Argument Structure in Hindi (Dissertations in Linguistics) [74 ed.] 1881526445, 9781881526445

This book argues for a conception of linguistic organisation involving the factorisation of syntactically relevant infor

139 90 2MB

English Pages 269 [288] Year 1994

Report DMCA / Copyright

DOWNLOAD FILE

Polecaj historie

Argument Structure in Hindi (Dissertations in Linguistics) [74 ed.]
 1881526445, 9781881526445

Citation preview

Contents Author’s Note: June 1993 / vii Preface / viii Contents /xii Abbreviations / xvi 1.

Introduction / 1 1.1. The Organization of Lexical Information / 1 1.2. The Specification of Predictable Information / 2 1.3. Co-presence of Information / 3 1.4. An Outline of the Work / 4

2.

An Introduction to Hindi Syntax / 5 2.1. Nominals / 5 2.2. Verbals / 7 2.3. Word Order / 11

3.

The Conceptual Framework / 15 3.1. The Organization of Information / 15 3.2. Semantic Structure / 18 3.2.1. Two Aspects of Meaning / 19 3.2.2. Are Elements of sem str Visible to Syntax? / 21 3.3. Argument Structure / 22 3.3.1. Valency and Relative Prominence / 22 3.3.2. Representation of Thematic Role Information / 23 3.3.3. The Hierarchy of Arguments / 28 3.3.4. The Mapping Between sem str and arg str / 31 3.3.4.1. Multiply Associated Entities / 32 3.3.4.2. Unassociated Entities / 36 3.3.5. The Notion Logical Subject / 38 3.3.6. The Notion Logical Object / 39 3.3.7. Summary / 41 3.4. Grammatical Function Structure / 42 3.4.1. The Internal Organization of gf str / 44 3.4.2. The Mapping Between arg str and gf str / 46 3.5. Grammatical Category Structure / 49 3.6. Co-presence of the Levels of Structure / 51 3.7. Summary / 54

4.

The Case System / 56 4.1. The Case Marking System / 60 4.2. Case Features: Their Syntactic and Semantic Substance / 65

4.3. The Syntax of Direct Case in Hindi / 70 4.3.1. ergative and nominative on the Subject / 70 4.3.2. genitive on the Subject / 79 4.3.3. accusative and nominative on the Object / 80 4.4. Case Preservation: Direct and Indirect Case / 92 4.4.1. Preservation of dative / 94 4.4.2. Accusative Preserving Dialects / 95 4.4.3. Objects of "Unaccusative Transitives" / 98 4.5. Summary / 100 5.

Verb Agreement and Word Order / 103 5.1. Verb Agreement : The Simple Facts / 103 5.2. Agreement and an Incorporation Construction / 107 5.2.1. The Incorporation Construction: Syntactic Properties / 107 5.2.2. An Analysis: A Noun+Verb Lexical Compound / 111 5.2.3. Phonological and Morphological Evidence / 113 5.2.4. Agreement with the Incorporated Object / 118 5.3. Concluding Remarks / 119

6.

Grammatical Subjecthood / 122 6.1. Semantic and Coding Properties of Subjects / 6.2. Behavioral Properties of Subjects / 6.2.1. The Reflexive apnaa / 6.2.2. Subject Obviation / 6.2.3. Participial Control Clauses / 6.2.4. Gapping in Coordinate Constructions / 6.2.5. A Summary of the Diagnostics / 6.3. Grammatical Subjecthood as Prominence / 6.4. Concluding Remarks /

7.

Indirect Case on Subjects / 7.1. Dative Subjects / 7.1.1. Dative Logical Subjects / 7.1.2. Their Grammatical Subjecthood / 7.2. The Instrumental Subject Construction / 7.2.1. Instrumental Logical Subjects / 7.2.2. Their Grammatical Subjecthood / 7.3. Locative Subjects / 7.3.1. Case and Spatial Relations / 7.3.2. Semantic Fields / 7.3.3. The Extensions of Case to Non-Spatial Relations / 7.3.4. Case Features, Case Markings, and Case Meanings / 7.3.5. Locative Logical Subject as Grammatical Subject / 7.4. Genitive Subjects / 7.4.1. Genitive Logical Subjects / 7.4.2. Their Grammatical Subjecthood / 7.5. Accusative Logical Subjects / 7.5.1. The Accusative Logical Subject Construction /

7.5.2. A Subjectless Construction / 7.5.3. A Formal Representation / 7.5.4. Implications / 7.6. Theoretical Consequences / 8.

Complex Predicates / 8.1. Introduction / 8.2. The Category Structure of Complex Predicates / 8.2.1. The Nominal as Part of the Verbal Constituent / 8.2.2. A Complex Predicate as a Phrasal Category / 8.2.3. The Nominal Host as a Lexical Category / 8.2.4. Summary / 8.3. Predicatehood in Complex Predicates / 8.3.1. The Nominal Host as an Independent Predicate / 8.3.2. The Complex Predicate as a Single Predicate / 8.3.3. Putting the Pieces Together / 8.4. The Argumenthood of the Nominal Host / 8.4.1. Verb Agreement / 8.4.2. Passivization / 8.4.3. Absence of Agreement: an apparent anomaly / 8.5. Summary: Structural Mismatches / 8.6. The Notion "Lexical" /

9.

Conclusion / 9.1. Levels of Structure / 9.2. Co-presence of the Levels /

References /

PREFACE

I came to Stanford four years ago committed to becoming a phonologist and terrified of syntax. The process of writing my first qualifying paper on Malayalam causatives, together with Joan Bresnan's first course on the Lexical Mapping Theory in Fall 1987, made me realize that syntax was not as forbidding and mysterious as I had thought. I began to explore grammatical subjecthood in Hindi as a result of the discovery, made during a conversation with John Paolillo, of a construction with peculiar (and at that time bewildering) semantic and syntactic properties. From that point on, I have become increasingly convinced that syntax can be as much fun as phonology! The outcome of the exploration is a phonologist's view of syntax. A common strategy of theory construction in linguistics is to design a theory on the basis of intuitions built up in the course of one's experience with linguistic data; to look for data and analyses that provide evidence for the theory; and to minimally modify the theory when required by new evidence. Another strategy is to construct a grammar that accounts for a large body of data from a language, unearthing the theoretical assumptions implicit in the grammar, and allowing the theory to emerge in the process of grammar construction. I have found myself adopting the latter strategy in my own research. This thesis articulates the theory of argument structure that I believe underlies the patterns of Hindi syntax. Theses cannot be written in solitude. There are numerous people who have contributed to this one in numerous ways, both direct and indirect. I cannot hope to name them all, or acknowledge them sufficiently. Many who have influenced or contributed to my work may have gone unacknowledged. For this, I beg indulgence. For most graduate students, the first year in graduate school is a harrowing experience. I would not have survived the ordeal but for the friendship and support of my classmates, particularly Cheri Garcia, Kathryn Henniss, and Marcia Hunt. I am also fortunate to have been part of the CSLI Room 210 research group, which included Alex Alsina, Ki-Sun Hong, Smita Joshi, Jonni Kanerva, Paul Kroeger and Tan Fu. Members of this group met regularly to share work in progress. This forum, which allowed us to try out finished analyses, half-baked ideas, or even wild thoughts that occur to the fevered minds of graduate students in half-sleep, gave me more training in syntax than all my

courses put together. We owe thanks to Joan Bresnan for forming the research nucleus, and for being its most regular member. During various stages that led to this thesis, I have profited from interaction with Yamuna Kachru, George Lakoff, Will Leben, Mithilesh Misra, Rajeshwari Pandharipande, John Paolillo, Gillian Ramchand, Tom Wasow, Steve Wechsler, Annie Zaenen, and Arnold Zwicky. I received invaluable comments on earlier versions from Alex Alsina, Lubna Alsagoff, Miriam Butt, Michael Inman, Tracy King, Paul Kroeger, and Yo Matsumoto. Akhil Gupta, Purnima Mankekar, and Ravi Oswal provided sensitive judgements on subtle distinctions in Hindi. I can't promise that I won't wake them up again at odd hours of the night with a linguistic crisis. Prathima Christdas performed an incredible feat of proofreading the thesis and checking the data at very short notice, in an unreasonably short time. Michelle Collette Murray and Gina Wein, with their friendly warmth, have generously helped me at every stage to avoid the administrative hassles of graduate school. Lubna Alsagoff and Jen Cole have been my rescuers in all kinds of emergencies. They, along with Aurora Ballester, Kathryn Henniss, John and Rozanne Stonham, and Linda Uyechi provided us with a thoughtful and caring community of friends on campus. The Kumar home was not only a source of Malayalam judgements but also a second home and a link with the world outside campus. Purnima and Akhil were not just informants, they bacame part of the family. Purnima was an indulgent audience as I thought out loud my conception of linguistic organization and philosophy of linguistics. As at MIT years ago, Professor and Mrs. Larrabee continued to be our “local guardians”, even from Long Beach. My work during the summer of 1988 and the academic year 1989-90 was funded through the Grammatical Theory and Discourse Structure project at the Center for the Study of Language and Information. A great many unacknowledged ideas in this thesis came from the members of my thesis committee. Paul Kiparsky never failed to provide stimulating challenges and to force me to consider alternative solutions. The section on noun incorporation, which would otherwise have been an uninteresting and insignificant set of observations, is the outcome of his penetrating questions. Peter Sells' enthusiasm for and interest in my work were an unending source of support and encouragement. With his "Japanese insights," he helped me articulate the intuitions I was struggling to express. He also guided me through the overwhelming mazes of syntactic literature. Bruce Pray cheerfully accepted the extra burden of serving as an outside member on my committee. With his scholarship and insights into the grammar of Hindi, he enriched my range of data

and analysis. Eve Clark and Carol Neidle were not officially involved with the thesis, but they read a complete draft, providing insightful comments, criticism, and corrections. There are some people without whom I would not have written this thesis. To Joan Bresnan, I owe my growth as a linguist. For four years, I relied on Joan's unfailing encouragement and interest. One of my proudest moments in graduate school came one day in my fourth year. During a discussion with Joan lasting an hour and a half, I was subjected to intense criticism, unsoftened by a word of approval or praise. For me, this was a sign that I had reached the stage in my growth at which I could withstand such criticism. Joan has not only been my thesis advisor, sharing her knowledge and insight, but also a teacher, mentor, and friend. She has often said, "If your idea is right, you will eventually find a simple way to express it." This I take with me as her blessing. To Dr. Jeff Torchia, I owe a greater debt of gratitude than I can express. His chiropractic help kept my back from collapsing under the strain of many long hours at the computer, and allowed me to focus my energy on my research. Jeff is a gentle healer and compassionate friend; he generously shared his good cheer and optimism, and gave me a more positive world-view. I see this thesis as the beginning of an exciting journey of exploration. During its writing, my husband, Mohanan, took charge of the kitchen, overlooked my unreasonable moods, pushed me relentlessly to higher standards, and was there to support me every time I was about to break down. Above all, he has been my constant companion, sharing the joy and wonder of the exploration, as well as the pain and frustration. The moral support and the prayers and blessings of our families half-way across the world gave me the strength to survive the struggles of graduate school. My brother Shashi's critique of a paper of mine has had a direct bearing on my writing style. The one who has made the greatest sacrifices so that this thesis could be written is our daughter Malavika. For four years, she patiently endured the endless hours of boring conversation and heated debates on linguistics, even during so-called "vacations". She has been the element of sanity and balance in our lives, actively reminding us on the one hand that there is more to life than work, and on the other that there are no excuses not to work. It is the angels she left on my shoulders every morning that have inspired this thesis to its completion.

1 Introduction At the heart of a great deal of research in syntactic theory lies the idea that given the meanings of individual verbs, the patterns of their syntactic distribution and alternation are largely predictable. This insight has been the driving force behind case grammar (Fillmore (1968)), generative semantics (Lakoff (1968; 1976), Lakoff and Ross (1976), McCawley (1968)), the projection principle in the Government Binding (GB) theory (Chomsky (1981)), and most recently, the various approaches that employ the "linking” mechanism (Hale (1983), Marantz (1984), L. Levin (1986), Rappaport and Levin (1986), Kiparsky (1987), Bresnan and Kanerva (1989)). A central concern in all these approaches is the role of verb meanings in constraining syntactic structure. Much of the research in this domain comes under the study of “argument structure". This dissertation presents an intensive study of some aspects of Hindi syntax. It explores the regularities governing the expression of the arguments of predicates in terms of case marking and agreement in Hindi, with special emphasis on the case of grammatical subjects, and the interaction of case and agreement with noun+verb compounds and noun+verb complex predicates. In doing so, it seeks to further our understanding of the nature of argument structure. Some of the main questions that have guided this research are: What are the levels of structure over which the semantic and syntactic information contained in the lexical entry of a predicate are distributed? Are there parts of this information that may be predicted from the other parts? What are the principles of correspondence that derive the predictable information? What is the organization of and the relation between the levels of structure required by the grammatical principles?

1.1. The Organization of Lexical Information A core component of the conception of grammar in this dissertation is the set of levels or dimensions of structure over which the information in the lexical entry of a predicate is distributed. We view the representation of a

1

2 / Arguments in Hindi

predicate as being organized into at least four levels of structure. Semantic structure (SEM STR) represents those aspects of meaning that are relevant for systematic morphological and syntactic regularities. This includes the meanings associated with the arguments of the predicate. Argument structure (ARG STR) encodes the valency (number of arguments) of the predicate in terms of argument slots, and their semantically determined relative prominence. Thus, the information in traditional theta role representations is distributed in this organization over SEM STR and ARG STR. Grammatical function structure (GF STR) expresses the grammatical functions of the arguments of the predicate in terms of notions like subject and object, and grammatical features such as case, number, gender, tense, and the like. Grammatical category structure (GC STR) represents information about grammatical categories such as noun and verb. SEM STR, ARG STR, GF STR, and GC STR form a set of interdependent, interacting levels of structure, linked to one another by general principles of grammar. The elements and relations at each of these four levels are drawn from a universal inventory. The four levels of structure associated with predicates at the word level are also present at the sentence level and discourse level organization. One may think of the representations at the sentence and discourse levels as being the result of putting together representations of words that carry partial specification of information at all the four levels of structure.

1.2. The Specification of Predictable Information An important goal of syntactic theory is to express the regularities of distribution within lexical representations, and the regularities of alternation between related pairs of lexical representations. One way of expressing these regularities is to factor out the idiosyncratic information in a predicate into an underspecified lexical representation, and to allow general principles of grammar to specify the predictable information in the completely specified lexical representation. In the lexical representation of a predicate, the grammatical functions, grammatical features, and categories associated with the arguments are largely predictable from the SEM STR and ARG STR information, although the reverse is not possible. Thus, there is an asymmetric dependence involved in the relation between the levels of structure. Therefore, we might express the SEM STR and ARG STR information in the underspecified repre-

Introduction / 3

sentation, and derive the GF STR and the GC STR properties of an argument through universal and language specific principles of correspondence. Discovering general principles of grammar that govern the mapping from argument structure to grammatical functions is the primary concern of the Lexical Mapping Theory (LMT) within Lexical Functional Grammar (LFG), the Direct Linking Theory in Kiparsky (1987; 1988), NP-movement within the Government Binding (GB) theory, and relation changing rules in Relational Grammar (RG). In this thesis, I articulate and defend certain specific assumptions about the relation between ARG STR and grammatical functions. However, the bulk of the work is devoted to discovering the principles that govern the mapping between SEM STR, ARG STR and grammatical functions on the one hand, and the grammatical features of case on the other, and placing these principles within the perspective of a theory of grammar. The case of an argument is predictable to a far greater extent than is often recognized, and may be dependent on the meaning of the argument, its grammatical function, or both. This study reveals that the principles of case selection can be quite independent of grammatical functions, even those such as subject and object.

1.3. Co-presence of Information Despite the derivationality involved in specifying predictable information via principles of grammar, I will argue that the four levels of structure are co-present in the resultant architecture of the representation. An assumption often implicit in the separation of levels of structure is that only principles that define the mapping or correspondence between two levels can refer to them simultaneously; all other principles are internal to a single level. In the course of the thesis, I hope to show that principles of grammar that state the regularities in natural languages may simultaneously access information in the various levels of structure, unless stipulated otherwise. The co-presence of levels of structure envisaged in this study crucially assumes that the relation between levels does not involve any structure changing mechanisms. Without this assumption, clearly, co-presence of levels of structure would involve a global mechanism.

4 / Arguments in Hindi

1.4. An Outline of the Work This study takes the following course of progress. Chapter 2 provides a preliminary introduction to the syntax of Hindi. Only those aspects are touched upon that are relevant for the discussion in subsequent chapters. Chapter 3 attempts to articulate the organization of grammar assumed in this thesis. In the history of this study, the theoretical assumptions emerged in the process of a detailed exploration of Hindi syntax in several cycles. In presenting the study, I reverse the order, and first spell out the independent assumptions that constitute the theoretical perspective. The subsequent chapters show how the theoretical assumptions and the analysis of the syntactic phenomena in Hindi illuminate each other. Chapter 4 outlines the case system of Hindi, and provides a partial analysis of the system. From this account of case emerges a possible set of assumptions in a universal theory of case, which postulates a distinction between direct and indirect case association, and between semantic and non-semantic case association. Chapter 5 shows that the facts of Hindi verb agreement follow from the statement that the verb agrees with the highest nominative argument. It also addresses a puzzle presented by the facts of verb agreement in a construction involving noun incorporation, and presents a solution to the puzzle by recognizing the distinction between a categorial word and a functional word. Chapter 6 outlines the properties of behavior whereby the grammatical subject in a clause in Hindi is identifiable. Chapter 7 employs the diagnostics derived in Chapter 6 to show that a grammatical subject in Hindi may bear almost any of the case markings available in the language. The choice of case is predictable on the basis of meaning. This demands principles of case selection independent of grammatical functions, supporting the theory of case outlined in Chapter 4. It also demonstrates the need for the syntactic principles of case association to have direct access to semantic information. Chapter 8 examines the organization of SEM STR, ARG STR, GF STR, and GC STR in constructions with complex predicates composed of a nominal host and a light verb, and offers an account of the apparently anomalous patterns of case marking and agreement in these constructions. Finally, Chapter 9 lays out the theoretical conclusions that emerge in the course of the study.

2 An Introduction to Hindi Syntax This chapter provides a general introduction to some aspects of Hindi syntax. Subsequent chapters will explore some of them in detail. I hope that the overall view of the relevant patterns of the language sketched here will help the reader to place in the larger picture of the language the patterns studied in detail in the later chapters.1 Before we proceed, a note on the language and the data. Hindi is an Indo-Aryan language spoken in parts of Northern India. In discussions about this language, one often comes across the terms Hindi, Urdu, Hindustani, and Hindi-Urdu. Within a scheme that distinguishes these varieties of the language, "Hindi" refers to the variety that exhibits a strong Sanskrit influence, and "Urdu", a strong Perso-Arabic influence, particularly in vocabulary. "Hindustani" refers to the variety that is neutral to Sanskrit and Persian, and avoids heavy borrowing from either. The term "Hindi-Urdu" is perhaps used to express what is common to Hindi and Urdu in a neutral way.2 The term Hindi in this thesis is not intended to differentiate one variety from another. In this study, I follow the assumption that the object of inquiry in generative grammar is the mental linguistic system of an individual language user. As Chomsky points out, the linguistic systems of no two individuals are identical, not even twins brought up in the same community (1988:36-37). Under this view, our object of inquiry is not the "Hindi" language, which is a social entity. Rather, it is the internal linguistic system of an individual Hindi speaker.3 It is to be expected that the internal linguis1 For detailed grammars of Hindi, see Guru (1922), Sharma (1958), Srivastava (1969), McGregor (1972), Kachru (1966; 1980). 2 For more detailed information on the language, see Kachru (1980; 1987). 3 I have based the description in this thesis on my own judgements. However, in order to eliminate the danger of judgments being influenced by the theory, I only use judgements that at least three other native speakers of Hindi agree on, and as far as possible, the same three speakers. The variety of Hindi spoken by these individuals is perhaps best described as an informal variety used by urban educated speakers of Hindi as a first language.

6 / Arguments in Hindi

tic systems of other individuals belonging to the Hindi speaking community will differ in minor as well as major ways from the system described here. Given the stated object of inquiry, the variability among speakers of the Hindi "language" does not invalidate the enterprise. In what follows, I briefly outline some of the basic properties of the nominal system, the verbal system, and word order in Hindi.

2.1. Nominals 2.1.1. One of the most notable features of the nominal system of Hindi is its rich sub-system of case. Consider the following example:4 (1)

raam-ne ilaa-se bacce-ko gaanaa sik¥vaayaa. Ram-E Ila-I child-A song- N learn-C-C-PERF Ram made Ila teach the child a song.

Each of the four nominals in (1) bears a distinct case. A subject in Hindi may bear nominative, ergative, dative, instrumental, locative or genitive case. The object of a diadic verb may be either accusative or nominative. A detailed account of the case system of Hindi is reserved for Chapters 4 and 7. 2.1.2. Hindi employs a two-way number system of singular (SG) and plural (PL), generally expressed through inflection. It also has a two-way gender system of masculine (M) and feminine (F). In animate nouns, grammatical gender corresponds to the sex of the referent. In inanimate nouns, however, gender is arbitrary. Thus, the gender of prad¥aan mantrii "prime minister" depends on whether the referent is male or female. In contrast, ããk¥ "eye" is feminine, while oÆ¥ "lip" is masculine. The pronominal system of Hindi does not encode gender distinctions, but it does reflect the two distinctions of number, as well as a three-way person distinction. 2.1.3. Possessive and adjectival modifiers agree with their head noun in number and gender. They also reflect the nominative versus non-nominative case distinction of the head nominal. Modifier agreement is briefly illustrated in (2). Not all modifiers exhibit surface alternations in the ending. 4 For expansions of the abbreviations in the glosses, see List of Abbreviations. If the patterns behind the glosses for case are unclear at this point, they will become apparent when we discuss the case system at length i n Chapter 4.

An Introduction to Hindi Syntax / 7

When they do, however, the aa ending signals masculine singular nominative ((2a)), e signals masculine plural ((2b)) or masculine non-nominative ((2c)), and ii signals feminine ((2d)). (2) a.

puraanaa old-M.SG-N

kap±aa cloth-M .SG-N

"old cloth"

b.

puraane old-M.PL-N

kap±e cloth-M .PL-N

"old clothes"

c.

puraane kap±e-se old-M.SG-NN cloth-M .SG-I

"with the old cloth"

d.

puraanii old-F.SG-N

"old chair (s)"

kursii ( yãã ) chair-F.SG-N (PL)

Modifier agreement is briefly discussed in the context of case agreement in Chapter 4. 2.1.4. Hindi is a pro-drop language, in which subjects and objects can be omitted. Thus, the question fayil b¥ejaa? "Sent file?" can mean, "Has she sent the file to him?" ilaa-ko b¥ejaa? "Sent to Ila?" can mean, "Has he sent it to Ila?" And b¥ejaa? "Sent?" can mean, "Has she sent it to him?"

2.2. Verbals 2.2.1. In the Hindi verbal system, auxiliaries and modals concatenate with either the verb root or its different inflected forms to yield various distinctions of tense, aspect, mood and voice. Consider first the forms derived through inflection from the verb root. The sentences in (3) contain verbs of the form V+(y)aa which is the perfective (PERF) aspect, V+taa which may be termed the imperfective (IMPERF) aspect, and V+naa which is a nonfinite (NF) form: (3) a.

raam-ne pahale b¥ii yah giit gaayaa hai. Ram-E before also this song- N sing-PERF be-PRES Ram has sung this song before.

7

8 / Arguments in Hindi

b. raam hameßaa yah giit gaataa hai. Ram-N always this song- N sing-IMPERF be-PRES Ram always sings this song. c.

raam-ne yah giit gaan aa Ram-E this song- N sing-NF Ram wished to sing this song.

caahaa. desire-PERF

Continuous or progressive (PROG) aspect is indicated by an auxiliary that is homonymous with the main verb rah "stay", cooccurring with the copula ho "be" which carries the tense information: (4)

raam gaa rahaa Ram-E sing PROG Ram was / is singing.

t¥aa / hai. be-PA be-PRES

The PERF marking, when not accompanied by a tense bearing auxiliary, signals past tense. The simple past and future forms in Hindi are illustrated in (5a, b). What corresponds to the simple present is given in (5c): (5) a.

raam-ne darvaazaa Ram-E door-N Ram opened the door.

k¥olaa. open-PERF

b. raam darvaazaa k¥olegaa. Ram-N door-N open-FU Ram will open the door. c.

raam darvaazaa k¥oltaa hai. Ram-N door-N open-IMPERF be-PRES Ram opens the door.

Like the auxiliary rah which expresses progressive aspect, an auxiliary that is homonymous with the main verb jaa "go" has several uses in Hindi. Two of the uses of this auxiliary are indicated in (6b, c): (6) a.

raam vah gaanaa Ram-N that song- N Ram will sing that song.

gaaegaa. sing-FU

An Introduction to Hindi Syntax / 9

b. raam vah gaanaa gaataa jaaegaa. Ram-N that song- N sing-IMPERF go-FU Ram will go on singing that song. c.

vah gaanaa gaayaa jaaegaa. that song- N sing-PERF go-FU That song will be sung.

(6a) and (6c) demonstrate the active-passive alternation in Hindi. Thus, while the auxiliary jaa in (6b) adds aspectual meaning to the main verb, in (6c) it expresses the voice alternation with (6a). Like jaa and rah , Hindi has a large number of verbs that serve an auxiliary or modal function. In addition, some of these verbs combine productively with other verbs, adjectives and nouns to form complex predicates. In doing so, the verb may lose to varying degrees the syntactic and semantic structure that is associated with its main verb counterpart. Such a verb has been called a "light verb", and the entity it combines with is its "host".5 In (7) are given instances of light verbs in combination with a verbal host, adjectival host, and nominal host respectively: (7) a.

raam gaa uƥaa. Ram-N sing rise-PERF Ram sang out spontaneously (burst into song).

b. raam-ne kamraa host) Ram-E room-N Ram cleaned the room.

(verbal host)

saaf kiyaa.

c. raam-ne ravii-kaa piic¥aa host) Ram-E Ravi-G Ram followed/chased Ravi.

clean

(adjectival do-PERF

kiyaa. pursuit-N

(nominal do-PERF

In the literature on Hindi, the verb+verb complex predicates are often referred to as compound verbs, and the noun+verb complex predicates as conjunct verbs. Complex predicates with nominal hosts like the one in (7c) are examined in detail in Chapter 8.

5

The terminology is discussed further in chapter 8.

9

10 / Arguments in Hindi

2.2.2. A large number of verbs encode transitive-intransitive alternations in terms of vowel alternations. Given in (8)-(10) are examples of verb roots that exhibit such an alternation: (8)

a.

k¥ul "open (intr.)"

b.

k¥o l "open (tr.)"

(9)

a.

ruk "stop (intr.)"

b.

rok "stop (tr.)"

(10)

a.

mu± "turn (intr.)

b.

mo± "turn (tr.)”

These may be treated as causative alternations. The verb also expresses affixal alternations between causative and non-causative forms. There are two causative morphemes, one which generally indicates direct causation, and the other, indirect causation. The verbs in (11b) and (11c) are the direct and indirect causative forms respectively of the verb in (11a). (11) a.

raam-ne gaanaa Ram-E song- N Ram learned the song.

siik¥ a a . learn-PERF

b. ilaa-ne raam-ko gaanaa sik¥ a a y a a . Ila-E Ram-D song- N learn-C-PERF Ila taught Ram the song. c.

maa-ne ilaa-se raam-ko gaanaa mother-E Ila-I Ram-E song- N Mother made Ila teach Ram the song.

sik¥vaayaa. learn-C-C-PERF

2.2.3. The Hindi verb exhibits agreement in number, gender, and person with its subject if it is nominative ((12a)). If the subject is not nominative, the verb agrees with the object if that is nominative ((12b)). The agreement features are expressed in the final vowel on the verb: (12) a.

raam kitaabã Ram-M.SG-N book-F.PL-N Ram will buy books.

b.

raam-ne Ram-M.SG-E Ram bought books.

kitaabã book-F.PL-N

k¥ariidegaa. buy-FU-M.SG

k¥ariidññ. buy-PERF-F.PL

An Introduction to Hindi Syntax / 11

The verb in (12a) agrees with the masculine singular nominative subject. In (12b), the subject is not nominative. The verb therefore agrees with the feminine plural nominative object. Since person agreement occurs only when the controller of agreement is a pronoun, we will ignore it for the purposes of this thesis. Verb agreement is discussed in detail in Chapter 5. In a verbal complex containing the main verb and auxiliaries, modals or light verbs, every element in the complex that is not in the root form reflects the agreement features, as in (13): (13) a

raam gaataa rahtaa t¥aa. Ram-M.SG-N sing-IMPERF-M.SG stay-IMPERF-M.SG be-PA-M.SG Ram used to keep singing.

b. ilaa gaatii rahtii Ila-F.SG-N sing-INPERF-F.SG stay-IMPERF-F.SG Ila used to keep singing.

t¥ii. be-PA-F.SG

2.3. Word order English is considered to be a "fixed word order" language in contrast to, say, Warlpiri, a Central Australian language, which is a "free word order" language (Nash (1980), Hale (1983), Simpson (1983)). Hindi, perhaps along with most other South Asian languages, lies somewhere between English and Warlpiri on the continuum of word order freedom. 2.3.1. In Hindi, the word order of grammatical functions like subjects and objects is largely free. Clause internal changes in precedence do not necessarily result in grammatical function differences. The freedom of word order is illustrated in (14): (14) a.

ilaa-ne anuu-ko haar b¥ejaa. Ila-E Anu-D necklace-N send-PERF Ila sent Anu a/the necklace.

b. ilaa-ne haar anuu-ko b¥ejaa Ila-E necklace-N Anu-D send-PERF Ila sent Anu the/*a necklace.

11

12 / Arguments in Hindi

c.

haar ilaa-ne anuu-ko b¥ejaa . necklace-N Ila-E Anu-D send-PERF Ila sent Anu the/*a necklace.

d.

ilaa-ne b¥ejaa anuu-ko haar. Ila-E send-PERF Anu-D necklace-N (It was) Ila (who) sent Anu the/a necklace.

Illustrated in (14) are some of the scrambling possibilities of sentences with no changes in the grammatical functions of the nominals.6 2.3.2. Of the various permitted orders of the sentences in (14), however, there is one, namely (14a), that is "canonical", "unmarked", or in the terminology of Jakobson (1963:268), "stylistically neutral". The other orders are permitted deviations from the canonical order in (14a). Such deviations are generally accompanied by non-neutral stress and intonation, differences in presupposition, and shifts in prominence, emphasis, and other discourse effects.7 One cue of non-canonical order easiest to tease out in Hindi is that of definiteness. For example, the preferred interpretation of the object haar "necklace" in (14a) is as indefinite; but it may be interpreted as definite, unless the context disallows such an interpretation. However, displacing this object from its canonical position immediately preceding the verb, as in (14b, c), tends to restrict the interpretation of the object to definite. In contrast, a subject which may marginally be interpreted as indefinite as in (15a), is obligatorily definite in the position immediately preceding the verb, because it is displaced from its canonical clause-initial position: (15) a.

sunaar-ne anuu-ko haar b¥ejaa. goldsmith-E Anu-D necklace-N send-PERF The/?a goldsmith sent Anu a/the necklace.

b. anuu-ko haar sunaar-ne b¥ejaa. Anu-D necklace-N goldsmith-E The/*a goldsmith sent Anu the/*a necklace.

send-PERF

6 I use the term "scrambling" to refer to the phenomenon of multiple possibilities of word order rather than to the formal operation of scrambling. 7 The notion of canonical word order is essential for an adequate analysis of a large number of syntactic phenomena involving definiteness effects, case marking, reversal, and locative inversion in Hindi (T.Mohanan (1989)).

An Introduction to Hindi Syntax / 13

The contrast in the definiteness effects of the subject and the object in identical positions shows that the effect is not a property of the position itself, but has to do with displacement from canonical position. 2.3.3. The scrambling effects witnessed in (14)-(15) are not restricted to simple sentences. They extend to more complex constructions including embedded non-finite clauses. For instance, the argument of an embedded non-finite clause may scramble with the arguments of the matrix clause. Consider (16), where the elements of a non-finite clause are in bold face: (16) a.

ilaa-ne mohan-se Ila-E Mohan-COM raam-kaa s a a m a a n kamre-mã rak¥ne-ko Ram-G luggage-N room-L keep-to Ila told Mohan to keep/put Ram's luggage in the room.

kahaa. tell-PERF

b. mohan-se raam-kaa saamaan ilaa-ne kamre-mã rak¥ne-ko kahaa. c.

mohan-se raam-kaa saamaan kamre-mã ilaa-ne rak¥ne-ko kahaa.

d.

ilaa-ne raam-kaa saamaan mohan-se kamre-mã rak¥ne-ko kahaa.

In (16a), all the constituents are in the canonical order. (16b, c, d) have constituents of the matrix clause intervening between the parts of the embedded clause. Thus, the domain of scrambling in Hindi, as in a language like Warlpiri, is the finite clause. Now, whereas parts of a non-finite clause may be scrambled with respect to parts of the matrix clause, this is not true for other constituents. Unlike in Warlpiri, the parts of a noun phrase, for instance, cannot be scrambled, as shown by (17b, c). Once again, the parts of the constituent are given in bold face: (17) a.

ilaa-ne raam-se anuu-kii puraanii k i t a a b k¥ariidii. Ila-E Ram-I Anu-G old book- N buy-PERF Ila bought Anu's old book from Ram.

b. * ilaa-ne

a n u u - k i i raam-se

puraanii k i t a a b

c. * raam-se

a n u u - k i i p u r a a n i i ilaa-ne

kitaab

13

k¥ariidii. k¥ariidii.

14 / Arguments in Hindi

2.3.4. We might account for the above facts by assuming that the direct daughters of S are order-free, and that what is functionally a non-finite clause need not correspond to a single constituent in terms of categories. In other words, the functional constituency and the categorial constituency of a sentence such as in (16a) are not identical. The two constituencies may be given as "co-present" structures as in (18). The structure above the sentence in (18) is the functional constituency, while that below the sentence is the categorial constituency: (18)

CLAUSE

OBL 8

SUBJ

OBJ

PRED

CLAUSE

________________________________ ilaa-ne Ila

mohan-se Mohan

raam-kaa saamaan Ram's luggage

kamre-mã rak¥ne-ko kahaa. in the room to keep told

_______________

NP NP

NP

NP

V

V

S

A consequence of (18) is that Hindi does not have a VP node that reflects the subject-non-subject asymmetry in terms of categories. I will not present an analysis of word order in this thesis. However, the observation that scrambling can apply across non-finite clauses, and the assumption that the direct daughters of S are order-free, has bearing on the categorial structure of complex predicates discussed in Chapter 8.

8 The function of the argument is given as OBLIQUE only for the sake of concreteness. Finding the exact function of the argument needs more study.

3 The Conceptual Framework This chapter articulates the conception of multidimensional organization of grammar I assume. This conception is not unlike the Halle and Vergnaud (1980) conception of organization in phonology. The purpose of this chapter is to give the reader an intuitive sense of the constructs used in the analysis of Hindi, and the need for these constructs.

3.1. The Organization of Information Around the argument structure of a predicate is built a rich collection of information, partly regular and partly idiosyncratic. Take for instance the English WORD STRING lifts, which is the result of the concatenation of the morphemes lift and -s.1 We know that the PREDicate lift is a Verb (information about Grammatical Category STRucture); that it takes two ARGuments (information about its ARGument STRucture); and that one argument, the "lifter", CAUSEs an UPward MOVEment (roughly speaking) of the other argument, the "liftee" (information about SEMantic STRucture). Suppose we schematize this information as follows:2 (1) [ X

EVENT CAUSE ] _____________ SEM STR [ Y [UP ] MOVE ] ……………………………………………………………………… ˘ ARG ARG ¿ PRED ARG STR ……………………………………………………………………… CAT: V GC STR

……………………………………………………………………… lift WORD STRING 1 I use the term WORD STRING to refer to a string of one or more words. 2 The convention of horizontal dotted lines is used throughout to separate levels of structure.

15

16 / Arguments in Hindi

We know that the morpheme -s is an affix that attaches to a verb, and that the verb is in PRESent tense. This affix also registers, through AGReement features, that the subject of the verb is third person singular (the verb agrees with the subject in these features). In (2) is given the representation of -s with all this information: (2)

TENSE: PRES AGR NUM: SG PER : 3

……………………………………………… CAT : V ……………………………………………… ———— -s

GF STR

GC STR WORD STRING

The information in (1) and (2) is idiosyncratic to the verb lift and the affix s. It must be represented in the lexical entry of these morphemes, and must be learnt individually. Let us call the representation that minimally specifies this information the underspecified lexical representation. We also know the following about the verb lifts. In a clause that contains this predicate, the lifter argument is the SUBJect, and the liftee argument is the OBJect (information about Grammatical Function STRucture). The liftee argument must follow the verb, while the lifter argument must precede the verb. This information is not discussed or represented here, because we will not be concerned directly with word order in this dissertation. The lifter argument is in NOMinative case, and is third person singular; the liftee argument is in ACCusative case. However, given the underspecified representation of lift and -s, we can predict this information. Such predictable information, we assume, need not be stipulated in the underspecified representation. It can be derived via principles of the grammar, which build the predictable information around the underspecified representation to yield the fully specified representation. The fully specified representation of lifts is as in (3). The representation in (3) results from the concatenation of the underspecified representations of lift and -s , and the principles of association that map the idiosyncratic information of these two morphemes to the fully specified lexical representation. Just as the principles of association yield the representation of words from the representation of morphemes, the concatenation of words and further principles of association yield the representations of phrases, sentences, and larger pieces of discourse.

The Theoretical Framework / 17

(3)

[

EVENT CAUSE ] _____________ SEM STR [ Y [UP ] MOVE ] ……………………………………………………………………………… X

˘ ARG ARG ¿ RED ARG STR ……………………………………………………………………………… SUBJ OBJ NOM ACC NUM: SG PER : 3

PRED GF STR TENSE: PRES AGR NUM: SG PER : 3

……………………………………………………………………………… CAT: V GC STR ……………………………………………………………………………… lifts

WORD STRING

The idea of association of two pieces of information by principles of grammar is what has come to be known as the linking mechanism. The term "linking" signifies a non-structure-changing relation between levels of structure. Linking in Hale (1983) (Government Binding Theory, henceforth GB) refers to the association between LS at s-structure and PS . In terms of the schema in (3), this is the association between GF STR and GC STR. Marantz (1984) assumes a similar relation between all related levels of structure. In the Lexical Mapping Theory (LMT) within Lexical Functional Grammar (LFG) (L. Levin (1986; 1987), Bresnan and Kanerva (1989), Bresnan and Moshi (1990), Bresnan (1990), Alsina and Mchombo (1988), Alsina (1990)), the linking mechanism is primarily used to capture the regularities in the mapping of arguments onto grammatical functions by virtue of their "thematic roles". In terms of (3), this is the association between ARG STR and SEM STR on the one hand and GF STR on the other. In Kiparsky (1988), "linking is understood as the licensing, in virtue of case, agreement, or position, of the association of Th-roles to syntactic arguments at a level of Lexical Structure (essentially analogous to the NP-structure of Riemsdijk and Williams 1981)." Given that Lexical Structure at sstructure in Chomsky (1981) and Hale (1983) and NP-structure in Riemsdijk and Williams (1981) are equivalent to the subsystem of grammatical functions in what we have called GF STR, the focus in Kiparsky's enterprise is identical to that in LMT, except that the linking must be licensed by the subsystem of grammatical features in GF STR. In the representation in (3), the information about the predicate lifts is distributed over four LEVELS OF STRUCTURE, called SEM(ANTIC)

18 / Arguments in Hindi

STR(UCTURE), ARG(UMENT) STR(UCTURE), G(RAMMATICAL) F(UNCTION) STR(UCTURE), AND G(RAMMATICAL) C(ATEGORY) STR(UCTURE). I as-

sume that this multistructural organization, present in the representation of morphemes as in (1)-(2), and that of words as in (3), is present in the representation of sentence and discourse structures as well. A primary purpose of this thesis is to discover the principles that govern the association of case with ARG(UMENT)s in Hindi. This association, as we will see, is rigorously conditioned by the SEM STR configurations that the ARG is associated with, and to some extent, by its GF as well. This calls for an examination of the relation between SEM STR, ARG STR, and GF STR in constructions involving simple as well as complex predicates. From this study emerge a number of fundamental assumptions about the organization of the levels of structure. The enterprise in this chapter is to spell out these theoretical assumptions, which form the basis for the analysis of Hindi in the subsequent chapters. The discussion progresses as follows. §3.2.- §3.5. lay out the internal organization of the four levels of structure over which the information in the lexical representation of a predicate is distributed, namely, SEM STR, ARG STR, GF STR, and GC STR. The rationale for the separation of these levels will become clear as we proceed. In addition, §3.3. also explores in some detail the relation between ARG STR and SEM STR, the two levels over which the information contained in traditional thematic role representations is distributed. §3.6. articulates an important claim in this organization of grammar, namely, the co-presence of the levels of structure: the information in each of these levels is simultaneously accessible to principles of the grammar.

3.2. Semantic Structure The level of SEM STR encodes all and only the linguistically relevant semantic distinctions that show systematic correlates in the morphology or syntax of natural languages. It is distinct from meanings in the real world involving entailments, and from the non-linguistic representations of concepts, situations, and so on. The entities represented at this grammar-internal level of structure are accessible to principles of grammar that regulate syntactic and morphological structure. The level of ARG STR (to be discussed shortly) on the other hand represents the number of ARGs a predicate takes, and their semantically-determined syntactically-relevant relative prominence. In other words, the information expressed in most traditional

The Theoretical Framework / 19

"theta role" representations is factored apart and distributed here over two levels, SEM STR and ARG STR. 3.2.1. Two Aspects of "Meaning" It is uncontroversial that the meanings of words must be represented in the lexicon in some fashion. Such a representation, which might be called the lexical conceptual structure (LCS) (cf. Jackendoff (1986), Hale and Keyser (1987)), must express all the elements of meaning that the speaker of a language associates with a word. In this representation, words that refer to distinct entities (objects, properties, activities, and the like) must have distinct lexical conceptual structures (e.g., red vs. pink, assassinate vs. murder, near vs. far, fall vs. pinch ). Given that meanings of predicates govern regularities of morphological and syntactic behavior, one might assume that these meaning-governed regularities are stated in terms of entities in LCS. Alternatively, one may assume that only "grammaticalizable" meanings can govern morphological and syntactic regularities in human languages, and that these meanings must be represented at a distinct, more abstract level of structure. These two alternatives are labelled by Pinker (1989:166) the "Unrestricted Conceptual Representation" and the "Grammatically-Relevant Subsystem" hypotheses respectively. In the former view, grammaticalizable meanings are not distinguished from real-world knowledge. In principle, then, any element of conceptual structure can govern systematic morphological and syntactic regularities. This is clearly an undesirable hypothesis. For example, the sentence John assassinated Bill. entails that (a) Bill is a prominent personality, (b) Bill is dead, and (c) Bill underwent a change of state. Under the "Unrestricted Conceptual Representation" hypothesis, these three components of meaning are equally likely to participate in morphosyntactic alternation. In actuality, however, we find only the third component, namely, CHANGE OF STATE, participating in such alternation crosslinguistically. This is consistent with the latter view, explicitly argued for by Wierzbicka (1988) and Pinker (1989), among others. Pinker (1989:166) holds that grammaticalizable meanings are "a set of semantic elements and relations that is much smaller than the set of cognitively available and culturally salient distinctions". Following this position, I assume that grammaticalizable meanings are represented at the grammar-internal level of structure called SEM STR, which is distinct from LCS.

20 / Arguments in Hindi

In order to illustrate the distinction between LCS and SEM STR, consider the meanings of red, black, redden, and blacken. The representation of the conceptual structure of each of these words must be distinct. However, the distinction between redness and blackness is irrelevant for SEM STR, as this distinction is unlikely to play a systematic role in the morphology and syntax of a language. In contrast, the meaning of change of state associated with the suffix -en in the distinction between red and black on the one hand, and redden and blacken on the other, involves a systematic morphological-semantic correlation, and must be represented in SEM STR: (4)

(5) redd-en

black-en

x change to red x change to black conceptual structure ……………………………………………………………………………… x CHANGE TO STATE x CHANGE TO STATE SEM STR

These examples illustrate that verbs that are distinct in LCS may be identical in SEM STR. Likewise, verbs that are identical with respect to the real-world activities that they refer to may be distinct in SEM STR. Thus, the statements Point A precedes point B, and Point B follows point A , have identical entailments, but are different in SEM STR, with corresponding differences in syntax, for instance, in subject selection.3 The distinction between the representation of LCS and SEM STR is analogous to the distinction between phonetic representation and phonological representation. While phonological representation and SEM STR are both strictly grammar-internal structures, phonetic representation and LCS both provide the interface between grammar and the real-world entities like the speech signal and the referent respectively.4 The relation between LCS and 3 The following example from Pinker (1989:358) speaks convincingly for the distinction between LCS and SEM STR: "… Georgia Green (1974) finds I carried / dragged / hauled / pulled / pushed him the box to be grammatical, but I do not. I would explain this in terms of these verb entries having different sets of semantic representations in her dialect than mine. Among the versions of drag in her lexicon is one that has the rough meaning "X causes Y to have Z b y causing Z to move in a dragging manner with X to Y"; my lexicon lacks such an entry and cannot attain it with any of my lexical rules. Yet surely Green and I do not have different conceptions of what dragging is." 4 Thus, one might imagine that the entity CHANGE TO in both LCS and SEM STR in (4)-(5) is a duplication of information. But the presence of the entity change in both LCS and SEM STR is analogous to the existence of the feature [± voice] i n

The Theoretical Framework / 21

SEM STR, like that between phonetic representation and phonological repre-

sentation, is certainly not arbitrary or random. However, the two must necessarily be distinguished. That there exists a small set of semantic primitives that participate in morpho-syntactic distribution and alternation patterns is an empirical hypothesis analogous to the hypothesis in phonology that there exists a small set of distinctive features that participate in morphologically sensitive phonological distributions and alternations. The discussion in this subsection is purely by way of clarifying the distinction between LCS and SEM STR, and defining the scope of SEM STR. Not being concerned with the representation of LCS per se, this thesis will not have anything specific to say about it. 3.2.2. Are Elements of SEM STR Visible to Syntax? An assumption implicit (or explicitly expressed) in many treatments of argument structure has been that while theta role information determines the argument structure of a verb, it is not directly visible to the rest of syntax, and is not available for the statement of syntactic generalizations (Burzio (1986), Zubizarreta (1987), Grimshaw (1988), Levin and Rappaport (1986), Belletti and Rizzi (1988:294), among others). Counter to this has been the assumption that syntax can in fact be directly constrained by the semantic substance of syntactic arguments (Lakoff (1968; 1976), Jackendoff (1972; 1978), Talmy (1972; 1985), Dowty (1979; 1986), Bresnan (1982a), Foley and Van Valin (1984), Fauconnier (1985), among others). Let us refer to these two views as the "Indirect Reference Hypothesis" and the "Direct Reference Hypothesis" respectively. The Indirect Reference Hypothesis is that semantic information is visible to syntax only to the extent to which it organizes argument structure. This may be interpreted in two ways. A strong interpretation is that no lexical semantic information is visible to any syntactic regularity unless encoded in argument structure. This interpretation, however, is clearly untenable. Meanings not necessarily encodable in argument structure, such as animacy and definiteness, different types of locative meanings, and so on, govern case selection in languages, even on subjects and objects. Therefore, those who subscribe to the Indirect Reference Hypothesis could not intend this strong interpretation. The weaker interpretation is that function selectboth phonetic representation and phonological representation. The entity has a different significance at each level. Therefore, it is not a duplication of information.

22 / Arguments in Hindi

ing meanings are exhaustively reflected in argument structure. However, this interpretation begs the question why function selecting meanings must be more privileged than all other grammatically relevant meanings. Having separated SEM STR from lexical conceptual structure, I assume that the latter is not directly visible to syntax. However, I also crucially assume that SEM STR information can be accessed by principles that govern syntactic and morphological regularities. With respect to SEM STR, then, I reject the Indirect Reference Hypothesis. The need for direct reference by syntax to meanings other than those relevant for function selection will be elaborated in §3.3.2., and demonstrated in our account of Hindi in the subsequent chapters.

3.3. Argument Structure Some of our central assumptions regarding ARG STR, similar to those articulated in Grimshaw (1988:1), for instance, may be explicitly stated as follows: (6) ARG STR a . contains information about the syntactic valency of a predicate. b. represents prominence relations among arguments. c. contains no thematic role meanings.

In what follows immediately, we will elaborate on these assumptions, which will prove to be crucial for the account of Hindi presented in the chapters that follow. 3.3.1. Valency and Relative Prominence It is well-known that predicates fall into various subclasses which place demands on the number of syntactic arguments they take, that is, their VALENCY. Thus, the verbs sleep, hit, and send belong to the subclasses of monadic, dyadic, and triadic verbs respectively. Valency information ((6a)) is represented in ARG STR in terms of abstract ARG slots, with which elements of SEM STR, GF STR, and GC STR can be associated. The notion of an ARG position corresponds to the notion "predicate argument" in Bresnan (1982b), to a "lexical role" as distinct from "thematic role" in Bresnan and Kanerva (1989), and perhaps to an "argument" in Kiparsky (1987).

The Theoretical Framework / 23

The relative prominence of ARGs, as we will see shortly, is strictly regulated by pair-wise statements of relative prominence on elements in SEM STR. In theta role representations, this is expressed as the Thematic Hierarchy (Gruber (1965), Jackendoff (1972)). I use the convenient notation of indicating the scalar relative prominence of ARGs with subscripted numerals as in (7).5 To exemplify, I use the predicate send in Paul sent his car to the garage. The simplified SEM STR in (7) may be read as: " x causes y to move to z." (7)

send [

X

EVENT

CAUSE

]

_______________ [ Y TO Z MOVE ] SEM STR ……………………………………………………………………… ARG1

ARG2

ARG3

PRED

ARG STR

The ARG STR in (7) expresses the idea that (a) the predicate has three ARGs; and (b) the variable X (the agent of sending) is the most prominent ARG, and the variable Z (the locational goal of sending) is the least prominent ARG. We will return shortly to the notion of such a hierarchy of ARGs. 3.3.2. Representation of Thematic Role Information The terms "agent", "goal", "patient", and the like are convenient labels widely used to refer to the semantic relations that arguments bear to their predicates, and have been variously called case relations (Fillmore (1968)), 5 The scalar relative prominence of ARGs may be represented either in terms of the grouping relation, as in (i), or the successor relation, as in (ii): (i) (ii) ˘ ARG < ARG < ARG ¿ PRED /\ ARG

ARG

ARG

PRED

There does not appear to be any crucial empirical difference between the two formalisms. The subscripted numerals on ARGs merely indicate relative prominence, and are not to be confused with labels. This is in contrast, for instance, to RG, where "1", "2", and "3" not only reflect a hierarchy of "grammatical relations" but also are labels for them. Therefore it is perfectly legitimate to have a structure in RG with 1 and 3 but no 2, or a 2 without a 1. These options are not available in ARG STR representations.

24 / Arguments in Hindi

semantic relations (Katz (1972)), thematic relations (Gruber (1965), Jackendoff (1972)), and currently the most familiar thematic roles, or †roles. In most theories that employ theta role labels, thematic role information is expressed in the predicate argument structure of a verb. While different theories may disagree on the structuring and representation of arguments, they agree that minimally reflected in the representation of the argument structure of a predicate are (a) the number of arguments the predicate takes; (b) the semantic relations they bear to it; and (c) their relative prominence. The inherent relative prominence of individual thematic roles is expressed in many theories in terms of a fixed language-independent hierarchy of roles, known as the THEMATIC HIERARCHY (Gruber (1965, 1976), Jackendoff (1972), Givón (1976; 1984), Ostler (1979), Foley and Van Valin (1984), Kiparsky (1987; 1988), Bresnan and Kanerva (1989; in press), among others). This hierarchy serves to order the arguments in a predicate argument structure in terms of a semantically determined prominence scale, much as in (7). A version of the Thematic Hierarchy is given in (8), where x < y means "x is higher on the hierarchy than y", as is the general convention: (8) Thematic Hierarchy agent < beneficiary < goal

< instrument < patient / theme < locative

Even though researchers might disagree on the specific details of the hierarchy,6 they generally agree that such a hierarchy plays a role in governing syntactic regularities. A primary function of the Thematic Hierarchy has been the formal expression of the notion LOGICAL SUBJECT (Kiparsky (1987) and related works). Another crucial function has been to identify the default associations between meanings and grammatical functions such as subject and object (Givón (1984), Foley and Van Valin (1984), Kiparsky (1987), Bresnan and Kanerva (1989), among others). The hierarchy has also been found relevant to characterize the asymmetries in idiom formation (Kiparsky (1987)). In many languages, it has been argued that the hierarchy also crucially serves to constrain word order (Uszkoreit (1984, 1986)). 6 For intance, there may be disagreements on the relative position of "themes" with respect to "goals" and "locatives", or even about whether the relative prominences can be reduced to a single hierarchy. See Bresnan and Kanerva (in press) for a discussion.

The Theoretical Framework / 25

Another type of prominence encoded in the representation of argument structure in some theories is that of the so-called "deep grammatical functions". Thus, many current GB frameworks employ the notions "external argument" and "direct internal argument" in argument structure. For instance, in Levin and Rappaport (1986), the lexical representation for the predicate send would be as in (9).7 The agent is the "external argument" in (9).8 (9)

agent theme

goal

send

DIR. ARG

The structural information expressed by the Thematic Hierarchy is not identical to that expressed by notions like external argument and direct internal argument. For instance, the cat in The cat chased the mouse is the logical subject as well as the external argument. In contrast, the cat in The cat feared the dog need not be the external argument, even though it is necesarily the logical subject.9 What is common to (7) and (9) is that they both encode the relation of arguments to their predicate at the level of structure that closely corresponds to our ARG STR. In short, most representations of †-roles serve to conflate the structuring of arguments (whether as in (7) or as in (9)) and the thematic content of the arguments in a predicate argument structure. In these representations, the meanings isolated by thematic roles are particularly those that 7 Levin and Rappaport use the typographical convention of bracketing for indicating the external argument and italics for direct internal arguments, e.g. send : agent < theme goal > 8 The notions "external argument" and "direct internal argument" are due t o Williams (1977) and Marantz (1981) respectively. The notion of external argument in Williams (1977), (1979), (1981), (1984), however, is very different from that found in most other works in GB, for instance, Rappaport and Levin (1986), Belletti and Rizzi (1988). Although Williams claims that it is "a notion drawn purely from the †-theory," (1984:642), he designates grammatical subjects of both passive and active sentences in Malayalam as "external arguments" in order to account for case assignment, reflexive binding, and control (1984:656-657). For Williams, then, "external argument" is identical t o "grammatical subject with a †-role". Therefore, it is not an argument structure notion. 9 See section 3.3.5. for a discussion of the notion logical subject.

26 / Arguments in Hindi

are relevant for their association with grammatical functions. The wider range of meanings that are relevant for other aspects of syntax (e.g., intentionality, volitionality, and telicity, relevant for auxiliary selection; containment, contact, and direction, relevant for case selection) are left out of these representations, and are available only in lexical semantics. Thus, these representations of argument structure conflate argumenthood and function selecting argument meanings as †-role labels. Now, according to (6a,b), we have assumed that the number and organization of the ARGs of a predicate are represented in ARG STR. Given the assumption that all (and only) grammaticalizable meanings must be represented in SEM STR, it should be possible to define thematic roles in terms of configurations of entities in SEM STR, say, along the lines in Jackendoff (1972, 1987:378). By way of illustration, consider the English examples and their representations in (10b), (11b), and (12b) below. The tentative SEM STR representations indicate the compositionality of lexical meaning in terms of primitives of the kind used by Jackendoff (1972 and subsequent works).10 In (10c), (11c), and (12c) are given approximate theta role labels equivalent to the combination of substructures in SEM STR and ARG positions in ARG STR.11 Thus, the constructs that correspond to theta role labels are recoverable from a combination of SEM STR and ARG STR, as in (10c), (11c), and (12c). The question then is, need function selecting meanings be singled out of SEM STR and represented in ARG STR as well, along with valency and relative prominence information? The motivation for this would be the need to encapsulate function selecting meanings in a level of representation distinct from all other grammatically relevant meanings. However, in the absence of similar encapsulation of case selecting meanings, auxiliary selecting meanings, and the like, there seems to be no reason to copy function selecting meanings onto ARG STR.

10 These representations are simplified and cannot be taken as complete semantic representations. They are not unlike some of the representations i n Foley and Van Valin (1984), Dowty (1987), and Pinker (1989). 11 "Theme" in (11c) corresponds to the entity that undergoes change of state or location (Gruber (1965), and others). In (10c), it is the entity in a state or location. In (12c), "Agent" corresponds to the causer of an event.

The Theoretical Framework / 27

(10) a.

The door is open. [open ]

b.

[

BE-AT STATE

X

˘ ARG ¿ c.

]

SEM STR

PRED

ARG STR

BE-AT STATE

X

=

Theme

ARG

(11) a.

The door opened. [open ]

b.

[

[

X MOVE

X BE-AT STATE ]]

˘ ARG ¿ c.

X

SEM STR

PRED

ARG STR

MOVE

=

Theme

ARG

(12) a.

John opened the door. [open ]

b.

[ ˘

c.

Y

Y CAUSE ARG

[

X MOVE

ARG ¿

[ X BE-AT

STATE

P RED

]]]

SEM STR ARG STR

CAUSE

=

Agent

ARG

I therefore assume that meanings relevant for function selection are part of the independently motivated SEM STR (Jackendoff (1972; 1987), Dowty (1986; 1987), Ladusaw and Dowty (1988), Talmy (1972; 1985)). Only their argumenthood is encoded in ARG STR.12 The statement of syntactic regu12 This idea of partitioning information into two levels of structure is already available in Hale and Laughren (1983), Grimshaw (1988), and Rappaport and Levin (1986), among others, who distinguish between the Predicate Argument

28 / Arguments in Hindi

larities may then refer independently to an ARG, or an element in SEM STR, or jointly to elements in both levels of structure. Our account of Hindi in the subsequent chapters provides compelling evidence for this factorization of information into two levels of structure. The exact number and characterization of the primitive elements and relations required in SEM STR is an empirical issue that I will not confront in this thesis. In the chapters that follow, I use familiar and convenient terms such as "agent", "goal", "instrument", and "theme" in portions of the text that do not demand precision. Where these terms fail to provide the required precision, I introduce semantic constructs required for the account of Hindi, explaining the terminology as and when necessary. 3.3.3. The Hierarchy of Arguments If the syntactically relevant prominence relations among arguments are semantically governed, and if semantic information is available for the statements of syntactic regularities, a justifiable question is: Why should these relative prominence relations be represented at ARG STR rather than at SEM STR? This section seeks to answer this question, and also to decipher how the effects of the traditional Thematic Hierarchy relevant for syntactic processes are captured within the conception being outlined here. I assume that for every lexical representation of a predicate, there exists an ordering relation among the ARGs expressed at ARG STR, which I refer to as the ARGUMENT HIERARCHY. This ordering is governed by universal statements of asymmetries in the relative prominence of ARGs in terms of the entities at SEM STR they are associated with. These statements may take the form in (13). The statements in (13a-d) refer to the inherent properties of semantic variables, while (13e) refers to the constituency in SEM STR.13

Structure (PAS), which expresses the syntactic relations of arguments to their predicates, and the Lexical Conceptual Structure (LCS), which includes the semantic relations arguments bear to their predicates. The separation between SEM STR and ARG STR differs from that between PAS and LCS in that while LCS contains all aspects of meaning, SEM STR contains only grammaticalizable aspects of meaning. 13 The statements in (13) are similar in spirit to those in Wechsler (1990).

The Theoretical Framework / 29

(13) a. b. c. d. e.

An ARG associated with a causer is more prominent than an ARG associated with a noncauser. An ARG associated with a sentient participant is more prominent than an ARG associated with a non-sentient participant. An ARG associated with a source is more prominent than an ARG associated with a goal. An ARG associated with an undergoer of change (of state/location) i s higher than an ARG not associated with change (of state/location). An ARG associated with a participant in a matrix SEM STR UNIT is more prominent than an ARG associated with a participant in an embedded SEM STR unit.

Viewed in the light of (13), the traditional notion of the Thematic Hierarchy is the cumulative result of a number of interacting relative prominence relations between semantic entities, which, in mapping into ARG STR, yield an ordering of ARGs.14 For instance, the effect of the agent ARG being more prominent than the theme ARG in (14b) is derived from (13a): (14) a.

John opened the door.

b.

[

Y

EVENT

CAUSE

]

___________________ [ X MOVE & BE-AT STATE ]

SEM STR

………………………………………………………………………… ARG1

ARG2

A-PRED

ARG STR

15 An important assumption underlying the relative prominence statements in (13) is that the various statements themselves have priority relations among them, governed by general principles. In other words, some have greater weight than others. Thus, when two of the statements are 14 This view of the relation between SEM STR and ARG STR prominences makes available the possibility of situations in which the semantic asymmetries underdetermine the argument hierarchy in a particular ARG STR. While I do not employ this possibility for the purposes of this thesis, it is an issue open for empirical substantiation. 15 Here (and henceforth), A-PRED/ARG-PRED refers to a syntactic predicate at ARG STR, and S-PRED/SEM-PRED refers to a semantic predicate at SEM STR.

30 / Arguments in Hindi

applicable in a given situation, and yield conflicting results, one statement may have priority over the other. For instance, by (13a) and (13e), the ARG associated with Y is the more prominent one. But by (13d), the ARG associated with X is the more prominent one. Thus, (13d) is in conflict with (13a) and (13e). I assume that, in the event of conflict, (13e), which refers to constituency relations rather than to inherent semantic properties, has priority over all other statements. Hence, Y is associated with ARG1, and X with ARG2. Another instance of conflict between the relative prominence statements in (13) is illustrated by the contrast in prominence between (15a) and (15c): (15) a. b. c. d.

Tracy threw Miriam the ball. Tracy threw the ball to Miriam. Tracy threw the ball to the tennis court. *Tracy threw the tennis court the ball.

According to the version of the Thematic Hierarchy in (8), Miriam in (15a) is higher than the ball by virtue of being a "goal". In (15c), in contrast, the tennis court is lower than the ball by virtue of being a "locative". The statements in (13) yield precisely this result in (15c). By (13d), the ball, which undergoes change of location, is higher than the tennis court. In (15a), however, we have a conflict. By (13d), the ball is higher than Miriam, exactly as in (15c). But by (13b), Miriam is higher, being sentient. This conflict may be resolved by assuming that a principle which involves the idea of sentience has priority over one which does not. (13b) would then have priority over (13d). By analysing the two occurrences of throw in (15a) and (15c) as involving a sentient goal and a non-sentient goal respectively, we capture the contrast in relative prominence between their ARGs. In addition, we also capture the common core in the two occurrences of throw: the only difference between them is in the additional specification of sentience in the goal of (15a). This common core is obscured in a traditional thematic role representation, which would encode the argument structure of throw in (15a) as ˘agent goal theme¿, and that of throw in (15c) as ˘agent theme locative¿.16 16 I must add two points of clarification. First, whether throw in (15b) i s analysed on par with that in (15a) or in (15c) is an open question. Second, this

The Theoretical Framework / 31

We can now answer the question that we raised at the beginning of the section: why should we represent the semantically determined relative prominence of ARGs at ARG STR rather than at SEM STR? Given the conception of pairwise relative prominence statements such as in (13), it is not possible to represent semantic relative prominence as a scale at SEM STR: the final hierarchy can be computed only as a cumulative result of the entire set of principles at the level of ARG STR, some principles reinforcing each other, and others conflicting with each other. As we proceed, we will accumulate further evidence for this view of the Argument Hierarchy. The conception of the Argument Hierarchy outlined above is analogous to that of the Sonority Hierarchy in phonology, ARG STR and SEM STR being somewhat parallel to the skeletal structure and melodic structure of phonological segments. For example, a skeletal slot (segment) associated with the melody [+sonorant] is higher on the Sonority Hierarchy than one associated with [-sonorant]; a slot associated with [+continuant] is higher than one associated with [-continuant]. Sometimes, the two statements yield conflicting results, for instance with respect to a nasal ([+son, -cont]) and a fricative ([-son, +cont]). In such an event, the conflict is resolved in favour of the statement in terms of [±sonorant], making the nasal higher on the hierarchy than the fricative. In short, although the Argument Hierarchy and the Sonority Hierarchy are scales of prominence, they can be derived from sets of pairwise prominence statements, with provision for conflict resolution. 3.3.4. The Mapping Between ARG STR and SEM STR An important consequence of partitioning information into two distinct levels of structure is that it allows us to formally express systematic distribution and alternation patterns at one level without involving the other level. Such partitioning of information lends itself to otherwise impossible formal possibilities of elements at one level being (a) multiply associated with elements at the other level, and (b) not being associated with any element at the other level. This was the central insight of factoring apart the representation of tonal information and segmental information in Autosegmental Phonology. With respect to SEM STR and ARG STR, the factorization opens up the possibility of having ARGs associated with more discussion should not taken as a commitment to the exact status of the notion of "sentience" in the dative alternation in English. Sentience may be a consequence of some kind of a possession relation as suggested by Pinker (1989).

32 / Arguments in Hindi

than one element of meaning, or an ARG not associated with any entity in SEM STR. Similarly, in the lexical representation of a predicate, there could also be elements of SEM STR that are unassociated with any ARG, or associated with more than one entity at ARG STR. The schematic representations in (16) illustrate the various mapping possibilities between SEM STR and ARG STR: (16) a.

b. x

y

A

B

c. x

y B

y A

B

d. x

y

\/ A

SEM STR ARG STR

The structure in (16a) appears to be the most natural and unmarked. However, there are numerous syntactic phenomena in natural languages that are best analysed as instances of the structures in (16b-d), as I briefly illustrate in what follows.17 3.3.4.1. Multiply Associated Entities Jackendoff (1972:32), using examples such as (17a,b), demonstrates the need for syntactic arguments to be associated with more than one theta role: (17) a. * The rock deliberately rolled down the hill. b . Mary deliberately rolled down the hill.

The subject argument in (17a) is only a theme, whereas the subject argument of (17b) is at once a theme and an agent, an instance of the structure in (16d). English does not provide morphological alternations illustrating the multiple association in (16d). Morphological evidence comes from the productive process of causativization in a number of South Asian languages, for example, in Malayalam, a Dravidian language of Southern India (T. Mohanan (1988), K.P. Mohanan (1988)). Causativization is conventionally seen as involving the embedding of a predicate inside a CAUSE predicate: the result is the addition of a CAUSER argument, as illustrated in the following Malayalam sentences:

17

The formalism in (16) allows for a fifth possibility: that of more than one being associated with a single SEM STR entity. Whether or not such a structure should be allowed is an open question. ARG

The Theoretical Framework / 33

(18) ravik'k'™ sa˚˚a§i

manass-il-aay-i. Ravi-D matter-N mind-L-become-PA Ravi understood the matter. (Lit. To Ravi, the matter became in the mind.)

(19)

alkka raviye sa˚˚a§i manass-il-aa-kk-i. Alka-N Ravi-A matter-N mind-L-become-C-PA Alka caused Ravi to understand the matter.

The verb in (19) is the morphological causative version of the verb in (18), and as expected, causativization is accompanied by an increase in valency. Ravi in (18) and (19) is the experiencer of understanding, and Alka in (19) the causer of understanding. Now, although such increase in syntactic valency is the most common pattern found in causatives across languages, it is not a necessary component of causativization. Consider the following causative, which does not involve any increase in syntactic valency: (20)

ravi sa˚˚a§i manass-il-aa-kk-i. Ravi-N matter-N mind-L-become-C-PA Ravi understood the matter.

The verb forms in (19) and (20) are identical. However, whereas the causer of understanding and the experiencer of understanding are distinct in (19), the two participants merge in (20). In other words, even though a causer is added in both, there is no corresponding increase in valency in (20). We may distinguish between (19) and (20) by representing them as (21) and (22) respectively. They are identical in SEM STR, but not in ARG STR, and consequently, the mapping between the two structures is also different. The internal SEM STR of the stem manass-il-aa "to understand" is not relevant for the issue under discussion. It is therefore simplified for ease of interpretation. Unlike predicates like CAUSE, MOVE, BE , and so on, predicates like understand are not atomic predicates drawn from the universal inventory of SEM STR. I use them in representations in this thesis for convenience. Henceforth, the notation of lower case indicates the status of such predicates.

34 / Arguments in Hindi

(21) manass-il-aa-kk "to cause to understand" [

((19))

EVENT CAUSE ] _________________ SEM STR [Y Z understand ] …………………………………………………………………… ˘ ARG1 ARG2 ARG3 ¿ A-PRED ARG STR X

(22) manass-il-aa-kk "to cause to understand" [X

((20))18

CAUSE ] _________________ SEM STR [Y Z understand ] …………………………………………………………………… ˘ ARG1 ARG2 ¿ A-PRED ARG STR EVENT

What (18)-(20) demonstrate is that the causative morpheme in Malayalam adds a causer to the SEM STR, which may be associated with an independent ARG, in which event there is an increase in valency ((19)). Or else it may be associated with an ARG that already has an entity in the embedded semantic predicate associated with it. In that event there is no change in valency ((20)).19 Precisely the same phenomenon appears in Hindi as well. Thus, the causativization in (23b) increases valency, as in the structure in (21). But that in (24b) adds a causer that merges with the experiencer, with no valency alternation, exactly as in (22): (23) a.

gaa±ii mu±ii. vehicle-N turn (intr.)-PERF The vehicle turned.

b.

±raivar-ne gaa±ii driver-E vehicle-N The driver turned the vehicle.

mo±ii. turn (tr.)-PERF

18 It is not the case that the X in (22) is unlinked to syntax, and is understood as ARG1 by some default interpretation. Under no circumstances as far as I know can the understander and the causer of understanding be distinct in (22). 19 The possibility of causativization without change in valency is not available to predicates whose highest argument is an agent. Such causativization would involve the merging of the agent added by causation and the agent of the caused event, resulting in a vacuous application of causatvization, which I assume is prevented by the no vacuous affixation condition in Marantz (1984).

The Theoretical Framework / 35

(24) a.

anuu-ko taaraa dik¥aa. Anu-D star-N be/become visible-PERF Anu saw the star. (Lit.: To Anu the star became visible.)

b.

anuu-ne taaraa Anu-E star-N Anu saw the star.

dek¥aa. see/look at-PERF

Now in (20) and (24b), the causer merges with the experiencer in a single argument. A causer may also merge with a theme. Consider the Hindi examples in (25b-d), which are morphological causatives of (25a): 20 (25) a.

baccaa kutte-se ±artaa child-N dog-I fear-IMPERF be-PR The child fears the dog.

b.

raam bacce-ko kutte-se ±araataa Ram-N child-A dog-I fear-C-IMPERF be-PR Ram causes the child to fear the dog. (Lit.: Ram frightens the child of the dog.)

c.

raam bacce-ko ±araataa hai. Ram-N child-A fear-C-IMPERF be-PR Ram causes the child to fear (something). OR Ram frightens the child.

d.

kuttaa bacce-ko ±araataa hai. dog-N child-A fear-C-IMPERF be-PR The dog frightens the child.

hai.

hai.

In (25b), Ram is the causer of fear, and the dog is the source of fear; (25b) has the structure in (26). In (25d), the dog is both the causer and source of fear; this sentence has the structure in (27). In (25c), Ram is the causer of fear, and can also be interpreted as the source of fear. The sentence is ambiguous between the structures in (26) and (27).

20 For an elaborate discussion of types of linking in (16b-d) between and SEM STR, see K.P. Mohanan (1988).

ARG STR

36 / Arguments in Hindi

(26) ±araa (X causes Y to fear Z. [X

X≠Z)

EVENT

((25b)) CAUSE

]

_____________ SEM STR [ Y Z fear ] ………………………………………………………………………… ˘

ARG1

ARG2

ARG3

(27) ±araa (X causes Y to fear Z. [X

¿

A-PRED

ARG STR

X=Z)

EVENT

((25d)) CAUSE

]

_____________ SEM STR [ Y Z fear ] ………………………………………………………………………… ˘

ARG2

ARG1

¿

A-PRED

ARG STR

In (26), the causer is an independent ARG. In (27), the causer of fear merges with the object of fear. The instances of the merging of two entities in SEM STR in a single entity at ARG STR in (22) and (27) exemplify the structure in (16d). Observe the relative prominence of ARGs in (27). By the relative prominence statement in (13b), the ARG associated with the sentient experiencer of fear in (25a) is higher than the ARG associated with the object of fear.21 In (25d), however, the object of fear (Z in (27)) merges in a single ARG with the causer of fear, which is therefore the higher ARG. This apparent "reversal" of the Argument Hierarchy follows directly from the interaction of the principles in (13). 3.3.4.2. Unassociated Entities Transparent examples of the structure in (16c), where an entity in ARG STR is unassociated with SEM STR, are found in the English constructions with a pleonastic there ((28a)) and it ((28b)), which are "non-thematic" (Bresnan (1982a: 72ff)):

21 It must be emphasized that the notion of sentience relevant in (13b) does not depend on the sentience of the referent, but on what is specified in the SEM STR of the predicate. Thus, whereas the experiencer of fear must necessarily be sentient (as part of the verb meaning), the sentience of the object of fear is not specified in the SEM STR. (cf. The child feared the storm. )

The Theoretical Framework / 37

(28) a. b.

There was a reindeer on the roof. It seems that Jay is late.

The pleonastic elements in (28) are generally acknowledged to have no theta roles, and are identified in Bresnan and Kanerva (1989:28) as "non-thematic lexical roles", and in Kiparsky (1987) as "non-thematic arguments". The SEM STR and ARG STR of the predicate in (28a) may be represented, by way of illustration, as follows: (29)

[ X LOC Y BE-AT ] SEM STR ……………………………………………………………… ARG

ARG ARG PRED

ARG STR

Corresponding to the three ARGs in (29), with which grammatical functions can be associated, there are only two variables in SEM STR. One of the ARGs is thus unassociated with any semantic content. Likewise, an entity of SEM STR in the lexical representation of a predicate may be unassociated with an ARG position ((16b)). That is, although present in SEM STR, it is unavailable for association to any grammatical function or overt expression through ARG STR. An example is the so-called "middle" construction. Keyser and Roeper (1984) explicitly argue for an "implicit agent" in this construction, classic examples of which are given in (30): (30) a. b.

Bureaucrats bribe easily. This book reads well.

The lexical representations of (30a,b) may be given as in (31), exemplifying the structure in (16b): (31)

[ X Y S-PRED ] SEM STR ………………………………………………………… ARG A-PRED ARG STR

The X in (31) (that is, the briber in (30a) and the reader in (30b)) has no corresponding entity in syntax. Once again, as with multiple associations of SEM STR to ARG STR in causatives, evidence for treating the lexical representation of middle verbs as having an agentive element in SEM STR that is

38 / Arguments in Hindi

not associated with an ARG comes from the morphological alternation of verbs in Hindi, discussed in detail in §7.5. 3.3.5. The Notion Logical Subject Following Bresnan (1982b:150), let us refer to the "thematic arguments" of a predicate, that is, those that are associated with elements of SEM STR, as LOGICAL ARGUMENTS. Non-thematic arguments such as pleonastic elements then will not be logical arguments. The most prominent logical argument in an individual ARG STR we will call the LOGICAL SUBJECT (LSUBJ). The intuitive pretheoretic notion of logical subject in traditional grammar is defined as the deep-structure subject in Chomsky (1965), and as the highest role of a Theta-structure in Kiparsky (1987). An L-SUBJ may be viewed as being prototypically associated on the one hand with the grammatical subject, and on the other with meanings such as actor, causer, or effector, typically bearing the properties of intentionality, volitionality, the conscious choice to perform an action, and the like. However, neither grammatical subjecthood nor these semantic properties can define logical subjecthood. Thus, Peter in both Peter ate the potatoes, and The potatoes were eaten by Peter, is an L-SUBJ. But it is a grammatical subject only in the former sentence. Similarly, John in John fell is an L-SUBJ, but it is not an agent or causer, nor does it bear the properties of volitionality or intentionality. We will see in Chapters 6 and 7 that the notion of L-SUBJ plays a crucial role in reflexive binding in Hindi: the antecedent of the reflexive apnaa in Hindi must be either an L-SUBJ or a grammatical subject. Now, consider the Hindi sentence in (32a), which is the causative of a causative: (32) a.

umaa-ne alkaa-se ravii-ko apnii daßaa samaj¥vaaii. Uma-E Alka-I Ravi-A self-g state-N understand-C-C-PERF Umai made Alkaj make Ravik understand self'si/*j/*k state/situation.

In (32a), only Uma is an eligible antecedent of the reflexive. In other words, the embedded causer Alka is neither a grammatical subject nor a logical

The Theoretical Framework / 39

subject: (32a) has only one L-SUBJ.22 The representation in (32b) captures this by representing the structure as a single clause at ARG STR. (32) b.

samaj¥vaa "to cause to cause to understand" [

M

EVENT

[

CAUSE

_____________________________ X EVENT CAUSE ]

]

SEM STR

_________________ [ Y Z understand ] ……………………………………………………………………… ˘ ARG1 ARG2 ARG3 ARG4 ¿ A-PRED ARG STR

Suppose we use the term "X STR clause" to refer to a unit at level X containing a predicative head and its dependents. In (32b), three SEM STR CLAUSES correspond to a single ARG STR CLAUSE. We may then make the definition of L-SUBJ more precise as follows: (33)

An L-SUBJ is the most prominent logical argument in an

ARG STR

clause.

According to (33), there will be as many L-SUBJs in a sentence as there are ARG STR clauses. In an ARG STR clause with only one ARG, that ARG will be the L-SUBJ. The account of subject selection and anaphora in Hindi in Chapters 6 and 7 shows that the notion L-SUBJ must be recognized as a legitimate theoretical entity, as is done in the Linking Theory in Kiparsky (1987), and in Bresnan and Kanerva (1989), and subsequent works in LMT. 3.3.6. The Notion Logical Object The notion of LOGICAL OBJECT (L-OBJ), when we have truly understood it some day, will perhaps eliminate a great deal of confusion that exists today in its conception, definition as well as terminology. This notion shares many of the pre-theoretic intuitions underlying various constructs such as Undergoer in Foley and van Valin (1984), Patient at the "action tier" in Jackendoff (1986), "patient" in Alsina and Mchombo (1988) and Alsina 22 When there are two causers, and the inherent semantic properties fail t o derive the relative prominence among ARGs, as in (32b), the statement in (13e) serves to establish this relative prominence.

40 / Arguments in Hindi

(1990), and "affected" in Ackerman (1990). Much research has gone into finding semantic correlates of these constructs (e.g. Tenny (1987), Krifka (1987), Heinrichs (1985), among others), semantic correlates such as "affectedness", "acted upon", "undergoer of change of state", "undergoer of change of location", "endpoint of action", "point of termination", "telicity", and so on. But the idea we are all trying to capture has notoriously evaded precise characterization. This suggests that perhaps the notion L-OBJ has no universal definition in semantic terms, and must be viewed as a syntactic notion at the level of ARG STR. For the purposes of this thesis, I will begin with the notion of L-OBJ implicit in Varma's Kerala Paa”iniiyam (1895), a grammar of Malayalam that draws on Pa”ini's grammar of Sanskrit. Interpreting Varma's kaarakam as "argument", kartaa as "logical subject", karmam as "logical object", vyaapaaram as "event", and p¥alam as "result", I translate as follows: "The logical subject is the leader of all other arguments. If we divide a verb into its two aspects, namely, event and result, that on which the event relies is the logical subject, and that on which the result dwells is the logical object. Sometimes, the event and the result may be in the same entity. Such verbs do not have a logical object. It is only when the event is in one entity and the result in another that the verb can have a logical object." (p. 183). According to this characterization, a logically transitive structure is one which has an entity towards which the action or event is directed, and an independent inceptor of the action or event. The logical object in this structure is the entity towards which the action or event is directed. This conception is not different from the idea of karman in Pa”ini, which is iipsitatama, or "that which is directly reached by the action" (Joshi and Roodbergen (1975)). In the Sanskrit equivalent of "Devadatta sat on the floor," the case on "the floor" is locative (ad¥ikara”a ). But "the floor" can be DESIGNATED as the karman , in which event it gets the accusative case of the karman . The idea of designating an entity as karman suggests that the notion is not purely semantic. According to Cardona (1976), karman is a category that is intermediate between "semantic relations and syntactic strings". Thus, "…there is no single set of homogeneous semantic features which can be said, in Pa”ini's system, to define a category such as karman … "(p.24) Now, that the notion L-OBJ (karman ) is a syntactic notion does not imply that it has no semantic correlates whatsoever. Thus, the properties of affectedness, undergoer of change of state or location, endpoint of action, and

The Theoretical Framework / 41

the like, are prototypically associated with this notion. The relation between L-OBJ and its semantic correlates is somewhat like the relation between a syllable nucleus and its segmental distinctive feature correlates in phonology. A syllable nucleus is not a distinctive feature notion, nor can it be defined in terms of distinctive features. However, there are universal as well as language particular constraints on the correlation between the nucleus and distinctive features. This analogy suggests that the semantic consequences of logical objecthood may vary in limited ways across languages. I believe that this is indeed true, but I will not pursue this idea here. For our purposes, I sum up the notion of L-OBJ as follows: An L-OBJ is an ARG STR construct which is associated with a SEM STR participant towards which an action or event is directed by an independent inceptor of the action or event.23 This notion plays a central role in the association of accusative case (Chapter 4), and in our account of complex predicates (Chapter 8) in Hindi. 3.3.7. Summary Our central assumptions about the level of ARG STR may be summarized as follows: ARG STR is a syntactic level of representation at which the number and relative prominence of the arguments of a predicate are expressed. The relative prominence of ARGs is constrained by statements of prominence in terms of association with entities at SEM STR. The mapping between SEM STR and ARG STR is not one-to-one: we allow multiply associated as well as unassociated elements at ARG STR, and also unassociated elements at SEM STR. An ARG STR clause is made up of a predicative head and its dependent ARGs. An ARG associated with a SEM STR configuration is a logical argu23 Pinker (1989) provides a detailed and extremely appealing account for dative alternations in English in terms of differences in semantic structure. By his account, the grammatically relevant meaning of John gave Mary a book would be: "John caused Mary to have a book," and that of John gave a book t o Mary would be: "John caused a book to go to Mary." However, as Pinker himself points out (p.83), the semantic contrast between the two constructions is weak. Presumably, the meaning of give is "x causes y to go from x to z, and z to have y." Pinker analyses differences such as the one between the two sentences given above as a "rearrangement of verb meanings" which have consequences for discourse. This rearrangement, I believe, involves shifting of focus, or what Fillmore (1977) calls "perspective", which may in fact be expressed as a shifting of logical objecthood. Thus, Mary is the L-OBJ in John gave Mary a book, and a book is the L-OBJ in John gave a book to Mary.

42 / Arguments in Hindi

ment. The most prominent logical ARG in an ARG STR clause is the LSUBJ. An ARG STR clause may also have an L-OBJ, which is the ARG associated with a SEM STR participant towards which an action or event is directed by an independent inceptor of the action or event. The construct L-OBJ does not participate in the relative prominence relations at ARG STR. The empirical advantages of this conception of ARG STR and its separation from SEM STR will find further justification in our analysis of case preservation (Chapter 4), the "middle" construction (Chapter 7), and complex predicates (Chapter 8) in Hindi.

3.4. Grammatical Function Structure The semantically-governed relative prominence among the ARGs of the verb send in the sentence Paul sent his car to the garage , given earlier as (7), is repeated below: (7)

send [

X

EVENT

CAUSE

]

_______________ [ Y TO Z MOVE ] SEM STR ……………………………………………………………………… ARG1

ARG2

ARG3

PRED

ARG STR

The general assumption in syntactic theories that appeal to the thematic hierarchy is that these relative prominence relations at ARG STR remain constant in (34a) and (34b): (34) a. b.

Paul sent the car to the garage. The car was sent to the garage by Paul.

However, syntactic theories also recognize that there exist phenomena such as anaphora and control with respect to which Paul is the most prominent argument in (34a), and the car in (34b). In other words, unlike the relative prominence according to the Argument Hierarchy, this prominence is not the same in (34a) and (34b), and must be expressed at some other level of syntactic structure. This level I call GRAMMATICAL FUNCTION STRUCTURE (GF STR).

The Theoretical Framework / 43

The relative prominence of the arguments in (34a) and (34b) at GF STR are given in (35a) and (35b) respectively.24 (35) a. b.

Paul < the car

the car < to the garage < to the garage < Paul

send send

The representations in (35) express certain relative properties of the arguments, such as the structural asymmetries holding among grammatical functions (GFs) such as SUBJECT and OBJECT. These relative prominence relations and asymmetries are variously expressed in the literature as the Accessibility Hierarchy (Keenan and Comrie (1972)), Relational Hierarchy (Perlmutter and Postal (1974)), Grammatical Hierarchy (Sag (1986), Sells (1988)), and Obliqueness Hierarchy (Pollard and Sag (1987)). In addition to the relative properties, most theories agree that representations of GFs must allow for reference to individual GFs, even though they differ widely in their formal encoding. I use the labels familiar from traditional grammar, such as subject and object to refer to individual GFs. I also assume as part of GF STR a subsystem that represents information about GRAMMATICAL FEATURES: inherent verbal features such as tense, aspect, mood, and so on; inherent nominal features such as number, gender, and person; and case features such as nominative, accusative, locative, and the like.25 Given our separation of the various pieces of information into different levels of structure, it might turn out on further investigation that we must also represent grammatical functions and grammatical features at distinct levels, thus adding yet another level in the general organization of grammar. For the purposes of this thesis, however, the two subsystems are represented at GF STR. Chapters 6 and 7 deal with the properties of grammatical subjects in Hindi. In this section, I lay out the atomic elements of GFs available at GF STR, as well as some derived notions useful for easy reference, and their relation to elements at ARG STR. As for the subsystem of grammatical fea24 As with relative prominence at ARG STR, relative prominence at GF STR may be expressed in terms of grouping or the successor relation. The choice between the representations may be empirically relevant. However, for our purposes, it i s a matter of detail. I therefore employ the successor relation to encode relative prominence in GF STR. 25 This is consistent with the position in much of the work in LFG (Andrews (1982), Neidle (1982; 1988), Simpson (1983), among others) that grammatical features are represented in f-structure.

44 / Arguments in Hindi

tures, our assumptions will be presented in chapters 4 and 7, which explore the case system in Hindi, and chapter 5, which lays out verb agreement. 3.4.1. The Internal Organization of GF STR In the lexical representation of a predicate, that is, at the level of word grammar, all the elements in the GF STR of a predicate are associated with ARGs. In contrast, at the level of sentence grammar, GF STR clearly must also include NON-ARGs, that is, ADJUNCTS, such as the underlined units in (36a, b): (36) a. b.

Mary sold the car for Susan yesterday . Tom went to church on Sunday for Pat's sake.

In describing GF STR, I assume the distinction between adjuncts and non-adjuncts. However, we limit our discussion to GFs associated with ARGs, called "central functions" in Relational Grammar (RG) (Perlmutter and Postal (1983:86)). Among the central functions, RG makes a crucial distinction between TERM functions and non-TERM functions.26 The functions "subject" or 1, "direct object" or 2, and "indirect object" or 3, comprise TERMs. The nonTERM functions are "obliques" (traditionally "oblique objects"), and "chômeurs" (which include, for instance, the agent in a passive). The distinction between TERMs and non-TERMs is also assumed in Kiparsky (1987): TERMs are "grammatically linked" arguments, and non-TERMs are "semantically linked". In LFG as well, although not explicitly distinguished with a label, TERMS are shown to form a natural class: for instance, only a term can be a controller in "functional control", or a null pronominal (PRO) (Bresnan (1982c)). I assume this distinction to be available at GF STR.27 I also assume the categorization in RG among terms into two partially overlapping subsets of nuclear functions and object functions. Subject and "direct object" are nuclear functions; "direct objects" and "indirect objects" form the class of object functions. "Direct" and "indirect" objects in RG appear to refer to entities in the "initial stratum". Instead, I use the terms 26 "Grammatical functions " here correspond to "grammatical relations " i n Perlmutter and Postal and in RG in general. 27 TERM is not a uniform notion across theories. We ignore the exact differences, which are not relevant for the purposes of this thesis.

The Theoretical Framework / 45

"primary" and "secondary" object, which seem to correspond roughly to entities of the "final stratum" in RG. The use of the terminology is illustrated in the following examples: (37) a.

John subj

gave Mary pr.obj

b.

John gave a book subj pr.obj

a book. sec.obj

to Mary. oblique

Note that the primary object in (37a) is Mary, while that in (37b) is a book. In traditional terminology, a book is the "direct object" in both (37a) and (37b). Thus, the terminology of "direct" and "indirect" objects is not identical to that of "primary" and "secondary" objects. The Lexical Mapping Theory (LMT) postulates a classification of GFs in terms of the features [±r] and [±o]. The [-r] or "thematically unrestricted" functions are subject and object, the two GFs that may be nonthematic (Bresnan and Kanerva (1989:25)). The nuclear functions in RG correspond to the [-r], or UNRESTRICTED functions. Similarly, [+o] or "objective functions" correspond to the object functions of RG. "The intuition behind the feature [+o] is that there are several objectlike functions that appear as arguments of transitive categories of predicators (Verb and Preposition) but not of the intransitive categories Noun and Adjective." (Bresnan and Kanerva (1989:25)). To go back to the distinction between primary objects and secondary objects, the former is equivalent to an unrestricted object (OBJ), and the latter to restricted object (OBJ†). To summarize, the GF distinctions we have discussed are the following: (a) TERM vs. non-term, (c) UNRESTRICTED vs. restricted function, and (d) OBJECT vs. non-object.28 These distinctions yield the following crossclassification among GFs:29 (38) a. b. c.

SUBJ(ECT)

unrestricted non-object term unrestricted object term SEC(ONDARY) OBJ(ECT) restricted object term PR(IMARY) OBJ(ECT)

28 To repeat, we ignore adjuncts which are not associated with ARG STR. 29 In a theory such as GB, GFs are encoded in terms of structural configurations at s-structure. For instance, a subject is an NP of S, (or an equivalent, e.g. SPEC of IP), and a primary object is an NP of VP adjacent to the V.

46 / Arguments in Hindi

d.

OBL(IQUE)

restricted non-object non-term

Now, there are certain elements, such as the agent of a passive, that are ARGs but resemble adjuncts in their behavior; Grimshaw (1988:2-3) refers to these elements as ARGUMENT-ADJUNCTs. I will use this term to refer to the GF of the agent in a passive. Exactly as at ARG STR and SEM STR, there are clauses at GF STR, headed by GF-PREDs. A GF STR clause corresponds to the "clause nucleus" in Bresnan (1982c:304), which is "the domain of lexical subcategorization in the sense that it makes locally available to each lexical form the grammatical functions that are subcategorized by that form." It also corresponds to the "complete functional complex" in Chomsky (1986:169), which contains a predicative head and "all grammatical functions compatible with its head". Familiar from the literature are two constraints that hold on the internal structure of a GF STR clause. The first is the "Final 1 Law" in RG, and the corresponding SUBJECT CONDITION in LFG, which states that: (39)

Every GF STR clause must have a SUBJ.

This demand is imposed in GB by the Extended Projection Principle (Chomsky (1981), Rothstein (1983)). In §7.5., we will investigate in detail whether or not the subject condition can be maintained as an inviolable principle. 3.4.2. The Mapping Between A RG S T R and GF ST R The second condition, known in RG as the "Stratal Uniqueness Law", states that only one dependent of a clause may bear any given TERM function (Perlmutter and Postal (1983:19-20)). This well-formedness condition forms an integral part of the Function-Argument Biuniqueness Principle in LFG (Bresnan (1982b)). The Function-Argument Biuniqueness principle states: (40)

Every expressed ARG in the lexical representation of a predicate must be associated with one and only one (non-adjunct) GF, and every (nonadjunct) GF with one and only one expressed ARG.

The Function-Argument Biuniqueness Principle rules out configurations such as in (41):

The Theoretical Framework / 47

(41) a . * ARG … /\…

b. * ARG ARG c. ……………

SUBJ OBJ

SUBJ SUBJ

* ARG

ARG

ARG STR

…\/… SUBJ

GF STR

The configuration in (41a) is ruled out by the first part of (40), and those in (41b, c) by the second part. Recall that multiple associations such as in (41c) were permitted in the mapping between SEM STR and ARG STR. This in fact provided motivation for the separation of the two levels. Now, if they are not permitted in the mapping between ARG STR and GF STR, why should these be considered distinct levels of structure? The configuration in (42), where prominence relations at the two levels are in conflict, provides the rationale for this separation. If ARG STR and GF STR were conflated into a single level, the result would be an intra-level conflict of prominences. (42)

o




o

GF STR

Function-Argument Biuniqueness is a general constraint on the mapping between the two levels of structure. There also exist constraints on the mapping between individual elements in these two structures. The relation between individual ARGs and GFs has been central to the investigation of phenomena such as the active-passive, causative, dative, and locative alternations, initially subsumed under Cyclic Transformations in early generative grammar, later under NP Movement in GB, and more recently in the "Linking" approaches in Kiparsky (1987; 1988), and in the Lexical Mapping Theory within LFG (L.Levin (1986; 1987), Bresnan and Kanerva (1989), Bresnan and Moshi (1990), Bresnan (1990), Alsina and Mchombo (1988), Alsina (1990)). We might view principles of association to be essentially of two types. The CANONICAL or "unmarked" (default) principles characterize the most natural or likely (= canonical) associations, distinguishing them from the less natural or less likely. The ABSOLUTE principles distinguish possible associations from impossible ones. Among the canonical mappings between ARGs and GFs, the most salient ones are the predisposition of the L-SUBJ to be the SUBJ, and that of the L-OBJ to be the PR.OBJ, stated in (43a, b):

48 / Arguments in Hindi

(43) Canonical a.

L-SUBJ

……… SUBJ

b.

L-OBJ

ARG STR

……… PR .OBJ

GF STR

An important property of canonical principles is that, unlike absolute principles, they may be preempted by other specifications in the grammar. Thus, in the passive construction, the L-SUBJ is "demoted" or "suppressed". That is, it is made unavailable for expression as a non-adjunct function, thereby precluding the effect of (43a). Given in (44a) and (44b) are representations that illustrate the active-passive alternation in John opened the door and The door was opened by John: (44) a.

[ X Y S-PRED ] SEM STR ……………………………………………………………………… ˘ ARG1 ARG2 ¿ A-PRED ARG STR ……………………………………………………………………… SUBJ OBJ GF-PRED GF STR

b.

[ X Y S-PRED ] SEM STR ……………………………………………………………………… ˘ ARG1 ARG2 ¿ A-PRED ARG STR ……………………………………………………………………… ADJ SUBJ GF-PRED GF STR

In (44a), the default mappings in (43) take effect, yielding a SUBJ and OBJ at GF STR. In (44b), on the other hand, the L-SUBJ (the open-er) is suppressed, as indicated by the notation of underlining. It is therefore an ARGUMENTADJUNCT. Interacting with the suppression of the L-SUBJ is the Subject Condition ((39)). Since every GF STR clause must have a SUBJ, the L-OBJ (the opened thing) is assocated with the SUBJ in (44b). The effect of (43b) is also thus precluded. In short, the mapping principles in (43a, b) are canonical principles that may be preempted by other principles, and take effect only by default. The notion of the preemption of one principle by another is crucially employed in the account of case in Hindi in Chapters 4 and 7. As mentioned at the outset, the focus of this thesis is the mapping between ARG STR and case. Therefore, I will not elaborate further on the mapping between ARG STR and GFs.

The Theoretical Framework / 49

3.5. Grammatical Category Structure Most syntactic theories distinguish the representation of surface grammatical functions from that of surface grammatical categories. For instance, surface GFs and their organization are represented in GB in LS (lexical structure) of S-structure (Chomsky (1981), Hale (1983), Williams (1984)) or NP-structure (Reimsdijk and Williams (1981)), and in LFG in f(unctional)-structure (Kaplan and Bresnan (1982:175)). Surface grammatical categories and their organization are represented at "surface structure" (a representation internal to the PF-component) in Chomsky (1981:18), P(HRASE) S(TRUCTURE) in Hale (1983), and c(onstituent)-structure in LFG. Chomsky (1981: 33) discusses four representations for the sentence, It is unclear who to see : (45) a. it b . it structure) c. it structure) d. it

is unclear [S who [S to see ]] is unclear [S whoi [S PRO to see ti ]]

(surface structure) (S-

is unclear [S COMP [S PRO to see who ]]

(D-

is unclear [S for which person x [S PRO to see x ]]

(LF-representation)

"Surface structure" representations ((45a)) share with PS and c-structure representations the following properties: they are the input to phonological interpretation; they encode linear precedence relations; they do not contain PRO or NP-trace.30 Another property of these representations is that they do not demand a "VP" constituent that contains a verb and all its complements except the subject. Roughly equivalent to surface structure, PS , or c-structure is what I call the GRAMMATICAL CATEGORY STRUCTURE (GC STR). Like PS and cstructure, GC STR contains information about GRAMMATICAL CATEGORIES such as noun, verb, adjective, and the like, and their constituency, as in (46): (46)

S NP

VP

GC STR

30 It is not clear to me how "configurational structure" in Chomsky (1981:127-135) is different from his "surface structure" or Hale's PS.

50 / Arguments in Hindi

/\ V NP ………………………………………………………………… Max admires Bob. WORD STRING

Parallel to the notions of a SEM STR clause, ARG STR clause, and GF STR clause is the GC STR clause, which is the same as S in (46). In this thesis we use the familiar label S rather than GC STR clause. A central property of surface structure, PS , and c-structure is that they do not contain "discontinuous constituents". Given the variability in word order across natural languages, therefore, it would follow that for the same GF STR, the GC STRs of different languages would exhibit variability in constituency and in linear precedence. Consider the GC STR and GF STR of a sentence in English ((47)) and a corresponding one in Warlpiri, an Australian language ((48)). The representations, adapted from Bresnan (1988: 2-3), are stripped of details: (47)

GF-CLAUSE

SUBJ

PRED

OBJ

GF STR

…………………………………………………………………………… _________________ __________ ______ The two small childrenare chasing that dog WORD STRING ………………………………………………………………………… V NP

GC STR

\/ NP

AUX

VP

S

(48)

GF-CLAUSE

SUBJ

PRED

OBJ

………………………………………………………………………………… wita-jarra-rlu ka-pala wajilipi-nyi yalumpu kurdu-jarra-rlu maliki 'the two small' 'are' 'chasing' 'that' 'children''dog' …………………………………………………………………………………

The Theoretical Framework / 51

NP

AUX V

NP

NP

NP

S

Whereas the GC STR representations in (47) and (48) differ, the GF STR representations are identical. The mismatch between GC STR and GF STR, shown in (47) and (48), was illustrated in Hindi in §2:(18). As is well-known, a non-finite clause need not correspond to an S in phrase structure in Warlpiri: that is, constituents of a matrix GF STR clause may be scrambled with the constituents of an embedded GF STR clause (Nash (1980), Hale (1983), Simpson (1983)). As shown in §2:(16), this property holds for Hindi as well. However, unlike Warlpiri, parts of a non-clausal constituent in Hindi cannot be scrambled with respect to elements outside the constituent, as shown by §2:(17). In contrast to both Warlpiri and Hindi is Malayalam, a Dravidian language, in which scrambling is restricted to the constituents of a single clause (Mohanan (1982)). Finally, in a language like Japanese, the head of a clause is fixed in the final position, while the other constituents may scramble with respect to one another (Farmer (1980)). Thus, English, Japanese, Malayalam, Hindi, and Warlpiri constitute examples of the various types of configurations possible in GC STR for the same GF STR configuration.

3.6. Co-presence of Levels of Structure We have so far assumed that the pieces of syntactically relevant information associated with a word string are organized into at least four levels of structure, namely, SEM STR, ARG STR, GF STR, and GC STR. These four levels of structure, I assume, are CO-PRESENT. That is, they co-exist as structures along multiple dimensions, related through structure building operations or through constrained association, rather than as transformationally related stages in a derivation. Therefore, principles of grammar can refer to them simultaneously, as is required by our analysis of case association and agreement in Hindi.31

31 This is analogous to syllable structure and morphological structure, which constitute two different levels of structure, and yet, because they are co-present, phonological principles can refer to the two structures simultaneously.

52 / Arguments in Hindi

It is the co-presence of multidimensional structures that underlies the conception of Systemic Grammar, as developed in Halliday (1970), and subsequent works. This is also the intuitive substance underlying the claim in Head Driven Phrase Structure Grammar (Pollard and Sag (1987) and related works), that linguistic structures constitute a single level of representation. The conception of co-present, interacting levels of structure is also articulated within LFG in Kaplan (1985). If two levels of linguistic structure are related through structure-changing (i.e. transformational) derivation, then the information at the two levels cannot exist simultaneously. If so, accessing information from both levels would require a global mechanism, that is, one that refers to two different stages in a derivation. Suppose that linguistic theory disallows such unconstrained global power. It would then follow that in a theory that employs structure changing derivations, a grammatical principle that accesses information from one level cannot access information from the other. The device of separating of transformationally related levels has been widely employed in both syntax and phonology to express the effects of non-interaction between subsets of information. An example of non-interacting levels of structure in a linguistic theory is that of LF and PF in GB. The GB model makes the prediction that principles of semantic interpretation in LF, for instance, quantifier scope, cannot have access to, say, phonological constituency in PF, and vice versa. In phonology, the idea that word-internal information is not available to principles of phrasal phonology and syntax is implemented through the structure-changing device of Bracket Erasure mediating between the word-internal and phrasal levels of structure. The separation of levels then yields the effect of non-interaction. In a system where levels of structure are co-present, the separation of levels cannot automatically yield the effect of non-interaction, although the effect can be derived through stipulation.32 We have not stipulated non-interaction between the four levels of structure discussed in this chapter. The co-presence of the levels of structure envisioned here may at first glance appear to allow for unconstrained access across levels by principles of 32 An example of such stipulation is in the relation between lexical conceptual structure and SEM STR, discussed earlier. The two levels are mutually constraining, and therefore co-present. But we stipulate that lexical conceptual structure is not available to any syntactic principles. Another example is the relation between prosodic structure and syntactic structure in Inkelas (1989). Prosodic structure and syntactic structure can mutually constrain each other, and therefore must be co-present. However, it is stipulated that phonological rules cannot directly refer to syntactic structure.

The Theoretical Framework / 53

grammar. How, for instance, does one rule out non-occurring grammatical principles such as the following? "A quantifier cannot have scope over an expression that contains a [+nasal] segment," "the antecedent of a reflexive must be bi-moraic," and so on. The answer to this problem has two aspects. The first part is the recognition that the problem is not a consequence of allowing the co-presence of various structures. The prohibition against inter-level principles implicitly claims that the separation of structural information into levels of representation is both sufficient and necessary to constrain the interaction of information available to principles of grammar. Thus, one might think that the prohibition against quantifier scope referring to nasality, or reflexives referring to the moraic structure of their antecedents, is an automatic consequence of non-interaction between LF and PF . This conclusion, however, is false. The specifications of nasality and number of moras are present in the lexical entries inserted at d-structure, and are carried over to s-structure and LF, unless explicit principles regulate what is carried over. Nothing in the theory therefore prevents a language from relating quantifier scope and reflexive binding to nasality and moraic structure. Therefore, all theories require substantive principles that constrain the interaction of clusterings of information in linguistic regularities, independent of separation of levels of representation. Given that such constraints are needed within levels, they may be extended to constrain inter-level interaction as well. Therefore, there is no justification to assume that non-interaction holds between all pairs of levels of structure. If non-interaction does not motivate the separation of levels of structure, what motivates the different levels assumed in this thesis? Could not the information in the various levels be viewed as part of a single level of representation? Our motivation comes from mismatches in the structures called for by different phenomena. When evidence from different sets of phenomena point to conflicting sets of conclusions, the separation of structures allows the conflict to be resolved. The conflict may involve conclusions regarding (a) the number, (b) the relative prominence, (c) the grouping, or (d) the labelling of entities in a linguistic object. To give an example, the intervocalic t in the word city behaves as [-voice] for certain phonological phenomena, but as [+voice] for others, thus involving a conflict in labelling. The apparent contradiction may be resolved by assuming that the entity is [+voice] at one level of structure, and [-voice] at another.

54 / Arguments in Hindi

One of the motivations for separating SEM STR and ARG STR is a difference in the number of entities at the two levels, which calls for the options of either multiply associated or unassociated entities.33 In the mapping between SEM STR and ARG STR, we allow both (see (16)), although such a mismatch is disallowed in the mapping between ARG STR and GF STR by the principle of Function-Argument Biuniqueness. Phenomena such as anaphora and control across languages which converge on two different sets of prominence relations motivate the separation of ARG STR and GF STR. Differences in grouping relations, for instance in the Warlpiri example in (48), motivate the separation of GF STR and GC STR. Grouping differences also play a role in the separation of SEM STR and ARG STR: in the causative structures in (21)-(22), two SEM STR clauses correspond to a single ARG STR clause. Conflicts in labelling would result if we did not separate functional and categorial labels. An entity that is a predicate in function is typically a verb in category. However, it is widely known that this is not always so. A telling example of such a mismatch is the cleft construction in Malayalam (Mohanan (1982)). In this construction, the predicate belongs to the nominal category, and one of its arguments to the verbal category. If we were to express predicatehood in terms of categorial labels, this would constitute a conflict in labelling.

3.7. Summary What I have outlined in the preceding sections is a conception of linguistic organization in terms of a set of interdependent, interacting levels of structure, linked to one another by principles of grammar. The syntactic information of a WORD STRING, be that a morpheme, a word, a sentence, or a discourse unit, is organized into at least four levels of structure: SEM STR, ARG STR, GF STR, and GC STR. The main proposals that will be validated in the course of the analysis of Hindi are summarized below. A key assumption in the organization of the levels of structure is that the levels are co-present, and principles of grammar may simultaneously access any one or more of these structures, unless stipulated otherwise. The 33 This is analogous to the separation of tones from tone bearing units i n autosegmental phonology, which allows multiple associations, and crucially relies on unassociated entities in the intermediate stages of a derivation, although such entities are disallowed in the final representation.

The Theoretical Framework / 55

need for simultaneous access to more than one level will be crucial for the account of case selection in Hindi. Traditional theta role representations conflate the number, relative prominence, and the semantic content of the arguments of a predicate. We factor this information into two separate levels of structure, ARG STR and SEM STR. This separation has several desirable consequences. First, it allows for the formal expression of what corresponds to "multiple theta roles" on the same argument, "theta roles" that are not arguments, and arguments without "theta roles". It also overcomes the inadequacy of "theta roles" to account for case selection. Further, bestowing a privileged status to function selecting meanings by representing them as "theta roles" is avoided. As we will see, case selection in Hindi depends not only on semantic information expressed by theta roles, but on a wider range of semantic configurations involving "conscious choice", "source", "goal", "containment", and "contact". The relative prominence relations currently expressed as the Thematic Hierarchy are viewed as the cumulative result of a number of interacting prominence relations stated on pairs of entities at SEM STR. These interacting prominence relations serve to order the ARGs in the ARG STR of a predicate. The resultant Argument Hierarchy plays a significant role in identifying the L-SUBJ, which is crucial for subject selection and anaphora in Hindi. The notion of L-OBJ participates in case preservation, the middle construction, and case and agreement in complex predicates. I will also argue that in Hindi, the phenomena of case selection (Chapter 4), as well as pronominal and reflexive anaphora, control, and gapping (Chapter 6) require reference to grammatical function information, regardless of how this information is represented.

4 The Case System In Hindi, as in most other South Asian languages, there is no one-to-one correspondence between grammatical functions and case. A nominative argument, for instance, may be either the subject or the object of a clause. Conversely, the subject may be nominative, ergative, dative, instrumental, genitive, or locative; a primary object may be accusative or nominative. On the other hand, there is often a systematic correspondence between meaning and case marking. Semantic entities that constrain or are constrained by case may be predicate role meanings such as agent, instrument, or location, inherent nominal meanings such as animacy or humanness, or discourse meanings such as definiteness. Conceptions of the function of case and case marking have varied widely among researchers. Within one view, the function of case marking at least on subjects and objects is merely to distinguish an agent NP, for instance, from a patient NP in a sentence (Comrie (1978:379), Dixon (1979:68-69)). In contrast is the view that cases have a positive semantic content that can be identified crosslinguistically, even when they may possibly involve a differentiating function (Jakobson (1958), Hopper and Thompson (1980), Wierzbicka (1980, 1981)). The present study of the Hindi case system holds that cases may be associated with specific grammatical functions, or specific meanings, or for that matter, with both.1 A distinction crucial in the discussion is that between CASE FEATURES and CASE MARKINGS.2 Case features are drawn

1 This view of case is by no means unique. It is close in spirit to that expressed, for instance, in Goddard (1982), 2 The distinction between case features and case markings has been repeatedly argued for in the literature (see Wierzbicka (1980; 1981; 1988), Goddard (1982), among others). The term "feature" has been used in the context of case in a very different sense from the one used here. Jakobson (1958), in a study of the case system of Russian, argues for a feature decomposition of case on the basis of semantic considerations and case syncretism. Neidle (1988) modifies and extends the three feature system proposed by Jakobson. What I call a case feature, for instance,

56

The Case System / 57

from a universal inventory, and demand language independent characterization. In the organization of grammar outlined in Chapter 3, the case feature system forms part of the subsystem of grammatical features at the level of GF STR, and the case features are associated with ARGs at ARG STR. Case markings are language specific morphological elements associated with the case features. The association of case features with ARGs in a language is distinguished along two dimensions in this analysis. If the association of a case feature makes reference to grammatical functions, it is DIRECT case association, as defined here.3 If no reference is made to grammatical functions, it is INDIRECT case association. Similarly, depending on whether or not the association of a case feature makes reference to meaning, it may be SEMANTIC case association or NON-SEMANTIC case association. For ease of reference, I use the label DIRECT CASE, INDIRECT CASE, SEMANTIC CASE, and NON-SEMANTIC CASE to refer to case features whose association is direct, indirect, semantic and non-semantic respectively.4 Nominative, ergative, genitive and accusative are direct case features in Hindi. Of these, ergative is semantic as well. Indirect case features are those not constrained by GFs. Among the indirect case features, genitive case may be semantic or non-semantic. All other indirect cases are semantic. This chapter presents an analysis of the direct case features in Hindi. A study of the indirect case features is provided in Chapter 7. The exact significance of this classification for a theory of case, and its consequences for existing proposals, will become clear only after an analysis has been presented. However, in order to avoid confusion with constructs available in existing proposals, a few clarificatory statements are in order. Indirect or semantic case on subjects does not have to be "quirky" or idiosyncratic. Rather, it is systematic and predictable. As will eventually be demonstrated in Chapter 7, indirect semantic case in Hindi is rigorously dative, refers in Neidle's system to the set of features [-Locational, +Quantifying, -Directional, -Partial]. 3 The term "direct case" has been used in the literature in many ways, different from the way it is defined here.. 4 This view allows for the possibility of the same case feature belonging t o two different classes in the same language, depending on the manner of association. Thus, as is well-known (Thráinsson (1980), Levin and Simpson (1981), Andrews (1982) and the works cited therein), accusative case in Icelandic may be either lexically stipulated or assigned by virtue of grammatical function. In our terminology, it may be either direct or indirect. In Hindi, we will see that GEN case may be direct non-semantic, indirect non-semantic, or indirect semantic, depending on the principle of association.

58 / Arguments in Hindi

conditioned by semantic principles, whether it appears on subjects or nonsubjects. The idea that regular, predictable case can be governed by meaning is not new. It is uncontroversial that the choice of case and prepositions on obliques and adjuncts (e.g. put the book on/under/near the table ) is governed by semantic factors. However, a general practice has been to treat instances like dative, accusative, and genitive case on subjects as lexically stipulated idiosyncratic case. For instance, Zaenen, Maling, and Thráinsson (1985:465) make a distinction between semantic case and idiosyncratic case. Semantic case appears on obliques and adjuncts, while idiosyncratic case appears on subjects and objects. "Idiosyncratic or lexical case marking is an idiosyncratic property of a lexical item, assigned by a verb, preposition or adjective. We assume that idiosyncratic case is associated with a particular thematic role, and that this case marking is assigned before thematic roles are associated with grammatical functions." The assumption that (in nominative-accusative languages) nonnominative case on subjects and non-accusative case on objects is lexically stipulated is built into GB case theory as well. These subjects and objects bear "inherent case", assigned at d-structure and associated with theta marking (Chomsky (1986:193)). Inherent case is "idiosyncratically selected by a verb" (Belletti and Rizzi (1988:343)). I argue in this chapter that the ergative case on subjects in Hindi is not idiosyncratic, but is semantically governed. In Chapter 7, I will demonstrate the semantic regularity of other non-nominative subjects as well. These conclusions show the need to recognize direct semantic case and indirect semantic case on subjects. It is likely that a careful examination of the alleged instances of "quirky" case in other languages will also lend to a similar treatment. Real idiosyncracy may then be reduced to a minimum. What I have called non-semantic case corresponds to structural case in GB. Direct case here corresponds to functional case in Zaenen et al (1985). But no treatment as far as I know recognizes direct semantic case, or distinguishes between direct and indirect non-semantic case.5 The need for these distinctions will become apparent to the reader in the course of the discussion in this chapter and in Chapter 7. 5 Kiparsky (1987) distinguishes between grammatical linking, which is the linking of syntactic arguments to term functions via case, agreement, or word order, and semantic linking, which is linking to non-terms. It must be pointed out that grammatical and semantic linking is orthogonal to our four-way classification of case.

The Case System / 59

Our analysis also results in several other important findings. First, even though Hindi has ergative and nominative case marking on subjects, and nominative and accusative case marking on objects, Hindi is not a typical example of ‘split ergativity’. The same verb can take combinations of either an ergative or a nominative subject with either a nominative or an accusative object. Second, the case of uninflected objects must be analysed as nominative case. It cannot be treated either as zero-marked accusative or as absolutive. Third, in spite of identity of case marking, dative and accusative must be distinct case features in Hindi. The need to differentiate between dative and accusative case features provides motivation for distinguishing between case features and case markings. These findings have consequences for a number of the prevalent ideas about case in GB. First, Hindi exhibits an ergative-nominative case alternation on agentive subjects, depending upon verb morphology. As we shall see in Chapter 7, the ergative and nominative case on agentive subjects alternates further with instrumental case under the meaning of capability. Neither ergative nor instrumental case on agents can be treated as "inherent" case in GB, since inherent case cannot be assigned to external arguments because they are not governed by the verb (Belletti and Rizzi (1988), Stowell (1989)). Nor are they "structural" case, because the structural case assigned to subjects by INFL is nominative. As far as I know, no mechanism is currently available in GB to predict ergative-nominativeinstrumental alternation on subjects in Hindi. Second, the case on the primary object in Hindi alternates between nominative and accusative, depending on its animacy and definiteness. Neither nominative nor accusative case could be inherent case, because they are not governed by theta roles. Given that the alternation is restricted to primary objects, both nominative and accusative case on objects must be structurally assigned. However, it is not clear how a verb would be sensitive to animacy and definiteness in order to assign the appropriate structural case. Third, in some dialects of Hindi, the nominative-accusative alternation on the object argument of the active is preserved on the corresponding subject argument of the passive. This nominative-accusative case could not be structural case, because structural case is not preserved under passivization. On the other hand, it could not be inherent case either, because, as stated earlier, the case on the argument is perfectly predictable from its being the object of the active. Furthermore, had nominativeaccusative case on the object been inherent, why should the argument be

60 / Arguments in Hindi

forced to move into the subject position when the attachment of the passive morphology absorbs the structural case assigned by the verb?

4.1. The Case Marking System I begin with an introduction to the case markings by means of which the case features in Hindi are signalled, and then go on to the distribution of the case features in terms of meanings and grammatical functions. In encoding case, Hindi employs three types of overt markings, distinguished here as stem forms, clitics, and postpositions. The motivation for the distinction as well as the terminology will become clear as we proceed. This section lays out the details of the case markings in Hindi, identifying the ones that will be most central for our purposes. 4.1.1. The first set of markings I will discuss are the "case clitics", given in (1)-(6), along with illustrative examples: (1)

-ne : as in bacce-ne b a c c e - n e kitaab pa±¥ii. child-ne book read-PERF The child read a book.

(2)

(3)

-ko : as in bacce-ko a.

niinaa b a c c e - k o uƥaayegii. Nina child-ko lift-FU Nina will pick the child up.

b.

niinaa-ne b a c c e - k o kitaab dii. Nina- ne child-ko book give-PERF Nina gave the child a book.

-se : as in ±a”±e-se, kamre-se a.

raam-ne ±a”±e-se sããp-ko maaraa. Ram-ne stick-se snake-ko kill-PERF Ram killed the snake with a stick.

b.

baccaa k a m r e - s e niklaa. child room-se emerge-PERF The child emerged from the room.

The Case System / 61

(4)

-kaa : as in bacce-kaa raam-ne b a c c e - k a a naam pukaaraa. Ram-ne child-kaa name call-PERF Ram called the child's name.

(5)

-mã : as in kamre-mã baccaa kamre-mã baiÆ¥aa child room-mã sit-PERF The child is sitting in the room.

(6)

hai. be-PR

-par : as in darvaaze-par raam-ne kurtaa darvaaze-p a r Æãã˚gaa. Ram-ne shirt door-par hang-PERF Ram hung (his) shirt on the door.

Following convention, I use the label NOMINATIVE to refer to the nominal form without a clitic (NOM: niinaa ((2a)), baccaa ((3b))). To refer to the nominal forms with the case clitics, I use the labels ERGATIVE (ERG: baccene ((1))), ACCUSATIVE (ACC: bacce-ko ((2a))), DATIVE (DAT: bacce-ko ((2b))), INSTRUMENTAL (INST: ±a”±e-se ((3a))), GENITIVE (GEN: bacce-kaa ((4))), and LOCATIVE (LOC: kamre-mã ((5)), darvaaze-par ((6))).6, 7 It is not crucial for our purposes whether these case markings are called clitics or affixes. One must note, however, that a pause may intervene between the nominals and their case markings. Furthermore, the case markings exhibit phrasal scope. Thus, when two nominals are coordinated, which is possible only if they are identically case marked, the scope of a case marking may extend over both nominals, as in madraas aur haiderabaadse "from Madras and Hyderabad", instead of madraas-se aur haiderabad-se. Therefore I assume that the case markings in (1)-(6) concatenate with the 6 Detailed discussions of the uses of these clitics can be found in traditional as well as modern linguistic descriptions of Hindi (e.g. Kellogg (1875), Guru (1922), Sharma (1958), Bahl (1967), Srivastava (1969), Pray (1970), McGregor (1972)). 7 The one other clitic is tak "until", illustrated below: raam k a l - t a k yahãã t¥aa. Ram yesterday -tak here be-PA Ram was here until yesterday. This clitic is not of special interest to us because, unlike other clitics, it does not appear on TERM ARGs as far as I know.

62 / Arguments in Hindi

noun phrasally and not lexically. Their structure (with only the essential details) may tentatively be given as in (7): (7)

/\ N

N CASE

Hence the term clitic rather than affix. 4.1.2. The nominal stems with clitics in (1)-(6) have forms such as bacce "child", kamre "room", and darvaaze "door". The corresponding NOM stems, that is, those without clitics, have the forms baccaa, kamraa, and darvaazaa respectively. This distinction constitutes the type of case marking we call "stem forms". Nominals exhibit three case stem forms, NOMINATIVE, NONNOMINATIVE, and VOCATIVE, even though in large classes of words, the distinctions are not visible on the surface.8 Although the case stem distinctions will not be the focus of any of our discussion, it will become obvious that the stem forms must carry partial case information.9 The stem form distinctions of a masculine nominal in the singular and plural are given in (8) for illustration. I ignore vocative case, as it does not participate in morphosyntactic regularities. 10 (8) singular

plural

a.

NOM

baccaa

bacce

b.

NONNOM

bacce

baccõ

8 The NOMINATIVE and NONNOMINATIVE stem distinction is described in the literature on Hindi as a distinction between direct (or bare) and oblique (e.g. Kachru (1965), Kachru (1980:26), Pray (1970), McGregor (1972), Hook (1979)). 9 It is by no means assumed that the stems that exhibit a case distinction are underived or morphologically unrelated forms. Rather, I assume that just as stems carry information such as SINGULAR or PLURAL, they also bear the case features NOM, NONNOM, or VOC. A detailed and insightful morphological analysis of the noun stem forms is given in Pray (1970). 10 The exact shape of the stem forms is dependent on (a) the final phonological segment, and (b) gender (there being a two way gender distinction in Hindi, masculine and feminine).

The Case System / 63

The NOM stems do not have any case clitics attached to them.11 Clitics can be attached only to NONNOM case stems. Conversely, NONNOM stems must take a case clitic after them.12 4.1.3. We refer to the third type of case markings in Hindi as "postpositions". The postpositions generally demand that their nominal object must bear the clitic -ke .13 Examples of these are given in (9): (9) a.

bacce -ke liye child(NN) G(NN) for 'for the child'

b.

bacce -ke dvaaraa child(NN) G(NN) through (the agency of) 'through (the agency of) the child'

The terminological distinction between clitics and postpositions, though not important for our purposes, deserves clarification. The two classes of case marking have not generally been distinguished in the literature on South Asian languages. One difference between the two classes lies in that the clitics are bound forms; nothing can intervene between them and their host stems. The postpositions, on the other hand, are free words. Therefore, it might perhaps be more accurate to think of the "clitics" as clitic postpositions and the "postpositions" as nonclitic postpositions. However, I 11 VOC stems are not relevant for any syntactic generalizations, and will therefore not be discussed any further. 12 An exception to this generalization in many dialects of Hindi are locative destinations, which appear in the NONNOM stem form without a clitic. That they must be treated as NONNOM even in the absence of surface differences between the NOM and NONNOM forms is shown by the facts of modifier agreement, discussed later in this chapter. According to McGregor (1972:50), a place, that is, "a geographical locality denoted by a place name, or is otherwise felt as a specific destination," is "best indicated by a noun in the oblique case without the following postposition." 13 The genitive clitic, like nouns, inflects for case, gender, and number features. The form -kaa in (4) is the masculine singular NOM form of the clitic, the plural form being -ke, and the feminine form -kii . The NONNOM form of kaa is also -ke , as in (9). Pray (1970) convincingly accounts for the NONNOM case on the genitive as part of the larger phenomenon of modifier-head agreement in Hindi, a detailed discussion of which is not relevant for our immediate purposes.

64 / Arguments in Hindi

assume that the nominals marked with clitics in (1)-(6) are NPs, not PPs.14 In other words, they do not carry categorial information, and therefore, cannot determine the category of their mother. Nominals with any of the clitics in (1)-(6) can be grammatical subjects. If we assume that only NPs, not PPs, can be subjects, it follows that the expressions in bold face in (1)(6) are NPs. The only nominal form with a postposition that I have found to function as a subject is N-ke paas "near N", e.g. bacce-ke paas "near the child". By the above argument, this expression must also be an NP. Now, it exhibits a difference of behaviour depending on whether or not it is in subject position. For instance, in subject position, paas is not separable from N-ke by intervening material.15 It could be, therefore, that paas may or may not carry category information, and that N-ke paas may be either an NP or a PP . As for the rest of the postpositions, their categorial status requires further study. Again, nothing in this thesis crucially hinges on the categorial status of what I have called clitics and postpositions. The tests of subjecthood and separability are given only by way of justifying the terminology. 4.1.4. At the heart of this study lie the syntax and semantics of the case clitics in (1)-(6). A major reason for making these clitics the focus of our investigation is the striking pattern in the syntax of Hindi that allows all the nominals with these case clitics to function as grammatical subjects of clauses. I give examples in (10), and will argue later that the nominals in bold face are indeed grammatical subjects. (10) a.

b.

ravii kelaa k¥aa Ravi-N banana-N eat Ravi was eating a banana.

rahaa

t¥aa. PROG

be-PA

ravii-ne kelaa k¥aayaa. Ravi-E banana-N eat-PERF Ravi ate the banana.

14 Gair and Wali (1989) assume that nominals marked for case with clitics and postpositions in Hindi are PP s, for an analysis of verb agreement. The inconsistencies that result from this assumption are discussed in Chapter 5. 15 For instance, when N-ke paas is not a subject, the expression mere g¥ar-ke bilkul hii paas "very near my house" is perfectly possible.

The Case System / 65

c.

ravii-ko kelaa k¥aanaa t¥aa. Ravi-D banana-N eat-NF be-PA Ravi was obliged to/needed to eat the banana.

d.

ravii-se kelaa k¥aayaa Ravi-I banana-N eat-PERF Ravi couldn't eat the banana.

not

nahññ go-PERF

e.

ravii-ke caar bacce t¥e. Ravi-G four children-N be-PA Ravi had four children.

f.

r a v i i - m ã bilkul dayaa nahññ t¥ii. Ravi-L at all mercy-N not be-PA Ravi had no mercy at all.

gayaa.

Our primary concern in Chapter 7 will be to provide evidence for the subjecthood of the nominals in bold face in (10c-f), identify the principles governing the distribution of case on subjects, and explore the consequences of these principles for formal syntactic theory.

4.2.

Case Features: Their Syntactic and Semantic Substance

4.2.1. As mentioned before, Hindi shows patterns of syntactic behavior that call for a distinction between the systems of case and case marking. A case system is the system of CASE FEATURES conventionally labelled as nominative, accusative, dative, instrumental, and so on. These case features are associated in individual languages with language specific CASE MARKINGS. I assume that the labels associated with case features are drawn from a universal inventory in a theory of case. Underlying the conventional terminology for case features is a history of usage that has established an association of each label with certain crosslinguistic properties of syntactic and semantic patterning. These properties can by no means define the case features, but they can certainly be used as prototypical clusterings that provide rational criteria for the identification of these features across languages. To elaborate the point, I quote Goddard (1982:169): "For instance, a genitive case must include among its functions that of indicating the possessor; an accusative case should include the patient of a transitive affirmative sentence; an instrument case should mark the weapon or tool

66 / Arguments in Hindi

used to accomplish an action. In any individual language a category given one of these names may IN ADDITION have certain other functions. For instance in Polish the genitive case extends to objects in negated sentences; the accusative is used in expressions indicating duration; certain postpositions require the genitive or the accusative case." In other words, case features are crosslinguistically characterizable in terms of a core set of properties, while allowing for language specific idiosyncratic extensions. But they are not reducible to any other kind of information. Like grammatical categories such as noun and verb, and grammatical functions such as subject and object, case features have an independent status, with constraints on their association with grammatical functions and meanings. There are two conditions that the use of case features in the grammar of a language must satisfy. First, the featural distinctions of case must be motivated by morphological and syntactic regularities internal to that language. Second, the labelling of these distinctions must be grounded in explicit criteria in a theory of case in universal grammar. In the absence of agreement among researchers who have used the labels, or even among those who have tried to articulate explicit criteria, I assume the statements in (11)-(12) as the core criteria. Despite disagreements, treatments of case within descriptive and theoretical work share many intuitions, implicit in the use of case terminology. The intuitions are based on pervasive generalizations found repeatedly across languages. The statements in (11)-(12) reflect some of these shared intuitions: (11) a.

NOM

b.

ERG

c.

ACC

case of subjects. If a language has two distinct cases associated with subjects, one inflected and the other uninflected, NOM typically refers to the case of the uninflected one. 16 case of inflected subjects of transitive verbs in a language where subjects of intransitive verbs are typically uninflected. case of objects.

16 By case "inflection", I mean an overt case ending, whether in the form of a stem ending, affix, clitic, or pre-/postposition.

The Case System / 67

(12) a. b. c. d.

DAT INST GEN LOC

case case case case

of goals/recipients of instruments used to accomplish an action of possessors / adnominal case of physical locations

The statements in (11)-(12) apply to the CORE uses. An individual language may either restrict or extend these uses. Thus, ERG in Hindi on the one hand is restricted to sentences with perfective aspect, and on the other is extended to subjects of intransitive verbs with an "agentive" meaning. Likewise, INST is extended to the "demoted" agents of passive and causative constructions. 4.2.2. Based on these observations, we relate the case markings in Hindi to the universal case features according to their core syntactic and semantic distribution. The case features employed in Hindi along with their markings, and some of their salient semantic and syntactic properties, are given in (13)-(14): feature

marking

syntax and semantics

(13) a. b.

NOM ERG

` -ne

c.

ACC

-ko

a. b.

DAT INST

-ko -se

c.

GEN

-kaa

d.

LOC1

-mã -par

subject ((2a)); inanimate prim. object ((1)). agentive subject with verb in perfective aspect ((1)). primary object ((2a)).

(14)

LOC2

goal ((2b), (10c)). (i) instrument ((3a)), (ii) source ((3b)), (iii) inter-mediary agent (causee), (iv) demoted agent of passive. possessor (involving ownership of something, or relationship to somebody ((4))). 'in', 'within' ((5), (10f)). 'on', 'at' ((6)).

68 / Arguments in Hindi

Given in (13)-(14) is a crude approximation of the most obvious patterns of semantic and syntactic distribution of the case features in Hindi.17 Our attempt will be to provide an explicit account of the distribution of these cases, the focus being their syntactic patterning. It must be observed that the marking -se in Hindi is used with more meanings perhaps than any other case marking. First, among other meanings, it indicates both instruments and sources. Thus, we may think of this marking as corresponding to two case features, namely instrumental and ablative. Or we might collapse them as a single case feature, and either refer to it as INSTRUMENTAL (as in most of the literature on Hindi) or as ABLATIVE (e.g. Kellogg (1875)). In this thesis, we will systematically refer to it as INST. This case marking also indicates the meaning of path, as in is raste-se "using this path" or "through this path". Once again, I refer to this use of the marking as INST. The meaning of path is alternatively indicated with the postposition dvaaraa "through". Yet another distinct use of -se is in the "comitative" sense, as in anune ilaa-se baat kii "Anu spoke with Ila." Whereas there appears to be a cross-linguistic semantic regularity underlying this use, this regularity is difficult to identify precisely. I will gloss this use of -se as COM(ITATIVE) when required by the data, but will have nothing more to say about it. A note about the absence of ABSOLUTIVE case in (11) and (13) is perhaps in order. ABSOLUTIVE case, viewed in the literature as the case of the uninflected object in the ergative-absolutive system, might appear to be relevant in the context of Hindi. Given that Hindi has an ERG inflection, and in addition, uninflected objects, a typical move would be to call the uninflected objects ABS, as has been claimed for Australian languages (e.g. Dixon (1979)). Now, Goddard (1982) argues that the distinction between ABS and ACC in many Australian languages is incorrect, and that absolutive case is not a case at all (Goddard (1982:183)). The uninflected objects in these languages must be analysed as uninflected ACC, as has been argued for languages like Russian and Polish (Wierzbicka (1980; 1981)). In Hindi, there is evidence against the analysis of the case of uninflected objects as ABS. Distinguishing the case of uninflected subjects 17

Implicit in the notation of LOC1 and LOC2 is the assumption that LOC is a cover feature within which additional features such as DIRECTIONALITY, CONTAINMENT, CONTACT, and PROXIMITY are required to make further distinctions. Thus, -mã "in" (LOC1 ) marks CONTAINMENT while -par "on"/"at" (LOC2 ) marks CONTACT.

The Case System / 69

and objects as NOM and ABS would not simplify the system in any way. Instead, it would only add complexity, and obscure significant generalizations. First, there would be a complete neutralization of NOM and ABS both on nominals as well as on modifiers of nominals, which agree with them. The statement of modifier agreement would have to refer to the disjunction of NOM and ABS on the one hand, and the rest of the case features on the other, instead of NOM on the one hand and the rest of the features on the other.18 It would also force a disjunction in the statement of verb agreement, discussed in Chapter 5. Third and most important, implicit in the treatment of the uninflected objects as bearing ABS case is the assumption that Hindi has a split case system. If so, we predict that when the subject of a transitive verb is ERG, the object must be ABS, and when the subject is NOM, the object must be ACC. This prediction is entirely false: both ERG and NOM subjects cooccur with both NOM and ACC objects. Therefore, I assume that the feature ABS is not part of the case system of Hindi. 4.2.3. Some of the language specific extensions of the core uses of case features are semantically motivated, and found repeatedly in natural languages. Thus, INST, the case of the means (mediating source) of action, is extended in numerous languages to express the source of action in passives and causatives. The semantic motivation for other extensions may be intuitively less obvious. For instance, in Bengali, the GEN of possessors is extended to mark experiencer-goals, while in Malayalam, the DAT of goals is extended to mark possessors. In contrast, many of the logically conceivable extensions of the core uses of case features are unattested in languages, for example, the use of same case to mark reason and locative goal, end point of time and locative source, or path and primary object. An adequate theory of case must predict the crosslinguistic patterning of natural, possible and impossible extensions of the core uses of case features. There have been serious attempts in this direction (e.g. Anderson (1971; 1977), Croft (1986)). Unfortunately, such studies have not sparked the kind of investigation this submodule of case theory deserves. A detailed articulation of such a submodule is, of course, far beyond the present study.

18 As we will see later, the statement cannot be in terms of the presence or absence of case inflection on the nominal.

70 / Arguments in Hindi

4.3. The Syntax of Direct Case in Hindi In Hindi, a statement of the distribution of a case feature on arguments may make reference either to the meaning of the argument or to its grammatical function. In what follows, we assume a classification of case governed by whether or not the principle that associates the case appeals to semantic or GF information. If the principle that associates a case makes reference to meaning, it associates a SEMANTIC CASE. If not, it associates NONSEMANTIC CASE. If the principle requires GF information, it associates DIRECT CASE. If not, it associates INDIRECT CASE. A case may be associated with an ARG purely on the basis of its GF. In that event, we have the association of a DIRECT NON-SEMANTIC CASE. If a principle refers to both meaning and GF, it associates DIRECT SEMANTIC CASE. Often, case is associated with an ARG entirely by virtue of its meaning, with no reference to its GF. We then have INDIRECT SEMANTIC CASE. On the other hand, a principle of case association may refer neither to meaning nor GF. It then associates INDIRECT NON-SEMANTIC CASE. The motivation for this four-way distinction will emerge as we proceed. This section provides an account of the distribution of the direct case features in Hindi. 4.3.1. ERGATIVE and NOMINATIVE on the Subject A nominal bearing almost any of the case markings in (13)-(14) can function as the grammatical subject of a clause in Hindi, a property that Hindi shares with most other South Asian languages. Among the different case marked subjects, the direct case subjects, namely, ERG and NOM, are accepted uncontroversially as grammatical subjects.19 The intuitive reason for this acceptance is that these are the two cases that appear on the "agent" of a base (that is, underived) verb: this agent is ERG when the verb is morphologically PERFECTIVE. 20 Elsewhere, it is the default NOM. 21 Let us examine the alternation in some detail. 19 The third case in (13), ACC case, is a direct case in the dialect of Hindi under discussion precisely by virtue of being associated with grammatical objects. Naturally, then, it can never appear on grammatical subjects. In fact, it is the only case feature in (13)-(14) that does not appear on subjects. 20 Kachru and Pandharipande (1979) show that the alternation is a purely morphological phenomenon; Hindi has no syntactic ergativity. 21 Moreover, the alternation is restricted to some Western Hindi dialects (Srivastava (1969)): ERG case is entirely absent in other dialects.

The Case System / 71

4.3.1.1. Perfective Aspect In the following transitive sentences, subjects are ERG if the verb is in the simple past tense ((15a)), or in perfective aspect ((15d-f)). Otherwise, they are NOM ((15b, c)). (15) a.

raam-ne ravii-ko Ram-E Ravi-A Ram beat Ravi.

piiÆaa. beat-PERF

b.

raam ravii-ko piiÆtaa hai. Ram-N Ravi-A beat-IMPERF be-PR Ram beats Ravi.

c.

raam ravii-ko Ram-N Ravi-A Ram will beat Ravi.

piiÆegaa. beat-FU

d.

raam- ne ravii-ko Ram-E Ravi-A Ram had beaten Ravi.

piiÆaa t¥aa. beat-PERF be-PA

e.

raam- ne ravii-ko Ram-E Ravi-A Ram has beaten Ravi.

piiÆaa hai. beat-PERF be-PR

f.

raam-ne ravii-ko piiÆaa hogaa. Ram-E Ravi-A beat-PERF be-FU Ram must have beaten Ravi.

In (15a) and (15d-f), the main verb piiÆ "beat" carries the inflection -aa, and the subject bears the ergative marking -ne, while in (15b, c), there is neither the ERG marking on the subject, nor the formative -aa on the main verb. Thus, a correlation holds between ERG case marking on the one hand and the form of the main verb in simple past and perfective aspect on the other.22 Assuming that the inflection -aa bears the morphological feature

22 This correlation has been repeatedly observed in the literature (e.g. Platts (1898), McGregor (1972)). Note that the inflection on the verb in (15b), glossed as IMPERFECTIVE, has the form -taa , not -aa.

72 / Arguments in Hindi

PERF, the surface condition for ERG case may be identified as PERF on the

main verb.23 4.3.1.2. Transitivity The NOM-ERG alternation is claimed in the literature, in spite of observed and catalogued exceptions, to be largely governed by the transitiveintransitive distinction, ERG being associated specifically with the subject of transitive verbs (Srivastava (1969), McGregor (1972), Kachru (1980)). This apparent correlation is misleading in Hindi, as there are instances of both transitive and intransitive verbs that take NOM or ERG subjects (Kachru and Pandharipande (1979), Amritavalli (1979)). "With some transitive verbs, such as samaj¥naa 'understand', and jannaa 'give birth to', the subject may optionally take ne. Subjects of other transitive verbs such as bolnaa 'speak', laanaa 'bring', do not take ne. The intransitive verbs that allow ne with their subject are, for example, nahaanaa 'bathe', c¥ññknaa 'sneeze', k¥ããsnaa 'cough'." (Kachru (1980:63-64)). Thus, verbs in Hindi fall into three classes, independently of transitivity: those that, given the required aspectual conditions, take (i) only NOM subjects, (ii) only ERG subjects, and (iii) either NOM or ERG subjects.24 Let us look at instances of intransitive verbs that fall into the three classes: gir "fall", soo "sleep", and jaa "go" take only NOM subjects ((16)); nahaa "bathe", and c¥ññk "sneeze" take only ERG subjects ((17)); and the subjects of cillaa "shout", ciik¥ "scream", naac “dance”, gaa “sing”, and k¥el "play" may be either ERG or NOM ((18)).

23 As mentioned in §2.2., the marking -aa in (15d-f) indicates perfective aspect; tense information is borne by the auxiliary ho "be". In (15a), however, it indicates simple past tense. 24 That more transitive verbs belong to class (ii) than to (i) or (iii), and more intransitives to class (i), is perhaps what led to the generalization in terms of transitivity. Further, there exists a certain amount of variation among speakers with respect to ERG marking. While many speakers of Hindi, including all my informants and myself, allow ERG on the subjects of a subset of monadic verbs, I believe there are some who restrict it to transitive verbs, as in other Indo Aryan languages, like Marathi. Whether transitivity partially or entirely governs ERG marking for those speakers requires further investigation.

The Case System / 73

(16) a.

raam giraa. Ram-N fall-PERF Ram fell hard.

a.

raam-ne nahaayaa. Ram-E bathe-PERF Ram bathed.

b.

* raam -ne Ram-E

*

raam nahaayaa. Ram-N

giraa.

(17)

(18)

raam / raam-ne Ram-N Ram-E Ram shouted loudly.

b.

jor-se loudly

cillaayaa. shout-PERF

(19)-(21) show a similar pattern among transitive verbs: (19) a.

raam-ne ßiißaa to±aa. Ram-E mirror-N break-PERF Ram broke the mirror.

b. * raam Ram-N

a.

raam ßiißaa laayaa. Ram-N mirror-N bring-PERF Ram brought the mirror.

b. * raam-ne Ram-E

ßiißaa

to±aa.

(20)

(21)

raam / raam-ne samj¥aa k i g¥ar Ram-N Ram-E think-PERF that house-N I-G Ram thought that the house was mine.

ßiißaa laayaa.

meraa hai. be-PR

Ergative marking cannot, therefore, be consistently associated with transitivity.25 We might use verbs that can choose between NOM or ERG subjects to look for a possible semantic basis for the choice. The contrast in case between (22a) and (22b) below clearly suggests a semantic conditioning: (22) a.

25

raam-ko acaanak ßer dik¥aa. vah / *us-ne cillaayaa. Ram-D suddenly lion-N appear-PERF he-N he-E scream-PERF Ram suddenly saw a lion. He screamed.

Also see Kachru (1981).

74 / Arguments in Hindi

b.

us-ne / * vah jaan buuj¥ kar he-E he-N deliberately He shouted deliberately.

cillaayaa . shout-PERF

4.3.1.3. Conscious Choice While the ERG subject in (22b) carries the meaning of deliberate action, the NOM subject in (22a) carries the meaning of inadvertent action. The case marking contrast between (22a) and (22b) thus correlates with a semantic contrast, and suggests that ERG case may require that the argument be associated with the semantic property I tentatively call CONSCIOUS CHOICE of performing the action.26 Based on the optionality of ERG marking on the subject of verbs like c¥ññk "sneeze" and k¥ããs "cough", Kachru and Pandharipande (1979) argue that ERG in Hindi cannot be correlated with agency or volitionality, and therefore, with any semantic property. They conclude that ergative case has no systematic synchronic correlates, and appeal to historical development as the sole explanation. Under their account, the choice between ERG and NOM on the subject must be stipulated synchronically on every verb. Such a stipulation, however, would miss the synchronic generalization that the action referred to by the class of verbs that can take only ERG subjects (class (ii)) must be deliberate, and the action referred to by the class of verbs that do not take an ERG subject (class (i)) is largely nondeliberate. Also, when a NOM subject cooccurs with verbs that have an option between NOM and ERG subjects, the action must be nondeliberate. Furthermore, whenever a new verb is created or introduced into the language, it is possible to predict on the basis of its meaning whether or not its subject can bear ERG case. Such predictability points to the conclusion that there must be a systematic synchronic correlate of ERG case in Hindi.27 Now, all instances of verbs like cillaa "scream" cooccurring with ERG subjects are not necessarily associated with deliberateness; there are a few 26 The property that is termed "conscious choice" is very similar to the property in Gair (1971:242) associated with the subjects of "active sentences" which "commonly express conscious and/or voluntary participation in the action of the sentence; that is, the sentence implies action involving choice and control so that one could conceivably decide to do it or not." This property has also been crucially employed in an account of complex predicates in Sinhala i n Paolillo (1989). 27 One might of course encode "conscious choice" as a grammatical feature i n order to make it available to syntax without referring to semantics. However, the result would be encoding an entity at a level where it doesn't naturally belong.

The Case System / 75

instances that do not always conform to the pattern in terms of conscious choice. There are also isolated instances of verbs like bol "speak" and laa "bring" which may be associated with deliberateness, but cannot take ERG subjects (Kachru (1980)). Faced with such counterexamples, one possible conclusion is that the NOM-ERG pattern is governed by a semantic contrast we have not been able to clearly identify yet. Further research may reveal a wider semantic generalization. An alternative position, which I believe is correct, is that these are instances of lexical exceptionality to the generalization we have identified, quite natural in lexical patterns. The notion conscious choice corroborates the intuition underlying the term "agentive marker" used to refer to the ergative (Kachru (1980), Kachru et al (1976)). This is supported by the use of an ERG subject with the verb jaan 'know', which cannot otherwise take an agentive subject ((23a)), to convey the specific meaning of deliberately acquiring knowledge ((23b)): (23) a.

raam jaantaa t¥aa Ram-N know-IMPERF be-PERF Ram knew that Sita was very ill.

k i siitaa bahut bimaar hai. that Sita-N very ill be-PR

b.

raam-ne jaanaa k i siitaa bahutbimaar Ram-E know-PERF that Sita-N very ill Ram found out that Sita was very ill.

hai. be-PR

Additional confirmation comes from constructions involving complex predicates, or predicates composed of two elements, where the first element may be a noun, an adjective, or the infinitival form of a verb. The second element is a verb which has undergone partial grammaticalization or lexical 'emptying' (Masica (1976)), thus having an impoverished semantic structure, and has been called a "light verb" (see §2.2.). The light verb in complex predicate constructions may determine the case of the subject. The main verb in (24) is pii "drink", which normally takes an ERG subject: (24) a.

ravii davaaii pii gayaa. Ravi-N medicine-N drink go-PERF Ravi (impulsively) drank up the medicine.

(*ravii-ne )

b.

ravii-ne davaaii pii ±aalii. Ravi-E medicine-N drink pour-PERF Ravi (deliberately) drank up the medicine.

(*ravii )

76 / Arguments in Hindi

The only difference between (24a) and (24b) is in the light verb. (24a) and (24b) share a sense of quickness of action. However, the verb jaa "go" adds to the action the meaning of "impulsively", or "without thinking" to the action described in (24a), while in (24b), the verb ±aal "pour" expresses violent action with emphasis on the "deliberateness" of the action. The case of the subject in (24), then, is determined by the light verb, which has the power to supply or to take away the semantic property responsible for ERG case.28 Having stated that the semantic factor conditioning ERG case is the notion of conscious choice, we find some apparent counterexamples. Thus, natural forces, when subjects, may appear in ERG case: (25) a.

havaa-ne patte bik¥er diye t¥e. wind-E leaves-N scatter give-PERF be-PA The wind had scattered the leaves.

b.

baadalõ-ne suuraj-ko g¥er liyaa hai. clouds-E sun-A surround take-PERF be-PR The clouds have surrounded the sun.

c.

tuufaan-ne ßiißaa storm-E glass-N The storm broke the glass.

to± diyaa. break give-PERF

Given that the wind, clouds, and storm are not sentient entities that can exercise conscious choice, the ERG SUBJs in (25) have two alternative accounts. One solution is to abandon the notion of conscious choice, and to look for some other semantic construct governing ERG case, for instance, SOURCE, INCEPTOR or CAUSE. However, we already have evidence for conscious choice as opposed to source, inceptor or cause, from the examples in (16)-(24). Furthermore, even though natural forces can take ERG case, other inanimate entities cannot, except, for example, in a fairy tale: (25) d. ?? patt¥ar-ne stone-E

ßiißaa glass-N

to± diyaa. breakgive-PERF

28 The relation between ERG case and the notion "conscious choice" i n complex predicates in Hindi is independently argued for in Butt (1990). G. Mahajan (1989) also gives an account of the syntax and semantics of complex predicates such as those in (24).

The Case System / 77

The stone/rock broke the glass.

Yet another piece of evidence in support of the conscious choice hypothesis comes from infinitival NPs, which can take all case markings, with the exception of ERG: (26) a.

[ilaa-kaa anuu-ko ci±¥aanaa ]NP mãã-ko acc¥aa na lagegaa. Ila-G Anu-A tease-NF -N mother-D good not strike-

FU

Mother didn't like Ila's teasing Anu. b.

[ilaa-ke anuu-ko ci±¥aane ]-se … Ila-G Anu-A tease-NF -I From/because of Ila's teasing Anu…

c.

[ilaa-ke anuu-ko ci±¥aane ]-par… Ila-G Anu-A tease-NF -L On Ila's teasing Anu… (= When Ila teased Anu…)

d.

[ilaa-ke anuu-ko ci±¥aane ]-kaa Ila-G Anu-A tease-NF -G result-N The result of Ila's teasing Anu…

e. * [ilaa-ke Ila-G

anuu-ko Anu-A

natiijaa…

ci±¥aane ]-ne tease-NF -E

This gap finds a natural explanation in the fact that infinitival NPs, which may be a source or a cause, nevertheless cannot be associated with conscious choice. The other solution is to assume that natural forces, although nonsentient in the real world, nevertheless are imbued at least partially with sentient properties, and this is reflected in linguistic patterning. Thus, while entities such as the sun, ocean, and rain qualify as natural forces, and hence can take ERG case, entities like sunlight, river, and water do not qualify as "natural forces", and cannot take ERG case. The special patterning of natural forces along with "agents" has been widely noticed across languages since Fillmore (1968), and is not particular to Hindi. I therefore assume that whatever general conditions are required to account for the partial patterning of natural forces with sentient entities will account for the appearance of ERG case on natural forces in Hindi as well. 4.3.1.4. Ergative Case Association

78 / Arguments in Hindi

The alternation in the case of the subject in (22)-(24), as in (16)-(21), shows that the distinction cannot be constrained in Hindi by transitivity. The same verb, whether transitive or intransitive, can take a NOM or ERG subject depending on the semantic environment, as in (22)/(23), and on the light verb it forms a complex predicate with, as in (24). Examples like (24) were advanced by Amritavalli (1979), who assumed that ERG case marking was indeed governed by transitivity, to argue that the notion of transitivity must be redefined as an inherent property of the verb, rather than as a contextual feature (of having or not having an object) as in Chomsky (1965). However, transitivity under such a redefinition would be an abstract diacritic feature without a substantive basis. I suggest that although the transitivity distinction does yield an apparent generalization, and for some speakers, it is perhaps the only relevant distinction, the wider generalization based on meaning accounts for contrasts such as in (22)-(24) as well.29 It is uncontroversial that, whereas not all grammatical subjects in Hindi are in ERG case, any argument that bears ERG case must be a grammatical subject.30 To summarize, then, an ARG bears ERG case if (i) it is the SUBJ at GF STR (ii) it is associated with the SEM STR configuration of CONSCIOUS CHOICE, and (iii) the main verb is associated with the feature [PERF]. Given the conception of interacting information structures outlined in Chapter 3, this is precisely the kind of situation we would expect to find. The association of ERG case with an ARG is constrained simultaneously by information from SEM STR and GF STR. The principle for associating ERG case with an ARG may be formulated as follows: (27) ERG association: Hindi [conscious choice] ˘

ARG

\



¿

SEM STR PRED

SUBJ ERG

ARG STR GF STR

PERF

29 This account of morphological ergativity in Hindi is very similar in spirit to the account of ergativity in Basque given in B. Levin (1983). 30 The ARG which bears ERG case when it is the SUBJ loses the ERG case when i t is no longer the SUBJ (under passivization and causativization).

The Case System / 79

(27) says: "If an argument is the grammatical subject, is associated with the meaning of 'conscious choice', and the predicate is associated with PERF aspect, then the argument is associated with ERG case." I will systematically use the convention of solid and dotted association lines to indicate the premise and consequence respectively of an if-then conditional. ERG case association requires information from both SEM STR and GF STR. So ERG case in Hindi is a direct semantic case. By the Elsewhere Condition (Kiparsky (1973)), the specific principle of ERG association in (27) takes priority over the general principle of NOM association in (28). NOM case association in (28) requires GF STR information but not SEM STR information. It is therefore one of direct non-semantic case. (28) NOM-SUBJ Association: Universal Canonical ARG

ARG STR

\ SUBJ

GF STR

NOM

(28) is a formal expression of the universal default principle whereby subjects are associated with nominative case. In contrast, (27) is a language particular principle. Recall the distinction made in Chapter 3 between CANONICAL and ABSOLUTE principles. Canonical principles characterize the most natural situations, distinguishing them from the less natural ones, and may be preempted. Absolute principles distinguish possible relations from impossible ones. (28) is a universal canonical principle. It is preempted by (27). (27) is not an absolute principle either. We will see in Chapter 7 that (27) may also be preempted, as a consequence of the interaction of principles. 4.3.2. G E N I T I V E on the Subject Consider the GEN nominals in (29): (29) a.

raam-ke baiÆ¥ne-par mãã-ne usko k¥aanaa Ram-G sit-NF -L mother-E pron-D food-N When Ram sat down, mother gave him food.

b.

raam-ke baiƥte hi

mãã-ne

usko k¥aanaa

diyaa. give-PERF diyaa.

80 / Arguments in Hindi

Ram-G sit-NF mother-E pron-D food-N As soon as Ram sat down, mother gave him food.

give-PERF

The SUBJ of the embedded non-finite verb baiÆ¥ "sit" in (29a, b) is in GEN case, as also in (26a-d). This looks like a typical instantiation of the GB idea that only INFL with AGR can assign NOM case to the subject. We might translate the substance of this idea as: only finite verbs can assign NOM case to the subject, even though case assignment to the object is not affected by finiteness.31 The effect of this statement is to make NOM case the special one, assigned under finiteness, with GEN case as default. Instead, we formulate the principle of GEN case assignment as the special one, shifting the burden from finiteness to non-finiteness:32 (30) Direct GEN Association : Universal Canonical ˘

ARG

\

… ¿

PRED

ARG STR

SUBJ GEN

GF STR NONFIN

(30) is an instance of direct non-semantic case association, like (28). By the Elsewhere Condition, (30) preempts (28). 4.3.3. ACCUSATIVE and NOMINATIVE on the Object Primary objects in Hindi either bear the ACC -ko ((31a)), or are uninflected for case ((31c)). I assume for now, and argue shortly, that the objects uninflected for case are in NOM case. 4.3.3.1. Distribution of Case on objects Consider the following examples: (31) a.

ilaa-ne e k Ila-E

one

bacce-ko uƥaayaa. child-A lift/carry-PERF

(= rise-CAUS-PERF)

31 ERG case, like NOM case, cannot appear on subjects of non-finite verbs. This does not require any special condition in Hindi, since ERG case requires PERF on the verb, and conscious choice of the ARG. In non-finite clauses, these two conditions are never simultaneously satisfied. 32 As we will see shortly, this allows us to express in a simple way the idea that in Hindi, nominative case is the default case of both subjects and objects.

The Case System / 81

Ila lifted a child. b.* ilaa-ne e k Ila-E c.

one

baccaa uƥaayaa. child-N lift -PERF

ilaa-ne e k haar uƥaayaa. Ila-E one necklace-N lift -PERF Ila lifted a necklace.

d.* ilaa-ne e k Ila-E

one

haar-ko uƥaayaa. necklace-A lift -PERF

A widely accepted generalization with regard to objects in Hindi is that the canonical case for animate objects is ACC ((31a)), and the canonical case for inanimate objects is NOM ((31c)) (e.g. Srivastava (1969), McGregor (1972)). This generalization is in fact true for many South Asian languages (Masica (1976)). Observe that the objects in (31) are all indefinite. In the absence of a determiner, nominals in Hindi may be interpreted as generic, definite, or indefinite. Now, even though the canonical case associations of objects are animate:ACC ((32)) and inanimate:NOM ((33a)), an inanimate object can be ACC if it is definite ((33c)) (McGregor (1972), Masica (1976), Comrie (1981)):33 (32)

ilaa-ne bacce-ko / * baccaa uƥaayaa. Ila-E child-A child-N lift -PERF Ila lifted the / a child.

(33) a.

ilaa-ne haar uƥaayaa. Ila-E necklace-N lift -PERF Ila lifted a / the necklace.

b.

ilaa-ne haar-ko uƥaayaa. Ila-E necklace-A lift -PERF Ila lifted the /*a necklace.

33 According to McGregor (1972), "-ko occurs in association with direct objects which are individualized to some extent, and to which a degree of contextual importance is thus attached; hence usually where direct objects refer to human beings, and certain animals, and quite frequently where they refer t o inanimate objects. Such words may be called definite direct objects." (p. 49-50)

82 / Arguments in Hindi

In (33b), the ACC inanimate object must necessarily be definite, while the NOM inanimate object in (33a) may be either definite or indefinite. In other words, inanimate primary objects, if ACC, must be interpreted as definite. We also encounter animate objects in NOM case. However, as we will see in Chapter 5, they are NOM only when they have a reading in which they are "incorporated" into the predicate.34 It is interesting that, although the lexical meaning of the nominal (animacy), and the meaning that a nominal acquires in discourse (definiteness), may constrain or be constrained by the choice between ACC and NOM, the choice itself may depend on individual verbs. A verb by virtue of its meaning may either require that its object be animate, or that it be inanimate. It may also be neutral to animacy. The choice between ACC and NOM is available only to the objects of those verbs that are neutral to the animacy of their objects. Thus, in contrast to the verb uÆ¥aa 'lift', the verb lik¥ 'write', can only take inanimate objects, and does not allow ACC objects even when they are definite: (34) a.

ilaa-ne

yah

k¥at

lik¥aa.

34 The pattern is actually more complex, because it involves further distinctions between specific and nonspecific within indefinites, and between human and nonhuman within animates. In contrast to the human object in (32), a nonhuman animate object may be NOM if it is nonspecific, as in (i):

(i)

ravii (ek) gaay k¥ariidnaa caahtaa hai. Ravi-N one cow-N buy-NF wish-IMPERF be-PR Ravi wishes to buy a cow (with no particular cow in mind).

(ii)

ravii ek gaay-ko Ravi-N one cow-A Ravi wishes to buy a (particular) cow.

k¥ariidnaa buy-NF

caahtaa hai. wish-IMPERF be-PR

The object in (i) is nonspecific (that is, the speaker cannot uniquely identify the referent); in (ii), it is ambiguous between specific and nonspecific readings. Now consider the contrasts in (iii)-(iv):

(iii)

ravii gaay-ko k¥ariidnaa caahtaa hai. Ravi-N cow-A buy-NF wish-IMPERF be-PR Ravi wishes to buy a particular cow.

(iv)

ravii us gaay-ko Ravi-N that cow-A Ravi wishes to buy that cow.

k¥ariidnaa buy-NF

caahtaa hai. wish-IMPERF be-PR

In (i), the animate object is nonspecific, with or without the determiner e k "one", because it is NOM. When the object is ACC , it may be either specific or nonspecific, as in (ii), but is indefinite because of the indefinite determiner. However, without the determiner ((iii)), the ACC object must be specific, though it may be either definite or indefinite. In (iv), with us "that", the object must be definite (that is, both the speaker as well as the listener can identify the referent). The issues raised by contrasts such as these require further investigation.

The Case System / 83

Ila-E this- N letter-N Ila wrote this letter. b.* ilaa-ne is Ila-E this-N N

k¥at -ko letter-A

write-PERF lik¥aa. write-PERF

Similarly, the verb piiÆ "beat" (more accurately, "spank") requires animate objects, and does not allow NOM objects. Thus, a verb may impose an animacy condition on its object, preempting the other conditions for alternation.35 In sum, the most salient conditioning factors of the ACC ª NOM case alternation on objects are animacy and definiteness, as has been widely noted in the literature. The effects of these conditioning factors can be captured in the grammar by postulating the following principles of case-argument association: (35) NOM-OBJ Association: Canonical in South Asian ARG

ARG STR

\ OBJ

GF STR

NOM

(35) is the canonical principle for assigning case to objects in South Asian languages, as in (31c). It associates a direct non-semantic case. The more specific principle in (36) takes priority over (35), as in (31a) and (32). (36) ACC Association: South Asian [+ anim]

SEM STR

ARG

ARG STR

\ OBJ

GF STR

35 Other verbs that take only NOM objects are banaa "make", pa±¥ "read", gaa "sing", and pii "drink". Verbs that are neutral to the animacy of their objects are giraa "drop", to± "break", k¥oj "search for", kaaÆ "cut", laa "bring", pahcaan "recognize", ubaal "boil", jalaa "burn", b¥ar "fill", pakaa "cook", and dek¥ "see". Verbs that can take only ACC objects are maar "kill" and bulaa "call". We must keep in mind that when speaking of individual verbs and verb classes, we must expect there to be considerable speaker variation and grey areas in speaker judgments.

84 / Arguments in Hindi

ACC

Notice that (36) makes reference to GF information. It therefore associates a direct case. It also refers to SEM STR information. However, this information is not part of the argument meaning but of the nominal meaning. In such an event, whether the case is semantic or non-semantic is an open issue. The specific principle in (36) applies obligatorily when the verb demands an animate object. It cannot apply when the verb demands an inanimate object. When the verb does not specify the animacy of its object, however, (36) may apply optionally to any object. As a consequence, inanimate objects of such verbs may be either associated with NOM by (35) as in (33b), or with ACC by (36) as in (33a). Now, if an inanimate object is associated with ACC, the principle in (37) imposes a restriction in discourse on the interpretation of the object:36 (37) ACC Inanimate OBJ Interpretation: Hindi [- anim]

SEM STR

ARG

ARG STR

ACC

[+definite]

GF STR DISCOURSE

(37) reads as: "An inanimate ARG associated with ACC case must be interpreted as definite." One might think of reversing the directionality of prediction and reformulating the statement as: If an object is inanimate and definite, it must be associated with ACC case. However, this generalization would be false, because a NOM object may be either definite or indefinite, as in (33a). Notice that the only difference between the principles of NOM case assignment to subjects in (28) and objects in (35) is in the specification of the GF. The two principles may therefore be collapsed as follows: 36 The NOM ª ACC alternation on objects is an areal property of South Asian languages by (35)-(36). The actual manifestation of the alternation may vary slightly from language to language, as indicated by (37). Wherever Hindi exhibits the alternation, Malayalam, a Dravidian language, does too. However, the only conditioning factor in Malayalam seems to be animacy. In other words, Malayalam allows only the default case associations ((35)-(36)) on objects.

The Case System / 85

(38) NOM Association: Canonical in South Asian ARG

ARG STR

\ UNRESTRICTED

GF STR

NOM

We might summarize the effects of (36)-(38) on object case as in (39): 37 (39)

Object Case

Verb

requires inanimate

Nominal

inanim

neutral inanim

anim

requires animate anim

37 One might ask why animacy is taken to be the conditioning factor in the canonical association of ACC and NOM, and in the three-way classification of verbs in (39). Let me take some of the possible alternatives and show that they do not condition the case of the object. The first factor is related to an aspectual theory of argument structure (Tenny (1987)): the case of the object might correlate with whether the object i s "capable of measuring out and delimiting the event" or not, a crucial test for which is the ability to cooccur with expressions like "in an hour" on the one hand and "for an hour" on the other. However, it can be shown that there is n o correlation between the delimiting property and the case of the object. Thus, an obligatorily ACC object in Hindi can cooccur with either a delimited event ("X killed Y in/*for an hour.") or a nondelimited event ("X spanked Y for/*in an hour."). Likewise, an obligatorily NOM object can cooccur with both types of events ("X read a book in/for an hour"). Another factor that might systematically correlate with the case of the object is how amenable an entity might be to individuation. Thus, while song i s individuatable, music is not. "Persons" are generally seen as being more "individual" than "things"; hence one might say that individuatable objects are associated with ACC case, and nonindividuatable objects with NOM case. However, in the sentence: "Ram read three books," the object is clearly individuated, and in "Ram drank water," it is not. Yet, both are obligatorily NOM in Hindi. Yet another possible factor is the thematic role structure of an individual verb: one might say that the obligatorily ACC objects are in fact not ACC but DAT, semantically determined by virtue of being the endpoint of the action. However, it is not clear to me how the thematic role of the object of the verb kaaÆ "cut", which may be either NOM or ACC , can be less of an endpoint of action than the object of piiÆ "spank", which must obligatorily be ACC .

86 / Arguments in Hindi

Case

NOM

ACC

Discourse

ACC

[+def]

Observe that the three principles of case association in (36)-(38) are recurrent across natural languages (Givón (1976), Comrie (1979), Hopper and Thompson (1980), Wierzbicka (1981), Croft (1988), among others). That is, case on the object in other languages is dependent on similar factors, even though not in exactly the same way. Therefore we must assume that these principles stem from universal grammar. However, each language must specify whether or not it makes use of these principles. In other words, (36)-(38) express natural but not absolute patterns of case. Recall that in (13)-(14), ACC and DAT are both associated with the marking -ko ((13c, 14a)). In the literature on Hindi, -ko has often been referred to as the accusative/dative postposition (e.g. Kachru (1980:27)). Hindi scholars, both traditional and modern, have either explicitly or implicitly assumed that -ko is the ACC marker when the nominal it is attached to is the primary object, and the DAT marker when the nominal it is attached to is, say, the "goal" of a 'give'-type verb (Greaves (1919:314)). The intuition underlying this assumption is essentially the same as that underlying the statements in (11)-(12) about the universal substance of the ACC and DAT features. Now, if a verb is ditransitive, with a goal object and a theme object, the case on the theme object is systematically NOM. The goal object is invariantly associated with the DAT marking -ko by virtue of its meaning. In its presence, the ACC -ko is precluded on inanimate objects even when they are definite, as in (40b) (cf. (33b)): (40) a.

ilaa-ne mãã-ko yah haar diyaa. Ila-E mother-D this- N necklace-N give-PERF Ila gave this necklace to mother.

b.* ilaa-ne mãã-ko Ila-E mother-D

is

haar-ko diyaa. this-N N necklace-A give-PERF

The Case System / 87

In (40a), the canonical association in (35) takes effect as an absolute association. In (40c) below, the canonical association in (36) is prevented from taking effect: even though the object is animate (cf. (32)):38 (40) c.

ilaa-ne mãã-ko baccaa / * bacce-ko Ila-E mother-D child-N child-A Ila gave a/the child to the mother.

diyaa. give-PERF

Thus, when there are potentially two ARGs associated with -ko in a predicate, the DAT -ko precludes the ACC -ko, because the language disfavours identical case marking on more than one nominal. The phenomenon in (40) is described informally by Masica (1982) as follows: "In the case of verbs taking both direct and indirect objects, the latter has priority and precludes the use of -ko with the direct object." (p. 20) Object case in ditransitives in Hindi shows that the association of case with arguments cannot be entirely context-free: the case of the theme object is not only governed by the semantic and syntactic properties of the argument, but also the presence of a goal object in the representation of the predicate. Recall that a direct case feature is one that makes reference to GF information for its association. An indirect case feature is not associated with any grammatical function information. By this criterion, DAT in Hindi is an indirect case. It is associated with goal ARGs obligatorily, regardless of the GF. ACC, on the other hand, is a direct case feature like NOM and ERG, because its association depends on the ARG being an OBJ. The preclusion of ACC in ditransitives then results from the general principle in (41), which I believe to be a recurrent universal: (41)

In the event of conflict, the association of indirect case takes priority over the association of direct case.

38 It must be pointed out that the ungrammaticality of the ACC theme object i n (40b) and (40c) is not the result of the adjacency of two -ko marked nominals. The examples remain ungrammatical even if the two nominals are separated b y other expressions: ilaa-nemaa-ko das din-ke baad baccaa / *bacce-ko diyaa. Ila-E mother-D ten day-G after child-N child-A Ila gave a/the child to the mother after ten days.

give-PERF

The constraint against identical case marking in Hindi is not an isolated instance of such a constraint: for instance, it is not unlike the double -o constraint i n Japanese.

88 / Arguments in Hindi

Recall from Chapter 3 the distinction between underspecified and fully specified lexical representations, employed to factor apart predictable and unpredictable information. We assume that principles of case association take effect whenever the information they specify is available. Now, the GF information referred to by principles of direct case association is available only in the fully specified lexical representations, because it is predictable. Indirect case on the other hand is associated with ARGs without reference to the GFs they are associated with. Thus, the information referred to by principles of indirect case association is available in the underspecified lexical representation. We also assumed that associations once established cannot be eliminated. Then, once indirect case has been associated with an ARG on the basis of information in the underspecified lexical representations, direct case association cannot take effect even if the conditions for the association are satisfied. It follows directly from this organization of the grammar that principles of indirect case association would preempt principles of direct case association. The principle in (41), then does not require independent stipulation, but is a consequence of this organization. We will encounter further effects of (41) in Chapter 7. 4.3.3.2. Why Are Uninflected Objects Not Zero-Marked ACCUSATIVE Why must we distinguish between the two object nominal forms in terms of the case features ACC and NOM, rather than case markings? That is, why cannot we treat the inanimate objects without overt case marking as zeromarked ACC, rather than as NOM? This is a perfectly justifiable question: inanimate objects have, in fact, been assumed to be "unmarked in the accusative" in the literature (Masica (1976:51), Kellogg (1875), Kachru (1980)). If this were so, the situation in Hindi would be another instance of case syncretism found in languages such as Russian and Polish, where certain lexical subclasses of nouns have identical case forms systematically associated with two (or more) distinct case features (Jakobson (1958), Neidle (1982; 1988), Wierzbicka (1980, 1981)).39 In what follows, however, I shall argue that the alternation between the inflected and uninflected objects in Hindi is not an instance of case syncretism, and that primary objects

39 This is similar to the account given for some Australian languages i n Goddard (1982).

The Case System / 89

uninflected for case in Hindi must be treated as NOM, not as ACC with zero marking. First of all, unlike what happens in languages like Russian, it is not that the ACC clitic -ko cannot be attached to inanimate nominals. We saw examples of inanimate objects with -ko when [+definite]. Therefore the choice between zero and -ko in Hindi is governed not by lexical subclasses of nouns but by their syntax and semantics, constituting evidence against a syncretism analysis.40 More crucially, recall that a nominal stem is in its NONNOM form if it is in any case other than NOM or VOC. Consider the following sentences: (42) a.

kaccaa kelaa sastaa hai. unripe-N banana-N inexpensive be-PR Unripe bananas are inexpensive.

b.

ravii-ne kaccaa kelaa kaaÆaa. Ravi-E unripe-N banana-N cut-PERF Ravi cut the / an unripe banana.

c.

ravii-ne kacce kele-ko Ravi-E unripe-NN banana-A cut-PERF Ravi cut the / *an unripe banana.

d.

kacce kele-mã kii±aa hai. unripe-NN banana-L worm-N be-PR There is a worm in the unripe banana.

e.

raam kacce kele-se kyaa banaaegaa? Ram-N unripe-NN banana-I what make-FU What will Ram make with the unripe banana?

kaaÆaa.

Whereas in (42a), the noun kelaa 'banana' is the subject, in (42b, c), it is the object. When the object bears -ko ((42c)), the stem is in the NONNOM form kele, as also in (42d, e). But when the object does not bear a clitic, the

40 A similar situation exists in other South Asian languages like Malayalam. As mentioned earlier, all animate objects in Malayalam are ACC , and inanimate objects NOM. Now, most postpositions in Malayalam demand that their objects be in the ACC form, making no distinction between animate and inanimate nominals. That is, an inanimate nominal bears the ACC marker when it is the object of a postposition. Therefore the areal pattern of overt marking on objects is not statable in terms of the presence or absence of overt morphological marking.

90 / Arguments in Hindi

stem is in the NOM form ((42b)), exactly like the subject in (42a). Now, one might argue that the case stem alternation is dependent not on the case feature, but on whether the noun is inflected or not: a nominal is in the NONNOM case stem form only if overtly marked with a case clitic or postposition. This is not true, however. Locative destinations in Hindi are not overtly marked with a clitic in most dialects. Yet, a nominal that is a locative destination is encoded in its NONNOM form. This is evident in (43). The place name kalkattaa is in the NOM form in (43a). In (43b), even though it is not overtly marked with a clitic, it is in the NONNOM form: (43) a.

k a l k a t t a a bahut duur hai. Calcutta-N very far be-PR Calcutta is very far away.

b.

raam kalkatte Ram-N Calcutta-NN Ram went to Calcutta.

gayaa. go-PERF

That objects without overt marking must be treated as NOM rather than zero-marked ACC is further endorsed by facts of modifier agreement. Modifiers agree with their head nouns in gender, number, and case (see §2.1.). Thus, the noun kelaa 'banana' in all the sentences in (42) is modified by an adjective which must agree in case stem form with the case stem form of the noun. Modifiers of objects not marked with -ko, as in (43b), systematically show NOM agreement, suggesting that the head noun is NOM. When the same inanimate head noun is overtly marked with a NONNOM case ((43c-e)), the modifier exhibits NONNOM case. Once again, it cannot be that the agreement is a purely morphological phenomenon. The case distinction is evident in modifier agreement even when the nominal does not show any surface distinction between NOM and NONNOM stem forms. Recall that differences in case stem form are not always accompanied by corresponding phonological differences. Consider the sentences in (44). The nominal g¥ar 'house' is the subject in (44a), an inanimate object in (44b), and a locative destination in (44c). Their stem forms are identical on the surface, and none of them is overtly marked with a clitic. Nevertheless, the modifier in (44a) and (44b) is NOM, and in (44c), NONNOM: (44)

The Case System / 91

a.

meraa g¥a r gããw-mã hai. I-G-N house village-L My house is in a/the village.

be-PR

b.

raam-ne meraa g¥a r k¥ariidaa. Ram-E I-G-N house buy-PERF Ram bought my house.

c.

raam mere g¥a r aayaa. Ram-N I-G-NN house come-PERF Ram came to my house.

Given the above facts, modifier agreement must be stated on the case feature, not on its morphological manifestation. The difference in the form of the modifiers in (44) can be accounted for by assuming that g¥ar 'house' in (44a, b) is NOM, and in (44c), it is NONNOM. If so, we must also assume that the object in (42b) is NOM. Further evidence comes from verb agreement. As we will see in Chapter 5, verbs in Hindi can agree only with a nominal in NOM case. It agrees with the subject in gender, number and person if the subject is NOM; if not, it agrees with the object if the object is NOM; if neither the subject nor the object is NOM, the verb is in the default (i.e., masculine singular) form. Verbs, as might be expected in the light of the above discussion, do agree with objects not overtly marked with the clitic -ko, as shown by the sample sentences in (45a, c): (45) a.

ilaa-ne kelaa uƥaayaa. Ila-E-F banana-N-M lift-PERF-M Ila picked up the/a banana.

b.

ilaa-ne roÆii-ko uÆ¥aayaa. Ila-E-F bread-A-F lift-PERF-M Ila picked up the bread.

c.

ilaa-ne roÆii uÆ¥aaii. Ila-E-F bread-N-F lift-PERF-F Ila picked up the/a bread.

Finally, coordination of two nominals is possible only if they have identical case. As can be expected, uninflected and inflected objects cannot be conjoined. Thus, whereas (46a) is acceptable, (46b) is ungrammatical.

92 / Arguments in Hindi

(46) a.

raam-ne bacce-ko aur uske juute-ko Ram-E child-A and pron-G shoe-A Ram picked up the child and its shoes.

b. * raam-ne bacce-ko Ram-E child-A and

aur

uskaajuutaa pron-G shoe-N

uƥaayaa. lift-PERF uƥaayaa. lift-PERF

The ungrammaticality of the NOM OBJ in (46b) is not because the verb requires an ACC OBJ. We have seen examples in which the verb uƥaa "pick up" takes NOM OBJs. In sum, the overt form of the object nominal is not determined by any morphological classification, but by syntactic and semantic conditions. Modifier agreement, verb agreement, and coordination show that the distinction between inflected and uninflected objects must be treated as a distinction in syntactically relevant case features, not merely in morphological case marking.41 We must therefore conclude that objects not overtly marked with -ko are NOM, not zero-marked ACC.

4.4. Case Preservation: Direct and Indirect Case In what preceded, we concentrated primarily on the marking -ko on primary objects, and charted its distribution. Now, in the discussion of ditransitives with potentially two -ko marked arguments, we saw that -ko exhibits two different types of behavior. This difference may be accounted for in terms of two case features, ACC and DAT, which syncretize on a single case marking, -ko . Or one might view -ko as being associated with a single case feature, ACC, which when associated with a primary object by virtue of its function behaves in one way, and when associated with a "goal/recipient" by virtue of its meaning behaves in another way. I follow traditional wisdom and adopt the former option. The reasons are both language internal and crosslinguistic. There are phenomena in Hindi that converge on the distinction between what might be neutrally called "primary object case" and "goal case". First, in the presence of both, the latter takes priority. Second, the ACC-NOM 41 In order to account for the absence of the ACC inflection on inanimate objects, one might suggest a case changing rule of the form: ACC ≠ NOM / [animate]. For an accurate account, however, this rule must be made to precede agreement and coordination. Given the facts of Hindi, the status of such a rule i s at best dubious.

The Case System / 93

alternation discussed earlier involves only primary objects, not goals.42 The condition that if inanimate and ACC, the argument must be [+definite] is also specific to primary objects, and does not affect goals. Relating the two sets of behaviour to a single case feature would obscure these differences. One might try relating the two patterns to two different aspects of the same case feature purely in terms of an available distinction, namely, direct and indirect. However, then we would expect all the differences in patterning to fall out from the principled differences between direct and indirect case features. But the effects of animacy and definiteness in the primary object case association are not properties of direct case. Rather, they are crosslinguistic patterns found in primary object case, which is what is traditionally called accusative. Separating ACC and DAT, although associated with the same marking, -ko, simplifies the statement of these generalizations. Crosslinguistic behavior patterns also suggest the separation of ACC and DAT case features. In the Dravidian languages spoken in India, there are two distinct case endings corresponding to the -ko in Hindi. One of them, traditionally called the DAT ending, appears on the "goal/recipient", regardless of the GF of the argument. In other words, DAT case is an indirect case. This is the case that appears on the subject of the widely known "dative subject" or "experiencer subject" construction, which shows a remarkable similarity in behavior across languages not only South Asia, but also in Romance, Slavic, etc. The other, namely the ACC ending, appears on animate primary objects, inanimate primary objects generally being in NOM case. This case feature is associated with information that includes a specific grammatical function, and is thus a direct case. In most of the Indo Aryan languages including Hindi, there is only one case marking corresponding to the two in the Dravidian languages. But the distribution and alternation patterns of this single case marking parallel those of the DAT and ACC cases described above for the Dravidian languages. The syntactic and semantic similarities in spite of the morphological difference between the two sets of languages could not have been an accident. Statement of the crosslinguistic generalizations in South Asian languages about the "dative subject construction", or of the pattern of case on objects, becomes far simpler if we distinguish the case features ACC and

42 "Primary object case", unlike "goal case" must be associated because of the GF. It cannot be equated with "patient/theme case", because a "patient/theme", if it is not the object, cannot get ACC case.

94 / Arguments in Hindi

DAT in the Indo Aryan languages as has been done by the traditional

grammarians. A notable pattern in the distribution of ACC and DAT case that appears in the "passive" construction supports the assumption that ACC is a direct case and DAT an indirect case. This pattern and its implications for a theory of case deserve special attention. In what follows, we will examine some of the differences in patterning between ACC and DAT, with a view to lending more substance to the distinction between direct and indirect case. 4.4.1. Preservation of D A T I V E Consider the sentence in (47a) and its passive version in (47b): (47) a.

raam anil-ko Ram-N Anil-A Ram will carry Anil.

uƥaaegaa. lift/carry-FU

b.

anil (raam-se) uƥaayaa Anil-N Ram-I carry-PERF go-FU Anil will be carried (by Ram).

jaaegaa.

The object nominal of the morpholexically underived verb in (47a) is in ACC case. The same nominal, when the subject of the passive in (47b), is in NOM case. That is, ACC case on the object is not "preserved" in passivization, precisely as expected. ACC case is associated with a specific grammatical function, namely, primary object. Therefore, when a nominal is not the object, it cannot be ACC. Now consider the ditransitive sentence in (48): (48) raam-ne

anil-ko haar b¥ejaa. Ram-E Anil-D necklace-N send-PERF Ram sent a/the necklace to Anil.

When (48) is passivized, either the DAT nominal Anil may be the subject ((49a)), or the NOM nominal haar 'necklace' ((49b)):43 (49) 43 Whichever nominal precedes the other in linear order is the grammatical subject.

The Case System / 95

a.

anil-ko haar b¥ejaa Anil-D necklace-N send-PERF Anil was sent a/the necklace.

gayaa. go-PERF

b.

haar anil-ko b¥ejaa necklace-N Anil-D send-PERF The necklace was sent to Anil.

gayaa. go-PERF

In contrast to the subject of the passive in (47b), which is NOM, Anil in (49a) is DAT even when it is the subject.44 In other words, DAT case is preserved on the nominal, exactly as expected if DAT case is not associated with any particular grammatical function, but with a specific meaning.45 The generalization relating from the type of case association on the one hand and its preservation under GF alternations on the other, may then be stated as: (50)

Indirect case is preserved under

GF

alternations.

(50), like (41), is in fact an automatic consequence of the organization of grammar envisioned here. Indirect case association with an ARG takes place prior to the association of the ARG with its GF. It is therefore not sensitive to the GF of the ARG. Nor can an association once established be altered. The result is the effect of case preservation. 4.4.2. A C C U S A T I V E Preserving Dialects In the earlier example (47b), the passive of (47a), the subject nominal is NOM. There exist speakers of Hindi who find (47b) unacceptable.46 The passive version of (47a) accepted by these speakers is given in (51) below: it

44 Diagnostics for subjecthood by which a nominal can be shown to be a subject are given in Chapter 6. 45 The idea of capturing this effect of case preservation under changes of grammatical function by relating these cases directly to argument meanings, i n contrast to cases that are not preserved, namely, the cases that are associated with grammatical functions, is found in Levin and Simpson (1981), Marantz (1981), Zaenen and Maling (1984) and others. In all these works, however, "quirky" case is treated as lexically stipulated idiosyncratic case, rather than as semantically regular case. 46 Yamuna Kachru (personal communication).

96 / Arguments in Hindi

apparently has exactly what is impossible in the dialects represented by (47b), namely, an ACC subject:47 (51) anil-ko

(raam-se) uƥaayaa jaaegaa. Anil-A Ram-I carry-PERF go-FU Anil will be carried (by Ram).

The generalization that ACC case is associated with the object function does not therefore hold true for these speakers: a nominal realized as the primary object of an active sentence is in ACC case even when it is realized as the subject in a passive construction.48 However, the distribution of the preserved ACC case obeys exactly the same conditions as in the other dialects: inanimate objects are not ACC, but NOM. Could ACC case in these dialects then be determined by the meaning, like DAT case? Let us consider the facts. On the one hand, ACC case appears on object arguments not necessarily associated with a single meaning. The object's case is preserved in the ACC preserving dialects regardless of argument meaning. However, nominals that are subjects of clauses without passive morphology, but have the same meanings, cannot be ACC. Consider the following sentences which illustrate the different argument meanings that could be associated with ACC case: (52) raam-ne

anil-ko Ram-E Anil-A Ram beat Anil.

piiÆaa. beat-PERF

(53) raam-ne

anil-ko giraayaa. Ram-E Anil-A fall-CAUS-PERF Ram dropped/caused-to-fall Anil.

(54) raam-ne

anil-ko

dek¥aa.

47 These dialects appear similar in their ACC preserving property to Marathi, another Indo Aryan language (Joshi (1987)). Lexically stipulated ACC case i s also preserved in Icelandic (Andrews (1982)). 48 Speakers of the dialects represented in (51) generally disallow the overt expression of the agent phrase in the passive. That the ACC nominal is the grammatical subject for these speakers seems evident from the behaviour of the nominal with respect to subjecthood tests involving reflexive binding and pronoun-antecedent relations (see Chapter 6). I shall therefore assume tentatively that these ACC nominals are indeed grammatical subjects. However, this dialect requires further investigation for conclusive evidence.

The Case System / 97

Ram-E Anil-A Ram saw Anil.

see-PERF

(55) raam-ne

anil-ko k¥uß Ram-E Anil-A happy Ram made Anil happy.

kiyaa. do-PERF

It would be generally accepted that the ACC nominal Anil in (52) is "affected" by the action without being an undergoer of a change of state or location. In (53) it is both affected and an undergoer of change of location. In (54), it is a "neutral" participant: neither affected nor an undergoer of change. In (55), it is an experiencer, and undergoer of change of state.49 The "passives" of (52)-(55) in the ACC preserving dialects are given in (56)-(59): 50 (56) anil-ko piiÆaa Anil-A beat-PERF Anil was beaten.

gayaa. go-PERF

(57) anil-ko

giraayaa gayaa. Anil-A fall-CAUS-PERF go-PERF Anil was dropped/caused-to-fall.

(58) anil-ko

dek¥aa Anil-A see-PERF Anil was seen.

gayaa. go-PERF

(59) anil-ko

k¥uß kiyaa gayaa. Anil-A happy do-PERF go-PERF Anil was made happy.

Now let us look at nominals that bear the same argument meanings as those in (52)-(55), but are subjects of non-passives: (60) anil piÆaa. Anil-N be beaten-PERF 49 The predicate of (55) is a complex predicate, k¥uß kiyaa "made happy". We will be concerned with complex predicates in Hindi in Chapter 8. 50 Three distinct constructions, discussed in Chapter 7, co-occur with what i s seen as "passive morphology" in Hindi. Regardless of which of these the ones i n (56)-(59) are, the generalization remains that in the ACC preserving dialects, ACC case is exceptionlessly preserved under passivization.

98 / Arguments in Hindi

Anil was beaten.

(61) anil

giraa. Anil-N fall-PERF Anil fell.

(62) anil k¥uß

huaa. Anil-N happy become-PERF Anil became happy.

(60) is what one one might view as the inchoative of (52). (61) is the uncausativized version of (53). (62) is the intransitive counterpart of (55). Thus, the NOM nominals in (60), (61), and (62) bear the same argument meanings as their ACC counterparts in (52), (53), and (55) respectively. Now, the one property shared by all the ACC arguments in (52)-(59) is that they are LOGICAL OBJECTS, defined in Chapter 3 as the entity towards which an action is directed by an independent inceptor. This is also the property that distinguishes them from the NOM arguments in (60)-(62), which otherwise have the same meanings as the former set. The difference between the two types of dialects exhibits itself only in the passive construction. In the passive, the logical object is the grammatical subject. It is ACC in the ACC preserving dialect and NOM in the other dialect. In short, in the ACC preserving dialects, ACC case is associated with the logical object. It does not require GF information for its association, and therefore is an indirect case feature. In contrast, in the dialect that is primary for our purposes, ACC case is associated with the grammatical object. Once again, we see here an instance of the general statement in (50): indirect ACC case is preserved under passivization. 4.4.3. Objects of "Unaccusative Transitives" The grammatical objects we have encountered so far may be in NOM or ACC case, depending on their animacy and definiteness. Now, there exist grammatical objects in Hindi that cannot be ACC even when these conditions are satisfied. Such objects are illustrated in (63a, b), which are examples of the "dative subject" construction, to be discussed at length in §7.1.51 51 Evidence for the objecthood of the NOM nominal in (63a) and (63b) comes from the facts of gapping in Hindi. As we will see at length in Chapter 6, in order to be gapped, an element must be identical to the gapper in both grammatical

The Case System / 99

(63) a.

ilaa-ko anu dik¥ii. Ila-D Anu-N appear-PERF Ila saw Anu. (Lit.: To Ila appeared Anu.)

b.

raam-ko ravii milaa. Ram-D Ravi-N encounter-PERF Ram met Ravi unexpectedly.

The object nominals in (63a) and (63b) are animate, even human. We would therefore expect them to be ACC. Yet they are invariantly NOM in all dialects. In order to account for this, one might simply stipulate that if the subject is DAT, the object must be NOM. However, this is incorrect. Adding the modality meaning of obligation, for instance, to a predicate that otherwise takes an ERG subject, can induce DAT case on that subject. In such an event, the subject is DAT and the object ACC. The prohibition is therefore not an absolute one. Furthermore, such a stipulation would miss a deeper generalization. We said earlier that in the ACC preserving dialects, ACC case is associated with the logical object. Recall that related to the notion of logical object is the semantic property of being the entity toward which a logically transitive action is directed. The grammatical objects in (63) do not have this property. The objects of dative subject constructions cannot be passivized either. In other words, they are not logical objects. We might then call the construction illustrated in (63) the "unaccusative transitive" construction; that is, one that has a grammatical object but not a logical object. It is transitive because it is diadic and has a grammatical object. It is unaccusative because its subject is "nonagentive",

function and case. The NOM argument of the verbs in (56) can both gap as well as be gapped by uncontroversial grammatical objects, as shown by (i) and (ii) respectively: (i) ilaa-kok¥at milaa aur anu-ne ——— pa±¥aa. Ila-D l etter-N get-PERF and Ila got a letteri, and Anu read ——— i.

(ii)

Anu-E

raam-ne aam mez-par rak¥a a aur Ram-E mango-N table-L put-PERF and Ram put the mangoi on the table, and Anu saw ——— i.

read-PERF

anu-ko ——— dik¥aa. Anu-D appear-PERF

Given that they can both gap and be gapped by grammatical objects, the NOM nominals in (56) must be grammatical objects. Evidence for the nonsubjecthood of the NOM nominal comes from facts of reflexive binding, pronouns, and control.

100 / Arguments in Hindi

and it has no logical object. It naturally follows that the object in this construction cannot be ACC in the ACC preserving dialects. In the dialect that is the primary focus of this thesis, we said that ACC case is associated with animate grammatical objects. The association was formally stated in (36). In order to account for the lack of ACC on the grammatical objects in (63) in the dialects that do not preserve ACC, we must then assume that ACC is associated with only those grammatical objects that are also logical objects. In other words, in order to receive ACC case, an argument must be the logical object in all the dialects. In this dialect, it must also be the grammatical object. In this light, the association in (36) must be reformulated as in (64), with the ARG specified as being the logical object ( LO): (64) ACC Association: Hindi 52 [+animate]

SEM STR

ARG LO

ARG STR

\ OBJ

GF STR

ACC

The NOM OBJs of the unaccusative transitive sentences will automatically receive NOM case by the principle in (38), which associates the default NOM case with ARGs bearing UNRESTRICTED GFs.

4.5. Summary To summarize, we distinguished CASE FEATURES, which are universal entities and demand a crosslinguistic characterization, from the language specific CASE MARKINGS which signal them. Within the conception of linguistic structure in this work, the universal case features are part of the 52 In the ACC preserving dialects, object case association would instead be stated as follows: [+animate] ARG LO

ARG STR

ACC

GF STR

The Case System / 101

subsystem of grammatical features represented at GRAMMATICAL FUNCTION STRUCTURE in syntax, while the case markings associated with them belong to the domain of morphology. This chapter was primarily concerned with the association of case features in Hindi, which are a subset of the universal inventory, with meanings and grammatical functions, even though the association of case features with case markings was also addressed. The principles of ERG, GEN, NOM, and ACC case association formulated in this chapter require GF information. Observe that the GFs that they referred to are subjects and primary objects, which are UNRESTRICTED functions (§3.4.1). Exploiting this observation, we may entertain the following universal condition: (65)

Principles of case association cannot refer to the class of FUNCTIONS.

RESTRICTED

A consequence of (65) would be that principles of case association may refer to subjects and objects, but not obliques, or adjuncts. Whether or not this condition is an absolute universal or has marginal exceptions must be determined on crosslinguistic grounds. We characterized ERG, GEN, NOM and ACC in Hindi as DIRECT CASE, as opposed to DAT, LOC, and so on, which are instances of INDIRECT CASE. Direct case is that which makes reference to GF information for its association to ARGs, whether or not it also requires reference to meaning. This does not yield a classification of case features. Rather, it is a classification of the principles of case association in a language. It does not rule out, for instance, a particular case feature participating in the grammar as both direct case and indirect case. We will see in Chapter 7 that the GEN case in Hindi exhibits precisely this duality. An immediate consequence of distinguishing between direct and indirect case is found in CASE PRESERVATION patterns in Hindi. A case that refers to GF information cannot be preserved on an ARG under an ARG-GF realignment. In contrast, a case that does not refer to GF STR information must be preserved. It follows therefore that indirect case is preserved under GF alternations ((50)). The notion L-OBJ allows for a simple analysis of a dialect difference in the preservation of ACC case in Hindi. In the ACC preserving dialect, ACC case is associated with the L-OBJ, while in the non-ACC preseving dialect, it must be associated with the L-OBJ at ARG STR and OBJ at GF STR (§4.4.2.).

102 / Arguments in Hindi

This analysis also accounts for why the OBJ in the "dative subject" construction must be NOM even if animate: it cannot be ACC because it is not an L-OBJ (§4.4.3.). The account of case in this chapter crucially depends on the concept of conflict resolution through the preempting of one principle by another, governed by a general universal principle and the organization of grammar. Thus, ERG and GEN case associations preempt NOM case association on subjects. ACC case association preempts NOM case association on objects. Both these instances of preemption follow from the Elsewhere Condition: the more specific principle preempts the general principle. The constraint against the appearance of two -ko marked nominals prohibits the occurrence of ACC and DAT together in ditransitives. In such an event, the conflict is resolved by the DAT preempting the ACC: the primary object gets the default NOM case even if animate. This effect follows from the general organization that derives the statement in (41). Finally, the analysis of direct case in Hindi brings into focus one of our central assumptions in the organization of grammar: given that case may be constrained by information at SEM STR, ARG STR, and GF STR, the various levels of structure cannot be strictly ordered with respect to one another so as to prevent grammatical principles from accessing information concurrently from more than one level. Rather, they are co-present structures that can simultaneously constrain syntactic phenomena. The account of case in Hindi given in this chapter conflicts with a number of assumptions in GB theory. In order to account for the ERG-NOM alternation on agentive subjects, depending upon verb morphology, we assumed that (a) NOM is the default case, (b) ERG case is associated with the SUBJ under certain semantic conditions. In order to provide an equivalent analysis in GB, it is necessary to acknowledge two types of structural case, those that are governed by meaning (ERG) and those that are not (NOM/GEN). Needless to say, this calls for abandoning the widely held assumption that syntax cannot access lexical semantics other than what is indirectly reflected in argument structure in terms of external argument, direct internal argument, and the like. The recognition of two types of structural case is necessary for objects as well. The nominal meaning of animacy conditions the choice between ACC and NOM on the object. The idea that NOM case is the default for both subjects and objects runs counter to the very spirit of case theory in GB, which is built on the structural dichotomy between subjects and objects. The default association of NOM case with ARGs bearing UNRESTRICTED GFs ((38)) unifies NOM

The Case System / 103

case association of various types: (a) NOM on SUBJ, (b) NOM on inanimate PR.OBJ, (c) NOM on inanimate or animate PR . OBJ which is not an L-OBJ, and (d) NOM on animate or inanimate PR.OBJ in ditransitives, which is also an L-OBJ. As far as I can see, GB case theory does not lend itself to an equally simple account. A similar problem appears in the analysis of ACC case preservation in passives in the ACC-preserving dialects. Under our analysis, this is an instance of the preservation of indirect case. Indirect case is not equivalent to inherent case, as the latter, unlike the former, is idiosyncratically stipulated in the lexical entry of the verb. ACC case on objects is not an inherent case, as it is entirely predictable, and not governed by theta roles. Therefore, the preservation of ACC case in the ACC-preserving dialects of Hindi calls for the replacement of inherent case with indirect case as defined in this chapter.

5 Verb Agreement The purpose of this brief chapter is two-fold. First, I briefly outline the facts of verb agreement in Hindi, as part of the background for the discussion of grammatical subjecthood in the next chapter. I argue that verb agreement in Hindi is controlled by the highest argument in ARG STR associated with nominative case. It is not controlled exclusively by grammatical subjects. Within the conception outlined in Chapter 3, then, agreement is stated in terms of the GF STR information associated with an ARG. Second, this chapter calls attention to a puzzling phenomenon that has to do with the dual nature of the argument that a verb agrees with. Hindi has a noun incorporation construction, in which an argument of the verb shows semantic as well as syntactic characteristics of having been lexically incorporated into the verb. Even when "incorporated", however, the verb may agree with this noun. In other words, the verb may agree with a verb-internal element. Under the assumption that what a verb agrees with is one of its arguments, this appears to be a baffling situation. I present a solution to this apparent conflict in the representation of incorporated noun in terms of a difference between the internal organization of ARG STR and GF STR on the one hand and that of GC STR on the other.

5.1. Verb Agreement: The Simple Facts Recall from §2.2. that a verb in Hindi agrees in number, gender, and person with its NOM SUBJ.1 Consider the agreement in (1):2

1 Here, and in the rest of the discussion, I ignore person agreement. 2 The convention adopted in the glosses is as follows: when a form uniquely expresses one member in a contrasting pair, the member is included in the gloss. Thus, the specification of number is omitted from the gloss if the singular and plural forms of a noun are identical on the surface. But number is indicated in the gloss if the forms are distinct. This applies to the specification of the NOMNONNOM case distinction as well. The specification of person, which is third person in all examples, is systematically left out.

103

104 / Arguments in Hindi

(1) a.

ravii baalak-ko uƥaaegaa. Ravi-N(M) boy-A(M) lift-FU .M .SG Ravi will lift up the boy.

b.

niinaa baalak-ko uƥaaegii. Nina-N(F) boy-A(M) lift-FU .F .S G Nina will lift up the boy.

The verb in (1a) and (1b) agrees with the NOM SUBJ. If the SUBJ is NONNOM, however, and there is no NOM argument in the sentence, the verb is in the default or non-agreeing form; it bears the third person singular masculine inflection: (2) a.

ravii-ne baalak-ko Ravi-E(M) boy-A(M) Ravi lifted up the boy.

uƥaayaa. lift-PERF.M .SG

b.

niinaa-ne baalak-ko uƥaayaa. Nina-E(F) boy-A(M) lift-PERF.M .SG Nina lifted up the boy.

c.

ravii-ne baalikaa-ko Ravi-E(M) girl-A(F) Ravi lifted up the girl.

uƥaayaa. lift-PERF.M .SG

d.

niinaa-ne baalikaa-ko Nina-E(F) girl-A(F) Nina lifted up the girl.

uƥaayaa. lift-PERF.M .SG

The verb does not agree with an ARG associated with ERG case, as shown by (2). We have assumed that an ERG argument must be associated with the SUBJ function (§4.3.1). We will see further evidence for its subjecthood in Chapter 6. Under this analysis, an argument may be the SUBJ and still not be the controller of verb agreement. In (2), the SUBJ is NONNOM, and so is the PR(IMARY) OBJ. In the absence of a NOM SUBJ, if the PR.OBJ happens to be NOM, the verb agrees with it, as in (3)-(4):

Verb Agreement / 105

(3) a.

ravii-ne roÆii k¥aayii. Ravi-E(M) bread-N(F) eat-PERF.F.SG Ravi ate bread.

b.

niinaa-ne roÆii k¥aayii. Nina-E(F) bread-N(F) eat-PERF.F.SG Nina ate bread.

c.

ravii-ne kelaa k¥aayaa. Ravi-E(M) banana-N(M) eat-PERF.M.SG Ravi ate a banana.

d.

niinaa-ne kelaa k¥aayaa. Nina-E(F) banana-N(M) eat-PERF.M.SG Nina ate a banana.

a.

ravii-ne niinaa-ko Ravi-E(M) Nina-A(F) Ravi fed Nina a banana.

kelaa k¥ilaayaa. banana-N(M) eat-C-PERF.M.SG

b.

ravii-ne niinaa-ko Ravi-E(M) Nina-A(F) Ravi fed Nina bread.

roÆii k¥ilaayii. bread-N(F) eat-C-PERF.F.SG

c.

niinaa-ko Nina-D(F)

(4)

(ravii-se) kelaa Ravi-I(M) banana-N(M)

diyaa gayaa. give-PERF.M go-

PERF.M .SG

Nina was given a banana (by Ravi). d.

niinaa-ko (ravii-se) roÆii dii Nina-D(F) Ravi-I(M) bread-N(F) give-PERF.F Nina was given bread (by Ravi).

gayii. go-PERF.F.SG

If both the SUBJ and the OBJ are NOM, then the verb agrees with the SUBJ, as in (5): (5) a.

ravii roÆii k¥aaegaa. Ravi-N(M) bread-N(F) eat-FU.M .SG Ravi will eat bread.

b.

niinaa roÆii k¥aaegii. Nina-N(F) bread-N(F) eat-FU.F.SG Nina will eat bread.

106 / Arguments in Hindi

c.

ravii kelaa k¥aaegaa. Ravi-N(M) banana-N(M) eat-FU.SG.M Ravi will eat a banana.

d.

niinaa kelaa k¥aaegii. Nina-N(F) banana-N(M) eat-FU.SG.F Nina will eat a banana.

The generalization about verb agreement in Hindi then appears to be the following. The verb agrees with the NOM SUBJ. If the SUBJ is NONNOM, the verb agrees with the NOM OBJ. If the OBJ is also NONNOM, the verb is in the neutral form, namely, masculine third person singular. This generalization, found in the literature (e.g. Kachru, Kachru, and Bhatia (1976:86)), adequately describes the facts of Hindi verb agreement, and might be captured by the following condition: The controller of verb agreement must be the highest available NOM UNRESTRICTED function. However, an ARG associated with NOM case in Hindi is either the SUBJ or OBJ. The NOM case on SUBJs and OBJs is accounted for by the principle of NOM case association in §4:(38). Now, given that verb agreement requires the ARG to be in NOM case, and that only SUBJs and OBJs can be in NOM case, it is redundant to stipulate that a verb can only agree with a SUBJ or a OBJ. An alternative characterization of verb agreement is the following: (6)

Verb Agreement in Hindi The verb agrees with the highest ARG associated with NOM case.

Gair and Wali (1989) (G&W), in a GB analysis of Hindi agreement in terms of the c-command relation, assume that nominals marked with clitics and postpositions are PPs rather than NPs. In their analysis, a NONNOM subject or object is a PP . Consequently, the c-command relations in structures with NP subjects and objects on the one hand, and PP subjects and objects on the other, are distinct. Nominative subjects and objects c-command the verb, unlike non-nominative subject and objects. G&W crucially rely on this distinction to account for verb agreement in Hindi: a verb agrees only with an NP, not a PP.

Verb Agreement / 107

However, in order to account for the facts of anaphora (reflexives and pronouns)3 within the GB theory, one is forced to assume that uninflected subjects (NPs in G&W) and subjects marked with the clitics and postpositions (PPs in G&W) have identical c-command relations, contradicting the earlier assumption. A similar contradiction appears in the application of the case filter in passivization. Since the case filter applies only to NPs and not to PPs, one must assume that the accusative object in Hindi, which in G&W’s analysis is a PP, is in fact an NP. Only then does it obligatorily become the subject under passivization. The only motivation for the PP analysis, then, is the view that agreement is exclusively governed by c-command, not by case. As G&W themselves note, however, a c-commanding locative NP (the uninflected locative destination mentioned in Chapter 4) does not agree with the verb. In order to account for this fact, G&W rule out verb agreement with non-nominative case in Hindi. Once the agreement principle is allowed to refer to a case feature, as G&W ultimately are forced to do, the inconsistencies observed above may be avoided by the simple assumption that all the case marked nominals are NPs, and that verbs cannot agree with NONNOM NPs.

5.2. Agreement and an Incorporation Construction Hindi has a construction in which an argument of a predicate appears incorporated into the predicate, similar but not identical to the incorporation constructions discussed in Mithun (1984), Sadock (1985), Baker (1988), and the references therein. This construction in Hindi poses an interesting question when juxtaposed with the condition in (6). 5.2.1. The Incorporation Construction: Syntactic Properties Consider the sentences in (7). Each of these sentences has a second reading, indicated in the sentence glosses. In this reading, which I refer to as the "incorporated" reading, the incorporated noun must obligatorily have a generic interpretation: 4, 5 3 See Chapter 6 (§6.2.1. and §6.2.2.) for details of reflexive binding and pronominal coreference in Hindi. 4 This is unlike languages such as Mohawk (Mithun (1984)), Nahuatl (Merlan (1976)), or Southern Tiwa (Allen, Gardiner, and Frantz (1984)), where an incorporated noun is not necessarily accompanied by genericity or indefiniteness.

108 / Arguments in Hindi

(7) a.

anil kitaabã Anil-N(M) book-N-PL(F) Anil will sell books. OR

becegaa. sell-FU .M .SG Anil will do book-selling.

b.

anil-ne kitaabã becññ. Anil-E(M) book-N-PL(F) sell-PERF.PL.M Anil sold books. OR Anil did book-selling.

c.

anil-ne k¥aanaa pakaayaa. Anil-E(M) food-N(M) cook-PERF.M Anil cooked food. OR Anil did food-cooking.

d.

raam-ne lak±ii kaaÆii. Ram-E(M) wood-N(F) cut-PERF.F Ram cut wood. OR Ram did wood-cutting.

In addition to the genericity requirement, the incorporated reading is possible only under severely restricted syntactic conditions. First of all, an incorporated noun cannot take a modifier. Thus, (8) cannot have the incorporated reading:

(8)

anil puraanii kitaabã becegaa. Anil-N(M) old-F book-N-PL(F) sell-FU .M Anil will sell a/the old books. (*Anil will do the selling of old books.)

Second, nothing may intervene between the noun and the verb. The intervening subject in (9a), the locative adjunct in (9b), and the negative in (9c) prevent the incorporated reading: (9) a.

kitaabã anil becegaa. book-N-PL(M) Anil-N(M) sell-FU .M Anil will sell the books. (*Anil will do book-selling.)

b.

anil kitaabã baazaar-mãbecegaa. Anil-N(M) book-N-PL(F) market-L sell-FU .M Anil will sell (the) books in the market. (*Anil will do book-selling in the market.)

c.

anil

kitaabã

nahññ

becegaa.

5 The different interpretations have corresponding differences in word stress and word melody patterns. We will return shortly to these correspondences.

Verb Agreement / 109

Anil-N(M) book-N-PL(M) not sell-FU .M Anil will not sell (the) books. (*Anil will not do book-selling.)

110 / Arguments in Hindi

Third, an incorporated object cannot be the subject of a passive:

(10)

anil-se kitaabã becii jaaãgññ. Anil-I book-N-PL sell-PERF go-FU-PL The books will be sold by Anil. (*Book-selling will be done by Anil.)

An incorporated noun cannot gap or be gapped in a coordination construction. Thus, (11) cannot have the incorporated reading either:

(11)

anil g¥o±e k¥ariidtaa hai aur raam bectaa hai. Anil-N horses-N buy-IMPERF be-PRES and Ram-N sell-IMPERF be-PR Anil buys and Ram sells horses. (*Anil does horse-buying and Ram does ——— -selling.)

Nor can either the incorporated noun or the verb be conjoined, as shown by the fact that (12a) and (12b) cannot have the incorporated reading: (12) a.

anil haat¥ii aur g¥o±e bectaa hai. Anil-N elephants-N and horses-N sell-IMPERF be-PR Anil sells elephants and horses. (*Anil does elephant- and horse-selling.)

b.

anil g¥o±e k¥ariidtaa aur bectaa Anil-N horses-N buy-IMPERF and sell-IMPERF Anil buys and sells horses. (*Anil does horse-buying and -selling.)

hai. be-PR

Finally, if a noun is inflected for case, it is not incorporated. Thus, exactly like (8)-(12), (13) cannot have the incorporated reading:

(13)

anil kitaabõ-ko Anil-N(M) book-A-PL(F) Anil will sell the books.

becegaa. sell-FU .M (*Anil will do the selling of the books.)

To summarize the conditions required for the incorporated reading, then, (i) the incorporated noun must have generic interpretation; (ii) there can be no material intervening between the noun and the verb; (iii) neither the noun nor the verb can be conjoined; (iv) the noun cannot be gapped; (v) it cannot be modified; (vi) it cannot be case marked; and (vii) it cannot be a subject.

Verb Agreement / 111

The incorporated nouns we have seen are all inanimate objects. Do animate objects, and non-objects get incorporated as well? First take animate objects, which are associated with ACC case (§4:(36)). Consider (14): (14) a.

ilaa bacce k¥ojtii rahtii hai. Ila-N children-N search-IMPERF PROG be-PR Ila keeps performing the act of searching for children. ( * Ila keeps searching for the/some children.)

b.

ilaa baccõ-ko k¥ojtii rahtii hai. Ila-N children-A search-IMPERF PROG be-PR Ila keeps searching for the/some children. ( * Ila keeps performing the act of searching for children.)

When an animate object is NOM, the incorporated reading is obligatory ((14a)). When it is ACC, on the other hand, the incorporated reading is impossible ((14b)).6 Now, one might translate (14a) as: "Ila keeps searching for children," and regard the effect of NOM case as forcing a [+generic] reading, rather than an incorporated reading. However, this could not be, because, whenever an animate object is NOM, the construction exhibits all the syntactic properties listed above:7 the object cannot take case marking ((14b)), modification ((14c)), conjoining ((14d, e)), or gapping ((14f)), and does not allow anything to intervene between itself and the verb ((14g)). (14) c. * ilaa Ila-N

c¥oÆe bacce k¥ojtii smallchildren-N search-IMPERF

d. * ilaa Ila-N

la±keaur boys- N

e. * ilaa Ila-N

bacce k¥otii aur k¥ojtii rahtii hai. children-N lose-IMPERF and search-IMPERF PROG be-PR

f. * ilaa Ila-N

c¥oÆe bacce k¥ojtii smallchildren-N search-IMPERF

PROG

rahtii hai. be-PR

la±kiyãã k¥ojtii rahtii hai. and girls-N search-IMPERF PROG be-PR

PROG

rahtii hai. be-PR

6 There are dialects of Hindi that allow animate objects to be NOM even when not incorporated. See, for instance, Verma (1971:104). 7 Incorporated objects in Turkish exhibit the same properties as the ones mentioned here for Hindi (Knecht (1986:99)).

112 / Arguments in Hindi

g. * ilaa Ila-N

bacce hameßaa children-N always

k¥ojtii search-IMPERF

rahtii hai. PROG be-PR

The animate objects in (14c-g) cannot be incorporated, because of the modification, conjoining, gapping, and the material intervening between the noun and the verb. And if not incorporated, the object must be ACC, not NOM. Hence the ungrammaticality of these sentences. One of the restrictions on incorporation, as we have seen, is that an incorporated noun cannot be case-marked. Now, other than nominals in NOM case, the only arguments that do not have any morphological marking to indicate case are locative destinations (see Chapter 4), such as in (15): (15) a.

raam-ne apne bete-ko ßahar b¥ejaa. Ram-E self-G son-A city-L send-PERF Ram sent his son to the city.

b.

baccaa skuul gayaa. child-N school-L go-PERF The child went to school.

The sentences in (15), like those in (7), can have an incorporated reading. In other words, the LOC ARGs in (15) can be incorporated as well: like the objects we have discussed, they exhibit all the properties listed as being peculiar to incorporated nouns. While I assume that locatives without morphological marking may be incorporated, I will not give a detailed demonstration of the incorporation of locatives.8 5.2.2. An Analysis: A Noun+Verb Lexical Compound A possible analysis of the incorporation discussed above is that the noun forms a lexical compound with the predicate. As we have seen, there are behavioral properties that correlate with the two readings, say, of (7b). What is the structural difference that allows for this systematic correlation? The correlation might be captured in terms of differences in category and con8 Mithun (1984), Allen et al (1984) and others suggest that noun incorporation exhibits a split in the Thematic Hierarchy, in that while patients, themes, locatives, and instrumentals can undergo incorporation, the incorporation of agents, beneficiaries and recipients is crosslinguistically ruled out. While incorporation in Hindi does not provide additional evidence for this claim, the facts of incorporation in Hindi are perfectly consistent with it.

Verb Agreement / 113

stituency (that is, in GC STR), and represented as in (16) and (17). (16) provides in simplified form the GC STR for (7b) with the regular reading, and (17) for (7b) with the incorporated reading: (16)

S N N V GC STR ………………………………………………………………………… anil-ne kitaabã becññ. WORD

STRING

(17)

S N

V V N

V

GC STR

………………………………………………………………………… anil-ne kitaabã becññ. WORD STRING

The incorporated noun in (17), the verb it forms a constituent with, and their mother node, are all assumed to be lexical (X0) categories.9 Recall from Chapter 4 the assumption that case clitics in Hindi are concatenated with the nominals not in the lexical but in the phrasal module. The restriction that incorporated nouns cannot have case morphology then follows from the assumption that in the incorporation construction, the incorporated noun and the verb together form a lexically concatenated compound, as represented in (17), and that phrasal material cannot intervene between them. A lexical category cannot be syntactically modified. Nor can it be conjoined (Simpson (1983)), or gapped (Simpson (1983), Booij (1985)). The representation in (17) then explains why the incorporated noun in Hindi cannot be modified or gapped, and why neither the noun nor the verb can be 9 In a theory of morphology, for instance, that in Selkirk (1982), X0 represents a maximal morphological structure, or morphological word. Smaller morphological constituents are represented with progressively larger negative bar levels. This is not important for our purposes.

114 / Arguments in Hindi

conjoined. It also explains why nothing can intervene between the noun and the verb: for a syntactic entity like a subject or an adjunct to intervene between the noun and the verb, their mother node must be a non-lexical category.10 The failure of an incorporated object to be the subject of a passive follows directly from the following generalization concerning verbal compounds: "The SUBJ argument of a lexical item may not be satisfied in a compound structure." (Selkirk (1982:34)) Thus, a compounded noun cannot be interpreted as the subject argument of the verb.11 In short, the syntactic peculiarities of the incorporation construction are a natural consequence under the assumption that the incorporated noun forms a lexical compound with the verb. I now turn to some phonological and morphological evidence that supports this assumption. For ease of reference, I will refer to the incorporated noun and the verb together as the N+V compound. 5.2.3. Phonological and Morphological Evidence 5.2.3.1. Similar to what has been observed in Malayalam (Mohanan (1982)), a sub(ordinate) compound of the form [head+modifier] in Hindi has a single primary stress and word melody, as opposed to a co(ordinate) compound of the form [head+head+ ..], which has as many primary stresses and word melodies as it has stems. Consider the contrast in the two compounds in (18): (18) a.

jáanakiinandan Janaki's son

b.

jáanakiinándan Janaki and Nandan

Within a level ordered conception of lexical phonology and morphology (Mohanan (1982), Kiparsky (1982), and others), the contrast in (18) can be accounted for by assuming that (i) subcompounding and cocompounding

10 This also explains why the construction cannot have the incorporated reading if the noun by itself or the verb by itself scrambles or topicalizes. However, the two together can scramble or topicalize, retaining the incorporated reading. 11 The insight underlying this condition on compounds was first formulated as the first sister principle in Roeper and Seigel (1978:208).

Verb Agreement / 115

constitute different lexical strata/levels, (ii) the former is an input to the latter, and (iii) stress and word melody assignment take place at the end of the subcompounding stratum. Now, the N+V compounds behave exactly like subcompounds in that they have a single stress and word melody shared by the noun and the verb. (19a) is the incorporated version: (19) a. g¥ó±e horse-PL to horse-sell

becnaa sell-NF

g¥ó±e horse-PL to sell horses

bécnaa sell-NF

b.

c. safed g¥ó±e white horse-PL to sell white horses

bécnaa sell-NF

d. g¥ó±õ-ko bécnaa horse-PL-A sell-NF to sell the horses

The contrast in stress and word melody in (19a) on the one hand, and (19b-d) on the other, follows if we assume that the N+V compound in (19a) is formed in the same lexical stratum as the nominal subcompound in (18a). Another piece of evidence for the lexical nature of the N+V compound comes from an optional phonological process of morpheme final high vowel shortening. For instance, the final ii of the word jaanakii in (18a) may be optionally shortened, whereas the same vowel in (18b) cannot undergo shortening. Now consider the forms of the nouns in (20): (20) a.

(i)

miƥaaii banaaii.

(ii) miƥaai

banaaii

116 / Arguments in Hindi

sweets-N make-PERF made sweets.

sweets-N

make-PERF

Verb Agreement / 117

b.

(i)

sabzii becii. (ii) sabzi becii vegetables-N sell-PERF vegetables-N sell-PERF sold vegetables.

While not all speakers find the vowel shortening acceptable, for those speakers who do shorten the vowel, the process is regular: when the vowel is shortened, the noun and the verb must have a single primary stress and word melody, and must also obey all the syntactic conditions laid out earlier. These facts follow from the assumption that vowel shortening is a rule that applies in the lexicon, in a module prior to co-compounding. Now, one might say that the noun and the verb are not concatenated lexically, but form a single phonological word, and that the domain of vowel shortening is the phonological word. However, a noun along with a case clitic forms a single phonological word: miÆ¥aaii-mã "in the sweets"; sabzii-se "with the vegetables". Yet, the vowel cannot shorten in these forms. Therefore the domain of the final high vowel shortening in (20) cannot be captured in terms of the phonological word, but must be accounted for in terms of a lexical submodule. 5.2.3.2. We come finally to the morphological evidence. The N+V compound can undergo derivational affixation. The suffix -vaalaa , which may be thought of as an agentive marker, can be attached to this compound as illustrated in (21): (21) g¥ó±e

bécne vaalaa

horse-PL

sell-NF

horse-seller

The sequence of N+V-vaalaa in (21) takes a single primary stress and word melody. The object nominal in (21) cannot be modified or conjoined, take case morphology, or be separated from the verb by, say, an intervening adverbial. It must also be mentioned that (21) cannot be analysed as [g¥o±e ] [becne vaalaa ]. While g¥o±e vaalaa "one who keeps/sells/employs horses" is a possible form, becne vaalaa "seller" as an independent compound is not. How does this show that the N+V compound is a lexical compound? In order to answer this question, we must examine further the suffix -vaalaa. The form vaalaa has at least three systematic uses in Hindi, other than its use in (21). First, it concatenates with nouns and functions as what may be called an occupation marker. This is illustrated in (22):

118 / Arguments in Hindi

(22) Vaalaa as an occupation-marker12 a.

g¥o±aa horse

g¥o±e - vaalaa a horse keeper/seller (M)

b.

sabzii vegetable

sabzii - vaalii vegetable vender/seller (F)

c.

miƥaaii sweets

miƥaaii - vaalaa sweet - seller/maker (M)

d.

Æ¥elaa cart

Æ¥ele - vaalaa cart pusher/owner (M)

e.

Ææksii taxi

Ææksii -vaalaa taxi driver (M)

f.

g¥ar house

g¥ar - vaalii wife (=house keeper) (F)

g.

kabaa±ii junk

kabaa±ii - vaalaa junk dealer

h.

baaÆlii bottle

baaÆlii - vaalaa a last name (etymology perhaps similar to ‘Goldsmith’)

The attachment of vaalaa illustrated in (22) is not productive. That is, vaalaa cannot be attached to say, kitaab "book", to derive "book seller", and so on. Also notice that the meaning that vaalaa contributes to the compound is not entirely systematic. The common thread of meaning in (22) is that of occupation. However, taxi vaalaa in (22e), for instance, cannot mean "seller" or "owner" of taxis. Clearly, then, the vaalaa in (22) must be lexically concatenated. It must be observed that a final high vowel of a noun in a noun+vaalaa form such as in (22) can undergo optional shortening. As pointed out in Verma (1971), the suffix vaalaa in Hindi has at least two other functions, where its attachment is not lexical. It marks a subject relative clause, as in (23):

12 It is interesting to note that if the nominal in an N+V-vaalaa form is also one that is part of an N-vaalaa form, then the verbs that it combines with to form the N+V form fall to a large extent within the range of meanings that the N+vaalaa form carries.

Verb Agreement / 119

(23) Vaalaa as a relative clause marker a.

Boston-se kaanfrens-ke liye aane vaalaa aadmii ... Boston-from conference-for coming man The man who is coming from Boston for the conference ...

b.

moÆe panne

aur

ßaandaar

tasviirõ

vaalii

kitaab... thick pages and grandpictures that That book, which has thick pages and grand pictures...

vah book

Vaalaa can also be an aspect marker indicating immediate future, as in (24): (24) Vaalaa as a marker of immediate future a.

raam kaanfrens-ke liye kæliforniya jaane vaalaa hai. Ram conference-for California g o about to i s Ram is going soon/about to go to California for a conference.

b.

raam kaccecaaval k¥aane vaalaa hai. Ram uncooked rice eat about to i s Ram is about to eat uncooked rice.

Consider now the sentence containing the suffix vaalaa in (25). The sentence has three possible interpretations:

(25)

yah ßer gaajar k¥aane vaalaa hai. this tiger-N carrot-N eat-NF be-PR (i) The tiger is about to eat a carrot. (immediate future) (ii) This tiger is one who eats carrots. (relative clause) (iii) This tiger is a carrot-eater. (N+V compound)

If the words gaajar k¥aane vaalaa in (25) are said with more than one primary stress and word melody, the incorporated reading in (25 iii) is not available. Nor is this reading available unless all the syntactic conditions for incorporation discussed earlier are met. In short, we have two types of vaalaa s, one attached lexically, as in (22), and the other non-lexically, as in (23) and (24). The question is, with respect to the properties that distinguish the two types, what is the status of vaalaa attachment to the N+V compound in (21)? The table in (26) summarizes the properties of the four different vaalaa s we have seen. (A = the

120 / Arguments in Hindi

N+V+ vaalaa ((20)); B = as the occupation marker ((21)); C = as the rela-

tive clause marker ((22)); and D = the immediate future marker ((23))).

Verb Agreement / 121

(24) A

B

C

D

genericity requirement





x

x

modification disallowed





x

x

case marking disallowed





x

intervening material disallowed





conjoining and gapping disallowed





x x x

separate stress and word melody on vaalaa disallowed





x

x

optional vowel shortening





x

x

x x

The pattern is obvious: the N+V compound+vaalaa in (21) systematically behaves like the lexically concatenated noun+vaalaa in (22), and not with the other vaalaa forms that are non-lexically concatenated. Therefore, the vaalaa in (21) must be lexically attached. In other words, the N+V compound must form a single lexical unit, exactly as represented in the category structure in (17). 5.2.4. Agreement with the Incorporated Object Now consider the facts of agreement in the incorporation construction. The verb agrees with its NOM subject, for instance, in (7a). But in (7b), (7c), and (7d), where the subject is NONNOM, verb agreement shifts to the incorporated object. Given the GC STR representation in (17), this presents an apparent puzzle. The verb kitaabã bec- "book-sell" agrees with an element internal to it, namely, kitaabã "books". However, this agreement with the object, which is internal to a compound created in the lexicon, takes effect only if agreement with the subject, which is concatenated in the phrasal module, is prevented by NONNOM case. The solution to this apparent puzzle lies in that even though the GC STR representation for the incorporated and the unincorporated readings are distinct, as in (16) and (17), the ARG STR and GF STR is common to both readings. The ARG STR and GF STR representation in (27) below is shared by both readings of (7b). The GC STR in (27) is that of the incorporated reading only (=(17)). The difference in structure between the two readings lies purely in the category and constituency information at GC STR in (16) and (17). Observe that the structure in (27) satisfies the requirements for agreement. Although "incorporated" at GC STR, the nominal is an ARG at ARG STR, and is

122 / Arguments in Hindi

associated with NOM case at GF STR by default (§4: (38)). It is therefore a legitimate controller of agreement. (27)

CLAUSE ARG ARG A-PRED ARG STR …… /……… /………/………………………………………… SUBJ OBJ ERG NOM

PRED PERF

GF STR

………………………………………………………………………… anil-ne

kitaabã

becññ

WORD

STRING

………………………………………………………………………… N

V

GC STR

V

V

N

S

The argumenthood of the incorporated OBJ in (27) also explains why the structure cannot contain another NP object, as is possible, say, in the English sentence: Max stage-managed the company . The incorporated object stage in this sentence does not fill an ARG position, hence the availability of the ARG position for the object company. In Hindi, on the other hand, the incorporated object occupies the ARG position, ruling out an independent object in the construction.

5.3. Concluding Remarks In sum, verb agreement in Hindi is simultaneously sensitive to case at GF STR and argumenthood at ARG STR. The controller of verb agreement is the highest ARG associated with NOM case. Given that only subjects and primary objects get NOM case, this condition automatically restricts agreement to subjects and primary objects. The principle of agreement holds even when the controller of agreement forms a lexical compound with the verb. In other words, agreement is not sensitive to the constituency or category information of the construction. Thus, in the absence of a NOM SUBJ, the verb agrees with the NOM OBJ whether or not the OBJ is a lexical category or a maximal projection, and whether or not it is dominated by a V0 or an S . In Chapter 8, we will see that it may also be dominated by a V.

Verb Agreement / 123

The analysis of object incorporation in Hindi provided in this chapter conflicts with the claim in Baker (1988) that noun incorporation must be analysed as the syntactic movement of a lexical head to combine with another X0. 13 In this analysis, incorporation takes place in the lexicon, and cannot be treated as a syntactic movement. When juxtaposed with our analysis of object incorporation, verb agreement in Hindi challenges the very notion of the lexicon as a unitary notion. What is a "lexical" phenomenon? Within a standard interpretation, it is one that takes place at the "word level". What, then, is a word? Take, for instance, a verbal predicate and its nominal argument. In a syntactic sense, they are two different words. However, in the incorporation construction, they form a single morphological word. Now, the term "lexical" is used to refer to phenomena internal to a "word" in the syntactic sense, as well as in the morphological sense, with the expectation that they converge. When the two diverge, so do theories of syntax. A syntactic analysis of incorporation finds its basis in the predicate and the argument being two syntactic units. A lexical analysis, such as the one proposed here, finds its basis in the predicate and its incorporated object being a single morphological word.14 Once we recognize the non-convergence of syntactic and morphological units, and separate the two notions of "lexical", agreement in the incorporation construction in Hindi ceases to be a puzzle. Suppose for now that we restrict the term lexical to refer to phenomena internal to morphological words. The noun incorporation in Hindi discussed in this chapter then is lexical. This is represented in the structure in (17) in terms of categories and constituency at GC STR, where the verb and its nominal object form a single morphological word (V0). Verb agreement in Hindi, however, is not sensitive to this information. For the purposes of agreement, the object is an argument of the predicate at ARG STR in the representation in (27), capturing the intuition that the predicate and its argument are two independent syntactic units. In short, the N+V compound is composed of two elements which form a single morphological structure but retain their syntactic independence. The grammatical features of case and agreement have been assumed in this thesis to belong to the same level of structure as grammatical func13 So does any incorporation construction that undergoes lexical phonological or morphological processes. 14 It is this non-convergence of morphological and syntactic units that the theory of Autolexical Syntax (Sadock (1985)) seeks to capture.

124 / Arguments in Hindi

tions. This is consistent with the claim in LFG that these grammatical features are part of f-structure (Kaplan and Bresnan (1982), Simpson (1983), Neidle (1982; 1988)). Another possibility might have been to consider these grammatical features to belong with grammatical categories in the level of GC STR. The grouping together of grammatical features with categories is found, for instance, in Generalized Phrase Structure Grammar (Gazdar et al (1985)), and in Head-driven Phrase Structure Grammar (Pollard and Sag (1987)). The account of agreement in the incorporation construction in this chapter provides evidence against grouping grammatical features with grammatical categories. Agreement is sensitive to case, and if case were part of GC STR, agreement in the incorporation construction would be forced to look at the internal structure of a word at GC STR, thus violating the lexical integrity hypothesis. Central to our account of incorporation in Hindi is the dual representation of categories and functions in (27), parallel to the representations of cstructure and f-structure in LFG. The facts and analysis of agreement in the incorporation construction in Hindi thus provide crucial evidence for the separation of the part-whole (or head-daughter) relationships in the representation of grammatical functions and that of grammatical categories: the two sets of relationships are not identical in the incorporation construction.

6 Grammatical Subjecthood In a search for universal properties that would identify the subject of any given sentence in any given language, Keenan (1975) distinguishes semantics, coding, and behavior as bases for criterial properties of subjecthood. SEMANTIC properties are properties such as agency, autonomous existence, and animacy. CODING properties are properties of word order, case marking, and verb agreement. BEHAVIORAL properties have to do with "deletion", "movement", and control of coreference. A number of researchers have employed tests provided by these three types of properties to identify grammatical subjects in Hindi and other South Asian languages (Kachru (1970), Kachru et al (1976), Sridhar (1976a), Verma ed. (1976), and so on). This chapter is concerned with reexamining subjects in Hindi with respect to these properties. Our purpose is to clearly formulate a set of diagnostics which would provide evidence for the subjecthood of a nominal in Hindi. In our terminology, coding properties mostly involve the realization of GFs in terms of grammatical features, behavioral properties involve syntactic principles that refer to GFs, and semantic properties involve the relation between GFs and entities at SEM STR. §6.1. briefly sketches the semantic and coding properties prototypically associated with grammatical subjects, and says that these properties do not yield any diagnostics for subjecthood in Hindi. §6.2. focuses on formulating diagnostics for grammatical subjecthood on the basis of the behavior of NOM and ERG subjects, which are uncontroversially recognized as grammatical subjects. These diagnostics are directly relevant for our purposes: they will be crucially employed in the next chapter to establish the subjecthood of the controversial subjects, namely, those involving indirect case.

6.1. Semantic and Coding Properties of Subjects Recall the principle whereby an L-SUBJ at ARG STR is canonically mapped onto a SUBJ at GF STR (§3: (43a)). The intuitive substance of this principle 122

Grammatical Subjecthood / 123

is that the highest "theta role" of a predicate is the prototypical subject. Another intuition shared by linguists is that an agent is the prototypical subject. This intuition is expressed in GB as the condition that agents are "external arguments", and in the Lexical Mapping Theory as the condition that agents cannot be objects. These two principles come closest to defining semantic properties of grammatical subjects. However, neither logical subjecthood nor agency are necessary or sufficient conditions for grammatical subjecthood. Therefore, while they are useful heuristics, they do not yield reliable diagnostics. As illustrated briefly in Chapter 2, Hindi is a language with considerable freedom of word order. Although word order changes may be associated with systematic alternations of GFs in a limited set of constructions (T.Mohanan (1989)), clause internal changes in precedence relations largely result only in changes of focus, emphasis, and other discourse meanings such as definiteness. In other words, without supporting evidence from various properties of behavior, word order cannot be taken as encoding GFs in Hindi, and provides no diagnostic for grammatical subjecthood. As we saw in Chapter 4, there is no one-to-one correspondence between GFs and case in Hindi. Not all NOM nominals are subjects, nor are all subjects NOM. Nominative case, which is prototypically the case of the subject, cannot therefore be used as a test for subjecthood in Hindi. However, among the various case features that appear on subjects in Hindi, ERG is uniquely associated with SUBJ. Thus, although all subjects are not ERG, all ERG nominals are subjects. The generalization then is: (1)

An ERG argument must be a SUBJ.

When verbs in a language agree with one and only one of their arguments, it has been assumed that that argument must be the grammatical subject; and further, that if the verb does not agree with an argument, that argument is not the subject. These assumptions lead to the conclusion that because NONNOM logical subjects in Hindi are not controllers of verb agreement, they do not qualify to be grammatical subjects (Perlmutter and Postal (1974)). But as we saw in Chapter 5, subjecthood is neither necessary nor sufficient for verb agreement in Hindi. Therefore, agreement cannot provide a defining criterion for grammatical subjecthood. Given the condition in Chapter 5 that the verb agrees with the highest ARG in NOM case, however, we may formulate the following agreementbased generalization:

124 / Arguments in Hindi

(2)

A NOM ARG that the verb does not agree with cannot be a SUBJ.

(2) can only be used to confirm the non-subjecthood of a NOM ARG. The fact that NONNOM ARGs do not control verb agreement does not preclude them from grammatical subjecthood. In sum, semantic and coding properties cannot be appealed to for establishing the grammatical subjecthood of a nominal in Hindi.

6.2. Behavioral Properties of Subjects This section tries to give a systematic account of some behavioral properties of nominals that are unequivocal instances of grammatical subjects. These properties are used to construct diagnostics that can provide a basis for examining the "deviant" instances of grammatical subjects, namely, the indirect case subjects. Exactly as with coding properties, there is no absolute one-to-one correlation between GFs and their behavior in Hindi. Yet, reliable unidirectional diagnostics may be formulated towards locating SUBJs. §6.2.1. looks at the reflexive apnaa, whose antecedent must be either a SUBJ or an L-SUBJ. §6.2.2. shows that the antecedent of a pronoun cannot be a SUBJ within its minimal finite clause. §6.2.3. identifies two control constructions in which the controller of the null NP of a participial adjunct must be a SUBJ. It is widely acknowledged in syntactic theory that obligatorily null control sites are grammatical subjects. However, because of interference from semantics, this does not serve as a reliable test for the subjecthood of indirect case subjects, and therefore, I will not deal with this property of subjects. §6.2.4. discusses certain necessary conditions for gapping (in the sense of Ross (1970)) in coordination: the GF and case of the deleted element must be identical to the GF and case of the trigger of the gap, a result which can be fruitfully employed to identify the GF of nominals, and to isolate the SUBJ. 6.2.1. The Reflexive

apnaa

For many Hindi speakers, the reflexive apnaa can take as its antecedent a subject, grammatical or logical, but no other argument. This is illustrated by the sentences in (3):

Grammatical Subjecthood / 125

(3) a.

ravii apnii saikil-par Ravi-N self-G bicycle-L Ravii sat on self'si bike.

baiƥaa. sit-PERF

In (3a), Ravi, is the L-SUBJ and the SUBJ. It is also the antecedent of the reflexive. b.

vijay-ne ravii-ko apnii saikil-par biƥaayaa. Vijay-E Ravi-A self-G bicycle-L sit-C -PERF Vijayj seated Ravii on self'sj/*i bike. (Lit.: Vijay caused Ravi to sit on Vijay's bike.)

When (3a) is causativized, as in (3b), Vijay , the causer, is the L-SUBJ and the SUBJ. It is also the only eligible antecedent of apnaa . Ravi , the causee, as we will see shortly, is neither an L-SUBJ nor a SUBJ. c.

raam-ne vijay-se ravii-ko apnii saikil-par biƥvaayaa. Ram-E Vijay-I Ravi-A self-G bicycle-I sit-C-C-PERF Ramk made Vijayj seat Ravii on self'sk/*i/*j bike. (Lit.: Ram caused Vijay to cause Ravi to sit on Ram's bike.)

(3c) is the causative alternant of (3b). In (3c), the new causer, Ram , is the only L-SUBJ and SUBJ, and again, is the only possible antecedent of apnaa. Now consider the passive alternant of (3b). d. ravii vijay-se apnii saikil-par biƥaayaa gayaa.1 Ravi-N Vijay-I self-G bicycle-L sit-C-PERF go-PERF Ravii was seated by Vijayi on self'si/j bike.

In the passive sentence in (3d), Vijay , the "demoted" agent, is the L-SUBJ. However, the causee Ravi is now the SUBJ. Either of them can be antecedents of apnaa . Observe that the causees in (3b) and (3c) are not SUBJs. Nor are they L-SUBJs. The conclusion that follows is that the causative construction in Hindi is a single ARG STR clause and a single GF STR clause. Thus, a sim-

1

The passive verb form in Hindi is discussed in §2.2.

126 / Arguments in Hindi

plified SEM STR, ARG STR, and GF STR representation of (3c) would be as follows:2 (4)

[

x causer

m

CAUSE ] EVENT

____________________ [ y z n CAUSE] causer

causee EVENT

________ [ z sit ]

SEM STR

………………………/………………………………………………… ˘ ARG

ARG

ARG

¿

A-PRED

ARG

STR

…………………………………………………………………………… [ SUBJ OBL Ram Vijay

OBJ ]

Ravi

GF-PRED sit-CAUS-CAUS

GF STR

The three SEM STR clauses in (4) correspond to one ARG STR clause and GF STR clause. Using the RG terminology of "clause union" (Perlmutter and Postal (1974)), this is a clause union in the mapping between SEM STR and ARG STR. When a structure contains more than one L-SUBJ or SUBJ, as in (5), any of the L-SUBJs or SUBJs may be the antecedent of the reflexive apnaa :

(5)

raajaa-ne mantrii-ko apne g¥ar jaane-kii king- E minister-A self-G house-L go-NF-G aagyaa dii. order-N give-PERF The kingi ordered the ministerj to go to self's i/j house.

In (5), the object of the matrix verb grammatically controls the subject of the embedded clause. "The minister" is an eligible antecedent of the reflexive by virtue of being the L-SUBJ and SUBJ of the embedded clause, and "the

2 In T. Mohanan (1988), I distinguish between "indirect causatives" and "direct causatives" in terms of dyadic versus triadic CAUSE predicates in SEM STR. Thus, "causer causes event" is indirect causation; "causer acts upon causee t o cause event" is direct causation. That is to say, common to all causatives is the structure [causer CAUSES event]. In direct causation, the structure [causer ACT UPON causee] is superimposed on this structure. The matrix CAUSE predicate in (4) is a simplified representation of indirect causation; the embedded CAUSE predicate is a simplified representation of direct causation.

Grammatical Subjecthood / 127

king" by virtue of being the L-SUBJ and SUBJ of the matrix clause. Now consider (6):

(6)

raajaa-ne mantrii-se apne g¥ar jaane-kaa king- E minister-I self-G house-L go-NF-G vaadaa kiyaa. promise-N do-PERF The kingi promised the ministerj to go to self's i/*j house.

In contrast to (5), the embedded subject in (6) is controlled by the matrix subject. "The minister" in (6) cannot be the antecedent of the reflexive because it is not an L-SUBJ or SUBJ either in the embedded or the matrix clause. (7) below shows that even though apnaa is not clause-bounded, the antecedent cannot be outside its finite clause:

(7)

raajaa-ne kahaa king- E say-PERF

k i mantrii that minister-N

apne self-G

g¥ar gayaa. house-L go-

PERF

The kingi said that the ministerj went to self's j/*i house.

As we might expect, the antecedent of apnaa must also c-command it.3 The non-c-commanding SUBJ in (8) cannot be the antecedent of apnaa :

(8)

raajaa-kaa hasnaa (*apne) mantrii-ko buraa lagaa. king- G laugh-NF self-G minister-D bad be struck-PERF The king's laughing made (*self's) minister feel bad.

"The king" is the SUBJ of the embedded nonfinite clause in (8). Hence, it is not an eligible antecedent of the reflexive. There being no other eligible antecedent, the sentence is ungrammatical. In the light of the above facts, the principle governing the binding of apnaa may be stated as: 4 3 I use the term "c-command" to refer to a configuration: node x c-commands node y if the branching node most immediately dominating x also dominates y . Although the term c-command is used in syntax to refer to a categorial configuration, I use it here to refer to a GF STR configuration (cf. f-command (Bresnan (1982c))). 4 On the basis of examples such as the following, Gurtu (1985) concludes that the generalization underlying (9), formulated in various ways in the literature (Subbarao (1971), Cohen (1973), Kachru and Bhatia (1975) and others), is false. jon-ne merii-ko

apnii kitaab dii.

128 / Arguments in Hindi

(9)

The reflexive apnaa must be bound by an L-SUBJ or minimal finite clause.

SUBJ

within its

The diagnostic for SUBJ-hood with respect to apnaa may be formulated as:

(10) The antecedent of apnaa must be an L-SUBJ or a SUBJ. 6.2.2. Subject Obviation The antecedent of a pronoun in Hindi cannot be the SUBJ of its clause: (11) a.

ravii uskii saikil-par Ravi-E pron(oun)-G bicycle-L Ravii sat on his * i bike.

b.

vijay-ne ravii-ko uskii saikil-par biƥaayaa. Vijay-E Ravi-A pron-G bicycle-L sit-C -PERF Vijayj seated Ravii on his i/*j bike.

c.

raam-ne Ram-E

vijay-se Vijay-I

ravii-ko Ravi-A

baiƥaa. sit-PERF

uskii saikil-par biƥvaayaa. pron-G bicycle-I sit-C -C -

PERF

Ramk made Vijayj seat Ravii on his i/j/*k bike. d.

ravii Ravi-N

vijay-se Vijay-I

uskii saikil-par biƥaayaa gayaa. pron-G bicycle-L sit-C-PERF go-

PERF

Ravii was seated by Vijayj on his j/*i bike.

In (11a-c), the L-SUBJ is also the SUBJ. The pronoun can be coreferent with any nominal other than the SUBJ in these sentences. In the passive in (11d), the L-SUBJ is Vijay . The pronoun can be coreferent with it in the passive because it is not the SUBJ. (12) shows that subject obviation, that is, the inability of the pronoun to take SUBJ antecedents, holds within the finite clause: John-EMary-D self-G book-N John gave Maryi self'si book.

give-PERF

As has been pointed out several times, the patterns of reflexive binding need not be identical across speakers of the same "language". Thus, the reflexive in the above example cannot refer to the non-subject Mary for many speakers. That there exist speakers of "Hindi" for whom the conditions of reflexive binding are different does not question the validity of (9) for the speakers for whom the generalization holds.

Grammatical Subjecthood / 129

(12) raajaa-ne mantrii-ko uske g¥ar jaanee-kii aagyaa dii. king-Eminister-A pron-G house-L go-NF-G order-N give-PERF The kingi ordered the ministerj to go to his *i/*j house.

The pronoun cannot be coreferent with either "the king" or "the minister": the former is the SUBJ of the finite clause, and the latter the SUBJ of the minimal clause, that contains the pronoun. In (13), "the minister" is not a SUBJ, and therefore, is an eligible antecedent of the pronoun: (13) raajaa-ne mantrii-se uske g¥ar jaanee-kaa vaadaa kiyaa. king-E minister-I pron-G house-L go-NF-G promise-N do-PERF The kingi promised the ministerj to go to his j/*i house.

The domain of subject obviation, as with reflexive binding, is the finite clause: the pronoun in a finite clause may be coreferent with the SUBJ of a higher clause: (14) raajaa-ne kahaa ki mantrii uske g¥ar gayaa. king-Esay-PERF that minister-N pron-G house-L go-PERF The kingi said that the ministerj went to his i/*j house

Although "the king" is the SUBJ of the matrix verb, the pronoun can be coreferent with it because it is outside the minimal finite clause that contains the pronoun. (15) below shows that subject obviation, like reflexive binding, involves the c-command condition: (15) raajaa-kaa hasnaa uske mantrii-ko buraa lagaa. king- G laughing pron-G minister-D bad The king'si laughing made hisi minister feel bad.

struck

The non-c-commanding SUBJ of the minimal finite clause, "the king" in (15), may be the antecedent of the pronoun. On the basis of these facts, we state the principle of subject obviation in Hindi as:

130 / Arguments in Hindi

(16) A pronominal cannot be bound by a clause.5

SUBJ

within its minimal finite

We can now extract the following diagnostic from (16):

(17) The antecedent of a pronoun cannot be a c-commanding

SUBJ

of the

minimal finite clause that contains the pronoun.

A result of the combination of (10) and (17) is the absence of a complete complementarity in the distribution of the reflexive and pronouns. An L-SUBJ that is not a SUBJ can be the antecedent of both a reflexive and a pronoun. This is precisely what happens in passives. The demoted agent is an L-SUBJ, and can therefore be the antecedent of a reflexive, as in (3d); it is not a SUBJ, and can therefore be the antecedent of a pronoun, as in (11c). 6.2.3. Participial Control Clauses As is common crosslinguistically, participial adjuncts in Hindi require their controllers to be grammatical subjects. We will consider two such participial adjuncts in this section. The sentences in (18)-(19) contain present participial adjuncts. (18) shows that the control site in (19) is obligatory. That is, the argument cannot be realized as an overt NP: (18) * ravii-ne vijay-ko [ apne / niinaa-ke muskuraate hue] biƥaayaa. Ravi-E Vijay-A self-G Nina-G smile-IMPERF be-NF sit-C-PERF *Ravi seated Vijay while self / Nina smiling.

(19) a.

ravii-ne vijay-ko [ ——— muskuraate hue] biÆ¥aayaa. Ravi-E Vijay-A smile-IMPERF be-NF sit-C -PERF Ravii seated Vijayj while ——— i/*j smiling.

5 The principle in (16) is stated in terms of "pronominal" rather than "pronoun" because it applies to null pronominals (small pro) as well. (i) raajaa-ne mantrii-ko [——— kitaab b¥ejne-kii ] aagyaa dii. (ii)

king-E minister-A book-N send-NF-G order-N give-PERF The kingi ordered the ministerj to send him *i/*j the book. raajaa-ne mantrii-se [ ——— kitaab b¥ejne-kaa ] vaadaa kiyaa. king-E minister-I book-N send-NF-G promise-N do-PERF The kingi promised the ministerj to send him *i/j the book.

Grammatical Subjecthood / 131

b.

raam-ne ravii-se vijay-ko [ ——— muskuraate hue ]

biÆ¥vaayaaa. Ram-E Ravi-I Vijay-A smile-IMPERF be-NF sit-C-C-PERF Ramk made Ravii seat Vijayj while ——— k/*i/*j smiling. c.

ravii-se vijay [ ——— muskuraate hue] biÆ¥aayaa gayaa. Ravi-I Vijay-N smile-IMPERF be-NF sit-C -PERF go-PERF Vijayj was seated by Ravii while ——— *i/j smiling.

The contrast in controllerhood between (19a, b) and (19c) shows that the controller of these present participial adjunct clauses must obligatorily be the matrix SUBJ.6 The same generalization holds with respect to the participial adjuncts in (20), where again, the control site is obligatory: 7 (20) a.

ravii-ne vijay-ko [ ——— darvaazaa k¥ol kar ] biÆ¥aayaa. Ravi-E Vijay-A door-N opendo sit-C -PERF ——— i/*j having opened the door, Ravii seated Vijayj .

b.

raam-ne ravii-se vijay-ko [ ——— darvaazaa

k¥ol kar ]

biÆ¥vaayaaa. Ram-E Ravi-I Vijay-A door-N open do sit-C-C-PERF ——— k/*i/*j having opened the door, Ramk made Ravii seat Vijayj . 6 The present participial clause in (19a-c) may also be interpreted as the complement of the verb with an idiosyncratically restricted set of verbs such as dek¥ 'see', and mil 'find'. See Harbert and Srivastav (1988) for a discusion of asymmetries in the binding of anaphors in complements and adjuncts in Hindi. The asymmetry is found in English as well: in (i), the participial clause is a complement, and in (ii), an adjunct: (i) John found Mary playing with the cat. (ii) John found Mary, (while) playing with the cat. In English, if the clause is a complement, the controller may be an object. However, it must then necessarily follow the object. If it is an adjunct, it may precede or follow the subject. Yet another difference between them is that the object and the participial clause are said as a single intonational phrase if it is a complement, but an intonational break is possible after the object if the participial clause is an adjunct. 7 The controllee in the embedded clauses in (19)-(20) must also be a grammatical subject. Examples like (19), which have been used in the literature on Hindi as a test for subjecthood, have been referred to as raising constructions (Kachru et al (1976). Note that they cannot be raising constructions, given that the controllee is assigned a theta role by the matrix verb. Control in participial adjunct clauses in (20) has been treated as conjunction reduction in some studies on subjects in South Asian languages (e.g. Kachru et al (1976)).

132 / Arguments in Hindi

c.

ravii-se vijay [ ——— darvaazaa k¥ol kar] biÆ¥aayaa gayaa. Ravi-I Vijay-N door-N open do sit-PERF go-PERF ——— *i/j having opened the door, Vijayj was seated by Ravii .

Once again, the only legitimate controllers of these adjunct clauses are the matrix SUBJs. The diagnostic offered by (19)-(20) may be formulated as follows:

(21) A nominal that can control a participial adjunct clause with an obligatory control site must be a

SUBJ.

It must be emphasized that the reverse of condition (21) is not true. That an NP cannot control the adjunct does not imply that the NP is a nonsubject. For example, consider the following sentence, with a stative predicate: (22) a.

mohan us kamre-mã Mohan-N that room-L Mohan is in that room.

hai. be-PR

Mohan in (22a) is the antecedent of the reflexive but not the pronoun in (22b): (22) b.

mohan apne / us-ke kamre-mã Mohan-N self's his room-L Mohan i is in self's i / his * i room.

hai. be-PR

However, it cannot be the controller of participial adjunct clauses ((22c)): (22) c. * mohan [ ——— muskuraate hue] us Mohan-N smile-NF be-NF * Mohan is in the room smiling.

kamre-mã that room-L

hai. be-PR

We must therefore conclude that not all SUBJs can be controllers of obligatory control sites. It remains, however, that if a nominal can be controller of a participial adjunct clause with an obligatory control site, it must be a SUBJ. Now, most syntactic theories acknowledge that an obligatory control site in a nonfinite clause must be a grammatical subject. This is illustrated in Hindi in the participial clauses in (23). (23a) is ungrammatical because

Grammatical Subjecthood / 133

all the arguments are overtly expressed: the participial clause in (23a) requires a control site: (23) a. * [raam-ne darvaazaa k¥ol Ram- E door-N open

kar ] anil-ne do Anil-E

us-ko pron- A

saaf kiyaa. clean do-

PERF

* Ram having opened the door, Anil cleaned it.

In (23b), the SUBJ of the matrix clause controls the SUBJ of the embedded clause. The OBJ pronoun in the matrix clause is coreferent with the OBJ of the embedded clause: (23) b.

raam-ne [ ——— darvaazaa Ram-E door-N

k¥o l kar ] us-ko open do pron- A

saaf kiyaa. clean do-

PERF

Ram i , ——— i having opened the door, cleaned it.

In (23c), the SUBJ of the embedded clause is the control site, exactly as in (23b), but the OBJ of the embedded clause is a null pronominal (pro): (23) c.

raam-ne darvaazaa [ ——— k¥ol kar ] saaf kiyaa. Ram-E door-N open do clean do-PERF Ram i , ——— i having opened the door j, cleaned (it) j .

(23d) is ungrammatical, because a control site must be a SUBJ: (23) d. * [raam-ne ——— k¥ol kar ] anil-ne darvaazaa saaf kiyaa. Ram-E open do Anil-E door-N clean do-PERF * Ram having opened the door i , Anil cleaned ——— i .

(23e) shows that the ungrammaticality of (23d) is not merely the consequence of the controller being a non-SUBJ. (23e) is ungrammatical in spite of a SUBJ controller: (23) e. * darvaazaa [raam-ne ——— k¥ol kar ] band h o door-N Ram-E open do closed * The door, Ram having opened ——— i , closed.

gayaa. be do-PERF

134 / Arguments in Hindi

In (23d, e), the OBJ cannot be the control site. If it is interpreted as a pro, as in (23c), the structures are ill-formed for lack of a control site in the embedded clause. The test for subjecthood that the facts in (23) yield is stated in (24):8 (24)

The obligatorily null ARG of a participial adjunct clause must be a SUBJ .

It must be pointed out at the outset that the constructions with indirect case subjects that we will examine in Chapter 7 do not yield participial adjunct clauses. Therefore we will not have occasion to employ this diagnostic in our investigation of their subjecthood. 6.2.4. Gapping in Coordinate Structures Gapping of entities in coordination constructions requires identity of case and grammatical function between the gapper and the gappee. Hence, by gapping an element A whose GF is uncertain with an element B whose GF we are certain about, we may determine the GF of A. 6.2.4.1. Case Identity In coordination constructions in Hindi, the gapped element in a coordinate constituent must be identical to the gapper in case. Consider the following sentences:9 (25) a.

ravii g¥ar gayaa Ravi-N home-L go-PERF Ravi went home and slept.

aur ——— soyaa. and sleep-PERF

8 This test would of course not hold in participial clauses that do not have an obligatory site. Consider, for instance, the following example, where the control site is not obligatory: anu-neilaa-ko [ mãã-ke us-ko / ——— ±ããÆne-par ]tasallii dii. Anu-E Ila-D mother-G pron-A scold-NF on comfort-N Anu i comforted Ilaj on the mother's scolding him/her/——— i/j/k .

give-PERF

9 Instead of treating the sentences in (23) as instances of subject gapping (conjunction reduction with subject ellipsis), one might think of them as instances of VP-coordination. See Bresnan and Thráinsson (1984) for arguments for two different analyses of similar constructions in Icelandic. The point still holds that the two nominals must be identical in GF and case. Moreover, a VPcoordination analysis is inconsistent with our assumption that Hindi does not have a VP in the sense of a GC STR constituent that contains the verb and the object but not the subject.

Grammatical Subjecthood / 135

b.

ravii-ne k¥aanaa k¥aayaa aur ——— pikcar dek¥ii. Ravi-E food-N eat-PERF and movie-N see-PERF Ravi ate his meal and watched a movie.

c.

ravii

g¥ar

gayaa

aur

us-ne / * ———

k¥aanaa

k¥aayaa. Ravi-N home-N go-PERF and he-E Ravi went home and he ate his meal. d.

food-N eat-PERF

ravii-ne k¥aanaa k¥aayaa aur vah / * ——— g¥ar gayaa. Ravi-E food-N eat-PERF and he-N home-N go-PERF Ravi ate his meal and (he) went home.

Ravi is the grammatical subject of both conjuncts in each of the coordinate structures. In (25c, d), gapping is disallowed, because there is a mismatch in the case requirements of the conjoined clauses ((25c, d)).10 6.2.4.2. Grammatical Function Identity In addition to identity of case, gapping also requires the identity of grammatical function in the gapper and the gappee. First consider gapping among non-subjects. A locative oblique cannot be gapped by a locative adjunct ((26)), but can be gapped by another locative oblique ((27)):

(26)

raam mez-pari baiÆ¥aa aur Ram-N table-L sit-PERF and ravii-ne us-par i / * ——— kitaab rak¥ii. ravii-E it-L book- N put-PERF Ram sat on the table and Ravi put the book on it.

(27)

raam-ne mez-pari paaliß agaayaa Ram-E table-L polish- N apply-PERF ravii-ne ———i kitaab rak¥ii. ravii-E book- N put-PERF

aur and

10 That gapping is not prevented by a transitivity distinction is clear from the following examples, where both conjuncts have intransitive verbs: (i) ravii-ne nahaayaa aur vah / * ——— so gayaa. Ravi-E bathe-PERF and he-N Ravi bathed and (he) went to sleep.

sleep

go-PERF

If, on the other hand, a transitive and intransitive verb both take ergative subjects, they can gap each other: (ii) ravii-ne nahaayaa aur usne / ——— k¥aanaa k¥aayaa. Ravi-E bathe-PERF and he-E Ravi bathed and (he) ate his meal.

food-N eat-PERF

136 / Arguments in Hindi

Ram polished the table and Ravi put the book (on it).

Similarly, the demoted subject in a passive cannot be gapped by an adjunct bearing the same postposition ((28)), although it can be gapped by the demoted subject of another passive ((29)).

(28)

ravii-ne raam-ke dvaaraa i niinaa-ko ciÆÆ¥ii b¥ejii Ram-E Ram-G through Nina-D letter-N send-PERF aur anil uske dvaaraai / * ——— i piiÆaa gayaa. and Anil-N pron-G through beat-PERF go-

PERF

Ravi sent Nina a letter through Rami , and Anil was beaten by himi .

(29)

anil raam-ke dvaaraai piiÆaa gayaa Anil-N Ram-G through beat-PERF go-PERF and ravii ———i d¥akelaa gayaa. Ravi-N push-PERF go-PERF Anil was beaten and Ravi was pushed by Ram.

aur

The demoted subject of the passive in (28)-(29) bears the postposition dvaaraa "through". It can alternatively be marked with the INST case clitic se. Now, the (semantically) embedded agent of a causative construction ((e.g. 3c)) is also marked with the INST case clitic -se. But it cannot gap the ARG-ADJ in the passive ((30)), although it can marginally gap an INST oblique in a non-causative construction ((31)): (30)

raam-ne ravii-se pe± kaÆvaayaaaur Ram-E Ravi-I tree-N cut-C-PERF and anil us-sei / * ——— i piiÆaa gayaa. Anil-N pron-I beat-PERF go-PERF Ram made Ravii cut the tree and Anil was beaten by himi .

(31) raam-ne ravii-se pe± kaÆvaayaaaur Ram-E Ravi-I tree-N cut-C-PERF and anil-ne us-sei / ? ——— i bahut savaal puuc¥e. Anil-N pron-I manyquestions-N ask-PERF Ram made Ravii cut the tree and Anil asked himi many questions.

The contrast between (30) and (31) is accounted for if we assume that the embedded INST agent of the causative is an OBL, and therefore differs in GF from the demoted INST agent of the passive.

Grammatical Subjecthood / 137

For many speakers, the primary object and the secondary object cannot gap one another.11 The PR.OBJ of a dyadic verb can gap the PR.OBJ of a triadic verb ((32)), but not a SEC.OBJ ((33)):

(32)

ravii-ne kitaab k¥ariidii, aur raam-ne niinaa-ko ——— dii. Ravi-E book-N buy-PERF and Ram-E Nina-D give-PERF Ravi bought a book, and Ram gave it to Nina.

(33) ?* ravii-ne niinaa-ko bulaayaa, aur

raam-ne ——— kitaab dii. Ravi-E Nina-A call-PERF and Ram-E book-N give-PERF Ravi called Nina, and Ram gave her a book.

Now consider the mismatch between subjects and nonsubjects. Even though there is no case conflict, gapping is prohibited in (34b) because an OBJ cannot gap a SUBJ: (34) a.

ravii-ne aam k¥ariidaa aur niinaa-ne ——— k¥aayaa. Ravi-E mango-N buy-PERF and Nina-E eat-PERF Ravi bought, and Nina ate, a mango.

b.

ravii-ne

aam

k¥ariidaa aur vah / * ——— k¥aÆÆaa

niklaa. Ravi-E mango-N buy-PA and it-N sour Ravi bought a mango and it turned out sour.

emerge-PERF

Recall from Chapter 4 that inanimate objects are NOM, and not zeromarked ACC. While 'mango' in the second conjunct in (34b) is a SUBJ, in the first conjunct it is an OBJ, and therefore fails to gap. Given that gapping demands identity of case as well as GF, one might ask if logical subjecthood is also relevant for gapping. That this is not so is demonstrated by (35): piiÆaa gayaa aur ——— hasne Ravi-N beat-PERF go-PERF and laugh-NF Ravi was beaten and started laughing.

(35) ravii

lagaa. struck-PERF

11 For some speakers, including myself, the condition on gapping does not provide clear tests to distinguish between primary and secondary objects. However, even for these speakers, there is an asymmetry in the gapping relation. While a primary object can gap a secondary object, the reverse is not possible.

138 / Arguments in Hindi

The first conjunct of (35) is a passive: the SUBJ is not the L-SUBJ. In the second conjunct, however, the null argument is both the SUBJ and the LSUBJ. An L-SUBJ can thus be gapped by a nominal that is not an L-SUBJ. The condition on gapping in Hindi can now be stated as follows: (36)

A gapped element must be identical in GF and case to the gapper.

The diagnostic for subjecthood offered by (36) is: (37)

If an argument gaps or is gapped by a SUBJ, it must itself be a SUBJ.

6.2.5. A Summary of the Diagnostics The diagnostics for grammatical subjects in Hindi we have formulated so far are repeated below: (38) a. (= (10)) The antecedent of apnaa must be an L-SUBJ or a SUBJ. b. (= (17)) The antecedent of a pronoun cannot be a c-commanding minimal finite clause that contains the pronoun

SUBJ

of the

c. (= (21)) A nominal that can control a participial adjunct clause with an obligatory control site must be a SUBJ. d. (= (37)) If an argument gaps or is gapped by a

SUBJ,

it must itself be a SUBJ.

While these tests as stated above are reliable tests of subjecthood, they are unidirectional, and not all tests are relevant or applicable in all contexts. Hence the need for them all. The generalizations in most of these tests are fairly well known, except perhaps the subject obviation test ((38b)). However, to the best of my knowledge, the crucial elements in these generalizations have not so far been isolated and stated precisely as has been done above. For instance, the obligatory control relation in a participial adjunct clause can be used in principle to formulate two different unidirectional diagnostics: (a) If an NP is a controller, then it is a subject; (b) if an NP is a subject, then it can be a controller. Only (a), given as (38c) above, is true in Hindi. Nevertheless, the unidirectional tests have often been collapsed into

Grammatical Subjecthood / 139

and used as a single bidirectional test. The result has been the dubious assumption that subjecthood is a continuum.

140 / Arguments in Hindi

6.3. Grammatical Subjecthood as Prominence As Keenan rightly points out, none of the properties listed by him as basic subjecthood properties (coding, behavior, and semantics) provide an absolute definition of the notion SUBJ. Taken in isolation, none of them are either necessary or sufficient conditions for subjecthood. The crosslinguistically attested clustering of characteristic properties does, however, serve as a basis for constructing language specific diagnostics for subjecthood. Keenan's solution to this problem is to take these properties as contributing towards a prototypical definition of subjecthood. By making such a move, Keenan and others are forced to conclude that an entity that has more of these properties is more of a subject, and one that has less of the properties is less of a subject. That is to say, subjecthood is a continuum, or a squish, and "not a discrete relational category" (Kachru et al (1976:79), Sridhar (1976b)). Rather than assuming that the properties outlined by Keenan are definitional of grammatical subjects, we might assume that these properties are symptomatic of various types of PROMINENCES. SUBJ is the most prominent entity at GF STR, referred to repeatedly in patterns of anaphora and control. Another prominence recognized as relevant for various linguistic phenomena is that of L-SUBJ, the most prominent entity at ARG STR. Prominence on the animacy scale (roughly: human < animate < inanimate), as pointed out by Keenan, is a semantic prominence relevant for subject selection, and, as we saw in Chapter 4, for ACC case selection for objects in South Asian languages. Prominence on the dimension of case has often been attributed to NOM case. This prominence is found in Chapter 5 to be relevant, for instance, for verb agreement in South Asian languages. Yet another type of prominence is that of person, shown to be relevant for agreement in Gujarati (Mistry (1978)). Viewed in this light, many of what Keenan identifies as prototypical properties of grammatical subjects may be associated with other prominences. A close examination reveals that there is interaction among the scales of prominence, and that linguistic patterns of coding, behavior, and meaning may appeal to prominences of more than one type. To cite an example, reflexives in natural languages may refer not only to grammatical subjects but also to logical subjects (Perlmutter (1978, 1981), Joshi (1987)), 'logophoric' nominals (Clements (1975), Sells (1987), Matsumoto (1988)), and to topics (Alsagoff (1989)). Similarly, the facts of Hindi

Grammatical Subjecthood / 141

agreement make sense if viewed as depending upon two types of prominence, that of case and GF. In the unmarked instance, the various prominences, including grammatical subjecthood, converge on a single nominal, creating the mistaken assumption that all properties of coding and behavior refer exclusively to GF prominence. The effects of apparent nondiscreteness in the notion SUBJ, then, are results of mismatches of prominence. For instance, if we assume that case marking is solely determined by GFs, and equate SUBJ with NOM, then the presence of NONNOM case on an argument that behaves like a SUBJ must be attributed to the argument being less of a SUBJ than a NOM SUBJ. Once these different prominences and their effects are recognized and factored apart, we are no longer forced to view subjecthood as a squish.

6.4. Concluding Remarks The binding of the Hindi reflexive apnaa appeals to L-SUBJ, the most prominent logical ARG, and to SUBJ, the most prominent GF: the antecedent of apnaa must be an L-SUBJ or SUBJ within its minimal finite clause. The demoted subject of a passive can be the antecedent of the reflexive apnaa by virtue of being an L-SUBJ. The notion L-SUBJ or an equivalent must therefore be expressed in any theory of syntax. In a theory such as GB, one might think that the binding of apnaa can be expressed in terms of the following statement: the antecedent of apnaa must be either an external argument or an NP of S at s-structure. However, as we will see in §7.5., an argument in the "accusative logical subject construction", which is neither an external argument, nor an NP of S at s-structure, is nevertheless an L-SUBJ, and can be the antecedent of apnaa.12 In short, L-SUBJ is distinct from the notion external argument, and must be incorporated into a theory such as GB for an adequate account of reflexive binding in Hindi. Now, there have been proposals in GB to the effect that all the arguments of a predicate originate within the VP (e.g. Koopman and Sportiche (1985), Sportiche (1988)). If 12

Another example of an L-SUBJ that is neither an external argument nor an of S at s-structure is the goal argument in "psych" verbs in Hindi. The analysis of inversion in psych verbs in GB (e.g. Belletti and Rizzi (1988)) involves the assumption that they do not contain external arguments. When the theme argument of the psych verbs in Hindi is a grammatical subject, the goal argument, which is not an external argument, can nevertheless be the antecedent of the reflexive apnaa by virtue of being the L-SUBJ. See T. Mohanan (1989) for details of this construction. NP

142 / Arguments in Hindi

so, perhaps the equivalent of L-SUBJ would be the highest argument in a VP at d-structure. A notion equivalent to L-SUBJ must be incorporated into a theory such as RG as well. The fact that the causative construction in Hindi has only one L-SUBJ motivates the structure in (4), in which there is only one ARG STR clause corresponding to three SEM STR clauses. As pointed out earlier, this merger is an instance of what has been called "clause union" in RG (Perlmutter and Postal (1974)). The separation of SEM STR and ARG STR as two distinct levels of representation, which allows a many-to-one association between them (§3: (16)), facilitates the formal representation of Hindi causatives in terms of a merger in the mapping between SEM STR and ARG STR, yielding only one L-SUBJ for a given causative verb. For an adequate account of the interaction between causatives and reflexive binding in Hindi, it is imperative that even the "initial" syntactic structure of a causative (= ARG STR) be "monoclausal". Now, clause union in RG is a relation between two syntactic structures. The principle governing the relation between the reflexive and its antecedent, such that it includes causative constructions, might be stated as follows: The antecedent of the reflexive must be an initial 1 or final 1 in the final clause that contains the reflexive. Such a statement would not only be unrevealing, but also require appealing to two different types of initial 1s, one computed prior to clause union, and the other after clause union. Whether or not RG would permit such a move, I do not know. In Chapter 7, we will employ the diagnostics established on the basis of NOM and ERG subjects to argue for the grammatical subjecthood of DAT, INST, LOC, and GEN nominals.

7 Indirect Case on Subjects A striking pattern in the syntax of Hindi is that arguments marked with any of the case features available in the language, with the exception of ACC case, can function as the grammatical subject of a clause. The sentences in (1) below are repeated from Chapter 4. They illustrate logical subjects in Hindi with NOM, ERG, DAT, INST, LOC and GEN cases, given in bold face: (1) a.

ravii kelaa k¥aa rahaa t¥aa. Ravi-N banana-N eat PROG be-PA Ravi was eating a banana.

b.

ravii-ne kelaa Ravi-E banana-N Ravi ate the banana.

c.

ravii-ko kelaa k¥aanaa t¥aa. Ravi-D banana-N eat-NF be-PA Ravi was obliged to/needed to eat the banana.

d.

ravii-se kelaa k¥aayaa Ravi-I banana-N eat-PERF Ravi couldn't eat the banana.

e.

ravii-ke caar bacce t¥e. Ravi-G four children-N be-PA Ravi had four children.

f.

ravii-mã bilkul dayaa Ravi-L at all mercy-N Ravi had no mercy at all.

k¥aayaa. eat-PERF

nahññ gayaa. not go-PERF

not

nahññ t¥ii. be-PA

As (1) demonstrates, a wide array of case features is available to the L-SUBJ of a verb. The question is, is every L-SUBJ in (1) also a SUBJ? Recall that the association of indirect case with an ARG may be governed entirely by the meaning of the ARG, and is independent of the GF of the ARG. Therefore an L-SUBJ, by virtue of its meaning alone, may be 142

Indirect Case on Subjects / 143

associated with indirect case. An L-SUBJ is prototypically also a SUBJ. The result is a SUBJ with indirect case. Thus, although indirect case is not directly associated with grammatical subjecthood, it is perfectly natural for indirect case to appear on grammatical subjects. One primary aim of this chapter is to argue that the L-SUBJs in (1c-f) are indeed SUBJs. We will apply to clauses with L-SUBJs associated with DAT, INST, GEN and LOC cases the diagnostics for subjecthood formulated in Chapter 6, and show that each of them is the only eligible candidate for subjecthood in its clause. Another equally important goal of this chapter is to show that DAT, INST, LOC, and GEN case on subjects is not idiosyncratic in Hindi. The semantic regularity underlying the association of indirect case to arguments has not been sufficiently explored in most syntactic theories. When faced with the grammatical subjecthood of nominals with indirect case, a general tendency has been to view the case of these subjects in all instances as being idiosyncratically assigned by the verb. Hence the terminology of irregular or "quirky" case subjects. In GB, for instance, indirect case subjects are acknowledged to bear "inherent" case (Chomsky (1986), Belletti and Rizzi (1988)). But other than pointing out that inherent case is assigned at d-structure, and associated with theta marking (Chomsky (1986:193)), there has been no systematic effort to relate meanings and case. I will argue that there are regular correspondences between case features, case clitics, and case meanings.1 A systematic study of these correspondences involves an investigation of the configurations in SEM STR, and the semantic fields in which they occur. Observe that in (1), we have examples of SUBJs associated with all the case features in Hindi except ACC. Recall that in some dialects of Hindi, the ACC case on logical objects is preserved under passivization (§4.4). In these dialects, in other words, a grammatical subject may be in ACC case. Now, even in the dialects in which ACC case is not preserved under passivization, there is a construction in which ACC case on the L-OBJ cooccurs with a verb form that appears to be the passive. I will argue in §7.5. that this construction is not the passive, that the ACC nominal is in fact the grammatical object, and that the construction does not have a grammatical subject. This

1 Saksena (1982) argues in a similar manner for systematic correspondences between meanings and some of the case markings in Hindi. See also Schwartz (1988). See Watkins (19..), and also Nikiforidou (1989), for the semantic regularities of GEN case in a number of Indo European languages.

144 / Arguments in Hindi

conclusion conflicts with the widely held assumption that every clause must have a SUBJ (§3: (39)). The account of the association of case in this chapter provides justification for the separation of the levels of SEM STR and ARG STR. Crucially, the analysis of case alternations on subjects in Hindi provides evidence for multiple associations between the two levels, and for an entity in SEM STR to be unassociated with ARG STR. It is not clear how corresponding analyses can be constructed within a theory that conflates the two levels of structure, as is done for instance in traditional thematic role representations.

7.1. Dative Subjects A phenomenon that has triggered a great deal of linguistic research is the 'dative subject' construction, also known as 'experiencer subject' or 'inversion' construction. This construction has drawn considerable attention in work on South Asian languages since Emeneau (1956). 7.1.1. Dative Logical Subjects 7.1.1.1. The dative subject construction in Hindi has been studied in some detail in Bahl (1974), Shapiro (1974), Hook (1976), Kachru (1966; 1970), Davison (1969), Kachru and Bhatia (1975), and Verma (1976), among others. The L-SUBJ in this construction is associated with DAT case. Examples are given in (2): (2) a.

tu‹aar-ko caand Tushar-D moon-N Tushar saw the moon.

dik¥aa. see/become visible-PERF (Lit: To Tushar the moon appeared.)

b.

tu‹aar-ko kitaab milii. Tushar-D book- N receive-PERF Tushar received a book.

c.

tu‹aar-ko vah kahaanii yaad aayii. Tushar-D that-N story-N memory-N come-PERF Tushar remembered that story. (Lit.: To Tushar the memory of that story came.)

d.

tu‹aar-ko Tushar-D

k¥ußii happiness-N

huii. happen-PERF

Indirect Case on Subjects / 145

Tushar became happy.

(Lit.: To Tushar happiness happened.)

e.

tu‹aar-ko cuhee-se ±ar lagtaa hai. Tushar-D mouse-I fear-N be struck-IMPERF be-PR Tushar is afraid of mice. (Lit.: To Tushar fear of mice strikes.)

f.

tu‹aar-ko miÆ¥aaii k¥aanii hai. Tushar-D sweets-N eat-NF be-PR Tushar wants to eat sweets. (Lit.: To Tushar is the urge to eat sweets.)

g.

tu‹aar-ko davaaii piinii pa±ii. Tushar-D medicine-N drink-NF fall-PERF Tushar had to drink the medicine. (Lit.: To Tushar fell the obligation to drink the medicine.)

DAT case on the L-SUBJ may be induced by any one of three types of predicates. The first is a small set of simple verbs, as in (2a, b). Belonging to the second type are noun+verb complex predicates, as in (2c-e). A third source of DAT case on the L-SUBJ are modality meanings such as urge or inner compulsion, as in (2f), and obligation as in (2g), derived from complex verbals involving auxiliaries. 7.1.1.2. The term 'experiencer subject' often used to refer to DAT SUBJs suggests that there is an exclusive association between the semantic notion experiencer and DAT case. However, as in other South Asian languages, the facts of Hindi show that the semantic basis for DAT case cannot be reduced to the notion experiencer. Take (2b): the subject of the simple verb "receive" cannot be called an experiencer. Yet, it has DAT case. Likewise, take (3a-d), which contrast with (2a), (2c), (2d), and (2e) respectively in meaning but not in "theta roles". The subject in each of these is clearly an experiencer, and yet, is in NOM or ERG case: (3) a.

tu‹aar-ne caand dek¥aa. Tushar-E moon-N see/look at-PERF Tushar saw the moon.

b.

tu‹aar-ne vah kahaanii Tushar-E that-N story-N Tushar remembered that story.

c.

tu‹aar k¥uß huaa. Tushar-N happy become-PERF

yaad kii. memory-N do-PERF

146 / Arguments in Hindi

Tushar became happy. d.

tu‹aar cuhee-se ±artaa hai. Tushar-E mouse-i fear-IMPERF be-PR Tushar fears mice.

Therefore, the notion experiencer is neither sufficient nor necessary for the characterization of the semantics associated with DAT case. 7.1.1.3. It has been proposed for other South Asian languages such as Malayalam (Mohanan and Mohanan (1988)) and Marathi (Pandharipande (1988)) that DAT case is associated with meanings that can be reduced to two abstract notions, which they refer to as GOAL and POSSESSION. These meanings may co-exist on a single ARG with meanings such as experience, as in (2d), and even agency, as in (2f, g). The meaning of GOAL is most transparent in the sentence in (2b): the L-SUBJ in (2b) is a "recipient". The semantic extensions that are encompassed by the notion GOAL become clear from the contrast between (2) and (3). First take (2c) and (3b). In both, the predicate involves remembering, and the L-SUBJ is the one who remembers. However, in the former, the remembrance comes to Tushar, while in the latter, the act of remembering proceeds from Tushar. Likewise, in both (2d) and (3c), Tushar experiences happiness. However, in the former, happiness moves towards Tushar; in the latter, Tushar moves towards the state of being happy.2 Similar observations apply to the pairs in (2a) and (3a), and (2e) and (3d). Extensive arguments for the relevance of abstract GOAL and POSSESSION for dative subjects in Malayalam and Marathi have been given in Mohanan and Mohanan (1988) and Pandharipande (1988). I assume the essential correctness of their position for Hindi dative subjects as well, although perhaps with minor variations in detail.3 The simplified semantic representations in (4a) for the sentence in (2d), and that in (4b) for the sentence in (3c), appeal to the notion of GOAL: 2 A syntactic difference between (2c, d) on the one hand, and (3b, c) on the other, is that the latter can be embedded in a clause containing the predicate koßiß kar "try", but not the former. 3 In Bangla, an Indo Aryan language, the L-SUBJ of this construction is often marked with the GEN case inflection. In other words, the GEN that marks possessors is extended to mark goals as well. In Malayalam, on the other hand, Mohanan and Mohanan argue that the DAT that marks goals is extended t o possessors as well. In this respect, Hindi appears to be closer to Malayalam than to Bangla.

Indirect Case on Subjects / 147

148 / Arguments in Hindi

(4) a.

[

Y

MOVES

TO X

]

TO Y

] [STATE]

HAPPINESS

b.

[

X

MOVES

___________ [ X HAPPY ] The entity X in the semantic configuration [MOVE TO X ] is what we call GOAL. X in both (4a) and (4b) is the experiencer of happiness. However, the ARG associated with X receives DAT case in (4a), but not in (4b). The adjective in (2c) associated with the entity Y in (4b) does not get DAT case because it is not an ARG.4 Within the conception wherein ARG-GF association and ARG-CASE association are governed by independent but interacting principles, it is possible to account for a DAT SUBJ without any special stipulation. The L-SUBJ is associated with SUBJ by the universal default principle in (§3: (43a)), repeated as (5):5 (5)

L-SUBJ

ARG STR

SUBJ

GF STR

!

The L-SUBJ is associated with DAT case by the general principle of association of DAT case to GOALs, independently of their GF. The principle of DAT case association given in Mohanan and Mohanan (1988) is formulated as (6): (6)

GOAL

SEM STR

ARG

ARG STR

DAT

GF STR

!

4 We will return in Chapter 8 to the mapping between the SEM STR representations such as in (4a, b) and their ARG STR. 5 Recall that in the association principles, the convention of solid lines and dotted lines is used to indicate the premise and consequence respectively of an ifthen conditional.

Indirect Case on Subjects / 149

(6) applies not only to the goal subjects in (2) but also to the goal objects of give -type verbs. The combination of (5) and (6) yields a DAT SUBJ. 7.1.1.4. The presence of DAT case on the L-SUBJ in examples such as (2) has been accounted for in the literature largely in terms of two types of analyses. Both types of analyses make predicate-particular stipulations either on the case or the GF of the L-SUBJ. One type of analysis treats the nominal as a grammatical subject by general principles of function assignment that assigns grammatical subjecthood to logical subjects. In such an analysis, quirky or idiosyncratic DAT case is assigned to the subject by stipulation (Kachru et al (1976), Verma (1976), Mohanan (1982a, 1983)). The assumption implicit in these analyses is that dative case on the subject is not predictable. In contrast, the principle of DAT case association in (6) claims that the distribution of DAT case is predictable both on subjects and non-subjects. In another type of analysis (Sridhar (1976a, b), Perlmutter (1983; 1990), Klaiman (1988), and much of the work on this subject within RG), the nominal gets dative case by general principles of case assignment. In these analyses, it is generally assumed that dative case can be assigned only to the indirect object. Therefore, the logical subject that receives dative case is associated with the indirect object by a special rule of "inversion". The evidence generally adduced in favour of the "inversion" analysis is that the argument bears dative case, the verb cannot agree with it, and that the construction, even when diadic, does not passivize. These, however, do not constitute valid arguments. There is no a priori reason to believe that grammatical subjects cannot be in dative case. Given that a verb does not agree with an ERG nominal which is uncontroversially a subject, and that it can agree with a NOM object, agreement is neither necessary nor sufficient for subjecthood (§5.1.). As for passives, the independently required semantic condition that only verbs with volitional subjects can be passivized (Pandharipande (1981a)) is sufficient to account for the non-passivizability of the semantically conditioned dative subjects. We will shortly see persuasive evidence (§7.2.) for the subjecthood of these dative nominals from the facts of reflexives, subject obviation, control, and gapping. I therefore assume that the inversion analysis cannot be maintained for Hindi, and perhaps for South Asian languages in general. Our analysis is applicable to a large body of data, including not only dative but also instrumental, locative, and genitive subjects, all of which have a simple account within the view proposed here.

150 / Arguments in Hindi

7.1.1.5. Recall from Chapter 4 that the direct case features on a SUBJ are ERG, GEN and NOM (§4:(27)-(28), (30)). Recall also that indirect case association preempts direct case association. Therefore the default principles of direct case association do not apply to the L-SUBJs in (2). Recall also that a verb that takes a NOM or ERG SUBJ in the finite clause takes a GEN SUBJ in the non-finite clause (§4.3.2). The preempting of direct case by indirect case obtains in non-finite clauses as well. The predicate in (7a) takes a NOM SUBJ, and the one in (8a), a DAT SUBJ: (7) a.

yah baccaa this child-N This child is ill.

bimaar ill

hai. be-PR

b.

[ is bacce-kaa bimaar honaa ] saad¥aara” nahññ this child-G ill be-NF ordinary not be-PR It is not normal for this child to be ill.

a.

isbacce-ko buk¥aar this child-D fever-N This child has fever.

hai.

(8) hai. be-PR

b. ? [ is bacce-ko buk¥aar honaa ] saad¥aara” nahññ this child-D fever-N be-NF ordinary not be-PR It is not normal for this child to have fever.

hai.

c. * [ is this

hai.

bacce-kaa buk¥aar child-G fever-N

honaa ] saad¥aara” nahññ be-NF ordinary not be-PR

In (7b), the non-finite SUBJ receives the direct GEN case, as expected. Indirect case on the embedded SUBJ in (8b) is somewhat awkward. But (8c) shows that direct case is impossible in the embedded clause if the embedded predicate requires indirect case. The facts of case association in examples involving the interaction of main verbs and modals (e.g. (2f, g)) provides additional substance to the idea that indirect case association preempts direct case association. The main verbs k¥aa "eat" ((2f)) and pii "drink" ((2g)), when they are on their own, take ERG or NOM case on the SUBJ, depending on whether or not the verb is in PERF aspect. When the verb is combined with an indirect case inducing modal, however, the subject takes indirect case, as shown by (2f, g).

Indirect Case on Subjects / 151

Now, one might seek an explanation for these facts by assuming that (i) the main verb + modal construction is syntactically a biclausal structure, with the modal as the predicate of the matrix clause, and (ii) the case assigned by the modal is what appears on the overt matrix subject. This analysis, however, cannot be correct, because when the main verb requires indirect case and the modal does not, the subject bears the indirect case required by the main verb. In (9), the modality of possibility, which has no indirect case requirements, is combined with the two predicates in (7a) and (8a), which otherwise take a NOM and DAT SUBJ respectively: (9) a.

baccaa bimaar h o saktaa hai. child-N ill be able-IMPERF be-PR The child could fall ill (e.g. from playing in the rain).

b.

bacce-ko buk¥aar ho saktaa hai. child-D fever-N be able-IMPERF be-PR The child could get a fever (e.g. from playing in the rain).

The case of the SUBJ in (9a) and (9b) depends on the main predicates bimaar ho "be ill" and buk¥aar ho "have fever", not the modal sak "be possible". Thus, when modality meanings create a conflict between direct case and indirect case on an argument, the conflict is resolved in favour of indirect case, whether the meaning associated with the indirect case is part of the main predicate or the modal. These syntactically monoclausal main verb+modal sequences contrast with truly biclausal verb+verb sequences that involve obligatory control of the embedded clause. Such a biclausal sequence does not allow in the embedded clause a predicate that requires a DAT SUBJ: (10) a.

baccaa [ ——— bimaar h o kar ] child-N ill be do Having been ill, the child became weak.

b. * baccaa child-N

[ ——— buk¥aar fever

ho be

kar do

]

kamzoor pa± gayaa. weak fall go-PERF kamzoor pa± gayaa. weak fall go-PERF

(11) a.

baccaa [ ——— bimaar honaa ] caahtaa child-N ill be-NFdesire-IMPERF The child wishes to be ill.

hai. be-PR

152 / Arguments in Hindi

b. * baccaa [ ——— buk¥aar honaa ] caahtaa child-N fever-N be-NFdesire-IMPERF

hai. be-PR

(10b) and (11b) show that a DAT SUBJ cannot be a controllee.6 Therefore, the analysis of (9a) and (9b) could not be in terms of a biclausal structure involving control, because then (9b) would require a DAT SUBJ to be a controllee. One might then entertain a possible analysis in terms of a biclausal structure without control. That is, the subjects of the modal sak , namely, baccaa biimar ho in (9a) and bacce-ko buk¥aar ho in (9b), are non-finite clauses. If so, the matrix verb does not influence the case on the subject of bimaar ho and buk¥aar ho. This analysis could not be correct either. As we saw in (7), the direct case of the SUBJ in a nonfinite clause in Hindi is GEN, not NOM as in (9a). Therefore, the sentences involving a main verb and a modal in (9) must be syntactically monoclausal. If so, the assumption that indirect case preempts direct case is essential for their analysis. 7.1.2. The Grammatical Subjecthood of D AT L- SUBJ s We now turn to the question of whether the L-SUBJs in (2a-g) are SUBJs. Recall that the antecedent of the reflexive apnaa must be either an L-SUBJ or a SUBJ (§6: (38a)).7 Consider the use of this reflexive in (12): (12) a.

b.

vijay-ko apnii k¥i±kii-se caand dik¥aa. Vijay-D self-G window-I moon-N Vijayi saw the moon from self's i/*j window. vijay-ko kitaab apnee g¥ar-mã Vijay-D book-N self-G house-L Vijayi found the book in self's i/*j house.

appear-PERF

milii. find-PERF

The only eligible antecedent of apnaa in (12) is Vijay, the L-SUBJ in both sentences. Now, if it is not also the SUBJ, apnaa should have another possible antecedent in the sentence. The DAT nominal being the only possible antecedent of apnaa points to two alternative conclusions: either the L-

6 (10b) and (11b) are ungrammatical with a DAT SUBJ as well, because the matrix predicate takes a NOM SUBJ. 7 The numbers used to refer to the diagnostics are those in the summary of the diagnostics in Chapter 6.

Indirect Case on Subjects / 153

SUBJ is also the SUBJ, or these sentences have no SUBJ. In order to choose

between these two conclusions, we must appeal to the other diagnostics. A pronoun, in contrast to the reflexive, cannot be coreferent with the SUBJ of its minimal finite clause (§6:(38b)). Consider the coreference facts in (13), where the reflexives in (12) are replaced with pronouns: (13) a.

b.

vijay-ko uskii k¥i±kii-se caand dik¥aa. Vijay-D pron-G window-I moon-N Vijayi saw the moon from his/her j/*i window. vijay-ko kitaab uske g¥ar-mã Vijay-D book-N pron-G home-L Vijayi found the book at his j/*i home.

appear-PERF

milii. find-PERF

Neither in (13a) nor in (13b) can the pronoun be coreferent with the DAT nominal. These facts of pronominal coreference in (13) follow from principle (§6:(16)) if we assume that the dative L-SUBJ in these sentences is the SUBJ as well. The assumption that these DAT nominals are SUBJs is further confirmed by the fact that they can be the controllers of participial adjunct clauses with obligatory control sites, a privilege available only to a SUBJ (§6:(38c)): (14) a.

vijay-ko apne b¥aayii-se [ —— muskuraate hue ] milnaa Vijay-D self-G brother-I smile-IMPERF be-NF meet-NF Vijayi must meet his brotherj while ——— i/*j smiling.

b.

ravii-ko [ —— kitaab paa kar ] ba±ii k¥ußii huii. Ravi-D book-N get-NF do-NF much joyhappen-PERF ——— i having got the book, Ravii was very happy.

hai. be-PR

Once again, the DAT L-SUBJs behave like SUBJs. Finally, recall that in coordinate structures, a gapped element must be identical in case as well as function to its gapper. The DAT nominals in sentences like those in (2) can gap one another. This means that they must be identical in GF:

154 / Arguments in Hindi

(15) ravii-ko

niinaa dik¥ii, aur Nina-N appear-PERF and usko / ——— ba±ii k¥ußii pron-D much joy-N Ravii saw Nina and he / —— i was very happy. Ravi-D

huii. happen-PERF

Naturally, given the requirement of case identity, DAT nominals can gap or be gapped only by other DAT nominals, and not by NOM or ERG subjects. The gapping in (15) does not provide any clues as to the GF of the DAT nominals. However, coordination with the active-passive alternants of ditransitive constructions provides support for the subjecthood of these DAT nominals. If the DAT nominals in the two conjuncts in (15) are subjects, they should not be able to gap nor be gapped by the DAT objects of ditransitive verbs. This prediction is borne out in (16): (16) a.

niinaa-ko ±igrii milii, aur hed maastar-ne Nina-D degree-N receive-PERF and head master-E us-ko/ * ——— skuul-mã naukrii dii. pron-D school-L job-N give-PERF Ninai received a degree, and the Head Master gave heri / * ——— a job in the

school. b.

ravii-ne Ravi-E

niinaa-ko gu±iyaa dii, aur Nina-D doll-N give-PERF and us-ko/ * ——— ba±ii k¥ußii huii. pron-D much joy-N happen-PERF Ravi gave Ninai a doll, and shei / * ——— was very happy.

However, when the DAT nominal in a ditransitive construction is the SUBJ in the passive, it can gap and be gapped by a DAT nominal: (17) a.

school.

niinaa-ko ±igrii milii, aur us-ko/ ——— Nina-D degree-N receive-PERF and pron-D skuul-mã naukrii dii gaii. school-L job-N send-PERF go-PERF Ninai received a degree, and shei / ——— i was given a job in the

Indirect Case on Subjects / 155

b.

niinaa-ko (ravii-se) gu±iyaa dii gaii aur Nina-D Ravi-I doll-N give-PERF go-PERF and us-ko/ ——— ba±ii k¥ußii huii. pron-D much joy-N happen-PERF Ninai was given a doll (by Ravi), and shei / ——— i was very happy.

These facts follow from principle (4:(34)), if (i) the DAT L-SUBJs of the predicates "receive" and "be happy" are SUBJ, (ii) the DAT goal of the verb "give" is a SUBJ in (17) but not in (16). Thus, the DAT L-SUBJs in (2) are legitimate antecedents of reflexives ((12)); they cannot be coreferent with pronouns within the minimal finite clause ((13)); they can be controllers of participial clauses ((14)); and they can be gapped by DAT SUBJs of the ditransitive passive ((17)) but not by the DAT OBJs of their active counterparts ((16)). We must therefore conclude that they are grammatical subjects. In sum, in the fully specified representation of a dative subject construction, the highest logical ARG at ARG STR is associated with a GOAL in SEM STR, and with the SUBJ function and DAT case in GF STR.

7.2. Instrumental Subjects What I refer to as the "instrumental subject" construction in Hindi is a far less explored and less well understood construction than the dative subject construction. In this construction, an otherwise NOM or ERG L-SUBJ is associated with the INST case marking -se in a specific semantic context. The verb morphology is identical to that of the "passive" construction. In what follows, I will argue that the INST nominal in this construction is a SUBJ, and that the construction is quite distinct from the passive in Hindi, although they bear a remarkable surface resemblance. 7.2.1. The Instrumental Subject Construction 7.2.1.1. For the reader to get a feel for the semantics of this construction, let me begin by giving pair-wise alternants with NOM/ERG and INST subjects in Hindi: (18) a.

umaa-ne k¥aÆÆaa aam Uma-E sour mango-N Uma ate a sour mango.

k¥aayaa. eat-PERF

156 / Arguments in Hindi

b.

umaa-se k¥aÆÆaa aam Uma-I sour mango-N eat-PERF Uma couldn't eat a sour mango.

a.

mohan uƥ rahaa hai. Mohan-N rise-NF stay-PERF Mohan is getting up.

k¥aayaa not

nahññ go-PERF

gayaa.

(19)

b.

be-PR

mohan-se uÆ¥aa nahññ jaa rahaa hai. Mohan-I rise-PERF not go-NF stay-PERF Mohan can't get up (finds himself unable to get up).

be-PR

(20) a.

kyaa baalak haat¥ii-ko nocegaa? what boy-N elephant-A pinch-FU Will the boy pinch the elephant?

b.

kyaa baalak-se haat¥ii-ko nocaa what boy-Ielephant-A pinch-PERF go-FU Will the boy be able to pinch the elephant?

jaaegaa?

7.2.1.2. I would first like to emphasize the surface similarities of the INST SUBJ construction and the passive. To begin with, the verb forms in the two constructions are identical: they involve adding the auxiliary jaa to the perfective form of the main verb. The auxiliary, which is homophonous with the main verb jaa 'go', carries the tense and aspect morphology. Secondly, the L-SUBJ of passives, like those in (18)-(20), is expressed with the INST case marking -se. The INST SUBJ construction may therefore appear at first glance to be just a passive. In fact, it has been assumed in the literature on Hindi to be identical to the passive for the purposes of syntax (Pandharipande (1979; 1981a; 1981b), Davison (1982)). The INST nominal in examples like (18b), (19b), and (20b) has been referred to as a passive agent (Kachru (1980)).8 Let me now draw attention to the significant differences between the INST SUBJ construction and the passive. Take the contrast illustrated in (21). (21b) is the INST SUBJ alternant of (21a), and (21c), the passive. 8 Rosen and Wali (1989) in an RG analysis of the two constructions i n Marathi distinguish them as "Regular Passive" and "Capability Passive", and argue that they are syntactically distinct structures. While they conclude that the INST nominal in the Capability Passive in Marathi is a chomeur, I argue that the INST nominal in the construction in Hindi is a grammatical subject.

Indirect Case on Subjects / 157

(21) a.

ravii-ne raam-ko Ravi-E Ram-A not Ravi didn't beat Ram.

nahññ piiÆaa. beat-PERF

b.

ravii-se raam-ko piiÆaa nahññ Ravi-I Ram-A beat-PERF not go-PERF Ravi couldn't (bring himself to) beat Ram.

c.

raam (ravii-se) piiÆaa Ram-N Ravi-I beat-PERF Ram wasn't beaten (by Ravi).

gayaa.

nahññ gayaa. not go-PERF

Consider first the semantics of the two constructions. Firstly, a peculiar property of the INST SUBJ construction is that it most characteristically involves a negative ((18b, 19b, 21b)), or a question ((20b)). An appropriate context may be contrived for the use of this construction in a declarative statement; however, it would be rather infelicitous. The passive is not subject to such a restriction. Secondly, the INST SUBJ construction is systematically associated with the meaning of capability: it is used to negate or to question (or express doubt about) the capability of the subject to be the instrument of the action, which Pandharipande (1979) calls "internally determined capability".9 The passive, on the other hand, bears no such special meaning.10 Corresponding to these semantic differences, we find several properties of surface realization that distinguish the two constructions. Firstly, the ERG SUBJ of (21a) has INST case in both (21b) and (21c). However, the ACC OBJ of (21a) retains its ACC case in (21b), but is NOM in (21c). Recall that ACC case is associated with an ARG that is both an L-OBJ and an OBJ (§4:(64)). Now, the ACC OBJ of an active sentence is invariantly NOM in the passive, because it is not an OBJ. But in the INST SUBJ construction, it retains ACC case; it must therefore be the OBJ. It must be noted that if the object of the active sentence is inanimate, and hence NOM, there may be no distinction on the surface between the passive and the INST SUBJ construction. Such a sentence would then be 9 Hence the terminology of Capability Passive in Rosen and Wali. 10 Because of identity in morphology, and the consequent failure t o distinguish the two constructions, it has been argued that the auxiliary jaa 'go' in the passive must be treated as a modal with "capabilitative meaning" (Saksena (1977)). One of the problems of such a treatment that Pandharipande (1979) points out is that if the agent is deleted, the capabilitative meaning is lost.

158 / Arguments in Hindi

ambiguous between the two structures. Similarly, in the dialects that preserve the ACC case of objects under passivization, this surface distinction is obliterated: the object in (21a) would be ACC in both (21b) and (21c), simply by virtue of being the L-OBJ. Secondly, the "demoted" agent in the INST case is optional in the passive, and is generally omitted ((21c)). In contrast, the INST nominal is obligatorily expressed in the INST SUBJ construction ((21b)). If the INST argument in (21b) is omitted, although the sentence is not ungrammatical, the meaning of internally determined capability associated with the INST SUBJ construction is entirely lost.11 Yet another difference between the passive and the INST SUBJ construction involves word order. In the former ((21c)), the INST argument is free to either follow or precede the NOM argument. In contrast, the INST argument in the latter ((21b)) must obligatorily precede the ACC argument, unless it is postverbal.12 A fourth difference is that the INST -se in the passive may be replaced by the postposition -ke dvaaraa 'through', but not in the INST SUBJ construction. Now, -ke dvaaraa generally appears only on adjuncts. Under the assumptions that (i) N-se is an NP, while N-ke dvaaraa is a PP , and (ii) only an NP, and not a PP , can be a SUBJ, the inability of N-ke dvaaraa to appear in the INST SUBJ construction follows from the assumption that the L-SUBJ in this construction is a SUBJ. Finally, while a passive requires that the verb be transitive, the INST SUBJ has no such requirement. The INST SUBJ construction in (19b) has no passive counterpart. Given these differences, we must conclude that the INST SUBJ construction and the passive in Hindi are two syntactically distinct constructions. In spite of the clear differences, the two constructions can easily merge on the surface. Thus, when the INST L-SUBJ is overtly expressed in the passive, is initial, and the L-OBJ is inanimate, and therefore NOM, the two constructions are identical on the surface. In addition, in the ACC preserving dialects, even the surface distinction between the NOM LOBJ of the passive and the ACC L-OBJ of the INST SUBJ construction, is absent. The possibility of such convergence appears to have prevented researchers from noticing the differences. 11 The resultant construction, distinct from both the passive and the INST SUBJ construction, is discussed in §7.5. 12 An account of this difference in word order is given in T. Mohanan (1989) in terms of the "freezing" of canonical word order in the INST SUBJ construction.

Indirect Case on Subjects / 159

A consequence of not distinguishing these two apparently similar constructions has been that grammatical subjecthood has been viewed as being shared by the "demoted subject" and the "promoted object" of the passive. Factoring apart the two constructions, which we have just seen to be necessary on independent grounds, would avoid an analysis involving shared grammatical subjecthood: the L-SUBJ is not the SUBJ in the passive, but it is the SUBJ in the INST SUBJ construction.13 7.2.1.3. The modality meaning of capability associated with the INST SUBJ construction is the property characterized by Pandharipande (1979) as "internally determined capability". The semantic predicate that bears this modality meaning I abbreviate as I(NTERNALLY)-ABLE. Verbs such as k¥ol "open (tr.), muskuraa "smile", cal "walk", ro "cry", de "give", dek¥ "see", and siik¥ "learn" freely appear in the INST SUBJ construction, in contrast to verbs like gir "fall", ±uub "drown", ±ar "dread/fear", pahõc "reach", dik¥ "appear", and sik¥ "become learnt": (22) a.

raam-ne darvaazaa Ram-E door-N Ram opened the door.

k¥olaa. open-PERF

13 Another nominal that has been recognized in the literature as an instrumental subject is one that occurs as the instrument of action in inchoatives (Verma (1976), Kachru (1988)), as illustrated in (b) below: (i) a. raam-ne ßiißaa to± diyaa. Ram-E bottle-N break(trans.)-NF give-PA Ram broke the bottle. b . raam-se ßiißaa ÆuuÆ gayaa. Ram-I bottle-N break (intr.)-NF go-PA The bottle broke (from/through/because of Ram). (ii) a. siitaa-ne galtii kii. Sita-E mistake-N do-PA Sita committed a mistake. b . siitaa-se galtii ho gayii. Sita-I mistake-N be-NF go-PA A mistake occurred (from/through/by Sita). This construction does not share any of the semantic peculiarities of what we have termed the INST SUBJ construction. It is not associated with the meaning of capability. Besides, it is perfectly felicitous as a declarative sentence. I believe that it is distinct from the INST SUBJ construction syntactically as well. An analysis of the construction, however, requires further investigation.

160 / Arguments in Hindi

b.

raam-se darvaazaa k¥olaa nahññ gayaa. Ram-I door-N open-PERFnot go-PERF Ram couldn't (bring himself to) open the door.

(23) a.

raam muskuraayaa. Ram-N smile-PERF Ram smiled.

b.

raam-se muskuraayaa nahññ gayaa. Ram-I smile-PERF not go-PERF Ram couldn't (bring himself to) smile.

a.

raam giraa. Ram-N fall-PERF Ram fell.

(24)

b. # raam-se giraa nahññ gayaa. Ram-I fall-PERF not go-PERF Ram couldn't (bring himself to) fall.

(25) a.

raam g¥ar pahõcaa. Ram-N house-L reach-PERF Ram reached home.

b. # raam-se g¥ar pahõcaa nahññ Ram-I house-L reach -PERF not Ram couldn't (bring himself to) reach home.

gayaa. go-PERF

That the ability involved in the INST SUBJ construction is indeed "internally determined capability" becomes clear from the contrast between (25b) and (26). The interpretation of (25b) is that Ram could not bring himself to reach home, which is odd at best. (26) also involves the modality meaning of ability. However, the ability is not necessarily internally determined: it could be dependent on external forces, for instance, his car breaking down: (26)

raam g¥ar pahõc nahññ sakaa. Ram-N house-L reach -NF not be able-PERF Ram couldn't reach home.

Indirect Case on Subjects / 161

An independent test for internal ability involves the formation of imperatives carrying the meaning of request/command (as opposed to a blessing or a wish, for instance). All and only those events that are compatible with the modal meaning I-ABLE yield imperatives: (27) a.

darvaazaa k¥olo. door-N open-IMPERATIVE Open the door!

b.

muskuraao. smile-IMPERATIVE Smile!

c. # giro. fall-IMPERATIVE Fall! d. # g¥ar pahõco. house-L reach-IMPERATIVE Reach home!

Now consider the following imperatives, which, although infelicitous in the positive, are perfectly acceptable when negative: (28) a. ?? g¥abraao! worry-IMPERATIVE Worry! b.

mat g¥abraao! don't worry-IMPERATIVE Don't worry!

c. ?? ±aro! fear-IMPERATIVE Be afraid! d.

mat ±aro! don't fear-IMPERATIVE Don't be afraid!

(28b, d), unlike (28a, c) are acceptable: preventing the event of worrying and fearing is within the internal ability of the participant, even though bringing about the event of worrying and fearing is not. The assumption that what

162 / Arguments in Hindi

underlies both the INST SUBJ construction and imperatives is the modal meaning of I-ABLE predicts that the contrasts in (28) will be found in the INST SUBJ construction as well. This prediction is borne out, as shown by (29): (29) a. ?? raam-se g¥abraayaa nahññ gayaa. Ram-I worry-PERF not go-PERF Ram couldn't (bring himself to) worry. b.

raam-se na g¥abraayaanahññ gayaa. Ram-I not worry-PERF not go-PERF Ram couldn't (bring himself not to) worry. (= Ram couldn't help worrying.)

c. ?? raam-se ±araa nahññ gayaa. Ram-I fear-PERF not go-PERF Ram couldn't (bring himself to) be afraid. d.

raam-se na ±ara nahññ gayaa. Ram-I not fear-PERF not go-PERF Ram couldn't (bring himself not to) be afraid. (= Ram couldn't help being afraid.)

In short, the modality meaning of I-ABLE, which induces INST case on its participant, can be superimposed on any event that is semantically compatible with the meaning of internally determined capability. Recall that ERG CASE is associated with the semantic configuration of CONSCIOUS CHOICE. We have just said that the INST SUBJ is associated with I-ABLE. Now, making a conscious choice to bring about or not bring about an event requires internally determined capability, although having such capability does not entail the availability of choice. Given that CONSCIOUS CHOICE entails I-ABLE, we conclude that all predicates that allow ERG SUBJs must freely participate in the INST SUBJ construction. As far as I know, this is correct: predicates that allow ERG SUBJs form a proper subset of those that allow the INST SUBJ. Now consider the examples in (30). (30a, c) are DAT SUBJ constructions, and (30b, d) are their INST SUBJ counterparts. (30b, d) are ungrammatical, not merely odd like (29b, d):

Indirect Case on Subjects / 163

(30) a.

raam-ko g¥ar Ram-D house-N Ram saw the house.

dik¥aa. appear-PERF (Lit.: To Ram appeared the house.)

b. * raam-ko/-se g¥ar dik¥aa nahññ gayaa. Ram-D/I house-N appear-PERF not go-PERF c.

raam-ko k¥at milaa. Ram-D letter-N receive-PERF Ram received a letter.

d. * raam-ko/-se k¥at milaa Ram-D/I letter-N

nahññ gayaa. receive-PERF not

go-PERF

In (30), the main verb requires DAT case, and the modality meaning of IABLE requires INST case. The two requirements are in conflict: a nominal cannot be both DAT and INST at the same time. Recall that in conflicts between direct and indirect case, the latter preempts the former: the syntactic representation settles on the indirect case. In a conflict between two indirect cases, on the other hand, there is no resolution of the conflict: the syntactic representation settles on two incompatible specifications which do not "unify". Therefore, the sentences in (30b, d) are not merely infelicitous but ungrammatical. In order to illustrate the representation of the INST SUBJ construction, I give in (31a) the structure of the active sentence in (21a), and in (31b) the structure of the INST SUBJ construction in (21b): (31) a.

[ X Y beat ] SEM STR ………………………………………………………………………….. ˘ ARG ARG ¿ A-PRED ARG STR ……\………….\………….\……………………………………… SUBJ OBJ GF-PRED GF STR ERG

ACC

………………………………………………………………………….. ravii-ne raam-ko nahññ piiÆaa WORD STRING

Ravi Ram didn't beat ………………………………………………………………………..… NP NP V GC STR S

164 / Arguments in Hindi

b.

[

X

Y

NOT

I-ABLE

________________________ [ X Y beat ] SEM STR ………………………………….\…………………………………….. ˘ ARG ARG ¿ A-PRED ARG STR ……\………….\………….………\………………………………. SUBJ OBJ GF-PRED GF STR ERG

ACC

………………………………………………………………………….. ravii-se raam-ko piiÆaa nahññ gayaa WORD STRING

Ravi Ram couldn't beat ………………………………………………………………………..… NP NP V GC STR S

In (31b), the participant of the semantic predicate I-ABLE merges on the LSUBJ with the highest participant of the embedded semantic predicate. The indirect case requirement of the modal meaning is satisfied, thus preempting direct case on the L-SUBJ. In sum, the L-SUBJ in the INST SUBJ construction is the SUBJ by the default principle §3:(43a). It gets INST case by a specific principle of INST case association, which might be formulated as in (32): (32)

[

X I-ABLE ARG

]

SEM STR14 ARG STR

! INST

GF STR

This principle of INST case association preempts both the language specific principle of ERG case association (§4:(27)) as well as the universal default principle of NOM case association (§4:(38)). This effect of preempting is the result of the general condition whereby principles of indirect case association preempt principles of direct case association (§4:(41)). This analysis hinges crucially on the separation of the levels of SEM STR and ARG STR as illustrated in (31). The "embedded" semantic clause ([X Y beat ]) in the SEM STR of (31) expresses the meaning that is common to the ERG SUBJ and INST SUBJ constructions (e.g. (21a) and (21b) respectively). This is also the meaning shared by a construction in which a modal14

We will see shortly, in §7.3.2., that this stipulation is in fact unnecessary.

Indirect Case on Subjects / 165

ity meaning induces DAT case on the agentive SUBJ, for example, in (2f, g). These patterns of semantic alternation corresponding to case alternation would be impossible to capture if we did not separate SEM STR and ARG STR, and allow multiple associations between them. Having seen that the INST SUBJ construction is quite different from the passive construction in semantics as well as in syntax, we must provide a representation of the passive such that it is contrasted with the representations in (31). A representation of the passive in (21c) is given in (33). We follow the notational convention outlined in §3.(44b). (33)

[ X Y beat ] SEM STR ………………………………………………………………………….. ˘ ARG ARG ¿ A-PRED ARG STR ……\………….\………….\……………………………………… ADJ SUBJ GF-PRED GF STR INST

NOM

………………………………………………………………………….. ravii-se raam piiÆaa nahññ gayaa WORD STRING

By Ravi Ram was not beaten ………………………………………………………………………..… NP NP V GC STR S

7.2.2. Their Grammatical Subjecthood We now turn our attention to the grammatical subjecthood of the INST ARGs in the INST SUBJ construction. First consider this ARG with respect to the reflexive apnaa. In the INST SUBJ construction in (34), apnaa takes the INST ARG as its antecedent, but not the ACC ARG. This contrasts with the situation in a passive, for instance, (§6:(3d)). (34) vijay-se ravii-ko apnii saikil-par biÆ¥aayaa nahññ gayaa. Vijay-I Ravi-A self-G bicycle-L sit-C -PERF not go-PERF Vijayj couldn't seat Ravii on self's j/*i bike.

166 / Arguments in Hindi

Like the DAT SUBJ, the INST L-SUBJ is the only eligible antecedent of the reflexive.15 Therefore, either it must be SUBJ, or the construction has no SUBJ (§6:(38a)). Let us consider the facts of pronominal coreference in this construction in (35). Once again, there is a sharp contrast between (35) below and the passive in (§6:(11d)). (35) vijay-se ravii-ko uskii saikil-par biÆ¥aayaa nahññ gayaa. Vijay-I Ravi-A pron-G bicycle-L sit-C -PERF not go-PERF Vijayj couldn't seat Ravii on his i/*j bike.

In the passive, a pronoun can be coreferent with the INST L-SUBJ, because it is not the SUBJ. In (35), the pronoun can be coreferent with the other ARG, but not with the INST L-SUBJ. Recall that a pronoun in Hindi cannot take a SUBJ antecedent in its minimal finite clause (§6:(38b)). The combination of the facts in (34)-(35) and the diagnostics (§6:(38a, b)) further suggests that the INST L-SUBJ in (35) is a SUBJ. By §6:(38c), a nominal that can control a participial adjunct clause with an obligatory control site must be a SUBJ. The subjecthood of the INST L-SUBJ is confirmed by the facts of control in participial adjunct clauses in (36) and (37). These facts once again contrast with those involving the passive, in §6:(19c) and §6:(20c). (36) ravii-se vijay-ko [ ——— muskuraate hue] piiÆaa nahññ gayaa. Ravi-I Vijay-A smile-IMPERF be-NF beat-PERF not go-PERF Ravii couldn't beat Vijayj while ——— i/*j smiling.

(37) ravii-se vijay-ko [ ——— darvaazaa k¥ol kar] piiÆaa nahññ gayaa. Ravi-I Vijay-A door-N open do beat-PERF not go-PERF Ravii couldn't beat Vijayj having ——— i/*j opened the door. (Lit: Ravi couldn't open the door and beat Vijay.)

15 For speakers of the ACC preserving dialects, (34) would be ambiguous. If (34) is interpreted as the passive, the ACC ARG will also be an eligible antecedent of the reflexive by virtue of being the SUBJ. This potential confusion must be borne in mind during the rest of the discussion.

Indirect Case on Subjects / 167

In both (36) and (37), the INST ARG is an eligible controller of the obligatory control site, and therefore, it must be a SUBJ. Recall that the INST SUBJ must obligatorily be expressed. This requirement is in conflict with its being an obligatory control site. No other ARG in the INST SUBJ construction can be obligatorily controlled either, since obligatory control requires subjecthood. In coordinate structures, recall that the gapper and the gapped element must be identical in case and GF (§6:(36)). Given the facts in (34)-(37), any nominal that can gap or be gapped by the INST nominal under discussion must be a SUBJ associated with INST case. In other words, we predict that the INST ARG in an INST SUBJ construction cannot be gapped by a NONSUBJ. This prediction is borne out in (38)-(40). The gapping relation with the INST SUBJ involves the demoted agent of a passive in (38), an oblique agent in a causative in (39), and an oblique source in (40): (38) a.

ravii raam-se piiÆaa gayaa aur Ravi-N Ram-I beat-PERF go-PERF and us-se / * ——— hãsaa nahññ gayaa. pron-I laugh-PERF not go-PERF Ravii was beaten by Ramj , and hei/j couldn't laugh.

b.

raam-se hãsaa nahññ gayaa aur Ram-I laugh-PERF not go-PERF and ravii us-se / * ——— piiÆaa gayaa. Ravi-N pron-I beat-PERF go-PERF Rami couldn't laugh, and Ravi was beaten by himi .

a.

raam-ne anil-se ravii-ko piÆvaayaa aur Ram-E Anil-I Ravi-A beat-C-PERF and us-se / * ——— hãsaa nahññ gayaa. pron-I laugh-PERF not go-PERF Rami made Anilk beat Ravij , and hei/j/k couldn't laugh.

b.

anil-se hãsaa nahññ gayaa aur Anil-I laugh-PERF not go-PERF and raam-ne us-se / * ——— ravii-ko piÆvaayaa. Ram-E pron-I Ravi-A beat-C-PERF Anili couldn't laugh, and Ram made himi beat Ravi.

(39)

(40)

168 / Arguments in Hindi

a.

anil-ne raam-se paisaa mãã˚gaa aur Anil-E Ram-I money-N ask for-PERF and us-se / * ——— diyaa nahññ gayaa. pron-I give-PERF not go-PERF Anil asked Rami for money, and hei couldn't give (it).

b.

raam-se anil-ko ±ããÆaa nahññ gayaa aur Ram-I Anil-A scold-PERF not go-PERF and anil-ne us-se / * ——— paisaa mãã˚gaa. Anil-E pron-I money-N ask for-PERF Rami couldn't scold Anil, and Anil asked himi for money.

Gapping is disallowed in (38)-(40) despite case identity. Given the evidence from the other diagnostics, we know that this is because of nonidentity of function: the INST nominal in the second conjunct in (38a), (39a) and (40a) and in the first conjunct in (38b), (39b) and (40b) is a SUBJ, unlike the other INST nominals in the sentences. To conclude, the L-SUBJ of the INST SUBJ construction is a SUBJ: it is an eligible antecedent of the reflexive apnaa ((34)), triggers subject obviation in pronouns ((35)), can control participial clauses ((36), (37)), and cannot be gapped by non-SUBJ functions with the same case marking ((38)-(40)). Finally, it is worth noting that the INST SUBJ construction is an areal property of South Asian languages, cutting across genetic differences. It has been attested in Marathi, Nepali, Panjabi, and Kashmiri (Pandharipande (1979; 1981a, b)). It is found also in Bangla, and Malayalam, a Dravidian language. It is, therefore, a promising area for further exploration into areal patterns of invariance and variability, like the DAT SUBJ construction.

7.3. Locative Subjects In addition to DAT and INST subjects in Hindi, Kachru (1969; 1980; 1988), McGregor (1972), and Pandharipande (1981b), among others, discuss nominals in LOC and GEN case that function as grammatical subjects. LOC case is signalled by the case clitics -mã "in", -par "at", and -ke paas "near". What follows is a discussion of constructions with subjects associated with LOC case. In order to understand the semantics of the various LOC SUBJs, it is essential that we briefly investigate the spatial semantics of the relevant case clitics. I shall therefore begin with a sketch of the atomic elements of meaning involved in the use of these clitics in their most tangible manifes-

Indirect Case on Subjects / 169

tations, namely, to express concrete spatial relations. LOC SUBJs, and as we will see, the DAT and INST SUBJs as well, are best understood as extensions of the spatial uses of case to non-spatial dimensions. After attempting to identify the semantic distribution of the LOC case clitics in Hindi in spatial and nonspatial relations, I argue that L-SUBJs with LOC case, like the DAT and INST SUBJs, must be analysed as grammatical subjects. 7.3.1.

Case and Spatial Relations

There has been a long-standing tradition of decomposing spatial relations expressed by cases and prepositions into their atomic elements (Gruber (1965), Jackendoff (1972) and subsequent works, Anderson (1977), Givón (1976; 1984), Talmy (1972; 1985), Croft (1986), Lakoff (1987)). The insight behind this decomposition of meaning, first demonstrated in detail in Gruber (1965), is that the semantics of motion and location provide a basis for extension to a range of "semantic fields". For our purpose of investigating indirect case subjects in Hindi, I adopt the atomic elements of spatial relations given in (41)-(42). The semantic configurations in (41) may intuitively be thought of as STATIC LOCATIONs, associated with the primitive predicate BE. Those in (42) express motion from or towards a location; they may be thought of as DYNAMIC LOCATIONs, associated with the predicate MOVE. (41) STATIC LOCATION a. b. c.

CONTAINMENT PROXIMITY

(42) DYNAMIC LOCATION a. b.

[ BE ]

CONTACT

SOURCE GOAL

[ AT X ] [ IN X ] [ NEAR X ]

[ MOVE ] [ FROM X ] [ TOWARDS X]

These constructs, alone and in combination, are illustrated with English prepositions in (43). The atomic elements in (43a) and (43e) are static locations. That in (43b) involves a movement, or dynamic location. Those in (43c, d, f) are movements that terminate in or originate from a static location. Together, [ MOVE TOWARDS x ] and [ BE AT x ], as in (43c), corresponds to [ TO x] in Jackendoff (1986).

170 / Arguments in Hindi

(43) a. b. c. d. e. f.

John remained at his post. at : [ j BE AT p ] John went towards the market. towards : [ j MOVE TOWARDS m ] John went to the market. to : [ j MOVE TOWARDS m] & ([ j BE AT m ]) John went into the room. into : [ j MOVE TOWARDS r] & [ j BE IN r ] John was in the market. in : [ j BE IN m] John came out of the market. out of : [ j BE IN m ] & [ j MOVE FROM m ]

Consider the meanings of case features in the Hindi examples in (44): (44) a.

raam apnii post-par Ram-N self-G post-L Ram stayed at his post.

rahaa. stay-PERF

b.

raam baazaar gayaa. Ram-N market-L go-PERF Ram went to the market.

c.

raam baazaar-mã t¥aa. Ram-N market-L be-PA Ram was in the market.

d.

raam baazaar-se niklaa. Ram-N market-I emerge-PERF Ram emerged from /came out of the market.

e.

raam baazaar-ke paas k¥el rahaa hai. Ram-N market-G-L play PROG be-PR Ram is playing near the market.

As a notation to distinguish between the different locatives in (44a-e), I will refer to them as LOC-par ((44a)), LOC-ko ((44b)), LOC-mã ((44c)), and LOC-paas ((44e)). Recall from Chapter 4 that a locative destination, for example baazaar "market" in (44b), is not overtly marked for case in most dialects of Hindi, although it is clear from modifier agreement that the nominal associated with the locative destination is in NONNOM case (§4.3.3.). In the dialects that do overtly mark locative destinations, the case

Indirect Case on Subjects / 171

marking used is -ko , the clitic associated with ACC and DAT case. Hence the label LOC-ko , even though it is not marked overtly with a case formative. I have so far assumed that the clitic -se signals what we have called INST case, even in examples such as (44d), because one of the most salient uses of this clitic is to mark instruments or tools of action. However, exactly as LOC-ko signals spatial destinations, it is possible to think of the -se in (44d) as LOC-se which signals spatial origins. Based on the examples in (44), we use the atomic elements of spatial relations in (41)-(42) to describe in (45) the semantics of LOC case in Hindi: (45) a. b. c. (44c)) d. (44a)) e. (44e))

LOC-se LOC-mã

m ] & [ r FROM m] m ] & ([ r AT / IN m ]) [ r IN m ]

LOC-par

[r

AT

LOC-paas

[r

NEAR

LOC -ko

[r [r

AT / IN

TOWARDS

p] m]

(= (44d)) (= (44b)) (= (= (=

For convenient reference, we might use the terminology of source, goal, etc., given in (41)-(42) to refer to the atomic constructs used to describe the concrete spatial meanings of case in (45). Thus, [FROM m] in (44d) describes a spatial SOURCE, [TOWARDS m] in (44b) a spatial GOAL, [IN m] in (44c) CONTAINMENT, [AT p] in (44a) CONTACT, and [NEAR m] in (44e), PROXIMITY. In the following section, we will see that it is when an L-SUBJ is associated with the semantic configurations in (45) in a non-spatial, abstract "semantic field" that we get grammatical subjects bearing the corresponding indirect case formatives. 7.3.2.

Semantic Fields

The notion of semantic fields is a traditional notion in theories of semantics (Lyons (1977) and the references therein). The extension of this notion to linguistic analysis by Gruber (1965), Jackendoff (1972, 1986), and others allows us to factor out the core meanings that are common to different dimensions of human experience. Consider the following English sentences: (46)

172 / Arguments in Hindi

a. b. c.

John flew from Boston to Chicago. John sang from dawn to dusk. John swung from deep depression to intense joy.

Indirect Case on Subjects / 173

It is well-known that temporal and locational uses of prepositions and case exhibit systematic correspondences (Anderson (1971), Clark (1978), Croft (1986)). It is not an accident that the preposition from in English, which refers to a spatial point of origin in (46a), also refers to the temporal point of origin in (46b). Likewise, the preposition to, which refers to a spatial point of termination in (46a), also refers to the temporal point of termination in (46b). In a language that has distinct forms for spatial origin and termination points, it is unlikely that the form used for temporal origin will merge with that for spatial termination point, and that for temporal termination point will merge with the one for spatial origin. These regularities can be expressed by separating the core semantic constructs such as those in (41)-(42) from the domain of experience, as is done in (47): (47) ((46a)) FIELD CONFIGURATION

from SPACE [ FROM X ] to SPACE [ TO X ]

((46b)) FIELD CONFIGURATION TIME TIME

[ FROM X ] [ TO X]

The semantic configurations expressed by the prepositions in (46a) and (46b) are identical. The difference between them lies in their semantic field. Just as semantic configurations can be common to spatial and temporal fields, as in (47), so also, these semantic configurations can participate in other dimensions, or semantic fields. Thus, in (46c), the prepositions from and to express abstract origin and termination points. The points of origin and termination are neither spatial nor temporal; they belong to the domain of abstract mental states. However, the semantic configurations are identical to those in the spatial and temporal fields, warranting the use of identical prepositions. The idea of the same semantic configurations existing in different semantic fields yields the possibility of multiple layers of semantic structuring along the different fields. Compare (48a) and (48b): (48) a. b.

John gave the book to Bill. John took the book from Bill.

As far as the physical transaction of the book is concerned, John is the SOURCE and Bill the GOAL in (48a). In (48b), John is the GOAL, and Bill the SOURCE. Yet, at a more abstract level, the action (whether of giving or

174 / Arguments in Hindi

of taking) originates in John and comes to rest on the book. Jackendoff (1987:394) refers to this abstract level as the "Action Tier". On the action tier, what moves is not a participant but the action itself: in both (48a) and (48b), the action moves from John to the book. In (48b), then, John is the X in [FROM X] with respect to the action of taking, but the X in [TO X] with respect to the movement of the book. This brief and extremely sketchy discussion of semantic fields is by way of laying the ground for examining the use of identical case markings to express identical semantic configurations in varying semantic fields in Hindi. It is not my intention to examine in detail the various semantic fields that have been proposed by Gruber (1965), Jackendoff (1972, 1986), and others. I shall, however, rely on the idea that the organization of spatial concepts and their correlation with case as in (45) is reflected in the organization of non-spatial fields. These include fields encompassing static and dynamic locations of abstract mental states as in the English example in (46c). 7.3.3. The Extensions of Case to Non-Spatial Relations 7.3.3.1. The semantic configuration [FROM X] or SOURCE is expressed in Hindi by the case clitic -se . In (44d), -se marks a source in the spatial field. The case feature in this use we called LOC-se ((45a)). Recall that we have assumed the case feature in the nonspatial fields of the use of the clitic -se to be INST. To get a flavour of the non-spatial uses of -se , here are some examples, all of which have the core semantic configuration [FROM X]: (49) a.

anil do din-se bimaar hai. Anil-N two day-SE ill be-PR Anil has been ill for two days (lit.: from two days ago).

b.

sardii-se saare patte j¥a± gaye. cold-SE all leaves-N fall-NF go-PERF All the leaves dropped off because of the cold (lit.: from the cold).

c.

ilaa-ne pensil-se ciÆÆ¥ii lik¥ ii. Ila-E pencil-SE letter-N write-PERF Ila wrote a letter with a pencil.

d.

anil raam-se piiÆaa gayaa. Anil-N Ram-SE beat-PERF go-PERF Anil was beaten by Ram.

Indirect Case on Subjects / 175

e.

anil-ne raam-se pe± Anil-E Ram-SE tree-N Anil made Ram cut the tree.

kaÆvaayaa. cut-C-PERF

The temporal point of origin in (49a), the cause or reason in (49b), the instrument in (49c), the demoted agent of the passive in (49d), and the agent of the caused event in the causative in (49e) share the property of being a SOURCE, although in different semantic fields (see Clark (1989) for a similar discussion). Recall that the modal meaning of the INST SUBJ construction, we said, is the internally determined capability of the L-SUBJ to bring about or prevent an action or event. Suppose we rephrase this meaning as: the ability (or the absence thereof) of the L-SUBJ to be the source of the action or event. The internal nature of the I-ABILITY would then follow from the specification of source. If so, INST case on the L-SUBJ in the INST SUBJ construction follows directly from its semantic configuration, namely, [FROM X]. We modify the modal meaning of the INST SUBJ construction as follows: [ [FROM X] EVENT I-ABLE ]. This guarantees INST case on the LSUBJ without the special stipulation given earlier in (32). Exactly like the SOURCE in an abstract field that is in INST case, we saw in §7.1. that a GOAL, that is, the X in [TOWARDS X] or [TO X], is in DAT case.16 The distinction between DAT case and LOC-ko, then is a distinction in semantic fields, not in the semantic configurations. This allows for the use of the same case clitic for both features.17 We might go one step further and say that it is a common core semantic configuration that allows ACC case and DAT case, and now LOC-ko as 16 Like the English preposition to, -ko in Hindi ((45b)) is neutral to whether the destination is reached or not, as shown by (i) and (ii): (i) raam-ne ravii-ko kitaab dii. Ram-E Ravi-D book- N give-PERF Ram gave Ravi a book. (ii) raam-ne ravii-ko kitaab b¥ejii lekin Ram-E Ravi-D book- N send-PERF but ravii-ko nahññ milii. Ravi-D not get-PERF Ram sent Ravi a book, but Ravi didn't get (it). In (i), [TOWARDS x] & [AT x] are part of the meaning of the verb give . It is clear from (ii), on the other hand, that [at x] is not obligatory for linking with -ko . 17 According to Saksena (1982), -se denotes non-targets. It marks "originating points of activity," and by "extension", originating points of place and time. Likewise, -ko denotes targets, or "culmination points of activity," and by extension, culminating points in place and time.

176 / Arguments in Hindi

well, to converge on a single case clitic, namely, -ko . Recall that ACC case is associated with logical objects. A logical object, we said, is an entity towards which an action or event is directed. In other words, a logical object is the ARG associated with Y in an action that moves FROM X TO Y, where X and Y are distinct. What is common to ACC, DAT, and LOC-ko, then, is that they all involve the SEM STR configuration [TO X], despite the differences in their syntactic patterning. The difference between DAT and LOC-ko lies in the semantic field, as we just said. The difference between DAT and LOC-ko on the one hand and ACC on the other lies in that ACC is a direct case, and is sensitive to the GF of the ARG as well. A fine-grained distinction among semantic fields would show that ACC also involves a semantic field distinct from those of DAT and LOC-ko. The discussion above dealt with the semantic configurations of dynamic locations in (42) and their use in non-spatial fields. The configurations of static location in (41) also participate in non-spatial fields. Before turning our attention to these configurations, however, a brief note on the interaction between semantic configurations, semantic fields, and logical subjecthood is in order. I rely on the semantic configuration [TO X] (which includes [TOWARDS X]) for illustration. Consider the Hindi sentences in (50). A simplified SEM STR is given alongside, using the configurations we have been discussing. (50) a.

b.

raam g¥ar aayaa. Ram-N house-L-ko come-PERF Ram came home.

[ r

TOWARDS

h] & [r

raam-ko gussaa aayaa. Ram-D anger-N come-PERF Ram was angry. (Lit.: To Ram anger came.) [ a TOWARDS r ] & ( [ a

AT

AT

h]

r])

The SEM STR representations of (50a) and (50b) are almost identical. However, the semantic fields are distinct. In (50a), the locations are concrete spatial elements. In (50b), on the other hand, the locations are part of an abstract field. The entity that moves is an abstract mental state. It requires that the point of termination be one where the mental state can come to rest. In other words, what we have is a sentient GOAL.

Indirect Case on Subjects / 177

Recall from Chapter 3 that the relative prominence of ARGs at ARG STR is severely constrained by relative prominence statements on the entities of SEM STR they are associated with. According to (§3:(13b)), a sentient participant is more prominent than a non-sentient participant. By the principles that determine the relative prominence of ARGs, Ram in both (50a) and (50b) is higher than the other ARG, which is non-sentient. In (50a), then, the undergoer of movement is the L-SUBJ, whereas in (50b), the GOAL is the L-SUBJ. A difference in the semantic field thus results in differing relative prominence relations among ARGs. 7.3.3.2. We now turn to the static locations in (41), the case clitics that express them, and their occurrence in non-spatial fields. Let us first take the clitic -mã, which expresses the spatial relation [IN X] in (44c). Consider its use in (51): (51) a.

niinaa-mã ba±ii mamtaa hai. Nina-L much affection-N be-PR Nina has a great deal of affection. (Lit.: In Nina there is much affection.)

b.

niinaa-mã ßaktii paidaa huii. Nina-L strength-N born happen-PA Nina became (mentally) strong. (Lit.: In Nina strength was born.)

The sentences in (51) involve the relation of CONTAINMENT of inherent characteristics such as kindness, courage, and patience. The location of these characteristics is the X in the semantic configuration [IN X], and thereby expressed with mã 'in'. The ARG associated with the X in this relation is the L-SUBJ, and by default, the SUBJ, as we will see shortly. Kachru (1970), Pandharipande (1981b). and others point out that the abstract location of some inherent characteristics, such as himmat 'courage', d¥airy 'endurance', and b¥ay 'fear', normally marked by the clitic -mã "in", may alternatively be marked by the clitic -ko . The alternation is illustrated in (52): (52) a.

niinaa-ko b¥ay hai. Nina-D fear-N be-PR Nina is afraid.

178 / Arguments in Hindi

b.

niinaa-mã b¥ay hai. Nina-L fear-N be-PR Nina is fearful.

While -ko encodes the abstract location of a temporary state, such as happiness or worry, or a temporary fear as in (52a), -mã expresses the location of a characteristic attribute that is relatively permanent, such as a fearful disposition, as in (52b). When the state is inherently temporary, as in the event of a cough or a fever, the use of -mã is disallowed, perhaps because abstract containment cannot be extended to temporary states: (53) niinaa-ko /*mã

bahut k¥ããsii hai. Nina-D much cough-N be-PR Nina has a bad cough.

Now consider the use of the clitic par which expresses the spatial relation [AT X] or CONTACT ((44a)). The abstract location of responsibility, burden, and mishaps, is marked with par : (54) a.

ravii-par un sab baccõ-kii jimmevaarii hai. Ravi-L those all children-G responsibility- N be-PR Ravi is responsible for (the care of) all those children. (Lit.: On Ravi is the responsibility of all those children.)

b.

ravii-par bahut ba±aa booj¥ hai. Ravi-L very big burden-N be-PR Ravi has a very big (abstract) burden (to carry). (Lit.: On Ravi is a very big burden.)

c. ravii-par aafat aaii. Ravi-L mishap-N come-PERF Ravi was the victim of mishaps. (Lit.: Mishaps came on Ravi.)

Finally, the locative -ke paas is associated with the semantic configuration of PROXIMITY, [NEAR X], in (45e). Its use in the spatial field is illustrated in (44e). This locative of proximity is also used to express ownership of concrete objects, as in (55):

Indirect Case on Subjects / 179

(55) a.

vijay-ke paas æsimov-ki sab kitaabã hãñ. Vijay-G near Asimov-G all books be-PR Vijay has all of Asimov's books. (Lit.: Near Vijay are all of Asimov's books.)

b.

raam-ke paas caar makaan hãñ. Ram-G near four buildings-N be-PR Ram has/owns four buildings.

The relation of possession and the different ways in which it is expressed in Hindi can be discussed further only in the context of genitive subjects. At this point, we can only note that the semantic correlate of the locative postposition -ke paas is the configuration [NEAR X] in SEM STR. In (55), the nominal that bears this postposition is the L-SUBJ. In sum, the relative prominence of ARGs, and hence logical subjecthood, are partially dependent on the semantic field. The case clitic on the nominal, on the other hand, is dependent on the semantic configuration. The same semantic configurations appear in different semantic fields. As a consequence, nominals bearing any of the case clitics may be the L-SUBJ, and hence, the SUBJ. 7.3.4. Case Features, Case Markings, and Case Meanings The general principles for the association of LOC case, regardless of the GF of the ARG, may be formulated as in (56): (56) a.

LOCATION

[

BE

PRED

X

]

ARG

SEM STR ARG STR

! LOC

b.

GF STR

LOCATION

[

BE

PRED

X ARG

]SPATIAL

SEM STR ARG STR

! LOC

GF STR

180 / Arguments in Hindi

The principle in (56a) accounts for the appearance of LOC on subjects ((51), (54), (55)) as well as nonsubjects ((44a, c, e)). The variable PRED in (56a) is a static location: it can be instantiated by IN, AT or NEAR. The variable PRED in (56b) is a dynamic location, namely, FROM or TOWARDS. The ARG associated with the participant X in (56b) gets the case feature LOC only if the configuration is in the spatial field. This principle accounts for the appearance of LOC on the nonsubjects in (44b, d). These principles do not make any direct reference to the case clitics themselves. Therefore, they are not sufficient for determining the choice from among the different locative case markings, say, between mã and par . So far in our discussion, both in this chapter and in Chapter 4, we have formulated principles of case association as the linking of case features to ARGs. We now know that the choice of individual case clitics depends upon semantic contrasts involving configurations such as those in (41)-(42). In order to guarantee the correct choice of case markings, one may assume that the lexical entries of the case clitics and postpositions specify their inherent meanings in terms of semantic configurations, as in (57): (57)

semantics a. b. c. d. e.

-se -ko -mã -par paas

[ FROM X ] [ TO X ] [ IN X ] [ AT X ] [ NEAR X ]

syntax INST / LOC ACC / DAT /

( LOC )

LOC LOC LOC

The principles of case association specify syntactic case features such as ACC, DAT, INST, and LOC. A general condition stating that the semantic configuration specified in the lexical entry of the clitic or postposition must match the semantic configuration in the SEM STR of a predicate guarantees the correct choice of the clitic or postposition on the actual form. LOC association through (56b) takes effect only in the spatial field. In a non-spatial field, say, that of action or of an abstract movement, therefore, (56b) does not hold. The effect of this specification is that LOC-se and LOCko do not appear on L-SUBJs, as spatial participants are generally the lowest on the Argument Hierarchy. The -se and -ko that appear on SUBJs then must be INST and DAT.18 18 The ARG asociated with the X in [FROM X] in a non-spatial field gets INST case, and that in [TO X] gets DAT case. In the light of our discussion of the semantic configuration of case clitics and the fields associated with case features,

Indirect Case on Subjects / 181

7.3.5. The Grammatical Subjecthood of LOC L- SUBJ s Given the narrow range of environments in which the LOC nominals appear as subjects, not all the diagnostics for grammatical subjecthood prove useful in determining their subjecthood. These nominals do, however, exhibit behavior that would be predictable if we assumed that they were grammatical subjects. Recall that the antecedent of the reflexive apnaa must be either an L-SUBJ or a SUBJ (§6:(38a)). The LOC L-SUBJs in (58) are all eligible antecedents of the reflexive: (58) a.

niinaa-mã apnii mausii-ke liye ba±ii mamtaa Nina-L self-G aunt-G for much affection-N Nina i has a lot of affection for self's i aunt.

hai. be-PR

b.

ravii-par apne parivaar-kii saariijimmevaarii hai. Ravi-L self-G family-G entire responsibility- N be- P R Ravi i is solely responsible for (the care of) self's i family.

c.

vijay-ke paas apne pitaajii-kii sab kitaabã Vijay-G near self-G father-G all books- N Vijay i has all of self's i father's books.

hãñ. be-PR

In each of the instances in (58), the LOC L-SUBJ is the only possible antecedent of the reflexive, suggesting that either the L-SUBJ is the SUBJ, or there is no SUBJ. Once again, the facts of pronominal coreference support the former hypothesis. Pronouns cannot be coreferent with the SUBJ in their minimal finite clause (§6:(38b)). The pronoun in the sentences in (59) cannot be coreferent with the LOC L-SUBJs:

the principles governing DAT case and INST case must be modified to include the specification of the semantic field in which they hold. As mentioned earlier, we are not concerned with the precise details of the specific fields, except by way of illustration. It is not clear, for instance, if the temporal field should be treated along with the spatial field as "concrete", and the specification "spatial" in (56b) be modified to "concrete". In other words, should the temporal -se and -ko be treated as LOC-se and LOC-ko respectively, or as INST and DAT? Similarly, it might be that DAT case is associated with an X in [TO X] not in all non-spatial fields but in a non-spatial, non-action field. Such details are beyond the scope of this discussion.

182 / Arguments in Hindi

(59) a.

niinaa-mã uskii mausii-ke liye ba±ii mamtaa Nina-L pron-G aunt-G for much affection-N Nina i has a lot of affection for her * i aunt.

hai. be-PR

b.

ravii-par uske parivaar-kii saariijimmevaarii hai. Ravi-L pron-G family-G entire responsibility- N be- P R Ravi i is solely responsible for (the care of) his * i family.

c.

vijay-ke paas uske pitaajii-kii sab kitaabã Vijay-G near pron-G father-G all books- N Vijay i has all of his * i father's books.

hãñ. be-PR

These facts of reflexive binding and obviation follow automatically if we assume that the LOC L-SUBJs in (58)-(59) are SUBJs. Finally, given that only a SUBJ can control obligatory control sites in participial adjuncts (§6:(38c)), (60) shows conclusively that the LOC L-SUBJs are SUBJs: [ ——— bacce-kii avast¥aa dek¥ kar ] Nina-in child-G condition-N see-NF do-NF mamtaa paidaa huii. affection-N born happen-PERF Nina i , ——— i/*j seeing the child's condition, began to feel

(60) niinaa-mã

affection (for the child).

Thus, the LOC L-SUBJs in (51)-(55), and those in (58)-(60), must also be analyzed as grammatical subjects.

7.4. Genitive Subjects GEN case in Hindi, which appears on nominal modifiers within NPs (§4.3.), may also be associated with an independent ARG. In this section, we will be concerned with the distribution of GEN case on L-SUBJs, and with the grammatical subjecthood of GEN L-SUBJs. The outline of the facts of GEN SUBJs is given primarily for descriptive completeness. First I discuss GEN CASE on ARGs that might broadly come under the rubric of possessors. Next we encounter instances of GEN L-SUBJs where it seems unlikely that the association of GEN case is governed by a single coherent semantic regularity. On the other hand, we do find a syntactic regularity: GEN case appears on ARGs that are dependents of nouns. We conclude that GEN case in these

Indirect Case on Subjects / 183

instances is assigned by a nominal, and is an instance of indirect non-semantic case. Now, it might be possible to account for the GEN case on possessors also as being assigned by nominals. However, a precise implementation of this idea requires further work. 7.4.1. Genitive Logical Subjects 7.4.1.1. There is no verb of possession in Hindi, parallel to the English verb have . In Hindi, the verb ho 'be/become', with a GEN SUBJ, indicates certain types of "possession", as illustrated in (61): (61) a.

raam-kaa e k beÆaa hai. Ram-G one son-N be-PR Ram has one son.

b.

raam-kii e k bahin hai. Ram-G one sister-N be-PR Ram has one sister.

c.

raam-ke tiin b¥aaii Ram-G three brothers-N be-PR Ram has three brothers.

hãñ.

d.

raam-kii tiin beÆiyãã Ram-G three daughters-N be-PR Ram has three daughters.

hãñ.

In (61a-d), the GEN nominal that refers to the possessor agrees in number and gender with the nominal that refers to the possessed entity. The GEN marking -kaa ((61a)) is the masculine singular form of the genitive clitic (see §4. n.13). The masculine plural form is-ke ((61c)), and the feminine form is -kii ((61b, d)). These are the NOM forms of the GEN clitic, the NONNOM form invariantly being -ke. The relation of "possession" in (61) is one of kinship, or a relation to persons, not one of ownership. The specific type of possession that does not involve ownership also extends to the relationship one has to animate beings, or to one's body parts. This relation has been referred to as "inalienable possession" (Kachru (1969, 1980)). In expressing the relation of inalienable possession as in (61), the NOMINATIVE form of the GEN clitic may optionally be replaced by the NONNOMINATIVE form -ke , as illustrated in (62):

184 / Arguments in Hindi

(62) a.

isbillii-kii / -ke ek hii ããk¥ hai. this cat-G one only eye-N(F) be-PR This cat has only one eye.

b.

is haat¥ii-kaa / -ke bahut c¥oÆaa mastak this elephant-G very small forehead-N(M) be-PR This elephant has a very small forehead.

hai.

As Jackendoff (1986:191) points out, the relation of possession can hold in different semantic fields, “since there are several distinct notions of possession". Thus, one can possess an eye ((62a)), books and buildings ((55)), or even a son or daughter ((61)). One can also possess inner qualities such as mercy ((51)), or responsibilities ((54)). In other words, the semantic configurations that are overtly expressed in terms of LOC and GEN case in (51)-(55) and (61)-(62) may all be viewed as expressing the relation of possession in the various semantic fields. To repeat what we said earlier, the semantic field interacts with the principles of relative prominence, to make the possessor the L-SUBJ. The semantic configuration determines the overt case marking. The L-SUBJ is by default the SUBJ. As a consequence, we have SUBJs in the LOC and GEN case. Whereas the use of NONNOM GEN -ke is restricted to the relation of inalienable possession, the NOM GEN -kaa has wider use, for instance, in the relation of ownership, as in (63) below. It must be noted that the GEN -kaa and the LOC -ke paas are mutually interchangeable in (63): (63) raam-kaa / raam-ke paas

e k hii makaan hai. Ram-G Ram-L one only building-N be-PR Ram has/owns only one building.

McGregor (1972) suggests that -kaa in instances such as (63) indicates a permanent or characteristic type of relationship, and -ke paas denotes a more "contingent" possession. However, Pandharipande (1981b) shows the inadequacy of this characterization. The clause, "which he is trying to sell," may modify the building in (63) without creating a contradiction even if the possessor is marked with -kaa. Likewise, the clause, "which he will hand down to his children," may modify the building in (63), again without a contradiction, even if the possessor is marked with -ke paas. Pandharipande argues that -kaa is used when the relation is one of emotional attachment or intimacy (or, one might add, inalienable possession), and -ke paas is used to express purely material ownership. Further, there are relations which con-

Indirect Case on Subjects / 185

ventionally involve intimacy, such as kinship or friendship, where -ke paas is disallowed, and others which are conventionally viewed as that of ownership, such as a pencil, where -kaa is disallowed. Either of the two markers may be used when convention doesn't select one, as in (63), and the relation may be ambiguous.19 It is reasonable to conclude that GEN case on the SUBJ in all these instances is governed by the semantic relation of possession. On the other hand, identifying the semantic configuration(s) responsible for association with GEN case in (61)-(63) requires further study. The idea that this GEN case may in fact not be governed by semantics but by syntax is worth exploring. Therefore, I will not attempt to provide formal statements of GEN case association analogous to the principles of DAT, INST, and LOC case association. 7.4.1.2. Certain complex predicates with predicative nominals such as vicaar ho "to think", daavaa ho "to claim", iraadaa ho "to intend", and vißvaas ho "to believe", are associated with GEN SUBJs. These predicates are perhaps characterizable as involving deliberate mental activity proceeding from the L-SUBJ. Such predicates are illustrated in (64)-(65):20 (64) raam-kaa [ g¥ar

lauÆne ] -kaa vicaar t¥aa. Ram-G home-N return-NF-G thought-N be-PERF Ram was thinking of returning home.

(65) anu-kaa [ mãã-ko Anu-G

sabkuc¥ bataane] -kaa iraadaa t¥aa. mother-D everything tell-NF-G intention-N

be-

PERF

Anu had the intention of telling mother everything.

In (64)-(65), the GEN L-SUBJ is the source of the mental activity of thought and intention. One might take the position that it is also the "possessor" of

19 According to Pandharipande (1981b), the choice between the markers -k o and -mã ((52)-(53)) and between -kaa and -ke paas, can be accounted for only i n terms of pragmatic factors. They cannot be determined by semantics alone, because the choice is dependent on the context of use. However, that the case i s determined by pragmatics is not a necessary conclusion. I suggest that, when certain predicates allow the use of more than one marker, if each marker i s associated with one or more specific semantic configurations, lexical meaning i s sufficient for determining case. Needless to say, the meaning selected must be compatible with the context. 20 Also see Kachru (1988).

186 / Arguments in Hindi

the activity, where "possessor " refers to an abstract relationship, the nature of which is not clear. Also associated with GEN SUBJs are a set of complex predicates that may crudely be characterized as involving irreversible events, or events that result in a relatively permanent state. Examples of such events are janm ho "be born", vivaah ho "get married", mÇtyuu ho "die", aaramb¥ ho "to begin", ant ho "to end" (Kachru (1988)). These predicates are illustrated in (66)-(68): (66) ßriikÇ‹”-kaa

mat¥uraa-mã janm huaa. Sri Krishna-G Mathura-L birth-N happen-PERF Sri Krishna was born in Mathura.

(67) anu-kaa

e k saal-se ad¥ikilaaj huaa. Anu-G one year- more than treatment-N happen-PERF Anu underwent treatment for more than a year.

(68) baccõ-kaa savere

vidyaaramb¥ huaa. children-G morning learning-beginning- N happen-PERF The children were initiated into learning in the morning.

In (66)-(68), the GEN L-SUBJ is the undergoer of the event. Unlike in (64)(65), however, viewing these L-SUBJs as the possessor of the event would be rather far-fetched. According to Kachru (1988), "the phenomenon of genitive subject is a result of a grammaticalization process, and is not predictable on the basis of the semantics of the predicates." The situation is not as random as may first appear. Let us separate the GEN case clearly governed by meaning ((61)(63)), and that which appears not to have any systematic correlates ((64)(68)). The latter set invariably involves complex predicate constructions of the form noun+verb, and the verb is systematically ho "be"/"become". This is not to say, of course, that all noun+ho complex predicates take GEN SUBJs, as shown by the DAT SUBJ in (2d). Having reduced the range of apparent randomness to the complex predicate construction, we find a syntactic regularity. As is hinted at by Kachru, we might conclude that the GEN in these instances is the default case assigned by a nominal. It is then an indirect non-semantic case, because its association makes reference neither to meaning nor to the GF of the ARG. As we will see in Chapter 8, the nominal assigns GEN case not only to subjects but to any ARG whose case is dependent on it unless a specific semantic

Indirect Case on Subjects / 187

configuration results in the association of an indirect semantic case, for instance, the DAT. In such an event, GEN case association is preempted by the association of semantic case. 7.4.3. The Grammatical Subjecthood of G EN L- SUBJ s Consider the genitive nominals in (61)-(68) in the light of the diagnostics for grammatical subjecthood laid out in §6:(38a-d). The GEN L-SUBJs in (69) are the only eligible antecedents of the reflexive apnaa (§6:(38a)), suggesting that they are SUBJs, unless, of course, the sentences have no SUBJ: (69) a.

raam-ke apnaa e k b¥ii beÆaa Ram-G self-G one even son-N not Ram didn't have even one son of his own.

b.

ßriikÇ‹”-kaa apne Sri Krishna-G self-G

nahññ be-PA

maamaa-ke mahal-mã janm huaa. uncle-G palace-L birth

t¥aa.

happen-

PERF

Sri Krishna i was born in self’si uncle's palace.

Consider the contrast between (66) and (70) below. (66) is repeated for ease of comparison. (66) ßriikÇ‹”-kaa

mat¥uraa-mã janm huaa. Sri Krishna-G Mathura-L birth-N happen-PERF Sri Krishna was born in Mathura.

(70) ßriikÇ‹”-kaa

janm mat¥uraa-mã huaa. Sri Krishna-G birth-N Mathura-L happen-PERF Sri Krishna’s birth took place in Mathura.

The only surface difference between (66) and (70) is in the order of words. However, consider the difference in meaning, clear from the sentence glosses. In (70), the GEN nominal is the modifier of janm "birth". In (66), on the other hand, it is an independent argument of the complex predicate janm ho "be born". If so, we would predict that the GEN nominal in (70), in contrast to the one in (66), cannot be the antecedent of the reflexive apnaa

188 / Arguments in Hindi

: it is not an ARG, let alone an L-SUBJ or SUBJ. This prediction is fulfilled, as shown by the contrast between (69b) and (71): 21 (71) * ßriikÇ‹”-kaa

janm apne maamaa-ke mahal-mã huaa. Sri Krishna-G birth self-G uncle-G palace-L happen-PERF Lit.: Sri Krishna's i birth took place in self’si uncle's palace.

As we said, the GEN L-SUBJs in (69) could be SUBJs. The facts of pronominal coreference (§6:(38b)) support this possibility. The pronoun in (72) cannot be coreferent with the GEN L-SUBJ:

(72)

ßriikÇ‹”-kaa uske Sri Krishna-G pron-G

maamaa-ke mahal-mã janm huaa. uncle-G palace-L birth

happen-

PERF

Sri Krishna i was born in his * i uncle's palace.

Once again, the pronoun may be coreferent with the GEN nominal if the nominal is the modifier of another nominal, as shown by (73):

(73)

ßriikÇ‹”-kaa janm uske maamaa-ke mahal-mã huaa. Sri Krishna-G birth pron-G uncle-G palace-L happen-PERF Sri Krishna'si birth took place in hisi uncle's palace.

The GEN L-SUBJ under discussion can be the controller of a participial adjunct clause with an obligatory control site: (74) ßriikÇ‹”-kaa [ ——— muskuraate

hue ] Sri Krishna-G smile-NF be-NF Sri Krishna i was born, ——— i smiling.

janm birth-N

huaa. happen-PERF

By the diagnostic in §6:(38c), the controller of a participial adjunct clause with an obligatory control site must be a SUBJ. Therefore, we must conclude that the GEN nominal in (74) is a SUBJ. This also explains its behavior with respect to reflexive binding and subject obviation.

21 (71) is not unacceptable to some speakers, for instance, Yamuna Kachru (p.c.). This implies that the reflexivization patterns of these speakers differ from those relied on in this thesis.

Indirect Case on Subjects / 189

7.5. Accusative Logical Subjects In the previous sections, we studied the syntax and semantics of L-SUBJs in DAT, INST, LOC, and GEN case, and provided evidence for their grammatical subjecthood. We come finally to a construction in which the L-SUBJ is in ACC case. 7.5.1. The Accusative Logical Subject Construction (75a) is an active sentence. (75b) is its passive version. (75c) is an example of the construction we will examine in this section. I refer to the construction in (75c) as the ACC L-SUBJ construction. (75) a.

raam-ne ravii-ko Ram-E Ravi-A Ram beat Ravi.

piiÆaa. beat-PERF

b.

ravii (raam-se) piiÆaa Ravi-N Ram-I beat-PERF Ravi was beaten (by Ram).

c.

ravii-ko (*raam-se) piiÆaa gayaa. Ravi-A Ram-I beat-PERF go-PERF Ravi got beaten.

gayaa. go-PERF

The passive in (75b), and the construction in (75c) have identical verb morphology, as does the INST SUBJ construction discussed in §7.2. Given below is a set of sentences illustrating all three uses of the "passive morphology" in Hindi: 22 (76) a.

pulis-ne cor-ko police-E thief-A The police caught the thief.

pak±aa. catch-PERF

b.

(pulis-se) cor pak±aa gayaa. police-I thief-N catch-PERF go-PERF The thief was caught (by the police).

c.

cor-ko pak±aa

gayaa.

22 Wallace (1985) observes a similar three-way semantic distinction in the use of passive morphology in Nepali.

190 / Arguments in Hindi

thief-A catch-PERF go-PERF The thief was caught. d.

pulis-se cor-ko pak±aa nahññ gayaa. police-I thief-A catch-PERF not go-PERF The police couldn't bring themselves to catch the thief.

(76a) is the active sentence, (76b), the passive, (76c), the ACC L-SUBJ construction, and (76d), the INST SUBJ construction. (76c) differs from the passive on the surface in that the ACC OBJ of (76a) is NOM in (76b), while ACC case is retained in (76c). (76c) differs from both the passive and the INST SUBJ construction in that whereas the agent ARG is obligatorily expressed in the INST SUBJ construction, and optionally expressed in the passive, it is obligatorily unexpressed in the ACC L-SUBJ construction. (76c) is ungrammatical with an expressed agent.23 The understood agent in (76c) is necessarily interpreted as arbitrary, that is, non-specific. Now, if the object in the active sentence were inanimate, and hence NOM, it would be NOM in both the passive and the ACC L-SUBJ construction. Moreover, in the dialects of Hindi that preserve ACC case under passivization, an ACC object in the active would remain ACC in both constructions. If, in addition, the agent in the passive (which is often unexpressed) were absent, the surface distinction between the passive and the ACC L-SUBJ construction would be entirely obliterated.24 This merging of the passive and the ACC L-SUBJ construction on the surface has prevented researchers from analysing them syntactically as distinct structures, exactly as with the passive and the INST SUBJ construction. In what follows, I show that the ACC nominal in (75c)/(76c) is an LSUBJ but not a SUBJ, and argue that the construction lacks a SUBJ altogether. I will then provide a formal representation for the ACC L-SUBJ construction 7.5.2. A Subjectless Construction

23 Thus the three distinct constructions, each with its own syntactic and semantic peculiarities, all employ the same verb morphology. Clearly, the 'passive' morphology in Hindi does not characterize a single syntactic phenomenon, and it is natural to ask what the unity of the three constructions is. The issue is left as a puzzle for future research. 24 In short, a surface distinction between the two constructions is available only (a) in the non-ACC preserving dialects, (b) if the object in the active is ACC .

Indirect Case on Subjects / 191

Recall once again that the antecedent of the reflexive apnaa must be either an L-SUBJ or a SUBJ (§6:(38a)). Consider the behavior of the passive and the ACC L-SUBJ construction in (77a) and (77b) respectively. Their behavior is identical with respect to the reflexive. (77) a.

ravii apne g¥ar-mã piiÆaa Ravi-N self-G house-L beat-PERF Ravi i was beaten in self's i house.

gayaa. go-PERF

b.

ravii-ko apne g¥ar-mã piiÆaa Ravi-A self-G house-L beat-PERF Ravi i got beaten in self's i house.

gayaa. go-PERF

Given that the reflexive can refer to the ACC nominal in (77b), we must conclude that this nominal must be either the L-SUBJ or the SUBJ (or both). Now consider pronominal coreference: a pronoun cannot be coreferent with the SUBJ within its minimal finite clause (§6:(38b)). (78) a.

ravii uske g¥ar-mã piiÆaa Ravi-N pron-G house-L beat-PERF Ravi i was beaten in his* i house.

gayaa. go-PERF

b.

ravii-ko uske g¥ar-mã piiÆaa Ravi-A pron-G house-L beat-PERF Ravi i got beaten in his i house.

gayaa. go-PERF

Given that the pronoun in (78b) can be coreferent with the ACC nominal, we are forced to conclude that this nominal is not a SUBJ. We have so far been referring to the ACC nominal in (75c) as an LSUBJ. The facts of reflexive binding in (77b) show that it must be either an L-SUBJ or a SUBJ. The facts of subject obviation in (78b) show that it is not a SUBJ. Therefore, the ACC nominal must be an L-SUBJ. As might be predicted, the ACC ARG cannot be the controller of participial adjuncts with obligatory control sites, where the controller must be a SUBJ, as in (79):

(79) * ravii-ko [ ——— muskuraate hue ] Ravi-A

smile-IMPERF

be-NF

piiÆaa beat-PERF

gayaa. go-PERF

192 / Arguments in Hindi

However, it is an acceptable controller of participial clauses which do not require the controller to be a SUBJ. (80a) is an example of a participial adjunct clause where either the subject or the object may be the controller. The control site is not obligatory in the embedded clause: it is a null pronominal (pro). In (80b), the ACC L-SUBJ is an eligible controller. (80) a.

ilaa-ne [ raam-ke / ——— darvaazaa k¥olne par] anu-ko ±ããÆaa. Ila-E Ram-G door-N open-NF on Anu-A scold-

PERF

Ila i scolded Anu j on Ram'sk / ——— i/j/k opening the door. b.

ravii-ko [ ——— muskuraane par ] piiÆaa Ravi-A smile-NF on beat-PERF go-PERF Ravi i got beaten on ——— i/j smiling.

gayaa.

In short, the construction in (75c) presents an instance of a clause without a SUBJ (also see Pandharipande (1981b)). The ACC nominal in (75c)/(76c) is an L-SUBJ but an OBJ. 7.5.3. A Formal Representation Given the differences in the syntactic properties of the passive and the ACC L-SUBJ constructions, we must assign to them appropriate lexical representations that will express these differences. Let us examine the association of meaning, GF, and case to ARGs in these constructions, using the kinds of representations introduced in Chapter 3. First take the verb in the active sentence in (76a). The fully specified representation of the verb is given in (81). The SEM STR representation in (81) is extremely simplified and skeletal. As we already know, the participant X bears the ARG meaning of CONSCIOUS CHOICE; and the action of catching moves from X to Y , thereby making Y an L-OBJ. These details are omitted from (81): (81) [

X

Y

catch

SEM STR

]

……………………………………………………………………… ˘ ARG ARG ¿ A-PRED ARG STR ……\………….\…………….\………………………………. SUBJ ERG

OBJ

GF PRED

GF STR

ACC

……………………………………………………………………… pulis-ne cor-ko pak±aa WORD STRING Police thief caught ……………………………………………………………………… NP NP V GC STR

Indirect Case on Subjects / 193

S

194 / Arguments in Hindi

The information in GF STR in (81) is entirely predictable. The default principles of function assignment associate the L-SUBJ with SUBJ and the L-OBJ with OBJ (§3:(43)): the ARGs associated with X and Y receive subjecthood and objecthood respectively. The SUBJ gets ERG case and the OBJ gets ACC case by the principles of case association formulated in Chapter 4. The representation of the passive in (76b) is given in (82). Recall that we use the convention of underlining to indicate the suppressed L-SUBJ: (82) [

X

Y

catch

]

SEM STR

……………………………………………………………………… ˘ ARG ARG ¿ A-PRED ARG STR ……\………….\…………….\………………………………. ADJ INST

SUBJ NOM

GF PRED

GF STR

……………………………………………………………………… pulis-se cor pak±aa gayaa WORD STRING Police thief was caught ……………………………………………………………………… NP NP V GC STR S

In the ACC L-SUBJ construction in (76c), as we have said, the "agent" is obligatorily unexpressed. We also said that the ACC nominal is the LSUBJ but not the SUBJ. Now, in order for the ACC nominal to be the LSUBJ, it must be the highest logical ARG in the ARG STR. In other words, the agent in this construction cannot be an ARG. If we consider the meaning of the verb paka± 'catch', however, we have to acknowledge the presence of an implicit nonspecific agent, exactly as in the "middle" construction, an archetypal example of which in English is Bureaucrats bribe easily.25 The notion of suppression, which expresses the relation between the active sentence in (76a) and the passive in (76b), holds in the mapping between ARG STR and GF STR. The L-SUBJ in the passive is then an IMPLICIT ARGUMENT. One way of expressing the systematic relation between (76a) and (76c) is to extend the notion of suppression to the mapping between SEM STR and ARG STR, and assume that the "catcher" in the SEM STR of (76c) is suppressed in the mapping to ARG STR. That is, it is an IMPLICIT AGENT: 25 For an extensive summary of the difference between middles and passives, see Roberts (1987).

Indirect Case on Subjects / 195

(83) [

X

Y

catch

SEM STR

]

……………………………………………………………………… ˘

ARG

¿

A-PRED

ARG STR

………………….\…………….\………………………………. OBJ ACC

GF PRED

GF STR

……………………………………………………………………… cor-ko pak±aa gayaa WORD STRING thief was caught ……………………………………………………………………… NP V GC STR S

In (83), the catcher X is unavailable for association with ARG STR. Therefore, although it would have been the highest ARG if associated with an ARG, it cannot be the L-SUBJ because it is not an ARG. This accounts for why the "patient", not the "agent", is the L-SUBJ in the ACC L-SUBJ construction. We have shown that the ACC nominal in the ACC L-SUBJ construction is not a SUBJ. As pointed out earlier, the "patient" ARG in this construction is NOM if inanimate, and ACC if animate (or inanimate definite). If we assume that the ARG is a PR(IMARY) OBJ, this alternation follows directly from the principles of OBJ case association formulated in Chapter 4. The grammatical objecthood of the ACC nominal is consistent with its logical objecthood. That it is an L-OBJ follows directly from the SEM STR representation: it is the entity towards which the action is directed from an independent source of the action. This would not be possible if there were no "agent" in the SEM STR of the construction. Evidence for the presence of an implicit agent in the SEM STR of the construction has another source. The adverb razaamandii-se may be interpreted either as "willingly", or as "with the consent (of)".26 Consider the use of this adverb in (84)-(87): (84)

raam-ne razaamandii-se ravii-ko piiÆaa. Ram-E Ravi-A beat-PERF (i) Ram willingly beat Ravi. (ii) Ram beat Ravi with the consent (of, say, the government).

26 For the latter interpretation, some speakers require that the giver of consent be explicitly specified.

196 / Arguments in Hindi

Willingness in (84.i) is attributed to the SUBJ, namely, Ram. The consent to act is given to the "agent", also Ram, in (84.ii). Ravi may be the giver of consent in (84), but cannot be the receiver of consent. (85) ravii

razaamandii-se (raam-se) piiÆaa gayaa. Ravi-N Ram-I beat-PERF go-PERF (i) Ravi was beaten willingly (by Ram). (ii) Ravi was beaten (by Ram), with consent (of, say, the government).

In the passive in (85), willingness to be beaten is attributed to the SUBJ, Ravi. It cannot be attributed to the L-SUBJ, Ram. On the other hand, the consent to act is given to the "agent", Ram. (85) cannot be interpreted to mean that consent was given to Ravi to be beaten.27 Now consider this adverb in the ACC L-SUBJ construction:

(86)

ravii-ko razaamandii-se piiÆaa gayaa. Ravi-A beat-PERF go-PERF (i) # Ravi got beaten willingly. (ii) Ravi got beaten, with the consent (of, say, the government).

Willingness can only be attributed to a SUBJ, as shown by (84)-(85). The ill-formedness of (86.i) follows from the assumption that the sentence has no SUBJ. In both (86) and (85), Ravi may be the giver of consent. But (86), like (85), cannot be interpreted to mean that consent was given to Ravi to be beaten. Consent is given to the performer of the action of which Ravi is the undergoer. In order for (86) to be well-formed, then, there must be an implicit agent in the SEM STR to whom consent can be given. (87) below is the "inchoative" version of (84), with no implicit agent. (87) shows that in the absence of an "agent" in the SEM STR, the interpretation of razaamandii-se as "with the consent (of)" is ill-formed.

(87)

ravii razaamandii-se piÆaa. Ravi-N consent-with get beaten-PERF (i) Ravi got beaten willingly. (ii) #? Ravi got beaten with consent (of, say , the government).

The contrast between (86) and (87) in the interpretation of the adverb thus provides support for the assumption that the SEM STR of the ACC L-SUBJ 27 Needless to say, one would expect variability across speakers in the interpretation of the adverb.

Indirect Case on Subjects / 197

construction ((86)) contains an implicit agent which is unavailable for argumenthood. The representation of (87) is given in (88): (88)

[ Y get beaten ] SEM STR …………………………………………………………………… ˘ ARG ¿ A-PRED ARG STR ……\………….\……………………………………………. SUBJ NOM

GF PRED

GF STR

…………………………………………………………………… ravii piÆaa WORD STRING Ravi got beaten …………………………………………………………………… NP V GC STR S

In (88), unlike in (83), Ravi is neither an L-OBJ nor an OBJ. It is an L-SUBJ and a SUBJ. It exhibits subjecthood properties with respect to reflexives, subject obviation, control, and coordination. The L-SUBJ gets NOM case by virtue of being a SUBJ. 7.5.4.

Implications

If the correct distinction between the passive and the ACC L-SUBJ construction is as proposed above, we are forced to recognize that the argumenthood and meaning of arguments cannot be collapsed into a single level of representation, as is done in traditional theta role representations. The passive ((82)) involves the suppression of a unit of ARG STR, whereas the ACC LSUBJ construction ((83)) involves the suppression of a unit at SEM STR. Our analysis of these two constructions justifies the separation of SEM STR and ARG STR. This analysis also justifies our definition of L-SUBJ as the highest logical ARG, rather than as the "highest theta role". In spite of an agent participant in SEM STR which is semantically higher the undergoer participant, the latter is the L-SUBJ in the ACC L-SUBJ construction. If the conclusion that the ACC L-SUBJ construction has no SUBJ is correct, it violates the widely accepted condition that every GF clause must contain a SUBJ (§3:(39)). Given the above analysis, the Subject Condition cannot be an absolute universal, and theories of syntax must allow violations of the condition, even though such violations might be rare and highly marked.

198 / Arguments in Hindi

One might argue that the construction does have a SUBJ, which is obligatorily unexpressed. That is, it is a null subject with the agent theta role, that is, an entity equivalent to pro or PRO in GB. This is in fact the account given for somewhat similar constructions in languages like Italian and Irish (e.g. Burzio (1986), Stenson (1989)). However, given that the ACC nominal must be the L-SUBJ in the ACC L-SUBJ construction in Hindi, the agent cannot be an ARG, and hence cannot be either pro or PRO. One might counter-argue that an L-SUBJ may be re-defined as the highest expressed ARG. The ACC ARG would then be the L-SUBJ only by virtue of being the highest expressed ARG, because the agent is obligatorily unexpressed. This cannot be true, however. Recall that in the passive, the agent, which is the L-SUBJ, is optionally expressed. Even when the L-SUBJ is not expressed, it can be the antecedent of the reflexive. Therefore, the notion LSUBJ cannot be re-defined as the "highest expressed argument". One may then postulate an obligatorily null subject without a theta role, in other words, a null expletive. The sole motivation for such a move, however, would be to prevent the violation of the Subject Condition. Furthermore, an expletive, that is, an argument without a theta role, characteristically correlates with an indefiniteness effect. Take for example an English sentence: There is a/*the reindeer on the roof. Analyses of covert or null expletives have also found this systematic correlation with indefiniteness (McCloskey (1985), Burzio (1986), Stenson (1989), among others). In the ACC L-SUBJ construction, such an indefiniteness effect is entirely absent. Thus, the construction does not fit the pattern of constructions that have been analysed as having expletive subjects. The status of the Subject Condition (§3:(39)) as an inviolable universal is thereby called into question by this construction. Given that the motivation for the Subject Condition is cross-linguistically robust, the condition can by no means be abandoned. The only conclusion open to us then is that the Subject Condition may be violated under certain special circumstances. How precisely to delimit the circumstances that allow the violation of the condition requires careful investigation.

7.6. Theoretical Consequences Lexical meanings can govern a wide range of syntactic phenomena such as grammatical function selection, case selection, anaphora, control, auxiliary selection, and the like. Our focus in this chapter has been on the role of meaning in case selection on SUBJs in Hindi. This study brings into focus

Indirect Case on Subjects / 199

several important questions. What are the meanings of predicates that can participate in syntactic regularities? How are these meanings represented? How do we formally state syntactic regularities that are governed by lexical semantics? 7.6.1. The systematic correspondences between case and meaning in Hindi show that the so-called "quirky" case is not necessarily quirky or idiosyncratic at all. The association of cases such as DAT and LOC to an ARG is heavily conditioned by semantic principles.28 As we saw in Chapter 4, dependence on semantics is not restricted to indirect case: ERG case association, although sensitive to the GF, is also governed by the semantic property we called CONSCIOUS CHOICE. Thus, direct case can also be governed by semantics. This dependence on SEM STR demands a theory of lexical semantics that factors out and expresses in SEM STR those semantic configurations that are relevant for case selection. Meanings relevant for case selection are not capturable in terms of theta roles. Thus, in the so-called "experiencer" subject construction in Hindi, the role of experiencer is neither necessary nor sufficient. In order to characterize case selecting meanings, we employed notions like SOURCE, GOAL, CONTAINMENT, CONTACT, and PROXIMITY, which are abbreviations for semantic configurations made up of atomic semantic predicates and variables. These semantic configurations may exist in any of the several semantic fields including those of space, time, possession, action, and the like. Selection of case features and their markings is governed by the semantic configurations and semantic fields. The need for the various semantic configurations and semantic fields for an account of DAT, INST, GEN, and LOC cases in Hindi presents a most convincing instance of the inadequacy of traditional theta roles for expressing syntactically relevant verb meanings. As indicated in Chapter 3, the intuitive substance behind theta roles is expressible in terms of the kinds of semantic configurations and semantic fields we have postulated. However, not all semantic distinctions expressed in terms of these configurations and fields are translatable into the theta role conception. Therefore, given the syntactic representation called ARG STR, and the independently motivated

28 That the relation between case and meaning is predictable does not, of course, rule out the possibility of lexical idiosyncrasies, where the case i s lexically stipulated.

200 / Arguments in Hindi

structured representations of meaning that we have called SEM STR, the notion of theta roles can be eliminated from syntax. Only a subset of the meanings associated with an ARG position can constrain GF selection. Again, as mentioned in Chapter 3, it is the GF selecting meanings that are captured by theta role representations. Now, the intuition underlying a proposal in Ramchand (1990) is that in order for an ARG to be a TERM, it must be one of three entities in an event structure: CAUSE, CHANGE, and COMPLETION. In the semantic configuration [ X MOVE FROM Y TO Z ] in the field of action, Y corresponds to CAUSE, X to CHANGE, and Z to COMPLETION. In other words, the meanings relevant for TERM-hood may be abstracted into these three notions. An ARG that does not fall into one of these three categories must be an OBLIQUE. If Ramchand's claim is correct, the meanings relevant for GF selection may be expressed more precisely at SEM STR in terms of semantic configurations and semantic fields such as those relevant for case selection. 7.6.2. We now turn to the issue of how syntactically relevant meanings are accessed by syntactic principles. The Indirect Reference Hypothesis (§3.2.2.) holds that meanings of verbs cannot be directly accessed by principles of syntax. Verb meanings play a direct role only in the organization of argument structure. Lexical semantic information is available to syntax only to the extent that it is reflected in argument structure. Our analysis of the Hindi facts clearly contradicts this hypothesis. Thus, given the same ARG STR, the interaction of semantic fields with semantic configurations such as SOURCE, GOAL, CONTAINMENT, and the like, and meanings such as IABILITY and CONSCIOUS CHOICE can determine the case on the SUBJ. 7.6.3. Recall from Chapter 4 our four-way classification of case, depending on the kinds of information its association requires. The classification gives us the following typology of case on SUBJs: (89) a.

DIRECT CASE:

(i) (ii) b.

non-semantic: semantic:

Default NOM and GEN on SUBJ. ERG on SUBJ.

INDIRECT CASE:

(i) (ii)

semantic: non-semantic

DAT, INST, LOC on SUBJ. GEN assigned by nominals.

Indirect Case on Subjects / 201

Direct non-semantic case ((89a i)) corresponds to "structural case" in GB. Indirect semantic case ((89b i)) may appear to correspond to "inherent case". But GB views inherent case as being idiosyncratic, and stipulated in the lexical entry of the predicate (Belletti and Rizzi (1988:332-333)). We have shown for Hindi that both direct and indirect semantic case are perfectly regular, and predictable from SEM STR. GB must therefore incorporate a mechanism to capture the regularities of semantically governed case. Kiparsky (1987) distinguishes between "grammatical linking" and "semantic linking". This distinction may at first appear to correspond to that between direct and indirect case. In order to find out whether or not this correspondence holds, it is essential that we understand what "grammatically linked" and "semantically linked" mean. "Grammatical linking" and "semantic linking" are the two ways in which an argument is linked to overt morphosyntactic expression in terms of case marking, agreement, and word order. A crucial claim in Kiparsky is that given the Theta-structure and the overt expression of arguments, and the distinction between grammatical and semantic linking, the functional distinction between subjects and objects (§3.4.1.) is redundant. A subject is simply defined as the "highest grammatically linked argument". A transparent interpretation of the terminology of semantic linking and grammatical linking is that the former refers to the selection of overt expression on the basis of meaning, and the latter refers to the selection of overt expression without referring to meaning, that is, entirely on the basis of Thematic Hierarchy and non-demoted status of an argument. If so, nonnominative subjects in Hindi must be semantically linked because their case is determined by meaning. But they must be grammatically linked because they are subjects. In order to avoid this contradiction, it is necessary to interpret "grammatically linked" and "semantically linked" as TERMS and NONTERMS, which are primitive GF constructs (§3.4.1.). The mode of selection of overt expression, (semantic vs. non-semantic) will then be an independent variable. That is to say, a subject, which is always grammatically linked, may bear either direct case or indirect case. Hence, the correspondence which we set out to examine cannot hold. Given our analysis of the ACC L-SUBJ construction, the definition of a subject as the highest grammatically linked argument is impossible. The ACC nominal in this construction is the highest grammatically linked argument, and yet, it is an object, not a subject.

202 / Arguments in Hindi

7.6.4. The separation of the two levels of structure, namely, SEM STR and ARG STR, allows us to express the co-variation of the DAT, INST, LOC, and GEN case on the SUBJ with certain aspects of meaning while keeping other aspects of meaning constant. The ARG slot in ARG STR provides the anchor for both the constant and the varying aspects of meaning, exploiting the possibilities of multiple association between ARG STR and SEM STR (§3:(16d)). The separation of levels also allows us to express the structural difference between the passive and the ACC L-SUBJ construction. In the former, an ARG is suppressed. In the latter, a semantic unit is suppressed, exploiting the possibility of elements at SEM STR that are unassociated with ARG STR (§3:(16b)). Furthermore, it allows for the possibility of L-SUBJ and L-OBJ converging on a single ARG, essential for the interaction of reflexive binding, subject obviation, and case marking in the ACC L-SUBJ construction. 7.6.5. A fundamental assumption that we have appealed to in our account of case association in this Chapter, as well as in Chapter 4, is that when two principles governing syntactic regularities are in conflict, one of them "preempts", or has priority over the other. In particular, the specific principle preempts the general principle (the Elsewhere Condition), and an indirect case association principle preempts a direct case association principle (§4:(41)). When a SUBJ ARG has CONSCIOUS CHOICE and the verb is PERF, either ERG case or NOM case could be associated with it (§4:(27)-(28)). However, the ERG association principle is more specific. Therefore, by the Elsewhere Condition, it takes priority over the more general NOM case association principle. ERG case association also takes priority because it is language specific, and therefore preempts the universal default principle of NOM association. However, this pattern of preempting does not require independent formulation, because its effect follows from the Elsewhere Condition. Recall the assumption (§4.5) that case association takes place whenever the information specified by the principles becomes available. Direct case association, which is dependent on GF information, can take place only after this information is available. Indirect case association, on the other hand, does not require GF information, and therefore takes place prior to association of GFs. Given our assumption that associations once established cannot be eliminated, it follows that indirect case association preempts direct case association. Thus, principles that associate DAT, and LOC cases, as well as indirect GEN case, preempt NOM case association. These cases do not

Indirect Case on Subjects / 203

compete either with one another, or with ERG case, because the semantic configurations with which they are associated do not overlap. INST case association in the INST SUBJ construction, however, competes with and preempts both ERG association as well as NOM association. Finally, it must be noted that neither of the ideas of preempting that we have used in our account of case in Hindi require independent stipulation. The idea of a specific principle preempting the general one is a general principle of cognitive organization. The preempting of direct case association principles by indirect case association principles is a consequence of the structure-building derivation of predictable GF and case information within our organization of grammar.

8 Complex Predicates In Hindi, as in most South Asian languages, the phenomenon of complex predicates is extremely pervasive, and encompasses a wide range of predicate types.1 Nouns, adjectives, and nonfinite forms of verbs can all combine with verbs fairly productively to form complex predicates (§2:(7)). This chapter is concerned with complex predicates of the form nominal+verb.

8.1. Introduction 8.1.1. I illustrate in (1) a complex predicate (henceforth CP) of the form nominal+verb (henceforth N+V): 2 (1) raam-ne

niinaa-kii madad Ram-E Nina-G help-N Ram helped Nina.

kii. do-PERF

We call the N+V sequence madad kar ‘help’ in (1) a "complex predicate" because the clause structure of the sentence is determined not by the verb alone, but jointly by the N and the V. Thus, as we will see shortly, the number of arguments, their meanings, and their case may all be determined by either the N or the V. Because a portion of the predicative burden in an N+V CP construction is borne by the N, the V in a CP has often been referred 1 See Bahl (1974) for a detailed discussion of previous work on Hindi complex predicates. 2 Some typical examples of such complex predicates in English (Cattell (1984:1-3)) are: i. to make a dash ( ª to dash ) ii. to have a bath ( ª to bathe ) iii. to do a dance ( ª to dance ) iv. to give a gasp ( ª to gasp ) Hindi has a relatively large repertoire of light verbs that can combine with nominal hosts, some of the most common ones being kar 'do', h o 'be' / 'become' / 'happen', aa 'come', de 'give', k¥aa 'eat', l a g 'be struck', maar 'strike', and rak¥ 'keep'.

201

202 / Arguments in Hindi

to as a "light verb" (Jespersen (1954), Cattell (1984), Grimshaw and Mester (1988), and others).3 The element that the light verb combines with to form a CP has been called the "host". This is the terminology I follow. Although there has been some sporadic work on complex predicates in English over the years (for instance, Ross (1967), Chomsky (1975), Jackendoff (1972, 1974), Higgins (1974), Oehrle (1975), Wierzbicka (1982)), it is only in very recent years, perhaps since Cattell (1984), that the phenomenon has accumulated a large body of theoretical work.4 In the descriptive grammars of the Indo Aryan and Dravidian languages of South Asia, however, CPs have been an area of interest for a long time. In the literature on Hindi, they were a major concern in traditional grammars as early as Gilchrist (1796), Kellogg (1875), Platts (1898), and more recently in Guru (1922) and Sharma (1958), and also in Bailey (1956), Burton-Page (1957), Hacker (1961), Kachru (1966), Verma (1971), McGregor (1972), Hook (1974), Bahl (1974), and Masica (1976), to name just a few.5 8.1.2. The focus of this investigation is the relation between the semantic structure of CPs and their syntactic structures. I argue that, like the incorporation construction (§5.2.), the analysis of Hindi CPs reveals the need to recognize at least two distinct notions of "lexicality". We will see that the light verb in a CP can be separated from its nominal host, and can occur clause initially as the topic. Therefore, by the lexical integrity hypothesis, which "prohibits syntactic reorderings into and out of lexical categories" (Bresnan (1982a:54)), the CP cannot be a lexical unit. On the other hand, as we will see, the argument structure of a clause containing a CP is regulated jointly by the light verb and the nominal host. Hence, by the principle of direct syntactic encoding, by which all alterations of argument-grammatical function associations must take place lexically (Bresnan (1982a:6-8)), the CP must be a lexical unit. This conflict forces us to recognize the non-identity of the morphologically relevant lexical unit (henceforth CATEGORIAL WORD) referred to by the lexical integrity hypothesis, and the lexical unit relevant for ARG-GF association (henceforth FUNCTIONAL 3 Gilchrist (1796) called them "subservient" verbs. 4 For instance, Grimshaw and Mester (1988), Jayaseelan (1988), Huang (1989), Cole (1989), Mahajan (1989), Matsumoto (1989), Paolillo (1989), Sells (1990), Butt (1990), Ramchand (1990), among others. 5 In these works, complex predicates of the form verb+verb have often been referred to as "compound verbs", and those of the form noun+verb as "conjunct verbs".

Complex Predicates / 203

WORD). I will argue that within our organization of grammar, categorial word is a GC STR notion, whereas functional word is an ARG and GF STR

notion. We thus avoid the logical contradiction by factoring apart the two notions of "lexical", and representing them at different levels of structure. CPs in Hindi reveal other apparent contradictions as well. The N in the CP is a part of the predicate of the clause; along with the light verb, it can determine the valency of the CP construction, as well as the meanings and cases associated with the arguments of the construction. At the same time, it is an argument on par with the other arguments of the clause: firstly, it may passivize; and secondly, the light verb may agree with it. This dual nature of the nominal host is expressed within our system of representation through multiple association of the nominal with an ARG as well as the ARG-PRED at ARG STR. This chapter proceeds as follows. §8.2. explores the categorial structure of CPs in order to establish its non-lexicality with respect to the notion categorial word. §8.3. argues for the lexicality of the N+V CP with respect to the notion functional word. §8.4. demonstrates the argumenthood of the nominal host, and shows how the representations we develop correctly express the multi-faceted nature of the N+V CP . §8.5. and §8.6. spell out the theoretical consequences of this analysis of the Hindi CPs for levels of structure and the notion of lexicon. 8.1.3. Providing an account of N+V CPs requires explicit criteria for telling apart N+V sequences that are CPs and those that are not. There has been considerable difference of opinion on this issue in the literature on Hindi. On the one hand are N+V sequences that clearly are not CPs, as in (2). On the other hand are instances of N+V sequences that are uncontroversially accepted as CPs, as in (3): (2) a.

raam-ne apnaa homwark kiyaa. Ram-E self-G homework-N do-PERF Ram did his homework.

b.

niinaa-ne raam-ko kitaab Nina-E Ram-D book- N Nina gave Ram a book.

dii. give-PERF

204 / Arguments in Hindi

(3) a.

raam-ne mohan-par b ¥ a r o s a a k i y a a . Ram-E Mohan-L reliance-N do-PERF Ram relied on Mohan.

b.

niinaa-ne kahaanii-par d¥y a a n diyaa. Nina-E story-L attention-N give-PERF Nina paid attention to the story.

Compare the structure of the sentences in (2) and (3). In (2a), we have the doer, Ram, and the done thing, homework. Likewise in (3a); the done thing is the act of relying. But in contrast to (2a), (3a) has a third argument, Mohan. This argument could not have been licensed by the verb kar 'do' which is dyadic. The only plausible account, then, is that the argument is licensed by the noun b¥arosaa ‘reliance’. In other words, b¥arosaa kar 'rely' contributes to the number of arguments in the clause. Similarly, (3b) contains the giver, Nina, the givee, the story, and the given thing, attention, exactly as in (2b). However, recall that the givee (the goal of giving) is normally in DAT case, for instance, raam-ko in (2b). But the givee in (3b) is in LOC case, with the marking -par 'at'. We have already seen that LOC-par is systematically associated with the semantic configuration [AT X] (§7.3.). This semantic configuration is not part of the SEM STR of the verb de 'give'. Therefore de cannot licence LOC case. Hence, we must assume that the LOC case in (3b) is licensed by d¥yaan de 'pay attention'. In other words, d¥yaan ‘attention’ contributes to the SEM STR associated with an ARG, and thereby, its case. Thus, the N in the N+V sequences in (3) has the capacity to constrain the number, meaning, and case of the ARGs in the clause.6 This capacity is normally characteristic of predicates. Therefore, b¥arosaa ‘reliance’ and d¥yaan ‘attention’ must be predicates, though categorially Ns. Under the assumption that a complex predicate is a unit composed of more than one predicative element, the N+V sequences in (3) must be CP s. Such uncontroversial CPs provide the basis for establishing the structural properties of CPs required for making our arguments.7 That the N can govern the argument structure of the CP does not imply that the V has nothing to contribute to the argument structure. Consider the 6 This is similar to what is suggested in Sells (1990) for Japanese. 7 These structural properties may then be used to construct diagnostics t o determine the complex predicatehood of unclear instances, even though I do not undertake this task here.

Complex Predicates / 205

difference between the light verbs ho ‘happen/become’ in (4a)/(5a) and kar ‘do’ in (4b)/(5b): (4) a.

kamre-kii safaaii hui. room-G cleaning-N happen/become-PERF The room got cleaned.

b.

raam-ne kamre-kii safaaii Ram-E room-G cleaning-N do-PERF Ram cleaned the room.

a.

kamraa saaf huaa. room-N clean happen/become-PERF The room became clean.

b.

raam-ne kamraa saaf Ram-E room-N clean do-PERF Ram cleaned the room.

kii.

(5)

kiyaa.

Whereas the light verb kar ‘do’ sanctions an agentive ARG in (4b) and (5b), the light verb ho ‘happen/become’ in (4a) and (5a) is incapable of doing so. In short, in Hindi, light verbs can also determine the ARG STR of the complex predicates, and the SEM STR entities associated with the ARGs, counter to what has been claimed in Grimshaw and Mester (1988) for the light verb suru in Japanese.8

8.2. The Category Structure of Complex Predicates The point of this section that is crucial to my subsequent arguments is that the light verb and its nominal host are concatenated phrasally; they do not form a categorial word (§8.2.2). In order to lend substance to this claim, we will try to characterize the category structure of CP s. For the sake of concreteness, I wil try to articulate the structure in detail, showing that the N in a CP (a) is not a direct daughter of S , but forms a constituent with the light verb (§8.2.1.); and (b) is itself a lexical category (§8.2.3.).

8

Also see Sells (1990) and Matsumoto (1989) on suru in Japanese.

206 / Arguments in Hindi

8.2.1.

The Nominal Scrambling

as Part of the Verbal Constituent:

Recall from §2.3. that in Hindi, all and only direct daughters of S may scramble. Consider (2a, b), which do not involve CP s, repeated below for easy reference: (2) a.

raam-ne apnaa homwark kiyaa. Ram-E self-G homework-N do-PERF Ram did his homework.

b.

niinaa-ne raam-ko kitaab Nina-E Ram-D book- N Nina gave Ram a book.

dii. give-PERF

Some of the various possible word orders of (2a) and (2b) are given in (6a,b) and (6c-e) respectively. In (6), crucially, the nominal immediately preceding the verb in the unmarked order in (2) need not appear in the immediately preverbal position: (6) a. b. c. d. e.

homwark raam-ne niinaa-ne kitaab niinaa-ne

raam-ne kiyaa. kiyaahomwark. kitaab raam-ko dii. niinaa-ne raam-ko dii. raam-ko dii kitaab.

Now consider the CP constructions in (3), repeated here for convenience: (3) a.

raam-ne mohan-par b ¥ a r o s a a k i y a a . Ram-E Mohan-L reliance-N do-PERF Ram relied on Mohan.

b.

niinaa-ne kahaanii-par d¥y a a n diyaa. Nina-E story-L attention-N give-PERF Nina paid attention to the story.

Complex Predicates / 207

In (3), all the nominals with the exception of the one internal to the CP enjoy freedom of word order just like those in (2). This is illustrated by (7)(8): (7) a. b. c.

mohan-par raam-ne mohan-par

raam-ne b¥arosaa k i y a a . b¥arosaa k i y a a mohan-par. b¥a r o s a a kiyaa raam-ne.

a. b. c.

kahaanii-par niinaa-ne d¥yaan kahaanii-par d¥yaan d i y a a niinaa-ne d¥yaan d i y a a

(8) diyaa. niinaa-ne. kahaanii-par.

The nominal within the CP, however, is not free to scramble, as shown by the ungrammaticality of (9) and (10): (9) a. b. c. d. e.

* raam-ne b ¥arosaa mohan-par k i y a a . * b¥arosaa raam-ne mohan-par kiyaa. * mohan-par b¥a r o s a a raam-ne k i y a a . * mohan-par raam-ne kiyaa b ¥ arosaa . * mohan-par k i y a a raam-ne b ¥ a r o s a a .

a. b. c. d.

* niinaa-ne d¥yaan kahaanii-par diyaa. * d¥yaan niinaa-ne kahaanii-par diyaa. * kahaanii-par d¥yaan niinaa-ne d i y a a . * niinaa-ne kahaanii-par diyaa d¥yaan.

(10)

The facts in (9)-(10) can be accounted for by assuming that the nominal host of a light verb is not a direct daughter of S, but forms a categorial constituent with the light verb: (11)

S

\

CAT: V

/\

CAT: N

b¥arosaa

CAT: V

kar

208 / Arguments in Hindi

The point of the structure in (11) is to encode that the CP-internal nominal is not a direct daughter of S, but is a part of the verbal constituent. In the following sections, we will determine whether the various nodes in (11) are lexical or phrasal categories, and spell out further details of the structure. 8.2.2. A CP as a Phrasal Category : Topicalization 8.2.2.1. A topic in Hindi occurs clause initially (Kachru (1980:130). In this section, I will show that unlike scrambling, topicalization in Hindi is not restricted to the direct daughters of S. Even though the light verb in a CP cannot be scrambled, it can be topicalized, providing evidence that a CP is not a categorial word. The difference in the behavior of the light verb with respect to scrambling and topicalization calls for a brief discussion of the two phenomena. Consider the following sentences, in which the genitive modifier of a noun phrase cannot scramble (§2.3.), but may occur clause initially through topicalization: (12) a.

aaj raat raam m o h a n - k i i today nightRam-N Mohan-G Ram will read Mohan's book tonight.

kitaab book- N

pa±¥egaa. read-FU

b. * aaj raat mohan-kii raam k i t a a b today nightMohan- G Ram-N book- N

pa±¥egaa. read-FU

c.

pa±¥egaa. read-FU

m o h a n - k i i aaj raat raam k i t a a b Mohan-G today nightRam-N book- N Ram will read Mohan's book tonight.

The GEN nominal in (12b) cannot be separated from its head through scrambling, because the genitive is not a direct daughter of S. The displacement of the GEN nominal in (12b) cannot be analysed as a result of topicalization either, because a topic must occur clause initially, as in (12c). Hence the ungrammaticality of (12b). In contrast, (12c) is perfectly grammatical; the GEN nominal is topicalized. Another difference between scrambling and topicalization is that while the former is not clause-bounded, the latter is:9 9 The examples of scrambling across clauses, given in Chapter 2 (§2:(16ac)), are repeated below:

Complex Predicates / 209

(13) a.

raam-ne siitaa-se Ram-E Sita-COM

mohan-kii Mohan-G

kitaab book-N

pa±¥ne-ko read-NF

kahaa. tell-

PERF

Ram told Sita to read Mohan's book. b.

mohan-kii

c.

raam-ne

kitaab siitaa-se

mohan-kii

kitaab

pa±¥ne-ko siitaa-se

raam-ne kahaa.

pa±¥ne-ko

kahaa.

(13b) and (13c) are instances of scrambling: raam-ne , siitaa-se, mohan-kii kitaab, pa±¥ne-ko and kahaa are direct daughters of the same S (cf. §2:(16)). Now consider (14), which shows that topicalization is clausebounded: (14) * mohan-kii Mohan-G

raam-ne Ram-E

siitaa-se k i t a a b Sita-COM book- N

pa±¥ne-ko read-NF

kahaa. tell-

PERF

Although the GEN modifier can be displaced to the initial position through topicalization in (12c), this is disallowed in (14). This restriction does not apply to scrambling, as shown by (13b, c). 8.2.2.2. We now turn to topicalization in CP constructions. The light verb in a CP may be displaced from the CP to the clause initial position for special discourse effects, as illustrated in (15): (15) k i y a a

raam-ne mohan-par b¥arosaa. do-PERF Ram-E Mohan-L reliance-N Ram relied on Mohan.

The light verb cannot be displaced from the CP to a non-initial position (see ((9e)), demonstrating that it is not a constituent that can be scrambled.

(i)

ilaa-ne Ila-E

mohan-se raam-kaa s a a m a a n kamre-mã Mohan-I Ram-G luggage-N room-L rak¥ne-ko kahaa. keep-to tell-PERF Ila told Mohan to keep/put Ram's luggage in the room. (ii) mohan-se raam-kaa saamaan ilaa-ne kamre-mã rak¥ne-ko kahaa. (iii) mohan-se raam-kaa saamaan kamre-mã ilaa-ne rak¥ne-ko kahaa.

210 / Arguments in Hindi

Therefore, (15) must be analysed as an instance of topicalization, not scrambling.10 By the lexical integrity hypothesis which disallows subconstituents of a lexical unit from undergoing syntactic movement, the N and the V of a CP cannot constitute a lexical unit. That is, it is not a categorial word. If all X0 categories are lexical units, then the CP is not an X0 category. 8.2.3.

The Nominal as a Lexical Category

It must be noted that even though the light verb in a CP can be topicalized, the nominal host cannot be topicalized by itself ((16)): (16) *

b ¥ a r o s a a raam-ne mohan-par kiyaa. reliance-N Ram-E Mohan-L do-PERF Ram relied on Mohan .

In order to account for this, we might say that the nominal host is a lexical category, and hence, not a maximal projection. The ungrammaticality of (16) would then follow from the general assumption that only maximal projections can topicalize. Below, I provide evidence from adjectival modification, conjoining, wh-questions, and relativization that converges on the analysis of the nominal host as a lexical category. 8.2.3.1. Adjectival Modification CP-internal nominals allow no modification, adjectival or otherwise, unless

of course it is adverbial modification of the entire predicate. Take the pairs of sentences in (17)-(18). The subject nominal in (17a) and the object nominal in (18a) are modified by an adjective and a numeral respectively. In (17b) and (18b), the same nominals, when internal to a CP, cannot be modified:11 (17) a.

aniyantrit krod¥ manu‹ya-ko unrestrained anger-N man-D Unrestrained anger will ruin man.

b¥ra‹Æ ruined

b.

ilaa-ko

aayaa.

( * aniyantrit ) krod¥

kar degaa. do-NFgive-FU

10 This correctly predicts that the light verb cannot be topicalized out of an embedded clause. 11 In the terminology of Grimshaw (1988), the same nominal is "thetamarking" in (a) and "non-theta-marking" in (b) in (17)-(18).

Complex Predicates / 211

Ila-D unrestrained (Ila got uncontrollably angry.)

anger-N

come-PERF

(18) a.

mohan-ne raam-ko e k k¥abar Mohan-E Ram-D one news-N Mohan gave Ram a (piece of) news.

b.

mohan-ne raam-ko ( * ek ) Mohan-E Ram-D one Mohan informed Ram.

dii. give-PERF

k¥abar news-N

kiyaa. do-PA

The restriction that a CP-internal nominal cannot be modified follows directly from the assumption that it is a lexical category. If syntactically modified, the composite form will no longer be a lexical category. 8.2.3.2. Conjoining The facts of conjoining in Hindi would also follow naturally if we assumed that the nominal host of a light verb is a lexical category. Consider the conjoining in (19): (19) a.

raam-ko kitaab aur pensil milii. Ram-D book- N and pencil-N get-PERF Ram got a book and a pencil.

b.

raam-ne homwark aur eksarsaais kiyaa. Ram-E homework-N and exercise-N do-PERF Ram did (his) homework and exercise.

This is not possible when the entities to be conjoined are nominal hosts in CPs: (20) a. * raam-ko kahaanii yaad aur p a s a n d aaii. Ram-D story-N memory-N and liking- N come-PERF Ram remembered and liked the story. b. * iilaa-ne Ila-E

raam-ko Ram-A

k‹ama pardon-N

PERF

Ila pardoned and accepted Ram.

aur sviikaar kiyaa. and acceptance-N do-

212 / Arguments in Hindi

To account for the contrast between (19) and (20), a plausible hypothesis is that in Hindi, lexical categories cannot be conjoined. The hypothesis is supported by (21):

Complex Predicates / 213

(21) laal kitaabã red (i) (ii)

aur kursiiyãã book-PL-N and chair-PL-N Chairs and red books. * Red books and red chairs.

Given our assumption that N in Hindi is a maximal projection, the structure of the noun phrase with an adjective will be: [N A N ]. The structure corresponding to (21.i) is: [[N laal kitaabã ] aur [N kursiiyãã ]]. The structure corresponding to (21.ii) would be: [N laal [N kitaabã ] aur [N kursiiyãã ]]. The latter structure is ruled out by the condition against conjoining lexical categories. This account correctly predicts that postpositions cannot be conjoined: (22)

* bacce-ke liye aur dvaaraa child-G for and through (Intended meaning: Through and for the child.)

8.2.3.3. Wh- Questions Wh - interrogatives in Hindi are formed by using the appropriate wh - word in situ, as illustrated in (23b-c) and (24b-d): (23) a.

raam-ne bahut kaam kiyaa. Ram-E much work-N do-PERF Ram did a lot of work.

b.

kis-ne bahut kaam kiyaa? who-E much work-N do-PERF Who did a lot of work?

c.

raam-ne kyaa Ram-E what do-PERF What did Ram do?

a.

niinaa-ne ravii-ko kitaab dii. Nina-E Ravi-D book- N give-PERF Nina gave Ravi a book.

b.

kis-ne ravii-ko kitaab dii? who-E Ravi-D book- N give-PERF Who gave Ravi a book?

kiyaa?

(24)

214 / Arguments in Hindi

c.

niinaa-ne k i s - k o kitaab dii? Nina-E who-D book- N give-PERF Who did Nina give a book to?

d.

niinaa-ne ravii-ko k y a a diyaa? Nina-E Ravi-D what give-PERF What did Nina give Ravi?

The nominal hosts in CPs do not yield wh-questions. (25b) and (26b) are thus unacceptable as interrogative counterparts of (25a) and (26a): (25) a.

ravii-ne mohan-par b ¥ a r o s a a kiyaa. Ravi-E Mohan-L reliance-N do-PERF Ravi relied on Mohan.

b.* ravii-ne mohan-par k y a a kiyaa? Ravi-E Mohan-L what do-PERF (*What did Ravi do on Mohan?)

(26) a.

niinaa-ne kahaanii-par d¥y a a n diyaa. Nina-E story-L attention-N give-PERF Nina paid attention to the story.

b.* niinaa-ne kahaanii-par kyaa diyaa? Nina-E story-L attention-N give-PERF (*What did Nina give to the story?)

If we assume that wh- phrases (kyaa ‘what’, kis-ne ‘who-E’, kaun ‘who’, etc., in Hindi) are maximal projections (X), and that the nominal host in a CP is a lexical category, the facts in (25b) and (26b) follow directly. 8.2.3.4.

Relativization

The behavior of CP-internal nominals with respect to relative clauses also supports the assumption that the nominal is a lexical category. The nominal host in a CP cannot be replaced by the gap of a relative clause, as shown by (28), in contrast to ordinary nominals, as in (27):

Complex Predicates / 215

(27) a.

[vah eksarsais [ jo raam-ne ——— kiyaa ]] bahut mußkil t¥aa that exercise-N that Ram-E do-PERF very difficult be-PA The exercise that Ram did was very difficult.

b.

[vah kitaab [ jo niinaa-ne raam-ko ——— dii ]] mere paas hai. that book-N that Nina-E Ram-D give-PERF I-G near be-PR The book that Nina gave Ram is with me.

(28) a. * [vah that

b¥arosaa [ jo raam-ne mohan-par ——— kiyaa ]]… reliance-N that Ram-E Mohan-L do-PERF

b. * [vah d¥yaan [ jo niinaa-ne kahaanii-par ——— diyaa ]] ... that attention-N that Nina-E story-L give-PERF

Another strategy for relativizing in Hindi is using the relative marker jo to mark the relativized NP, with a pronoun as the head of the relative clause, as illustrated in (29): (29) a.

raam-ne jo eksarsais kiyaa, vah bahut mußkil t¥aa. Ram-E that exercise-N do-PERF it very difficult be-PA The exercise that Ram did was very difficult.

b.

niinaa-ne raam-ko jo kitaab dii , vah mere paas hai. Nina-E Ram-D that book-N give-PERF it I-G near be-PR The book that Nina gave Ram is with me.

CP-internal nominals cannot be relativized in this manner either:

(30) a.* raam-ne Ram-E

kahaanii story-N

j o yaad kii, that memory-N do-PERF

b.* niinaa-ne Nina-E

kahaanii-par story-L

vah ... it

j o d¥y a a n diyaa, that attention-N give-PERF it

vah ...

It might be argued that nominals that can host light verbs are prevented by their inherent abstract nature from being relativized. However, consider (31b), where the unrelativizable nominal in (30a) becomes relativizable when it is not part of a CP:

216 / Arguments in Hindi

(31) a.

b.

raam-kii yaad mãã-ko sataatii Ram-G memory-N mother-A torture-NF The memory of Ram tortures mother.

hai. be-PR

jo yaad mãã-ko sataatii hai vah hame b¥ii ... that memory-N mother-A torture-NF be-PR it we-A The memory that tortures mother, it (...) us also.

also

We must conclude that the failure of the nominal to relativize in (30a) is because it is internal to a CP. Under the assumption that only maximal projections can be relativized, the failure of the nominal within CPs to participate in relativization further converges on the analysis of the nominal as a lexical category. In short, evidence from adjectival modification, conjoining, wh - questions, and relativization point to the conclusion that the N in a CP is a lexical category.12, 13 This conclusion runs counter to the assumption in many syntactic theories that an argument must be maximal projection. As we will see in §8.4., the nominal host of the light verb must be an ARG, even though it is a lexical category. 8.2.4.

Summary

In the preceding discussion, we used scrambling to demonstrate that the nominal host of a light verb is not a direct daughter of S . Topicalization shows that the CP is not a categorial word. The impossibility of adjectival modification, conjoining, wh -question formation, and relativization show that the nominal host is a lexical category. Recall that the inability of the nominal host to topicalize follows from the assumption that only maximal projections can topicalize. Needless to say, the light verb, which can topi12 One might argue that the facts of wh -questions and relativization could be made to follow by treating the nominal host as a non-lexical category that is not a maximal projection. The facts of adjectival modification cannot, however, be made to follow from such an assumption. As mentioned earlier, I assume two bar levels for Hindi: X0 and XMAX. 13 The conclusion that the nominal within a CP is a lexical category is not t o be taken as applying to CP -internal nominals across languages. In Malayalam, an areally related language of South Asia, for instance, the CP -internal nominal allows modification by adjectives, and even relative clauses, and is analysed as a maximal projection (K. P. Mohanan (1983)). Hence, the evidence presented above must be viewed as a structural regularity that allows language specific variation.

Complex Predicates / 217

calize, must then be a maximal projection. The categorial structure of N+V CPs that emerges from these conclusions is the following:14 (32)

S \ N

V /\

V

8.3. Predicatehood in Complex Predicates The previous section concentrated on developing the categorial structure of CPs, and showing that a CP is not a categorial word. In this section, we will focus on the relation between their SEM STR and ARG STR representations, and show that a CP is a functional word. The section falls into two parts. The first part develops a formal representation to express the joint predicatehood of the N and the V in the CP. The second part provides evidence for this representation from gapping in coordinate structures and other ellipses. 8.3.1. The Nominal Host as an Independent Predicate 8.3.1.1. We saw in §8.1.3. that the SEM STR and ARG STR of the CP construction is determined jointly by the nominal host and the light verb. In order to develop formal representations of the CP construction such that they reflect this property, let us explore in detail a single light verb, aa 'come', in combination with various nominal hosts. As a full verb, aa 'come' has two arguments, the entity that comes, and its locative destination: (33) raam

mere g¥ar aayaa. Ram-N I-G house-L come-PERF Ram came to my house.

Let us schematize the skeletal SEM STR of the verb aa ‘come’ in (33) as in (34). An accurate SEM STR must separate the configurations [TOWARDS X] and [AT/IN X] which I have collapsed as [TO X]. It must also specify that the 14 The categorial structure in (32) may appear to wrongly allow CP formation to be recursive. However, CP formation requires a "light verb", that is, a semantically bleached predicate. When a CP has been formed, the predicate is not bleached any more, and therefore does not allow further CP formation. Thus, the undesirable recursion is prevented by the SEM STR.

218 / Arguments in Hindi

verb ‘come’ involves the direction of movement towards the deictic center, which marks the speaker's actual or empathized position.15 I ignore these details in (34). (34) [

TO Y MOVE ]SPATIAL SEM STR …………………………………………………………………………

X

˘

ARG1

ARG2

¿

A-PRED

ARG STR

Recall from §7.3.3. the idea of semantic fields: (34) involves a spatial field. Also recall that the subscripts on the ARGs in (34) indicate their relative prominence, governed by relative prominence statements on the SEM STR entities they are associated with. X in (34) is the L-SUBJ, and by default, the SUBJ (§3: (43a)). Now look at the same verb aa 'come' in the CP constructions in (35): (35) a.

mohan-ko h a s i i aaii. Mohan-D laughter-N come-PERF Mohan had the urge to laugh. (Lit.: Laughter came to Mohan.)

b.

mohan-ko kahaanii y a a d aaii. Mohan-D story-N memory-N come-PERF Mohan remembered the story. (Lit.: Memory of the story came to Mohan.)

c.

mohan-ko bacce-kaa d¥y a a n aayaa. Mohan-D child-G thought-N come-PERF Mohan thought of the child. (Lit.: Thought of the child came to Mohan.)

d.

mohan-ko raam-par krod¥ aayaa. Mohan-D Ram-L anger-N come-PERF Mohan was angry with Ram. (Lit.: Anger on Ram came to Mohan.)

In (35a), exactly as in (33), we have two participants, the entity that comes, namely, laughter, and its destination, namely, Mohan. But the semantic field in (35) is different from that in (33). The entity that moves is 15

See Sinha (1972) for a discussion of the deixis of the verb ‘come’ in Hindi.

Complex Predicates / 219

an abstract mental state, and requires a sentient destination for it to come to rest at. In short, in contrast to the locational goal in (33), (35a) requires a sentient goal. Consequently, by the principles of relative prominence (§3:(13b)), it is higher on the argument hierarchy than the entity that moves. It is therefore the L-SUBJ, and by default, the SUBJ. A skeletal SEM STR and ARG STR representation of the light verb aa ‘come’ in (35a) would involve extending the spatial semantic field in (34) to the abstract field. This can be done by eliminating the field specification from the representation of the full verb ((34)), to derive the representation of light verb by adding the specification of abstract field ((36)): (36) [

TO Y MOVE ]ABSTRACT SEM STR …………………………………………………………………………

X

˘

ARG2

ARG1

¿

A-PRED

ARG STR

When the verb is extended into the abstract field, and the destination is a sentient entity, the relative prominence of the ARGs gets realigned by the principles of relative prominence (§3:(13)). This realignment in (36) is reflected in (35a) in both the default association of ARGs to GFs, as well as the canonical word order of the sentence. ‘Mohan’ (i.e. X ) is the SUBJ in (35a). It precedes ‘laughter’ (i.e. Y ) in the canonical word order. Both these properties of (35a) follow as a natural consequence of the relative prominence of the ARGs in (36). Now, although the semantic and syntactic properties of a light verb in a CP are not identical to those of its full independent counterpart, it is clear that there is a systematic synchronic relation between the two.16 In what follows, I make the methodological assumption that this relation, wherever it exhibits regularity, must be formally expressed in the grammar. One way of expressing the relation is to assume that specifications in the SEM STR of the full verb may be deleted in the SEM STR of the light verb, or new specifications may be added. For instance, the specification of the semantic field

16 The process of a verb becoming 'light' has been described in Masica (1976:141) as a "shift in the semantic center of gravity" from the light verb t o its host, "with concomitant lexical emptying or grammaticalization" of the light verb. The extension of the spatial field to the abstract field in (34)-(36) corresponds to the intuition of "emptying".

220 / Arguments in Hindi

in (34) is eliminated to extend the verb to non-spatial fields, and the specification of abstract field is added in (36).17 The lexical representation of the predicative nominal hasii ‘laughter’ in (35a) may be given as in (37): (37) [

laughter ] SEM STR …………………………………………………………………………

Y

˘

ARG

¿

A-PRED

ARG STR

The representations in (36) and (37) may be combined to derive the representation of the CP hasii aa ‘laughter come’ in (38). The participants of the semantic predicate MOVE in the SEM STR of (38) have been reordered merely for ease of understanding: (38) [

TO Y

MOVE ]ABSTRACT

X

SEM STR

[ Y laughter ] ………………………………………………………………………… ˘

ARG1

ARG2

¿

A-PRED

ARG STR

In (35b, c, d), not only is the semantic field an abstract one as in (35a), but there is an additional argument, not licensed by the verb aa 'come'. The additional ARG must therefore be an ARG of the N in the CP. This ARG is associated with a different case in each of these sentences: NOM in (35b), GEN in (35c), and LOC in (35d). The semantic configurations that determine the case associated with these additional ARGs must also therefore be distinct. The representation of the predicative nominal krod¥ ‘anger’ in (35d) is given in (39a), and that of the CP krod¥ aa ‘anger come’ in (39b): (39) a.

[ Y AT Z anger ] SEM STR ……………………………………………………………………… ˘

ARG1 ARG2

¿

A-PRED

ARG STR

17 I believe that the possible deletions and additions of specifications are strictly governed by a set of constraints. However, given those constraints, the relation between a full and light pair of verbs cannot be derived by rule. The relation is non-random only to the extent that it is subject to the set of constraints on possible deletions and additions. A precise characterization of such a set of constraints remains a rich domain for further investigation.

Complex Predicates / 221

b.

[

TO Y

MOVE ]ABSTRACT SEM STR

X

[ Y AT Z anger ] ……………………………………………………………………… ˘

ARG1

ARG3

ARG2

¿

A-PRED

ARG STR

In the representation in (39b), the ARG associated with the participant Z is dependent on the N in the CP , unlike the participant Y , which is shared by the two predicates. It is therefore not surprising that the case of the ARG is also dependent on the N, as also in (3a), (35b), and (35c). The ARG solely dependent on the N in (35d)/(39b) is associated with Z in the configuration [AT Z]. Therefore, the case associated with it is LOC-par (§7:(56a)). The corresponding ARG in (35c) is not associated with any of the semantic configurations associated with the various semantic case features. Therefore, it is associated with the default indirect non-semantic case assigned by the nominal, namely, GEN case. 8.3.1.2. Similar to the representations of the verb aa ‘come’, a representation of the verb kar ‘do’ in (2a)/(3a) might be given as in (40): (40) [

Y do ] SEM STR …………………………………………………………………………

X

˘

ARG1 ARG2

¿

A-PRED

ARG STR

The SEM STR and ARG STR of b¥arosaa ‘reliance’ ((3a)) is given in (41): (41) [

ON Z reliance ] SEM STR …………………………………………………………………………

X

˘

ARG1

ARG2

¿

A-PRED

ARG STR

The example in (3a) with the CP b¥arosaa kar ‘to rely’ is repeated below as (42a). The representations of the predicative nominal ((41)) and the verb ((40)) are combined in (42b) to get the representation of the CP: (42) a.

raam-ne mohan-par b ¥ a r o s a a k i y a a . Ram-E Mohan-L reliance-N do-PERF

222 / Arguments in Hindi

Ram relied on Mohan. b.

[

X

[ ˘

do ]

Y

SEM STR

X ON Z reliance ] ……………………………………………………………………

ARG1

ARG3

ARG2

¿

A-PRED

ARG STR

In the representation in (42b), there are two SEM-PREDs, b¥arosaa ‘reliance’ and kar ‘do’, with two participants each. However, there is only one ARGPRED, b¥arosaa kar ‘rely’, and three ARGs. Thus, the CP is a single ARGPRED formed through the fusion of two SEM-PREDs. 8.3.1.3. In sum, the N in a CP can contribute to the number of ARGs of the entire predicate, and the meaning and case associated with the ARGs. This property may be viewed as the uncontroversial overt symptom of complex predicatehood. It is expressed in the representation of CPs by assuming that the nominal host is an independent SEM-PRED which constitutes part of an ARG-PRED. Most treatments of complex predicate formation involve the combination of two predicates in syntax. The combination is done in terms of syntactic restructuring (Chomsky (1975)), merger of subcategorization frames (Jackendoff (1974)), theta-role promotion (Jayaseelan (1988)), “Argument Transfer" (Grimshaw and Mester (1988)), or control (Huang (1989)). The analysis of the N+V CPs in Hindi presented here departs from other treatments of complex predicates in a significant way. I assume that the Hindi N+V CPs involve the combination of the SEM STRs of the two predicates, yielding an ARG STR clause with a single ARG-PRED, that is, a single "functional word".18 This analysis crucially relies on the separation of traditional theta role information into SEM STR and ARG STR, and the consequent provision for multiple associations between SEM STR and ARG STR. If theta role information is not thus factored into two separate levels of structure, 18 Lefebvre (1988) proposes an analysis of serial verb constructions in Fon, and Mahajan (1989), an analysis of V+V CPs in Hindi, that are somewhat similar in spirit to the analysis of N+V CPs given here. They show that the predicates must be combined at "Lexical Conceptual Structure", in order to account for the regularities in combination. It is not clear to me, however, if the complex predicates constitute single predicates at a level corresponding to ARG STR i n their analyses.

Complex Predicates / 223

the consequence would be the need for multiple associations between arguments and grammatical functions, and thus, a violation of FunctionArgument Biuniqueness (§3:(40)), which is stated on the relation between ARG STR and GF STR. 8.3.1.4. As an alternative to (42), in which there is a single ARG-PRED, one might wish to analyse CPs as involving the concatenation at ARG STR of two ARG-PREDs. In such an analysis, the multiple association in b¥arosaa kar ‘rely’ as in (42) would be replaced by a coindexing formalism of the type used, for instance, in Alsina and Mchombo (1989) and Alsina (1990). Such a representation is given in (43): (43) [

X

do ]

Y

SEM STR

[

reliance ] ………………………………………………………………………… ˘

X

ARGi

ON Z

ARG

˘

ARGi

ARG

¿

¿

A-PRED

ARG STR

A-PRED

Now, the prediction that (43) makes is that since it has two ARG STR CLAUSEs, it should have two L-SUBJs as well. In that event, both L-SUBJs must be eligible antecedents of the reflexive apnaa (§6:(16)). In the example in (43), the two L-SUBJs are co-indexed, and therefore the prediction cannot be verified. A test case would be a CP construction in which the L-SUBJs of the light verb and the host are not coindexed. Then they must both be eligible antecedents of the reflexive. I have not been able to find appropriate test structures among N+V CPs. However, ADJ(ECTIVE)+V CPs show this prediction to be incorrect: (44) a.

ilaa apne gaane-se k¥uß huii. Ila-N self-G song- I happy become-PERF Ilai became happy with self'si song.

b.

raam-ne ilaa-ko apne gaane-se k¥uß kiyaa Ram-E Ila-A self-G song- I happy do-PERF Ramj made Ilai happy with self'sj/*i song.

As shown by (44a), k¥uß ‘happy’ is a monadic adjective. In (44b), the light verb kar ‘do’ provides an additional agentive argument, which is not merged

224 / Arguments in Hindi

with the experiencer-goal of happiness. If the representation of k¥uß kar had two ARG STR CLAUSEs, it should be possible for the reflexive in (44b) to also refer to Ila, the L-SUBJ of k¥uß. Since this is not possible, I conclude that Ila is not an L-SUBJ, and hence, that the CP has a single ARG-PRED.19 There are two further reasons why the analysis in (42b) is preferrable to the analysis in (43). First, as stated above, (43) has two ARG STR clauses, while (42b) has only one. However, the sentence in (42a) has a single GF STR clause. Therefore, the representation in (43) entails a "clause union" in the mapping between ARG STR and GF STR. That is, two ARG STR clauses would correspond to a single GF STR clause. The representation in (42b), on the other hand, does not involve such a clause union. Therefore, if we adopt the analysis in (42b), we may universally prohibit the merger of two ARG STR clauses into a single GF STR clause, thereby making the strong prediction that no language would exhibit structures that have two L-SUBJs but only one SUBJ. In the absence of counterexamples to this prediction, we must prefer the more restrictive theory that prohibits such a merger, which forces us to adopt (42b) in favour of (43) as the representation for (42a). Second, (43) employs the coindexing device in (45a), which is, in fact, a notational variant of the multiple linking device in (45b). The latter is explicitly prohibited by the function-argument biuniqueness principle (Bresnan (1982b:163)): (45) a.

ARGi ARGi GF

b.

ARG ARG

GF

ARG STR

GF STR

ARG STR

GF STR

If (45b) is prohibited, (45a) should also be prohibited. If not, the prohibition against (45b) becomes empirically vacuous: every instance of (45b), which the theory forbids, can be trivially translated as (45a), which the coindexing device sanctions.

19 Recall that the "merger" in causative construction in Hindi, as in the case of the complex predicate construction, requires a "monoclausal" ARG STR, as the causative construction in Hindi does not have two L-SUBJs either.

Complex Predicates / 225

Thus, the advantages of adopting the representation of CPs in (42b) are the following. First, it allows us to account for why there is only one LSUBJ in CP constructions in Hindi ((44b)). Second, it allows us to maintain function-argument biuniqueness. Third, it allows us to forbid the "argument merger" (i.e., two ARGs corresponding to a single GF) in (43), and thereby make the strong universal prediction that there cannot be a structure with two L-SUBJs but one SUBJ. As we will see in the concluding section, this prohibition may be viewed as the empirical substance of the direct syntactic encoding principle, proposed in (Bresnan (1982a)). 8.3.2. The Complex Predicate as a Single Predicate In what preceded, we saw that the nominal host in a CP is a SEM-PRED. We also argued that the CP is a single ARG-PRED. In this section, we provide further evidence for this assertion from the facts of gapping in coordination, and from ellipsis in answers to yes-no questions. 8.3.2.1. Gapping It appears that in Hindi, you can gap the syntactic predicate of a clause, an argument of the predicate, or the head of an argument. Gapping of a predicate is shown in (46): (46) a.

raam-ko kitaab milii, aur niinaa-ko pensil Ram-D book- N get-PERF and Nina-D pencil-N Ram got i a book and Nina ——— i a pencil.

b.

raam-ne

homwark

kiyaa,

aur

———.

niinaa-ne eksarsaais

———. Ram-E homework-N do-PERF and Nina-E exercise-N Ram didi (his) homework, and Nina ——— i (her) exercise.

In (47), an argument is gapped: (47) ilaa-ne Ila-E

laal kitaab k¥ariidii, aur niinaa-ne ——— red book- N buy-PERF and Nina-E

PERF

Ila bought [a red book]i , and Nina sold ——— i .

and in (48), the head of an argument:

becii. sell-

226 / Arguments in Hindi

(48) ilaa-ne

laal kitaab k¥ariidii, aur red book- N buy-PERF and niinaa-ne niilii ——— becii. Nina-E blue sell-PERF Ila bought a red booki , and Nina sold a blue ——— i . Ila-E

(49) shows that the nonhead in an NP cannot be gapped. The adjective in the first conjunct cannot be interpreted as being gapped in the second conjunct. (49) ilaa-ne Ila-E

laal kitaab k¥ariidii, aur red book- N buy-PERF and niinaa-ne laal /* ——— pensil becii. Nina-E red pencil-N sell-PERF Ila bought a redi book, and Nina sold a red/ * ——— i pencil.

Consider the coordination of CP constructions: both conjuncts in (50a, b) have the same light verb, and both conjuncts in (51) have the same nominal host. You can gap neither the verb alone ((50)), nor the noun alone ((51)): (50) a.

raam-ko Ram-D

kahaanii yaad aaii, aur story-N memory-N come-PERF and niinaa-ko kavitaa pasand aaii / * ———. Nina-D poem-N liking-N come-PERF Ram remembered the story, and Nina liked the poem.

b.

ilaa-ne raam-ko k‹ama kiyaa, aur Ila-E Ram-D pardon-N do-PERF and niinaa-ne us-kaa svaagat kiyaa / * ———. Nina-E pron-G welcome-N do-PERF Ila pardoned Ram, and Nina welcomed him.

(51) raam-ne

bacce-par d¥yaan rak¥aa, aur niinaa-ne child-L attention-N keep-PERF and Nina-E ilaa-kii aavaaz-par d¥yaan/ * ——— diyaa. Ilaa-G voice-L attention- N give-PERF Ram kept his attention i on the child, and Nina paid attention/*— — — Ram-E

i to

Ila's

voice.

The entire complex predicate, of course, can be gapped, as we would expect:

Complex Predicates / 227

(52) raam-ko

kahaanii yaad aaii, aur Ram-D story-N memory-N come-PERF and niinaa-ko kavitaa ———. Nina-D poem-N Ram rememberedi the story, and Nina ——— i the poem.

The facts in (50)-(51) would follow from the assumption that a part of a predicate cannot be gapped. The predicate of the second conjunct in (50a) is pasand aaii 'approved', in (50b) it is svaagat kiyaa 'welcomed', and in (51) it is d¥yaan diyaa 'paid attention'. The gapping of the subparts of these predicates in (50)-(51) is ungrammatical. Gapping may be thought of as the absence of a categorial node corresponding to the predicate of a clause, an argument, or the head of an argument. If the subpart of a predicate cannot be gapped, it follows from the representations in (39b) and (42b) that neither the N nor the V of the CP, which are subparts of the predicate at ARG STR, can be gapped.20 8.3.2.2. Ellipsis in Answers toYes-No Questions In answer to a yes-no question in Hindi, the predicate of a clause can stand for the entire clause. Consider (53)-(54) for illustration. Typical answers to the yes-no interrogative in (53a) and (54a) are given in (53b) and (54b), with everything except the predicate omitted: (53) a.

raam-ne aaj bahutkaam kiyaa? Ram-E todaymuch work-N do-PERF Did Ram do a lot of work today?

b.

hãã, kiyaa . yes do-PERF Yes, (he) did.

a.

niinaa-ne ravii-ko kitaab dii? Nina-E Ravi-D book- N give-PERF Did Nina give Ravi the book?

b.

nahññ dii.

(54)

20 It seems that gapping may also be regulated by phonological or prosodic constraints (Booij (1985)). I have not found evidence so far for such constraints in Hindi.

228 / Arguments in Hindi

not give-PERF (She) didn't give (Ravi the book).

However, the N in a CP cannot be omitted in this manner, as shown by (55)-(56): (55) a.

raam-ne mohan-par b¥arosaa kiyaa? Ram-E Mohan-L reliance-N do-PERF Did Ram rely on Mohan?

b.* hãã, yes

kiyaa. do-PERF

(56) a.

niinaa-ne kahaanii-par d¥yaan diyaa? Nina-E story-L attention-N give-PERF Did Nina pay attention to the story?

b.* nahñ ñ diyaa. not give-PERF

The facts in (55)-(56) can be accounted for by assuming that if the response to a yes-no interrogative contains the ARG-PRED, it must contain the entire ARG-PRED. The N in the CP cannot alone be omitted, because, as represented in (39b) and (42b), it is a part of the ARG-PRED. In sum, the inability of subparts of a CP to be omitted separately, both in gapping and in ellipsis, indicates a CP constitutes a single ARG-PRED, and is accounted for by the representation in (42b), which expresses the joint predicatehood of the N and the V at ARG STR. 8.3.3.

Putting the Pieces Together

It is now time to piece together the semantic structure, argument structure, and categorial structure of CPs. The SEM STR, ARG and GF STR, and GC STR of CPs may be related as follows: (57)

GC STR

ARG / GF STR

S …………………CLAUSE

\

/\

\

V …………………PRED

SEM STR

Complex Predicates / 229

/

/

V ……………………………………SEM PRED

N …………………………………………SEM PRED

The nominal host is a lexical category; it cannot be modified by adjectives or quantifiers, it cannot be relativized or conjoined, nor can it be replaced by a wh -word in-situ. A light verb and its nominal host form a single verbal category at GC STR; they cannot be scrambled away from each other. They do not form a single categorial word. That is, they are phrasally concatenated; topicalization can separate the light verb from its nominal host. The light verb and the nominal host do, however, form a single functional word, expressed at ARG and GF STR. Given in (58b) is the representation in (42b), along with its GC STR and GF STR correspondences. Evidence for this GF STR of the CP will be provided shortly. (58) a.

raam-ne mohan-par b¥arosaa kiyaa. Ram-E Mohan-L reliance-N do-PERF Ram relied on Mohan.

b.

[

X

do

Y

(= (3a))

] SEM

STR

[ X ON Z reliance ] ……/……………………………………………………………………… ˘ ARG1 ARG3 ARG2 ¿ A-PRED ARG STR ……\………..\……………..\………..\……………………… SUBJ OBL OBJ GF-PRED GF STR …………………………………………………………………………… raam-ne mohan-par b¥arosaa kiyaa WORD STRING …………………………………………………………………………… N N N V GC STR V S

In (58b), the nominal host of the CP is an ARG in ARG STR, and the OBJ in GF STR. Evidence for the argumenthood and objecthood of the CP-internal nominal is provided in the next section.

230 / Arguments in Hindi

Now, multiple associations such as that of the ARG-PRED to the two semantic predicates reliance and do in (58b) has a straightforward meaning: the two semantic predicates together compose the ARG-PRED. The multiple association of the semantic predicate reliance with an ARG as well as an ARG-PRED, however, is unfamiliar in syntax, and the meaning of such multiple associations may not be transparent. In order to explain what I mean by this multiple association, let me appeal to the uses of this formalism in phonology. One of the uses of the formalism is to express the shared properties of two entities. For instance, the shared melodic substance of two segments, such as geminate consonants and long vowels, is represented as in (59a) and (59b) respectively (Clements and Keyser (1983)): (59) a.

C \/

C (= pp)

b.

p

V V \/ ( =uu ) u

Multiple linking has also been used to express dual membership of a single entity. An example is the ambisyllabicity of a single segment, as when an intervocalic consonant is simultaneously the coda of one syllable and the onset of the following syllable (Kahn (1976)), as in (60): (60)





/\/ s i

t i

(city )

The multiple association of reliance in SEM STR with an ARG and an ARGPRED in ARG STR is another instantiation of the use of this formalism in (60): it means that the semantic predicate is simultaneously an argument in its own right and a part of the syntactic predicate.21

8.4. The Argumenthood of the Nominal Host A descriptively as well as theoretically puzzling phenomenon exhibited by N+V CPs in Hindi has to do with verb agreement and passivization. As we will now see, a light verb within the CP may agree with its nominal host. If 21 Such multiple uses of a single notation is admittedly undesirable. However, in the absence of more transparent notation, I rely on existing notational conventions.

Complex Predicates / 231

what the verb agrees with is an ARG of the clause, the CP-internal nominal must be an ARG. The argument status of the CP-internal nominal is endorsed by the fact that it may undergo passivization. However, we have already established that a CP functions as a single predicate. We thus face a baffling situation: the predicate of a clause agrees with one of its parts . I will outline a solution to this apparent paradox, drawing on the duality already implicit in the representation of CPs we have developed so far. 8.4.1. Verb Agreement in Complex Predicates Recall from Chapter 5 that the verb in Hindi agrees with its highest NOM ARG. Now, the nominal host in a CP cannot carry any case marking.22 When the subject of an N+V CP construction is not NOM, and the nominal host is NOM, the light verb agrees with the host. Consider the agreement on the light verb kar ‘do’ in (61)-(65): (61) a.

raam mohan-kaa apmaan karegaa. Ram-N(M) Mohan-G insult-N (M ) do-FU-M Ram will insult Mohan.

b.

ilaa mohan-kaa apmaan Ila-N(F) Mohan-G insult-N (M ) Ila will insult Mohan.

karegii. do-FU-F

22 This follows from our analysis of the nominal host as a lexical category. Recall that other than NOM nominals, the only nominals uninflected for case are locative destinations. In our discussion so far, we have only encountered NOM hosts. However, exactly as locative destinations can form N+V compounds in the incorporation construction, we might expect that locative destinations can be hosts of light verbs as well. Bruce Pray has pointed out to me precisely such an instance: (i) yah kitaab raam-ke kaam aaegaa. this book- N Ram-G-NN work come-NF This book will be useful to Ram. The verb aa ‘come’ in (i) takes two ARGs, the one who comes, and the locative destination. The nominal host kaam ‘work’ fills the position of the locative, and is therefore uninflected. It is NONNOM, as evidenced by the genitive agreement. (i) exhibits the syntactic properties we found to be characteristic of complex predicate constructions. Our analysis predicts that the light verb cannot agree with this LOC nominal. For this to be overtly visible, we need a feminine nominal host, which I have not been able to find.

232 / Arguments in Hindi

The verb in (61a, b) agrees with the NOM subject. In (62)-(65), the subject is not NOM. The light verb in these sentences agrees with its nominal host. (62) a.

b.

ilaa-ne mohan-kaa apmaan Ila-E Mohan-G Ila insulted Mohan.

kiyaa. insult-N (M )

do-PERF-M

ilaa-ne mohan-kii praßamsaa k i i . Ila-E Mohan-G praise-N (F) Ila praised Mohan.

do-PERF-F

ilaa-ne mohan-par b¥arosaa kiyaa. Ila-E Mohan-L reliance-N (M) Ila relied on Mohan.

do-PERF-M

ilaa-ne mohan-par kripaa kii. Ila-E Mohan-L favour-N (F) Ila showed kindness to Mohan.

do-PERF-F

(63) a.

b.

(64) a.

ilaa-ne mohan-se bair kiyaa. Ila-E Mohan-I enmity-N (M) do-PERF-M Ila showed enmity towards Mohan.

b.

ilaa-ne mohan-se Ila-E Mohan-I Ila hated Mohan.

a.

unhõ-ne gehõõ-ke vitara”-mã vilamb kiyaa. they-E wheat-G distribution-L delay-N (M) do-PERF-M They delayed the distribution of wheat.

b.

unhõ-ne gehõõ-ke utpaadan-mã pragati they-E wheat-G production-L progress-N (F) They showed progress in the production of wheat.

nafrat kii. hatred-N (F) do-PERF-F

(65)

kii. do-PERF-F

In each of the sentences in (62)-(65), the light verb shows masculine agreement in (a) and feminine agreement in (b).23 23 The phenomenon of agreement in (62)-(65) must be distinguished from the spreading of agreement features within verbal complexes. A predicate in Hindi

Complex Predicates / 233

In short, the situation presents a structural paradox. For the light verb to agree with the nominal host, the nominal must be an ARG. However, as we know, the nominal is part of the predicate. A CP construction must then have two different structures. Given in (66) are the two different groupings of the elements in (63a). (66a) is the grouping required by phenomena such as gapping, responses to yes-no questions, etc.; (66b) is the grouping required by the facts of agreement. (66)

_________ ___________ a.

b.

ilaa-ne _________

mohan-par ___________

_________ ilaa-ne _________

___________ mohan-par ___________

__________________ b¥arosaa kiyaa __________________ __________ ________ b¥arosaa kiyaa __________ ________

The intuitive substance of (66a, b) is in fact expressed in the formal representation in (58b), where the nominal host of a light verb is simultaneously an argument and a predicate. 8.4.2.

Passivization

The ARG status of the nominal host is endorsed by the fact that it may undergo passivization. Given in (67b) is an example of an ordinary passive, like the ones we saw in earlier chapters: may involve more than one verbal element, e.g. auxiliaries, modals, and the like. In such an event, all the non-final verbs reflect the agreement features of the final verb, which, as we know, agrees with its highest NOM ARG. The spreading of the agreement features within the predicate is illustrated in (i) and (ii) below: (i) raam gaataa jaa rahaa t¥aa. Ram-N(M) sing-NF(M ) g o stay-NF(M) be-PA(M) Ram was going on singing. (ii) siitaagaatii jaa rahii t¥ii. Sita-N(F) sing-NF(F) g o stay-NF(F) be-PA(F) Sita was going on singing. The auxiliary jaa ‘go’ is in its root form, and cannot reflect agreement. All the other parts of the predicate bear the same agreement features. One might ask why the agreement in (62)-(65) cannot be treated as part of this phenomenon. The reason is that exactly as in the agreement between a verb and an ARG, the light verb agrees with the subject if the subject is in NOM case. Agreement shifts to the nominal host only when the subject is in NONNOM case. Therefore the agreement in N+V CPs can only be treated as part of the phenomenon of agreement between a verb and one of its ARGs.

234 / Arguments in Hindi

(67) a.

niinaa-ne kitaab k¥ariidii. Nina-E book- N buy-PERF Nina bought a book.

b.

niinaa-se kitaab k¥ariidii gaii. Nina-I book- N buy-PERF go-PERF The book was bought by Nina.

The object kitaab ‘book’ shows an OBJ-SUBJ alternation in the activepassive pair. Now consider the CP-internal nominal apmaan 'insult' in (62a), repeated as (68a), and its passive counterpart in (68b): (68) a.

b.

ilaa-ne mohan-kaa apmaan Ila-E Mohan-G Ila insulted Mohan.

kiyaa. insult-N

do-PERF

ilaa-se mohan-kaa apmaan kiyaa gayaa. Ila-I Mohan-G insult-N do-PERF (Lit.) Insult was done to Mohan by Ila.

go-PERF

Neither Ilaa-se nor mohan-kaa is the subject of (68b), as can be shown from the facts of subject obviation and control.24 We must therefore conclude that the nominal apmaan 'insult' is the subject. The fact that it is available for selection as the SUBJ in the passive provides additional support for the conclusion that the CP-internal nominal is an ARG, and furthermore, an OBJ.25 8.4.3. Absence of Agreement: an apparent anomaly

24 For example, if we add the phrase us-ke g¥ar-mã ‘in his/her house’ to (68a) and (68b), the pronoun us can refer to Mohan but not to Ila in (68a). Recall from chapter 4 that a pronoun cannot be coreferent with the SUBJ in its minimal finite clause. In (68b), the pronoun can refer either to Ila or to Mohan, indicating that neither of them is the SUBJ. 25 One might ask why (68b) cannot be treated as an instance of the subjectless construction in §7.5. This could not be so, because the subjectless construction §7.5. does not allow the expression of the implicit agent, unlike (68b). Furthermore, the L-SUBJ of the subjectless construction is the "patient" argument. But in (68b), the L-SUBJ is the "agent". If we add the phrase apne g¥armã ‘in self's house’ to (68a) and (68b), Ila (but not Mohan) is the antecedent of the reflexive apnaa .

Complex Predicates / 235

8.4.3.1. As noted by Bailey (1956) and others, there exists a small set of N+V CPs in which the light verb cannot agree with its host. Instances of the failure of agreement are illustrated in (69): (69) a.

ilaa-ne mohan-ko pasand kiyaa. Ila-E Mohan-A liking-N (F ) do-PERF-M Ila liked (approved of) Mohan.

b.

ilaa-ne mohan-ko yaad Ila-E Mohan-A Ila remembered Mohan.

kiyaa. memory-N(F)

ilaa-ne mohan-ko k‹amaa Ila-E Mohan-A Ila pardoned Mohan.

kiyaa. forgiveness-N (F) do-PERF-M

c.

do-PERF-M

The CP-internal nominals in (69) are the only nominative nominals in the sentences. We would therefore expect the light verb to agree with them. These nominals are feminine, as is evident from the facts of modifier agreement in (70) below. Yet, the inflection on the light verbs in (69) is that of the masculine singular, indicating absence of agreement. (70) a.

ilaa-kii pasand vicitr t¥ii. Ila-G-F liking-N (F ) strange be-PA-F Ila's choice (taste) was strange.

b.

bacce-kii yaad mãã-ko sataatii t¥ii. child-G-F memory-N(F) mother-A torment-NF be-PA-F Memory of the child used to torment the mother.

c.

raam aap-kii k‹amaa caahtaa Ram-N you-G-F forgiveness-N (F) desire-NF be-PR Ram wishes for your forgiveness.

hai.

As we have already seen, when the light verb agrees with its nominal host, the nominal is an OBJ ARG. In what follows, I argue that when the light verb fails to agree with the nominal host, it is not an ARG of the clause. The argument is based on an intriguing correlation between the ability of the light verb to agree with the host, and the case associated with the ARG provided by the host.

236 / Arguments in Hindi

8.4.3.2. When the light verb can agree with its nominal host, the ARG provided by the host bears indirect case (GEN in (62), INST in (64), LOC in (63) and (65)). But precisely when the light verb cannot agree with the host, the ARG provided by the host bears direct case associated with PR.OBJ, namely, ACC when animate ((69)), and NOM when inanimate ((71) below). When it is NOM, the light verb agrees with that ARG:

Complex Predicates / 237

(71) a.

ilaa-ne kissaa yaad kiyaa. Ila-E incident-N(M) memory-N do-PERF-M Ila remembered/recalled the incident.

b.

ilaa-ne kahaanii yaad kii. Ila-E story-N (F) memory-N do-PERF-F Ila remembered/recalled/memorized the story.

Therefore, we must conclude that the ARG provided by the nominal host in (69) and (71) is a PR.OBJ. This conclusion is supported by the fact that this ARG is the subject of the passive counterpart. Given in (72) is the passive of (69a):26 (72) ilaa-semohan

pasand kiyaa gayaa. Ila-I Mohan-N(M) liking- N do-PERF-F go-PERF-F Mohan was liked (approved of) by Ila.

The correlation between predicate internal agreement and the case of the ARG provided by the nominal host is brought out most strikingly by the following contrast, taken from Bahl (1974: xxix): (73) a.

us-ne mohan-ko yaad kiyaa. pron-E Mohan-A memory-N(F) do-PERF-M He/she remembered Mohan.

b.

us-ne mohan-kii yaad kii. pron-E Mohan-G memory-N(F) do-PERF-F He/she remembered Mohan.

In (73a), Mohan bears ACC case, associated with the OBJ; kiyaa ‘did’ fails to agree with the predicate internal yaad ‘memory’. In (73b), on the other

26 Consider a context in which Ila approves of Mohan as a partner for either Ila's or Mohan's friend. If we attach to the active sentence in (69a) the phrase uskii frend-ke liye ‘for pronoun's friend’, the pronoun can be coreferent with Mohan but not with Ila. If we attach the phrase to the passive in (72), the pronoun can be coreferent with Ila but not with Mohan. On the other hand, if we attach the phrase apnii frend-ke liye ‘for self's friend’, the reflexive can refer only to Ila in (69a), but to both Ila and Mohan in (72). These facts of subject obviation and reflexive binding show that Mohan is the subject in (72).

238 / Arguments in Hindi

hand, Mohan is in GEN case; the light verb agrees with the predicate internal nominal. 8.3.4.3 The light verb kar ‘do’ has two semantic participants, as in the representation in (58b). One of these participants, the "doer", is agentive, and ends up as the L-SUBJ, and by default, the SUBJ. The other participant, the "done thing", is the L-OBJ, and by default, the PR.OBJ. Given that the PR.OBJ in (69a-c), (71a,b), and (73a) is the ARG provided by the nominal host, we must conclude that the nominal host is not itself a semantic participant of the light verb. That is, it is not the "done thing". This structure may be represented as in (74): (74) The light verb does not agree with the nominal host: [

X

S-PRED

Y

] SEM STR

[ X Z S-PRED ] ………………………………………………………………………… ˘

ARG

ARG

¿

A-PRED

ARG STR

As for the CP constructions where the light verb may agree with the nominal host, we have shown that the nominal host is the PR.OBJ. By the reasoning given above, we must conclude that it is the "done thing", and therefore the L-OBJ. A generalized representation of the structure of the CPs in (62)-(65) may then be given as in (75): (75) The light verb agrees with the nominal host: [

X

S-PRED

Y

] SEM STR

[ X Z S-PRED ] ………………………………………………………………………… ˘

ARG

ARG

ARG

¿

A-PRED

ARG STR

In (75), the light verb sanctions ARG-hood to the nominal host by association with the participant Y . In (74), in contrast, the ARG sanctioned by the light verb through Y is filled by the participant Z provided by the nominal host. Hence the nominal host itself cannot be an ARG. Therefore, the light verb cannot agree with the nominal host in (74). This analysis of the contrast in agreement within N+V CPs relies on an idiosyncratic property of the nominal host. This nominal may or may not

Complex Predicates / 239

be associated with a participant of the light verb. Once we do make this assumption, however, the facts of predicate internal agreement, passivization, and the case of the ARG provided by the nominal host follow automatically.

8.5. Summary: Structural Mismatches Let us summarize the discussion in this chapter. We have accounted for two sets of apparently puzzling behavior of N+V CPs in Hindi in terms of structural mismatches. The first puzzle centers around the grouping of the N and V in the CP . For the purposes of scrambling, the CP behaves as a single unit. The N in the CP is not on par with the other arguments of the clause. On the other hand, for the purposes of agreement and passivization, the N in the CP may behave like an independent entity with the same status as the other arguments of the clause. This behavior is accounted for as the result of a mismatch in constituency at GC STR and ARG and GF STR. At GC STR, the nominal within the CP , unlike the other nominals in the clause, is not a direct daughter of S , and therefore, cannot scramble. The corresponding entity at ARG and GF STR is a direct daughter of the clause, and therefore behaves as an independent ARG. (76) (=3a) CLAUSE

ARG

raam-ne

N

ARG

ARG-PRED

ARG

mohan-par b¥arosaa

N

ARG STR

kiyaa.

V

N

GC STR

V

S

The second puzzle concerns the status of the nominal within the CP, which is a semantic predicate. On the one hand, it is part of the predicate of the clause; it can determine the number of arguments in the clause, their

240 / Arguments in Hindi

meanings, and their cases. On the other hand, it can be an argument of the clause; it can passivize, and the light verb can agree with it. This dual behavior is accounted for by assuming that it is multiply associated at ARG STR with an ARG as well as an ARG-PRED. As mentioned earlier, this means that a single entity at SEM STR (and in the word string) has a dual function at ARG and GF STR: (77)

S - PRED

SEM STR

……………………………………………………… ARG

¿

A-PRED

ARG STR

The behavior of the N in the CP with respect to phenomena such as gapping and ellipses follows from this analysis. These phenomena essentially involve ARGs that do not have corresponding category nodes at GC STR. Predicate internal entities at ARG STR do not have this freedom to be unassociated. The N in the CP cannot undergo gapping or ellipsis because it is predicate internal. Recall that in Chapter 3, we argued for the separation of the levels of SEM STR and ARG STR in order to allow elements at one level to be multiply associated or unassociated with elements at the other level, as in §3:(16a-d). The one logically possible structure that we failed to provide illustration for was one in which an element in SEM STR is multiply associated with elements in ARG STR. This is precisely the structure in (77), employed to account for the simultaneous argumenthood and predicatehood of the nominal host. This analysis of CPs thus provides substantial evidence for the conception in which the grammatically relevant semantics of a predicate are factored apart and represented separately from argumenthood. Although the CP in Hindi is not a lexical category, as shown by topicalization, the phrasally concatenated N within it is a lexical category, as shown by the impossibility of adjectival modification and conjoining. The need for such structures, motivated in this chapter, calls for a re-examination of the assumption in many syntactic theories that an argument is necessarily a maximal projection.

8.6. The Notion "Lexical" Twice in the course of this thesis, Hindi verb agreement handed syntactic theory a puzzle to solve, once in the incorporation construction in Chapter

Complex Predicates / 241

5, and for a second time, in the complex predicate construction in this chapter. Both times, the solution to the puzzle lay in the argument status of the controller of verb agreement, which is independent of its categorial structure. An incorporated object and an agreement inducing nominal host in a CP are similar in that they are both ARGs on par with the other ARGs of the clause. There is, however, a crucial difference between them. Whereas the nominal host can be the subject in a passive (§8.4.2.), the incorporated object cannot (§5.2.1.). This difference follows directly from the difference between them in categorial structure. The N+V compound is a categorial word; it is input to derivational morphology and processes such as word stress assignment (§5.2.2.). The inability of the incorporated object to be the subject of the passive follows from the condition that a subject must be an open argument (Selkirk (1982)), which has the effect of prohibiting the nominal within a verbal compound from being interpreted as the subject. The complex predicate, on the other hand, is phrasally concatenated, and is not a single categorial word; the verb within the complex predicate can be topicalized (§8.2.2.). Not being lexically concatenated, the nominal host in a CP is not subject to the condition that it cannot be the subject. Hence, it can be the subject of a passive. That the parts of a complex predicate are phrasally concatenated, and yet the result is a single argument structure necessitates a careful examination of the notion "lexical" that underlies the lexical integrity hypothesis and the direct syntactic encoding principle. The lexical integrity hypothesis "prohibits syntactic reorderings into or out of lexical categories (such as V, N, etc.)." (Bresnan (1982b:54)) By this hypothesis, an N+V complex predicate (unlike an N+V compound) cannot be a lexical category; the relevant notion of lexicality here is that of the CATEGORIAL WORD.27 The principle of direct syntactic encoding holds that only lexical rules can alter argument-grammatical function associations (Bresnan (1982a:6-8)). In terms of structure building associations between ARGs and GFs, this means that ARG-GF associations are established internal to a "lexical" unit. By this principle, a complex predicate must be a lexical unit, since the ARGGF associations can be made only after the N and the V are concatenated. Clearly, the notion of lexicality relevant for this principle is that of FUNCTIONAL WORD, not the categorial word. Thus, the complex predicate construction in Hindi illustrates that the notions functional word and catego27 Assuming that "categorial word" is equivalent to "morphological word", i t might be useful to refer to this hypothesis as the morphological integrity hypothesis.

242 / Arguments in Hindi

rial word do not necessarily converge. In this construction, a single functional word corresponds to more than one categorial word.28 The reverse situation is exemplified by the morphological causatives in Japanese, formed by the suffixation of -sase to a verb. The form V-sase constitutes a single categorial word (Kitagawa (1986)); yet a sentence containing this word has a biclausal structure for the purposes of reflexive binding and other syntactic phenomena (Kuroda (1965), Kuno (1973)), including disjoint reference (Dalrymple (1982)). In other words, this causative construction must have two ARG STR clauses, and two GF STR clauses (Ishikawa (1985)). Thus, V-sase is a single categorial word but forms two functional words. That the notion functional word does not coincide with the notion categorial word should not be surprising in the light of the finding that the construct "morphological word" does not always coincide with the notion "phonological word" (Liberman and Prince (1977)).29 There has been a growing recognition since Marantz (1981) that "morphological word" does not always correspond to a single word-like unit for syntactic purposes either. Just as "phonological word" means a word-like unit for phonological purposes, thus, a functional word is a word-like unit relevant for the purposes of the principle of direct syntactic encoding,30 and a categorial word is a word-like unit relevant for the purposes of morphology and the lexical integrity hypothesis. It must be pointed out that the lack of convergence between functional word and categorial word does not make the principle of direct syntactic en28 This situation is found to a limited extent in the unproductive phrasal verbs (verb+particle) and idiom chunks in languages like English. Thus, the single functional word look up is made up of two categorial words, and can be discontinuous, as in look the word up. Similarly, the idiom keep tabs on is a single functional word, made up of several categorial words, containing a noun that can even be passivized. 29 Thus, John's in John's here is a single phonological word, but not a morphological word. In contrast, the lexical compound blackbird is a single morphological word but contains two phonological words. The term phonological word as used in Liberman and Prince has now come to be known as the metrical word (Inkelas (1989)). The term phonological word now refers to a domain of application for phonological rules, including those which build syllables, feet, and metrical words (Selkirk (1984), Sproat (1985), Inkelas (1989)). 30 Every functional word that is an ARG-PRED heads a single ARG STR clause.

Complex Predicates / 243

coding empirically vacuous. As stated earlier, it predicts that that there cannot be a merger of two ARG STR clauses into a single GF STR clause. That is, there cannot be constructions with two ARG STR clauses, and therefore two L-SUBJs, corresponding to a single GF STR clause, and a single SUBJ. This prediction contrasts sharply with the incorporation theory in Baker (1988), which claims that all GF changing phenomena are syntactic. The claim that all GF changing phenomena are instances of incorporation, and moreover, that all such incorporation is syntactic, leads to the conclusion that a GF changing affix retains its identity as a functional word. This predicts that if a GF changing affix has its own argument structure (for instance, a causative or applicative affix), the result must be a biclausal ARG STR. However, as shown by the facts of reflexive binding and disjoint reference in causatives in languages like Hindi (see Chapter 6) and Malayalam (K.P.Mohanan (1983)), this prediction is incorrect. Reflexive binding shows that a Hindi causative has only one L-SUBJ. It must therefore be monoclausal at ARG STR. The verb and the causative affix do not form two functional words. In short, the syntactic incorporation analysis fails to account for Hindi causatives, and allows for constructions with two logical subjects and a single grammatical subject.

9 Conclusion From the earliest of grammatical traditions, regardless of the language described, and of the terminology used to describe it, grammarians have recognized that a clause involves a predicate and its dependents. The meaning of the predicate regulates the number of its dependents, and their syntactic properties and relations. The regularities of the distribution and alternation in the syntax of these dependents have been central to much of the work in theories of syntax. In current terminology, this domain of research comes under the study of argument structure. This study has sought to add to our understanding of argument structure by closely examining a large body of facts from Hindi. The process of analysing these facts has led to the particular way of thinking about the organization of syntactic and semantic information around argument structure in natural languages, which I have tried to articulate in this thesis. The perspective that has emerged from the study is not inconsistent with many of the existing theories of argument structure. Yet, this conception of the organization of predicates has led to the confluence of a particular set of theoretical assumptions. Many of the individual assumptions are already available in existing theories, either implicit or explicitly stated. Some of the assumptions have been proposed in view of the specific analysis of Hindi.

9.1. Levels of Structure Central to our conception of argument structure has been the way information in a lexical entry is distributed over various levels of structure, and the way these levels interact. The four specific levels of structure employed in the analysis are SEM STR, ARG STR, GF STR, and GC STR.1 The enterprise of this thesis has been to examine some of the individual and joint contribu-

1 This is not to deny other levels of structure built around which represents topichood in a lexical entry.

242

ARG STR,

say, that

Conclusion / 243

tions of the different levels of structure to the total structure of a predicate and its dependents. 9.1.1. Semantic Structure The level of SEM STR as conceived of here has the following properties. First, it is the representation of all and only those meanings that can condition syntactic or morphological regularities. Second, it is built out of atomic semantic constructs drawn from a universal inventory. Third, the organization of elements at the level of SEM STR is not identical to the organization of event structures in the real world. To re-emphasize what was said in chapter 3, SEM STR is a strictly grammar-internal level of structure. It is not to be confused either with verb meanings in terms of entailments, or with lexical conceptual structure, both of which are non-linguistic representations of concepts, situations, and the like. A revealing example of the distinction comes from the organization of the SEM STR representations of the predicates in Hindi, k¥uß honaa "become happy", and k¥ußii honaa "become happy". We have adduced language internal evidence for their representation as [X MOVES TO HAPPINESS] and [HAPPINESS MOVES TO X] respectively (see §7:(4a,b)). Nevertheless, there is no reason to believe that they involve different sets of entailments. Nor is there reason to believe that speakers of Hindi conceptualize happiness in two different ways, unlike, say, speakers of English. To repeat our analogy in chapter 3, SEM STR on the one hand, and lexical conceptual structure and sets of entailments on the other, are parallel to phonological representations on the one hand, and phonetic representations and the speech signal on the other. 9.1.2. Argument Structure The information encoded in the argument structure in terms of theta role representations in many theories is factored apart within our conception into SEM STR and ARG STR. This factorization finds motivation in the various constructions that require multiply associated and unassociated elements at the two levels. An ARG at ARG STR may be multiply associated with participants at SEM STR in constructions involving modality meanings, causatives, and complex predicates. A predicate at ARG STR may be multiply associated with predicates at SEM STR in complex predicates and causatives. Likewise, a predicate at SEM STR may be multiply associated with an ARG as well as a predicate at ARG STR in N+V complex predicates. What we called

244 / Arguments in Hindi

the ACC logical subject construction in Hindi has a participant in SEM STR that is unassociated with any ARG. Such representations, without which a satisfactory account of causatives, complex predicates, and the ACC logical subject construction is difficult to imagine, amply justify the separation of SEM STR and ARG STR. The two grammatical constructs available at ARG STR, namely, LSUBJ and L-OBJ, play a significant role in Hindi syntax. The notion L-SUBJ is essential for the default association of an ARG to grammatical subjecthood (§3:(43a)). An L-SUBJ is also called upon as an eligible antecedent of the reflexive apnaa. Having separated logical subjecthood from grammatical subjecthood, we discover that the reflexive may choose as its antecedent either the L-SUBJ or the SUBJ. The ACC logical subject construction provides evidence for defining L-SUBJ as the highest logical ARG rather than the highest theta role. Exactly as the L-SUBJ at ARG STR is the default subject at GF STR, the L-OBJ at ARG STR is the default primary object at GF STR. The ACC case association in Hindi appeals to the notion L-OBJ. Thus, ACC case in Hindi requires the ARG it is associated with to be an L-OBJ. In some dialects, the ARG must, in addition, be the PR.OBJ as well. 9.1.3. Grammatical Function Structure Whereas ARG STR represents the number of syntactic arguments dependent on a predicate, GF STR represents the grammatical functions of these dependents. The elements at these two levels may differ in their relative prominence. Thus, in the passive construction, the L-SUBJ, which is the most prominent element at ARG STR, is suppressed, and thereby prevented from being the SUBJ, which is the most prominent element at GF STR. The elements at the two levels may also differ in their intrinsic properties. Thus, the L-SUBJ of the ACC logical subject construction in Hindi (§7.5.) is a PR.OBJ, although generally, the L-SUBJ of a monadic verb is a SUBJ. The difference in the GFs of the two L-SUBJs is not reducible to a difference at ARG STR. The atomic entities of GF STR are elements such as TERM, UNRESTRICTED, and OBJECTIVE. We have not motivated or defended these features in this thesis. The one entity of GF STR that has received close attention is the notion of grammatical subjecthood (UNRESTRICTED NONOBJECTIVE TERM). The intuition of the traditional notion subject has predominantly been based on a notion of subjecthood that conflates prominence of various kinds

Conclusion / 245

on a single entity. The fundamental insight that emerges from the patterns of Hindi analysed in chapters 6 and 7 is that it is necessary to recognize more than one grammatically relevant notion of prominence, and that grammatical subjecthood is just one of these prominences. Once we factor apart grammatical information into different levels of structure, it becomes clear that syntactic generalizations can be conditioned by information at any of these structures or their combination, and that languages may vary with respect to the prominences that can govern the same syntactic phenomenon. Once we recognize this, apparent confusions and conflicts in signals attributed to subjecthood are no longer conflicts. Rather, they are the result of a complex interaction between several parallel information structures. GF information forms a subsystem of GF STR. Another subsystem at GF STR is that of grammatical features such as case, number, gender, person, tense, aspect, and the like. A primary focus in this thesis has been on the principles whereby case is associated with ARGs. The principles of case association in Hindi appeal both to grammatical functions and to meanings. If the principle that associates a case to an ARG is conditioned by the GF of the ARG, the case is direct. Otherwise, it is indirect. Similarly, if the principle that associates a case to an ARG is conditioned by the SEM STR of the predicate, the case is semantic. Otherwise, it is non-semantic. This classification yields four types of case, all of which are instantiated in Hindi: direct non-semantic case (NOM, ACC, GEN assigned by non-finite verbs), direct semantic case (ERG), indirect non-semantic case (GEN assigned by nominals), and indirect semantic case (DAT, INST, GEN, LOC). Since direct case requires reference to specific grammatical functions, it follows that direct case will not be preserved under GF alternations. Indirect case, on the other hand, remains constant, regardless of the GF of the ARG. Thus, it is not only idiosyncratic or quirky case, stipulated in the lexical entries of individual verbs, that is preserved under GF alternations. All indirect case is preserved, even when entirely regular and predictable from argument meaning. An example of the preservation of indirect case is the preservation under passivization of DAT case in all dialects of Hindi, and that of ACC case in those dialects where ACC is associated with L-OBJ regardless of its GF. The possibility of associating case with ARGs not only on the basis of their GFs but also their SEM STR properties provides a simple mechanism for analysing the many-to-many relation between GFs and case. Grammatical subjects in Hindi appear not only with NOM case, but also with ERG, DAT, INST, GEN, and LOC case, and in some dialects, with ACC case as well. A

246 / Arguments in Hindi

theory that does not allow case assignment to be governed by SEM STR would miss the significant generalizations that govern the patterning of case in Hindi. 9.1.4. Grammatical Category Structure The level of GC STR represents information about grammatical categories and constituency. In this thesis, the organization of this level of structure has been of interest to us mainly for the purpose of demonstrating that the part-whole relations at GC STR can be quite distinct from those at ARG STR and GF STR, and that the notion "lexical" is not a unitary notion. 9.1.5. What is "Lexical"? The identification of the notion "lexical" as "word level" leads to the difficult question raised previously in chapters 5 and 8: what is a "word"? The notion categorial word does not always correspond to a word-like unit for syntactic purposes. A single categorial (GC STR) word may correspond to two independent functional (ARG STR/GF STR) words (§5.2.). Conversely, a sequence of two categorial words may form a single functional word (e.g. morphological causatives in Japanese). The lexical integrity hypothesis appeals to the notion categorial word. In contrast, the notion of lexicality relevant for the principle of direct syntactic encoding (Bresnan (1982a:6-8)) is that of the functional word. The factorization of the notions of CATEGORIAL WORD and FUNCTIONAL WORD promises to be a fruitful basis for the resolution of many of the current debates on lexical and non-lexical treatments of syntactic phenomena. 9.1.6. The Concept of Conflict Resolution The interaction between different principles of case association is crucially governed by statements of preemption that appeal to our classification of case association. When two principles of case association are in conflict, one of them preempts the other. A principle of indirect case association preempts that of direct case association. For instance, the association of DAT case preempts that of ERG and NOM case (§7.1.). The principle of semantic case association preempts that of non-semantic case association. For instance, the association of ERG case on subjects preempts that of NOM case (§4.3).

Conclusion / 247

The notion of preemption as a means to conflict resolution is central to the conception of grammar in this thesis. The specific statements of preemption in the analysis of case in Hindi follow from general principles. Thus, the effect of semantic case association preempting non-semantic case association follows from the Elsewhere Condition, which holds that a specific principle preempts a more general principle. The effect of indirect case association preempting direct case association follows from our general assumptions about the organization of grammar. The first assumption is that predictable information can be factored away and built on unpredictable information. A consequence of this assumption is that there exists a prior stage in the structure building where predictable information is not available, but is supplied at a later stage. The SEM STR of a predicate, being unpredictable, is available prior to its GF STR information, which is predictable. A second assumption is that principles of grammar take effect as and when the relevant information is available. A third equally general assumption is that associations once established cannot be changed. That is, derivations are structure building, not structure changing. The combination of these three asumptions yields the result that principles that appeal to information available at a prior stage preempt principles that appeal to information available at a later stage. Hence, indirect case association preempts direct case association. This conception of grammar, as we have already said, prohibits structure changing (i.e. transformational) operations. While structure can be built through principles of correspondence, the mapping cannot destroy or change structure. When two principles of mapping lead to conflicting results, the conflict is resolved through universally regulated preemption of one principle by the other. Furthermore, we disallow extrinsic rule ordering which is equivalent to stipulating preemption on a language specific basis.2 If mono2 If the theory disallows conflict resolution, the effects of preemption will have to be stipulated in each principle of the grammar. For instance, the principle of NOM case association on the subject (§4:(28)) would have to be revised as: Associate NOM case with a SUBJ ARG unless the ARG bears the meanings of CONSCIOUS CHOICE, [TO X], [X I-ABLE], [IN X], [AT X], and so on. Such a statement duplicates the environments for ERG, DAT, INST, and LOC case association. The principle of ERG case association on the subject (§4:(27)) would also have to be restated as: Associate ERG case with a SUBJ ARG associated with CONSCIOUS CHOICE if the verb bears PERF, unless the meanings of [X I-ABLE] or [TO X] are present o n the ARG. In a model that allows conflict resolution through preemption, there i s no need to make language particular stipulations in terms of "unless …" clauses. The architecture of the theory yields the preemption of NOM by ERG, and of ERG by DAT and INST.

248 / Arguments in Hindi

tonicity of grammars means absence of extrinsic rule ordering and structure changing devices, our conception is monotonic.

9.2. Co-presence of the Levels 9.2.1. Two levels of structure are co-present if principles of grammar can refer to both levels simultaneously. These principles include not only mapping principles that define the correspondence between the two levels, but also non-mapping principles that condition phenomena such as agreement, coordination, control and anaphora.3 If two levels L1 and L2 are related through structure building derivation, information expressed in L1 and L2 are co-present, unless stipulated otherwise. In contrast, if L1 and L2 are related through a structure changing derivation, then part of the information at one level is destroyed in the course of the derivation, and hence there is no stage where all the information from both levels co-exist. Hence in a structure changing mapping, a principle that accesses information from both L1 and L2 would involve a global mechanism. Global mechanisms are widely acknowledged to be undesirable. In a theory that disallows principles to simultaneously access more than one level, and restricts principles to strictly level-internal information, an alternative to the global mechanism would be to copy information from one level to the other. This is the strategy behind the idea of traces in GB. Thus, when an entity undergoes movement, its trace acts as a place holder, carrying information about its structural position prior to movement. Copying of information is also the strategy involved in encoding semantic information such as animacy, volitionality, and the like as syntactic features, in order to make them available to syntactic principles. The strategy has two undesirable consequences. First, it duplicates information. Second, it might represent information in a level where it does not naturally belong, for instance, semantics in syntax. Both these consequences may be avoided by allowing principles simultaneous access to the different levels of structure. The conception of co-present levels of structure, related through principles of association is central to this thesis. In claiming that SEM STR, ARG 3 Thus, in GB, move alpha is a device for mapping from d-structure to sstructure and from s-structure to LF. In LFG, annotations on phrase structure rules define the correspondence between c-structure and f-structure. In contrast, anaphora and control hold only level-internally.

Conclusion / 249

STR, GF STR, and GC STR constitute co-present levels of representation, the

claim is that the relation between them may not involve structure changing, and that a principle of the grammar of natural languages can simultaneously access information from more than one of these level. The need for co-present structures in Hindi is illustrated, for instance, by the principle of ERG case association. This principle requires simultaneous reference to SEM STR (CONSCIOUS CHOICE) as well as GF STR (SUBJ; PERF). Co-presence of the levels of structure in Hindi is also motivated by the phenomenon of verb agreement, which refers simultaneously to ARG STR and GF STR. The verb agrees with the highest ARG with NOM case. Likewise, reflexive binding requires reference to ARG STR and GF STR: the reflexive can be bound either to L-SUBJ or SUBJ. Both levels must therefore be accessed in order to determine whether or not the condition is satisfied. 9.2.2. The co-presence of the levels of structure envisioned here may at first blush appear to allow for unconstrained access across levels by principles of grammar. However, as we said in chapter 3, this problem is not a consequence of allowing the co-presence of various structures. The prohibition against inter-level principles implicitly claims that the separation of structural information into levels of representation is both sufficient and necessary to constrain the interaction of information available to principles of the grammar. However, we need to express a number of prohibitions that hold against interaction of information internal to a level. Hence, inaccessibility of information across levels of representation (except of course to mapping principles) is not sufficient to constrain possible interactions of information. Moreover, the same mechanism that expresses intra-level prohibitions can be used to constrain inter-level interaction as well. Therefore, inaccessibility is not necessary either. Furthermore, given the mechanism of copying, any information in one level may in principle be copied into another level in order to satisfy the demands of inaccessiblity across levels. The excessive freedom that is potentially a problem for a simultaneous access model is equally a problem for a theory that allows copying of information. Any stipulations that constrain the copying of information from one level to another in order to restrict principles to a single level can equally well be translated as stipulations that constrain the clusterings of information across levels as well, and moreover, do so without the duplication involved in copying. A more fundamental issue lies in a substantive claim underlying the conception of co-present, simultaneously accessible levels of structure.

250 / Arguments in Hindi

Theories that seek to constrain interaction of information in terms of levels of representation implicitly assume that the clustering of different pieces of information as required by different phenomena converge on clear lines of demarcation between levels. A central premise underlying the claim of simultaneous access is that there is no such convergence, and that clusterings of information may depend upon the phenomena in question. Thus, even though natural languages exhibit patterns of correlation holding between phonological properties like stress and intonation on the one hand, and syntactic constituency and entities like topic and focus on the other, they do not exhibit correlations between phonological segmental features and, say, syntactic constituency. In contrast, grammatical categories in a language may indeed exhibit systematic correlations with phonological segmental information.4 These generalizations cannot be captured by postulating appropriate non-interacting levels of representation. Similarly, whereas case selection in natural languages may be governed by specific semantic configurations such as those discussed in chapter 7, verb agreement cannot be sensitive to such information. Once again, levels of representation do not provide an adequate mechanism for the expression of such contrasts. Therefore, the separation of levels and prohibition against inter-level principles by themselves cannot constrain the space of possible grammatical statements. In order to accomplish this task, one has to identify substantive principles that govern the possible clusterings of information for the purposes of specific grammatical phenomena. All theories require such constraints level-internally. In a simultaneous access model, the constraints can be extended to hold across levels as well.

4 For example, Malayalam has the condition that short mid vowels cannot occur in non-initial syllables in a morpheme except in adverbs.

Index absolutive case 59, 68-69 accusative case and animacy 83 and definiteness 84 zero-marked 59 accusative logical subject construction formal representation 192 vs. instrumental subject construction 188 vs. passive 188 accusative logical subjects 187 and control 189 and reflexives 189 and subject obviation 189 Ackerman 39 action 197 action tier 172 affected 96 agreement 1, 123, 248 and phrase structure 106 complex predicates 230 modifier 6 verb 10 verb agreement 106 with incorporated object 118 Allen 107, 111 Alsagoff 140 Alsina 17, 39, 47, 222 alternations (see syntactic alternations) Amritavalli 72, 78 anaphora 54, 106, 248 Anderson 69, 168, 171 Andrews 43, 57, 95 animacy 59

argument non-thematic 36, 37, 38 relative prominence 22, 24, 36 asymmetries 28 representation 24, 31 thematic 38 valency 22 argument hierarchy 28 argument meanings 83 capability 59 semantic extensions 146 argument structure 1, 2, 22, 24-42, 242, 243 parallels in phonology 31 aspect 7 perfective 8, 71 progressive 8 Australian languages 68 autolexical syntax 120 autosegmental phonology 31, 53 auxiliary selection relevant for intentionality 25 telicity 25 volitionality 25 Bahl 61, 144, 202, 235 Bailey 202, 233 Baker 107, 120, 241 Basque 78 Belletti 21, 25, 58, 59, 140, 143, 198 Bengali 69, 146, 167 Bhatia 127, 144 bleached 216 Booij 112, 226 bracket erasure 52

Bresnan 1, 17, 21, 22, 24, 36, 37, 39, 44, 45, 46, 47, 50, 121, 127, 202, 223, 224, 239, 246 Burton-Page 202 Burzio 21, 195 Butt 76, 202 c-command 127 capability 197 meaning internally determined 156, 158, 159, 173 capability passive 155, 156 Cardona 40 case ablative 68 absolutive 68 accusative 81 and agreement interaction 1 modifier agreement 90 verb agreement 91 and definiteness 82 and specificity 82 association 57 direct 57, 58, 70, 83, 101, 198, 245, 247 direct non-semantic 58, 79, 80, 83 direct semantic 58, 79 indirect 57, 70, 101, 198, 245, 247 indirect non-semantic 58, 180, 184 indirect semantic 184 non-semantic 57, 58, 70, 198, 245 preemption 199 priorities 87 semantic 57, 70, 83, 198, 245 with meaning and UG 86 recurrent 86

comitative 68 default 102 ergative 70-79 feature 56-57 decomposition of 56 functional 58 genitive 79 grammatical subjects 1 Hindi 6 individuation 81, 85 inherent 58, 59, 102, 143, 198 instrumental 68 lexically stipulated 58 idiosyncratic 58 nominative versus non-nominative 62 patterns of absolute 86 natural 86 phrasal scope 61-62 preservation 59 primary object 59, 80-92 nominative-accusative 59 zero-marked accusative/nominative 88 principles of association 78, 80, 83, 84, 100, 147, 163, 177 quirky 57, 58, 95, 143, 196 selection 3, 250 relevant for contact 26 containment 25 direction 26 semantics of extensions 167 spatial relations 168 structural 59, 102, 198

subject 59 theta roles 59 case features 65, 177 core uses 66 syntactic and semantic distribution 67 universal inventory 65 case grammar 1 case marking 1, 56-57, 177 clitics 60, 167 ditransitive 86 postpositions 63 stem 89, 90 stems 62 non-nominative 63 case meanings 177 case preservation 92-98, 101, 102, 164 accusative 95 dative 94 case syncretism 88, 92 categorial word 202, 210, 228, 239, 246 Cattell 202 causative 33, 141 (also see syntactic alternations) cause predicate 32, 197 causee 125 causer 125 change 197 change of state 19 Chomsky 1, 17, 38, 46, 49, 58, 78, 143, 202, 221 Clark 171, 173 clause union 126, 141, 223 Clements 140, 229 co-presence of information 3, 14, 18, 51-54, 248-250 Cohen 127 Cole 202 complete functional complex 46 completion 197

complex predicates 9, 18, 75, 145, 183, 201 agreement in 230 analysis as argument transfer 221 merger of subcategorization frames 221 syntactic restructuring 221 theta-role promotion 221 and conjoining 211 and gapping 224 and modification 210 and passives 232 and relative clauses 213 and scrambling 206-208 and topicalization 208, 210 and wh-questions 212 and yes-no questions 226 categorial structure 216 clause structure 201 compound verbs 9 conjunct verbs 9 diagnostics for 203-205 host 202 argumenthood of 229 predicatehood of 216 structure of 227 vs. noun incorporation 239 compound coordinate 113 noun+verb 113 subordinate 113 verbal 113 Comrie 43, 56, 81, 85 conflict 87, 199 conflict resolution 29, 31, 101, 246-247 preemption 246 vs. unification 162

conscious choice 161, 190, 196, 199, 249 consent 193 constituency 14 categorial 14 functional 14 contact 168, 170, 197 containment 168, 170, 197 control 54, 248 convergence 250 coordination 225, 248 Croft 69, 86, 168, 171 cyclic transformation 47 Dalrymple 240 dative subjects 93, 98, 144-154 and control 152 and gapping 152 and reflexives 151 and subject obviation 152 grammatical subjecthood of 148 Davison 144, 155 deep grammatical functions 24 definiteness 59 direct internal argument 25 direct linking theory 3 direct syntactic encoding 202, 224, 239, 240, 246 discontinuous constituents 50 disjoint reference 240 Dixon 56, 68 Dowty 21, 26, 27 Dravidian 93 elsewhere condition 79, 80, 101, 247 Emeneau 144 emptying 75 endpoint 40 ergative case 123 and conscious choice 74-77

and perfective aspect 71-72 and transitivity 72-73 predictability of 74 experiencer subject 93, 97, 144 vs. dative subject 145 experiencer verbs 35 (also see psych verbs) extended projection principle 46 external argument 24, 25, 123, 140 vs. grammatical subject 25 f-command 127 factorization (see levels of structure) Farmer 51 Fauconnier 21 Fillmore 1, 23, 41, 77 final 1 law 46 final stratum 45 first sister principle 113 Foley 21, 24, 26, 39 Fon 221 Frantz 107 function-argument biuniqueness 46, 47, 54, 221, 223 functional control 44 functional word 202, 221, 228, 239, 246 Gair 64, 74, 106 gapping 98 and case identity of 134 and grammatical function identity of 135 Gazdar 121 gender 6 generalized phrase structure grammar 121 generative semantics 1 genitive subjects 79, 180 and control 186 and reflexives 186

and subject obviation 186 grammatical subjecthood of 185 Gilchrist 202 Givón 24, 85, 168 global mechanism 248 goal 146, 168, 170, 172, 173, 197 Goddard 56, 65, 68, 88 government binding theory 1, 3, 17, 58, 106, 248 Gradiner 107 grammatical category structure 2, 246 constituency 49, 119 grammatical features 17, 43, 120 grammatical function 17 adjunct 44 argument-adjunct 46, 48 chômeur 44 non-thematic 45 nuclear 44 object 244 direct 44 indirect 44 primary 44, 45 secondary 44, 45 unrestricted 45 oblique 44 representation 43 restricted 101 subject 44 term 44, 197, 244 unrestricted 45, 100, 106, 244 grammatical function structure 2, 42-48, 244 grammatical subject 38 and case 64 diagnostics for 122-139 PP as 64 prototypical 123, 139

grammatical subjecthood and control 130-134 and gapping 134-138 and prominence 139 and reflexive binding 124-128 and subject obviation 128-130 properties behavioral 122, 124 coding 122-124 semantic 122-124 grammaticalization 184 "grammatically-relevant subsystem" 19 Greaves 86 Grimshaw 21, 22, 27, 202, 205, 210, 221 Gruber 23, 24, 26, 168, 170, 172 Gujarati 139 Gurtu 127 Guru 5, 61, 202 Hacker 202 Hale 1, 11, 17, 19, 27, 49, 51 Halle 15 Halliday 51 Harbert 131 head-driven phrase structure grammar 51, 121 Heinrichs 40 Higgins 202 Hook 62, 144, 202 Hopper 56, 85 Huang 202, 221 Icelandic 57, 95 imperatives 160 implicit agent 37 inchoatives 98, 158, 194 indirect case (see under case association) grammatical subject 142 indirect refernce hypothesis 197

inherent case (see under case) semantic regularity of 143 initial stratum 44 Inkelas 52, 240 instrumental subject construction 173 vs. passive 155-158 instrumental subjects 154 and control 165 and gapping 166 and reflexives 164 and subject obviation 165 grammatical subjecthood of 164, 167 invariance and variability 167 inversion 144, 148 Ishikawa 240 Jackendoff 19, 21, 23, 24, 26, 27, 32, 39, 168, 170, 172, 182, 202, 221 Jakobson 56, 88 Japanese 204, 205, 240, 246 Jayaseelan 202, 221 Jespersen 202 Joshi 95, 140 Kachru 5, 62, 70, 72, 73, 75, 86, 88, 122, 131, 139, 144, 148, 167, 184, 202, 208 Kahn 229 Kanerva 1, 17, 22, 24, 37, 39, 45, 47 Kaplan 52, 121 Kashmiri 167 Katz 23 Keenan 43, 122, 139 Kellogg 61, 68, 88, 202 Keyser 19, 37, 229 Kiparsky 1, 3, 17, 22, 24, 37, 38, 39, 44, 47, 58, 79, 113, 198 Kitagawa 240 Klaiman 148 Knecht 110 Koopman 141 Krifka 40

Kuno 240 Kuroda 240 Ladusaw 27 Lakoff 1, 21, 168 language as an individual’s system 5 Laughren 27 Lefebvre 221 levels of structure 1, 3, 17, 18, 43, 242-250 argument structure 18, 30 c(onstituent)-structure 49 discourse 2, 18 f(unctional)-structure 43, 49 factorization 28, 31, 53, 54, 121, 163, 199, 243 grammatical category structure 18, 49-51 grammatical function structure 18 lexical conceptual structure 22 lexical structure 17, 49 melodic structure 31 np-structure 17, 49 predicate argument structure 24 prosodic structure 52 semantic structure 17 relation between 20 skeletal structure 31 surface structure 49 Levin, B., 1, 21, 25, 78 Levin, L., 1, 17, 47, 57, 95 lexical conceptual structure 19, 221 and semantic structure distinction between 20 lexical entry 1 predicate 1 lexical integrity 121, 202, 210, 239, 240, 246 lexical mapping theory 3, 17, 45, 47, 123 lexical representation 37 completely specified 2, 16, 87, 190

predicate 18, 25, 44 underspecified 2, 16, 87 lexical semantics 196 lexical strata 113 lexical-functional grammar 3, 17, 49, 52, 248 lexicality 202, 238-241, 246 Liberman 240 light verb 9, 75, 202, 203, 216 linear precedence 49 linking 1, 17, 39, 47 case 18 grammatical 58, 198 mismatch 54 one-to-many 32 principles 47 absolute 47 canonical 47 default 47 semantic 58, 198 unassociated entities 36 location dynamic 168 static 168 locative destination 89 locative subjects 167 and control 180 and pronouns 179 and reflexives 179 grammatical subjecthood of 179-180 logical object 39-41, 98, 99, 101, 174, 244 acted upon 40 affected 39 patient 39 semantic correlates 39-41 undergoer 39 logical subject 24, 25, 38-39, 40, 123, 140, 195, 244

deep-structure subject 38 demoted 48 in GB 141 suppressed 48 vs. highest expressed argument 195 logophoricity 140 Lyons 170 Mahajan 76, 202, 221 Malayalam 32, 40, 54, 69, 84, 89, 113, 146, 167, 215, 241 Maling 58, 95 mapping 3, 247 argument structure and case 48 argument structure and gf structure 3, 17, 46-48, 192, 223 default 48 semantic structure and argument structure 29, 31-38, 41, 47, 126, 141, 192 Marantz 1, 17, 25, 34, 95, 240 Marathi 72, 95, 146, 155, 167 Masica 75, 81, 87, 88, 202, 218 Matsumoto 140, 202, 205 McCawley 1 McCloskey 195 McGregor 5, 61, 62, 63, 71, 72, 81, 167, 182, 202 Mchombo 17, 39, 47, 222 meanings auxiliary selecting 26 case selecting 26 function selecting 26 merger 223 Merlan 107 Mester 202, 205, 221 metrical word 240 Mistry 139 Mithun 107, 111 modal meaning 158, 173 Mohanan 32, 51, 54, 113, 123, 126, 146, 148, 157, 241 Mohawk 107

monoclausality of modal+V 150 monotonicity 247 mood 7 morphological causatives 240 morphological integrity hypothesis 239 morphological word 120, 239, 240 Moshi 17, 47 move alpha 248 multiple associations 32, 37, 47, 228 multistructural organization 18 Nahuatl 107 Nash 11, 51 natural forces 76 Neidle 43, 56, 88, 121 Nepali 167, 187 noun incorporation 107-119, 238 and case 109 and compounding 113-115 and conjoining 109 and gapping 109 and modification 108 and passives 109 and phrase structure 112 and scrambling 113 and separability 108 and topicalization 113 generic interpretation 107 incorporated noun lexical category 112 lexicality of 120 np-movement 3 np-trace 49 null expletive 195 null pronominal 44, 130, 190 number system

Hindi 6 oblique 197 Oehrle 202 open argument 239 Ostler 24 Pa”ini 40 Pandharipande 70, 72, 146, 155, 156, 158, 167, 182 Panjabi 167 Paolillo 74, 202 participial adjunct 130 passive 48 formal representation 191 morphology 187 Perlmutter 43, 44, 46, 126, 140, 141, 148 phonological constituency 52 phonological word 240 Pinker 19, 26, 30, 41 Platts 71, 202 pleonastic 36, 37 point of origin abstract mental state 171 spatial 171 temporal 171 point of termination 174 abstract mental state 171 spatial 171 temporal 171 Polish 66, 68, 88 Pollard 43, 51, 121 possession 146, 181, 182, 197 inalienable 181 Postal 43, 44, 46, 126, 141 Pray 61, 62, 63 predicate argument structure 25

preemption 246 Prince 240 principles absolute 79 canonical 79, 83 pro 49 pro-drop Hindi 7 projection principle 1 prominence 244 prominences and subjecthood 139 factorization 245 pronominal system Hindi 6 pronouns 106 proximity 168, 170, 197 psych verbs 35, 140 quantifier scope 52 raising construction 131 Ramchand 197, 202 Rappaport 1, 21, 25 recipient 146 reflexive binding 38, 240, 249 and finiteness 128 reflexives 106, 139, 222 relational grammar 3, 44 relative prominence 175, 218 Riemsdijk 17 Rizzi 21, 25, 58, 59, 140, 143, 198 Roberts 192 Roeper 37, 113 Romance 93 Rosen 155, 156 Ross 1, 202 Rothstein 46

rule ordering 247 Russian 56, 68, 88 Sadock 107, 120 Sag 43, 51, 121 Saksena 143 Sanskrit 40 scales of prominence 31, 139 animacy scale 139 argument hierarchy 28-31 grammatical function hierarchy accessibility hierarchy 43 grammatical hierarchy 43 obliqueness hierarchy 43 relational hierarchy 43 mismatch 43, 47 sonority hierarchy 31 thematic hierarchy 23-24 Schwartz 143 scrambling 12, 13, 206 English 51 Japanese 51 Malayalam 51 typology 51 Warlpiri 51 Seigel 113 Selkirk 112, 113, 239, 240 Sells 43, 140, 202, 204, 205 semantic field 168, 170, 172-175, 182, 197, 218 semantic primitives 168 semantic structure 2, 18-22, 243 and syntax relation between direct reference hypothesis 21 indirect reference hypothesis 21, 22 configurations 18 parallels in phonology 243

vs. conceptual structure 243 vs. entailments 243 Shapiro 144 Sharma 5, 61, 202 Simpson 11, 43, 51, 57, 95, 112, 121 Sinha 217 Slavic 93 sonority hierarchy 31 source 168, 170, 172, 173, 193, 197 South Asian languages 11 Southern Tiwa 107 split ergativity 59 Sportiche 141 Sproat 240 Sridhar 122, 139, 148 Srivastav 131 Srivastava 5, 61, 70, 72, 81 Stenson 195 Stowell 59 stratal uniqueness law 46 stress 113, 114 structural mismatches 237 structure building 88, 247, 248 structure changing 52, 247 Subbarao 127 subject condition 46, 48, 195 as non-absolute 195 subject obviation and c-command 129 and finiteness 129 and logical subject 130 subject selection 139 subjectless construction 233 syntactic alternation 1, 2, 242 case dative-accusative 93

ergative-nominative 59 and instrumental 59 causative 10, 32, 38, 125, 136 causatives direct vs. indirect 10, 126 middle 37 passive 9, 93, 94, 97, 125, 136 transitive-intransitive 10 syntactic distribution 1, 2, 242 systemic grammar 51 Talmy 21, 27, 168 telicity 40 Tenny 40, 85 tense 7 past 8 thematic role 18, 22, 23 hierarchy 23, 24, 30 word order 24 representation 23, 30 thematic roles 17 theta role 2, 23, 26 merging 34, 35 representation inadequacies 32 theta role representations 194 theta-marking vs. non-theta-marking 210 Thompson 56, 86 Thráinsson 57, 58 topicalization 208 topichood 140, 242 trace 248 transitivity 72 Turkish 110 unaccusative transitives 98, 100 undergoer 96, 97 understood agent 188, 192

"unrestricted conceptual representation" 19 Uszkoreit 24 valency 2 increase 33, 34 Van Valin 21, 24, 26, 39 Varma 40 verb agreement 106 (also see agreement) verb meanings 1, 19 grammaticalizable 19 Vergnaud 15 Verma 110, 116, 122, 144, 148, 202 voice 7 active-passive 9 vowel shortening 114 Wali 64, 106, 155, 156 Wallace 187 Warlpiri 11, 13, 50, 54 Wechsler 28 Wierzbicka 19, 56, 68, 86, 88, 202 Williams 17, 25 willingness 193 word melody 113, 114 word order 11-14, 123 canonical 12, 157, 218 fixed 11 free 11 freezing 157 word order and definiteness 12 grammatical functions 11, 12 non-finite clauses 13 stress and intonation 12 word string 15 Zaenen 58, 95 Zubizarreta 21