<<

March 19, 2003

0. Introduction

Many languages form predicates productively by combining a N with a V, which combine to form a unit with properties of both the N and the V; usually the combination has the thematic structure of N and the case-marking properties of V. Languages with this construction include English, Japanese and Hindi/Urdu (1a-c)

1) a. We made them a promise [that we would be on time] [English]

b. keikoku suru 'do warning, warn'

john-wa murahito-ni [ookami-ga kuru-to] keikoku-o shita [Japanese] John-top villager-to wolf-nom come-comp warning do-pf

'John warned the villagers [that the wolf was coming] Grimshaw and Mester 1988:207.

c. yaad kar-naa 'do memory, remember' [Hindi/Urdu]

....woo .. apnee priya vastuooN-koo deekh-kar apnee bacpan-kii yaad kar lee-taa 3s self's dear objects -dat see-prt self's childhood memory do take-impf

'Seeing his cherished objects, he would remember his childhood. (Bahl 1974 72-3)

The combination of N and V as a single is widespread and productive in the languages which have this process. . One such language is Hindi/Urdu. In this paper I base an analysis on the N+P predicates of this language to inquire more closer into the general syntactic and semantic questions which they raise, especially in the context of syntactic analyses conforming to the constraints of the (Chomsky 1986, 1995). In particular I want to show the effect of the Principle of Full Interpretation on syntactic derivations well as semantic interpretation at the LF interface. This condition requires that every part of the (LF) representation must be licensed, by being case and theta marked if is an , or by assigning its theta roles if it is a predicate (Chomsky 1986: 97-103). One instantiation of Full Interpretation could be found in Higginbotham’s condition (1985) that arguments (receiving a theta roles) must be saturated, that is, they must not have to assign theta roles.

The N component of N+ V combinations has a paradoxical nature. The Ns of this construction are event nominals having a complex event structure and a theta array, more like V category heads than Ns. Yet the N is clearly the syntactic object of V. Further, the properties of N +V are strongly

1 influenced by N rather than V. These properties have been explained in much of the previous literature as the result of a lexical composition of a fully specified N with a specially designated ‘light’ lexical version of V, which lacks the normal argument structure of V. I will propose a syntactic rather than lexical derivation for N+V, which relies on a fully specified N merging as the syntactic object of V. But N is not fully saturated , as it has its own theta roles of assign. Successful merger of N as a syntactic object of V requires that the event structures of N and V have to merge. If unification of their event properties takes place, then the remaining arguments are all licensed and no theta roles are left undischarged, satisfying Full Interpretation at LF. The goal of this syntactic proposal is to show that the construction is an instance of a very general syntactic process of merger, theta discharge and case licensing, without special lexical stipulations, other than the specifications of lexical information encoded by N and V. I want to simplify greatly the role of the lexicon in the formation of N+V, to show the effects of Full Interpretation. One consequence is that syntactic licensing by case checking is not identical to semantic licensing via theta assignment. Another is that there are no special lexical ‘light’ , but some verbs are lighter than others, and these are the ones which participate in N+V combinations. Finally I want approach the problem of complex N + V predicates form the point of view of syntactic composition, defining the contribution of syntactic structure to interpretation. The verbal projection is constituted of sub-events and so N+V combinations have an accomplishment interpretation by virtue of N in syntactic object position, corresponding to a sub-event defining a result or end point.

1. The nature of N+V complex predicates These constructions have long been understood to present interesting problems of , and syntax, from a variety of descriptive and theoretical viewpoints. There is a vast array of papers on complex predicates in many languages. See Cattell 1984, Mohanan 1994, Butt 1995, Dubinsky 1997, Karimi-Doostan 1997 and other papers for references to much of this literature. The combination of V with N or another V in Hindi/Urdu has been discussed by Hook 1973, 1979, Verma 1993, Gambhir 1993, Mohanan 1994, Butt 1995 and Butt and Ramchand 2001. It is interesting that there has been from the first discussions, substantial on the properties of the construction, though there are many different approaches to its derivation. Jackendoff’s 1974 paper on English complex predicates of the general type (1) called attention to two features: the construction as a whole has the theta roles and subcategorization properties of N, not V, and the combination of N +V depends on some sort of compatibility in argument structures between N and V. Grimshaw and Mester’s 1988 study of Japanese N + suru, as in (2), conclude that the accusative-marked N forms a NP phrase with its thematic object (internal argument), while the and other arguments are outside the NP. These two points are well-grounded in empirical evidence and are important properties which must enter into accounts of the construction, including this one.

1.2 Interface conditions The conditions on the LF interface restrict the syntactic object so as to exclude combinations which

2 cannot have a conceptual-intensional interpretation. What these conditions are is less clear. Referentially dependent expressions such as anaphors must somehow be associated with syntactic antecedents conforming to the conditions of the language in question. The syntactic head- specifier relation may be invoked if anaphors are cliticized in some way to a functional head such as TENSE or AGR, identified the DP in the specific position of the projection (Cole and Sung 1995, Cole, Hermon and Huang 2001). But not much has been said about the consequences of LF interface conditions and what derivations would create violations leading to lack of interpretability at LF. . There is a condition on semantic interpretation originally proposed by Higginbotham 1985, and discussed and somewhat restructured by Speas (1990). Though proposed in the context of an earlier theory of syntax, I propose that it constrains LF representations in a way which is appropriate to the Minimalist Program. This condition constrains theta discharge and can be summed up as

2) Arguments are fully saturated (Higginbotham 1985:561) A constituent is saturated if it contains no undischarged theta roles in the array associated with its head. For an argument to be assigned a by a head bearing a theta-array, the argument must not itself have any theta roles which are not discharged.

Higginbotham's original proposal is grounded in the concept of government, which defines a local relation of syntax and construal. in D-structure and S-structure, levels of representation abandoned by the Minimalist Program because they are not motivated by interface conditions (phonetic and semantic). Some of the functions of government in defining a close syntactic relation are found in the operation MERGE. Heads merge with phrases which are arguments of the head and discharge one of the head's theta roles. Theta discharge takes place at the same time as MERGE (Chomsky 1995, Ura 2000). Here I propose that it is a condition on LF which constrains possible executions of the operation MERGE, and guarantees a well-formed LF representation. If MERGE takes place combining a theta-marking head with an unsaturated argument phrase, the result is an uninterpretable and ill-formed LF structure.

An example of normal MERGE of a head with a theta array and its arguments is shown in ( ). The theta related operations are theta discharge and theta projection (Higginbotham 1985: 560ff, Speas 1990). :

3) a. A friend remembered Sam b. remember <1,2,e> MERGES with Sam, discharging the 2 argument position

c. The V head PROJECTS upward, with its theta grid containing one discharged position

V<1,2*,e> PROJECT

3 3 V<1,2,e> DP remember Sam MERGE

d. The combination in ( c) MERGES with a friend:

V<1*,2*,e> PROJECT 3 DP V <1,2*,e> MERGE a friend 3 V<1,2,e> DP remember Sam

The theta grid projects upward, allowing the projected phrase to combine with other arguments until all theta roles are discharged. . Finally, the event argument e in the theta array is theta bound by TENSE (Higginbotham 1985). The syntactic object formed by these operations can be interpreted at LF. No theta positions are unsaturated. 1 A rough approximation of a the LF representation is given in (2), in a kind of English syntactic order for legibility. The principal feature of this representation, however, is that each position in the theta array is related to some component of the sentence, and this relation will be represented as coindexing, as Higginbotham originally proposed (1985); this coindexing is a means of cross- referencing phrases with the heads which license them, identifying their syntactic and semantic roles. Any equivalent notation or conditions will serve the same purpose. ______| | 4) TENSE(I) [ DP(j) ...[V<1(j), 2(k), e(I)> DP(k)] |______| |______|

The theta grid of V contains no open positions. DP(j) and (DP(K) discharge the 1 and 2 theta roles respectively. The event argument e is theta bound by the TENSE. This schema is meant as an example of a representation which is well-formed at the LF interface. It conforms to a condition which can be stated as Higginbotham's 1985 condition, quoted in (2) above.

A violation of (2) would contain open, unsaturated theta positions, or phrases not constructed with some theta position. Either would be violations of Full Interpretation (Chomsky 1986, 1995), doing duty for the earlier Theta Criterion, but not requiring an exact 1:1 correspondence of arguments and theta roles2

2. Complex predicates in Hindi/Urdu, and some syntactic assumptions.

4 N + V predicates are commonplace in many languages, often doubled in the lexicon by a denominal --make us of, use. in English or khooj kar-naa, khooj-naa 'make a search/search' in Hindi/Urdu. They are a very productive source of new predicates, in Hindi/Urdu allowing the borrowing of , and even verbs from other languages, including English, but especially from and the Urdu Perso-Arabic vocabulary. They are not morphologically transparent in Hindi/Urdu, and they resist direct affixation of verbal morphology. They are similar to the complex event nominals which Grimshaw 1990 of which Grimshaw (1990) provides a detailed lexical account.

2.1 Essential problematic properties of N+ V compounds

In this section, I note the morphological and syntactic evidence which supports the assumptions in (5):

5) a. The N is a syntactic argument of V, in fact a direct object. b. The N combines with the internal thematic object of N and assigns it case and a theta role within the N projection. c. The case of the external argument is determined by V. d. The syntactic structure of the N+V projection is:

VP 3 DP -SUBJ V' <1,2*, e> Theta Identification 3 NP <1,2*,e> V [] <1,2,e> 'do' 3 DP- Gen/Loc N <1,2, e>

These points are the basis for the course of the syntactic derivation I propose. The N merges first with its internal argument DP, and theta discharge is immediate. Case assignment is determined within NP. The nominal projection NP then merges with V, as a possible candidate as the internal argument of V. It is at this crucial point where identification takes place between the N and V argument structure, with categorial projection of V, borrowing the theta array of N. In the next sections, I will present the evidence for the assertions in (5). These generalizations are crucial to defining the constituent structure of complex predicate sentences, and the sequence of merger and theta discharge, as well as to explain a variation on (5d).

2.1.1. The N is syntactically the direct object of V. In Japanese , the N of the N + suru complex predicate is marked by the accusative -o:

6) taroo-wa ziroo-ni [yuusyooku-no teian -o sita [Japanese] Taro-top Ziro-dat dinner -gen propose-acc did

5 'Taro proposed dinner to Ziro.' (Dubinsky 1997: 630)

Hindi/Urdu has no direct counterpart of the Japanese accusative marker, as direct objects are either nominative or dative (preferred for direct objects with specific or animate reference). But the verbal agreement in Hindi/Urdu, which is a property only of nominative arguments (Khan 1988), provides the same kind of evidence for object status as the -o suffix in Japanese.

The agreement morphology in (7) makes this fact clear. The agreement on the verb-inflection complex reflects the feminine singular gender and number of the N component of the complex predicate, rather than the gender of the syntactic subject (3sm) or the masculine plural thematic argument yuddh-kee shaktishaal aayudh. 'powerful weapons of war'.

7) us-nee yuddh-kee shaktishaal aayudhooN -kii khooj kii hai 3s(m)-erg war - gen powerful weapons(mpl)-gen search(fs) do-pf-fs) is

'He researched powerful weapons of war (cf. Bahl 1974:222)

The general rule of agreement for verbal inflection in Hindi/Urdu is that the single set of agreement features are valued by a nominative argument, which may be the subject or object. Nominative case is indicated by the absence of a postposition and oblique morphology. A nominative subject (8a) requires agreement; here the subject is 1plm..

8) a. ham yah kitaab / in aadmiyooN -koo khooj-eeNgee we-nom(mpl) this book(fs)-nom/ these men (ms) -dat search-fut(1plm)

'We will search for this book/these men.' [Verb agrees with subject]

b.. ham -nee yah kitaab nahiiN khooj-ii We (mpl) -erg this book(fs) -nom not search-pf-fs

We did not search for this book.' [Ergative subject, V agreement with direct object]

c. ham-nee in aadmiyooN -koo nahiiN khooj-aa we-mpl-erg these men (mpl) -dat search-pf-ms

'We did not search for these men.' [Default agreement]

Postpositional cases like the ergative (5b) and dative (5c) block agreement. If the subject is not nominative, (5b) then a nominative object like yah kitaab triggers feminine singular agreement. Direct object This pattern of agreement is obligatory, and if there is no nominative argument, then the

6 verb-INFL complex has the default 3sm morphology (8c). In (7), the verbal inflection shows agreement, but not with either the subject or the thematic object. It agrees with the N khooj(f) 'search. We must conclude from this obligatory agreement that khooj is the syntactic direct object of kar-naa in (7).

2.1.2. The N determines the case of the internal argument The second striking fact, also common to complex predicates in other languages, is that there is a semantic direct object which does not get the kind of object structural case assigned within the V projection. Instead it gets the genitive case in Japanese (1b) and in Hindi/Urdu sentences such as (1c, 10).There is another case possibility, a locative case selected by the complex predicate. The N in (9b) selects one locative, while a different N in (9c) selects a different locative. This case choice is an instance of lexical case, in contrast to the genitive in (9a), which is the normal case linking a N with another nominal phrase.

As in English, the thematic object may be marked as genitive or have a locative case, such as Hindi/Urdu -par 'on'. The N+V takes a thematic complement which is marked with genitive (2a). Other thematic objects are marked by -par 'on' or -see 'from, with' (9b,c)

9) a. N. isteemaal ‘use’ + kar-naa ‘so,make’ =‘make use, use’ b. N hamlaa (m) ‘use’ +kar-naa' = 'make an attack on, attack' c. N. inkaar (m)' ‘refusal, denial’ +kar-naa '' = 'refuse, deny’

10) a. N = isteemaal 'use' coor -nee apnee dimaag-kaa sahii isteemal nahiiN ki-yaa thief-erg self's mind(m)-gen right use(m) not do-pf

'The thief did not make right use of his mind.' (Hook 1979:159)

The genitive case (-kaa, inflected like an ) is the normal way that a DP is linked to NP, which otherwise would lack structural case assigning properties.

11) a. amriikaa -kii khooj America-gen search(f)

The discovery of America (Bahri 1992: 138)

b. khoo-ee hu-ee baccooN-kii khooj lose-pf be-pf children -gen search (f)

7 'The search for missing children (Ibid)

c. [us-kii/*nee likh-ii hu-ii] kitaab 3s-gen/*erg write-pf be-pf book(f)

'The book [which he/she wrote]\

The genitive is used for objects of N (11a,b), and for subjects where the ergative case would be ill-formed (11c). I will assume that the genitive is the default structural case within the projection of N. If genitive is a structural case, the DP must undergo some movement to a specifier position to license this case.

Locative case marks the thematic object in (12) and (13). The choice of locative depends on the N:

12) . N= hamlaa 'attack'

churee -kii tarah teez daaNtooN-see tairaakooN-par/*see hamlaa kar dee-tii haiN knife-gen manner sharp teeth-with swimmers-on /*from attack (m) do give-impf are

They attack swimmers with sharp teeth like knives.' (Bahl 1978: 173)

13). N= inkaar 'denial, refusal' akbar kii janmjaat pratibhaa-see /*par kooii inkaar nahiiN kar sak-taa Akbar-gen inborn talent(f) from/ *on someone (m) denial (m.) not do can-impf-ms

'No one can deny Akbar's inborn talent.' (Bahl 1974: 264)

The genitive marker on the thematic object (10) seems to be the default choice of case. Locative cases are selected by specific Ns, as well as adjectives such as mazbuur ‘forced’.3 Other adjectives which combine with kar-naa 'do' include maaluum 'known' mahsuus 'perceived' pasand 'approved, liked', tay ‘deciding’ .

The verb kar-naa by itself does not assign these cases to its internal argument (14), though other verbs do. There is a genitive object in ( 15a), a -see ‘from’ locative object in (15b,c), and a -par ‘on object in (15d):

14) . us-nee bahut kaam *-kaa/*see/*par ki-yaa 3s-erg much work-ms-nom -gen/from/on do-pf-ms

' He/she did a lot of work.'

8

15) a. woo [PRO wahaaN jaa-nee]-kii sooc rahaa hai 3s there go-inf- gen-f think prog is

'He is thinking of [PRO going there].' b. un aadmiyooN-meeN nahiiN hai joo un dhamkiyooN-see Dar jaa-y those men-in not is who those threats -from be afraid go-cont

'There is no one among those men who is afraid of those threats (Nespital 1997:625) c. hariiš-nee daaroogaa -see sab sac -sac kah-aa Harish-erg police-inspector-from all truth-truth say-pf 'Harish told the police inspector the whole truth.' (Nespital 1997:212) d. . shiilaa apnii galtii-par ab pachtaa rahii hai Shila self's error-on now regret prog is

'Shila is now regretting her mistake.' (Nespital 1997: 769)

The N determines the case of the thematic direct object, as we have seen by keeping the V constant, but varying N in (12)-(13). This property of N is very much like the case-determining property of the main V in (15)a-d.

The normal direct object structural case for simplex verbs is the nominative (8b) or dative (16)

16) kuttee-koo patthar-see mat maar-oo, nahiiN too tum-koo kaaT-eegaa dog- dat rock-with not hit-imper not then you-dat cut-fut-3sm

'Don't hit the dog with a rock, otherwise he will bite you.' (Nespital 1997: 225)

Verbs like maar-naa 'beat, kill' and kaat-naa 'cut, bite' require the dative case on animate, specific direct objects; I’ll assume this case is checked by movement to a functional projection within VP, the light verb or some aspect-related head (Davison to appear).

This case is unavailable for most complex predicates, though there is a third case possibility for some N+V complex predicates, in which the N combines with V so that N+V functions just like a simplex verb, and the thematic direct object checks structural object 4case (17).

17) ... maaanoo us sparsh -koo ab bhii anubhav kar-tee hu-ee...

9 as though that touch-dat now also experience do-impf be-pf-obl

'...As though still feeling that touch even now...' (Bahl 1974:250)

This option is limited to a small number of N +V combinations which mark the thematic direct object in the same way as the syntactic direct object of a simplex verb (8).5 This is another instance in which the case on the thematic object is a property not of V alone but of the specific N+V combination.

2.1. 3. V determines the case of the subject The contribution of V can be seen when the N is kept constant, but V varies. Some complex predicates can be formed with more than one light verb, forming complex predicates which overlap in meaning. The nominal yaad 'memory' combines with kar-naa 'do' as well as rakh-naa 'place, aa- naa 'come', paR-naa 'fall' and hoo-naa 'be, become'.

18) a. yaad kar-naa 'remember, recall, learn by heart' b. yaad rakh-naa 'remember, keep in mind' 19) a. yaad aa-naa 'remember, recall, miss' b. yaad hoo-naa 'remember, have in mind'

These complex predicates all overlap in the meaning 'remember', with an experiencer theta role for the external argument, although they have somewhat different ranges of meaning (consistent with the idea that natural languages never have perfect synonyms (Carstairs-McCarthy 1999). They all can be used to express involuntary experiences, even when the subject is marked with the ergative case.

20) a. N. yaad 'memory' + kar-naa 'do' = ergative subject -nee

maiN-nee/*mujhee us-kii yaad kii 'I remembered/recalled him/her.' I-erg /*dat 3s-gen memory do-pf b. N yaad 'memory' + aa-naa 'come = dative subject -koo (or dative ) mujhee/*maiN-nee us-kii yaad aa-ii 'I remembered/missed him/her.' I-dat *I-erg 3s-gen memory come-pf

The ergative case is required for kar-naa 'do', but impossible with aa-naa 'come'; the dative case is impossible with the external argument of kar-naa 'do'. The theme external argument of aa-naa 'come' is nominative, but a goal/experiencer argument may be added in many complex predicates.

21) a. us-nee/*usee kaam ki-yaa 3s-erg/*3s-dat work do-pf

10 'He/she did the work.'

b. woo/ *us-nee ghar aa ga-yaa 3s-nom /*3s-erg house come go-pf

'He came home.'

c. usee /*woo bukhaar aa -taa hai 3s-dat *3s-nom fever(m) come-impf is

'He/she has a fever.' (Bahri 1992:41)

The light verb in (20a,b) has the same case requirements for the external argument as the same verb used alone (21a,b) or in abstract possession predicates (21c). The verb in the N + V combination retains its morphological properties of being able to assign case to arguments. The verb kar-naa ‘do’ differs from aa-naa ‘come’ in being able to assign ergative subject case and dative direct object case. Which I associate with the presence of a functional projection above VP (Davison, to appear).

2.1.3. The N determines the thematic role of the external argument The third striking fact complex predicates is that the thematic roles of the sentence arguments are determined by N, not V. In many cases it is not possible to tell, since the thematic roles of N are not dissimilar from those of V . In the case of khooj kar-naa 'make a search', both the N and the V have agentive external arguments and a theme internal argument. Other nominals assign the role of experiencer to the external argument, for example anubhav 'experience', mahsuus kar-naa 'experience' and afsoos 'regret'. Even though the subject is marked with the ergative -nee and kar- naa ‘do’ is normally agentive, the combinations in (22) and (23) are not agentive:6

22) a. anubhav kar-naa 'experience'

.kanyaa-nee mahsuus ki-yaa [ki woo jiivan-meeN pahlii baar kisii puruS-kee Kanyaa-rrg feeling do-pf that 3s life-in first time some man-gen aakarSaNpaash-meeN puurii taur-par bandh ga-ii hai. attraction-snare-in full state-in tied go-pf

'Kanya felt that for the first time in her life she was completely snared in the trap of attraction to some man.' (Bahl 1974: 253). [Experiencer, not agentive subject]

23) afsoos kar-naa 'regret, repent'

11 yee gumraah aurateeN peeshtar nahiiN [ too sharaab-kaa nashaa utar -nee- kee these wayward women earlier not then liquor-gen intoxication come down-inf-gen baad] apnii haalat-par afsoos kar-tii haiN. after self's condition-on regret do-impf are

'These wayward women, not at first, but after coming down from the intoxication of liquor, feel regret about their condition.' (Bahl 1974:98)

In each of these three sentence (9-c), the external argument has the role of experiencer rather than agent. This role has to come from the N component of the N + V compound, rather than from V.

2.2. In summary, I have given language-internal evidence for the syntactic and semantic assumptions outlined in (5) above. The V component of N+V takes N as its syntactic internal argument, with consequences for agreement just like other nominative arguments in the same syntactic context. V determines subject case as well as the case of the syntactic object (which may vary, as in (7) vs (17). The N has a greater influence than V on the thematic role of the subject, and it appears to theta mark its object directly, sometimes in with theta-related locative case.

3. Merger and theta discharge in complex predicates The demands of both N and V in a N+V combination form a complex construction, if both N and V have theta-grids and argument requirements. This combination is inherently contradictory, as N is both an argument of V (by the agreement evidence) and a head which requires arguments which receive theta roles. Ths ere are three distinct ways in the grammar of Hindi/Urdu which reconcile the conflicting requirements to produce a unified and interpretable sentence.

3.1 Three syntactic solutions These solutions are summed up in (24). DP represents the internal argument of N, and the thematic object of the N+V combination:

24) a. [DP-gen N] +V = examples (1c.7, 10 ) [Solution I] b. [DP-loc N] +V = examples (12, 13 ) [Solution II] c. DP-nom/dat [N+V] = (example (17) [Solution III]

A fourth possibility is that the N+V combination remains 'heavy' so that it is simply a transparent instance kar-naa or similar verb with a nominalization object, retaining the argument of kar-naa. In that case, the derivation proceeds as for a sentence with a direct object, the equivalent of the English sentence (3).

A construction with the three realizations summed up in (24)) is inherently contradictory. The crux

12 of the matter seems to be in the status of N. It is both a syntactic direct object, and part of the complex predicate itself, contributing selection and theta-marking properties. Since this construction is found in many languages and is very productive where it is found, some process intervenes to keep the N from simply receiving the theta role assigned by V to its inner argument, the normal case or ‘heavy’ as ‘opposed to ‘light’ interpretation of the V..

The condition requiring only fully saturated arguments discussed in (2 ) above seems to have a natural application in a Minimalist view of syntax as a condition on the LF interface. I will adopt this restriction as a central component of a syntactic account of complex predicates to be outlined below. . It will be the motivation for complex predicate formation within the course of the syntactic derivation of a sentence, with the effect of creating a well-formed LF representation for the sentence in which the predicate is a complex predicate.. In the well-formed outcome, the argument structure ends up with a single argument structure and theta array. This theta array is made up of theta and selection properties of N, while V determines the structural case of N:

25) a. N <1,2,e> default genitive or lexical locative selected for argument 2 b. V <1', 2', e'> selects case possibilities for external argument 1, case-licenses N c. N+V <1,2, e'> imports the theta array of N, preserving the case possibilities of V

The internal argument of N is syntactically somehow 'extra', getting a structural genitive or lexical locative case. The question is what sort of process derives (13) from (12), and at what point in the derivation of a sentence it applies

Lexically based solutions first of all assume that the lexicon specifies special light version of the verb ‘do, make’, like the Japanese suru. This light verb has some verbal properties, but its theta array is empty. It has argument structure but cannot specify theta roles or the specific number of arguments. The N transfers all of its arguments or some of them to V, in a process of lexical derivation, s as in Grimshaw and Mester 1988, or else N and V match in linked sub-components of the verbal structures, as in Jackendoff 1974, Jackendoff 1990, Rosen 1990, Butt 1995, with the N or other internal category supplying information missing from the light V. In either case, the light V is devoid of some specification which is present in its ‘heavy’ counterpart, and there is some derivational process of composition. Earlier analyses have placed this process in the lexicon, though some later accounts like Butt 1979 place it in the syntax.

3.2 Syntactic derivations In this section, I will propose a syntactic solution, making use of lexical idiosyncrasies only for N and only where no general principle is available. I propose that complex predicates are formed in the course of a syntactic derivation consisting of operations of MERGE,AGREE and MOVE, which allow for theta discharge to arguments and checking of features (Chomsky 1995, 2000, Ura 2000). With introduction of Minimalist assumptions, D-structure has no status as a syntactic representation

13 (Chomsky 1995). I will assume that the complex predicate sentences in Hindi/Urdu is formed by the operation of MERGE which combines the nominal N with one (or more of its arguments, and with V. Either V combines with a fully saturated N, all of whose theta roles are discharged by that point in the derivation, or N and V combine in a different way, resulting in the combination of argument structure.

3.2. 1 MERGE and theta discharge

The basic structure I propose for Hindi/Urdu is given in somewhat simplified form in (26a-c), showing.successive stages of MERGE and other operations (later tree structures will give just the results up to that point in the derivation. The event noun N has full categorial specification as N, but untypically for N, it has a theta array of its own (Grimshaw 1990)

26) a NP <1,2*,e> Theta projection . 3 DP- Gen/Loc N <1,2, e> Theta discharge

The N first merges with a DP, which gets the internal argument theta role. The genitive or locative case is checked within the nominal projection which are formed by MERGE. The theta grid with one discharged role is projected upward to the phrasal node.

26)b V' <1,2*, e> Theta merger (identification) 3 NP <1,2*,e> V [Light verb] <1,2,e> 'do' 3 DP- Gen/Loc N <1,2, e>

At this point the derivation can take two distinct paths, one of which is shown in (26b). Here the NP with an open theta position merges with V, but does not discharge the internal theta role of V. Instead the event arguments of N and V are identified and the theta grid of N with a discharged theta position is projected in the verbal phrase.

Finally, in (26c), the DP external argument merges with the V projection, discharging the external argument theta role. The DP checks its case in Spec/TP if it is nominative; if it is ergative its case is determined both within VP and Spec/TP (Davison to appear).

26) c. VP <1*,2*,e> Theta discharge 3 DP [Subj] V' Theta merger (identification)

14 3 NP<1,2*, e> V [Light verb] <1,2, e> 'do' 3 DP- Gen/Loc N <1,2, e>

The alternative is for the NP to count as the internal object of V, projecting V's theta grid. The NP has an open theta position for the external argument. The results are ill-formed as a LF representation (27), in contrast to the well-formed (28), which contains no open positions: ______27) | | * TENSE(I) [ DP(j) ...[V<1(j), 2(k), e(I)> NP(k) <1 , 2(m), e> DP(m) |______| |______| |______|

28) = (4) ______| | TENSE(I) [ DP(j) ...[V<1(j), 2(k), e(I)> DP(k)] |______| |______|

In the ill-formed representation, the N discharges the internal argument role of V. N has dis charged its own internal object role for DP(m), but its external argument role is undischarged, as is its event argument, which is not bound by TENSE. Arguably, the sentence TENSE cannot bind into NP with a separate argument structure from V. In the well-formed derivation, the theta merger (or event identification) of N and V means that the external position of N is projected as the external argument position of V, and TENSE binds the common event variable.

3.3 Syntactic derivation: case checking, theta discharge and theta merger It has been shown in previous sections that (a) N discharges a theta role for the internal argument, and determines its case within the NP projection , and (b) N determines the theta role of the external argument but V determines whether it will have ergative structural case or a lexical case such as the dative. This difference corresponds to the course of the syntactic derivation, which will be outlined below for the three possible solutions to the contradictions posed by combining two theta-assigning heads.

3.3.1. Genitive In Solution I, the thematic object gets default genitive case in NP, following the course of merger and projection outlined above in (26)

29). VP <1*,2*,e> Theta discharge Solution I 3

15 DP [Subj] V' <1,2*,e> Theta merger 3 NP <1,2*,e< V [Light verb] <1,2, e> kar 'do' Theta discharge 3 DP-Gen N <1,2,e > khooj (f) 'search'

30). ganapat singh-nee [eek naii bimaarii-kii khooj] kii hai Ganpat Singh-erg one new illness-Gen search(fs)-Nom do-pf-fs. is

'Ganpat Singh has discovered a new disease.' (Bahl 1974:222)

I am assuming that the genitive -kaa forms a Kase Phrase (KP) with the thematic object. The KP is a specifier of a nominal projection, agreeing with the head N.7 The N itself gets nominative case, which is checked by Tense in the relation AGREE, which does not require (Chomsky 1999), Ura 2000: 48-9, 123 and passim).

3.3.2. Locative object Solution II is like solution I, except that DP gets a locative case The choice of locative case is determined by specific Ns, from among the possibilities in (31):

31) a. -see 'from b. -par 'on' c. meeN 'in' b. -see selected by inkaar 'denial', preem 'affection, love' c. -par selected by afsoos 'regret', vishvaas 'trust d. meeN selected by vishvaas 'belief'

The structure (32) represents in condensed form the derivation of the sentence (33)

32) VP <1*, 2*,e> Theta discharge Solution II 3 DP [Subj] V' <1,2*,e> Theta Merger 3 NP[<1,2*> V [Light verb] <1,2> kar 'do' Theta discharge 3 DP-see 'from' N <1,2,e> 'refusal' -see

32) akbar kii janmjaat pratibhaa-see kooii inkaar nahiiN kar sak-taa Akbar-gen inborn talent from someone denial not do can-impf

'No one can deny Akbar's inborn talent.' (Bahl 1974: 264)

16 I am assuming that the locative case is a lexical case, associated with specific Ns and theta roles. I am also assuming that it is assigned at the same time as theta discharge, so that the DP in this instance does not have to be in some construal relation with a functional head to license case, in contrast to the licensing of structural case. While not much is said about lexical case, it appears from the contrast with structural case that lexical case checking is linked to theta discharge (Chomsky 1995). Ura (2000) note that a parametric value for structural case is that it can be checked in the same position where an argument receives a theta role, the same assumption made about lexical case.

3.3.2.1 The inconsistency of N and locative case This analysis suggests that DP-loc and N form a constituent which can occur freely on its own, just as DP-genitive N does. But this is not the case. The locative case does not freely combine with nominals alone (Jayaseelan 1988, Verma 1993.

33) a. siitaa-kaa preem b. *siitaa-see preem Sita-gen love Sita-from love

'the love of Sita' 'love with Sita

This would suggest that in Solution II structures, the thematic object does not combine immediately with N, as it does in the Solution I genitive-marking construction. Jayaseelan 1988 proposes that there is a delay in theta discharge in the derivation of complex predicates in Malayalam; so that the theta grid projects upward to combine with the V, then the locative argument is combined syntactically and its theta role is discharged.8

34) VP Solution II -Alternative 3 NP [Subj] V' <1,2*,e> Theta discharge 3 DP Dir Obj]-loc V' <1,2,e> Theta Merger 3 N <1,2,e> preem V kar <1,2,e> 'do' -see 'love'

While case licensing may or may not be postponed, theta assignment/discharge is never postponed; it is always a property of MERGE involving a theta-assigning head and a phrasal argument This postponement of theta discharge and lexical case assignment goes directly against the assumption that theta discharge is not a matter of checking features on DP , which can be postponed until an appropriate head H is merged. The features of H can enter into a feature-checking relation with DP’s features, as with structural case assignment is. (Chomsky 1995) If theta roles are not features which motivate movement, then theta assignment is strictly local and immediate, dependent on merger

17 of DP and a theta-marking head. For this reason, I will retain the original version of Solution II in (34), which assigns lexical case and a theta role before theta merger, looking further for an explanation of the ill-formedness of (33). .. 3.1.3 Lexical representation The verb kar-naa in all these compounds is the same, but it does not select either genitive or locative case for its arguments, unlike the verbs in (15) a-d. . The licensing of structural genitive case is a property of N, which in other instances may select a locative case. So I'll assume that N may either have a lexical case choice linked to an argument (35a) or not (35b):

35) a. N inkaar 'denial' <1, 2, e> + kar-naa 'do <1, 2, e> loc

b. N khooj 'search' <1,2, e> + kar-naa 'do' <1, 2, e>

If no lexical case is selected, as in (20b), the DP object of N gets structural genitive case. Some of these Solution II predicates may also adopt Solution I, with structural genitive case for the object of N, so that they have both representations (33 ). 9

3.3.4. Structural object case on DP Solutions I and II described above have in common that N forms a phrase with its thematic object. There is a third solution which differs from the other two in that N and V behave as a unit, excluding the thematic object of N. The difference lies in the case-marking of the internal argument, not the external argument. Instead of genitive case on the thematic object, there is the structural case which would be normal for the verb alone (the 'disappearing genitive' construction of Hook 1979, 158-9) . There are several possibilities, which were illustrated in (5) above for simplex verbs. . The structural dative is used for objects of complex predicates with kar-naa 'do', and nominative with combinations of N and aa-naa 'come'. I call this option Solution III: It is an option for a small number of N+V which would otherwise mark the thematic direct object with the genitive.

These pattern of case marking and agreement are expected if the N and V form a single unit, like a simplex V. The thematic object is also the syntactic direct object of N+V. N does not have the status of a syntactic object, so it does not trigger agreement. Solution III is represented by the derivation producing the structure in (36 ):

36) VP Solution III 3 NP [Subj] V' <1,2*,e> Theta discharge 3 DP Dir Obj]-Nom V' <1,2,e> Theta Merger

18 3 N <1,2,e> isteemal / yaad V kar <1,2,e> 'do' 'use'/'memory'

37) a. maiN-nee raam -koo yaad ki-yaa 'I remembered Ram' (Solution III) I-erg Ram-dat memory-f do-pf-ms

b. mujhee raam yaad aa-yaa 'I remembered Ram' (Solution III) I-dat Ram-nom memory-f come-pf-ms

I represent Solution III as the result of Theta merger taking place at the Xo level, before N has combined with a thematic DP object, as it would in solutions I and II. Compare (34) with Solution I (38):

39) VP Solution I 3 NP [Subj] V' Theta Merger <1,2*,e 3 NP <1,2*.e> V [Light verb] <1,2,e> kar 'do' Theta Discharge 3 DP-Gen N <1,2,e> isteemaal 'use'/ yaad 'memory

In Solution I, theta discharge takes place at the merge of DP and N, followed by forced theta merger. In Solution III, theta merger takes first, perhaps in the lexicon, followed by theta discharge of the two argument positions. If N+V formed a syntactic unit with two separate argument structures, the same problem would arise in syntactic derivation as in Solutions I and II, resulting in a structure like the ill-formed LF representation (27); the two syntactic arguments would discharge only two of the four available theta roles.

N has argument status in solution I, but not in solution III. Its argument status is neatly reflected in the differences of agreement in Hook's minimal pair in (37a,b):

39)a. diidii-nee [apnii aqal - kaa] isteemaal ki-yaa [Solution I] elder-sister-erg. self's wits-fs of use(ms) do-pf.ms

'Sister used her wits'. (V agrees with DP-kaa isteemaal 'use of DP') (Hook (1979:158))

b. bhaiyaa-nee [apnii taaqat] isteemaal kii [Solution III] brother-erg. self's strength-nom-fs use-ms do-pf-fs.

19 'Brother used his strength' (Hook (1979:158)) (V agrees with the thematic object

The difference between (39a) and (39b) lies in the case-marking of the thematic object, with consequences for what triggers agreement on the verb. The contrast underlines the reality of the syntactic difference between Solutions I/II and Solution III: only in Solution III is the thematic object marked as a syntactic object of V. In Solution I, the thematic object DP is very clearly dependent on the N, because of its genitive case. In Solution II, the locative lexical case is determined by N (12)- (13). In Solution III, the thematic object is not grouped with N, so that both theta assignment and case licensing are postponed to the N+V phrase. This is what makes Solution III different from Solutions I and II. The alternative form of Solution II (34) also postpones theta discharge and case checking.

3.3.4.1 Accounts of Solution III The vast majority of N + V complex predicates have the structures I or II. Solution III is a very marked option, in that it is not freely available for all the event nominals allowing Solution I.10 The Ns which allow Solution III include the following, and while no exhaustive enumeration has been made of this option, it would probably consist of not many more than these Ns.:

40) The Ns which allow Solution III include the following . a. yaad (f) memory b. isteemal (m) use' c. qatal (m) 'murder' d. anubhav (m) 'experience, feeling' ahsaas (m) perception, feeling' e. talaaš (f) 'search' cf. V talaaš-naa 'search for' f. kharc (m) expense V. kharc-naa 'spend' g. niilaam (m) auction h. naql (f) copy, imitation i. kalpanaa (f) imagination j. daan (m) gift k. pariikSaa (f) examination; niriikSan (m) test

The lexicon would list this relatively small number of Ns combined with kar-naa ‘do’ (and in some cases some other verb like aa-naa ‘come’), with a composite argument structure. The verb would still determine the case of the subject (18)-(19). If the alternative Solution II is chosen, it cannot be differentiated easily from Solution III, and it would be difficult to explain why Solution II is so productive, while Solution III is highly restricted.

3.3.4.2 Alternatives to Solution III An alternative to theta merger in the lexicon would be Noun Incorporation (Baker 1988, Hale and Keyser 1993). This solution would put the N+V combinations of N with a on a par with

20 denominal verbs like laugh, formed by incorporating N into a null V position. a derivation in Hale and Keyser’s l-syntax, a syntactic representation representing lexical relations rather than the actual sentential syntactic structure (s-syntax).

This lexical incorporation analysis in fact is a reasonable conception of HU verbs like those listed in (41), which coexist with the N+V Solution I complex predicate in (42) . Many verbs in Hindi/Urdu have verb stems which are formally identical to N, such as Dar-naa .’ fear’ and the N Dar (m) ‘fear’.There is an interesting natural experiment available in Hindi/Urdu.

41) a. khooj-naa 'search for' 42) a. khooj kar-naa 'search, discover' b. bahas-naa 'argue' b. bahas kar-naa 'argue, debate' c. talaaš-naa 'search according to c. talaaš kar-naa 'search for a concrete object' abstract criteria' d. Dar-naa ‘fear’ d. Dar hoo-naa ‘fear, be afraid of’

For many complex predicates consisting of N +V, there exists a simplex verb (Gambhir 1993). Such verbs could be formed by merging a N or unspecified root with a null V head, ‘conflation’ in Hale and Keyser (2002). The syntactic consequences of this approach remain to be explored for HU, especially in view of the revisions to it in Hale and Keyser 2002. A problem to be accounted for is that the N+V combination differs slightly in meaning from the simplex verb (Gambhir 1997:90- 92)::The N+V combination has a telic interpretation, implying a discrete and structured object which delimits a completed act.11

Hindi/Urdu does not have the characteristics of syntactic polysynthetic languages which incorporate objects as a method of case licensing (Baker 1995). Instead of treating complex predicates as syntactic incorporation, we can make use of lexical incorporation (Hale and Keyser 1993, 2002) for another kind of verb. We need to distinguish the simplex V khooj-naa (41) from the complex predicate (42) , because they differ in range of meaning and also in verbal aspect. The simplex verb could be lexically derived by associating the N khooj ‘search’ with a null V head (Hale and Keyser 1993). If we were to assume that Solution III N+V combinations were derived in the same way as the simplex V derived from N, we would not be able to explain why the N+V combination has a telic interpretation which is not a lexical property of the simplex V (see note 7).

Mohanan 1994 proposes a very intricate incorporation account of the difference between the Solution I/II verbs and my Solution III. She requires the addition of a level of representation to the Lexical-Functional Grammar framework she uses. The N component of N +V is both a semantic argument and a grammatical-function argument in the combination which I call Solution I; the index of N is linked to a syntactic argument position. In the Solution III combination, she treats N as a semantic argument of V, but its index is not linked with a syntactic argument position. Only syntactic arguments trigger agreement. The syntactic subject of V is linked (semantically) to the external

21 argument of N. Her account of Solution III is essentially a N incorporation account, which she extends to derive generic activities from accomplishments (Mohanan 1997).

The syntactic incorporation view of Solution III would also require that there are light versions in the lexicon of kar-naa ‘do’ and aa-naa ‘come’, equivalent to null V heads. This assumption would undermine the assumption that Vs are always lexically specified for an argument structure. If N and the light V are not in a complement-head relation, then there is no explanation in terms of LF interface conditions. The necessity for N+V combination to form a syntactic unit would be unmotivated.

Another alternative solution is to adopt Dubinsky's proposal for Japanese (1997) that some event terms are not fully specified as N, so they are like Adj in not triggering agreement or licensing genitive/locative case. In HU, some bivalent adjectives do have case-licensing properties, including locative case:

43) Adjectives licensing case a. jaa-nee -koo taiyaar ‘ready to go’ b. gadhee jaisii aawaaz ‘voice like a donkey’ go-inf dat ready donkey-obl like voice c. apnee-see/meeN sharmindaa hoo-naa ‘to be ashamed of oneself’ (Bahri 1992: 609) self-from/in ashamed be-inf.

The Dubinsky 1997 solution is closer to the spirit of this proposal in that the lexical exceptionality of the construction lies in the properties of the N (as less specified categorially), though he also requires the V to have the light interpretation. I do not decide the case here between my Solution III and Dubinsky’s categorial underspecification, as either one captures the lexical exceptionality of these cases, and preserves the generality of the other two types. A strictly lexical solution does not easily distinguish the productive syntactic combinations from the very restricted N+V units behaving as a single head.

4. 'Light' verbs I am making the claim that complex predicates in Hindi/Urdu are made of productive syntactic combinations of N and V, both of which have argument structures, and do not differ from their independent uses in other contexts. . Well-formed derivations lead to interpretations in LF which meet the interface condition defined above. This conception of the construction clearly would risk overgenerating combinations which are not used as complex predicates. Lexical accounts must specify each combination in the lexicon, with all the exceptional properties associated with it. Here I ask what the properties are of N and V which make them suitable for forming complex predicates if the process is productive and syntactic.

4.1 There are too many Vs for them all to be special ‘light’ verbs.

22 While 'do' verbs normally enter into these combinations in many languages, Hindi/Urdu allows combinations with a range of different verbs, both transitive and intransitive. Some, such as dee-naa 'give' and rakh-naa 'place', are ditransitive.

44) Verbs in Hindi/Urdu which can be used as light verbs with N Transitive a. kar-naa 'do, make' b. dee-naa 'give' c. rakh-naa 'put, place' d. khaa-naa 'eat' (in combinations with the meaning 'suffer, endure, go through N') e. lagaa-naa ' cause to strike, put' f. uThaa-naa 'raise' g. pahuNcaa-naa 'cause to arrive' h. maar-naa 'beat' (as in haar maar-naa 'to accept defeat', lit. 'to beat defeat'). i. paa-naa 'find'

45) Intransitive a. aa-naa 'come' b.. hoo-naa 'be, become' c, lag-naa 'be attached, strike' d. cal-naa 'walk, go to' e. pahuNc-naa 'arrive'

Compare the list in Hindi/Urdu (44)-(45) with Persian, which has the following commonly used verbs (from Karimi-Doostan 2002)

46) a. kard-an 'do' b. šod-an 'become' c. da:d-an 'give' d. ya:ft-an 'find' e. da:st-an 'have' (statives only) f. zad-an 'strike' g. xord-an 'collide, eat'

There is a certain similarity between verbs used in the two languages for complex predicates. suggesting there is some cluster of properties which makes a verb able to enter into a complex predicate combination..Butt 1995 remarks on the similarities among many languages.

4.2 Some verbs are ‘lighter’ than others. DiSciullo and Rosen (1990) derive light verbs from ordinary verbs which are optionally projected

23 into a verb structure with unspecified arguments The analysis proposed here assumes that there are no special 'light' verbs derived from ‘heavy’ verbs. Rather, the V part of N+V is an ordinary verb, with argument taking, case asigning and aspectual properties like other verbs, which undergoes event identification and theta merger in the syntactic derivation . (see also Rosen 1990, Butt 1995 and Karimi Doostan 1997). I would suggest that these properties include the lack of specific selection properties (or argument ‘qualia’ in Pustejovsky 1995, such the object properties associated with bake)12. Complex predicates are in some sense inherently ‘light’; they have argument structures and aspectual specification, but don’t impose many restrictions on their arguments. On this view, all languages have some verbs which are ‘lighter’ than others, not as completely specified for meaning by selection properties.13

Verbs which can be used as a ‘light’ verb are also of a particular aspectual type; they are all able to have telic senses The list of verbs in Hindi/Urdu (44)-(45) includes verbs of transfer, and change of location, which can be understood metaphorically as change of state and leading to a resulting state; in other words they are accomplishments. The accomplishment interpretation of an event is often signaled syntactically by the presence of a quantized object or goal phrase.

In an even more general way, the interpretation of the event is heavily influenced by the nature of the complement (Pustejovsky 1995:87- 88).14 So if a verb is not heavily specified for qualia and has a telic event structure, it can serve as a light verb without having to undergo special deprivation of lexical properties. This characterization preserves the observations about which there is substantial agreement but avoids the elaborate specification outlined in Grimshaw and Mester 1988 and subsequent lexical accounts of special, thematically empty light verbs.

5. Event nominals and argument structures

If the V component of N+V is not lexically different from ordinary verbs, what about the N components of N +V? Complex predicates are listed by N in the dictionaries of Hindi and Urdu (such as McGregor 1995, Bahri 1992). I think this is because the N+V combination owes much of its semantic content to N. The Ns in N+V all have complex event interpretations, as well as a result interpretation, like the English nominalizations discussed in Grimshaw 1990.

Event nominals in Hindi/Urdu are not as morphologically transparent as the ones in English. They come from Hindi/Urdu, and have the form of stems, such as khooj ‘search’, pahcaan ‘identification’; or they come from Sanskrit, pariikSaa ‘examination’, Persian afsoos ‘regret’, or Arabic ahsaas ‘feeling’, or English Teelifoon ‘telephone call’.15

It is not possible to distinguish formal sub-classes of N, such as -ation nominalizations and -ing .16

24 The closest counterpart of -ing in Hindi/Urdu is the suffix -naa, which is actually a tense form, and the V-naa combination is syntactically verbal rather than nominal (pace Butt 1995), even though some adjectival agreement is possible within the infinitival suffix. But the restrictions on argument projection which Grimshaw 1990 notes for English also hold for Hindi/Urdu. There is a restricted set of cases which N can assign on its own, basically the genitive forms and by in English and Hindi/Urdu -genitive and kee liyee ‘for’17. Event nominalizations in Hind/Urdu do not form binding domains in which the subject of N can be coindexed with another argument of N. Binding is possible only with a verbal host for N (Davison 1999, 2000). The question is whether the N is defective in ability to assign cases, or in the capacity to assign theta roles to syntactic arguments (as opposed to modifiers).

N in Hindi/Urdu very clearly lacks the ability to assign structural cases, nominative, ergative and direct object dative. these cases are licensed in HU by Tense/Aspect and functional projections within the verbal projection (Davison to appear). Lexical case as well as structural case is dependent on a verbal projection (32, 33). I will assume that event nominals are lexically specified for event structure/ argument structure, but require a verbal host (as a syntactic argument of V). Only V licenses higher verbal projections and functional projections like Tense.

Event identification, theta merger and the semantic structure of N and V In previous sections, I have focused on the predicate-argument structure of verbs, using the ‘theta grid’ notation to keep track of theta discharge on syntactically licensed arguments. In this next section, I will adopt a different and more finely-grained representation of the syntax and of light verbs and event nominals. What follows is highly speculative.

To represent accomplishment events, I will adopt the notation resolving accomplishments into three subevents (Higginbotham 1999, Hale and Keyser 1993, Ramchand and Butt (2001), Folli 2002):18

47) a. e1 -->

A complex event (47a) consists of an initial or instigating event e1, which logically/temporally precedes the telic pair , linked process and result events which (normally) share the same individual as argument. . The three subevents correspond to projections of three heads (48b), where v is the head of the instigating event, V is the process, and Rv the resulting state. The internal argument corresponds to the arguments of both e2 and e3 (48c).

48) a. Gianni ha chiuso la finestra b. chiudere ‘close’ : +v +VP +Rv John has closed the window Folli 2001:107, c. E --> e1 --> & e1 causes and Initiator (x, e1), and Resultee (y, , and

25 Cause (e1) and Process (e2) and Final State (e3). Folli 2001:114.

Hale and Keyser 1993 propose that theta roles are derived from structural relations. An argument has a semantic role by virtue of its relation to a verbal head, representing a sub-event in a complex event structure (47). The theta grid in (49a) thus corresponds to the event structure in (49b), for a verb with a theme argument undergoing a change of state, and an instigating external argument:

49) a. <21, 22, e> b. e = e1 -->

I am assuming that the transitive V component of N + V has this accomplishment structure (44), or just the process-result subpart of it if the V is intransitive (45).

For example V. hoo-naa ‘be. become’ combines with N talaash ‘search’ The result is the intransitive counterpart of talaash kar-naa ‘search for’19. Interestingly, this is not otherwise available in the ‘conflated’ verb paradigm talaash-naa ‘search, which has neither an intransitive nor a causative counterpart (55b):

56) a.. talaash hoo-naa talaash kar-naa talaash kar-vaa-naa 'be searched' 'search for' 'have someone search for'

b. *talash-naa talaash-naa *talash-vaa-naab. 'be searched' 'search for' 'have someone search'

The argument of e1 of talaash ‘search’ is not projected as a syntactic subject (though it may be expressed as an with -see ‘with’):

58) (pulis-see) us-kii talaash (nahiiN) hu-ii CHECK police-by 3s-gen search not be-pf

‘He/she was/could not be searched for (by the police)

Complex event structure is projected syntactically into a complex verbal projection with three heads. Butt and Ramchand (2001) propose the following syntactic structure to project the sub-events of a combination of an activity verb V1 with a ‘vector’ verb V2, which yields a telic meaning. V1 V2 49) nadya-nee xat likh li-yaa Nadya-erg letter write take-pf

‘Nadya wrote a letter (completely).’ Batt and Ramchand 2001)

26

Simplex verbs like likh ‘write’ are not always specified for telicity, even if they are perfective (Singh 1999). In V1-V2 combinations, a V2 like lee-naa ‘take’ adds manner and aspectual information, as well as a telic interpretation (Hook 1974, Butt 1995, Chatterjee 1988). The accomplishment interpretation is syntactically composed as in (50).

50) VP [Verb-Verb compounds] 3 DP v' 3 VP v [cause in some broadly conceived sense] 3 DP(I) V' object of change 3 RvP V = V2 change ‘take’ 3 DP(I) Rv =V1 result state ‘written’ object of result-xat ‘letter’

The light verb v project corresponds to e1, while the telic pair is projected as VP and RvP (result phrase), sharing an argument (xat ‘letter’).

Solution III combining N + Vt involves some sort of lexical specification which differentiates it from from (50) and also. Solutions I/II. I’ll propose complete event identification for Solution III, as in (51)-(52), which is stipulated as a marked (ie non-default) kind of lexical entry.20

51) VP [Solution III] 3 DP v' 3 VP v [cause in some broadly conceived sense] 3 DP(i) V' object of change 3 RvP V Process 3 DP(i) N-V -result object of result 3

27 N V

This combination clearly blocks any possibility of N being theta marked as an argument of V, as N is not the specifier of a verbal head which represents a sub-event. This is the option option for Persian, which marks the thematic object with the marker -ra, a case assigned by verbs, not nouns (see Karimi Doostun 1997 for an account of event identification in Persian).

52) yaad ‘memory’ e1 --> kar ‘do’ e4 --> | | |______|______|____| | |______|______| |______|

The structure for complex predicate of types I or II must be fifferent from (51)-(52) in that the combination is not lexically specified, and the identification of events proceeds by stages corresponding to the syntactic structure (53). The V kar-naa 'do' or other in tposition of the head of VP, the event N is the head of the Result P. This structure is very slightly different from the notation employed in the early sections of this paper.

53) VP e1 --> idetification of e1, e4 [Solutions I and II] 3 DP v' 3 VP v e4 Idenftification of telic pairs 3 DP(i) V' object of change 3 NP V = kar-naa change 3 DP(i) N e2, e3> object of result

This sequence of event identification proceeds in stages, corresponding to the the syntactic structure in (53). The N merges first its DP internal argument, and binds it into the event structure e2 The DP is coindexed with the argument of e2, yield the reading in which the DP is the argument of the telic pair .

54) (raam-kii yaad ) Translation, associated with (54c) NP: 8e1, e2, e3, x, y z[remembered (e3) & Final state (e3) and resultee (x, e3) & raam (x) & Process(e2) & Undergoer(y, e3) & raam(y) [& cause (e1, & Initiator (z, e1]

28 Here the indices for raam as Resultee and Undergoer are bound into the event structure, and this DP raam is associated with the theta roles determined by N. The NP itself does not combine in a function-argument structure with V (as the object of Process and Result) because NP is not fully saturated. It contains the variable z (Initiator (Z,e1)) not associated with an argument, violating the Interface Condition discussed above in (27)-(28).

This pair of events (e2, e3> is identified with associated with kar-naa 'do'. The identification of telic pairs only avoids this outcome, because it associates only part of the N's event structure with V. N is in syntactic object position but does not get associated as a constituent with the event structure of V.

The final step involves the identification of the external argument positions. The derivation is summed up as (55):

55) a. N e1 --> V | DP b. . N e1 --> V | | DP DP

c. N e1 --> V | | ______|_____| |______|

d. N e1 --> V | | | ______|______|_____| | |______|______| |______|

. I propose that the alternative is the identification of e1 and e4.

55) yaad ‘memory’ e1 --> kar ‘do’ e4 --> |______|

The dative-experiencer predicates (20b)and even the 'intransitive' version in (47), (49) will be derived in the same sequence in (54). I am assuming that N always has the same semantic contents, including an institigating event. Its argument is projected with the lexical case -koo in (20b) and as an optional instrumental argument in (49). 21

29 5. Conclusion

Complex predicates of the form N+V present a significant challenge to syntactic theory, because of the productivity of these combinations in a language (such as Hindi/Urdu, discussed in this paper), and the commonalities among languages which have the construction. Th N is not a thematic complement of V, because N is a complex event nominal which has its own thematic structure . N determines the interpretation of the arguments of N+V, but V retains its aspectual and case-marking properties. Syntactic and morphological evidence shows that in at least the principal expression of the construction, the N combines to form a constituent with its thematic object. The combination has an accomplishment interpretion by virtue of its syntactic structure, as the N represents a result in combination with V, which may alone have an acitivity interpretation.

The goal of this study has been to derive and explain these properties as a the result of general conditions on syntactic derivations and the interpretative interface, without appeal to special lexical stipulations or derivational procedures. I have represented the telic event structure of both N and V as composed of three subevents, which define the theta role of their arguments. The external argument is defined as the subject of the instigating event, the internal argument undergoes a procress or change which leads to a resulting state, the telic pair of events . The N merges with DP the subject of its resulting state and process, which is marked with genitive or lexical case.

N at this point in the derivation is still not fully saturated because it is unable to license its external argument syntactically. but not with its external argument.The NP cannot merge with V in a function- argument relation, theta-marked by V. If it did it would violate a condition on theta marking and the subsequent LF representation. This representation would be ill-formed, violating Full Interpretation, because it would contain sub-events (defining theta roles) not associated with some argument. This condition on Full Interpretation is one of the principal means of explaining why the syntactic derivation does not treat N as an argument of V.

Instead, the N’s internal sub-events are identified with the internal sub-events of V, a process of sub- event construal which is found in a range of constructions with telic interpretations. Finally the two external argument sub-events are idenified. A syntactic variant of event identification takes place in a N-V compound structure, identificying all sub-events at once. This exploitation of a general process of theta-role/sub-event identification is the second general principle which derives the properties of N+V and explains its productivity.

Instead of assuming a special lexical derivation of ‘light’ verbs, I account for the greater semantic influence of N as the greater inherent specification of N as denoting an event of a specific type, compared with the relatively sparse specification of the Vs which can occur in N+V combinations. These verbs have an argument structure and a telic event structure, but not much in the way of ‘qualia’ specifications. In any case, many ‘lighter’ verbs show the effect of their own thematic

30 arguments on the interpretation of their arguments. The lighter Vs will tolerate additional arguments specified by N, but they must be licensed by lexical case.

In place of the painstaking and well-observed lexical stipulations of earlier accounts, I have tried to offer an explanation derived from general and well-grounded principles of event composition and syntactic derivation.

References

Bahl, K.C. (1974) Studies in the semantic structure of Hindi.Vol. I. Delhi: Motilal Banarsidass. Bahl, K.C. (1979) Studies in the semantic structure of Hindi, Vol. 2. Delhi: Manohar. Bahri, Hardev (1992) Learner's Hindi Dictionary. Delhi: Rajpal and Sons. Baker, Mark (1988) Incorporation. Chicago: University of Chicago Press. Baker, Mark (1995) The Polysynthesis parameter. Oxford: Oxford University Press. Butt, Miriam (1995)The structure of complex predicates in Urdu. Stanford: CSLI Publications Butt, Miriam and Gillian Ramchand (2001) Complex aspectual structure in Hindi-Urdu. In B Jensen at al, Oxford Working Papers in Linguistics 6. Oxford: Committee for Comparative Philology and General Linguistics Carstairs-McCarthy, Andrew. (2000) The derivation of complex languages. Oxford: Oxford University Press. Cattell, Ray (1984) Composite predicates in English. Syntax and semantics, 21. Sydney: Academic Press. Chatterjee, Ranjit (1988) Aspect and meaning in Slavic and Indic. Amsterdam: benjamins. Chomsky, Noam (1981) Lectures on Government and Binding. Dordrecht: Foris Publications Chomsky, Noam (1986) Knowledge of language. New York: Praeger Chomsky, Noam. (1995) The Minimalist Program. MIT Press. Chomsky, Noam (1999) Derivation by phase. MIT Occasional Papers in Linguistics 18. Davison, Alice (2000) Lexical and anaphors in Hindi/Urdu. In B. Lust, K. Wali, J. Gair and K.V. Subbarao (eds.) Lexical anaphors and pronouns in selected South Asian languages. Berlin: Mouton deGruyter, 397-471. Davison, Alice (to appear) Structural case, lexical case and the verbal projection. In Veneeta Dayal and Anoop Mahajan (eds) structures in South Asian languages. Dordrecht: Kluwer Academic Publications. DiSciullo, Anna-Maria and Sarah Thomas Rosen (1990) Light and semi-light verb construction. In K. Dziwirek, P. Farrell and E. Mejias-Bikandi (eds) Grammatical relations: a cross-theoretical perspective. Stanford: CSLI, 109-127. Dubinsky, Stanley (1997) Syntactic underspecification and light-verb phenomena Linguistics 35, 627-677. Folli, Rafaella (2001) Constructing telicity in English and Italian. Wolfson College, Oxford University unpublished dissertation.

31 Gambhir, Vijay (1993) Complex verb phrase: a diachronic and synchronic view. In M. Verma (ed) The complex predicates in South Asian languages. Delhi: Manohar Publications, 77-96. Grimshaw, Jane (1990) Argument structure. Cambridge: MIT Press. Grimshaw, Jane and Armin Mester (1988) Light verbs and 2 marking. Linguistic Inquiry 19, 205- 232 Hale, Kenneth and Samuel Jay Keyser (1993) On argument structure and the expression of syntactic relations. In Kenneth Hale and Samuel Jay Keyser (eds) The view from Building 20: essays in honor of Sylvain Bromberger. Cambridge: M.I.T. Press. Hale, Kenneth and Samuel Jay Keyser (2002) Prolegomena to a theory of argument structure. Cambridge: M.I.T. Press. Higginbotham, James (1985) On semantics. Linguistic Inquiry 14, 545-597. Higginbotham, James (1999) Accomplishments. Unpublished paper. Higginbotham, James (2000) On events in linguistic semantics. In James Higginbotham, Fabio Pianesi and Achille Varzi (eds.) Speaking of events. Oxford University Press, 49-80. Hook, Peter (1973) The in Hindi. Ann Arbor: University of Michigan Center for South and Southeast Asian Studies. Hook, Peter (1979) Intermediate Hindi Structures. Ann Arbor: University of Michigan Center for South and Southeast Asian Studies. Jackendoff, Ray (1974) A deep structure projection rule. Linguistic Inquiry 5, 481-506. Jayaseelan, K.A. (1988) Complex predicates and 2 -theory. In. W. Wilkins (ed.) Syntax and semantics 21, Thematic relations. San Diego: Academic Press, 91-111. Karimi-Doostan, Go. (1997) Khan, Baber S. (1989) A note of disagreement about agreement in Hindi-Urdu,. Linguistics 27, 71- 87. Masica, Colin P (1991) The Indo-Aryan languages. Cambridge: Cambridge University Press. McGregor, R.S. (1995) Oxford Hindi dictionary. Oxford: Oxford University Press. Miyagawa, Shigeru (1989) Light verbs and the ergative hypothesis. Linguistic Inquiry 20, 659-668. Mohanan, Tara (1994) Arguments in Hindi. Stanford: CSLI Publications. Mohanan, Tara (1995) Wordhood and lexicality: Noun incorporation in Hindi. Natuiral Language and Linguistic Theory 13, 75-134. Nespital, Hellmut (1997) Dictionary of Hindi verbs. Allahabad: Lokbharati Publishers. Pesetsky, David (1995) Zero syntax. Cambridge: M.I.T. Press. Postal, Paul (1971) Cross-over phenomena. New York: Holt, Rinehart and Winston. Pustejovsky, James (1995) The generative lexicon. MIT Press. Pustejovsky, James (1998) The semantics of lexical underspecification. Folia Linguistica. Rosen, Carol Thomas (1990) Argument structure and complex predicates. New York: Garland. Shapiro, Michael C (1989) A primer of modern standard Hindi. Delhi: Motilal Banarsidass. Speas, Margaret.(1990) Phrase structure and the lexicon. Dordrecht: Kluwer Academic Publications. Ura, H (2000) Checking theory and grammatical functions in Universal grammar. Oxford University

32 Press. Verma, Manindra K (1993) Complex predicates and light verb in Hindi. .In M. Verma (ed) The complex predicates in South Asian languages. Delhi: Manohar Publications, 197-215..

1. Previous versions of this paper were presented at the University of Iowa Department of Linguistics Colloquium and the Conference on Argument Structure, Delhi University (Janaury 2003). I thank the audience at these presentations for helpful comments. This analysis owes much to work by Dubinsky 1997, Pustejovsky 1995, Mohanan 1994, Butt 1995, Karimi-Doostan 1997, Folli 2001 and Butt and Ramchand 2000.

2. I'm assuming that null arguments are either null pronominals, or identified by unselective binding and existential closure, but these issues are not relevant to the questions discussed here.

3. This A + V combination selects locative case on an infinitive complement: i) unhooN-nee DaakTar-koo [PRO fiis lee-nee]-par mazbuur ki-yaa they-erg doctor-dat fees take-inf-on forced do-pf

‘They forced the doctor [PRO to accept his fee].’ Bahl 1979:37

4. For arguments that the ergative postposition -nee is not a marker of agency, see Masica 1990, Davison 1999 and to appear,

5. The direct object marker -koo dative marks the DP as prominent in discourse. The text examples in Bahl 1974 all take the -koo option for DPs which are anaphoric or otherwise prominent in discourse, such as: i) aur us-see mujhee bahut-sii baateeN yaad aa ga-yii thiiN [jinheeN maiN yaad and 3s-from I-dat many-like things memory come go-pf be-past Rel-pl-dat I -nom yaad nahiiN kar-tii memory not do-impf

And from that I had remembered many things [which I don’t (want to) remember/recall.’ Bahl 1974:73

6. The experience referred to is not instigatedby an agentive subject; in fact it is caused by some outside event, as in (i): i) [kaaminii-kii pahlee din kii baat sun-tee hii] maiN apnee-meeN eek ajiib hiintaa Kamini-gen firs day-gen matter hear-impf emph I-nom self-in one strange inferiority

33 kaa anubhav kar-nee lag-aa gen experience do-inf begin-pf

After what Kamini said the first day, I began to feel within myself a strange sense of inferiority. (Bahl 1974: 248)

7. A full account of NP/DP structure presents a challenge which I will not explore further here. The genitive KP is a phrase which can be displaced from DP/NP, a reason not to identify -kaa directly with a D head..

8. Grimshaw and Mester 1988 describe the lexical process of N+V argument combination as argument transfer, which has two options. The combination of N with a genitive object, my Solution I, is partial argument transfer; the N projects one of its own arguments, but the subject argument and some others are transferred to the V. The other option, full transfer, transfers all the N arguments to V, which amrks them with V-case, corresponding to my alternate Solution II. Because of the Japanese- Hindi/Urdu differences of case morphology and the status of finite , the only positive evidence for full transfer is the incompatibility of lexical case and N. .

9. Japanese also has lexical cases case on the thematic object in sentences such as: i) john-wa tookyoo-kara shuppatsu sita John-top Tokyo- from departure did

‘John departed from Tokyo.’ Grimshaw and Mester 1988: 207

10. Japanese has an option like Solution III for some N+V combinations. Unlike the counterpart of Solution I (I)a, the N has null case marking, and its thematic object is accusative or null, but not genitive (I) b.. Some N+V combinations allow only this version (Miyagawa 1989). i) a. taroo wa ziroo-ni [tunami ga kuru to] -no keikoku-o sita Taro-top Ziro-dat tidal-wave nom come-comp-gen warning-acc did

‘Taro warned Ziro [that a tidal wave is coming] Dubinsky 1997:628

b. taroo wa ziroo-ni [tunami ga kuru to] keikoku- sita Taro-top Ziro-dat tidal-wave nom come-comp warning did

‘Taro warned Ziro [that a tidal wave is coming] Dubinsky 1997:628

34 11.The N khooj 'search' is given a telic interpretation 'discovery' which the verb khooj-naa 'search for' does not have. (ii). i) kalambas-nee amrikaa-kii khooj kii [khooj kar-naa 'do search'] Columbus-erg America-gen search do-pf

'Columbus discovered America.' (Hook 1974: 240) ii) kalambas-nee amrikaa-koo khooj-aa [khooj-naa 'search'] Columbus-erg America-dat search -pf

'Columbus searched for America.' iii) kalambas-nee amrikaa-koo khooj li-yaa [khooj lee-naa 'search successfully'] Columbus-erg America-dat search take-pf

'Columbus searched for America (and found it).'

The N+V combination is interpreted as telic in (I), as 'discovery' implies success, an achievement interpretation. It appears that the N+V structure itself encodes a telic interpretation (I), in contrast to the simplex verb (ii) The simplex V, though its default interpretation might be telic ('searched for and found'), is only telic in a nondefeasible way if it is combined with a 'vector' verb lee-naa 'take' (iii) . Vector verbs follow a main verb; they are verbs which are used productively in other sentences as main verbs (Hook 1974, Nespital 1997). Interestingly, the plain N+V combination in (I) implies that the discovery was unanticipated, while the N+V+V structure in (iii) suggests that it was foreseen (Hook 1974:240ff).

12. An example verbs with highly specific selection properties include kill, bake. The verb kill specifies that its second argument is an animate individual, with the formal property of being a physical object. The qualia specifications state that the result is that the object becomes dead, and this comes about by an agentive act (Pustejovsky 1995: 208). :A change of state verb like bake is an agentive act of baking, requiring an animate, physical object referent of the first argument, while the second argument is a physical object in the form of a mass.(Pustejovsky 1995:121).

13. Hale and Keyser (2002) compare verbs like put with more fully specified verbs such as shelve, bottle. The more specific verbs are the result of ‘conflating’ a noun shelf or bottle with V. If conflation does not occur, the V is expressed by put in the context: put the wine in a bottle, bottle the wine. (Thanks to Paula Kempchinsky for reminding me of the relevance of the Hale and Keyser proposal.)

35 14. The verb kar-naa ‘do, make’ on this view does not have a volitional agentive theta role for its subject. (Causative verbs also are not necessarily agentive, as they may have abstract subjects with a cause/source interpretation.) The V kar-naa can combine with experiencer Ns such as afsoos ‘regret’. Even with activities the subject of V kar-naa has a different kind of volitional agency depending on objects of kar-naa such as kaam ‘work’ and badmaashii ‘bad deeds, mischief’: i) [kaun-saa kaam /kaun-sii badmaashii kar-nee]-kee li-yee tum aa-ii hoo? what-like work / what-like mischief do-inf- gen for you come-pf be-pres

‘What kind of work/mischief have you come to do?’

15. Though the noun in English refers (only to) to a physical object, it refers to an event in Hindi/Urdu. The sentence in (I) describes a crank call, not a defective instrument. i) us-nee eek buraa Teelifoon ki-yaa ‘He made a bad telephone call. 3s -erg one bad telephone do-pf

16. See Pustejovsky 1995: 160, 169- 172 for a close comparison of the difference between inflected verbs in sentences with the corresponding V, V-ing and V-ation nominals. The differences are in event structure, in which there are two subevents: one or the other or neither may be the ‘head’ of the whole event, determining how the arguments are syntactically projected.

17.Interestingly, Japanese has a lot more latitude with Ns because it is possible to suffix the genitive -no on some of the case clitics. The case marker to ‘with’ is possible with NP if combined with -no (I), as is kara ‘from’, e ‘to’. i) [john-no bill -to -no aiseki] machigai datta John-gen Bill-with gen table-sharing mistake was

‘[John’s sharing a table with Bill] was a mistake.’ Grimshaw and Mester 1988: 223

18.The telic pair allows the VP to express a final state or telos (e3). The resulting state of an accomplishment can be modified or identified independently of the process V. Activity verbs without a telos project only one subevent

The sub-event structure in Pustejovsky (1995) is expressed somewhat differently; complex events consist of two sub-events and an ordering relation (precedence, overlap, ordered overlap). A complex event with an accomplishment aspectual interpretation consists of :

36 i) a. E1 =e1:process Restriction: <" (precedence) b. E1 = e2: state Head : e2 (focussing ont he final state) Achievement verbs have the second event as head. The exact relation of the process event to the result (eg ‘causative’ ) is as expressed as a value predicated of the event, rather than as a separate instigating event in( 47). (Pustejovsky 1995

19. Interestingly, Japanese combines unaccusative N with otherwise transitive suru ‘do’, yielding unaccusative predicates (Miyagawa 1989). Only the theme of N is projected.

20. Folli 2001: 166 uses this multiple event identification for break the vase into pieces, in which break and the accomplishment PP have the same kind of process-result sub-event structure

21. This structure represents the experiencer phrase NP-dat as a specifier to VP rather than a perhaps more plausible result phrase; the experiencer represents a terminus or goal on one possible reading of the meaning of experiencer predicates. Representations of experiencer predicates have to deal in some way with this ‘reversal’ (e.g Postal 1970, Pustejovsky 1995), including giving ‘frighten’ a causative analysis with a different theta role from ‘fear’ Pesetsky (1995).

In fact the experiencer is an object of the causative verb Dar-aa-naa ‘cause to fear, frighten’ But the dative experiencer (iii) passes the syntactic subject tests (reflexive binding and other tests, Davison to appear). It is more like the nominative subject of Dar-naa ‘fear, be frightened of’‘(I) than the dative object of Dar-aa-naa (ii). i). woo hamaaree kuttee-see Dar ga-yaa ‘He was frightened of our dog.’ (Bahri 1992: 266) 3s our dog-from fear go-pf ii) kuttee-nee [PRO bhauNk-kar] baccee-koo Dar-aa di-yaa dog-erg bark-prt child-dat fear-caus give-pf

‘The dog frightened the child with its barking.’ Bahri 1992:266) iii) mujhee apnee kuttee-see Dar nahiiN lag -taa hai ‘I am not afraid of my own dog.’ I-dat self’s dog -from fear not strike-impf be-pres

I will take the dative experiencer to be a higher rather than lower constituent of VP.

37