Visibility as formulated in (84) continues to hold. 12 12 The discussion does not explain why the raising subject in (i) is realized with two different Structural Cases, accusative and nominative. One way around this problem is to assume the Inverse Case Filter (Bo kovi 1997, 2002: 170 171). More specifically, there seems to be a requi rement that an accusative Case licenser in this case, the verb expected check/value the Case feature of an element. This requirement does not seem to be as restrictive with respect to nominative Case. Consider (ii) and (iii), for example. The standard assumption is that here is not Case marked. Nevetheless, it is allowed in a subject position (e.g., (ii), but not in an object position (e.g., (iiia). In other words, I 0 in (ii) does not obey the Inverse Case Filter in that it does not have to put the nom inative Case it bears into effective use. In (iiia), on the other hand, the verb love needs to make use of the accusative Case it bears. Note that (iiib) is grammatical simply because the verb has the chance to value the Case of it (Bo kovi 2002: 171, fn. 4). (i) Tom expected her to seem as if she is sick. (ii) Here is a good place to live. (iii) (a) *I love here. (b) I love it here.

PAGE 147

147 3.7 C onclusion This chapter presented a detailed analysis of Forward and Backward Control structures in Telugu and Assamese. Both types of Control are analyzed as instances of Sideward Movement. They both have the same derivational history. The difference between the two lies in the phonologic al component where Chain Reduction applies for the purpose of linearization. Forward Control is licensed in both languages and it obtains when Chain Reduction chooses to delete the CNP subject NP. Backward Control, on the other hand, obtain when Chain Redu ction deletes the matrix subject NP. This latter option is available only for Telugu. The reason is that the CNP clause in Telugu checks the Structural Case feature of the subject. In this way, the CNP subject and the matrix subject are on equal footing wi th regard to feature checking and consequently Chain Reduction is free to mark either copy for deletion. An Assamese CNP clause, on the other hand, does not check the Structural Case feature of its subject. This puts the CNP subject at a disadvantage, maki ng it a more susceptible victim of Chain Reduction. In addition to the analysis, the chapter dealt with the theoretical problem of multiple Case checking that the movement approach to control faces. Section 3.5 showed that there is enough empirical evidenc e to rule out any challenge that traditional Case Theory might present to the movement approach. Section 3.6 dealt with the empirical side of multiple Case checking. Considering that the movement approach derivationally puts control in the same category as raising, the section focused on the discrepancies in the Case behavior of s ubject NPs in raising vs. control structures and puts forth a solution that can be summarized in a principle I call Theta Role Visibility.

PAGE 148

148 CHAPTER 4 COPY ADJUNCT CONTR OL IN TELUGU AND ASSAME SE: THE ANALYSIS 4.1 Introduction Chapter 2 provided evidence for the different types of Adjunct Control allowed in Telugu and Assamese. Chapter 3 outlined the mechanisms involved in the derivation of two types of Adjunct Control: F orward and Backward Control. This chapter analyzes a third type of Adjunct Control that both languages allow: Copy Control. As (1) illustrates, Copy Control constructions involve two coreferential subjects that are phonologically realized. The sentences i n (2) and (3) are exampl es. They show that the CNP subject and the matrix subject are o bligatorily coreferential. D isjoint subjects result in ungrammaticality. The two subjects may be Case marked differently (e.g., (2a b) and (3a b)), or the same (e.g., (( 2c d) and (3c d)). 1 The m atrix subject maybe realized as a pronoun, an epithet, or an R expression. 2 1 In (3d), the CNP subject is Case marked experiential nominative, while the matrix subject is agentive nominative. Althou gh the types of Case involved correspond to different thematic roles, the phonological forms are the same. 2 The sentences in (2 3) exclude instances of control that have received inconsistent judgments from my consultants. These are limited to the Telugu Adjunct Control structures with a dative subject in the matrix clause (e.g., (ia b)), and to the Assamese Adjunct Control structures with a non experiential predicate in the CNP clause (e.g., (iia b).Chapters 2and 3 offers more details concerning why these sentences are degraded: (i) Telugu a. ??[Kumari aa maaTa win i] atanu ki koopamu waccin di [Kumar NOM that matter hear CNP] he DAT anger came 3.F.S b. ??[Kumar ki juttu uuDipooy i] atanu ki picci paTTin di [Kumar DAT hair lose CNP] he DAT craziness caught 3.F.S (ii) Assamese a. /*[Ram e kukur tu heru i] tar/ besera tu r dukh lagil [Ram NOM dog CL lose CNP] he.GEN/poor guy sad felt (b ) /* [Ram e loteri jik i] xi notun ghor kinil e [Ram NOM lottery win CNP] he.NOM new house bought 3 aving won the lott

PAGE 149

149 (1) Copy Control [ Matrix [ Subordinate Subject i ] [ Matrix Subject i ]] (2) Telugu a. [Kumar ki i jwaram wacc i] [Kumar.DAT fever come CNP ] atanu i/*k /Kum ar hospital weLLaa Du he.NOM /Kumar.NOM hospital went 3.M.S b. [Kumar ki i koopamu wacc i] [Kumar DAT anger come CNP] atanu i/*k / aa pichooDu i/*k /Kumar akkadi nunci wellipoyinaa Du he.NOM / that idiot.NOM /Kumar.NOM there from left 3.M.S c. [Kumar i sinima cuus tuu] [Kumar.NOM movie watch CNP] atanu i/*k / aa pichooDu i/*k /Kumar Nidra pooyaa Du he.NOM/ that idiot.NOM /Kumar.NOM as leep fell 3.M.S d. [Sarita i enimid inTiki bhojanamu tayaru ceesikun i] [Sarita.NOM eight time dinner prepare do for self CNP] aame i/*k /Sarita tommid inTiki tinnaa di she.NOM/Sarita.NOM nine time ate 3.F.S (3) Assamese a. [Ram Or i bhagar lag i] [Ram GEN exhaustion feel CNP] etiya xi i/*k /Ram xui thakil now he.ABS/Ram ABS sleep kept b. [Ram Or i khub bhok lag i] [Ram GEN very hunger feel CNP] xi i/*k / beseratu e i/*k /Ram e posa bhat khal e he.NOM /the poor guy NOM/Ram NOM stale rice ate 3

PAGE 150

150 c. [Ram Or i e ta bhal buddhi khela i] [Ram GEN one CL good idea play CNP] tar i/*k /Ram Or phurti lagil he.GEN /Ram Or exhilaration felt d. [Ram e i khong kor i] [Ram EXP NOM anger raise CNP] xi i/*k / gadha tu e i/*k /Ram e mor ghorto bhangil e he.NOM/ the donkey NOM /Ram NOM my house destroyed 3 A s I mentioned in Chapter 2, Copy Control obtains only if two conditions are met. T he CNP claus e has to be sentence initial, and the CNP subject has to be a non pronominal R expression. If either of these two conditions is violated, Copy Control becomes unacceptable. For example, the CNP clause in (4a) and (5a) is not sentence initial; both sentences are ungrammatical. The CNP clause in (4b) and (5b) is sentence initial, but the CNP subject is a pronominal; again, both sentences are ungrammatical. (4) Telugu a. atanu/Kumar [atanu ki/Kumar ki aakali wees i] he.NO M/Kumar.NOM [he DAT/Kumar.DAT hunger fall CNP] bhojanamu tayaru ceesikunaa Du dinner prepare did for self 3.M.S b. *[atani ki aakali wees i] [he DAT hunger fall CNP] Kumar/ata nu bhojanamu tayaru ceesikunaa Du Kumar.NOM/he.NOM dinner prepare did for self 3.M.S (5) Assamese a. xi/Ram e [tar/Ram Or ananda lag i] nasil e he.NOM/Ram NOM [he/Ram GEN happines s feel CNP] danced 3

PAGE 151

151 b. [tar ananda lag i] Ram e/xi nasil e [he GEN happiness feel CNP] Ram NOM/he.NOM danced 3 Section 4. 3 presents the derivational history of the Copy Con trol structures under examination. The presentation is based on the assumption that Copy Control is movement. There is a possibility, however, that Copy Control obtains as a result of base generated resumption. In this case, the matrix subject NP is a base generated resumptive copy rather than a copy of a moving element. Section 4. 2 examines this possibility and concludes that Copy Control must be movement. Sections 4. 4 through 4.6 address problems that the derivation presented in Section 4. 3 faces. Section 4. 7 is a conclusion. 4.2 Copy Control as Movement There are two types of resumptive element s argued for in the literature. These are base generation resumptive elements that relate to their antecedent via binding, a nd resumptive elements that are the result of movement ( Aoun, Choueiri, and Hornstein 2001 and Boeckx 2003 among others). 3 There are at least three reasons to believe that Copy Control is the result of movement and to rule out the possibility of base generation To begin with, base generated resumptive elements show up in positions out of which movement is illegal. For example, they show up in complex noun phrases, wh islands, and adjunct islands (Aoun, Choueiri, and Hornstein 2001: 372; McCloskey 2005: 11 12). Let us assume t hat movement out of CNP clauses including s ideward m ovement, is illegal for the simple reason that CNP clauses are adjuncts In this case, one would expect a resumptive element to be realized inside the CNP clause all the time, contrary to fact s Copy Con trol obtains 3 Aoun, Choueiri, and Hornstein (2001) labe l base generated resumption as true resumption, and they consider resumption that is the outcome of movement as apparent resumption. Beockx (2003), building on Sells 1984, generally agrees with Aoun, es the two types of resumption differently. He labels base generated resumption as intrusive resumption and resumption that is the outcome of movement as true resumption.

PAGE 152

152 only if the CNP clause is sentence initial. If the CNP clause is sentence internal, Copy Control is unacceptable. This implies that sideward movement out of CNP clauses is legal and that no base generated resumptive elements are involved. T he conclusion is in line with the assumption that the CNP subject undergoes sideward movement before the CNP clause acquires the status of an adjunct; that is, before it actually adjoins to the matrix c lause ( Rodrig uez 2004: 114 and works within). In addi tion, McCloskey (2005: 1 3) observes that a resumptive element may be either a pronoun (clitic, strong pronoun, or even pro ( Cinque 1990)) or s in Section 4. 1 and in Chapter 2, both pronounced subjects in a Copy Control construction may be R expressions. This observation is not r estricted to proper nouns ( Kumar, Ram, etc.) ; it extends to all types of NPs, as (6 7) illustrate. Longer cop ies are normally judged as redundant, but they are not ungrammatical. (6) Telugu [boss ki / Naa boss ki pooyina waaram koopamu wacci] [boss DAT /my boss DAT last week anger come] boss / Naa boss ibbandi peTTa Du boss.NOM / my boss trouble put 3.M.S (7) Assamese [boss tu r /mor boss tu r bhal lag i] [boss CL GEN / my boss CL GEN good feel CNP] Boss tu e /mor boss tu e employee burok bonus dil e Boss CL NOM / my boss CL NOM employee ACC bonus gave 3 McCloskey does not mention whether his observation targets based generated resumption or resumption that is the outcome of movement or both. Assuming that it targets both, the

PAGE 153

153 exceptions to this observation in (6) and (7) are more easily accommodated if the matrix subjects are the outcome of movement. The reason is that movement is understood as copy plus merge. T he CNP subject copies out of the CNP clause and merges in the matrix clause a s an exact copy. Compare to the view that (6) and (7) are the outcome of base generated resumption. Under this view, it is hard to explain how a resumptive element base generates as an exact copy of the element it refers to. Further, resumptive elements no rmally appear in what otherwise is the locus of a gap or a trace (Boeck x 2003: 14, McCloskey 2005: 94 Sells 1984 and Shlonsky 1992 among several others). 01: 327 329) convincingly argues, this means that gap s occupy a subordinate structure 4 Similarly, resumptive elements must be restricted to subordinated domains. If Copy Control were the result of base generated resumption, one would expect t he subordinate/CNP subject to be a resumptive pronominal and the matrix subject to be an R expression. This is not the case, however. In Copy Control constru ctions, t he subordinate subject has to be an R expression, otherwise the structure is ungram matical. It is hard to imagine how a base generated resumptive pronominal could be realized as an R expression that has to relate to an antecedent possibly a pronoun or an epithet via binding. Notice that this restriction does not pose a problem for th e movement approach to Copy Control. Under the Copy Theory of Movement (Chomsky 1995), the CNP subject copies out of 4 Nunes (2001: 327 Evidence comes from structures like ( i) below ( Nunes 2001: 327 328; (62a) and (66a b)). Sentence ( i a) is ungrammatical because by the time borrow requires which book to undergo Sideward Movement, PP without finding which book is al ready an island out of which movement is prohibited as PP in (ib) shows system could first start building the matrix derivational workspace before building an embedded derivational workspace, the sentence in [ ( i a) ] for instance, wo uld be incorrectly ruled in by a derivation where sideward movement proceeded from the object of borrow to the object of finding (i) a. *Which book did you borrow after leaving the bookstore without finding? b. [CP PRO [vP[vP leaving t he bookstore] [PP without PRO finding [which book] ]]] c. borrow

PAGE 154

154 the CNP clause and merges in the matrix clause as an exact copy of the same token. Decisions concerning which copy should be pronounced as an R expression and which copy should be a pronominal are made in the phonological component, most probably in accordance with precedence relations as we will see in Section 4. 5. 5 This section has tried to show that the pronominal subject in Copy Control constructions is unlikely to be a based generated resumptive element. The reaso ns can be summarized as follows. First, b ase generated resumptive pronominals show up in islands (adjuncts, NP clauses) that are immune to movement, and they show up all the tim e. Copy Control, on the other hand, is not restricted by the adjunctive nature of the CNP clause. Rather, it is restricted by the position of the CNP clause with respect to the matrix clause that is, whether the CNP clause is sentence initial or sentenc e internal. Second, r esumptive elements are strictly pronominals (pronouns or epithets). Although the matrix subject in Copy Control structures may be a pronominal, it may also be realized as an R expression that is a non distinct copy of the CNP subject. Resumption R expressions can be straightforwardly accounted for if they are considered as the outcome of movement, but not as straightforwardly if they are considered to be base generated. Finally, r esumptive pronominals only show up in subord inated domains that usually fail to function as launching sites for movement. When a pronominal subject is pronounced in a Copy Control construction, it shows up in the landing site: the matrix clause. The above discussion is not meant to spell out a theo ry of resumption. This task is beyond the scope of the present work. It is simply a brief diagnosis in order to show that Copy Control is 5 The cases of resumption as movement presented by Aoun, Choueiri, and Hornstein (2001) and Boeckx (2003) also seem to obey the restriction that resumptive elements are pronominals that occupy a subordinated domain. This is because the authors argue for a stranding approach to resumption that is reminiscent of quantifier floating. I return to this issue in Section 4.6 where I suggest a non stranding alternative.

PAGE 155

155 unlikely to be an instance of base generated resumption. The chapter proceeds by analyzing Copy Control as movement. 4.3 Copy Control: The Derivational History The derivation of Forward/Backward Control presented in Chapter 3 applies, with minor movement According to Nunes, movement comprise s fou r independent operations: Copy, Merge Form Chain and Chain Reduction. chain in accordance with Form Chain as formulated in (8). At PF, the structure is linearized in order to satisfy the Linear Correspondence Axiom in (9). The main idea behind (9) is that in real time a syntactic object cannot follow and precede itself at the same time. This is when Cha in Reduction, (8) Two constituents X and Y form a chain iff a. X and Y are non distinct; b. X commands Y. (9) Let X, Y be nonterminals and x, y terminals such tha t X dominates x and Y dominates y. Then if X asymmetrically c commands Y, x precedes y. (Kayne 1994: 33) (10) Delete the minimal number of constituents of a nontrivial chain CH that suffices for CH to be mapped into a linear order in accordance with the LCA ( Nunes 2004: 27 (44)). Based on the above, the Telugu Copy Control structure in (11) will have the derivational history outlined in (12). The CNP clause and the m atrix clause form independently in (12a). The CNP subject copy plus merges with matrix vP in ( 12b). In (12c), t he matrix subject moves from Spec,vP to Spec,IP to check the EPP feature Following, t he CNP clause merges with the matrix clause at CP, as ( 12d i) shows. The tree in ( 12d ii) is a snapshot of the derivation up to this point. It mainly hig hlights c command. As the dotted arrows show, the two matrix copies of Kumar {[ NP Kumar d ], [ NP Kumar c ]} enter a c command relationship and form a chain The two CNP

PAGE 156

156 copies {[ NP Kumar b ], [ NP Kumar a ]} also enter a c command relationship and form a chain No CNP copy, on the other hand, enters a c command relation with a matrix copy. At PF, Chain Reduction applies for the purpose of Linearization. As (12e) illustrates the lower copy of Kumar in each chain is deleted. Thus, [ NP Kumar c ] and [ NP Kumar a ] un dergo deletion. Two copies, [ NP Kumar b ] and [ NP Kumar d ] survive deletion, resulting in Copy Control. (11) Telugu [Kumar sinima cuus tuu] Kumar popkorn tinnaa Du [Kumar.NOM movie watch CNP] Kumar.NOM popcorn ate 3.M.S (12) a. i. [ CNP [ NP Kumar] [ vP [ NP Kumar] sinima cuus tuu]] [ CNP [ NP Kumar.NOM] [ vP [ NP Kumar.NOM] movie watch CNP]] ii [ Matrix vP popkorn tinnaa Du] [ Matrix vP popcorn ate 3.M.S] b. i. [ NP Kumar] ii [ Matrix vP [ NP Kumar] popkorn tinnaa Du] c. [ CP [ Matrix IP [ NP Kumar] [ Matrix vP [ NP Kumar] popkorn tinnaa Du] d. i. [ CP [ CNP [ NP Kumar b ] [ vP [ NP Kumar a ] sinima cuus tuu]] [ CP [ Matrix IP [ NP Kumar d ] [ Matrix vP [ NP Kumar c ] popkorn tinnaa Du]]]] ii. CP qp CNP P CP qu 3 SUBJ I' IP C Kumar b 3 3 vP I SUBJ I' 3 Kumar d 3 SUBJ 6 vP I Kumar a sinima cuus tuu 3 SUBJ 6 Kumar c popkorn tinnaa Du e. [ CP [ CNP [ NP Kumar b ] [ vP [ NP Kumar a ] sinima cuus tuu]] [ CP [ Matrix IP [ NP Kumar d ] [ Matrix vP [ NP Kumar c ] po pkorn tinnaa Du]]]]

PAGE 157

157 The main difference between (12) and the derivational history of Forward/Backward Control structures is the merging site of the CNP clause. In Copy Control constructions, the CNP clause merges clause initially at CP, as (12d) shows. I n Forward/Backward Control constructions, however, the CNP clause merges clause internally at vP. The exact same steps apply to Assamese. Thus, the Copy Control structure in (13) has the derivation in (14). The CNP clause merges with the matr ix clause at CP. The dotted arrows in (14) indicate that the two matrix copies of Ram {[NP Ram e d ], [NP Ram e c ]} enter a c command relationship and form a chain They also indicate that the two CNP copies {[NP Ram Or b ], [NP Ram Or a ] } enter a c comman d relationship and form a chain N one of copies of the CNP subject enters a c command relationship with the matrix subject. At PF, Chain Reduction applies for the purpose of Linearization. As a result, the lower copies of Ram [ NP Ram Or a ] and [ NP Ram e c ], are deleted. The two higher copies, [ NP Ram Or b ] and [ NP Ram e d ] ,survive deletion. T he outcome is Copy Control. (13) Assamese [Ram Or bhok lag i] Ram e bhat khal e [Ram GEN hunger feel CNP] Ram NOM rice ate 3 (14) CP qp CNP P CP qu 3 SUBJ I' IP C Ram Or b 3 3 vP I SUBJ I' 3 Ram e d 3 SUBJ 6 vP I Ram Or a bhok lag i 3 SUBJ 6 Ram e c bhat khal e

PAGE 158

158 One que stion that comes to mind is: How is a con trol interpretation established without c command? In other words, how does the matrix subject determine the identity of the CNP subject or vice versa if the two do not enter a c command r elationship? The simple answer, and probably the only one, is movement. The two copies are co referential because they are copies of the same token derived via movement. The analysis prese nted in this section faces three problems. These can be summarized as follows: Problem 1: The two pronounced subjects in a Copy Control structure escape deletion due to the lack of c command and Form Chain. Nevertheless, according to the Linear Correspondence Axion as stated in (9), they are still non distinct copies in a precedence relationship. Accordingly, they still indu ce a violation. Problem 2: The CNP clause becomes an adjunct and thus an island for movement only when it adjoins to the matrix clause. In a Copy Control structure, the CNP clause merges with the matrix clause at CP rather than at vP. This means that the CNP clause is accessible for movement for a longer period of time, which may result in the overgeneration of undesired structures. Problem 3: The analysis in this section does not explain why the CNP subject has to be an R expression while the matrix subject can be either an R expression or a pronominal (a pronoun or an epithet). 6 The following sections explain the problems in more detail and suggest solutions. I begin with the issue of linearization. 4. 4 Copy Control and Linearization The Linear Corr espondence Axiom as formulated in (9), repeated here as (15), predicts that the derivations in (12) and (14) above must not converge. The reason resides in the definition of precedence. As (1 5 ) indicates, a terminal x precedes a terminal y if x and y are i n a c command relationship or if the non terminal X that dominates x c commands y 6 Another problem is related to movement: Why does movement take place? Chapter 5 provides a detailed answer to this question.

PAGE 159

159 (15) Let X, Y be nonterminals and x, y terminals such that X dominates x and Y dominates y. Then if X asymmetrically c commands Y, x precedes y. (Kayne 1994: 33) If we apply ( 15) to the derivations in (12) and (14), we realize that at the end of the derivation the non terminal CNP clause asymmetrically c commands the matrix subject, as (16) illustrates. Therefore, the CNP subject must precede the matrix subject. Since the two s ubjects are copies of the same token that is, they are non distinct then the same element precedes and follows itself in the same structure, inducing a violation of the Linear Correspondence Axiom. Therefore, one of the copies must be deleted in orde r for the structure to converge. (16) CP qp CNP CP 3 3 SUBJ 6 IP C 3 SUBJ 6 reason is t hat they involve two non distinct copies that ar e in a precedence relationship; thus, they need to be linearized and one of them has to be deleted. Nevertheless, the two copies do not form a chain because neither copy c commands the other. Consequently, Ch ain Reduction cannot apply at PF and the structure cannot be mapped into a linear order in accordance with the Linear (2004: 51 52, 159) holds that Form Chain, although an independent operation, is mandatory. If Form Chain does not apply, the derivation crashes. The derivations in (12 ) and (14) do converge, however. T wo copies escape Chain Reduction and are actually pronounced. This means that the theory must be able to accommodate PRECEDES

PAGE 160

160 Knowing that Copy Control may be grouped with other instances of mult iple copy spell out in which more than one copy of the same token are phonologically realized, it is only reasonable to examine some analyses proposed for the phenomenon before putting forward a separate analysis. Section 4. 4.1 undertakes thi s task and presents two analyses, one proposed by Nunes (2001, 2004) and one by Fujii (2005). Building on Section 4. 4.1 and on work related to cyclic linearization, Section 4. 4.2 shows that if Multiple Spell Out (Uriagereka 1999) is added to whereby a structure is transferred to the phonological component multiple times throughout the derivation rather than once at the end of the derivation, Copy Control may receive an analysis similar to the analysis offered for the other instances of multi ple copy spell out. 4.4.1 Multiple Copy Spell Out Multiple copy spell out is attested in several languages. For example, Afrikaans allows multiple copies of a wh chain to be pronounced in a question (e.g., (17) ) (du Plessis 1977). Vata allows multiple c opi es of a verb chain to be pron ounced (e.g., (18) ) (Koopman 1984). Kandybowicz (2006) analyzes pronominal resumption, verbal re petition, and predicate clefting in Nupe as multiple copy s pell out Sentence (19) is an example of pronominal resumption (17) Afrikaa ns Met wie het jy nou weer ges met wie With who did you now again said with who het Sarie gedog met wie gaan Jan trou? did Sarie thought with who go Jan marry? (18) Vata Li li d a zu sak. eat 1 P eat PAST yesterday rice

PAGE 161

161 (19) Nupe Gana Musa kpe ganan uu gi bise o Gana Musa know COMP 3 S eat hen o Before trying to propose an explanation for the Telugu and Assamese Copy Control phenomenon, I survey the literature for possible accounts. 4.4.1.1 Nunes analysis The multiple copies realized in each of (17 19) above are non distinct copies in a c command relationship. Ba chain and, accordingly, they must undergo Chain Reduction. But they do not. How do two copies of the same token escape Chain Reduction? According to Nunes (2004: 40) this is possible only if one of the copies h ides inside another word thus, becoming invisible to the Linear Correspondence Axiom More specifically, if a copy in a given chain adjoins to another head, both the copy and or a sing I n the theory of Distributed Morphology ( Halle and Marantz 1993) this process is called Fusion T he Linear Correspondence Axiom cannot see into fused links and, consequently, the lower copy escapes deletion. To illustrate, the low er copy of li da heads are fused into a single terminal and are reanalyzed as a new phonological word. In this way, the Linear Correspondencce Axiom does not detect the two occurrences of li opies of the same token. This means that the structure can be linearized without the deletion of either copy. Consequently, neither copy undergoes Chain Reduction, given that Chain Reduction is a costly operation that applies minimally for the purpose of l inearization and convergence, as (20) explicitly states.

PAGE 162

162 (20) Delete the minimal number of constituents of a nontrivial chain CH that suffices for CH to be mapped into a linear order in accordance with the LCA. (Nunes 2004: 101 (31)) [my emphasis] Not all in stances of multiple copy spell out are analyzed as involving an invisible fused copy or a new phonological word. One such case is copy raising in English. The following copy raising as an instance of multiple copy spell out. 4.4.1.2 Fujii analysis English allows structures like (21) which involve a raising verb (e.g., seem ) and two subjects that are obligatorily coreferential and pronounced. These are known in the literature as copy raising constructions (Rogers 1971, Postal 1974, among others). (21) a. Tom i seems as if he i/*k is seeing someone. b. *Tom seems as if he is seeing someone. Fujii (2005) analyzes such instances of copy raising as m ovement. Details aside, Fujii presents empirical evidence to show that the sub ordinate clause in (21a) is a CP and that the structure invloves movement. 7 As he points out, however, this type of movement seems to be different from, say, wh movement as exemplified in (22). In both cases, an argument moves out of a lower CP into a hig her clause. The obvious difference between the two structures is that two copies of the moving element are pronounced in (21a) while only one is pronounced in (22). (22) What did you say that Tom bought what ? In order to account for this difference, Fujii hol ds that the movement of the subject in (21a) does not land in Spec,CP on its way to the matrix clause, while the movement of the wh element in (22) does. This idea is schematized in (23a b). Evidence for (23a) comes from structures like 7 The argument excludes structures like (i) which allow disjoint subjects. Following Potsdam and Runner (2001), Fujii considers the matrix subject in such constr uctions to be a thematic argument of the matrix verb. Accordingly, the structure does not involve raising. (i) Tom looks like someone has punched him in the face.

PAGE 163

163 (24). The argument is that if Tom landed in Spec,CP on its way to Spec,IP of the matrix clause, wh movement out of the subordinate CP would not be possible (Fujii 2005: 10 12). (23) a. [ CP Tom seems [ CP [ C' as if [ IP Tom is seeing someone]]]]] b. [ CP What did [ IP yo u say [ CP what [ C' that [ IP Tom bought what]]]]] (24) a. Who does Tom seem like he has met? b. [ CP WHO does [ IP Tom seem [ CP WHO [ C' like [ IP Tom/he has met WHO]]]]] How does landing vs. not landing in Spec,CP affect the phonological realizations of copies at PF? The answer resides in the formulation of Chain Reduction. In Nunes 2004, Chain Reduction applies at the end of the deriva tion Fujii resorts to Cyclic Chain Reduction, whereby Chain Reduction applies cycle by cycle or, more specifically, pha se by phase. Following Chomsky (2000, 2001), Fujii assumes that a phase is a vP or a CP. His formulation of Cyclic in (26) which states that when a phase is spel led out, its edge can still take part in further syntactic operations. Fujii formulates (27) as a consequence of (26) (25) Delete all copies of chain CH [within a phase] but the highest one. (Fujii 2005: 1 2; (1)) (26) Phase Impenetrability Condition At the phase ZP containing phase HP, the domain of H is not accessible to operations, but only the edge of HP. (Chomsky 2001:11, 2004: 108) (27) The Role of Phase Edge in Chain Reduction: The highest copy in phase PH that has not undergone deletion can be deleted later onl y if it is in the edge of PH. (Fujii 2005: 3; (7)) Let us have a closer look at (23 24), repeated with more details as (28 a b ), in the light of Cyclic Chain Reduction. In both (28a) and (28b), the subordinate CP forms first. In (28b), but

PAGE 164

164 not in (28a), th e moving element uses Spec,CP as an escape hatch. By the time the following phase head is introduced, the lower phase is spelled out and Cyclic Chain Reduction (CCR1) applies, deleting the lower copy of Tom in (28a) and of what in (28b). At this point, all of lower CP minus its edge is spelled out, as the boldface part of each sentence shows. When the following phase (the higher CP) is spelled out, Cyclic Chain Reduction (CCR2) applies. What in Spec,CP of (28b), being at the edge of the lower phase, undergo es deletion. The subordinate copy of Tom in (28a), on the other hand, survives deletion in accor dance with (26) and (27), or, operation [CCR2] since [it is] not c (28) a. [ CP [ IP Tom seems [ CP [ C' as if [ IP Tom is [ vP Tom [ VP seeing someone]]]]]]] b. [ CP What did [ IP you say [ CP what [ C' that [ IP Tom [ vP Tom [ VP bought what ]]]]]] ant, but it is not without problems. First, the structure in (28a) violates the Linear Correspondence Axiom. The two copies of the subject are non distinct and in a precedence relationship. 8 If both are pronounced, this means that the same element c comman ds itself, which is a violation of linearization. with respect to the operation Form Chain. Let us assume that Form Chain may apply across phases. This means that the two copies of the subject in copy raising constructions do form a chain Therefore, Chain Reduction must apply. When Chain Reduction (CCR2) applies in the higher CP, it cannot mark the lower copy of the subject for deletion because it is in a differen t 8 Fujii (2005: 22 26) offers an explanation concerning why the lower copy is a pronominal ra ther than an exact copy of the matrix subject. Section 4.6 provides a slightly different view. CCR1 CCR2 CCR1 CCR2

PAGE 165

165 phase; but it should be able to mark the higher copy since it is within its jurisdiction, so to speak. Yet, we never see structures like (29) where the higher subject is deleted. (Notice that deletion of copies at PF takes place regardless of Case and vi sibility.) (29) Tom seems as if Tom is seeing someone. Fujii tacitly avoids the problem exemplified in (29) by limiting deletion to the lower copies only as (25) above explicitly states Evidence from Backward Control not only in Telugu, but also in Japan ese (Kuroda 1965, 1978), Tsez (Polinsky and Potsdam 2002), Malagasy (Polinsky and Potsdam 2003), and Korean (Monahan 2003) shows that the higher copy is also susceptible to deletion, which makes the formulation in (25) a stipulation. Note, however, that according to the Phase Impenetrability Condition in (26), the domain of the phase is not accessible to any operation, including Form Chain. This means that Form Chain cannot operate across phases. Instead of solving the problem in (29), however, the failu re of Form Chain to apply creates an even more complicated problem. The reason is that at PF linearization is able to detect that the two copies of Tom are non distinct and in a precedence relation. Yet, the failure of Form Chain to apply means that no Cha in Reduction is possible. The result is again a violation of the Linear Correspondence Axiom. This said, it is imp o r following section suggests that they can be avioded by adopting Uriagereka Spell Out approach whereby every time a part of the structure is spelled out, it is converted into a phonological word that linearization cannot see into. 4.4.2 Multiple Copy Spell Out and Multiple Spell Out Impenetrability Condition, as stated in (26) above, follows from the assumption that a structure is transferred to the phonological component or spelled out phase by phase, whereby a phase is a vP or a CP. This means that a structure undergoes

PAGE 166

166 spell out several times throughout the derivation. Every time a phase is spelled out, which takes place when a new phase head is introduced, its complement is no longer transparent to further syntactic operations. To elaborate, when a CP phase is spelled out, IP complement of C 0 (but not the edge of CP: Spec,CP and C 0 ) becomes opaque to all syntactic operations. Empirical support of this approach comes from and Frank 2001 and Fox and Pesetsky 2005 among several others. and Frank present evidenc e from Bulgarian clitic ordering to argue in favor of multiple spell out by phase. Fox and Pesetsky build a theory of cylic linearization and order perservation based on phases, and they use it to account for the constraints on Object S hift ( eralization ) and Quantifier Movement (Inverse Holmberg effect) in Scandinavian 9 10 Uriagereka (1999) also argues that Multiple Spell Out is part of the computational system. He holds that spell out applies, not only at the end of the derivation, but multi ple times throughout the derivation. 11 According to Uriagereka (1999: 256), every time a domain is 9 phase (or spell out domain) is determined at the end of the phase by the operation Spell Out, and that this order cannot be altered or contradicted later in the derivation. This approach works for the Telugu and Assamese structures under investigation. Nevertheless, the details are orthogonal to the argument in the rest of this section, which is why I do not present them here. What is important for the purpose of this study is that Multiple Spell Out is needed on independent grounds and is not an ad hoc stipulation that is used to account for the phe nomenon of Copy (i) The authors argue that Move is actually Re merge rather than Copy plus Merge (Fox and Pesetsky 2005: 41); evidence from Copy Control, as well as other instan ces of Multiple Copy Spell Out, shows that Copy plus Merge is superior to Re merge. (ii) The function of Spell within each spell out domain is preserved throughout the derivation. While this idea works well for Telugu and Assamese Copy Control, an additional function of Spell Out is needed: turning a spell out domain into a lexical compound or a giant word, as Uriagereka (1999) and the rest of this section argues. If (i) and (ii) with the present analysis without becoming incompatible with the Scandinavian data that the authors examine. 10 ic linearization and order perservation to scrambling in Korean. 11 According to Uriagereka (1999), spell out applies to specific syntactic objects he calls Command Units A Command Unit is a syntactic object that is n of Merge hat is, through the

PAGE 167

167 spelled out, it is converted into a non phrasal structure or a giant lexical compound that is interpretable, yet inaccessabile to further syntactic operations Spell out transfers a phas e to the phonological component, and l inearization takes place in the phonological component. This means that every time a phase is spelled out, it is also linearized. Subsequently, the spelled out phase is converted into a gia nt word that is transparent to interpretation but opaque to all syntactic operation. The following sections examine the influence of Multiple Spell Out on the analysis of copy raising and Copy Control. Section 4. 4.2.1 shows that Multiple Spell Out is supe rior to Cyclic Chain Reduction. Section 4. 4.2.2 extends the analysis to Copy Control. 4.4.2.1 Multiple Spell Out and copy raising As I mentioned in Section 4. copy raising structures like (30 ) faces a problem. T he two pronounced subjects in (30 b), [ NP Tom c ] and [ NP Tom b ], are in a precedence relationship and, consequently, one of them has to be deleted in order for the structure to linearize. Yet, this does not happen due to the cyclicity of Form Chain and Chain Reduct ion. The result is a violation of the Linear Correspondence Axiom. (30) a. Tom seems as if he is seeing someone. b. [ CP [ IP Tom c seems [ C P [ C' a s if [ IP Tom b is[ vP Tom a [ VP seeing someone]]]]]]] Let us observe the derivational history of ( 30 ) in the light of Multiple Spell Out as described in the previous section. Subordinate CP forms in ( 31 a). [ NP Tom a ] in Spec,vP extension of the same syntactic object via the Merge of a new element For example, [x[y]] is a Command Unit. By (i.e., the Merge of two already formed Command Units) does not result with a Command Unit. For example, merging [x [y]] and [a[b]] results with [ [x [y]] [a[b]] ], which is not a Command Unit. Following Chomsky (2000, 2001, 2004), however, I assume that multiple spell out is phase based rather than Command Unit bas ed, and that a phase maybe vP or CP, but not IP. The main reason behind this choice is that phase theory, at least as I understand it, offers more precise specifications concerning the edge of a spelled out domain. For the porpose of my analy sis, it is important that the edge of a spelled out domain remains active. Uriagereka, on the other hand, seems to imply that the whole Command Unit is syntactically inactive once spelled out. CCR1 CCR2

PAGE 168

168 copy plus merges as [ NP Tom b ] in Spec,IP. [ NP Tom b ] and [ NP Tom a ] are non distinct and in a c command relationship; thus, they form a chain The m atrix IP is introduced in ( 31b). [ NP Tom b ] copy plus merges as [ NP Tom c ] in Spec,IP of the matrix clause, crossing over Spec,CP of the subordinate clause. The m atrix CP is introduced in ( 31 c). This is when subordiante CP is spelled out Chain Reduction applies to the chain {[ NP Tom b ], [ NP Tom a ]} and the lower copy is deleted. The CP phase is linearized and converted into a phonological word, as the grey font signifies. As Uriagereka (1999: 256) puts it, the spelled essence, [it] is like a giant lexical compound, whose syntactic terms are obviously interpretable (31) a. [ CP [ C' as if [ IP Tom b is [ vP Tom a [ VP seeing someone]]]]] b. [ IP Tom c seems [ CP [ C' as if [ IP Tom b is [ vP Tom a [ VP seeing someone]]]]] c. [ CP [ IP Tom c seems [ CP [ C' as if [ IP Tom b is[ vP Tom a [ VP seeing someone]]] ]]]] How does this approach solve the violation of the Linear Correspondence Axiom induced nverting the subordinate CP to a lexical compound, not only is the subordinate subject inaccessible to any syntactic operation, including Form Chain and Chain Reduction, but it is also invisible to linearization. Remember from Section 4. 3.1.1 that in Nunes with another head. The situation here is similar. The lower copy hides within the big phonological word that is produced by the operation spell out. Linearization can not see into fused elements. Therefore, any word that hides within a spelled out domain is not a problem for linearization or the Linear Correspondence Axiom. An almost identical conclusion has been arrived at by Nunes and Uriagereka (2000). One major poin t makes the approach to Multiple Spell Out as adopted here different. I consider a CCR1

PAGE 169

169 spelled out domain categorically inaccessible to any syntactic operations. A spelled out domain computational system, despite the fact that its constituent parts are, in a sense, gone; thus, for instance, [a spelled command is obtained by the composition of the elementary relations of sisterhood and containment, as Nunes and Uriagereka adopt this mysterious operation of Form Chain in order to account for (32 ) below. If Form Chain operates into a spelled out domain, then [which paper d ] may form a chain with [which paper b ] inside the spelled out PP [PP after [CP [which paper b ] reading [which paper a ] ]]]]. If Form Chain cannot o perate across phases, [which paper d ] and [which paper b ] cannot form a chain and thus Chain Reduction fails to apply. In this case, the system fails to account for the deletion of [which paper b ]. (32) [ CP [which paper d ] did [ IP John [ vP [ vP file [ which paper c ]] [ PP after [ CP [ which paper b ] reading [ which paper a ] ]] ]] F ortunately, there is way to account for (32 ) while still considering spelled out domains inaccessible to Form Chain and Chain Red uction. If the PP adjunct in (32 ) is considered a CP, as Grohm ann (2003) suggests, then [which paper b ] will occupy the edge of the phase, as (3 3 ) illustrates. In this way, when the adjunct is transferred to the phonological component, [which paper b ], being at the edge of CP, will still be accessible to further synt actic operations, including Form Chain. When the following phase is spelled out, [which paper b ] is already part of a chain {[which paper d ], [which paper b ] } and, consequently, undergoes Chain Reduction for the purpose of linearization. 12 12 Another solution is to consider PP a phase; in this case, [wh ich paper b ] moves to the edge of PP.

PAGE 170

170 (33) [ CP [which paper d ] did [ IP John [ vP [ vP file [ which paper c ]] [ CP [ which paper b ] after reading [ which paper a ] ]]]] 4. 4.2.2 Multiple Spell Out and Copy Control Let us have another look at the derivation of the Telugu example in (12) above, repeated here as ( 34 ). The ste ps of the derivation are delineated in ( 35 ). 13 The CNP clause and the m atrix clause form independently in (35a), before t he CNP subject copy plus merges with matrix vP in (35b). Following, t he matrix subject moves from Spec,vP to Spec,IP to check the EPP fe ature as sketched in (35c) The two non distinct copies of Kumar enter a c command relationship and form a chain {[ NP Kumar d ], [ NP Kumar c ]}. In (35d), m atrix CP is spelled out and linearized. Chain Reduction applies and marks the lower copy of Kumar, [ NP Kumar c ], for deletion. The spelled out domain is converted into a phonological word that is opaque to further syntactic operations, as symbolized by the grey font. Finally, t he CNP clause merges with the matrix clause at CP. Although Matrix CP is spelled o ut, its edge is still accessible to such an operation. The two non distinct CNP copies of Kumar enter a c commanding relationship and form a chain {[ NP Kumar b ], [ NP Kumar a ]}.The whole structure is spelled out and linearized again. Chain Reduction applies a nd marks the lower CNP copy of Kumar, [ NP Kumar a ], for deletion. The structure converges as (35e) (34) Telugu [Kumar sinima cuus tuu] Kumar popkorn tinnaa Du [Kumar.NOM movie watch CNP] Kumar.NOM popcorn ate AGR (35) a. i. [ CNP [ NP Kumar b ] [ vP [ NP Kumar a ] sinima cuus tuu]] ii. [ Matrix vP popkorn tinnaa Du] 13 The discussion in this section depends on the standard assumption that CNP clauses are IPs, but crucially not CPs (Jayaseelan 2004). In other words, they are not phases. Further, for the purpose of the presentation h ere, I only focus on CP as a spell out domain. Commonly, vP is also considered a phase. However, the study is concerned with subject NPs; these occupy the edge of vP and, thus, are still accessible even after vP is spelled out.

PAGE 171

171 b. i. [ NP Kumar] ii. [ Matrix vP [ NP Kumar c ] popkorn tinnaa Du] c. [ CP [ Matrix IP [ NP Kumar d ] [ Matrix vP [ NP Kumar c ] popkorn tinnaa Du] d. [ CP [ Matrix IP [ NP Kumar d ] [ Matrix vP [ NP Kumar c ] popkorn tinnaa Du] e [ CP [ CNP [ NP Kumar b ] [ vP [ NP Kumar a ] sinima cuus tuu]] [ CP [ Matrix IP [ NP Kumar d ] [ Matrix vP [ NP Kumar c ] popkorn tinnaa Du]] ]] T he same steps apply to Assamese. Consider s ente nce ( 36 ) The tree in (3 7 ) is a summary of its derivational history. As the dotted arrows show, the two matrix copies of Ram enter a c command relationship and form a chain {[ NP Ram d ], [ NP Ram c ]}. The matrix clause is s pelled out and li nearized and the lower copy is marked for deletion. The complement of matrix CP becomes a non phrasal lexical compound, as indicated by the grey box. Following, the CNP clause m erges at matrix CP. The two CNP copies of Ram enter a c command relationship a nd form a chain {[ NP Ram b ], [ NP Ram a ]} and the lower copy is deleted. The structure converges with two copies of Ram phonologically realized. (36) Assamese [Ram Or bhok lag i] Ram e bhat khal e [Ram GEN hunger feel CNP] Ram NOM rice ate AGR (37) CP qp CNP P CP qu rp SUBJ I' IP C Ram Or b 3 3 vP I SUBJ I' 3 Ram e d 3 SUBJ 6 vP I Ram Or a bhok lag i 3 SUBJ 6 Ram e c bhat khal e

PAGE 172

172 T he derivations in (3 5) and (37 ) do not violate the Linear Corr espondence Axiom. Linearization is not able to detect [ NP Kumar d ]/ [ NP Ram d ] of the matrix clause as a token of the same element as [ NP Kumar b ]/ [ NP Ram b ]. The reason is that the matrix copy is hidden inside a spelled out domain that behaves like a phonolo gical word, and linearization cannot see into phonological words. Consequently, precedence in the sense of Kayne 1994 is not detected and no violation is induced. has to do with the timing of the spell out of matrix CP in (35) and (37). As Section 4. 3 points out, the two pronounced copies in Copy Control structures do not form a chain for an independent reason: they do not enter a c command relationship. In Nunes syste m, Form Chain is obligatory in order to serve one purpose: linearization. The lack of Form Chain is a violation if linearization and the Linear Correspondence Axion are not satisfied. Stated differently, if linearization detects two non distinct copies, on e of them has to be deleted. In order for deletion or Chain Reduction to apply, the two non distinct copies have to form a c hain. If the two copies are no longer non distinct (because one of the copies is in a fused word or in a spelled out domain), Fo rm Chain is no longer an essential, derivation saving operation. Therefore, the fact that the two pronounced subjects in Copy Control constructions are not in a c command relationship and do not form a chain is no longer an issue. This might sound li ke an ad hoc stipulation that only serves the analysis of Copy Control structures. Nevertheless, we have seen that the two pronounced copies in copy raising constructions like (38) do enter a c command relationship and, ideally, are able to form a chain I f this happened, the only copy that would be able to escape deletion is the subordinate copy. The reason is that it is inside a phase. The matrix copy, on the hand, would be marked for deletion

PAGE 173

173 before the whole structure is spelled out. This does not happe n, however, which indicates that Form Chain must fail to apply. In this sense, the lack of Form Chain is not only tolerated but also required. (38) [ CP [ IP Tom seems [ CP [ C' as if [ IP Tom is seeing someone] ]]]] ultiple copy spell out that are the result of movement minus Chain Reduction. Such cases involve duplication of focalized elements in Brazilian Sign Language and clitic duplication in some dial ects of Arge n tinean Spanish ( Nunes 2004: 38 61). One might wond er, however, why his work does not involve similar cases that are the result of movement minus Form Chain. The reason is that spell place one time at the end of the derivation. This means that linearization happens only once aft er the syntax; that is, after all movement, feature checking, and crucially Form Chain take place. Multiple Spell Out, however, dictates that parts of the derivation (arguably, phases) be spelled out before others. This means that linearization happe ns several times. This also means that a structure might involve one or more copies of a certain token which are well beyond the syntax, including Form Chain. At the same time, the structure might involve other copies of the same token that are still in th e syntax and subject to Form Chain. This situation does not exist in movement minus Form Chain. ory formulates movement as comprising four independent steps: Cop y, Merge Form Chain and Chain Reduction. And just as Chain Reduction does not apply under certain circumstances, Form Chain may also do the same.

PAGE 174

174 As a matter of fact, the theory is so power ful that it allows Merge also not to apply. In other words, movement may, under certain circumstances, comprise only Copy. To illustrate, Nunes holds that instances of Across The Board movement take place if the numeration is exhausted but one more instanc e of Merge is needed in order for a structure to converge. One may argue that a token that is already in the structure does Copy, but it only merges if (i) it needs to check a feature of its own or (ii) the numeration is exhausted and it is the only elemen t that can check a feature on a target. It is hard to think of empirical evidence to test this claim, but it is a fair theoretical prediction. This leaves us with the spell out timing of CP in (35) and (37). The standard approach is that a phase is spelled out wh en another phase head is introduced or at the end of the derivation. The spell out of CP in Copy Control seems to fall in neither category. When CP is spelled out, neither a new phase head is introduced, nor is it the end of the derivation (an adjun ct still awaits merge). Clo ser examination shows that m atrix CP is actually spelled out at the end of the derivation. This observation follows from the properties of adjuncts. As Chom sky (2004: 117) points out, adjuncts are not selected by the head of the structur e they adjoin to, and other words, when the matrix CP is complete, the computation processes the structure as if it is the end of the derivation, and CP is spelled out. What may be considered new here is that the edge of matrix CP is still accessible to further computation (namely, the merge of the CNP clause) despite being spelled out in an end of the derivation fashion. This is not a totally b izarre idea. If we consider the edge of CP as responsible for linking CP to other structures in discourse (Rizzi 1997, Chom sk y 2004), then it is fair to assume that this edge is still active even after the final spell out.

PAGE 175

175 I mentioned at the beginning of S ection 4. 4 that there is a way for the theory to account for the section showed t hat Copy Control, as well as copy raising is derivationally only slightly different from other instances of multiple copy spell out. According to Nunes, occurrences of multiple copy spell out involve two non distinct copies, one of which has become distin ct due to fusion. At PF, linearization cannot see into the fused element. Accordingly, no precedence relationship is detected, and no deletion/Chain Reduction takes place. The same mechanism applies in the case of Copy Control. The steps are summarized as follows: Step 1: Two subject NPs are non distinct copies of the same token. Step 2: Due to Multiple Spell Out, one copy hides within a spelled out domain and becomes part of a giant phonological word. This outcome is, in essence, identical to fusion. Step 3: As a result, two copies escape deletion without inducing a violation of the Linear Correspondence Axiom simply because no precedence is detected. The only difference between Nunes multiple copy spell out and Copy Control is that the former involves m ovement minus Chain Reduction, while the latter involves movement minus Form Chain and Chain Reduction. Both types of movement are allowed only if no violation of the Linear Correspondence Axiom is induced. Forward/Backward Control is unaffected by this a ddition to the theory. Consider the Forward Structu re in (39) and its derivation in (40). The CNP clause and the matrix clause form ind ependently, and t copies out of the CNP clause (40a ). Kumar merges in Spec,vP of the matrix clause (40b) Following, the CNP clause adjoins to matrix vP and becomes an island (40c). the EPP feature and C 0 projects (40d) At this point, assuming that vPs are phases, matrix vP is spelled out, as indicated by the grey font. T he domain/complement of v 0 is linearized and

PAGE 176

176 COPY converted into a phonological word, but the edge of vP (including v 0 spec,vP, and the CNP clause/adjunct of vP) is still accessible to further computation. ma c commands the copy in the CNP clause and the copy in Spec,vP, forming a chain with each as the dotted arrows in (40d) indicate Step ( 40f ) takes place at PF. T he lower copy in each chain is deleted in accordance with Chain Reduction. Finall y, the whole structure is spelled out. (39) Telugu [ CP [ IP Kumar [ vP [ CNP Kumar ki jwaram wacc i] [ vP Kumar hospital weLLaa Du]]]] [ CP [ IP Kumar.NOM [ vP [ CNP Kumar DAT fever come CNP ] [ vP hospital went 3.M.S]]]]] (40) a. i. [ CNP [ NP Kumar ki] jwaram wacc i] [ NP Kumar] [ CNP [ NP Kumar DAT] fever come CNP] ii. [ Matrix vP hospital weLLaa Du] [ Matrix vP hospital went 3.M.S] b. [ Matrix vP [ NP Kumar] hospital weLLaa Du] c. [ Matr ix IP [ vP [ CNP [ NP Kumar ki] jwaram wacc i] [ vP [ NP Kumar] hospital weLLaa Du]]] d. [ CP [ Matrix IP [ NP Kumar] [ vP [ CNP [ NP Kumar ki] jwaram wacc i] [ vP [ NP Kumar] hospital weLLaa Du ]]] e. CP 3 IP C qp SUBJ I' Kumar qp vP I qp CN P P vP 3 3 SUBJ 6 SUBJ 6 Kumar ki aakali weesi Kumar popkorn tinnaa Du f. At PF: [ Matrix IP [ NP Kumar] [ vP [ CNP [ NP Kumar ki ] jwaram wacc i] [ vP [ NP Kumar ] hospital weLLaa Du]]]

PAGE 177

177 N otice that matri x CP is spelled out after the CNP clause has adjoined to matrix vP. In other words, the CNP clause is part of the spelled out domain and is, accordingly, part of the linearized structure. This explains why the CNP subject (or the matrix subject in Telugu) suffers deletion. The conclusi on can be summarized as follows. If the CNP clause merges at matrix vP, it is spelled out and linearized with matrix CP. Accordingly, Forward/Backward Co ntrol obtains. If the CNP clause merges at matrix CP instead, matrix CP i s spelled out and linearized independently. As a result, Copy Control obtains. One problem remains. B y merging at CP, the CNP clause acquires a prolonged lifespan as a non island, which may result with the overgeneration of ungrammatical structures. The f ollowing section spells out the details and suggests a solution. 4.5 Adjunction to CP and Unwanted Instances of Sideward Movement The CNP clause becomes an island only when it adjoins to the matrix clause. Prior to that, sideward movement out of the CNP cl ause is allowed. In Forward/Backward Control structures, the CNP clause adj oins to the matrix clause at vP right after the CNP subject undergoes sideward and merges i n Spec,vP of the matrix clause ( 42a) before the CNP clause adjoins to matrix vP ( 42b). After that, the CNP clause becomes an adjunct out of which movement is disallowed. (41) Telugu Kumar [ Kumar ki aakali wees i] popkorn tinnaa Du Kumar.NOM [ Kumar DAT hunge r fall CNP ] popcorn ate 3.M.S (42) a. CNP P 3 SUBJ 6 Kumar ki aakali wees i v P 3 SUBJ 6 Kumar popkorn tinnaa Du S IDEWARD M OVEMENT

PAGE 178

178 b. CP 3 IP C qp SUBJ I' Kumar qp vP I qp CNP P vP 3 3 SUBJ 6 SUBJ 6 Kumar ki aakali weesi Kumar popkorn tinnaa Du In Copy Control structures, on the other hand, the CNP clause adjoins to the matrix clause at CP. This means that the CNP clause is av a i lable for further instances of sideward movement that may overgenerate ungrammatical structures. Take the ano malous derivation in (43), for example. The derivation starts with the numeration in (43a). The CNP clause and matrix vP form independently in (43b). Notice that Spec,vP of the matrix clause is already filled and that the thematic requirements of matrix vP are satisfied. In other words, no sideward movement of Kumar is necessary. When matrix I 0 projects in (43c), either Sarita in Spec,vP or Kumar in the CNP clause must be able to occupy Spec,IP, especially that neither copy is closer to Spec,IP than the oth er. The CNP clause is not an island yet, so sideward movement may happen. Let us assume that it does, as demonstrated in (43d e). Following, the CNP clause adjoins to the matrix clause at CP ( 43f). Ideally, th e structure should not converge as (44 ). But it does (43) a. {Kumar 1 Sarita 1 sinima 1 cuus 1 tuu 1 popkorn 1 tinn 1 Tense 1 A gr 1 } {Kumar, Sarita, movie, watch, CNP, popcorn, eat, Tense, Agr} b. i. [ CNP [ NP Kumar] [ vP [ NP Kumar] sinima cuus tuu]] [ CNP [ NP Kumar.NOM] [ vP [ NP Kumar.N OM] movie watch CNP]] ii. [ Matrix vP [ NP Sarita] popkorn tinnaa di] [ Matrix vP [ NP Sarita.NOM] popcorn ate 3.N.S] c. [ Matrix IP [ Matrix vP [ NP Sarita] popkorn tinnaa di]

PAGE 179

179 d. [ NP Kumar] e. [ CP [ Matrix IP [ NP Kumar] [ Matrix vP [ NP Sarita] popkorn tinnaa di] f. [ CP [ CNP [ NP Kumar] [ vP [ NP Sarita ] sinima cuus tuu]] [ CP [ Matrix IP [ NP Kumar] [ Matrix vP [ NP Kumar] popkorn tinnaa di]]]] (44) Telugu *[Kumar sinima cuus tuu] [Kumar.NOM movie watch CNP] Kumar Sarita popkorn tinnaa di Kuma r.NOM Sarita.NOM popcorn ate 3.N.S The easy and quick answer concerning why (43) is not a possible derivation is that Kumar being an argument, has to merge in a thematic domain before it mo ves to the higher functional layers. Unfortunately, this answer does not do the trick because Kumar has already merged in a thematic domain and taken on a theta role by the time sideward movement takes place. This scenario is meant to point out that, unde r the current assumptions, sideward movement becomes so permissive that it overgenerates. In the old days, an argument merges in a thematic position only if it has not taken on a theta role yet. By the same token, it is allowed to merge in a Case position only if it has not checked Case yet. Over the last decade or so, research has shown that an NP may take on more than one theta role and more than one Case, two assumptions that are argued fo r on independent grounds ( 1994 and Bejar and Massam 1999, among others). In this sense, movement allows an NP to copy plus merge just anywhere regardless of its Case and thematic characteristics. This is true of sideward as well as intra clausal movement. Intra clausal movement seems to be less problematic, howe ver. Such movement takes place within a single derivational workspace and always targets a higher c commanding position.

PAGE 180

180 Under the current assumptions about multiple theta role and multiple Case checking, the landing site cannot be enforced by the feature characteristics of the moving element. Rather, it is enforced by the Extension Condition (Chomsky 1995: 248) which holds that merge extends the structure by applying at the root. Unlike intra clausal movement, sideward movement involves movement from one d erivational workspace to another. Such movement does not involve a c command relationship between the launching site and the landing site (Hornstein and Kiguchi 2001:11). This means that neither the feature characteristics of the moving element nor the Ext ension Condition may restrict this kind of movement or dictate its landing site. The former is not restrictive under the view that an NP may bear multiple theta roles and multiple Case. The latter is obeyed without resorting to c command; in other words, a sideward moving element targets the root of a particular structure, and thus extend s the structure, without actually moving to a higher c commanding position. As a solution I suggest that sideward moving elements undergo merge in the same f ashion they undergo their first merge. This idea is formulated in (45). (45) undergoes sideward movement. A domain X can be a thematic domain, a phi domain, or a discourse do main. The restriction in (45) dictates that a sideward moving element behave like an element selected from the numeration. For one thing, they both obey the Extension Condition and extend the structure without resorting to c command. In other words, c com mand relations are orthogonal to both. In (45), I suggest that sideward moving elements and elements selected from the numeration also behave in the same way with respect to the locus of merge. 14 14 This restriction in (45) h Locality of Movement. Grohmann (2003: 309 314) provides a similar, though not identical, formulation.

PAGE 181

181 If (45) is on the right track, then the adjunction of the CNP clause to matrix CP no longer overgenerates structures like (44). To illustrate, consider the derivation of (44) revisited in (46). The numeration in (46a) is selected from the lexicon. The CNP clause and the matrix clause form independently in (46b). As (46b ii) shows, matrix vP already has all its theta roles satisfied, leaving no room for another argument to merge. By the time Kumar Copies out of the CNP clause ( 46c), the only landing site available for merge is Spec,IP ( 46d). According to (45), Kumar c an only merge in a thematic position. The reason is that its first merge in the CNP clause also targets a thematic position. Merging in Spec,IP is a violation of (45), which is why the derivation crashes. (46) a. {Kumar 1 Sarita 1 sinima 1 cuus 1 tuu 1 popkorn 1 tinn 1 Tense 1 A gr 1 } {Kumar, Sarita, movie, watch, CNP, popcorn, eat, Tense, Agr} b. i. [ CNP [ NP Kumar] [ vP [ NP Kumar] sinima cuus tuu]] [ CNP [ NP Kumar.NOM] [ vP [ NP Kumar.NOM] movie watch CNP]] ii. [ Matrix vP [ NP Sarita] popko rn tinnaa di] [ Matrix vP [ NP Sarita.NOM] popcorn ate 3.N.S] c. [ Matrix IP [ Matrix vP [ NP Sarita] popkorn tinnaa di] d. [ NP Kumar] Note that the sideward movement of Kumar in (46c) is not optional. Otherwise, based on the numeration in (46a), (47) b elow with disjoint subjects should be a possible outcome, but it is not. I address this issue of obligatory movement in details in Chapter 5. (47) Telugu *[Kumar sinima cuus tuu] [Kumar.NOM movie watch CNP] Sarita popkorn tinnaa Du Sarita.NOM popcorn a te 3.M.S

PAGE 182

182 Another point is in order. In principle, it is possible to satisfy the restriction in (45) by having Kumar merge in the object position of the matrix clause since this position also belong s to the thematic domain. Yet, Object Control structures like (48) are ungrammatical. (48) [Kumarki i jwaram wacc i] [Kumar DAT i fever come CNP] Sarita atani ki i mandulu iccin di Sarita him to i medicines gave 3.N.S Here I consider movement to take place as a last resort, derivation saving operation (or set of operations). To elaborate, the last available landing site in the matrix clause that the CNP subject may move to is matrix Spec,vP. Since move ment is not optional the CNP subject moves once matrix v 0 projects because this is the last chance If the CNP subject does not move to matrix Spec,vP, it cannot move at all and the derivation crashes. By the same token, if the CNP subject moves earlier, say to the object p osition, movement is no longer a last resort, derivation saving operation. The following section deals with the nature of the pronounced subjects in Copy Control constructions. It offers an explanation concerning why each of th e subjects takes the form it does, that is, as an R expression or a pronominal. 4.6 Phonological Realization of Copies In Section 4. 2, I ran a diagnosis to show that the phonological realization of t he t wo subjects in Copy Control constructions is the resu lt of movement rather than base generated resumption. The conclusion was based on three mismatches between the characteristics of base generated resumptive elements on the one hand and the behavior and type of overt subjects allowed in Copy Contr ol structures on the other hand. First, b ase generated resumptive pronominals show up in islands (adjuncts, NP clauses) that are immune to movement, and they

PAGE 183

183 show up all the time. Copy Control, on the other hand, is not restricted by the adjunctive nature of the CNP clause. Rather, it is restricted by the position of the CNP clause with respect to the matrix clause that is, whether the CNP clause is sentenc e initial or sentence internal. Second, r esumptive elements are strictly pronominals (pronouns or e pithets). Although the matrix subject in Copy Control structures may be a pronominal, it may also be realized as an R expression that is a non distinct copy of the CNP subject. Resumption R expressions can be straightforwardly accounted for if the y are considered as the outcome of movement, but not as straightforwardly if they are considered to be base generated. Third, r esumptive pronominals only show up in subordinated domains that usually fail to function as launching sites for movement. When a pronominal subject is pronounced in a Copy Control construction, it shows up in the landing site: the matrix clause. This section is mainly concerned with the second and third points. We know for a fact that the matrix subject in Telugu and Assamese Copy Control structures may be realized as a pronoun, an epithet, or an R expression, and that it has to be co referential with an R expression in the subject position of the CNP clause. The section will try to answer two questions. First, i f (Copy) Control inv olves the movement of the CNP subject out of the adjunct into the matrix clause, wh y can the two copies be phonologically distinct? In other words, what allow s the matrix subject to be realized as either a pronoun or an epithet or an R expression? Second, why does the CNP subject have to be an R expression? I begin by addressing the first question. 4. 6.1 Movement and the PF Realization of Copies In this section I will refer to pronouns and epithets that are residues of movement as PF pronominals. Aoun, Cho ueiri, and Hornstein (2001: 372; (3)) analyze PF pronominals (more specifically, strong pronouns and epithets) as appositives ad j oined to R expressions. The R expression moves and the PF pronominal is stranded. This idea is illustrated in ( 49 ).

PAGE 184

184 (49) R expression [ DP R expression [ DP strong pronoun/epithet phrase]] Although slightly different from Aoun, approach leads to the same result. He holds that PF pronominals merge as part of the DP containing the R expression. When movement applies, only the R expression undergoes movement, as (50) indicates. (50) R expression [ DP R expression [ D' [ D PF pronominal ]] Both approaches fail to account for Telugu and Assamese Copy Control. If Nu argument that movement only takes place from a subordinated to a subordinating domain is on the right track ( fn. 4 ), then the subject copies out of the CNP clause and merges into the matrix clause rather than the other way around. In this case, (49) and (50) predict that the R expression in (51) below must copy out of the CNP clause (51a) and merge into the matrix clause (51b). The stranded CNP copy must be a PF pronominal. (51) a. CNP b. Matrix vP 3 3 SUBJ 6 SUBJ 6 3 6 R Expression 6 R Expression PF Pronominal This prediction is not born out, as (52 53) show. I n (52a) and (53a), the subject has the structure proposed in (49 50). The R expression copies out of the CNP clause (52a, 53a) and merges in the matrix clause (52b, 53b). The PF pronominal is stranded in the CNP clause. The outcome is the ungrammatical str uctures in (52c) and (53c). The reason is that the CNP subject has to be an R expression rather than a PF pronominal.

PAGE 185

185 (52) Telugu a. CNP b. Matrix vP 3 3 SUBJ 6 SU BJ 6 3 6 Kumar ki 6 Kumar atani ki/ aa pichooDu ki he DAT/that idiot DAT c. *[ atani ki / aa pichooDu ki koopamu wacc i] [ he DAT / t hat idiot DAT anger come CNP] Kumar akkadi nunci wellipoyinaa Du Kumar.NOM there from left 3.M.S (53) Assamese a. CNP b. Matrix vP 3 3 SUBJ 6 SUBJ 6 3 6 Ram Or 6 Ram e tar/ beseratu r he GEN / the poor guy GEN c. *[ tar / beseratu r khub bhok lag i] [ he.GEN / the poor guy GEN very hunger feel CNP] Ram e posa bhat khal e Ram NOM stale rice ate 3 It can be assumed that linear order dictates that the R expr ession appear first. Since linearly the CNP clause is realized first, the R expression appears in the CNP clause. In this way, instead of moving the R expression to the matrix clause and stranding the PF pronominal, the computational system moves the PF pronominal to the matrix clause and strands the R expression, as illustrated in (54a b) and (55a b). The result is the grammatical structures in (54c) and (55c).

PAGE 186

186 (54) Telugu a. CNP b. Matrix vP 3 3 SUBJ 6 SUBJ 6 3 6 Kumar ki 6 atanu / aa pichooDu atani ki/ aa pichooDu ki he DAT/that idiot DAT c. [ Kumar ki koopamu wacc i] [ Kumar.DAT anger come CNP] atanu / aa pichooDu akkadi nunci wellipoyinaa Du he.NOM / that idiot.NOM there from left 3.M.S (55) Assamese a. CNP b. Matrix vP 3 3 SUBJ 6 SUBJ 6 3 6 Ram Or 6 xi / beseratu e tar/ beseratu r he GEN / the poor guy GEN c. [Ram Or khub bhok lag i] [Ram GEN very hunger feel CNP] xi / beseratu e posa bhat khal e he.NOM / the poor guy NOM stale rice ate 3 Although the outcome in (54 55) is grammatical, the stranding analysis still fails to account for the instances of Copy Control in which both copies are R expressions, as illustrated in (56a) and (57a). In order for (49) and (50) to be able to account for such structures, the CNP subject must start as an R expression whose appositive (or adjunct) is an exact copy, as (56b, 57b) show. Subsequently, one of the copies moves into the matrix clause and the other is stranded (56c, 57c).

PAGE 187

187 (56) Te lugu a. [ Kumar ki koopamu wacc i] [ Kumar DAT anger come CNP] Kumar akkadi nunci wellipoyinaa Du Kumar.NOM there from left 3.M.S b. CNP c. Matrix vP 3 3 SUBJ 6 SUBJ 6 3 6 Kumar ki 6 Kumar Kumar ki (57) Assamese a. [ Ram Or khub bhok lag i] [ Ram GEN very hunger feel CNP] Ram e posa bhat khal e Ram NOM stale rice ate 3 b. CNP c. Matrix vP 3 3 SUBJ 6 SUBJ 6 3 6 Ram Or 6 Ram e Ram Or Appositives as presented in (49 50) may take the form of an epithet (or a st rong pronoun), 15 as (58a) and (59a) illustrate. I do not know of a case where the appositive is an exact copy of the expression it attaches to. At least in Telugu and Assamese, such structures are unacceptable, as (58b) and (59b) indicate. 15 A pronominal appositive is attested in American Sign Language. If a s igner is talking about a person or an space in front of the body. This sets that location as a representation of the original noun. Fro m that point on in the and Wong 2004: 459).

PAGE 188

188 (58) Telugu a. koopamu wacc i Kumar aa pichooDu akkadi nunci wellipoyinaa Du anger come CNP Kumar that idiot.NOM there from left 3.M.S b. *koopamu wacc i Kumar Kumar akkadi nunci wellipoyinaa Du anger come CNP Kumar Kumar.NOM there from left 3.M.S (59) Assamese a. xomoi no thak i Ram gadhatu e bhat o na khal e time NEG keep CNP Ram the donkey NOM rice even NEG ate 3 b. *xomoi no thak i Ram Ram e bhat o na khal e time NEG keep CNP Ram Ram NOM rice even NEG ate 3 A non stranding alternative of Aoun, (2003) approach can help account for the Telugu and Assamese data. Instead of only copying a part of the CNP subject (e.g., the R expression) and stranding the rest (e.g., the PF pronominal), I suggest that the whole CNP subject copy plus merges int o the matrix clause, as (60 61) show. (60) Telugu a. CNP b. Matrix vP 3 3 SUBJ 6 SUBJ 6 3 3 Kumar ki 6 Kumar 6 atani ki/ aa pichooDu ki atanu / aa pichooDu (61) Assamese a. CNP b. Matrix vP 3 3 SUBJ 6 SUBJ 6 3 3 Ram Or 6 Ram e 6 tar/ beseratu r xi/ beseratu e

PAGE 189

189 The outcome of (60 61) is (62 63) below where both the CNP subject and the matrix subject comprise an R expression with an appositive. At PF, the CNP subject is pronounced as an R expression for reasons to be discussed in the following section. Further, the phonological component decides ho w t he matrix subject is pronounced. T hree options are available: an R expre ssion, a pronoun, or an epithet. 16 (62) Telugu [[Kumar ki [atani ki / aa pichooDu ki]] koopamu wacc i] [[Kumar DAT [ he DAT / that idiot DAT]] anger come CNP] [Kumar [atanu / aa pichooDu]] akkadi nunci wellipoyinaa Du [Kumar.NOM [ he.NOM / that idiot.NOM]] there from left 3.M.S (63) Assamese [ [Ram Or [tar / beseratu r]] khub bhok lag i] [[Ram GEN [ he.GEN / the poor guy GEN]] very hunger feel CNP] [Ram e [xi / beseratu e]] posa bhat khal e [Ram NOM [ he.NOM / the poor guy NOM]] stale rice ate 3 T his approach is superior to Aoun, Choueiri, and I t accounts for the Telugu and Assamese Copy Control data. At the same time, it is more in line with the Copy Theory of Movement. No stipulative stranding is involved One question remains: Why does the CNP subject have to be an R expression? The following section presents a possible answer. 4.6.2 Lack of Cataphoricity and the Nature of the CNP Subject Recall from Section 4. 3 that Copy Control structures involve a CNP clause that is adjoined to the matrix clause at CP, resulting in two subjects in a non c command relationship. This idea is illustrated in ( 64 ). Lack of c command implies that whatever enforces an R expression in the 16 A fourth option is a silent copy pro Assamese disfavors this option because structures wit h a matrix pro can also be parsed as Backward Control constructions. Such constructions are marginal in Assamese.

PAGE 190

190 CNP clause and allows a PF pronominal in the matrix clause is unlikely to be Condition C. This is certainly a desired conclusion. Here is why. If we consider ( 65a) and (66 a) ungrammatical due to a Condition C violation, then we m ust find a way to explain why (65b) and (66 b) are grammatical despite a Condition C violation. According to Condit ion C, an R expression is simply free (Chomksy 1986: 164 165). That is, it cannot be bound by any element; this includes a PF pronominal as well as another R expression. (64) CP qp CNP CP 3 3 SUBJ 6 IP C 3 SUBJ 6 (65) Telugu a. *[atanu Sarita too naTyamu cees tuu] [he.NOM Sarita with dance do CNP] Kumar aami ki kata ceppaa Du Kumar.NOM her DAT story told 3.M.S b. [Kumar Sarita too naTyamu cees tuu] [Kumar.NOM Sarita with dance do CNP] Kumar aami ki kata ceppaa Du Kumar.NOM her DAT story told 3.M.S (66) Assamese a. *[tar PrOxad Or oporot kh ong uth i] [he.GEN Proxad GEN on anger come CNP] Ram e tar lagot kaziya koril e Ram NOM him.GEN with fight did 3

PAGE 191

191 b. [Ram Or PrOxad Or oporot khong uth i] [Ram GEN Proxad GEN on anger come CNP] Ram e tar lagot kaziya koril e Ram NOM him.GEN with fight did 3 It is worth noting that the same observation applies to the object NPs in (65b) and (66b). The CNP object is an R exp ression and the matrix object is a pronoun. Obviously the former does not c command the latter. Yet, if the CNP subject is a pronoun and the matrix subject is an R expression, as (67a) and (68a) exemplify, the result is ungrammaticality under the designate d reading. By the same token, both object NPs can be co referential R expressio ns without inducing a violation ( e.g., (67b) and (68b) ) (67) Telugu a. *[aami too i naTyamu cees tuu] [her with dance do CNP] Kumar Sarita ki i kata ceppaa Du Kumar.N Sarita DAT story told 3.M.S b. [Sarita too naTyamu cees tuu] [Sarita with dance do CNP] Kumar Sarita ki kata ceppaa Du Kumar.NOM Sarita DAT story told 3.M.S (68) Assamese a. *[tar i oporot khong uth i] [him.GEN on anger come CNP] Ram e PrOxad Or i lagot kaziya koril e Ram NOM Proxad.GEN with fight did 3 b. [PrOxad Or oporot khong uth i] [Proxad GEN on anger come CNP]

PAGE 192

192 Ram e PrOxad Or lagot kaziya koril e Ram NOM Proxad GEN with fight did 3 The examples in (67) and (68) indicate that the case of the subject NP s in Tel ugu and Assamese Copy Control is not control related. R ather, it follows from a more ubiquitous restriction on cataphoricity. Unlike English and other similar languages, Telugu and Assamese lack cataphoricity. Each of the sentences in (69 72) invol ves an R expression and a pronominal that are not in a c command relationship. In the (a) sentences, the R expression l inearly precedes the pronominal. T he sentences are grammatical. In the (b) sentences, the pronominal lin early precedes the R expression. T he sentences are ungrammatical under the designated reading. Notice that the English equivalents are considered acceptable, at least by some speakers. I suggest that lack of cataphoricity also disallows Copy Control structures with a PF pronominal linearl y preceding a coreferential R expression. 17 (69) Telugu a. pillalu i cus ina aa movie waalla nu i bayapettin di children.NOM see INF that movie them ACC frighten 3.N.S b. *waallu i cus ina aa movie pil lala ni i bayapettin di they.NOM see INF that movie children ACC frighten 3.N.S (70) Telugu a. ninna nenu pillala ni i cusaa nu yesterday I.NOM children ACC saw 1.S appaDu waallu i aDukuntunna ru at the time they.NOM were playing 3.M.P b. *ninna nenu waalla nu i cusaa nu yesterday I.NOM them ACC saw 1.S 17 Napoli (1992) builds on work by Larson (1988) and Jackendoff (1990) to show that c command is not the only way to establish domains (in the sense of binding domains) and that linear precedence plays a role as well.

PAGE 193

193 appaDu pillalu i aDukuntunna ru at the time the children.NOM were playing 3.M.P (71) Assamese a. lora sowali bur e i sow a sini ma khon e boy girl CL(P) NOM see INF movie CL NOM tahat ok i bhoy khwal e them ACC fear made 3 b. *tahat e i sow a sinima khon e they NOM see INF movie CL NOM lora sowali bur ok i bhoy khwal e boy girl CL(P) ACC fear made 3 (72) Assamese a. moy lora sowali bur ok i kali dekhil o I. NOM boy girl CL(P) ACC yesterday saw 1 tahat e i tetiya kheli asil they NOM at the time play were b. *moy tahat ok i kali dekhil o I.NOM them ACC yesterday saw 1 lora sowali bur e i tetiya kheli asil boy girl CL(P) NOM at the time play were On e final point before I conclude. T he CNP subject in Telugu and Assamese Copy Control structures has to be an R expression. It can not be a pronominal, even if the matrix subject is itself a pronominal. This is exemplified in the ungrammaticality of (73 74). (73) Telugu *[atani ki koopamu wacc i] atanu akkadi nunci wellipoyinaa Du [ he DAT anger come CNP] he.NOM there from l eft 3.M.S

PAGE 194

194 (74) Assamese *[ tar khub bhok lag i] xi posa bhat khal e [ he.GEN very hunger feel CNP] he.NOM stale rice ate 3 I take the ungrammaticali ty of (73 74) as a language specific behavior. Sentences (75 76) from Dakkhini and Karnataka Konkani suggest that other languages of South Asia might not have this restriction (Arora and Subbarao 2004: 40: (80 81)). 18 (75) Dakkhini [us ku bukhaar aa ke] uno m ar.gayaa [he DAT fever come CNP] he.NOM died (76) Karnataka Konkani [tak ka taap yewa nu] tO gellO [he DAT fever come CNP] he died Based on the above discussion, we can conclude that the forms of the subjects, R expressions vs. PF pronominals, in Copy Control structures follow from language specific fact ors of cataphoricity and economy rather than from the nature of the derivation or the nature of movement. If this conclus ion is correct, then movement or the operations it comprises becomes a more straightforward operation relieved from the burden of stranding. 4.7 Conclusion This chapter provided an analysis of Copy Control into CNP clauses in Telugu and Assamese. It showed that Copy Control is derivationally similar to Forward/Backw ard Control except for one step. W hereas the CNP clause in Forward/Backward Control merges at vP of the matrix clause, in Copy Control it merges at CP. In order to account for the Telugu a nd Assamese data, the analysis 18 Arora and Subbarao briefly suggest that the CNP pronominal subject in (116 117) is a phonological realization of PRO.

PAGE 195

195 and Uriagereka (1999) among others, I suggested that linearization takes place phase by phase. As a result, the matrix clause, being a phase, is spell ed out, linearized, and transformed into a phonological word prior to the adjunction of the CNP clause. After adjunction, linearization cannot detect the CNP subject and the matrix subject as non distinct copies of the same token. The reason is that the ma trix subje ct is now part of a bigger word, the spelled out domain of the matrix clause. Consequently, two subjects escape deletion, resulting in Copy Control. Finally, the chapter introduces a non stranding alternative to Aoun, Choueiri, and The authors argue that a PF pronominal is a part o f an R S ubsequently, the R expression moves, and the pronominal is stranded. I agree with the idea th at the PF pronominal and the R expression undergo their first merge as one phrasal structure. Nevertheless, the Telugu and Assamese data suggest that, not only the R expression, but the whole phrasal structure copies and merges in the subordinating domain. Decisions conc erning which part to pronounce, the R e xpression or the PF pronominal, are made at PF. Lack of cataphoricity in Telugu and Assamese plus other language specific restrictions enforces the CNP subject to be realized as an R expression wh ile the matrix subject is allowe d to take on one of three forms. These are a pronoun, an epithet, or an R expression.

PAGE 196

1 96 CHAPTER 5 TRIGGER: WHY MOVEMEN T IN CONTROL? 5.1 Introduction The discussion in the previous chapters assumed that the movement of the su bject out of the CNP clause in Adjunct Control constructions is possible. There is no mention, however, about why the subject moves. Within the Minimalist Program, movement does not happen for free, which brings us face to face with the topic of this chapt er: why movement? The chapter is organized as follows. Sections 5. 2 through 5. 4 survey the literature for a possible answer. Section 5. 2 checks whether Enlightened Self Interest as proposed by Lasnik (1995) is the reason why the CNP subject moves. Section 5. 3 (2006) Event based analysis of Obligatory Control and the idea that control involves movement only if the two clauses in a control structure do not express separate events. Section 5. 4 checks whether the size of the adjunct (IP vs. CP) is decisive with respect to movement. The three accounts fail to explain why movement in Telugu and Assamese Adjunct Control structures takes place. In Section 5 .5 ( 2006 ) proposal that a probe goal relationship must be established between any two elements undergoing merge. I suggest that CNP clauses in Te lugu and Assamese are non goals. The movement of the CNP subject makes up for this imperfection and makes it possible for the adjunct to merge with the matrix clause. In Section 5. 6, I extend the analysis to English. Section 5. 7 is a conclusion. 5.2 Enlightened Self Interest and Control The Minimalist Program holds that movement takes place as a Last Resort for the purpose of feature chec king. Originally, Last Resort assumed that movement is greedy, as the formulation in (1) indicates (Chomsky 1995). According to (1), an element can only move to check a feature

PAGE 197

197 of its own. Lasnik (1995 2002: 28 (21b)) argues that Last Resort must take the form of Enlightened Self Interest as formulated in (2). In this sense, an element moves to satisfy a need as Suicidal Greed. (1) Greed (2) Enlightened Self Interest The following subsections examine the two possibilities offered in (2) i n order to see if they can explain why the CNP subject moves. Section 5. I n other words, the CNP subject moves in order to check a feature of its own. Section 5 hat is, the CNP subject moves in order to check a feature on the target. 5.2.1 It can be speculated that the CNP subje ct in Adjunct Control structures moves to the matrix clause in order to check its Case feature. This idea may work for Assamese in which CNP clauses do not seem to license Structural Case on the subject. Thus, it can be argued that the CNP subject moves to the matrix clause to check Structural Case. Evidence comes from the degradation of Assamese Backward Control structures like (3). At a certain point in the derivation, the two non command relationship and form a chain as the dotted arrow in (4) signifies. When linearization applies at PF, Chain Reduction has to delete one of the two copies. The victim of deletion is normally the CNP subject. The reason is that it has at least one unchecked feature, namely, Structural Case

PAGE 198

198 (3) Assamese ?? Ram e [Ram Or khong uth i] mor ghorto bhangil e Ram NOM [Ram NOM anger raise CNP] my house destroyed 3 (4) CP 3 IP C qp SUBJ I' Ram e qp vP I qp CNP P vP 3 3 SUBJ 6 SUBJ 6 Ram Or khong uth i Ram e mor ghorto bhangil e This line of reasoning does not work for Telugu, however. Telugu CNP clauses do license Structural Case marked subjects. Evidence comes from grammatical instances of Backward Co ntrol constructions like (5) below. As (6) shows, at a certain point in the derivation two non distinct copies of Kumar the matrix subject and the CNP subject, enter a c command relationship and form a chain At PF, linearization applies and Chai n Reduction marks one of the copies for deletion. Because both copies have checked all their features, including Structural Case, Chain Reduction is free to choose either copy. If the matrix copy is deleted, the result is Forward Control; otherwise, Backwa rd Control. 1 1 Another type of evidence comes from Backward Control struc tures in which the CNP clause licenses a nominative subject while the matrix clause is a dative subject (e.g., (i)). As mentioned in Chapter 2, such structures are considered marginal because Telugu disfavors a dative subject in the matrix clause of Adjunc t Control structures. The marginality of (i) is independent of Backward Control, as the degradation of the Forward Control counterpart in (ii) shows. (i) ?? Sarita ki [Sarita aa maaTa win i] koopamu waccin di Sarita DAT [Sarita.NOM that matter hear C NP] anger came 3.N.S (i i) ??Sarita ki [ Sarita aa maaTa win i] koopamu waccin di Sarita DAT [ Sarita.NOM that matter hear CNP] anger came 3.N.S

PAGE 199

199 (5) Telugu Kumar [Kumar ki jwaram wacc i ] hospital weLLaa Du Kumar.NOM [Kumar DAT fever come CNP ] hospital went 3.M.S (6) CP 3 IP C qp SUBJ I' Kumar qp vP I qp CNP P vP 3 3 SUBJ 6 SUBJ 6 Kumar ki jwaram wacc i Kumar hospit al weLLaa Du The main idea is that Case cannot be the reason why the CNP subject moves to the matrix clause, at least not in Telugu. Apart from Case, I cannot think of another uninterpretable feature that the CNP subject might bear and that might trigger movement. A subject normally plays a role in the checking of uninterpretable phi features on functional layer s (I 0 ) and/or, arguably, the theta role feature on lexical layer s (v 0 ). T hese, however, are not uninterpretable features on the subject itself. Ac work. Section 5. 2.2 checks whether the movement of the CNP subject results in the checking of a feature on the target. Since Assamese CNP subjects might move for their own pu rposes, namely, to check Case, the focus of the following section will be only on Telugu. 5.2.2 The first landing site of a sideward moving CNP subject is Spec,vP of the matrix clause. If moves in order to check a feature on v 0 Hornstein (1999, 2003) as well as Lasnik ( 2002 )

PAGE 200

200 ( 1994 ) ( 1998 ) among others argues that theta roles are fea tures, just like Case and phi features If this is correct, then the movement of the CNP subject to Spec,vP results in the satisfaction of the thematic requirement of v 0 This seems like the end of the story. Nonetheless, there is a reason to believe that the theta role feature of the matrix predicate does not necessarily trigger movement all the time Here is why. Altruistic sideward movement of the type depicted in this subsection takes place when an instance of merge is needed for the derivation to conve rge, yet there are no tokens left in the n umer ation to satisfy this need ( Nunes 2001, 2004). What t his amounts to is the following. T he CNP subject moves to Spec,vP in order to check the theta role feature on v 0 only if there is no token left in the numera tion that can undergo merge in order to salvage the derivation. To elaborate, matrix vP can satisfy its theta role requirement either via External Merge, whereby an item selected from the numeration merges as an argument, or via Internal Merge, whereby an item that is already in the structure copy plus merges in Spec,vP. The latter option applies in order to sav e the derivation only if the former option is not possible because the n umeration is already exhausted. If this is correct, we should expect struct ures with CNP clauses to allow disjoint subjects. These would be structures where the matrix predicate has its thematic requirements satisfied by an element from the numeration rather than by the CNP subject. This does not happen, however. Consider (7) for example. The derivation starts with the numeration in (7a) which consists of the two potential arguments, Kumar and Sarita The indices in the numeration indicate the number of copies of each token. The CNP clause and matrix vP form independently in (7b), reducing the indices of most of the items in the Numeration to zero as (7 c ) shows. Notice, however, that the copy Sarita has not been used up yet This means that it can check (it actually has to check) the

PAGE 201

201 theta role feature of matrix vP, as (7 d ) shows. This makes the sideward movement of Kumar unnecessary. Accordingly, a structure like (8) should be possible, contrary to fact s The bottom line is that the CNP subject has to move and that sideward movement does not only happen when the numeration is exha usted (7) a {Kumar 1 Sarita 1 koopamu 1 wacc 1 i 1 akkadi 1 nunci 1 wellipoy 1 Tense 1 Agr 1 } b. i. [ CNP [ NP Kumar ki] koopamu wacc i ] [ CNP [ NP Kumar DAT] anger come CNP] ii. [ Matrix vP akkadi nunci wellipoyinaa di] [ Matrix vP there from left 3.N.S] c. {Kumar 0 Sarita 1 koopamu 0 wacc 0 i 0 akkadi 0 nunci 0 wellipoy 0 Tense 1 Agr 1 } d. [ Matrix vP [ NP Sarita] akkadi nunci wellipoyinaa di] (8) *[Kumar ki Koopamu wacc i] Sarita akkadi nunci wellipoyinaa di [Kumar DAT anger come CNP] Sarita .NOM there from left 3.N.S It is important to note that a scenario like the one delineated in (7), although not viable for structures involving a CNP clause, is possible if the structure involves a different type of adjunct. Recall from Section 2. 3 that Telugu, as well as Assamese, has other types of non finite adjuncts which I called INF clauses. These are similar to Telugu CNP clauses with regard to Case and agreement. I n other words, both CNP and INF claus es license Inherent and Structural Case marked subjects and the verb in both types of clauses shows no overt agreement. INF clauses stand out as different in two ways, however. First, they may take an overt complementizer. Second, they allow disjoint subj ects. These differences are illustrated in the Telugu examples in (9); compare to sentence (8) above. (9) a [Kumar sinima cuus tunna appuDu] Sarita popkorn tinnaa di [Kumar.NOM movie watch INF while] Sarita.NOM popcorn ate 3.N.S

PAGE 202

202 b [Kumar ki k oopamu wacc ina anduku] [Kumar DAT anger come INF because] Sarita akkadi nunci wellipoyinaa di Sarita.NOM there from left 3.N.S Although structures with INF clau ses allow disjoint subjects, if the subject of the non finite adjunct is not pronounced, Obligatory Control applies. In other words, the Sarita No matter how much cont ext is provided, a reading with disjoint subjects is infelicitous. 2 (10) Telugu a. i/*k Koopamu wacc ina anduku] anger come INF because] Sarita i akkadi nunci wellipoyinaa di Sarita.NOM there from left 3.N.S b i/*k sinima cuus tunna appuDu] Sarita i popkorn tinnaa di movie watch INF while] Sarita.NOM popcorn ate 3.N.S Once an overt pronoun is used, it can refer either to Sarita (granted it i s 3 rd person singular feminine) or to another character mentioned earlier in discourse, as the sentences in (11) indicate. (11) Telugu a. [aame i/k Koopamu wacc ina anduku] [she.NOM anger come INF because] Sarita i akkadi nunci wellipoyinaa di Sarita. NOM there from left 3.N.S b. [aame i/k sinima cuus tunna appuDu] Sarita i popkorn tinnaa di [she.NOM movie watch INF while] Sarita.NOM popcorn ate 3.N.S 2 Sen gupta (1999: 302 303) makes the same observation about Bengali.

PAGE 203

203 I take the structures in (10) as instances of optional control 3 S entence (10a) starts with the numeration in (12a) By the time the INF clause and matrix vP are f ormed in (12 b), no argument is left in the numeration to check the theta role feature of mat rix vP as (12c ) shows This is why Sarita undergoes sideward m ovement in (12d) The derivation converge s as (12e). (12) a. {Sarita 1 koopamu 1 wacc 1 ina 1 aanduku 1 akkadi 1 nunci 1 wellipoy 1 Tense 1 Agr 1 } b. i. [ INF [ NP Sarita ki] koopamu wacc ina and uku] [ INF [ NP Sarita DAT] anger come INF because] ii. [ Matrix vP akkadi nunci wellipoyinaa di] [ Matrix vP there from left 3.N.S] 3 The distinction between required and optional control as used here is different from the distinction between Obligatory vs. Non Obligatory Control made by Hornstein (1999, 2001, 2003) or Lan dau (2000, 2004). A required control interpretation arises when the movement of the subordinate subject has to happen all the time. In other words, if movement fails to happen, the derivation does not converge. This is the case of the Telugu and Assamese C NP subject which has to move due to a certain restriction to be specified in Section 5.5. Optional control, on the other hand, is the case of INF clauses whose subject moves to the matrix clause only if the numeration is exhausted and there is no other way to satisfy the thematic requirement of the matrix predicate. If the numeration is not exhausted, no control interpretation is required. Rodrigues (2004) analyzes a similar case of optional control in Brazilian Portuguese and Finnish. The distinction betw een Obligatory and Non Obligatory Control is different. According to Hornstein, Obligatory Control is the result of movement. Non Obligatory Control, on the other hand obtains when the subordinate clause is a non finite CP that is also an island out of whi ch movement is disallowed. In this case, the subordinate subject is pro and no movement is involved (Hornstein 2001: 56 59). For example, treating herself to an ice cream in (i) is a subject gerund out of which movement is not allowed. This is why the subj ect is pro (i) Tom said that [ pro treating herself to an ice cream pleased Mary] Landau offers a different analysis. He holds that Obligatory Control is the result of an ana phoric PRO that is c commanded by a matrix functional layer (i.e ., the matrix v 0 or I 0 ) and can thus enter an Agree relation with these layers ; for example, PRO in (ii) is c commanded by the functional layers in the matrix clause. Non obligatory Control (NOC) constructions are sentences in which PRO is logophoric I n o ther words, the identity of PRO is determined not by an antecedent finite clause is extraposed to a position where it is not visible to the functional layers in the matrix clause PRO gain s the status of logophors PRO in (iii) has this status. (ii) Tom likes [PRO to treat himself to an ice cream every now and then]. (iii) Sue likes ice cream; Tom knows that, and he lik es to see her happy. [PRO treating herself to an ice cream every now and then] pleases Tom. I will not address the distinctions (Obligatory vs. Non Obligatory) made by Hornstein and Landau in my study mainly because they do not seem to apply to the Telug it is not clear why the subject of INF clauses would be an unpronounced pro when a control interpretation is available, while it is a lexical subject when no such interpretation is required. Conce rning Landau, his system predicts that when the CNP clause is realized sentence initially, the CNP subject will be logophoric, contrary to facts. An Obligatory Control interpretation is required all the time.

PAGE 204

204 c. {Sarita 0 koopamu 0 wacc 0 ina 0 aanduku 0 akkadi 1 nunci 0 wellipoy 0 Tense 1 Agr 1 } d. [ NP Sarita] e. [ Matrix vP [ NP Sarita] akkadi nunci wellipoyinaa di] E xamples (9 11) suggest that only takes place when the numeration is exhausted and there is no other token that may satisfy the formal requirements of grammatic al only if they receive an Obligatory Control interpretation. If control is movement, this means that movement in structures with CNP clauses has to take place all the time and not only when the numeration is exhausted This kind of movement is responsible for the Obligatory Control interpretation that structure s with CNP clauses strictly require. All in all, Enlightened Self Interest is not able to explain why the sideward movement of the CNP subject is mandatory. The following section tries to find an exp lanation somewhere else. 5.3 Event and Control Another proposal is offered by Dubinsky and Hamano (2006) who examine control into a certain type of adjuncts in Japanese. These adjuncts involve a locative ni as exemplified in (13a b) ((4a b) in original). The controller is the matrix subject, and the controllee is the unpronounced possessor of the ni marked NP in the adjunct. (13) Japanese a. Mari wa [tue o yoko ni] tatiagatta Mari TOP [cane ACC side at] stood.up b. Mari wa [tue o yoko ni si te] tatiagatta Mari TOP [cane ACC side at do.TE] stood.up OR

PAGE 205

205 Sen tence (13a) is an instance of Obligatory Control. The identity of the possessor of the ni marked NP must be determined by the matrix subject. Sentence (13b), which involves the light verb si te does not have this requiremen t, as the English translation shows. Dubinsky and Hamano consider (13a) as an instance of movement, as illustrated in (14a). The adjunct is an aspectual phrase (AspP). The possessor Mari of the ni marked NP yoko the subject position in the matrix clause. At PF, the lower copy is deleted. Sentence (13b), on the other hand, does not involve movement. As (14b) shows, the adjunct is a vP/TP, and the possessor of yoko pro which might or might not refer to the matrix subject. The presentation in (14) is based on (14) in original. (14) Japanese a. Mari wa [ AspP tue o [Mari poss yoko] ni] tatiagatta Mari TOP [ AspP cane ACC [Mari poss side ] at] stood.up b. Mari wa [ TP [ vP [ AspP tue o [pro poss yoko] ni] si] te] tatiagatta Mari TOP [ TP [ vP [ AspP cane ACC [pro po s s side] at] do]TE] stood.up Another difference between (13a) and (13b) is that the former depicts one event while the however, refers to a sequence of events: the cane is put on it s side first, and then Mari stands up. Based on this observation, Dubinsky and Hamano conclude that the adjunct in (13a) is part of the event of the matrix clause, while (13b) comprises two independent events which allow different temporal expressions. The y attribute this difference to the presence of si te in (13b). 4 4 Notice that in both (13a) and (13b) the tens e of the adjunct is dependent on the tense of the matrix clause. B oth have to be in the past if the matrix temporal information is past. Dubinsky and Hamano (2006) and Dubinsky (2007) use this observation to argue, contra Landau (2000, 2004) that Event ra ther than Tense is responsible for the Exhaustive Control Partial Control distinction. According to them, only Exhaustive Control structures involve movement which is possible due to the lack of an Event Phrase in the subordinate clause. Partial Control st ructures comprise a subordinate clause that has an independent Event Phrase. As Dubinsky and Hamano put it, checking multiple theta roles is possible within the same Event frame but not across Events.

PAGE 206

206 Based on the event splitting quality of te Dubinsky and Hamano maintain that an adjunct with te projects an Event Phrase which assigns an event index EN on an agreeing NP. Conversely, an a djunct without te does not project an Event Phrase and, consequently, the possessor NP is not assigned an event index. They further hold that an NP may check more than one theta ith index EN cannot further check theta possessor in (13a), which has not been marked with EN yet, is allowed to move out of the adjunct to check the theta role feature of the matrix vP, while t he possessor in (13b) freezes for theta role related movement due to the EN that it has received from te Assamese. The reason is that CNP clauses in the two South Asian lang uages denote their own events, as illustrated in (15 16). Each of (15a b) and (16a b) contains a CNP clause with a temporal expression that is distinct from that of the matrix clause. This is an explicit indication that the CNP clause depicts its own event Sentences (15d) and (16d) do not include any temporal expressions. Semantically, however, they only mean that the particular couple kissed and then left. They do not mean that the couple left (while) kissing. (15) Telugu a. Sarita [enimidinTiki bhojanamu tayaru ceesikun i] Sarita.NOM [at 8:00 dinner prepare do for self CNP] tommidinTiki tinnaa di at 9:00 ate 3.N.S b. [Naa boss ki pooyina waaram koopamu wacc i] My boss last wee k anger come CNP] atanu muuDu roojulu waraku tina leedu he three days until eat

PAGE 207

207 c. [Kumar unnu Sarita muddu peTTukon i] [Kumar.NOM and Sarita.NOM kiss put for each other CNP] waallu iddaru wellipooyaa ru they both left 3.M.P NOT : (16) Assamese a. [January t loteri jik i] [January in lottery win CNP] Ram e Marc h ot notun ghor kinil e Ram NOM March in new house bought 3 b. [jua bosor kukur tu heru i] [past year dog Cl lose CNP] boss tu r dui xoptah r karone dukh lagil boss Cl G EN two weeks GEN for sad felt c. [suma kha i] Ram aru Prajakta gusi gol [Kiss eat CNP] Ram and Prajakta away went NOT: own Event Phrase does not involve movement for the purpose of control. The possessor of the ni marked NP is a pro whose ide ntity may be determined by the matrix subject, tho ugh it does not have to be ( e.g., (13b)). The reason is that an argument that is assigned an Event Index (EN) is not allowed to move into a new theta role position. The examples in (15) and (16) suggest tha t CNP clauses in Telugu and Assamese do project an Event Phrase. Still, they all enforce an Obligatory/Exhaustive Control interpretation and, most probably, mov ement Therefore, the event structure of Telugu and Assamese control constructions is not respon sible for the sideward movement of the CNP subject. The following section examines yet another possibility.

PAGE 208

208 5.4 CP vs. IP and Control The difference in Control behavior between CNP clauses and INF clauses seems to be related to the size of the adjunct. The main difference is that INF clauses may involve an overt complementizer (17a, 17b), while CNP clauses categorically disallow an overt complementizer (17b, 18b). (17) Telugu a. Kumar [ Kumar ki aakali wees ina anduku] Kumar.NOM [ Kumar DAT hunger fall INF because] bhojanamu tayaru ceesikunaa Du dinner prepare did for self 3.M.S b. Kumar [ Kumar ki aakali wees i (*anduku)] Kumar.NOM [ Kumar DAT hunger fall CNP ( because)] bhojanamu tayaru ceesikunaa Du dinner prepare did for self 3.M.S (18) Assamese a. [ Ram Or phurti lag a r karone] [ Ram GEN exhilaration do INF GEN because] Ram e pagolor nisena nasil e Ram NOM crazy person like danced 3 b. [ Ram Or phurti lag i (*karone)] [ Ram GEN exhilaration do CNP ( because)] Ram e pagolor nisena nasil e Ram NOM crazy person like danced 3 Therefore, it is fair to assume that INF clauses are CPs. By the same token, it is not unreasonable to consider CNP clauses as IPs, as they are conventionally assumed to be (Jayaseelan 2004). This observation sug ges ts that the size of the adjunct has a say in the obtainment of c ontrol: an IP adjunct enforces c ontrol, while a CP adjunct does not.

PAGE 209

209 This intuition is captured by San Martin (2004) who provides evidence from several languages ( e.g., Romanian and Serbo Croatian) to show that the subordinate clause in Exhaustive Control just like in raising is an IP. The sentences in (19 20) are examples from Romanian. Sentences (19a c) are raising structures (San Martin 2004: 93 (53 55), from Roussou 2001) They show t hat raising is only possible when the subordinate clause is an IP ( 19a). If the subordinate clause has an overt complementizers (i.e., if it is a CP), the subject has to remain downstairs ( 19b), because raising results in ungrammaticality ( 19c). The same a pplies to control. Sentence (20) is an instance of Exhaustive Control, in which case the subordinate clause may not be realized with an overt complementizer (San Martin 2004: 94 (54), from Alboiu and Motapanyane 2000) (19) Romanian a. sau nimerit fie bolnavi] All boys the REFL have happened [SUBJ be sick] b. Sa nimerit [ca fie bolnavi] It has happened [COMP all boys the SUBJ be sick] c. sau nimerit fie bolnavi] *All boys the REFL have happened [COMP SUBJ be sick] (20) Romanian Mio arai a nceput [(*ca) sei de plecare] Mioara has started [( COMP) SUBJ REFL prepare 3sg of departure] San Martin excludes Partial Control from her analysis. 5 What Partial Control structures have in common is that they allow an overt complementizer, in which case c ontrol does not apply. Sentence (19a) is an example of Partial Control from English. Sentence (19b) shows that 5 The appendix at the end of the dissertation explains the distinction between Exhaustive and Partial Control.

PAGE 210

210 an overt complementizer in the subordin ate clause is possible, and no c ontrol requirement is enforced. (21) a. Tom wants [to meet at 7:00]. b. Tom wants [for Sue to win.] bservation seems to apply to the c ontrol structures analyzed in this chapter. A CNP clause may never occur with an overt complementizer, as (22) shows. (22) a. Telugu [ Kumar unnu Sari ta okari ki okaru kathalu ceppukum tuu (*appudu)] [ Kumar and Sarita.NOM one to one stories tell CNP ( while)] Kumar unnu Sarita nawwukunnaa Du Kumar.NOM and Sarita.NOM laughed 3.M.P b. Assamese [ Ram aru Prajakta e Suma kha i (*pasot)] [ Ram and Prajakta NOM Kiss eat CNP ( after)] Ram aru Prajakta gusi gol Ram and Prajakta ABS away went In addition, sentences wi th CNP clauses are Exhaustive Control structures, as exemplified in (23a) and (24a). When an INF clause is used, Partial Control is allowed, as (23b) and (24b) show. Note that in (23b) and (24b), the matrix subject has to be coreferential with ( a part of) the subject of the CNP clause. Complete obviation is not allowed. This is why, no matter how much context is provided, (23c) and (24c) are infelicitous under the designated reading. (23) Telugu a. [ Kumar unnu Sarita okari ki okaru kathalu ceppukum tuu] [ Kumar and Sarita.NOM one to one stories tell CNP] Kumar nawwukunnaa Du Kumar.NOM laughed 3.M.S

PAGE 211

211 b. [okari ki okaru kathalu ceppukum tunna appudu] Kumar nawwukunnaa Du [one to on e stories tell INF while] Kumar.NOM laughed 3.M.S c. [okari ki okaru kathalu ceppukum tunna appudu] [one to one stories tell INF while] Rao nawwukunnaa Du Rao.NOM laughe d 3.M.S (24) Assamese a. *[ Ram aru Prajakta e Suma kha i] Ram gusi gol [ Ram and Prajakta NOM Kiss eat CNP] Ram ABS away went b. [su ma khaw a r pasot] Ram gusi gol [Kiss eat INF GEN after] Ram ABS away went c. *[suma khaw a r pasot] Proxad gusi gol [Kiss eat INF GEN after] Proxad ABS away went Let us assume with San Martin that Exhaustive Control structures include IP subordinate clauses while Partial Control structures contain CP subordinate clauses. T he sentences in (22) through (24) seem to sug gest that CNP clauses are IPs as I continue to assume, while INF clauses are CPs. This assumption is based on two facts : First, CNP clauses never take an overt compl ementizer, while INF clauses do 6 Second, CNP clauses enforce Exhaustive Control (Appendi x) while INF clauses do not. The question is: What can CP do that IP cannot? And how does CP interfere with control? At least two answers are available in the literature. These will be discussed in the following subsections. 6 We can also assume with Rizzi (1997) that what is originally referred to as CP is actually divided into at least two parts: (i) one facing the outside and is referred to as ForceP, and (ii) one faci ng the inside and is called FinP. Complementizers are realized in ForceP, while information about finiteness resides in FinP. Knowing that CNP clauses are non finite, we can be quite certain that they project as high as FinP. Since they do not allow an ove rt complementizer, however, it is reasonable to assume that they do not project as high as ForceP.

PAGE 212

212 5.4.1 IP as Defective for [Per son] a version of the PRO theory of control. She argues that control structures only license PRO and that PRO and lexical subjects are in complementary distribution. She also maintains that the size of the complem ent clause is decisive in the licensing process: (25) 7 According to San Martin, IP complements have an incomplete I 0 with [+Tense] feature but no [Person] feature. I 0 [+Tense, Person] is able to check Case on the embedded subject (evidence for Case comes from elements such as floating quantifiers in Icelandic (Sigur sson 199 1 )), but it is not able to license a lexical/overt subject. An indic ative CP, however, licenses a lexical subject because C 0 endows I 0 with a [Person] feature. Consequently, a CP complement has a complete I 0 [+Tense, +Person] which can check the [Person] feature of its subject, allowing it to be overt. The lexical subject of a CP complement does not have to be corefe rential with the matrix subject. Thus, an Obligatory Control interpretation is not required This analysis becomes suspect when examined in the the light of the Telugu and Assamese data. Chapter 3 shows that Tel ugu and Assamese Adjunct Control structures qualify as instances of Obligatory Control or, more specifically, Exhaustive Control (Appendix) system, this is only possible if the CNP clause is an IP. At the same time, Telugu and Assamese A djunct Control structures may be realized as instances of Copy Control. In other words, what is Assamese. This should be possible only if the CNP clause is a CP. T his contradiction leads to o ne of the following conclusions. The first conclusion is that CNP clauses are CPs, which is why 7 Landau (2006: 167; fn. 9) provides a brief argument against this view.

PAGE 213

213 they may license a lexical subject. In this case, it remains to be explained why the lexical subject has to be obligatorily corefere n tial with the matrix subject. Another possibility is that CNP clauses are IPs, which is why a control interpretation is required In this case, we need to explain how a defective I 0 [ Person] can license a lexical subject. Any attempt at explaining either conclusion has to resort to a stipulation and will not capture be made here. The following section gives the issue of size another chance and examines the difference between CPs and IPs from a slightly different perspective. 5.4.2 IP as Defective for [Tense] A ccording to Chomsky (2000: 124; Chomsky 2004 ), clauses that project no higher than IP are defective in the sense that they cannot license a Structural Case marke d subject. He holds that I 0 derives tense and agreement from C 0 I 0 without C 0 has defective [Tense] that does not 8 This idea is analyzed at length in Pesetsky and Torrego (2001) who argue that Structural Case is an uninterpretable [Tense] feature on NPs/DPs. I n the environment of I 0 with defective or unvalued [Tense], the subject does not check Structural Case. Assuming that CNP clauses ar e IPs while INF clauses are CPs, the conclusion is that CNP clauses do not license a Structural Case marked subject because they have a defective I 0 Consequently, the subject has to move. INF clauses, on the other hand, have a complete I 0 This means that the INF subject checks its Structural Case feature, which is why it does not have to move. 8 ructions. The subordinate clauses of control structures have a non defective Tense feature that can check Null Case on the verb.

PAGE 214

214 This approach works for Assamese. As Section 5. 2 mentions Assamese CNP clauses do not seem to check Structural Case, which is why Backward Control constructions ar e degraded. The same does not apply to Assamese INF clauses which do license Structural Case marked subjects. The approach does not work for Telugu, however. The fact that Forward and Backward Control structures are perfectly interchangeable and equally a cceptable in Telugu indicates that the CNP subject and the matrix subject have an equal status with regards to Case. Otherwise, Chain Reduction would always favor the matrix copy and mark the CNP copy for deletion. Under such circumstances, Forward Control structures would be at least more acceptable than their Backward Control counterparts, which is not true. To summarize thus far, Sections 5. 2 shows that Enlighte ne d Self Interest is less likely to be the reason why the CNP subject moves to the matrix cla use. Section 5. 3 explores Dubin s k y and role position is possible only if the subordinate clause and the matrix clause comprise the same event. This possibility has also been ruled out on empirical grounds: Telug u and Assamese CNP clauses project their own event phrases. Still, movement takes place. Finally, Section 5. 4 examines the possibility that the size of the adjunct (CP vs IP) has an effect on movement. All three sections share one main idea: the movement of v 0 ) is not triggered by formal requirments or features (Case, theta role, Event Index, etc.) re left with two possibilities. no movemen t at all B oth subjects in Telugu and Assamese Adjunct Control st ructures are base generated. Or, has to move to Spec, vP of the matrix clause to satisfy a requiremen t neither of its own, nor of v 0

PAGE 215

215 The former option is certainly plausible. As a matter of fact, it is a restatement of the PRO Theory of Control. Adopting this option however, means revising some of the major premises of the PRO Theory of Control, a proj ect that is worth a separate study in its own right. The latter o ption, on the other hand, is an invitation to dig a little deeper in order to see what else other than resides in the nature of the adjunct itself, not as a CP or a n IP, but more importantly as an appropriate goal for merge. 5. 5 Probe Goal Relationship s Pesetsky and Torrego (2006) maintain that before two syntactic objects (heads and/or established goal relation is triggered by a n unvalued feature on the probe. I for merge and for the semantic or phonological relations Requirement on Merge. (26) The Vehicle Requirement on Merge Whereas the requirement is traditionally accep ted for instances of movement (or Internal Merge) Pesetsky and Torrego apply it to External Merg no instance of merge is fre Adger ( 2003: 91 ) Chomsky ( 2000: 132 135 ) and Hornstein ( 2001: 56 ) arrive at a similar conclusion Pesetsky and Torrego recognize two types of External Merge External Merge that result s in agree ment and External Merge that do es not result in agreement They concern themselves with the latter. Both types of merge involve a probe goal relationship. The former type results in

PAGE 216

216 Take the noun category N, for example. As (27 28) illustrate, N may take a CP complement with that (27a) or a PP complement (28a). It does not take a CP complement without that (27b) or a DP complement (28b). Notice that the verb counterpart V may take a complement without that (27c) or a DP complement (28c). (27) a. the belief that Tom has gone bankrupt b. *the belief Tom has gone bankrupt c. I believe (that) Tom has gone bankrupt. (28) a. b. c. The company appointed Tom. B ased on their earlier work (Pesetsky and Torrego 2001), the authors attribute the facts in (27 28) to the feature characteristics of the complementizer that and the preposition P. They hold that that and P are bearers of a [Tense] feature. Eviden ce for this feature comes from the behavior of the two elements. The argument goes a s follows. Both that and P show the X trace effect exemplified in (29 30) (in original (12), from Perlmutter 1971) Sentences (29a b) involve a wh element that undergoes fi rst merge in the object p osition of a subordinate clause; (29a) does not include that while (29b) does. Both sentences are grammatical. Sentences (29c d), on the other hand, involve a wh element that moves out of the subject p osition of a subordinate clau se. Sentence (29c) does not include that ; the sentence is grammatical. Sentence (29d) includes that ; the sentence is ungrammatical (29) a. What do you think [Mary read what ]? b. What do you think [that Mary read what ]? c. Who do you think [ who read the book]? d. *Who do you think [that who read the book]? Prepositions behave in the same way as (30) illustrates (in original (13), from Kayne 1979) Sentences (30a b) below involve a wh element that moves out of a subordinate clause not heade d by a preposition. In (30a), the wh element starts out in the object position,

PAGE 217

217 while in (30b) it starts out in the subject position B oth sentences are accepta ble. When a preposition is involved wh movement out of the object po sition is acceptable (e.g., (30c)), while wh movement out of the subject position results in a degraded question (e.g., (30d)). (30) a. How much headway did he anticipate [Mary making how much on the issue]? b. How much headway did he anticipate [ how much bei ng made on the issue]? c. How much headway did he talk [about Mary making how much on the issue]? d. ??How much headway did he talk [about how much being made on the issue]? The X trace effect demonstrated in (29d) and (30d) corresponds to the behavior of dummy do In questions where the subject undergoes A' movement in the form of a wh element, dummy do cannot move to C 0 For example, (31a) is ungrammatical because do move to C 0 Compare to (31b) in which no such movement is involved and to (31c) in whi ch the wh element is an object NP. This behavior of do in (31a) patterns with the behavior of that in (29d) and P in (30d). A ll three elements are banned in C 0 when the subject undergoes A' movement. Pesetsky and Torrego consider the movement of do as an i nstance of T to C movement, with do bearing a [Tense] feature. They further hold that since that and P behave in the same way, they must bear the same feature: [Tense]. (31) a. *Who did buy a new car? b. Who bought a new car? c. What did Tom buy? Back to the p uzzle of nouns in English. N may take as a complement a CP with that or a PP, but not a CP without that or a DP. Recall that the Vehicle Requirement on Merge as stated in (26) enforces a probe goal relationship prior to the implementation of the operation merge. According to Pesetsky and Torrego, what is traditionally considered an uninterpretable Case feature on N is actually an uninterpretable tense feature [ u Tense]. When N enters a probe goal relation with the four possibilites in ( 32) below, only (a) an d (b) fulfill the Vehicle Requirement on Merge. The reason is that they are appropriate goals for the probe N since they bear an

PAGE 218

218 interpretable [Tense] feature. Options (c) and (d), on the other hand, are not appropriate goals. They do not bear an interpret able [Tense] feature and, thus, they cannot value [ u Tense] on N. (32) a. CP with that [Tense] b. PP [Tense] c. CP without that d. DP Interestingly, this approach provides a solution to a long standing problem of English infinitival r elatives. It is a fact about English that N can take an infinitival relative that is introduced by a pied piped wh PP (e.g., (33a b)), but it cannot take an infinitival relative that is introduced by a bare DP ( who which ) (e.g., (34a b)). According to Pes etsky and Torrego, the Vehicle Requirement on Merge provides an answer to this puzzle. DPs are not an appropriate goal for N. N has [ u Tense] feature that finds its match on the head of PPs but not DPs. This is why (33a b) are acceptable, but (34a (33) a. a person [with whom to speak at the conference] b. a topic [on which to work] (34) a. *a person [who to invite to the conference] b. *a book [which to read] Notice that if N merges with CP or PP, no agreement takes place. Normally, if an element X wi th an uninterpretable feature [ u F] merges with another element Y that bears an interpretable feature [F], the expectation is that the unvalued feature [ u F] on X will be valued, a step that is usually manifested by agreement. Yet, this does not happen in (3 3a b). Pesetsky and Torrego (2006: 24) solve this puzzle with a speculation at this point. They maintain that although the probe goal relation of the type discussed above is initiated for the purpose of the valuation of a feature on the probe, no valuation /agreement takes place because those instances of merge N [ u Tense]

PAGE 219

219 marking and ea can be summarized as follows. P robe goal relations have to be established before any operation of merge. If merge takes place at a lexical layer (e.g., in the thematic domain vP), no agreement takes place. If merge takes place at a functional layer (e.g., in the phi dom ain IP), agreement takes place. Another instance of merge that does not involve valuation of features or agreement is adjunction. The cases discussed by Pestesky and Torrego (2006), however, only handle the merge of complements. Si nce the main topic of this dissertation is Adjunct Control into CNP clauses, the following s ection upgrades the Vehicle Requirement on Merge so that it accommodates adjunction. 5.5.1 The Vehicle Requirement on Merge Revisited Unlike complements, adjuncts do not have to meet the selectional requirements of the head they merge with (Chomsky 2004: 1 17). This means that adjuncts do not enter a probe goal relation with the head of the structure they adjoin to, and accordingly they do not value features on probes. Still, adjunction is a type of merge, which means that the Vehicle Requirement on Merge mu st be a requirement on adjunction as well. C omplements and adjuncts undergo different types of merge. One difference is that c omplements are selected by the head of the structure they adjoin to, while adjuncts are not Chomsky (2004: 117). A probe goal rela tion is subject to the selectional requirements of the probe. In this sense, in (35) below, is limited to Set Merge only. This is why I suggest restating it as (36). (35) The Vehicle Requi rement on Merge

PAGE 220

220 (36) The Vehicle Requirement on Set Merge What about adjuncts? A lthough they are not selected, they undergo merge all the same. Accordingly, they must obey some requireme nt on merge which I name the Vehicle Requirement on Pair Merge: (37) The Vehicle Requirement on Pair Merge a. potential goal for some head b. A (potential) goal is a syntactic object with some feature F. A non goal is a syntactic object with no features. N. In English, a PP may function as an adjunct an d undergo Pair Merge with a vP ( e.g., (38a) ) According to (37), it qualifies for such an instance of Pair Merge because it is a potential goal for a probe N, as (38b) shows. An instance of Set Merge between N and PP means that PP is a goal with some feature F that matches F on N. (38) a. I [ vP [ vP heard the news] [ PP in that restaurant]]. b. [ DP my [ NP interest [ PP in that restaurant]]] ure that makes it suggest that it can undergo merge only if at the end of the derivation it is predicated of a constituent in the syntactic object it merges with Section 5. 5.2 shows that this is possib ly the case of Telugu and Assamese CNP clauses. As a non goal, a CNP clause has to be predicated of an element in the matrix clause in order to undergo merge. First, however, it is important to note that t his speculation is not entirely new,

PAGE 221

221 although its i mplementation probably is. It is reminiscent of Predication Theory as proposed by Rothstein (2001) and much earlier work. 9 A brief overview of this theory is important in order to formalize the speculation and clarify the details of its implementation. Ac cording to Predication theory, every predicate has to have a subject. Predicate and subject are defined as (39) and (40) respectively. As the definitions indicate, predication is a syntactic relation between an open constituent or a predic ate and its subject for the purpose of forming a maximal constituent or a proposition. It is a purely structural condition, independent of the thematic requirements of the predicate (Rothstein 2001: 47, 59 60). (39) A predicate is an open constituent/a non proposition. It has to be predicated of or syntactically saturated by a subject in order to become a maximal constituent/a proposition. A predicate may take one of the following forms: a. A maximal projection (e.g., a VP) : When syntactically saturated, this kind o f predicate is externally predicated of a subject. That is, the predicate and its subject form a proposition whose properties are not dictated by the head of the predicate (e.g., [ vP Subject [ VP predicate]] ) ; or b. I' : This is the only predicate that is a non maximal projection. When syntactically saturated, I' is internally predicated of a subject. In other words, I' and its subject IP Subject [ I' I 0 ]). ( Roth stein 2001: 19, 47 4 9, 118 120) (40) A subject is a non predicate. It is a structural construct that may take one of the following forms: a. A n argument ; for example, an agent functioning as a subject of a transitive predicate, or an experiencer functioning as a subject of a psyc h predicate b. A non argument; for example, an expletive functioning as a subject of an unaccusative or a raising predicate ( Roth stein 2001: 18 20, 60 65) Despite the obligatoriness of the subject, Rothstein (2001 : 121) does recognize the fact that re are IPs in which subjects do not occur IP (41a) comprises a predicate I' but no subject. According to Rothstein, the subject position is 9 because it is m ore syntactic in nature.

PAGE 222

222 non obligatory only if the subordinate IP is indirectly predicated of an element in the matrix clause through co indexation, as the indices in (41b) illustrate. This observation may be formulated as (42). Within Predication Theory, (42) accounts for Obligatory Control. 10 The fact that the subordinate predicate in (41 b ) is predicated of the subject in the matrix clause explains Tom On this view, if a lexical DP fills the subject position (e.g., 41c), the subordinate predicate becomes a proposition and no longer needs to be predicated of an element in the matrix clause. Consequently no control interpretation will be required. (41) a. Tom wants [ IP [ I' to win] ] b. Tom i wants [ IP [ I' to win] i ] i c Tom wants [for [ IP Sue [ I' to win] ] (42) A predicate may remain an open constitue nt if it is indirectly predicated of an element in a higher clause. Evidence from Backward and Copy Control presents a challenge to (42). The Backward and Copy Control structures in (43) and (44) show that that the empty subject position in the subordina te clause may in fact be filled with a lexical subject. I take this as a sufficient reason to reject (42) and the idea that indirect predication in control structures follows from the fact that the subordinate predicate is necessarily syntactically unsatur ated. (43) Telugu Kumar [Kumar ki jwaram wacc i ] hospital weLLaa Du Kumar.NOM [Kumar DAT fever come CNP ] hospital went 3.M.S (44) Assamese [Ram Or khong uth i] Ram e mor ghorto bhangil e [Ram NOM ange r raise CNP] Ram NOM my house destroyed 3 10 This conclusion is known as the predicational approach to control. Despite the different formulations of Predication Theory (Rothstein 2001, Williams 1980 Clark 1990 among others), something along the lines of (42) is common to all.

PAGE 223

223 At the same time, the i ntuition behind (42) holds. The subordinate clause although a proposition that has its own subject is in a way predicated of the subject i n the matrix clause. I adopt this intuition and place it within the framework of the Movement Theory of Control. I suggest that the dual property of the subordinate clause ( the property of being a proposition and at the same time be predicated of a subject in the matrix clause) follows from the fact the subordinate subject and the matrix subject are non distinct copies of the same token brought about by movement. Back to the main question of this chapter: Why does movement take place? Sections 5. 2 through 5 4 showed that the movement of the subordinate subject cannot be triggered by a formal requirement of the moving element or of the target. Here the Vehicle Requirement on Merge comes into the picture. Unlike Rothstein who holds that the subordinate clause is allowed to remain an open constituent (i.e. it is allowed to not have its own subject) i f it is indirectly predicated of an element in the matrix clause, I suggest that the subordinate subject copies out of the subordinate clause and merges in the matr ix clause in order to allow the subordinate clause to be predicated of an element in the matrix clause. M ovement that takes place to this end happens if the s ubordinate clause is a non goal. T hat is, if the subordinate clause does not qualify for an instan ce of merge In this sense, the movement of the subject licenses the merge of the non goal, as (45) illustrates. Based on this, I suggest that there should be a Vehicle Requirement on the Merge of Non Goals as formulated in (46). (45) [ [Subject] Predicate [ NON GOAL [Subject] Predicate]] (46) The Vehicle Requirement on the Merge of Non Goals 11 11 Rothstein (2001: 122 123) distinguishes between two types of subordinate predicates: primary and secondary. One example of primary predicates is subordinate ECM small clauses. These are considered primary predicates

PAGE 224

224 Let us examine how (46) applies to Telugu and Assamese adjunct Control. Take the Telugu example in (47). Assume that the CNP clause is a non goal IP (I postpone the pr oof till the next section). To make up for this deficiency, it has to be predicated of an element in the matrix clause. That is, it has to function as a predicate /an open constituent I' of the type presented in (39b). Nevertheless, the CNP clause Kumar ki jwaram wacc i already a proposition ( or a maximal constituent IP) by virtue of being predicated of its own the CNP subject is not an o ption for at least two reasons. First, this kind of deletion is n (Chomsky and Lasnik 1995: 44). Further, Backward and Copy Contr ol structures are empirical evidence that the CNP subject is physically there. (47) Telugu [Kumar ki jwaram wacc i ] Kumar hospital weLLaa Du [Kumar DAT fever come CNP ] Kumar.NOM hospital went 3.M.S Fortun ately, there is a way for the CNP clause to be predicated of an element in the matrix clause without having to go through a mysterious operation of subject deletion. The key word is because they are predicated of a subject that is theta marked by a head inside their boundaries. For example, the predicate sweet in the ECM small clause complement in (i) is predicated of the subject the coffee that is theta marked by a head contained in the small clause. On the other h and, subordinate IPs of the type exemplified in (41), repeated here as (ii), are secondary predicates. These IPs are predicated of an element that is theta marked by a head outside their boundaries. For example, subordinate I' in (ii) is predicated of Tom which is theta marked by matrix v 0 (i) Tom considered [ SC th e coffee sweet ] (ii) Tom i tried [ IP [ I' to win] i ] i I do not draw such a distinction here. I believe that if the subordinate clause is a non goal, the movement of its subject to the matrix cl ause allows it to be predicated of an element in the matrix clause and thus licenses its merge. In other words, the requirement in (46) applies regardless of whether the subordinate clause is a complement of a control verb, a n ECM complement or an adjunct Also, it applies regardless of whether the moving subject is a theta marked argument at all, as the discussion on Expletive Control in the following chapter suggests. I leave it for future research to determine whether the requirement should be more rest rictive.

PAGE 225

225 in the matrix clause. The two copies are non distinct. This means that the CNP predicate is predicated of its own subject which at the same time is an NP of the matrix predicate. In this way, the movement of the subject licenses the CNP clause for an insta nce of merge. This implementation may be informally dubbed as the movement approach to indirect predication as proposed in (42). The movement of the subject is trigg ered by the non goal characteris tic of the clause that hosts it. If the non goal CNP clause does not qualify for a legal instance of merge because it violates the Vehicle Requirement on the Merge of Non Goals in (46). This explain s why (48) is unacceptable. (48) Telugu *[Kumar ki jwaram wacc i ] [Kumar DAT fever come CNP ] Sarita (atani ki) mandulu iccin di Sarita.NOM (him DAT) medicines gave 3.N.S Sentence (48) becomes grammatical if an INF clause is involved, as (49) shows. Let us assume that Telugu and Ass amese INF clauses are goal s This means that an INF clause can undergo Pair Merge with the matrix clause without further ado. In other words, the movement of the subject is not necessary. It does not have to move for its own purposes, and it does not have to move in order to license the merge of the INF clause; the INF clause is alr eady a goal. One reason remains. I t moves to check a feature on the target. As Section 5. 2 shows, this seems to happen only when the numeration is exhausted and the thematic requ irement of the matrix predicate have not been satisfied yet. In (49), Sarita satisfies the thematic requirement of the

PAGE 226

226 (49) Telugu [Kumar ki jwaram wacc ina anduku] [Kumar DAT fever c ome INF because ] Sarita (atani ki) mandulu iccin di Sarita.NOM (him DAT) medicines gave 3.N.S If the above analysis is correct, the different parts of the Vehicle Requirement on Merge in (36), (37 ), and (46) may be combined into (50): (50) The Vehicle Requirement on Set and Pair Merge a. b. goal for some probe c. a non an d. A (potential) goal is a syntactic object with some feature F. A non goal is a syntactic object with no features. The following section examines the requirement in (50) against the Telugu and Assamese data. It confirms the assumptions made above, namely, that Telugu and Assamese CNP clauses are non goals, while their INF counterparts are goals. Section 5. 6 extends the analysis to English. 5.5.2 Telugu and Assamese Adjunct Control and the Vehicle Requirement on Set and Pair Merge Obser ve the following sentences one more time. The structures in (5 1 52 ) include an INF clause each. Consequently, a control interpretation is not required. The structures in ( 53 54 ) are the equivalent of (5 1 52 ), except that the subordinate clause is a CNP adj unct. In this case, a control interpretation is required. (51) Telugu a. [Kumar sinima cuus tunna appuDu] Sarita popkorn tinnaa di [Kumar.NOM movie watch INF while] Sarita.NOM popcorn ate 3.N.S

PAGE 227

227 b. [Kumar pani ki weLL ina tarwaata] Sarita kaafi taagin du [Kumar.NOM work to go INF after] Sarita.NOM coffee drink 3.N.S (52) Assamese a. [Ram Or bhal lag a r karone] Proxad e nasil e [Ram GEN good feel INF GEN because] Proxad NOM danced 3 b. [Ram Or khong uth a r korone] Proxad gusi gol [Ram NOM anger raise INF GEN because] Proxad ABS went away (53) Telugu a. [Kumar sinima cuus tuu] atanu/*Rao popkorn tinnaa Du [Kumar.NOM movie watch CNP] he/*Rao.NOM popcorn ate 3.M.S b. [Kumar pani ki weLL i] atanu/*Rao kaafi taagaa Du [Kumar.N OM work to go CNP] he/*Rao.NOM coffee drink 3.M.S (54) Assamese a. [Ram Or bhal lag i] xi/*Proxad e nasil e [Ram GEN good feel CNP] he/*Proxad NOM danced 3 b. [Ra m Or khong uth i] xi/*Proxad gusi gol [Ram NOM anger raise CNP] he/*Proxad ABS went away If control is movement, this means that the CNP subject has to move to the matrix clause, which is why control is required. No such restric tion applies to the INF subject. This is why control is optional ( fn. 3 ) As Section 5. 2 shows, the INF subject arguably undergoes movement only if there is no argument left in the numeration to satisfy the thematic requirement of matrix pre dicate. As Sections 5. 2 through 5. 4 showed, the difference between CNP and INF clauses does not reside in the the formal requir ements (e.g., Case or Person ) of the subordinate subject, for both the CNP and the INF subject seem to check all their fe atures. The difference is not due to the formal requirements of the target either since the thematic needs of matrix v 0 do not vary.

PAGE 228

228 Further, both types of clauses can express their own events ; therefore, the difference does not reside in the event dependency on the matrix clause The difference seems to be due to the goal According to the Vehicle Requirement on Set and Pair Merge, a clause that is al so a goal may und ergo Set Merge, without involving movement The reason is that it has some feature F that matches F on a probe, and this is all that is needed for merge to take place. A clause that is a non goal, on the other hand, cannot undergo Set Merge without involvi ng movement. The reason is that it has to be predicated of an element in the syntactic object it merges with For this to happen, it has to share its subject with that syntactic object. Let us examine whether INF and CNP clauses may undergo Set Merge witho ut having to share their subject. Consider the Telugu sentences in ( 55 ). Each sentence has a NP complement. Knowing that a NP qualifies as a potential goal (it is a syntactic object with phi features), a syntactic object that can replace the complement in ( 55 a b) must itself be a goal. The sentences in (5 6 ) show that INF clauses can be complements in the same position. Th us it is fair to assume that they are goals. Notice that the INF clauses used in (5 6 a b) as complements are the same INF clauses used in ( 51 a b) as adjuncts. In both cases, no movement or control are involved. 12 (55) Telugu a. [ NP samayaM] anTee [ NP dhanam e] [ NP time] mean [ NP wealth EMPH] 12 Note that anTee in (53) is a filler. The sentences are also acceptable without it, especially that Telugu allows equational sentences like (i) and (ii). Nevertheless, the native speakers prefer the sentences in (55 56) with anTee which is clos e in meaning to the French ca veut dire (i) Sarita cakkani pilla Sarita.NOM pretty girl (ii) naa ku ays kriim (anTee) iShTam me to ice cream (means) liking

PAGE 229

229 b. [ NP aalaysam] anTee [ NP na Tam e] [ NP delay] mean [ NP l oss EMPH] (56) Telugu a. popkorn tina Daaniki sari ayina samayam anTee popcorn eating for proper happening time mean [cinimaa cuus tunna appuD e] [movie watch INF while EMPH] co rn is while watching a movie. b. Sarita kaafi taaga Daaniki sari ayina samayam anTee Sarita.NOM coffee drinking for proper happening time mean [Kumar pani ki weLL ina tarwaat e] [Kumar.NOM work to go INF after EMPH] The same observation applies to Assamese INF clauses. The structure in (5 7 a) includes two NPs, one in the subject position ( Ram Or jibon tu position ( kam aru pois a as sentences (5 7 b c) illustrate. T he INF clause employed in (5 7 b) functions as an adjunct in ( 52 a) above Assuming that only goals can fill the locus of a subject, we can deduce that Ass amese INF clauses are goals. (57) Assamese a. [ NP Ram Or jibon tu] [ NP kam aru poisa] hoi gois e [ NP Ram GEN life CL] [ NP work and money] be go 3 b. [Ram Or bhal lag a r karon tu] [Ram GEN good feel INF GEN becaus e CL] tar ghoniyak hoi goise his wife be go 3 c. [Ram Or dur goti ho a r karon tu] [Ram GEN bad phase be INF GEN because CL]

PAGE 230

230 tar jua khela hoi gois e his gambling playing be go 3 C oncerning CNP clauses, the prediction is that they do not qualify as goals which is why the movement of the subject is obligatory This prediction is also borne out. The sentences in (5 8 ) and (5 9 ) are the equivalent of (5 6 a b) and (5 7 b c) above, except that the INF clauses are replaced by CNP clauses. The result is ungrammaticality. (58) Telugu a. *popkorn tina Daaniki sari ayina samayam anTee popcorn eating for proper happening time means [cinimaa c uus tuu( e)] [movie watch CNP( EMPH)] b. *kaafi taaga Daaniki sari ayina samayam anTee coffee drinking for proper happening time mean [pani ki weLL i( e)] [work to go CNP( EMPH)] (59) Assamese a. *[Ram Or bhal lag i ( tu)] tar ghoniyak hoi goise [Ram GEN good feel CNP( CL)] his wife be go 3 b. *[Ram Or dur goti ho i( tu)] [Ram GEN bad phase be CNP( CL)] tar jua khela hoi gois e his gambling playing be go 3 T he sentences in (5 8 5 9 ) suggest that CNP clauses in Telugu and Assamese are non goals. One can argue that they qualify as goals in different structures. According to Masica (2005: 127), the only other structures that a Telugu and Assamese CNP type clause/verb can par ticipate in are serial verb

PAGE 231

231 constructions as exemplified in ( 60 61 ). These constructions are different from the Adjunct Control structure s examined in this dissertation in that the CNP type ver b and the finite verb behave like a mono clausal predicate, exp ressing a single event (Aikhenvald 2003: 1 4). The two are similar, however, in that the CNP part of a serial verb construction and the CNP clause in an Adjunct Control structure can never be realized with a (pronounced or implied) subject that is not core ferential with the subject of the matrix clause. In other words, they both are non goals, and both have to be syntactically saturated by the matrix subject. In this sense, they satisfy the Vehicle Requirement on Set and Pair M erge by obeying restriction (5 0c ): goal, it 13 (60) Telugu Kumar nawwu tuu maaTLaaDataa Du Kumar.NOM smile CNP talked 3.M.S (61) Assamese Ram e ghortu sa i thakil e Ram NOM house watch C NP kept 3 In this section, I built on work by Pesetsky and Torrego (2006) and Rothstein (2001) to suggest that the CNP subject moves in order to compensate for a defect in the adjunct that hosts it. A CNP clause in Telugu and Assa mese is defective in the sense that it do es not qualify as a goal To make up for this defect, the subject moves to the matrix clause. As a result, the same NP that predicates the CNP clause also belongs to the matrix clause. In this way, the CNP clause be comes syntactically saturated by an element in the matrix clause thus satisfying ( 50c ). 13 See Post 2006 for an analysis of serial verbs in Assamese, Steever 1988 for a discussion on serial verbs in Dravidian languages, and Joseph and Zwicky 1990 and Aikhenvald and Dixon 2003 for two collections of articles on serial verb constructions.

PAGE 232

232 The immediate advantage of this approach is that it divorces control from any features or needs of the subordinate subject (or of the target). This explains why the C NP subject is free to be pronounced (at least in Telugu) when the phonological component allows it. On a larger scale, it explains why we do not see (Adjunct) Control structures with the controllee/subordinate argument as an object. In principle, if sidewa rd movement allows the subject to copy out of one clause and merge into another, it should allow the object to do the same. 14 The reason why this does not happen should be clear by now. The non goal adjunct has to be predicated of an element in the matrix c lause in order to qualify for merge. To do this, the CNP clause has to share the NP that syntactically saturates it. This NP is the subject; it cannot be the object. Two points are in order. First, note that the Vehicle Requirement on Set and Pair Merge do es not exclude the possibility of movement that takes place primarily for the purpose of feature checking. For example, the movement of the subordinate subject in Adjunct Control structures that involve INF clauses takes place when the numeration is exhaus ted in order to check the theta role feature on matrix v 0 This kind of movement is triggered by Enlightened Self Interest rather than the Vehicle Requirement on Set and Pair Merge. In addition, the formulation in (50c) does not explicitly state what exac tly happens when a non goal (e.g., a CNP clause) is predicated of a constituent in the syntactic object it merges with (e.g., the matrix clause). In other words, how does predicating the CNP clause of an element in the matrix clause prepare the CNP clause for an instance of merge ? 14 Here it might be argued t hat movement happens through an escape hatch at the edge of the Command Unit, and that the subject is closer to the escape hatch than the object. While this might be true, it is not clear that all sideward escape hatch. Let us assume for a moment that no escape hatch is needed. Given that distance is quite blurry in instances of sideward movement, the choice of a subordinate subject over a subordinate object as a co referential argument in Adjunct Control mu st be restricted by factors other than distance.

PAGE 233

233 Within Predication Theory, the CNP clause would be considered as an open constituent indirectly predicated of an element in the matrix clause through co indexation. Within the movemen t approach adopted here the CNP clause is a pr oposition rather than an open constituent. Still, it is indirectly predicated of an element in the matrix clause. The paradox of this statement is resolved by movement. A propositional CNP clause can be indirectly predicated of an element in the matrix cla use only if the subject that saturates it moves to the matrix clause. Rather informally, I speculate that through movement the non goal CNP clause acquires a dual status. It is a proposition as far as interpretation is concerned. At the same time, it is an open constituent as far as merge is concerned. Assuming that CNP clauses are IPs, the CNP clause will be an IP for the purpose of interpretation, but it will be viewed as I' for the purpose of merge. We know that I' of the CNP clause is a goal because it may undergo merge with a subject NP. Admittedly, this is not the most attractive solution, but it does answer one crucial question: What does indirect predication do to the CNP clause so that it becomes legitimate for an instance of merge? Another question is: Does the Vehicle Requirement on Set and Pair Merge apply to control in languages other than Telugu and Assamese? A comprehensive answer is too ambitious at this point, but an attempt of an answer is viable. 5. 6 Adjunct Control in English and the Vehi cle Requirement on Set and Pair Merge The Vehicle Requirement on Set and Pair Merge in ( 50 ), repeated here as (6 2 ) may provide an explanation to the dilemma of Adjunct Control in English. (62) The Vehicle Requirement on Set and Pair Merge a. b. goal for some probe

PAGE 234

234 c. a non an d. A (potential) goal is a syntactic object with some feature F. A non goal is a syntactic object with no features. adds, Control into adjuncts is no t simple [Obligatory Control], contrary to what is frequently claimed (Mohanan 1983, Borer 1989, Clark 1990, Hornstein 1999). Implicit agents do control into these adjuncts and for clause initial adjuncts, a grammatical controller is not even obligatory. M oreover, it is also not generally the case that when a controller is logophoricity, an observation originally made by Williams (1992). Once logophoricity is separat ed from subjecthood, one can easily get non subject controllers with adjuncts. (178) 1992), Adjunct Control is divided into two categories: Logophoric and Predicative Th ese are exemplified in (6 3) and (64) respectively (Williams 1992: 300 307; (10) and (29a b) ) In (63 a b), ome such, whose thoughts or feelings ar e Sentence (63 a) shows that even without an implicit argument, the 300). Sentence (6 4 ), on the other hand, involves an adjunct that is c commanded by the antecedent. To Williams, such sentences do not involve PRO at all, b ut rather theta role assignment. T role it needs to assign, and it assigns it to John 1992: 29 7). (63) a. Having traveled all day, the hotel was a vision indeed. b. c ontrol, when listening to Larouche. (64) John loses control of his fears when listening to Larouche.

PAGE 235

235 The adjuncts in all three sentences (63a b, 64) are CPs; (alt hough the adjunct in (6 3 a) does not have an overt complementizer, it certainly can take one; e.g., after having travelled all day ). CPs, at least in English, are goals. This means that a CP adjunct does not have to be predicated of an element in the matrix clause in order to undergo merge. Therefore, movement for the purpose of merge is not obligatory (although it might take place for other reasons such as satisfying the thematic requirement of the matrix predicate ). In this case, it may be argued that the adjuncts in (63a b) involve pro similar to the pro licensed in pro drop languages T his pro is [+Topic Oriented], referring to a non commanding antecedent, an NP that is mentioned or implied earlier in discourse (Kawasaki 1993 : Chapter 5 ). 15 15 Kawasaki, building on Borer 1989, holds that English c ontrol structures in general (these also include control into complements) involve pro as a subordinate subject. Pro is licensed by non finite T 0 insid e a CP complement. Its content is identified by a functional head Agr. If C 0 is empty, Agr raises to CP and inherits th e phi features of an antecedent. These phi features are later copied to pro Concering control into adjuncts like sentence (i), a napho r i c Agr raises to C 0 leaving a trace behind. It inherits the phi features of John and copies them to pro (Kawasaki tacitly considers the adjunct a PP with a CP complement.) (i) [ CP [ IP John i felt old [after [ CP Agr i [ IP pro i t i seeing himself i in the mirr or ]]]]] In sentences like (ii iii), on the other hand, Agr cannot be co indexed with a c commanding antecedent. This is why it is [+Topic Oriented] rather than [+Anaphoric]. (ii) Suddenly the pirates i sh owed up from behind the rocks. [After [ CP Agr i [ IP p ro i robbing the passengers ]] ], the ship was sunk. (iii) [After [ CP Agr i [ IP pro pitching the tents ]]] darkness fell quickly (adopted from Kawasaki 1993: 172 174 (23a) and (24)) Of course, it remains to be determined why the logophoric or topic oriented i nterpretation is missing in (iv). For this reason, I consider structures like those in (i iv) and (63 64) to involve movement which takes place only when the numeration is exhausted and the thematic requirements of the matrix predicate have not been satisf ied yet. (iv) Telugu a. i/*k Koopamu wacc ina anduku ] anger come INF because] Sarita i akkadi nunci wellipoyinaa di Sarita.NOM there from left 3.N.S b. i/*k sinima cuus tunna appuDu] Sarita i popkorn tinnaa di movie wa tch INF while] Sarita.NOM popcorn ate 3.N.S

PAGE 236

236 The discussion does not mean to offer a solution to Adjunct Control in English I t only means to show that the Vehicle Requirement on Set and Pair Merge may be able to explain why English Adjunct Control does not display the consistency that Adjunct Control into CNP cla uses in Telugu and Assamese does. Assuming that CNP clauses are IPs, t he discussion thus far seems to point out that CPs are goals, but IPs Assuming that CPs are phases while IP s are not, it is tempting to redefine 2 ) in terms of phase s (and heads). While this is a reasonable assumption, I remain agnostic about it, especially that evidence from clausal gerunds (Pires 2005) shows that a syntactic object might not be a phase (or a head) but still be a goal. Consider the clausal gerund in (6 5 a). Pires (2005: 44 47) convincingly argues that clausal gerunds are not CPs. For example, they do not allow an overt complementizer (6 5b). Compare to (65 c) where a to infinitive is used and an overt complementizer is allowed. In addition, clausal geru nds do not allow indirect questions with short wh movement (6 6b). Compare to (66 c) where a to infinitive is involved and a short wh movement is allowed. Accordingly, Pires deduces that clausal gerunds are IP s. (65) a. Tom prefers [going to work by bus]. b. *T om prefers [for Sue going to work by bus]. c. Tom prefers [for Sue to go to work by bus]. (66) a. Tom forgot [calling Sue]. b. *Tom forgot [who calling]. c Tom forgot [who to call]. Despite being IPs, c lausal gerunds qualify as goals Pires (2005: 35) ob serv es that clausal gerunds only appear in Case positions. For example, they occupy the subject position in passive constructions ( 67 a). Also, th ey can be complements in PPs ( 67 b). (67) a. [Going to work by bus] was preferred by Tom. b. Tom is keen on [calling Sue ].

PAGE 237

237 Since clausal gerunds are go als, the Vehicle Requirement on Set and Pair Merge predicts that they do not have to share their subject in order to undergo merge. That is, movement for the purpose of merge is not required Therefore, if Englightened Self Interest does come into play, we should expect structures that consist of a clausal gerund to allow disjoint subjects. Th is prediction is correct, as (68 a b) sh ow (Pires 2005: 35 36 (14a, 15b)) (68) a. [Frank reading this book] was preferred. b. Mary talked abo ut [John moving out]. It might be that the deverbal/ nominal quality of the verb is what qualifies clausal gerunds as goals. Accordingly, whether an IP may be inherently a goal or not remains an open question. 5.7 C onclusion This c hapter set off with one purpose, to find out why the CNP subject in Telugu and Assamese Adjunct Control structures moves to the matrix clause. After examining a number of possibilities from the literature, I came to the conclusion that the CNP subject moves in order to make up fo r a defect in the CNP clause. Telugu and Assamese CNP clauses are non goals. 16 Goals are bearer of feature s that license their merg e. Non goals do not have such feature s In order to undergo merge, they have to be predicated of an element in the syntactic o bject that they merge with. This idea is formalized as the Vehicle Requirement on Set and Pair Merge in (72). (69) The Vehicle Requirement on Set and Pair Merge a. feature F on b. goal for some probe c. a non d. A (potential) goal is a syntactic object with some feature F. A non goal i s a syntactic object with no features. 16 Note that the movement of the CNP subject in Assamese may be triggered by Structural Case as well.

PAGE 238

238 As non goals, Telugu and Assamese CNP clauses have to satisfy the requirement in (72c). Within Predication Theory, the requirement is satisfied through co indexation. If control is movement as the evidence from Backw ard and Copy Control seems to indicate we can conclude that a non goal is licensed for merge through the movement of its subject to the matrix clause. 17 Further, t he Vehicle Requir e ment on Set and Pair Merge seems to explain why Adjunct Control in English is not as consistent as Adjunct Control in Telugu and Assamese. The reason is that English adjuncts are CP goals. Thus, their merge with the matrix clause does not have to involve movement. I n Telugu and Assamese CNP clauses are non goals. They need to b e syntactically saturated by an element in the matrix predicate in order to be licensed for an instance of merge. When the CNP subject becomes an NP of the matrix clause, it also becomes the licenser. No such licenser is needed for the merge of English ad junct CP s I close with an afterthought : Does the Vehicle Requirement on Set and Pair Merge resolve the current controversy in control theory? goal characteristics of subordinate clauses is all that there is to con trol theory. And assuming that goals which is not true will leave this as an open que stion, hoping that the proposal I have made here has solved at least a part of the puzzle. 17 It would be interesting to test this requirement a gainst the Japanese data in Dubinsky and Hamano 2006. Nevertheless, testing Japanese ni and te adjuncts to determine whether they are goals or non goals is no trivial task. These particles are all over the place and have multiple functions. For example, te may be used as postposition or an imperative marker. I do not feel that I have enough knowledge about the particles to make an informed observation.

PAGE 239

239 CHAPTER 6 EXCEPTIONS TO ADJUNC T CONTROL AS NON EXCEPTIONS 6.1 Introduction C hapters 3 and 4 presented an analysis of Telugu and Assamese Adjunct Control structures in which Obligatory C ontrol obtains all the time. Exceptions do exist, however. These are structures that pattern the same as Adjunct Control structures (i.e. they involve a matrix clause and a CNP clause), yet they allow disjoint subjects. Sentences (1 a b) and (2a b) are examples. The main purpose of this chapter is to show that the se structures may be analyzed as instances of Obligatory Control and that they do not present a challenge to the analysis of Adjunct Control offered in the previous chapters (1) Telugu a. [bombu pel i] caala mandi canipoyaa ru [bomb.NOM explode CNP] many people.NOM died 3.M.P b. [war am paD i] cetlu/mokkalu periga yi [rain.NOM fall CNP] trees/plants.NOM grew 3.N.P (2) Assamese a. [e ta ghor ot zui lag i] bohut manuh moril [one CL house LOC fire ABS happen CNP] many people ABS died. b. [dhumuha ah i] bohut gos bhangil [storm ABS come CNP] many trees ABS broke Before proceeding, it is important to note that the phenomenon is not unique to Telugu a nd Assamese. As the following examples show, similar structures are allowed in Marathi (Pandharipande (1997: 446 (1277)), Tamil (Linholn 1975: 81 (3.38)), Hindi (Davison 1981: 122 fn. 5 (i)), and Bengali (Klaiman (1981: 114 (4.57e)), among other (South Asi an) languages.

PAGE 240

240 (3) Marathi [pa auus paD uun] dhaanya pikl a [rain fall CNP] crops grew 3.N.S (4) Tamil [maze penj u] aatu le taNNi ooduccu [rain fall CNP] river LOC water ran (5) Hindi [Diwaar gir kar] patthar gir gaee [wall fall CNP] stones fell went (6) Bengali [ce aar bheNge giy e] Modhu pore gaelo [chair break down CNP] Modhu fell down Davi son (1981: 122 fn. 5) consid ers these structures mysterious. A hard to see exactly what factors must be present for the like subject condition [or control] to not tri bute them purely to semantic factors without any reference to syntax. Pandharipande (1997: 445 446) briefly indicates that such structures are allowed when there is a cause effect relation ship between the e agents of the matrix and the participial with more details Linholn (1975) attributes the occurrence in part to a cause effect relation between the matrix and the subordinate clauses, and he adds a nother fa ctor whi According to natural relevance, it is not enough to have a cause effect relation between the CNP and the matrix clauses; the relations must also follow naturally. For example, the CNP and the matrix cl auses in (7) exhibit a cause effect relation, but the sentence is ungrammatical because the relation lacks natural

PAGE 241

241 relevance (Linholn 1975: 80 (3.37)). Compare to (6) where the relation between rain and the flowing of the river is a cause effect relation t hat is naturally relevant (Linholn 1975: 75 83). (7) Tamil *[ maze penj u] kaDe le ellaam koDe vittu pooccu [rain fall CNP] shop LOC all umbrella sell went becomes problem atic when examined in the light of (8) and (9). Examples (8a) and (9a) are a repetition of (1a) and (2a) respectively. The sentences indicate that some disaster happened leading to the death of a lot of people. The cause effect relation between the two inc although the idea that bomb explo suspect) and, as expected, b oth sentences are grammatical. If this analysis is correct or sufficient one would expect (8b) and (9b) to be grammatical as well. The y also indicate that some disaster happened leading to a sad outcome. The only difference is that (8b) and (9b) mention the agent behind the disaster, which is apparently why the sentences are ungrammatical. (8) Telugu a. [bombu pel i] caala mandi canipoya a ru [bomb.NOM explode CNP] many people.NOM died 3.M.P b. [Kumar bombu ni pelc i] caala mandi canipoya ru [Kumar.NOM bomb ACC explode CNP] many people.NOM died 3.M.P (9) Assamese a. [e ta ghor ot zui lag i] bohut manuh moril [one CL house LOC fire ABS happen CNP] many people ABS died. b. [Ram e ghor to t zui laga i] [Ram NOM house CL LOC fire cause to h appen CNP] bohut manuh moril many people ABS died

PAGE 242

242 (1981) analysis is a possible solution to the problem in (8) and (9) She holds that structures like (8a) and (9a) are allowed only when both the CN P and the matrix clauses express a non volitional activity. If one of the clauses expresses a volitional activity, disjoint subjects result in ungrammaticality. This is ex actly the case of (8b) and (9b). I n both structures, the CNP clause expresses a volit ion activity, which is why the sentences are unacceptable. The same is true if the matrix clause expresses a volitional activity, as (10 11) illustrate. (10) Telugu [bombu pel i] ambulens waccin di [bomb.NOM explode CNP] ambulance.NOM came 3.N.S (11) Assamese [e ta ghorot zui lag i] [one CL house LOC fire happen CNP] bohut manuh e police aloi phone koril e many people.NOM police DAT phone did 3 s analysis is purely seman t i c. She explicity rules out syntax and the possibility may be translated in syntactic terms without undermining the semantic natur e of the account. The following section s set out to do this and to show that what appears to be an exception to Adjunct Control is not an exception. Section 6. 2 re non volitional distin ction as accusative/unaccusative. Section 6. 3 suggests that the structures in (1 2) above are instances of Expletive Control that are possible only when the CNP and the matrix predicates are both unaccusative. This idea becomes problematic when examined in the light of English which does not allow similar patterns of Expletive Control. Section 6. 4 proposes an explanation and shows that the difference with regards to Expletive Control between English and the two South Asian languages under examination follow s from the Vehicle Requirement on

PAGE 243

243 Set and Pair Merge as proposed in the previous chapter. The conclusion summarizes the main points of this chapter and highlights aspects of Expletive Control that are left unexamined. 6.2 Non Volitional as Unaccusative A c loser look at the sentences in the previous section shows that what Klaiman describes as non volitional activities correspond in the syntax to unaccusative structures. Each of the grammatical sentences in (1 6) contains two unaccusative predicates, one in the CNP clause and one in the matrix clause. By comparis o n, the ungrammatical structures (8b), (9b), and (10 11) contain at least one clause that is not unaccusative. 1 1 mean, not only transitive and unergative, but also experiential predicates. F or example, the sentences in (i) and (ii) are ungrammatical because each contains one experiential predicate. Note that (ia) is ruled out for an independent reason; Telugu does not allow a matrix dative subject in Adjunct Control structures. I include (ib c) in which the matrix experiential predicate licenses a nominative subject. The sentence is still ungrammatical. (i) Telugu a. *[bombu pel i] kumar ki koopam waccin di [bomb.NOM explode CNP] Kumar DAT anger came 3.N.S b. *[Kumar ki jwaram wacc i] Sarita baada paDin di [Kumar DAT fever come CNP] Sarita.NOM sad felt 3.N.S Kumar had fever; Sarita was sad. c. *[ammaayi putt i] andaru santoo incaa ru [girl.NOM born CNP] family.NOM became happy 3.M.P A girl was born; the family was happy. (ii) Assamese a. *[e ta ghor ot zui lag i] Ram Or khong uthil [one CL house LOC fire ABS happen CNP] Ram GEN anger raised b. *[Ram Or khong uth i] bohut manuh moril [Ram GEN anger raise CNP] many people ABS died. am got angry This is an important point because Kl volitional predicates seems to include experiential predicates. She presents the two examples (iii iv) which include one experiential predicate in the matrix clause and two disjoint subjects (Klaiman 1981: 113, (4.55a b)) As far as I know, non e of the grammatical examples in her study include a CNP experiential pred icate and two disjoint subjects. (iii) Bengali a. [Taeks bere giy e] aneke r kasTo hoyece [tax increase CNP] many GEN difficulty became

PAGE 244

244 The standard assumption is that unaccusative structures license themes that are base generated low in the structure. Themes, along with goals and patients, are considered the lowest of all arguments. T hey are generated below causers, which in turn ar e generated below experiencers (Landau 2001: 120 and works within) This im plies that the non volitional subjects in (1 6) are themes that are generated low in the structure, probably as complements of V 0 It is desirable to have independent evidence that the unaccusative predicates under investigation comprise themes that are r ealized low in the structure, probably in the locus of their first merge. E vidence comes from structures that contain an unaccusative predicate and a locative expression. Although Telugu and Assames e are SOV languages, with the subject canonically accupyin g a sentence initial position (e.g., (12a) and (13a)) i f an unaccusative predicate is involved the locative expression is realized sentence initially (e.g., (12b c) and (13b c)). These examples are not unexpected, given that the locus of locative express ions is higher than the locus of themes (Grimshaw 1990: 24). This said, it is important to note that, owing to the free word order in Telugu and Assamese, u naccusative structures with a sentence initial theme followed by a locative expression are also acce ptable (e.g., (12d e) and (13d e)) (12) Telugu a. Kumar maa uuri loo bombu ni pelcaa Du Kumar.NOM my town LOC bomb ACC exploded 3.M.S b. maa uuri loo caala mandi canipoya ru my town LOC many people died 3.M. P b. [brisTi por e] caaside r laabh holo [rain fall CNP] farmers GEN profit became The analysis offered in this chapter accounts for the Telugu and Assamese data. Concerning th e Bengali examples in (iii), I do not have an explanation.

PAGE 245

245 c. pollalu loo cetlu/mokkalu periga yi field LOC trees/plants.NOM grew 3.N.P In the field t d. caala mandi maa uuri loo canipoya ru many people my town LOC died 3.M.P M any people in my tow n, c. cetlu/mokkalu pollalu loo periga yi trees/plants.NOM field LOC grew 3.N.P T he trees/plants in the field, (13) Assamese a. Ram e ghorto t zui lagal e Ram NOM house LOC fire caused to happen 3 b. prithibi t xanti ahil world LOC peace ABS came c. Guwahati t bohut gos bhangil Guwahati LOC many trees ABS broke d. xanti prithibi t ahil peace ABS world LOC came e. boh ut gos Guwahati t bhangil many trees ABS Guwahati LOC broke This obs ervation extend s to the structures in (1) and (2) above, repeated here with locative expressions. (14) Telugu a. [maa uuri loo bombu pel i] caala mandi canipoya ru [my town LOC bomb explode CNP] many people died 3.M.P b. [pollalu loo war am pad i] cetlu/mokkalu periga yi [fields LOC rain.NOM fall CNP] trees/plants.NOM grew 3.N.P

PAGE 246

246 (15) Assamese a. [e ta ghor ot zui lag i] bohut manuh moril [one CL house LOC fire ABS happen CNP] many people ABS died. b. [Florida t dhumuha ah i] bohut gos bhangil [Florida LOC s torm come CNP] many trees broke The following section shows how accusativity is releva nt to control or the lack of it. 6.3 Unaccusative Predicates and Expletive Control I consider structures that involve unaccusative predicates in the CNP and the matrix clauses as having null expletive s as subjects In other words, the sentences in (1 2) have the structure in (16 17), with the subject positions in each sentence being occupied with pro EXP (16) Telugu a. [pro EX P bombu pel i] [pro EXP bomb.NOM explode CNP] pro EXP caala mandi canipoyaa ru pro EXP many people.NOM died 3.M.P b. [pro EXP war am paD i] pro EXP cetlu/mokkalu periga yi [pro EXP rain.NOM fall CNP] pro EXP trees/plants.NOM grew 3.N.P (17) Assamese a. [pro EXP e ta ghor ot zui lag i] [pro EXP one CL house LOC fire ABS happen CNP] pro EXP bohut manuh moril pro EXP many people ABS died. b. [pro EXP dhumuha ah i] pro EXP bohut gos bhangil [pro EXP storm ABS come CNP] pro EXP many trees ABS broke The structures in (16 17) are based on the assumption that the theme maintains its position low in the clause, allowing an expletive to fill the subject position. The expletive is null because

PAGE 247

247 neither Telugu nor Assamese has overt expletives, whic h is expected in pro drop languages in general. This idea is confirmed by Subbarao and Murthy (1999: 217) who maintain that Telugu it or there Similarly, Rao (2002: 37 39) gu are obligatorily null (e.g., (1 8a d ) ungrammatical if th e null expletive is lexicalized (e.g., (19a d ) ) (Sentences (18a c) and (19a c) are from Rao 2002: 37 38 (17 18) ) 2 (18) Telugu a. pro EXP iiroozu callagaa un di. pro EXP to day cold is 3NS b. pro EXP occe nela veDigaa unTun di. pro EXP coming month hot will be 3NS c. pro EXP akkaDa ciikaTigaa unDin di. pro EXP there dark was 3NS d. kolkata loo oka illu ammak aaniki un di Calcutta LOC one house sale for is 3.N.S 2 This chapter does not try to prove that null expletives are a reality. It simply assumes that they exist. For a convincing argument that null expletives are a psychological reality, see Oshita ( 2004 and works within). Another piece of evidence comes from languages in which null expletives may be phonologically realized as clitics. In Standard Arabic, for example, null expletives trigger agreement on the verb, which is normally the default 3.M.S, as (i) shows. Notice that a pre verbal subject does not trigger full agreement on kaan a waaDiH an would normally do with other non raising verbs/expression. Now observe (iii) in which the null expletive takes on a phonological form. Thi s is possible only when the null expletive is in the vicinity of an element that may host a clitic (e.g., anna). (i) pro EXP kaan a waaDiH an anna l awlaad a fariH uun pro EXP was 3.M.S clear ACC that the children ACC happy NOM.3.M.P It was clear that the children were happy. (ii) l awlaad u pro EXP kaan a waaDiH an anna hum fariH uun the children NOm pro EXP was 3.M.S clear ACC that 3.M.P happy NOM.3.M.P It was clear that the children were happy. (iii) qult u anna hu kaan a waaDiH an anna l awlaad a fariH uun said 1.S that 3.M.S was 3.M.S clear ACC that the children ACC happy NOM.3.M.P It was clear that the children were happy.

PAGE 248

248 (19) Telugu a. *adi iiroozu callagaa un di. It.NOM today cold is 3NS b. *adi occ e nela veDigaa unTun di. It.NOM coming month hot will be 3NS c. *adi akkaDa ciikaTigaa unDin di. It.NOM there dark was 3NS d. *adi kolkata loo oka illu ammak aaniki un di it.NOM Calcutta LOC one house sale for is 3.N.S Expletives are obligatorily n ull in Assamese as sentences (20 a d ) suggest (Jason 2004, sentences (23), (89), (110), and (175)) If an overt pronominal i s used as an expletive, the sentences become ungrammatical ( e.g., (21a d) ) (20) Assamese a. pro EXP ghoriyal hati xap aru goru as e pro EXP crocodile elephant snake and cow are 3 b. pro EXP xei dex at bhekuli dzura eta as e pro EXP that country LOC frog pair one was 3 c. pro EXP Sita nam Or e dzoni dukhi mohila asil e pro EXP Sita name GEN one CLS.F sad woman was 3 d. pro EXP gos at moumakhi r bah asil e pro EXP tree LOC honeybee GEN nest was 3 (21) Assamese a. *xi/tai ghoriyal hati xap aru goru as e s/he.NOM crocodile elephant snake and co w are 3

PAGE 249

249 b. xi/tai xei dex at bhekuli dzura eta as e s/he.NOM that country LOC frog pair one was 3 c. xi/tai Sita nam Or e dzoni dukh i mohila asil e s/he.NOM Sita name GEN one CLS.F sad woman was 3 d. xi/tai gos at moumakhi r bah asil e s/he.NOM tree LOC honeybee GEN nest was 3 In addition, I consider the apparent exceptions in (16 17) above as instances of Expletive Control with the subordinate and the matrix null expletives as copies of the same token. In other words, I consider the structures to have a derivational history that is similar t o the more common instances of Adjunct Control analyzed in the previous chapters, except for one major difference: whereas the two non distinct subjects in the more common Adjunct Control structures are copies of the same argument, the non distinct subject s in Expletive Control structures are copies of the same expletive. For example, structures (22a) and (23a) have the derivational history presented in (22b) and (23b) respectively. In both structures, the CNP clause and the matrix clause form independently Notice that the thematic domain in both clauses is an unaccusative VP. The subject of the CNP clause, pro EXP undergoes sideward movement. I t copies out of the CNP clause and merges in Spec,IP of the matrix clause. Following, the CNP clause adjoins to th e matrix clause at VP (or CP) and the structures converge as (22c) and (23c). 3 (22) Telugu a. [war am paD i] mokkalu periga yi [rain.NOM fall CNP] plants.NOM grew 3.N.P 3 In (22b, 23b) and in the rest of this chapter, I used the terms SUBJ and subject position for any NP/DP (argument or expletive) whose function is to syntactically saturate the predicate in the sense o f Rothstein 2001.

PAGE 250

250 b. c. CP qp IP C qp SUBJ I' pro EXP qp V P I qp CNP P V P 3 6 SUBJ VP mokkalu periga yi pro EXP 6 war am paD i (23) Assamese a. [dhumuha ah i] bohut gos bhangil [storm ABS come CNP] many trees ABS broke b. CNP P 3 SUBJ VP pro EXP 6 war am paD i IP qp SUBJ I' pro EXP qp VP I 6 mokkalu periga yi S IDEWARD M OVEMENT CNP P 3 SUBJ VP pro EXP 6 dhumuha ah i IP qp SUBJ I' pro EXP qp VP I 6 bohut gos bhangil S IDEWARD M OVEMENT

PAGE 251

251 c. CP qp IP C qp SUBJ I' pro EXP qp V P I qp CNP P V P 3 6 SUBJ VP bohut gos bhangil pro EXP 6 dhumuha ah i The sideward movement of the CNP subject to the matrix clause is made obliga tory by the Vehicle Requirement on Set and Pair Merge in (24), a repetition of (50) in Chapter 5. We saw in Chapter 5 that the CNP clause is not a goal. According to (24c), it can merge with the matrix clause only if the CNP predicate is syntactically satu rated by an element in the matrix predicate. Nevertheless, the CNP predicate is already syntactically saturated in accordance with the defini tions of predicate and subject in (25 26). Assuming that the CNP clause is an IP, I' as a predicate is internally p redicated of the non argument subject pro EXP This is why, as a non goal, the CNP clause may satisfy the Vehicle Requirement on Set and Pair Merge only if it shares its subject with the matrix clause. In this way, the same element that saturates the CNP cl ause will be an element in the matrix clause. Note that this requirement is independent of the thematic content or lack of content of the CNP subject. Whatever saturates the CNP predicate moves in order to license the merge of the CNP clause with the matrix clause. (24) The Vehicle Requirement on Set and Pair Merge a. b. goal for some probe c. non

PAGE 252

252 d. A (potential) goal is a syntactic object with some feature F. A non goal is a syntactic object with no features. (25) A predicate is an open constituent/a non proposition. It has to be predicated of or syntactically saturated by a subject in order to become a maximal constituent/a proposition. A predicate may take one of the following forms: a. A maximal projection (e.g., a VP) : When syntactically saturated, this kind o f predicate is externally predicated of a subject. That is, the predicate and its subject form a proposition whose properties are not dictated by the head of the predicate (e.g., [ vP Subject [ VP predicate]]; or b. I' : This is the only predicate that is a non maximal projection. When syntactically saturated, I' is internally predicated of a subject. In other words, I' and its subject IP Subject [ I' I 0 ]). (Roth stein 2001: 19, 47 49 118 120) (26) A subject is a non predicate. It is a structural construct that may take one of the following forms: a. A n argument ; for example, an agent functioning as a subject of a transitive predicate, or an experiencer functioning as a subject of a psych predicate b. A non argument; for example, an expletive functioning as a subject of an unaccusative or a raising predicate (Roth stein 2001: 18 20, 60 65) If the CNP predicate or the matrix predicate is not unaccusative, the result is ungrammaticality, a s the examples in (8 11) above illustrate. Here is why. Observe the Telugu example in (27a). The CNP predicate is transitive, with Kumar as an external argument/subject. In order to satisfy (24c), the CNP clause must have its subject move to the matrix pre dicate. As (27b) shows, by the time Kumar undergoes sideward movement, the only available position is Spec,IP. This is not an appropriate landing site for the argument Kumar for reasons to be specified in the following section. The movement of the CNP s ubj ect is not optional, however. I t has to take place in order to license the merge of the CNP clause. In this case, sidewa rd movement fails to take place. C onsequently, the structure does not converge. Note that the structure would converge if the CNP subjec t merges in the theme position in the matrix clause, as (27c) shows. The result is (27d).

PAGE 253

253 (27) Telugu a. *[Kumar bombu ni pelc i] [Kumar.NOM bomb ACC explode CNP] caala mandi canipoya ru many people.NOM died 3.M.P b. c. d. [Kumar bombu ni pelc i] [Kumar.NOM bomb ACC explode CNP] atanu/ aa pichooDu/ Kumar canipoyaa Du he.NOM/ that idiot.NOM/Kumar.NOM died 3.M.S Kumar Conversely if the matrix predicate is transitive (or unergative) while the CNP predicate is unaccusative, as exemplified in (28a), the nul l expletive pro EXP in the CNP clause undergoes sideward movement to Spec,IP of the matrix clause as (28b) illustrates. Yet, the derivation crashes It suffices to say or stipulate that neither Assamese nor Telugu allows transitive exple tive construc tions of the type attested in Icelandic. In this sense, the argument in Spec,vP of CNP P 3 SUBJ vP Kumar 6 Bombu ni pelc i IP qp SUBJ I' qp VP I wo THEME V caala mandi canipoya ru S IDEWARD M OVE MENT CNP P 3 SUBJ vP Kumar 6 Bombu ni pelc i VP wo THEME V Kumar canipoya a Du S IDEWARD M OVEMENT

PAGE 254

254 the matrix clause ha This is why the sideward movement of the CNP expletive does not lead to a convergent derivation (28) Assamese a. [pro EXP e ta ghorot zui lag i] [pro EXP one CL house LOC fire happen CNP] bohut manuh e police aloi phone koril e many people.NOM police DAT phone did 3 b. It is worth noting that the theme of an unaccusative CNP predicate may raise to the subject position and syntactically saturate the CNP predicate. Accordingly, in the event of a transitive or unergative matrix predicate, t he theme of the CNP clause may it self move to the matrix clause ( unless, of course, the theme and the matrix predicate are semantically incompatible ) In this way, the derivation is salvaged, resulting in an Adjunct Control structure similar to the ones an alyzed in Chapters 3 and 4. Sentence (29a) is an example. As (29b) shows, war am starts out in the theme position of the CNP clause, after which it moves to the subject position intra clausally. Following, it undergoes sideward movement to Spec,vP of the matrix clause, thus licensing the merge of the CNP clause in accord ance with part (c) of the Vehicle Requirement on Set and Pair Merge in (24). CNP P 3 SUBJ 6 pro EXP e ta ghorot zui lag i IP qp SUBJ I' qy vP I 3 SUBJ V P bohut manuh e 6 police aloi phone koril e S IDEWARD M OVEMENT

PAGE 255

255 (29) Telugu a. [war am pad i] adi wiidhula ni subram ceesin di [Rain fall CNP] it streets ACC clean did 3.N.S b. The derivational history of the Expletive Control structures in (22) and (23) differs from that of the control structures analyzed in Chapters 2 and 3 in two more ways. These are related to the la nding site of pro EXP and to the late merge of the CNP clause. Sections 6. 3.1 and 6. 3.2 lay out the details. 6.3.1 Adjunct Control and the Target of Sideward Movement Consider the Forward Control structure in (30). As (30b) shows, when the CNP subject under goes sideward movement, it copies out of the CNP clause and merges in matrix Spec,vP. Compare to (22b) and (23b) where the CNP subject merges in Spec,IP of the matrix clause. (30) Telugu Kumar [ Kumar ki jwaram wacc i ] hospital weLLaa Du Kumar.NOM [ Kumar DAT fever come CNP ] hospital went 3.M.S b. CNP P 3 SUBJ 6 Kumar ki jwaram w acc i v P 3 SUBJ 6 Kumar hospital weLLaa Du S IDEWARD M OVEMENT CNP P 3 SUBJ VP war am 3 THEME V war am pad i vP 3 SUBJ VP war am 3 THEME V wiidhula ni subra m ceesin di S IDEWARD M OVEMENT

PAGE 256

256 This difference between (30) and (22 23) should not be a problem if the restriction on the landing site of sideward movi ng e lements formulated in (31), a re petition of (45) in Chapter 4, is correct. In (22) and (23), the CNP subject, pro EXP undergoes first merge in a phi position According to (31 ), when pro EXP undergoes sideward movement, it should land in the same position, which it does. Compare to (30 ) in which the CNP subject undergoes first merge in a thematic position W hen it undergoes sideward movement, its landing site is a thematic position. (31) undergoes sideward movement. A position X can be a thematic position, a phi position, or a discourse position. The restriction is bu ilt on the following assumption. Wh en an element moves intra clausally, it has to copy out of its position and merge in a higher c commandin g landing site that is a root (i.e. a landing site that is c commanded by no other node ) This is in accordance with the Extension Condition which hol extending the structure (Chomsky 1995: 248). When an element undergoes sideward movement, however, it obeys the extension condition without having to merge in a c commanding position. As a matter of fact, it cannot merge in a c commanding position (Hornstein and Kiguchi 2001: 11). Rather, the sideward moving element makes itself available in the computational workspace in the same way that an element in the numeration is available in the computatio nal workspace. I assume that it undergoes merge in the same way an element selected from the numeration undergoes merge. That is, they both target the same locus. The principle in (31) also explains why sen tence (27a) above, repeated as (32), is not a pos sible structure. As an argument, the subordinate subject Kumar undergoes first merge in a thematic position According to (31), it must also target a thematic position when it undergoes sideward movement. This explains why matrix Spec,IP is not an appropri ate landing site.

PAGE 257

257 (32) Telugu a. [Kumar bombu ni pelc i] [Kumar.NOM bomb ACC explode CNP] caala mandi canipoya ru many people.NOM died 3.M.P many b. Note that Case and theta role cannot account for the ungrammaticality of (32a). As I explained in Chapter 4, by the time the subordinate subject undergoes sideward movement, it has alr eady taken on Case and a theta role. Therefore, it has a need for neither. At the same time, under the view that multiple Case checking and multiple theta roles are possible, either Spec,vP or Spec,IP of the matrix clause must be an appropriate landing sit e for the sideward moving subject. Yet, only the former leads to a convergent structure. The restriction in (31) is able to explain why this is so. The following section deals with the issue of late merge that is brought on by Expletive Control. 6. 3.2 Exp letive Control and Late Merge In addition to Set vs. Pair Merge, merge can be further divided into cyclic merge and late merge. Cyclic merge obeys the Extension Condition in that it applies at the root of the structure, extending it and forming a new categ ory. For example, consider the structure M in (33 ). I f there max because CNP P 3 SUBJ vP Kumar 6 Bombu ni pelc i IP qp SUBJ I' qp VP I wo THEME V caala mandi canipoya ru S IDEWARD M OVEMENT

PAGE 258

258 this the only category that is not c commanded by any other category. After Merge applies, the structure is extended, yielding a new syntactic object L, (33) M = X max 3 X Y max 3 Y Z (34) L = max 3 X max 3 X Y max 3 Y Z h Y max in (33 ), r esulting in the structure in (35 )? Since Y max is not a root (i.e., it is c commanded by another category X), such an instance of merge is considered as counter cyclic merge, also known as late merge. (35) M = X max 3 X Y max 3 Y max 3 Y Z The Adjunct Control structures examined in the previous chapters only involve cyclic merge in the sense that all merge, including the merge of the CNP clause with the matrix clause, applies to the root. E xpletive Control, on the other hand, may involve late merge. Consider (23), repeated below as (36). By the time the CNP subject undergoes sideward movement in (36b), the matrix clause has already projected as high as IP. F ollowing, the CNP clause either ad joins to the m atrix clause at VP (36c) or waits till matrix C 0 projects and merge at CP (36d). The latter

PAGE 259

259 scenario is not a problem since no late merge is involved. In (36c), however, VP is no longer a root because it is dominated by IP. That is, the CNP c lause undergoes late merge. (36) Assamese a. [dhumuha ah i] bohut gos bhangil [storm ABS come CNP] many trees ABS broke b. c. IP qp SUBJ I' pro EXP qp VP I qp CNP P VP 3 6 SUBJ VP bohut gos bhangil pro EXP 6 dhumuha ah i d. CP qp CNP P CP 3 3 SUBJ VP IP pro EXP 6 3 dhumuha ah i SUBJ I' pro EXP 3 VP I 6 bohut gos bhangil CNP P 3 SUBJ VP pro EXP 6 dhumuha ah i IP qp SUBJ I' pro EXP qp VP I 6 bohut gos bhangil S IDEWARD M OVEMENT

PAGE 260

260 An easy way out of this dilemma is to assume that the CNP clause in Expletive Control structures can only undergo merge at CP of the matrix clause in order to avoid la te merge, and that (36d) is the only possible derivation. In this respect, Expletive Control structures will be similar to Copy Control structures. Nevertheless, considering that nothing crucial hinges on this assumption, and given that late merge will be needed for the discussion of English Expletive Control in the following section, I will continue to assume that (36c) is a possible derivation. and argues that all merge including the merge of adjuncts, is cyclic. The theoretical gain behind Everything being equal, a grammar with cyclic merge is better than a grammar with cyclic and late merge. On the empirical side, cancelling late merge fro m the grammar helps avoid unwanted occurrences of sideward movement like (37) below, which is an adjunct island violation (Nunes 2004: 116 118). (37) *Which man did you call your brother before meeting? In order to see how the violation is induced, let us assu me that late merge is allowed. The derivational history of (37) will be (38). The adjunct clause and matrix vP form independently in (38a). Matrix vP extends into IP in (38b) and then CP in (38c). At this point, the WH feature in matrix CP causes the wh ph rase [which man] to undergo sideward movement. I t copies ( 38d), and merges at Spec,CP ( 38e). Finally, the adjunct undergoes late merge at matrix vP, and the structure converges, contrary to fact. (38) a. i. [ P P before [ CP PRO meeting [which man]]] 4 ii. [ vP you call your brother] b. [ IP you [ vP you call your brother]] c. [ CP did [ IP you [ vP you call your brother]]] 4 Nunes labels the adjunct in (33a) as a PP and the unpronounced subject as PRO.

PAGE 261

261 d. [which man] e. [ CP [which man] did [ IP you [ vP you call your brother]]] f. [ CP [which man] did [ IP you [ vP [ vP you call your brother] [ P P before [ CP PRO meeting [which man]]] ]]]] In order to avoid such derivations, Nunes (2004: 119 121) maintains that all merge should be cyclic. In other words, if the adjunct in (38a i) is supposed to merge at matrix vP, it should do that right after step (38a ii). Consequently, the derivation proceeds as in (39). The adjunct and matrix vP form independently ( 39a), and undergo merge ( 39b). Matrix vP extends into IP ( 39c), and then CP ( 39d). The wh feature of matrix CP triggers the movement of [which man] in to Spec,CP. Nevertheless, the adjunct containing [which man] has already become an island upon merge. Movement out of an island is disallowed. Consequently, the wh feature of matrix CP does not get checked, and the structure crashes. Unlike late merge, cyc lic merge does not overgenerate ungrammatical structures. (39) a. i. [ P P before [ CP PRO meeting [which man]]] ii. [ vP you call your brother] b. [ vP [ vP you call your brother] [ P P before [ CP PRO meeting [which man]]] ]] c. [ IP you [ vP [ vP you call your brother ] [ P P before [ CP PRO meeting [which man]]] ]]] d. [ CP did [ IP you [ vP [ vP you call your brother] [ P P before [ CP PRO meeting [which man]]] ]]]] Let us assume for a moment that late merge is a necessity. This assumption may have a chance of survival if it can be supported with empirical evidence independent from Expletive Control and paras itic gap constructions, and if the problem in (37 38) can be solved without resorting to cyclic merge. Section 6. 3.2.1 presents evidence from the literature that late mer ge is needed on independent grounds. Section 6. 3.2.2 proposes a solution for (37 38) that does not require the banning of late merge.

PAGE 262

262 6.3.2.1 Late m erge: empirical e vidence Several researchers have argued that late merge of adjuncts is possible (e.g., Leb eaux 1988, 1991; Fox a nd Nissenbaum 1999 ) Stepanov 2001, building on work by Lebeaux (1988, 1991) and Chomsky (1995, 2000) among others, holds that adjuncts can only undergo late merge. He proposes an algorithm of phrase structure building that forces lat e merge of adjuncts. 41) (Stepanov 2001: 102 104 (16, 22)): (40) Least Tampering Given a choice of operations appl @ (X): a set of c command relations in a syntactic object labeled X. (41) Merge at the root when possible, where a root is a category that is c commanded by no other category. Here is what (40) tries to say. Consider the structures in (42 44). If merge needs to apply to the syntactic object M in (42), and there are two options, ( which yields the structure in (43) ) ( which yields the structure in (44) ) Least Tamper dictates that Set Merge should win command relations in M. As (43) shows, X still c commands chang es the c command relations in M. T he structur e in (44) indicates that X now c commands (42) M = X max 3 X Y max 3 Y Z

PAGE 263

263 (43) L = max 3 X max 3 X Y max 3 Y Z (44) M = X max 3 X Y max 3 Y max 3 Y Z Adjunction is not totally banned. As (41) indicates, if merge at the root, or Set Merge, is t take place (45) L = max 3 X max 3 X Y max 3 Y max 3 Y Z Stepanov uses thi s algorithm to account for constructions that involv e raising across an experiencer ( e.g., (46) ) Sentence (46a) can only mean that Tom enjoys shopping, which indicates that the subject of the embedded clause is itself Tom and that it has raised across the experiencer to Sue to land in Spec,IP of the matrix clause, as (46b) shows. The sentence cannot mean that Sue enjoys shopping. Although the experiencer position is a closer c commander to the subject of the embedded clause, r aising into this position is i mpossible, as the ungrammaticality of (46c) illustrates.

PAGE 264

264 (46) a. Tom seems to Sue to enjoy shopping. b. Tom seems to Sue [ IP Tom to enjoy shopping]. c. *Tom seems to Sue [ IP Sue to enjoy shopping]. Stepanov accounts for these facts by considering the experienc er to Sue as an adjunct. On his account, in order for (46) to converge with the intended meaning, (47a) has to form before t he adjunct is inserted in (47b). (47) a. Tom seems [ IP Tom to enjoy shopping]. b. Tom seems [to Sue] [ IP Tom to enjoy shopping] Hende rson ( 2007 ) independently arrives at the same conclusion that adjuncts can only compulsory, to the exclusion of verbal adjuncts of the type exemplified in (37 38) above. According to Henderson, a possible reason why nominal adjuncts and verbal adjuncts behave the site of adjunction in each case Verbal adjuncts must be adjoined to a position in the middle of the clause structure (vP); as such, verbal adju nction is an intermediate step in the derivation of a full clause. On the other hand, nominal adjuncts are presumably not adjoined to some lower projection within DP, but are merged to the DP itself. Therefore, nominal adjunction may be the final step in t he derivation of a DP. This allows nominal adjuncts to be merged anytime after the construction of a DP, giving them the appearance of late merger. This conclusion is preliminary, however, and I leave further exploration of this division for further work. (Henderson 2007: 218 ) perspective, however, it is preferable if all adjuncts merge in the same manner. From an empirical perspective, the Telugu and Assamese Expletive C ontrol structures examined in this chapter as well as the English Expletive Control introduced in Section 6. 4, seem to allow adjuncts to undergo late merge. What remains to be done is to show that the unacceptability of (37) above is not due to late merge

PAGE 265

265 6. 3.2.2 Parasitic g aps minus c yclic m erge Observe the structure in (37), repeated here as (48a), one more time. This structure is minimally different from the grammatical parasitic gap construction in (49a). Whereas which man in (48a) is the object of m eeting in (49a) it is the object of call and meeting. Comparing (48b) and (49b) shows that which man starts out in the thematic position of the adjunct in both sentences. When the wh element undergoes sideward movement, it has the chance to land in a them atic position in the matrix clause in (49b) but not in (48b). (48) a. *Which man did you call your brother before meeting? b. [ CP [which man] did [ IP you [ vP [ vP you call your brother] [ PP before [ CP PRO meeting [which man]]] ]]]] (49) a. Which man did you cal l before meeting? b. [ CP [which man] did [ IP you [ vP [ vP you call [which man]] [ PP before [ CP PRO meeting [which man]]] ]]]] If the restriction in (31) above, repeated below as (50), is correct, then we are able to account for the ungrammaticality of ( 48) without resorting to cyclic vs. late merge. The structure in (48) is ungrammatical because the sideward moving wh element is not able to target the position it normally targets when it undergoes first merge. The reason is that the position is already f illed with the NP your brother Notice that the restriction is not violated in the grammatical structure in (49). Which man undergoes first merge in the thematic domain of the adjunct. When it undergoes sideward movement, it targets the same domain (50) If an undergoes sideward movement. A position X can be a thematic position, a phi position, or a discourse position. This approach to parasitic gaps is certainly oversimplistic. Any attempt at a thorough examination of parasitic gaps goes bey ond the scope of this work ( Nunes 2001, 2004, and works within). The main points of section 6. 3.2 are the following. First, l ate merge is needed to account

PAGE 266

266 for Expletive Control in Telugu an d Assamese. Second, late merge seems to b e needed on independent grounds. Last, the structures that are believed to be overgenerated by late merge in Nunes 2004 can be avoided by adopting the restriction in (50). This chapter has assumed that Expletive Cont rol is a fact about Telugu and Assamese, and that Expletive Control exists. This assumption is problematic when looked at from the perspective of English which does not allow Expletive Control of the type encountered in Telugu and Assamese. The following s ection spells out the details and tries to show that the difference between Telugu and Assamese on the one hand and English on the other hand follows from the nature of expletives and from the Vehicle Requirement on Set and Pair Merge. 6.4 Expletive Contr ol in English vs. Telugu and Assamese There are three types of expletives in English. T hese are weather it (51a) extraposition it (51ab) and there (51c) (Chomsky 1981, Svenonius 2002, among many others). (51) a. Weather it It always rains in August. b Extraposition it It is clear that Tom is going to win. (c) there There is one secretary in that office. Of the three types, the first two may participate in Adjunct Control structures similar to (52). In other words, Expletive Control structures of the typ es exemplified in (53 a ) (Svenonius 2002: 6 (5a)) and (53b) (Williams 1994: 91 (45)) are possible (52) Tom left [without saying goodbye]. (53) a. Weather it It often clears up here [right after snowing heavily]. b. Extraposition it It can seem that someone is gui lty [without seeming that they actually committed a crime].

PAGE 267

267 Concerning the third type in (51c), Lasnik (1992) observes that Expletive Adjunct Control structures in which there is the expletive are not possible in English (e.g., (54a b)) unless the expletiv e is also phonologically realized in the subordinate clause (e.g., (55a b)). Note that Adjunct Control structures that involve an argument do allow (act ually require) a silent subordi n a te subject (e.g., (56a b)) (Lasnik 1992: 244 (51 54) ) (54) a. *[Having been a robbery] there was an investigation b. *There was a crime [without being a victim] (55) a. [There having been a robbery] there was an investigation b. There was a crime [without there being a victim] (56) a. [Having witnessed the robbery] John aided the inv estigation b. Harry was a witness [without being a victim] The Telugu and Assamese Expletive Control structures examined in this chapter seem to involve an expletive that is more similar to there than to weather it or extraposition it For example, in E nglish there expletive constructions, the verb does not sh ow agreement with the expletive. R ather, the verb agre es with another NP that is asso c i ated with the expletive. This NP is standardly referred to as the associate. To illustrate, in (57a) the verb a grees with the singular associate one secretary while in (57b) the verb shows plural agreement with two secretaries Compare to English extraposition it expletive constructions in (57c) in which the verb shows singular agreement with the expletive. This i s true even if the extraposed elemen t, which is not an NP but a CP, is plural by virtue of being a conjunct (e.g., (56d)). Radford ( 2004: 291 307 ) provides an overview. (57) a. There is one secretary in this room. b. There are two secretaries in this room. c. I t is clear that Tom is going to win. d. It is clear that Tom is going to win and that Sue is going to feel jealous.

PAGE 268

268 Like English there Telugu null expletives, which I assume to exist in unaccusative structures, do not enter an agreement relation with the verb. The verb agrees with an associate NP, as (58a b) show. In (58a), war am erb shows singular agreement. In (58b), war a alu this respect because unaccusative verbs do not show agreeme nt with their absolutive argument 5 (58) Te lugu a. pro EXP war am paDin di pro EXP rain.NOM fell 3.N.S b. pro EXP war aalu paDaa yi pro EXP rain.NOM fell 3.N.P This observation gives rise to three que stions delineated below. The first two are most central for our purposes: Question 1 : Why is there Expletive Control banned in English? Question 2 : Are the Telugu and Assamese structures examined in this chapter real instances of Expletive Control? Ques tion 3 : Given that there Expletive Control is banned in English, why is it Expletive Control allowed? I address these questions in the following sections. Section 6. 4.1 presents an analysis of (54 55) within the movement approach to control as proposed by Hornstein (2001). Section 6. 4.2 presents an alternative analysis, suggesting that the lack of there Expletive Control in English 5 I did not have the chance to test whether Telugu and Assamese allow Expletive Control structures that pattern with the English it Expletive Control struct ures in (52). Even if such structures exist, the problem that this section addresses persists. In other words, it will still be mysterious why Telugu and Assamese would allow Expletive Control structures in which the null expletive pro EXP behaves like Engl ish there while English does not allow the same type of structures. This section is meant to address this issue.

PAGE 269

269 does not necessarily mean that this type of control does not exist. Section 6. 4.3 briefly outlines the factors that make it Exp letive Control possible in English. 6. 4.1 The Movement Approach to Expletive Control Lasnik (1992) a nalyzes (54 55), repeated as (59 60), within the PRO Theory H e concludes that (59a b) are ungrammatical with a subordinate while PRO needs to be indexed. 6 1992: 247). Since this study is based on the conviction that control is movement, the discussion in the rest of this section is limited to the movement approach. (59) a. *[PRO having been a robbery] there was an investigation b. *There was a crime [PRO without being a victim] (60) a. [There having been a robbery] there was an investigation b. There was a crime [without there being a victim] H ornstein recognizes the problem presented in Lasnik 1992 and provides an analysis within t he framework of the Movement Theory of Control. He argues that the unacceptability of (59a b) follows from the generalization in (61), combined with the assumption th at all merge including the merge of adjuncts is cyclic. (61) Movement from the adjunct must proceed through a theta position in the matrix. (Hornstein (2001: 120 (119)) To illustrate, observe the derivation of (59a) as presented in (62) below. The adjun ct clause and the matrix clause form independent ly. Following, t he subordinate subject undergoes sidewar d movement to the matrix clause and t he adjunct clause undergoes merge with the matrix 6 examples like (i ii), he assu mes that expletives more particularly, there are unindexed. If there is an indexed NP, this means that it binds a man/men resulting in a Condition C violation. If there and a man/men are contra indexed, there is no way to account for the agreement on the verb. An unindexed expletive, on the other hand, seems to solve distinctness is satisfied between two items when k (1992: 246). (i) There is a man in the room. (ii) There are men in the room.

PAGE 270

270 clause, assumingly at VP. According to Hornstein, if the former s tep takes place first, it blocks the latter The reason is that the subordinate subject, being an expletive, can only move to Spec,IP of the matrix clause. This means that I 0 will have already projected by the time the adjunct has to merge with the matrix clause. In this case, the adjunct cannot undergo merge at VP of the matrix clause without violating the Extension Condition. (62) By the same token, if the adjunc t merges cyclically at VP, the expletive subject can no longer move out of the subordinate clause, as (63) shows. This is so because upon merge the subordinate clause becomes an island. (63) I P qy SUBJ I' 3 I VP qp VP ADJUNCT P 3 6 V 6 there having been a robbery was an investigation Two points ma 6. 2 and 6. 3. First, the generalization in (61) does not leave room for expletives. By pos it ing the ADJUNCT P 3 SUBJ 6 there having been a robbery IP 3 S UBJ I' there 3 I VP 3 V 6 was an investigation S IDEWARD M OVEMENT

PAGE 271

271 restriction that all sideward movement has to touch down in a thematic position, Hornstei n only includes arguments. Expletives are generally considered as non arguments and thus they do not merge in a thematic position. In this sense, his theory is not able to account for the it Expletive Control structures in (53) above. Compare to (31), repe ated below as (64), where no such restriction applies. A argument. (64) undergoes sideward movement. A position X may be a thematic position, a phi position, or a discourse position. In ad dition, Hornstein assumes that all merge is cyclic. Section 6. 3.2.1 presents empirical evidence that late merge is possible. Let us assume that Telugu and Assamese allow Expletive Control. This assumption requires late merge of adjuncts and sideward movem ent into a phi position. Interestingly, the lack of there Expletive Control in English is attributed to the impossibility of these two prerequisites. If Expletive Control is a fact about Telugu and Assamese, and if the analysis presented in Sections 6. 2 an d 6. 3 is on the right track, the ques t i on is: How can we account for the lack of there Expletive Control in English without banning late merge and sideward movement into a phi position and, more importantly, without banning Expletive Control completely ? F urther, g iven that it Expletive Control exists, and assuming that it Expletive Control structures derivationally require late merge and sideward movement into a phi position, we can infer that the failure of the expletive to move to the matrix clause in En glish there Expletive Control structures must be attribute to factors other than late merge or to movement into a phi position.

PAGE 272

272 Recall from Chapter 5 that adjuncts in English are CP goals. According to the Vehicle Requirement on Set and Pair Merge, it is enough for a syntactic object to be a goal in order to undergo Pair Merge. In other words, no movement of the subordinate subject is required in order to qualify the adjunct for merge. This may explain why structures like (65) below are possible. Such sent ences do not involve movement at all. The subject of the adjunct is most likely pro similar to the pro licensed in pro drop languages, as Borer (1989) and Kawasaki (1993) suggest. Notice that the identity of the unpronounced subject is determined via logop horicity. That is, it refers to are reported by the (Williams 1992: 300 (8) ). (65) Having just arrived in town, the new hotel seemed like a good place for a stop. The Vehicle Requirment on Set and Pair Merge is not the only reason why movement takes place. Movement may take place to check a feature of the moving element or to check a feature of the target. Both follow from one principle: Enlightened Self Interest. Let us check whether ei ther of the two reasons may trigger the movement of the subordinate expletive in the English structure (59a), repeated below as (66a). First of all, the fact that the subordinate expletive has to be pronounced indicates that it has checked its Case feature if it has one, and that it does not need to move for its own purposes. Note that this is true, not only of (66a) where the matrix subject is itself an expletive, but also for structures with disjoint subjects like (66b c). (66) a. There having been a robbery there was an investigation. b. Y ou dont get that big without there being some condition (be it physical or mental.) c. No business shall be transacted without there being at least two Officers and two ordinary members present. 7 7 Sentence (58b) and (58c) are from the following webpages respectively (last retrieved May 2007): http://training.fitness.com/members lounge/im watching show tlc 23098.html http://www.psychology.nottingham.ac.uk/bns/Constitution.htm

PAGE 273

273 Now we need to che ck whether the expletive may move to check a feature on the target. Such movement is not possible in the case of (66b c). Take (66b), for example; the adjunct and the matrix clause form independently as (67a b) indicate. At this point, Spec,vP of the matri x clause needs to be filled with an argument. The subject of the adjunct cannot satisfy this requirement because it is not an argument. Thus, the expletive cannot move. Note that the failure of the subordinate subject to move does not cause the English der ivation to crash, as the grammaticality of (66b) indicates. A similar scenario in Telugu or Assamese, however, does not result in a convergent structure. The reason is that, unlike English adjuncts, Telugu and Assamese CNP adjuncts are non goals whose merg e is licensed by the movement of the CNP subject to the matrix clause. (67) a. [ Adjunct there being some condition ] b. [ Matrix vP get that big] The movement of the expletive in (66b) is blocked by its inability to satisfy the thematic requirements of the ma trix predicate. Sentence (66a) is a different story, however. The adjunct and the matrix clause form independently, as indicated in (68a b). At this point, Spec,IP of the matrix clause needs to be filled with an expletive. If there is no expletive in the n umeration that can do the job, the expletive in the adjunct must be able to satisfy this requirement. If this is true, then we expect the subordinate expletive to undergo sideward movement to Spec,IP of the matrix clause, as the solid arrow in (69) signifi es. The two copies of there end up in a c command relationship, symbolized by the dotted arrow. Accordingly, they must form a chain that is subject to Chain Reduction for the purpose of linearization. At PF, one of the copies must be deleted. This does not seem to happen, however. (68) a. [ Adjunct There having been a robbery]

PAGE 274

274 b. [ Matrix IP [ VP was an investigation] (69) I P qy SUBJ I' 3 I VP qp VP ADJUNCT P 3 6 V 6 there having been a robbery was an investigation Several solutions/stipulations may be proposed to explain why (69) is not a possible d erivation. The following section offers one. 6.4.2 Expletives Are Inserted a s Needed In the Lectures on Government and Binding Chomsky (1981: 86) assumes a there there to be inserted in any position fre to mean that expletive insertion is an operation that comes for free in order to satisfy the EPP (Chomsky and Lasnik 1995: 123; Sabel 2000: 412). Within the framework of the Minimalist Program (Chomsky 1995 and subsequent work), the rule is banned by the Inclusiveness Condition. According to Chomsky (2004: 107), introducing no new elements but only rearranging those of the position. optimal if the computation has access only to those elements that have already been selected in the numeration.

PAGE 275

275 The word optimally seems to suggest that there are cases that allow or even favor the insertion of new elements in the process of the derivation that have not been selected in the numeration. One such case is dummy do insertion, as exemplified in (70). (70) Who did Sue help? Such off line insertions seem to be possibly only if they do not violate Full Interpretation whi ch requires that all material be legible at the interfaces. To elaborate, Full Interpretation precludes introducing material with phonological content at LF or material with semantic content at PF, either of which woul nks 2005: 2). Based on nothing necessarily prohibits the insertion of phonologically silent material in LF or semantically vacuous material in PF. do as semantically vacuous, then its insertion must not in duce a violation of Full Interpretation at PF. The same applies to expletives; they are generally considered semantically vacuous, used to fill the subject position (Sabel 2000: 412, fn. 1). Therefore, expletive insertion at PF should be allowed. Allowing dummy do insertion, although theoretically undesirable, seems not to create empirical problems. Allowing expletive insertion, however, means failing to explain why both (71a) and (71b) are allowed. As (72) illustrates, if expletive insertion comes for free it should always win over movement, which is generally considered a more costly operation. (71) a. There are three men in the room. b. Three men are in the room. (72) [ IP _____________ [ I' are [three men [in the room]]] Structures like (71a b) have led to the assumption that expletives respec t the Inclusiveness Condition (i.e. they undergo merge only if they are in the numeration ) The difference between there insertion

PAGE 276

276 (71a) and (71b) is then attributed to the fact that the two sentences start out with two differe nt numerations. These are (73a) and (73b) resp ectively. T he former contains an expletive, while the (73) a. { there 1 are 1 three 1 men 1 in 1 the 1 room 1 } b. {are 1 three 1 men 1 in 1 the 1 room 1 } The assumption behind this approach is that (71 a) and (71b) carry the same meaning and that the only difference between the two is that (71a) has an extra meaningless element that appears in a subject position. This is not necessarily true, however. As no ted by several researchers ( Dikken 1995, Felser and Rupp 2001, and Sabel 2000, among many others), lexical NPs in sentences with expletives always h ave a non specific reading or narrow scope. Consider the sentences in (74). Whereas ( 74b ) a specific or non specific reading of a man (74a) allows only a non specific reading. Besides, many has wide or narrow scope with respect to not whereas many in (74d) that is it cannot take scope over the negat (Sabel 2000: 413 (6)). (74) a. There is a man in the ga rden b. A man is in the garden c. There are not many men in the garden d. Many men are not in the garden Williams (1994 ) uses the scope distinction between (74c) and (74d) to argue that the NP many men in (77c) is a predicate saturated by the subject th ere He holds that if the NP many men were a subject, it would undergo Quantifier Raising a nd have wide scope. The fact that it can ot subject to Quantifer Raising (136). Altho ugh Williams does not really specify whether there is inserted off line or selected in the numeration his analysis suggest s that the difference between (71a) and (71b) resides in the

PAGE 277

277 behavior of the NP three men In (71a), the NP remains downstairs. Synta ctica lly it functions as a predicate. S emantically it allows a strictly non specific reading. In (71b), the NP three men moves to Spec,IP. In this way, it functions as a subjec t, and it allows a specific or non specific ion, I suggest that the difference between (71a) and (71b) should not be attributed to whether the numeration contains an expletive. Rather, both structures start with the same numeration without an expletive In (71a), the subject position is filled with an expletive via insertion because the man behaves as a predicate. In (71b), the man functions as a subject. No expletive insertion takes place Rothstein 2001: 79 arrives at a similar concl usion In addition, expletive there although phonolog ically pres ent, syntactically plays no role at all. For one thing, it does not have a Case feature (Belletti 1988, Lasnik 1995, Chomsky 1995, 2000, and Hornstein 2007, among others). In addition, the verb behaves as if there is not there and agrees with there iate. This further indicates that the assumption that there is inserted rather than selected is not totally unreasonable. 8 Let us assume that some thing along these lines is true. H ow does this reflect on the lack of Expletive Control in English? And why d Expletive Control in Telugu and Assamese? Concerning the first question, if expletive insertion is allowed, this means that expletives are always available for an instance of merge that satisfies the subj ect requirement of a clause. For example, at some point in the derivation of (66a) above, repeated below as (75a), Spec,IP of the matrix clause needs to be filled with an expletive. As (75b) illustrates, this need may be satisfied by merge, whereby an expl etive is inserted as needed, or by sideward movement. 8 The assumption that there is insert ed rather than selected, combined with the observation that there is not syntactically present in a structure, raises the following question: When does insertion take place? I do not have a definitive answer to this question. However, given that the signif icance of there is in its phonological content, it can be speculated that insertion takes place at PF, probably before the domain in which there occurs is converted into a phonological word in the sense of Uriagereka (1999).

PAGE 278

278 Expletive insertion comes for free and, in this sense, it is less costly than movement. Therefore, the expletive in the subordinate clause does not move. (75) a. There having been a robbery, there was an inv estigation. b. I P qy SUBJ I' 3 I VP qp VP ADJUNCT P 3 6 V 6 there having been a robbery was an investigation This is not to say that expletives never undergo movement. Observe the raising structure in (76). The derivation of (76) is (77). The numeration in (77a) is selected from the lexi con; no expletive is included. The subordinate clause forms in (77b). Spec,IP is still empty and needs to be filled, a requirement that may be satisfied either by the movement of the NP a man or by expletive insertion. The choice depends on semantic requir ements concerning specificity and/or scope. Let us assume that in this instance the choice is to have the NP a man unspecific, functioning as a predicate. This means that the NP remains in situ. Accordingly, expletive insertion takes place, as (77c) shows. Following, the matrix clause projects in (77d), at which point Spec,IP of the matrix clause also needs to be filled. The expletive in Spec,IP of the subordinate clause satisfies this requirement, as indicated in (77e). The two copies of there form a chain A t PF, the chain undergoes Chain Reduction for the purpose of linearization, and the lower copy is deleted. The result is (77f). (76) There seems to be a man in the room. (77) a. {seems 1 to 1 be 1 a 1 man 1 in 1 the 1 room 1 } b. [ IP ______ [ I' to [ VP be [ N P a man ] in the room]]] there insertion

PAGE 279

279 c. there insertion => [ IP there [ I' to [ VP be [ NP a man] in the room]]] d. [ IP _______ [ I' [ VP seems [ IP there [ I' to [ VP be [ NP a man] in the room]]]]]] e. [ CP [ IP there [ I' [ VP seems [ IP there [ I' to [ VP be [ NP a man] in the r oom]]]]]]] f. [ CP [ IP there [ I' [ VP seems [ IP there [ I' to [ VP be [ NP a man] in the room]]]]]]] Obviously, the step in (77e) contradicts the a nalysis proposed for (75b) The derivation in (75b) arrives at the same crossroads experienced in (77d). That is, Spec,IP of the matrix clause needs to be filled, and two possibilities are available. These are merge via expletive insertion or the movement of the subordinate expletive. In (75b), expletive insertion wins because merge is less costly. In (77), movement seems to prevail, as indicated by the fact that no copy of there is phonologically realized in the subordinate clause. The quick and easy answer is that movement takes place in (77e) because non finite I 0 does not check the Case feature of there This is a controversial idea considering the standard assumption that there does not have a Case feature to check (Belletti 1988, Lasnik 1995, Chomsky 1995, 2000, and Hornstein 2007, among others). Another possible reason why (75 ) is saved by expletive in sertion while (76 ) resorts to movement follows from the nature of the subordinate clause and the Vehicle Requirement on Set and Pair Merge. In (75), the subordinate clause is a CP, which is a goal. Movement out of a goal may only be triggered by Englighten ed Self Interest. In other words, an element in a goal may move to check a feature of its own or a feature of the target. The non control sentences (66b c) in Section 6. 4.1 suggest that the expletive in English adjuncts does not need to move for its own pu rposes. This means that any movement should be triggered by a need of the target, which in this case is the EPP feature of matrix I 0 This kind of movement takes place only if pure merge is not available; that is, only if the numeration does not have an el ement that fulfill s the need of the

PAGE 280

280 target, or if insertion is not availabe. This section assumes that expletive insertion is available as needed. Therefore, no movement is necessary. In (76), on the other hand, the subordinate clause is a non finite to IP As we saw in Chapter 5, such IPs are non goals (at least not in English). According to the Vehicle Requirement on Set and Pair Merge, in order for a non goal to undergo merge, it has to be predicated of an element in the matrix clause. The movement of th e expletive in (76) helps fulfill this requirement. The expletive copies out of the subordinate clause and merges in the matrix clause. The two copies are non distinct. In this way, the expletive that syntactically saturates the subordinate predicate is al so the expletive in the matrix clause. Concerning Telugu and Assamese Expletive Control, in Section 6. 3 I suggested that the null expletive in the CNP clause undergoes sideward movement for the same reason the expletive in English raising structures under goes movement. Both expletives move in order to license the merge of the non goal subordinate clause that hosts them. If this analysis is correct, it explains why Expletive Control is possible in Telugu and Assamese but not in English. The question that au tomatically follows is: How can we be sure that the two pro EXP Telugu and Assamese constructions u nder examination are related by movement? In other words, can pro EXP in the CNP clause and pro EXP in the matrix clause be copies of two separate toke ns? If this were the case, then we should also be able to find evidence for a structure with only one pro EXP in either the CNP clause or the matrix clause, but not necessarily both. That is, we should be able to find a Telugu or Assamese Expletive Control structure that may be realized as non control construction similar to the English sentence in (78). As (79 80) repetition of (10 11) above show, such structures are not acceptable. (78) Y ou dont get that big without there being some condition (be it phy sical or mental.)

PAGE 281

28 1 (79) Telugu [pro EXP bombu pel i] ambulens waccin di [pro EXP bomb.NOM explode CNP] ambulance.NOM came 3.N.S (80) Assamese [pro EXP e ta ghorot zui lag i] [pro EXP one CL house LOC fire h appen CNP] bohut manuh e police aloi phone koril e many people.NOM police DAT phone did 3 The following section briefly explains why it Expletive Control is possible in English 6.4.3 It vs. There Expl etive Control in English Unlike there Expletive Control, it Expletive Control is allowed in English as (53a b), repeated here as (81a b), illustrate. (81) a. It often clears up here [right after snowing heavily]. b It can seem that someone is guilty [without seeming that they actually committed a crime]. One way to account for this difference is to consider the expletive it as some kind of argument, an idea that has been argued for in the literature. Chomsky (1981: 323 325) considers weather it as a quasi arg ument, and so does Svenonius (2002: 6). Further, several researchers have argued that extraposition it is referential in the sense that it is co indexed with the extraposed CP (Bennis 1986, Kiss 2002); Stroik (1996) even argues that it is base generated in the specifier of the exptraposed CP. These views indicate that it Expletive Control structures are derivationally similar to other Adjunct Control structures in which the subjects are arguments. This means that, unlike the derivation of there Expletive Control structures, the derivation of it delineated in Section 6. 4.1 suffices to explain why there Expletive Control is not allowed.

PAGE 282

282 What makes this view su spect is that the same argument can be made for there For example, Bowers (2002) argues that there does not originate in Spec,IP. I t is base generated in Spec,VP before it moves to Spec,IP. If this is correct, then the derivation of there Expletive Contro l structures must also proceed cyclically. To avoid this dilemma, let us assume that both it and there when used as expletives, are semantically vacuous, non referential element that may not take on a thematic role. Further, let us assume that neither exp letive may b e b ase generate d as an argument or in the position of an argument. T hey both undergo first merge in Spec,IP. If this correct, then the acceptability of it Expletive Control vs. the ungrammaticality of there Expletive Control in English mus t be attributed to other factors, namely, the inherent differences between it and there One distinctive feature of expletive it is that it is not only phonologically but also syntactically present in a structure (Long 1961, in Postal and Pullum 1988: 650 ). The most obvious syntactic presence is manifested in Case and agreement. Expletive it is Case marked. I t can occur as the object of preposition in (82) (Postal and Pullum 1988: 649 (39a)) Further, as (83a b) illustrate, the verb always agrees wi th Expletive it even if the extraposed CP is a conjunct. (82) What do you make of it, then, that his mother is so evasive? (83) a. It is obvious that Tom loves Sue. b. It is obvious that Tom loves Sue and that Sue loves him back. Expletive there on the other han d, seems to be only phonologically but not syntactically present. As I mentioned earlier it does not have a Case feature (Belletti 1988, Lasnik 1995, Chomsky 1995, 2000, and Hornstein 2007, among others). In addition, the verb behaves as if there is not t here and agrees with there (e.g., (84a b))

PAGE 283

283 (84) a There is a man in the room. b. There are two men in the room. Based on this comparison, it is reasonable to deduce that expletive it unlike expletive there cannot be inserted R ather, it has t o be selected in the numeration in order to take part in it although an expletive par excellence, behaves like arguments in that it is subcategorized for. If this id ea is on the right track, this means that (81a) above, repeated here as (85a), has the derivation in (85b). The adjunct and matrix clauses form independently. At some point in the derivation, the numeration is exhausted and Spec,IP of the matrix clause sti ll needs to be filled with an expletive. Since it insertion is not an option, the only available expletive is it in the adjunct. Accordingly, it copies out of the adjunct and merges in matrix Spec,IP. The two copies form a chain At PF, Chain Reduction app lies, and the adjunct copy is deleted. (85) a. It often clears up here [right after snowing heavily]. b. I P qy SUBJ I' 3 I VP qp VP ADJUNCT P 6 6 clears up right after it snowing heavily 6.5 Conclusion The main purpose of this chapter was to show that structures that are normally referred to in the literat ure as exceptions to Adjunct Control into CNP clauses are not really exceptions. They are Expletive Control structures that are allowed only if the CNP clause and the matrix it insertion

PAGE 284

284 clause involve unaccusative predicates. The reason is that unaccusative predicates allow a null expletive to fill the subject position. The chapter is certainly not an exhaustive account of Telugu and Assamese Expletive Control structures. Several issues remain unaddressed. For example, once unaccusative structures come into the pictur e, there is always the possibility that locative expressions play the role of the subject, resulting in what is usually referred to as locative inversion (Bresnan and Kanerva 1989, Bresnan 1994). As a matter of fact, this idea is not far fetched. Take sen tence (86a) for example. It contains a locative expression. According to my consultants, both incidents are assumed to occur in the same location. In other words, the two clauses cannot be simply relating two events that happened in two different locations To illustrate, sentence (86b) is ungrammatical because the CNP clause and the matrix clause involve two different locative expressions. It might be argued that (86b) is ungrammatical because the two clauses do not express a cause effect relation. It is i mportant to note that not all structures that involve a CNP clause express a cause effect relation. Sentence (86c), for instance, does not mean that Kumar and Sarita left because they kissed each other; rather, it relates two incidents: kissing and leaving (86) Telugu a. [Kolkata loo bombu pel i] [Calcutta LOC bomb.NOM explode CNP] caala mandi canipoyaa ru many people.NOM died 3.M.P b. [Kolkata loo bombu pel i] [Calcutta LOC bomb.NOM explod e CNP] Ha yderabad lo o caala mandi canipoyaa ru Hyderabad LOC many people.NOM died 3.M.P in Calcutta and people died in H yderabad

PAGE 285

285 c. [muddu peTTu kon i ] Kumar unnu Sarita well i pooyaa ru kiss put each other CNP Kumar and Sarita.N OM left 3.M.P Another issue that is involved in Expletive Control constructions is the role of semantics. Although unaccusative CNP and matrix clauses are a prerequisite for Expletive Control to obtain, su ch structures seem to be limited to disasters and natural phenomena. The reason might be because speakers look at such incidents as whole events rather than a topic and a comment. In other words, a structure like (1a), repeated here as (87), does not depic t a bomb or certain i ndividuals and talks about them. R ather, it depicts two events: a bomb explosion and casualties. In this sense, the themes in (87) lack the quality of a topic. If we consider subjects to be topic like (Rizzi 2005), then it is expected that the themes in (87) do not to raise to a subject position. Consequently, the subject position is filled with an expletive. (87) Telugu [bombu pel i] caala mandi canipoyaa ru [bomb.NOM explode CNP] many people.NOM died 3.M.P I n conclusion, the topic of Expletive Control in Telugu and Assamese, as well as in other languages in the Indian Subcontinent, deserves an extensive study in its own right. This chapter suggests that these structures do not po se a challenge to the analysis of Adjunct Control proposed in chapters 3 through 5. Nonetheless, the structures are themselves a challenge, as this conclusion hopes to have shown.

PAGE 286

286 CHAPTER 7 SUMMARY AND CONCLUSION 7.1 Summary I set out to explore Adjunct Control in two S outh Asian languages, Telugu and A ssamese, within the Minimalist Program of syntactic theory (Chomsky 1995 and subsequent work). Adjunct Control structures comprise two obligatorily coreferential subjects, one in the matrix clause and one in the adjunct/subordinate clause. Telugu and Assamese have non finite Conjunctive Participle (CNP) clauses that function as adjuncts Both languages show evidence of Adjunct Control into CNP clauses. Three types of Adjunct Control were examined. These are Forward Control, Backward Control, and Copy Control. As (1) illustrates, in Forward Control structures the matrix subject is pronounced while the CNP subject is implied. Conversely, in Backward Control structures the CNP subject is pronounced and the matr ix subject is implied. Copy Control contains two pronounced subjects. (1) Types of Adjunct Control a. Forward Control [ Matrix [ Matrix [ CNP Subject b. Backward Control [ Matrix [ Matrix Subject [ CNP c. Copy Control [ Matrix [ CNP Subje [ Matrix Subject In Chapter 2, I present ed evidence for the different types of control. As (2 4) demonstrate, Telugu licenses all three types of control delineated in (1), while Assamese licenses only Forward and Copy Control. Backward Control str uctures in Assamese are judged as degraded or unacceptable. (2) Forward Control a. Telugu Kumar [ Kumar ki aakali wees i] sandwic tinnaa Du Kumar.NOM [ Kumar DAT hunger fall CNP] sandwich ate 3.M.S

PAGE 287

287 b. Assam ese Ram e [ Ram Or bhok lag i] posa bhat khal e Ram NOM [ Ram GEN hunger feel CNP] stale rice ate 3 (3) Backward Control a. Telugu Kumar [Kumar ki aakali wees i] sandwic tinnaa Du Kumar.NOM [Kumar DAT hunger fall CNP] sandwich ate 3.M.S b. Assamese ?? Ram e [Ram Or bhok lag i] posa bhat khal e Ram NOM [Ram GEN hunger feel CNP] stale rice ate 3 (4) Copy Control a. Telugu [Kumar ki aakali wees i] [Kumar DAT hunger fall CNP] atanu / aa pichooDu / Kumar sandwic tinnaa Du he.NOM / that idiot.NOM / Kumar.NOM sandwich ate 3.M.S b. Assamese [Ram Or bhok lag i] [Ram GEN hunger feel CNP] xi / beseratu e / Ram e posa bhat khal e he.NOM / the poor guy NOM / Ram NOM stale rice ate 3 Further, as (1c) and (4a b) tacit ly indicate, Copy Control obtains only if conditions (a) and (b) in (5) bel ow apply. An addition condition (5c) is needed for Assamese. (5) Conditions on Copy Control a. The CNP clause must be sentence initial. b. The CNP subject must be an R expression. c. T he CNP subject is licensed by an experiential predicate. (Only in Assamese) If these conditions are met, three subcategories of Copy Control become available, depending on the nature of the matrix subject. These are listed in (6).

PAGE 288

288 (6) Sub types of Copy Contr ol a. [ Matrix [ CNP R expression [ Matrix Pronoun b. [ Matrix [ CNP R expression [ Matrix Epithet c. [ Matrix [ CNP R expression [ Matrix R expression In Chapters 3 and 4, I propose d an analysis of Adjunct Control as movement (Hornstein 1999) Following Nunes ( 2004 ) I argue d that the subject starts out in the subordinate clause and undergoes sideward movement to the matrix clause resulting in non distinct copies of the same element in both clauses Decisions regarding the pronunciation of co pies take place on the phonological side of the computation for the purpose of linearization. According to Kayne (1994), a structure is considered linearized only if all c command and precedence relations are asymmetrical. In other words, if X c commands or precedes Y, Y necessarily does not c command or precede X. Nunes (2004), building on Kayne (1994), argues that a structure that comprises multiple non distinct copies of the same token can be properly linearized only if all but one copy are deleted at P F. Otherwise, the same token will precede and follow itself, which is not allowed. I adopted this approach for the analysis of Forward and Backward Control in Chapter 3, laying out the derivational history of each. Chapter 4 presented an analysis of Copy C ontrol as an instance of multiple copy spell out in which more tha n one copy of the same token is pronounced at PF. According to Nunes, such cases are allowed only if one of the non distinct copies that are derived by movement undergoes fusion with another head. The copy and the head form a new phonological word. Since linearization cannot see through words, the fused copy becomes invisible and escapes deletion. In the case of Copy Control, I suggest ed that one of the copies escapes deletion by becoming a p art of giant phonological word, a spelled out domain. This scenario assumes that Multiple Spell Out is derivation ally possible (Uriagereka 1999) and that linearization applies cyclically, phase by phase (Fujii 2005).

PAGE 289

289 Finally, Telugu and Assamese allow str uctures that are considered exceptions to Adjunct Control. These are structures that pattern the same as Adjunct Control structures, but they comprise disjoint subjects. I presented an analysis of these exceptions, suggesting that they are instances of Exp letive Control that have the same derivational history as the other more common instances of Adjunct Control. The main difference is that Expletive Control structures are allowed only if the CNP and matrix clauses contain unaccusative predicates, in which case the subject positions are filled with null expletives. 7.2 Theoretical Implications Analyzing Telugu and Assamese Adjunct Control as movement gives rise to thr ee theoretical issues. These are related to multiple Case checking, the nature of the prono unced copies in Cop y Control structures, and the trigger for movement. The preceding chapters dealt with these issues and suggested solutions. The following sections highlight the main points of each. 7. 2.1 Multiple Case Checking Under the analysis adopted in this study, the CNP subject in an Adjunct Control structure checks Case in the CNP clause before it moves to the matrix clause where it checks Case again. Although empi rical evidence from other languages suggests that multiple Case checking is possible (Bejar and Massam 1999, Merchant 2006, among others), one problem persists. Analyzing control as movement means arguing that control and raising are derivationally similar. Unlike control, however, raising does not involve multiple Case checking. To solve this problem, I proposed a principle called Theta Role Visibility. As the formulation in (7) indicates, an argument is allowed to check one round of Case only if it takes on a theta role. A moving element in a raising structure takes on one theta role; th is is why it checks one round of Case. A

PAGE 290

290 moving element in a control structure takes on two theta roles; this is why it checks two rounds of Case. (7) Theta Role Visibility a. An argument is visible for one round of Case checking iff it Merges into a Thematic position. b. A round of Case comprises Inherent C ase followed by Structural Case, depending on the availability of an appropriate licenser for each. The following section deals with the second issue: the nature of the pronounced copies in Copy Control structures. 7.2.2 R Expressions vs. Pronominals in Copy Control In Copy Control structure s, the CNP subject has to be an R expression, while the matrix subject may be a pronominal (a pronoun or an epithet) or an R expression. Concerning the CNP subject, I showed that the restriction is language specific, due to the fact that neither Telugu nor Assamese allows cataphoricity. In other words, given a structure S that has two co referential NPs, NP1 and NP2, such that NP1 linearly precedes NP2, if one of the NPs has to be an R expression while the other has to be a pronominal, NP1 will be the R expression and NP2 the pronominal, but not the other way around. Regarding the matrix subject, I showed that the different choices are made poss ible by movement. I followed Aoun, Choueiri, and Hornstein (2001) and Boeckx (2003) and suggested that pronounced pronominals (pronouns and epithets) that are the outcome of movement start out as adjuncts or appositives to another element, say, an R expres sion. Unlike the aforementioned authors who argue that these pronominals are stranded after the element they adjoin to has moved, as illustrated in (8), I suggested that they move along with the moving element, as (9) shows. Decisions concerning whether to pronounce the R expression or the pronominal are made at PF in accordance with the language specific rules of cataphoricity and economy.

PAGE 291

291 (8) [ [ SUBJ R expression [ [ SUBJ R expression [Pronominal] (9) [ [ SUBJ R expression [Pronominal] [ [ SUBJ R expression [Pronominal] 7.2.3 Why Movement Minimali stically, any approach that adopts movement as a central derivational mechanism has to explain why syntactic objects move. The more common instances of movement take place to satisfy a formal fea tural requirement of the moving element or of the target. I showed that the movement of the CNP subject in Telugu and possibly Assamese Adjunct Control is less conventional in this respect. I suggested that the CNP subject moves in order to make up f or a deficiency in the adjunct. Building on work by Pestesky and Torrego (2006) and on Predication Theory as proposed by Rothstein (2001), I showed that the adjuncts under investigation are defective in that they do not automatically qualify for an instanc e of merge with the matrix clause. I suggested that the movement of their subjects licenses their merge by allowing them to be predicated of an element in the matrix clause. I call this requirement the Vehicle Requirement on Set and Pair Merge. (10) The Vehicl e Requirement on Set and Pair Merge a. b. goal for some probe c. a non an d. A (potential) goal is a syntactic object with some feature F. A non goal is a syntactic object with no features.

PAGE 292

292 The power of (10) is that it is able to explain why Forward and Backward Control are in free variation in Telugu. In bot h types of control, the subordinate subject must move to the matrix clause. Yet, it does so, not to make up for a defect of its own, but rather to make up for a defect of the clause that hosts it. This is why when decisions concerning the pronunciation of copies are made at PF, the subordinate subject has an equal chance of survival as its matrix copy. (see Polinsky and Potsdam 2006: 14 for a similar conclusion.) 7.3 Concluding Remarks This study hopes to h ave provided some answers to long standing problem s in control theory. At the same time it has certainly left several questions unanswered and issues unaddressed. In my opinion, two issues are especially worth researching. First, knowing that Telugu and Asamese license Copy Control, it will be interestin g to examine how prevalent Copy Control is in the Indian Subcontinent. This includes other Dravidian and Indo Aryan languages, as well as Munda and Sino Tibetan languages. On a broader scale, if Telugu and Assamese Adjunct Control is movement, does this me an that control is movement cross linguistically, or is there a room for variation? And if variation exists, is it parametric, or is it due to interaction among a defined set of principles, as Henderson (2006), building on Newmeyer 2004 and Roberts and Hol mberg 2005, would argue? It is my conviction that control is movement, but only more research on more languages can provide a more definitive answer.

PAGE 293

293 APPENDIX ADJUNCT CONTROL AS EXHAUSTIVE CONTROL It is worth noting that Telugu and Assamese Adjunct Con trol can be classified, not only as Obligatory Control, but also as Exhaustive Control, whereby the identities of the CNP and matrix subjects fully coincide. Stated differently, Telugu and Assamese do not allow Partial Adjunct Control into CNP clauses. Acc ording to Landau (2004: 833), Partial Control is the case when from English, in ( 1 the manager plus a particular group that the manager decided to work with. Compare to ( 1 b) in which PRO can only refer to the manager I t cannot include other individuals. Sentence ( 1 a) is an example of Partial Control and sentence ( 1 b) of Exhaustive Control. (1) a. The manager decided to work on the project together] b. to work on the project (* together ) ] Only Exhaustive Control is allowed in Telugu and Assamese. Take the Telugu sentences in (2), for instance E ve n with enough context (e.g., (2 a)), only the Exhaustive Control structure (2b) is felicitous. O nce presented as an insta nce of Partial Control as in (2 c), acceptability dro ps dramatically. Notice that (2 c) becomes acceptable if an INF claus e is used, as exemplified in (2 d). (2) Telugu a. Kumar unnu Sarita okari ki okaru kathalu ceppukunnaa ru Kumar and Sarita one to one stories told 3.M.P b. [ Kumar unnu Sarita okari ki okaru kathalu ceppukum Tuu] [ Kumar and Sarita.NOM one to one stories tell CNP] Kumar unnu Sarita nawwukunnaa ru Kumar and Sarita.NOM laughed 3.M.P

PAGE 294

294 c. ??/* [ Kumar un nu Sarita okari ki okaru kathalu ceppukum tuu] [ Kumar and Sarita.NOM one to one stories tell CNP] Kumar nawwukunnaa Du Kumar.NOM laughed 3.M.S d. [okari ki okaru kathalu ceppuku m tunna appudu] Kumar nawwukunnaa Du [one to one stories tell INF while] Kumar.NOM laughed 3.M.S Assamese displays the same characteristics. Ev en with enough context (e.g. (3 a)) o nly the Exhaustive structure (3b) is acceptable. Sentence (3 c) is infelicitous because it is an instance of Partial Control. The idea in (3 c) may be expressed grammatica lly if an INF clause is used ( 3 d). (3) Assamese a. Ram aru Prajakta e suma khal e Ram a nd Prajakta NOM kiss ate b. [ Ram aru Prajakta e Suma kha i] [ Ram and Prajakta NOM Kiss eat CNP] Ram aru Prajakta gusi gol Ram and Prajakta ABS away went c. *[ Ram aru Prajakta e Suma kha i] Ram gusi gol [ Ram and Prajakta NOM Kiss eat CNP] Ram ABS away went d. [Suma khuw a r pasot] Ram gusi gol [Kiss eat INF GEN After] Ram ABS away went

PAGE 295

295 LIST OF REFERENCES South Asian language. In Experiencer subjects in South Asian languages ed. Manindra K. Verna and K.P. Mo hanan, 253 268. Stanford: CSLI Publication s Adger, David. 2003. Core syntax: A Minimalist approach New York: Oxford University Press. Aikhenvald, Alexandra Y. 2003. Serial verb constructions in typological perspective. In Serial verb constructions: A cro ss linguistic typology ed. Alexandra Y. Aikhenvald and R.M.W. Dixon, 1 68. New York: Oxford University Press. Aikhenvald, Alexandra Y. and R.M.W. Dixon, eds. 2003. Serial verb constructions: A cross linguistic typology New York: Oxford University Press. Alboiu, Gabriela and Virginia Motapanyane. 2000. The Generative approach to Romanian grammar: An overview. In Comparative studies in Romanian Syntax ed. Virginia Motapanyane, 1 48. Amsterdam: Elsevier. Allerton, David J. 1982. Valency and the English verb New York: Academic Press. Amritavalli, R. and Partha Protim Sarma. 2002. A case distinction between unaccusative and unergative subjects in Assamese. Snippets 5:6 7 Andrews, Avery D.1990. The VP complement analysis in modern Icelandic. In Modern Icelan dic Syntax ed. Joan Maling and Annie Zaenen, 165 185. San Diego, CA: Academic Press. Aoun, Joseph, Lina Choueiri, and Norbert Hornstein. 2001. Resumption, movement, and derivational economy. Linguistic Inquiry 32:371 403. Arora, Habir Kaur and Karumuri Ve nkata Subbarao. 2004. Syntactic change and convergence. In Non nominative subjects Vol. 1, ed. Peri Bhaskararao and Karumuri Venkata Subbarao, 25 48. Amsterdam: John Benjamin Publishing Company. Bejar, Susana and Diane Massam. 1999. Multiple Case Checking Syntax 2:65 79. Belletti, Adriana. 1988. The case of unaccusatives. Linguistic Inquiry 19:1 34 Bennis, Hans. 1986. Gaps and dummies. Dordrecht: Foris. Boeckx, Cedric. 2003. Islands and chains: resumption as stranding Amsterdam: John Benjamin Publishing Company. Boeckx, Cedric and Norbert Hornstein. 2004. Movement under control. Natural Language Semantics 14:1 56.

PAGE 296

296 Boeckx, Cedric and Norbert Hornstein. 2006. Control in Icelandic and theories of control. Linguistic Inquiry 37:591 606. Boeckx, Cedric, Norb ert Hornstein, and Jairo Nunes. 2007 Overt copies in reflexive and control structures: A movement analysis In University of Maryland Working Papers in Linguistics 15 ed. Anastasia Conroy, Chunyuan Jing, Chizuru Nakao and Eri Takahashi, 1 46. College Pa rk, MD: UMWPiL. Borer, Hagit. 1989. Anaphoric AGR. In The null subject parameter ed. Osvaldo and Kenneth Safir Jaeggli, 69 110. Dordrecht: Kluwer. Positions. Linguistic Analysis 24:247 286. The Syntax of Nonfinite Complementation: An Economy Approach MIT Press, Cambridge, Mass. movement and the EPP. Syntax 5: 167 218. Linguistic Inquiry 29:347 366. Bowers, John. 2002. Transitivity. Linguistic Inquiry 33:183 224. Bresnan, Joan. 1994. Locative inversion and the architecture of Universal Grammar Language 70:2 131. Bresnan, Joan and Jonni Kanerva. 1989. Locative inversio n in Chichewa: a case study of factorization in grammar Linguistic Inquiry 20:1 50. Brody, Michael. 1993. Theta theory and arguments. Linguistic Inquiry 24:1 23. Brody, Michael. 1995. Lexico Logical Form: A radically minimalist theory Cambridge, Mass.: MIT Press. Caldwell, Robert. 1856. A comparative grammar of the Dravidian or the South Indian family of languages. London: Harrison. Chomsky, Noam. 1965. Aspects of the theory of syntax. Cambridge, Mass.: MIT Press. Chomsky, Noam. 1981. Lectures on Governm ent and Binding. Dordrecht: Mouton de Gruyter. Chomsky, Noam. 1986 a Barriers Cambridge, Mass.: MIT Press. Chomsky, Noam. 1986b. Knowledge of Language New York: Praeger.

PAGE 297

297 Chomsky, Noam. 1995. The Minimalist Program. Cambridge, Mass.: MIT Press. Chomsky, N oam. 1998. Minimalist inquiries : The framework MIT Occasional Papers in Linguistics 15. Cambridge, Mass.: MITWPL. Chomsky, Noam. 2000. Minimalist inquiries: the framework. In Step by Step ed. Roger Martin and David Michaels and Juan Uriagereka, 89 155. C ambridge, Mass.: MIT Press. Chomsky, Noam. 2001. Derivation by phase. In Ken Hale: A life in language ed. Michael Kenstowicz. Cambridge, Mass.: MIT Press. Chomsky, Noam. 2004. Beyond explanatory adequacy. In Structure and beyond: The cartography of syntac tic structures, Vol. 3, ed. Adriana Belletti, 104 131. New York: Oxford University Press. Chomsky, Noam and Howard Lasnik.1995. The theory of principles and parameters. In Chomsky 1995. Chung, Sandra. 1978. Case marking and grammatical relations in Polynes ian. Austin: University of Texas Press. Cinque, Guglielmo. 1990. Types of A' dependencies Cambridge, Mass.: MIT Press. Cinque, Guglielmo. 1999. Adverbs and functional heads: A cross linguistic perspective. New York: Oxford University Press. Clark, Robin. 1990. Thematic theory in syntax and interpretation New York: Routledge. Culicover, Peter and Ray Jackendoff. 2001. Control is not movement. Linguistic I nquiry 32:493 512. Davison, Alice. 1981. Syntactic and semantic indeterminacy resolved: a mostly pragmatic analysis for the Hindi conjunctive participle. In Radical pragmatics ed. Peter Cole, 101 128. New York: Academic Press. Davison, A lice. 1993. Controlled experiencer subjects; implications for phrase structure. South Asian Language Review 3:46 58. Dikken, Marcel den. 1995. Binding, expletives, and levels. Linguistic Inquiry 26: 347 354. du Plessis, Hans. 1977. Wh movement in Afrikaan s. Linguistic Inquiry 8: 723 726. Dubinsky, Stanley. 2007. On the syntax of exhaustive co ntrol and the calculu s of events. Paper presented at LSA 81, Anaheim, California.

PAGE 298

298 Dubinsky, Stanley and Sho ko Hamano. 2006. A window into the syntax of control: Event o pacity in Japanese and English. In University of Maryland Working Papers in Linguistics 15 73 97. College Park MD : UMWPiL Dwarikesh, Dwarika Prasad. 1971. The historical syntax of the conjunctive participial phrase in the New Indo Aryan Dialects of the Ma Doctoral dissertation, University of Chicago, Illinois. Edwards, Keri Michon. 2003. Agency, individuation, and affectedness: basic clause structure in Assamese with special attention to grammatical relations and the ir semantic correlates. Masters thesis, University of Oregon. Evans, Nicholas. 2005. Multiple case in Kayardild: Anti iconic suffix ordering and the diachronic filter. In Double Case: Agreement by Suffixaufnahme ed. Frans Plank 396 428. New York: Oxford University Press. Felser, Claudia and Laura Rupp. 2001. Expletives as arguments: Germanic existential sentences revisited. Linguistische Berichte 187:289 324. Fox, Danny and David Pesetsky. 2005. Cyclic linearization of syntactic structure. T heoretical Linguistics 31:1 45. Fox, Danny and Jon Nissenbaum. 1999. Extraposition and scope: A case for overt QR. In In WCCFL 18 Proceedings ed. Sonya Bird, Andrew Carnie, Jason D. Haugen, and Peter Norquest, 132 144. Somerville: Cascadilla Press. Franks Steven. 2005. What is that ? In Indiana University Working Papers in Linguistics 5, ed. Yoshihisa Kitagawa and Dorian Roehrs, 33 62. Bloomington, Indiana: Indiana University. Linguistic Inquiry 32:174 183. Fujii, Tomohiro. 2005. Cycle, linearization of chains, and multiple case checking. In Proceedings of Console XIII ed. Sylvia Blaho, Luis Vicente and Erik Schoorlemmer, 39 65. Leiden, Netherlands: Leiden University Centre for Linguistics. Ghomeshi, Jila, Ray Jackendoff, Nicole Rosen, and Kevin Russell. 2004. Contrastive focus reduplication in English (The salad salad paper). Natural Language and L inguistic Theory 22:307 357. Goswami, Golockchandra. 1982. Structure of Assamese Calcutta: Gauhati University, Department of Publication. Goswami, Golockchandra and Jyotiprakash Tamuli. 2003. Asamiya. In The Indo Aryan Languages, ed. George Cardona and Dh anesh Jain, 391 443. New York: Rou tledge.

PAGE 299

299 Grimshaw, Jane. 1990. Argument structure Cambridge, Mass.: MIT Press Grohmann, Kleanthes. 2003. Prolific domains: On the anti locality of movement dependencies. Amsterdam: John Benjamins Publishing Company. Halle, Morris and Alec Marantz. 1993. Distributed Morphology and the pieces of inflection. In The view from building 20; Essays in honor of Sylvain Bromberger ed. Kenneth Hale and Samuel Jay Keyser, 111 176. Cambridge, Mass.: MIT Press. Harley, Heidi. 1995. Hug a tree: Deriving the morphosyntactic feature heirarchy. In Papers on Phonology and Morphology ed. Andrew Carnie and Heidi Harley, 189 230. Cambridge, Mass.: MIT Working Papers in Linguistics. Haspelmath, Martin. 1995. The converb as a cross linguisticall y valid category. In Converbs in cross linguistic perspective ed. Martin Haspelmath and Ekkehard Konig, 1 56. New York: Mouton de Gruyter. Henderson, Brent. 2006. The syntax and typology of Bantu relative clauses. Doctoral dissertation, University of Ill inois at Urbana Champaign. Henderson, Brent. 2007 Matching and raising unified. Lingua 117:202 220 Hiemstra, Inge. 1986. Some aspects of wh questions in Frisian. In North Western European Language Evolution 8: 97 110. Hornstein, Norbert. 1999. Movement a nd control. Linguistic Inquiry 30:69 96. Hornstein, Norbert. 2001. Move! A Minimalist theory of construal: Generative syntax United Kingdom: Blackwell. Hornstein, Norbert. 2003. On control. In Minimalist syntax ed. Randall Hendrick, 6 81. United Kingdom: Blackwell. Hornstein, Norbert. 2007. A very short note on existential constructions. Linguistic Inquiry 38: 410 411. Hornstein, Norbert and Hirohisa Kiguchi. 2001. PRO gate and sideward movement. Paper presented at the Conference on the S yntax and Semantics of Semitic Languages and associated workshops, University of Southern California Hornstein, Norbert, Jairo Nunes, and Kleanthes Grohmann. 2005. Understanding Minimalism United Kingdom: Cambridge University Press. Jackendoff, Ray 1990. On Larson's treatment of the double object construction. Linguistic Inquiry 21: 427 456.

PAGE 300

300 Jackendoff, Ray and Peter Culicover. 2003. The semantic basis of control in English. Language 79: 517 56 Jansen, Joana Worth. 2004. Complementation strategies in Assamese with particular consideration of the historical sources and development of the Assamese infinitive. Masters thesis, University of Oregon. Jayaseelan, K.A. 2004. Th e serial verb construction in Malayalam. In Clause structure in South Asian languages ed. Veneeta Dayal and Anoop Mahajan, 67 91. Dordrecht : Kluwer. Joseph, Brian D. and Arnold M. Zwicky (ed.). 1990. When verbs collide: Papers from the 1990 Ohio State min i conference on serial verbs Colu mbus: The Ohio State University. Kandybowicz, Jason. 2006. Conditions on multiple copy spell out and the syntax phonology interface. Doctoral dissertation, UCLA. Kawasaki, Noriko. 1993. Control and arbitrary interpretation in English. Doctoral dissertation, University of Massachussetts. Kayne, R. (1979) Two notes on the NIC. In Theory of markedness in generative grammar ed. Adriana Belletti, Luciana Brandi and Luigi Rizzi, 317 346. Pisa: Scuola Normale Superiore di Pisa. K ayne, Richard. 1994. The antisymmetry of syntax Cambridge, Mass.: MIT Press. Kehler, Andrew. 1996. Coherence and the coordinate structure constraint Paper presented at BLS 22, Univer sity of California, Berkeley. Kiss, Katalin 1985. Parasitic gaps and case. Paper presented at MIT, Cambridge, Mass. Kiss, Katalin 2002. The EPP in a topic prominent language. In Subjects, expletives, and the EPP ed. Peter Svenonius, 107 124. New Yor k: Oxford University Press. Klaiman, M. H. 1981. Volitionality and subject in Bengali: A study of semantic parameters in grammatical processes. Doctoral dissertation, University of Chicago, Illinois. Ko, Heejeong. 2007. Asymmetries in scrambling and cyclic linearization. Linguistic Inquiry 38:49 83 Koopman, Hilda. 1984. The syntax of verbs Dordrecht: Foris. Krishnamurti, Bh. 1997. Telugu. In The Dravidian languages ed. Sanford Steever, 202 240. New York: R outledge. Krishnamurti Bh 2003. The Dravidian l anguages United Kingdom: Cambridge University Press.

PAGE 301

301 Krishnamurti, Bh ., and J.P.L. Gwynn. 1985. A grammar of modern Telugu Delhi: Oxford University Press. Kuroda, S. Y. 1965. Generative grammatical studies in the Japanese language. Doctoral dissertation, MIT, Cambridge, Mass. (reprinted by New York: Garland, 1979). Kuroda, S. Y. 1978. Case marking, canonical sentence patterns, and counter equi in Japanese. In Problems in Japanese syntax and semantics ed. John Hinds, 30 51. Tokyo: Kaitakusha. Lakoff, Geor ge. 1986. Frame semantic control of the coordinate structure constraint. Paper presented at CLS 22, Chicago. Landau, Idan. 2000. Elements of control Dordrecht: Kluwer. Landau, Idan. 2001. Control and extraposition: The case of super equi. Natural Language and Linguistic Theory 19:109 152. Landau, Idan. 2003. Movement out of control. Linguistic Inquiry 34:471 498. Landau, Idan. 2004. The scale of finitene ss and the calculus of control. Natural Language and Linguistic Theory 22:811 877. Landau, Idan. 2006. S evering the distribution of PRO from Case. Syntax 9:153 170. Larson, Richard K. 1988. On the double object construction. Linguistic Inquiry 1:335 391. Lasnik, Howard. 1992. Two notes on control and binding. In Control and grammar ed. Richard K. Larson, Sa bine Iatridou, Utpal Lahiri, and James Higginbotham, 235 252 Dordrecht: Kluwer. Lasnik, Howard. 1995. Last Resort and Attract F. In Proceedings of the Sixth Annual Meeting of the Formal Linguistics Society of Mid America, 62 81. Bloomington, Indiana: Indi ana University Linguistics Club. Lasnik, Howard. 2002. Minimalist investigations in linguistic theory. New York: Routledge. Lebeaux, David. 1988. Language acquisition and the form of the grammar. Doctoral dissertation, University of Massachusetts, Amherst Lebeaux, David. 1991. Relative clauses, licensing, and the nature of the derivation. In Syntax and semantics 25: Perspectives on phrase structure ed. Susan Rothstein, 209 239. New York: Academic Press. Lee, Felicia. 2003. Anaphoric R expressions as boun d variables. Syntax 6:1:84 114.

PAGE 302

302 Linholm, James Milton. 1975. The conceptual basis of the Tamil adverbial participle. Doctoral dissertation, University of Chicago, Illinois. Long, Ralph B. 1961. The sentence and its parts. Chicago, Illinois: The University of Chicago Press. Marantz, Alec. 1995. The Minimalist Program. In Government and Binding theory and the Minimalist Program: Principles and Parameters in syntactic theory ed. Gert Webelhuth, 351 382. Cambridge, Mass.: Blackwell. Martin, Roger. 1996. A Min imalist theory of PRO and control. Doctoral dissertation, University of Connecticut. Martin, Roger. 2001. Null Case and the distribution of PRO. Linguistic Inquiry 32:141 166. Masica, Colin P. 1991. The Indo Aryan languages United Kingdom: Cambridge Unive rsity Press. Masica, Colin P. 2005. Defining a linguistic a rea New Delhi: Chronicle Books. Massam, Diane. 1985. Case theory and the Projection Principle. Doctoral dissertation, MIT, Cambridge, Mass. McCloskey, James. 2005. Resumption. In The Blackwell Com panion to Syntax, Vol. 4, ed. Martin Everaert and Henk van Riemsdijk, 94 117. Oxford: Blackwell. McDaniel, Dana. 1986. Conditions on wh chains. Doctoral dissertation, CUNY. Merchant, Jason. 2006. Polyvalent case, geometric hierarchies, and split ergativity Paper presented at CLS 42, Chicago. Mohanan, Karuvannur P. 1983. Functional and anaphoric control. Linguistic Inquiry 14: 641 674. Monahan, Philip. 2003. Backward object control in Korean. Paper presented at WCCFL22, UCSD. Monahan, Tara. 1994. Argument st ructure in Hindu Stanford: CSLI Publications. Mller, Gereon 2000. Shape conservation and remnant movement. Paper presented at NELS 30, Rutgers University. Napoli, Donna Jo. 1992. The d ouble o bject c onstruction, d omain a symmetries, and l inear p recedence Linguistics 30 : 837 871. Nath, Diganta. 2003. Case in Assamese. MPhil dissertation, Center Institute of English and Foreign Languages, Hyderabad, India..

PAGE 303

303 Newmeyer, Frederick J. 2004. Against a parameter setting approach to language variation. Linguistic Var iation Yearbook 4:181 234. Nunes, Jairo. 1995. The copy theory of movement and linearization of chains in the Minimalist Program. Doctoral dissertation, University of Maryland, College Park. Nunes, Jairo. 2001. Sideward movement. Linguistic Inquiry 32:303 344. Nunes, Jairo. 2004. Linearization of chains and sideward movement Cambridge, Mass.: MIT Press. Nunes, Jairo, and Juan Uriagereka. 2000. Cyclicity and extraction domains. Syntax 3:20 43. O'Neil, John. 1995. Out of control. Paper presented NELS 25, Un iversity of Massachusetts. Program. Doctoral Dissertation, Harvard University. Oshita, Hiroyuki. 2004. Is there anything there when there is not there: Null e xpletives and second language data. Second Language Research 20:95 130. Ouhalla, Jamal. 1999. Introducing transformational grammar: From Principles and Parameters to Minimalism New York: Oxford University Press. Pandharipande, Rajeshwari V. 1997. Marathi New York: Routledge. Perlmutter, David M. 1971. Deep and surface structure constraints in syntax New York: Holt, Rinehart and Winston. Pesetsky, David. 1998. Some optimality principles of sentence pronunciation. In Is the best good enough? ed. Pilar Ba rbosa, Danny Fox, Paul Hagstrom, Martha McGinnis and David Pesetsky, 337 384. Cambridge, Mass.: MIT Press. Pesetsky, David and Esther Torrego. 2001. T to c movement: Causes and consequences. In Ken Hale: a life in language ed. Michael Kenstowicz, 355 426. Cambridge, Mass.: MIT Press. Pesetsky, David and Esther Torrego. 2006. Probes, goals, and syntactic categories. Paper presented at the 7th annual Tokyo Conference on Psycholinguistics, Keio University, Japan Pires, Acrisio. 2005. The Minimalist syntax of defective domains: Gerunds and infinitives Ms., University of Michigan, Ann Arbor. [Published in 2006, Amsterdam: John Benjamin Publishing Company.] Polinsky, Maria and Eric Potsdam. 2002. Backward Control. Linguistic Inquiry 33 :245 282.

PAGE 304

304 Polinsky, Maria and Eric Potsdam. 2003. Control in Malagasy. Cornell Working Papers in Linguistics 19:173 187. Polinsky, Maria and Eric Potsdam. 2004. Malagasy control and its theoretical implications. Paper presented at BLS 30, University of California, Berkeley. Polins ky, Maria and Eric Potsdam. 2006. Expanding the scope of control and raising. Syntax 9: 171 192. Post, Mark. 2006. Grammaticalization and the distribution of serial verbs in Assamese. Paper presented at SEALS 14, Thammasat University, Bangkok Thailand. Postal, Paul M. 1974. On raising: One rule of English grammar and its theoretical implications Cambridge, Mass.: MIT Press. Postal, Paul M. and Geoffrey K. Pullum. 1988. Expletive noun phrases in subcategorized positions. Linguistic Inquiry 19 :635 670. Potsdam, Eric. 2006. Backward object control in Malagasy and principles of chain reduction. Ms., University of Florida, Gainesville. http://www.clas.ufl.edu/users/p otsdam/manuscripts/MalagasyBOC.pdf (last retrieved, June 2007) Potsdam, Eric and Jeffrey Runner. 2001. Richard returns: Copy raising and its implications. CLS 37:453 468. Chicago. Radford, Andrew. 2004. Minimalist syntax: Exploring the structure of Englis h United Kingdom: Cambridge University Press. Rao, Anuradha. 2002. Null objects in Telugu and English: Merge, Binding Theory, and pro drop. Doctoral dissertation, Center Institute of English and Foreign Languages, Hyderabad, India. Rizzi, Luigi. 1997. The fine structure of the left periphery. In Elements of grammar ed. Liliane Haegeman, 281 337. Netherlands: Kluwer. Rizzi, Luigi. 2005. On some properties of subjects and topics. Paper presented at Incontro di Grammatica Generativa XXX Roberts, Ian and An ders Holmberg. 2005. On the role of parameters in Universal Grammar: a reply to Newmeyer. In Organizing grammar: Linguistic Studies in honor of Henk van Riemsdijk ed. Hans Broekhuis, Norbert Corver, Riny Huybregts, Ursula Kleinhenz, and Jan Koster, 538 55 3. Berlin: Mouton de Gruyter. Rodriguez, Cilene A.N. 2004. Impoverished Morphology and A movement out of case domains. Doctoral dissertation, University of Maryland, College Park.

PAGE 305

305 Rogers, Andy. 1971. Three kinds of physical perception verbs. Paper presente d in CLS 17, University of Chicago, Illinois. Rosenbaum, Peter S. 1967. The grammar of English predicate complement construction Cambridge, Mass.: MIT Press. Ross, John R.1967. Constraints on variables in syntax. Doctoral dissertation, MIT, Cambridge, Mas s. Rothstein, Susan. 2001. Predicates and their subjects. Dordrecht: Kluwer. Roussou, Anna. 2001. Control and raising in and out of subjunctive complements. In Comparative syntax of Balkan languages ed. Maria Luisa Rivero and Angela Rali, 74 104. New York : Oxford University Press. Sabel, Joachim. 2000. Expletives as features. In WCCFL 19 Proceedings ed. Roger Billerey and Brook Danielle Lillehaugen, 411 424. Somerville: Cascadilla Press. San Martin, Itziar. 2004. On subordination and the distribution of P RO. Doctoral dissertation, University of Maryland, College Park. Sells, Peter. 1984. Syntax and semantics of resumptive pronouns. Doctoral dissertation. University of Massachusetts, Amherst. Shlons ky, Ur. 1992. Resumptive pronouns as Last Resort. Linguisti c Inquiry 23:443 468. Sigur sson, Halldr rmann. 1991. Icelandic case marked PRO and the licensing of lexical arguments. Natural Language and Linguistic Theory 9 : 327 363 Sigur sson, Halldor. 2004. Icelandic non nominative subjects. In Non nominative subje cts Vol. 2, ed. Peri Bhaskararao and Karumuri Venkata Subbarao, 137 160. Amsterdam: Jo hn benjamin Publishing Company. Steever, Stanford. 1988. The serial verb formation in the Dravidian languages Delhi: Motilal Banarsidass. Steever, Sanford. 1997. Introdu ction to the Dravidian languages. In The Dravidian languages ed. Sanford Steever, 1 39. New York: Routledge. Stepanov, Arthur. 2001. Late adjunction and Minimalist phrase structure. Syntax 4:94 125. Stowell, Tim. 1982. The tense of infinitives. Linguistic Inquiry 1:561 570 Stroik, Thomas S. 1996. Extraposition and expletive movement: A Minimalist account. Lingua 99:237 251.

PAGE 306

306 Subbarao, Karumuri Venkata 2004. Tense and case in backward control in Mizo, Telugu, and Assamese. Handout, University of Delhi and Un iversity of Hamburg. Subbarao, Karumuri Venkata and B. Lalitha Murthy. 1999. Lexical anaphors and pronouns in Telugu. In Lexical anaphors and pronouns in selected South Asian languages ed. Barbara Lust, Kashi Wali, James Gair, and Karumuri Venkata Subbara o, 217 276. Berlin: Mouton de Gruyter. Svenonius, Peter. 2002. Introduction. In Subjects, expletives, and the EPP ed. Peter Svenonius, 107 124. New York: Oxford University Press. Tserdanelis, Georgios and Wai Yi Peggy Wong. 2004. Language Files: Materials for an introduction to language and linguistics. Columbus: The Ohio State University Press. Uriagereka, Juan. 1999. Multiple Spell Out. In Working Minimalism ed. Samuel David Esptein, and Norbert Hornstein. Cambridge, Mass.: MIT Press. Vergnaud, Jean Rog er. 1982. Dpendances et niveaux de reprsentation en syntaxe. Doctoral dissertation, Universit de Paris VII. Watanabe, Akira. 1996. Case absorption and wh agreement Dordrecht: Kluwer. Williams, Edwin. 1980. Predication. Linguistic Inquiry 15:203 238. W illiams, Edwin. 1992. Adjunct Control. In Control and grammar ed. Richard K. Larson, Sabine Iatridou, Utpal Lahiri, and James Higginbotham, 297 322 Dordrecht: Kluwer. Williams, Edwin. 1994. Thematic structure in syntax. Cambridge, Mass.: MIT Press. Woolf ord, Ellen. 2006. Lexical case, inherent case, and argument structure Linguistic Inquiry 37:111 130. Yoon, James H. 2004. Non nom inative (major) subjects and case stacking in Korean. In Non nominative subjects: Volume 2 ed. Karumuri Venkata Subbarao and Peri Bhaskararao, 265 314. Amsterdam: John Benjamin Publishing Company.

PAGE 307

307 BIOGRAPHICAL SKETCH Youssef A. Haddad was born in Beirut, Lebanon on September 12 th 1972. In 1997, he graduate d from the Lebanese University, Faculty of Letters and Human Sciences with a Licence D'Enseignement in English Language and Literature. He earned a Diplme D'tudes Suprieures in Education from the Leb anese University, Faculty of Education in 2001. He was awar d ed a Fulbright scholarship to University of Texas at Austin in 2000. He embarked on a Ph.D. degree in linguistics at the University of Florida in August 2004. During his study, he taught for the D epartment of African and Asian Languages and Literatures from 2004 to 2005. He was a research assistant to Dr. Eric Potsdam from 2005 to 2006. He taught for the Linguistics Program from 2006 to 2007. He received the College of Liberal Arts and Sciences Awa rd for International Student with Outstanding Academic Achievement in 2006 He was also awarded the Russell Dissertation Fellowship in Spring 2 007 He will join the Department of Modern Languages and Li nguistics at Florida State University upon his graduat ion in August 2007.