Reverse Complex Predicate in Hindi

Reverse Complex Predicates in Hindi

Shakthi Poornima

Department of Linguistics

SUNY University at Buffalo

[email protected]

September 22, 2008

Abstract

Hindi, like other South Asian languages, has a rich set of complex predicatesthat are formed when nouns, verbs or adjectives combine with a light verb.This paper focuses on verb-verb complex predicates and illustrates that theorder of verbs in the complex predicate construction can be reversed. I dis-cuss the two types of complex predicate constructions and show that theydiffer not only in terms of linear order but also in terms of which verb isthe constructions head. I argue that Hindi complex predicate constructionssupport the claim that case assignment constraints are lexical rather thanphrasal; in fact, the same case assignment constraints that are operative forother verbal constructions can model the facts for both types of construc-tions. In both cases, the constraints apply to the argument-structure of thehead. Working within Head-driven Phrase Structure Grammar, I proposethat the light verb inherits its argument-structure from the main verb in thestandard, but not in the reverse complex predicate construction.

Contents

1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2

2 Light Verbs Semantics . . . . . . . . . . . . . . . . . . . . . . 6

3 Syntactic Structure . . . . . . . . . . . . . . . . . . . . . . . . 10

3.1 Constituency Tests . . . . . . . . . . . . . . . . . . . . 10

3.2 Auxiliaries . . . . . . . . . . . . . . . . . . . . . . . . 16

3.3 Summary . . . . . . . . . . . . . . . . . . . . . . . . . 22

4 Case Assignment . . . . . . . . . . . . . . . . . . . . . . . . . 24

5 An HPSG Analysis . . . . . . . . . . . . . . . . . . . . . . . . 37

5.1 Case Assignment Rules . . . . . . . . . . . . . . . . . 37

5.2 Clause Union Analysis . . . . . . . . . . . . . . . . . . 44

6 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . 50

7 Appendix: List of Light Verbs . . . . . . . . . . . . . . . . . . 52

Bibliography . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 55

1

1. INTRODUCTION

1 Introduction

In a complex predicate construction, two or more verbs form a monoclausal

syntactic structure and share a single array of grammatical relations. Each

predicate composing the construction contributes part of the information

that is ordinarily associated with a head. In the generative literature, such

structures were first described for Romance languages (Aissen and Perlmut-

ter (1976) for Spanish, Emonds (1978) for French, Rizzi (1978) for Ital-

ian). Additionally, complex predicates have been noted to occur extensively

in South Asian languages (Masica, 1991; Abbi and Gopalakrishnan, 1991;

Hook, 2001; Butt, 2005).

For Hindi, an extensive descriptive study of compound verbs was first

undertaken by Hook (1974). Hindi aspectual complex predicates are formed

when a verb bearing the main predication (i.e., the main verb) combines

with a semantically-bleached light verb which modifies the main verbs

semantics by expressing subtle semantic notions. Light verbs are homopho-

nous with form-identical full lexical verbs and the term light is used to

suggest that their meaning is more abstract in comparison to their full coun-

terpart (Hook, 1977). A list of the twelve most common Hindi light verbs

and their semantics is presented in Section 2. The following examples illus-

trate a single predicate and a complex predicate respectively.1

1The gloss used for a light verb refers to its meaning as a full verb. Abbreviationsare as follows: MV = main verb, LV = light verb, F = feminine, M = masculine; Erg= ergative, Nom = nominative, Gen = genitive, Dat = dative, Acc = accusative, Inst =instrumental, Loc = locative; Inf = infinitive; Perfv = perfective, Impfv = imperfective;Pres = present; Pron = pronoun; Sg = singular, Pl = plural. The marker - indicatesa morpheme boundary, = separates a clitic from a lexical item. Following : is listedwhether a verb is main or light. Most examples in this paper were created by the authorand cross-verified by 3 native speakers from northern India.

2

1. INTRODUCTION

Single Predicate Construction

(1) ShyamShyam.M

Leela=seLeela.F=Inst

lad-aafight-Perfv.M.Sg

Shyam fought with Leela.

Complex Predicate Construction

(2) ShyamShyam.M

Leela=seLeela.F=Inst

ladfight:MV

baith-aasit-Perfv.M.Sg:LV

Shyam fought with Leela unwillingly.

In both examples, the subject Shyam has a fight with Leela and their

translation is near-equivalent. However, the light verb in the complex pred-

icate in (2) creates the interpretation that the fight between the two people

was not deliberate on the part of the subject and was an unwilling act.

Although the semantics of (2) suggests that light verbs function like an

adverbial modifier, the light verb assigns case to the subject in one type

of complex predicate construction, thereby indicating its status as a verb

(see Section 4 for details). A list of Hindi aspectual light verbs and their

meanings is listed in Section 2.

A typical analysis for complex predicates is to consider the light verb

a verb that subcategorizes for the main verb (or the VP that is headed by

the main verb) as shown, for instance, for auxiliaries in French (Abeille and

Godard, 2002) or Spanish (Abeille and Godard, 2007). (An early analysis

of this kind was provided by Grimshaw and Mester (1988) for Japanese

complex predicates with the verb suru (do).) In a head-final language such

as Hindi, a standard complex predicate involves a finite light verb followed a

3

1. INTRODUCTION

non-finite main verb.2 This was illustrated in (2), and similarly in (3) where

the finite light verb de (give) follows the non-finite main verb maar (hit).

But, surprisingly, the order of the main and light verbs can be reversed.

This is shown in (4); the light verb is now non-finite and precedes the main

verb, which is finite.

Standard Complex Predicate Construction

(3) Shyam=neShyam=Erg

mujheme

tamaachaaslap.M.Sg

maarhit:MV

di-yaagive-Perfv.M.Sg:LV

Shyam slapped me (hit me with a slap).

Reverse Complex Predicate Construction


mujheme

tamaachaaslap.M.Sg

degive:LV

maar-aahit-Perfv.M.Sg:MV

Shyam slapped me (hit me with a slap).

The inflection is carried by the light verb in the standard construction

and by the main verb in the reverse construction. The semantics of ex-

amples (3) and (4) differ only in that the latter carries an indication of

suddenness. The two constructions differ in more than just linear ordering.

The two constructions also differ in terms of which verb is the constructions

head. The prime evidence for this claim is that in the two constructions,

different verbs govern whether the subject is marked ergative or not. Case

assignment is governed by the second verb of the complex predicate, i.e.

2Finite verbs in Hindi are inflected for number and gender, with a -aa/-ii ending (if theverb root ends in a vowel, a glide precedes e.g. -yaa). The exception is the verb ho (be),which also inflects for number. Nonfinite verbs in Hindi can either be a base infinitive,which consists only of the stem, or the so called to-infinitive, consisting of the stem +suffix -naa/nii for masculine and feminine gender respectively.

4

1. INTRODUCTION

by the light verb in the standard construction and the main verb in the

reverse construction. In other words, case assignment is positional and is

not always determined by the light verb. I suggest that the same case as-

signment constraints that are operative for other verbal constructions can

model the case assignment in both standard and reverse constructions, if one

assumes that the constraints apply to the argument-structure of the head.

Hindi complex predicate constructions provide evidence for the claim that

case assignment constraints are lexical rather than phrasal. Working within

Head-driven Phrase Structure Grammar, I show that the light verb inherits

its argument-structure from the main verb in the standard but not in the

reverse construction.

5

2. LIGHT VERBS SEMANTICS

2 Light Verbs Semantics

This section describes the semantics of light verbs and the kind of the main

verbs that they select. Light verbs that function as aspectual markers and

express subtle semantic notions form aspectual complex predicates (Butt,

1994).Native speakers will insist that the action seems incomplete or unsit-

uated when only a [single] verb is used (Butt, 1994, p84, original emphasis).

As an aspectual complex predicate, the standard construction can express

either perfective or imperfective aspect, whereas the reverse construction

can only express perfective aspect. Other subtle semantic notions such as

suddenness, benefaction, regret, violence, are also expressed.3 A list of Hindi

aspectual light verbs and their semantics is shown in Table 2.

In addition to the semantics listed in Table 2, the light verb in the

reverse construction also adds the meaning of extreme suddenness or non-

volitionality of the action. The following examples contrast the standard

and reverse constructions with respect to suddenness.

(5) a. us=nePron.3.Sg=Erg

jhatke=seforceful=Inst

lagaamrein.F.Sg

khiichpull:MV

di-igive-Perfv.F.Sg:LV

He pulled the reins forcefully.

3Butt (1994) also shows that aspectual complex predicates contribute information interms of inception and completion; I have not examined this claim in this paper.

6


Light verbsa Semanticsb Reversec

Transitive verbs

baith (sit) regret N

dal (put) thoroughness, violence N

de (give) benefaction (self) Y

le (take) benfaction (others), contempt Y

maar (hit) suddenness, violence Y

nikaal (remove) contempt, violence N

Intransitive verbs

aa (come) completion Y

jaa (go) completion Y

pad (fall) suddenness N

nikal (leave) suddenness Y

uth (rise) regret, suddenness Y

aDetails on how the list was created can be found in Appendix A.bAbbi and Gopalakrishnan (1991); Hook (1974)cN indicates that there were no examples either in the literature or in the EMILLE

corpus, and the native speakers could not construct a grammatical sentence that involveda reverse construction. EMILLE corpus: www.ling.lancs.ac.uk/corplang/emille/ Availablefrom ELRA. The corpus totals approximately 12,390,000 words and consists of articlesfrom news web sites, 20th century Hindi literature, and other miscellaneous documents.

Table 1: Aspectual Light Verbs

b. us=nePron.3.Sg=Erg

aahiste=seslow=Inst

lagaamrein.F.Sg

khiichpull:MV

di-igive-Perfv.F.Sg:LV

He pulled the reins slowly.


jhatke=seforceful=Inst

lagaamrein.F.Sg

degive:LV

khiich-iipull-Perfv.F.Sg:MV

He pulled the reins forcefully.

7


b. *us=nePron.3.Sg=Erg

aahiste=seslow=Inst

lagaamrein.F.Sg

degive:LV

khiich-iipull-Perfv.F.Sg:MV

*He pulled the reins slowly.

While the standard construction is compatible with both forcefully (5a)

and slowly (5b), the reverse construction is compatible only with the adverb

forcefully (6a). It is not possible to use a reverse construction to say that the

reins were pulled slowly. Thus, the reverse construction has certain semantic

restrictions that are not present in the standard construction.

Similar restrictions are seen with respect to the types of verbs that can

appear in the reverse construction. As the table suggests, fewer light verbs

can appear in the reverse construction than in the standard construction.

And with respect to main verbs, Hook (1974) notes that typically verbs

of running, throwing, hitting and breaking are all able to form a reverse

complex predicate and verbs that do not fall into these categories do not

form a reverse construction. For instance, it is possible to say that the glass

was broken by using both standard (7a) and reverse (7b) complex predicate

constructions, but to say that the glass was made, we can only use a standard

(8a) but not a reverse construction (8b).


gilaasglass.M.Sg

toRbreak:MV


He broke the glass.


gilaasglass.M.Sg

degive:LV

tor-aabreak-Perfv.M.Sg:MV

He broke the glass.

8



gilaasglass.M.Sg

banaamake:MV


He made the glass.

b. *us=nePron.3.Sg=Erg

gilaasglass.M.Sg

degive:LV

banaa-yaamake-Perfv.M.Sg:MV

He made the glass.

Although a complete analysis of the semantics of the reverse construc-

tion is beyond the scope of this paper, the above data indicates that the

semantics of the light verb in the reverse construction differs from the stan-

dard construction and expresses something akin to enhanced suddenness.

The reverse construction is also more restricted with respect to the range of

main verbs and light verbs that can appear in it. I will henceforth focus on

the syntactic structure of the standard and reverse construction.

9

3. SYNTACTIC STRUCTURE

3 Syntactic Structure of the Complex Predicate

In this section, I examine the syntactic structure of the complex predicates

in the standard and reverse construction and show that the two verbs in

both constructions form a single constituent. They do differ, however, in

the placement of the focus particle, which can appear between the two verbs

in the standard construction but not in the reverse construction. The re-

verse construction is also more restricted in comparison to the standard

construction with respect to combination with auxiliaries. In spite of the

more restricted nature of the reverse construction, the two constructions

involve the same phrase-structure configuration.

3.1 Constituency Tests

In this section, I show that Hindi complex predicate constructions are mon-

oclausal and the two verbs in it form a single constituent, by using tests

such as movement, coordination, modification, and the insertion of clitics.

3.1.1 Movement

Although Hindi has a relatively free word order, the main verb and the light

verb in an aspectual complex predicate move together; this is demonstrated

for the reverse construction in (9).

10


(9) a. [Leela=ne][Leela.F=Erg]

[Shyam=ko][Shyam.M=Dat]

[citthii][letter.F.Sg]

[maar[hit:LV

likh-ii]write-Perfv.F.Sg:MV]

Leela wrote a letter to Shyam.

b. [Shyam=ko] [Leela=ne] [citthii] [maar likhii]

c. [Leela=ne] [maar likhii] [citthii] [Shyam=ko]

d. [maar likhii] [Leela=ne] [Shyam=ko] [citthii]

e. [maar likhii] [citthii] [Shyam=ko] [Leela=ne]

f. [citthii] [maar likhii] [Leela=ne] [Shyam=ko]

g. [citthii] [maar likhii] [Shyam=ko] [Leela=ne]

h. *[citthii] [likhii] [Shyam=ko] [Leela=ne] [maar]

i. *[citthii] [likhii] [Shyam=ko] [maar] [Leela=ne]

The scrambling possibilities in (9a)-(9g) demonstrate that the light verb

and the main verb can move as a unit. However, the light verb maar (hit)

that precedes the main verb likh (write) cannot be moved away, as shown in

(9h) and (9i). Similar evidence can be presented for the standard construc-

tion as well. This indicates that the main verb and the light verb in a Hindi

complex predicate construction move together as a unit.

3.1.2 Coordination

The main verb and the light verb in a standard or reverse complex predicate

construction cannot be a coordinated structure in either the standard or

reverse construction, as shown in (10a) and (10b) respectively.

11


(10) a. *Leela=neLeela.F=Erg

Shyam=koShyam.M=Dat

citthiiletter.F.Sg

likhwrite:MV

aurand

degive:MV

maar-iihit-Perfv.F.Sg:LV

*Leela wrote and gave a letter to Mohan.

b. *Leela=neLeela.F=Erg

Shyam=koShyam.M=Dat

citthiiletter.F.Sg

maarhit:LV

likh-iiwrite.Perfv-F.Sg:MV

aurand

di-igive.Perfv.F.Sg:MV

*Leela wrote and gave a letter to Mohan.

Again, this indicates that the main verb and the light verb form a unit.

3.1.3 Modification

Adverbial modifiers can only take scope over the elements to their left.

For instance, kal (yesterday/tomorrow) can appear in various positions to

the left of the reverse complex predicate, as indicated in (11a) and (11b).

However, it cannot appear between the main verb and the light verb (11c),

since it cannot modify only the light verb.

(11) a. Leela=neLeela.F=Erg

kal

yesterdaysaaraaall

dinday.M

gappochats.M.Pl

meinin

[maarhit:LV

bitaay-aa]spend-Perfv.M.Sg:MV

Leela spent all day yesterday chatting.

b. Leela=ne saaraa din gappo mein kal [maar bitaay-aa]

c. *Leela=ne saaraa din gappo mein [maar kal bitaay-aa]

The above examples show that an adverbial modifier cannot appear

within the reverse complex predicate construction itself because syntacti-

cally it cannot intrude between the two verbs; Butt (1994)[p. 99] illustrates

12


the same for the standard construction.This shows that there are restrictions

on the modification of the complex predicate construction.

3.1.4 Insertion of discourse clitics

The previous three tests (movement, coordination, and modification) sug-

gest a strong degree of cohesion or bondedness between the light verb and

the main verb. In spite of this fact, Butt (1994) argues that the (standard)

complex predicate construction cannot be analyzed as a single morpholog-

ical unit because discourse clitics such as hii (exclusive contrastive focus

only) and bhii (inclusive contrastive focus also) can be inserted between

the verbs in a standard complex predicate construction (pp. 91-93). I show

that while these discourse clitics can be used as evidence for the standard

construction not being a morphological unit, they cannot be used as a test

for the reverse construction on distributional grounds.

Hindi discourse markers are syntactic clitics and not morphological af-

fixes, and they can take scope only over the constituents to their left (Sharma,

1999). The arrow below represents the scope of the focus clitic.

(12) a. [S] [O bhii] [V1 (V2)]

b. [S] [O] [ [V1 bhii] (V2)]

In (12a), the clitic bhii can only take scope on the object but not the

verbs; in (12b), the scope can only be on the first verb. This is illustrated

in the single predicate construction below. The can appear anywhere after

the unmarked argument (13a) or to the immediate right of the verb that it

is modifying (13b).

13



citthiiletter.F.Sg

bhii

alsobhej-iisend-Perfv.F.Sg:MV

He sent a letter also (along with other things).


citthiiletter.F.Sg

bhej-iisend-Perfv.F.Sg:MV

bhii

alsoparbut

mujheI

parvaahcare

nahiiNEG

He sent a letter (in addition to doing other things) but I dontcare (about the letter).

Note that the semantics of (13a) and (13b) differ. A similar distinction

is seen with a standard complex predicate construction. In the following

examples, the focus is on the unmarked object citthii (letter) or the main

verb bhej (send) depending on the placement of bhii. Since bhii can appear

between the two verbs in (14b), it suggests that the two verbs in the standard

complex predicate construction are separate lexical items that combine in

the syntax. As focus particles do not appear at the end of the clause, when

the focus particle appears after the light verb in the standard construction,

example (14c) is rendered ungrammatical. This indicates that only when

bhii appears between the two verbs in the standard construction can it

modify the main verb.


citthiiletter.F.Sg

bhii

alsobhejsend:MV


He sent a letter also (along with other things).


citthiiletter.F.Sg

bhejsend:MV

bhii

also


14


He sent a letter (in addition to doing other things).

c. *us=ne citthii bhej di-yaa bhii

The restriction on focus particles in the reverse complex predicate is dif-

ferent. Here, bhii can only precede the complex predicate (15a) but cannot

be inserted between the two verbs (15b) or, as indicated previously, appear

at the end of the clause.


citthiiletter.F.Sg

bhii

alsodegive:LV

bhej-aasend-Perfv.M.Sg:MV

He also sent off a letter (in addition to doing other things).

b. *us=ne citthii de bhii bhej-aa

The semantics of (14a) and (15a) are essentially the same. As shown in

(14a), the clitic bhii modifies the main verb only when it appears between

the two verbs in the standard construction. Since the first predicate in

the reverse construction is a light verb, it cannot be modified. As shown

in (14c), the focus clitic can also not appear at the end of the complex

predicate construction. Therefore, the fact that a discourse clitic cannot

appear between the two verbs in the reverse construction is neither evidence

for nor against its syntactic structure being different from the standard

construction. In what follows, I show that the only syntactic difference

between the two constructions concerns their combination with auxiliaries.

15


3.2 Auxiliaries

Hindi auxiliaries place no restrictions on the kind of verb they combine

with. They concatenate with either the verb root or its different inflected

forms to yield distinctions of tense, aspect, mood, and voice. In spite of

the relatively free word order in Hindi, the language has a tendency to have

verbal constructions in sentence-final positions in the presence of auxiliaries.

Auxiliaries can only appear after both the main verb and the light verb.

(16)

(17) Main Verb (Light Verb) (Passive) (Progressive) (Be Auxiliary)

Butt and Lahiri (2002)[pp. 11-12] show that auxiliaries and light verbs

show distinct syntactic behaviors with regard to case marking, word or-

der, reduplication, and topicalization. Further, Hindi auxiliaries are aspect

markers that also distinguish themselves from light verbs by not making any

other semantic contributions (like volitionality, etc.) to the argument struc-

ture. This section looks at the combinatorial possibilities of the standard

and reverse complex predicate construction and shows the the reverse con-

struction is restricted with respect to the auxiliaries that can appear with

it.

3.2.1 Tense Auxiliary

Hindi auxiliaries can co-occur but the tense auxiliary ho (be) must appear

after all other verbs. The auxiliary ho also inflects for gender and number

in the past (th-aa and th-ii) and number and person in the present (hai)

16


and future (ho-g-aa and ho-g-ii).4 An auxiliary cannot appear within (18b)

or before (18c) the complex predicate. Thus, we have the order [MV LV

(Auxbe)] as shown in (18a).

(18) a. voPron.3.sg

aacome:MV

ga-yiigo-Perfv.F.Sg:LV

th-ii

be.Past-F.3.Sg

She came.

b. *voPron.3.sg

aacome:MV

th-ii

be.Past-F.3.Sg


*She came.

c. *voPron.3.sg

th-ii

be.Past-F.3.Sg

aacome:MV


*She came.

The above examples show that the tense auxiliary must always appear

after the MV-(LV) combination. I show below that this holds true for the

reverse complex predicate construction as well. As shown below, it is only

obligatory in the presence of the progressive auxiliary.

3.2.2 Progressive Auxiliary

The verb rah (stay), which can function as a main verb, also serves as the

progressive. An example of the progressive auxiliary is shown in (19)5.

(19) ShyamShyam.Sg

gharhouse.M.Sg

beechsell:MV

rah-aastay-Imperfv.M.Sg

haibe.Pres.3.Sg

Shyam is selling a house.

4Butt and Lahiri (2002) explain that the mixed paradigm of the tense auxiliary is dueto the fact that this derived from a former past participle and the ho form is based onan old Sanskrit verbal inflectional form.

5Modal verbs sak (able), caah (want) and paa (able) can also appear in the sameposition as the auxiliary rah (stay).

17


In the above example, the main verb beech (sell) combines with the pro-

gressive auxiliary. The tense auxiliary is obligatory with the progressive.

However, the combination of the progressive auxiliary with a standard com-

plex predicate construction has been claimed to be pragmatically odd in

most contexts (Butt, 1994, p. 97). While this is true, I show below that a

combination of the progressive with the passive to form a passive progressive

is frequent with the standard construction.

3.2.3 Passive Auxiliary

In Hindi, a sentence is passivized by adding the verb jaa, which is form-

identical to the light verb jaa (go). The passive auxiliary can follow a single

predicate as the following example indicates.

(20) Shyam=kaShyam.M=Gen

gharhouse.M.Sg

beech-aasell-M.Sg

ga-yaago-Perfv.M.Sg

(th-aa)(be.Past-Masc.3.Sg)

Shyams house has been sold.

In the above single predicate construction, the main verb beech (sell) is

inflected for person and gender -aa. In a complex predicate construction as

in (21), the main verb beech (sell) is uninflected and it is the light verb de

(give) and the passive auxiliary jaa that carry the inflection.


gharhouse.M.Sg

beechsell:MV


gay-aago-Perfv.M.Sg

(th-aa)(be.Past-Masc.3.Sg)

Shyams house has been sold off.

18


As mentioned above, the progressive is pragmatically odd with complex

predicate constructions but in contrast, the passive progressive occurs fre-

quently in the EMILLE corpus. In example (22), the inflected progressive

rah (stay) follows the auxiliary jaa to contribute a progressive reading.


gharhouse.M.Sg

beechsell:MV

di-yaagive-M.Sg:LV

jaa

go


haibe.Pres.3.Sg

Shyams house is being sold off.

Additionally, in a passive progressive construction, the passive auxiliary

must be non-finite. As shown below, it cannot be inflected.

(23) *Shyam=kaShyam.M=Gen

gharhouse.M.Sg

beechsell:MV

di-yaagive-M.Sg:LV

ga-yaa

go-M.Sg


haibe.Pres.3.Sg

*Shyams house is being sold off.

The fact that jaa (go) cannot be finite in (23) is a test for whether jaa

in this case is a light verb or an auxiliary since, as noted in Section 1, the

light verb in a standard construction must be finite. No other verb (except

perhaps aa (come)) can occur in the position of jaa as a non-finite verb

preceding the progressive rah (stay). This indicates that jaa in this position

can only be a passive and not a light verb. If the non-finite jaa in (22)

were to be replaced by a light verb such as de (give), the sentence would be

ungrammatical.

19


(24) *Shyam=kaShyam.M=gen

gharhouse.Sg

beechsell:MV

de

give


haibe.Pres.3.Sg

*Intended reading: Shyams house is being sold off.

While the passive auxiliary jaa can be non-finite following the main verb

in (22), the light verb de in (24) cannot be.

Auxiliaries also adhere to a specific order; they must always follow the

complex predicate construction. The passive auxiliary precedes the pro-

gressive auxiliary and the tense auxiliary is obligatory with the progressive.

The tense auxiliary must be last. Any attempt to reorder the auxiliaries

produces an ill-formed result; this is shown in (25).

(25) a. Shyam=kaShyam.M=Gen

gharhouse.M.Sg

beechsell:MV

di-yaagive-M.Sg:LV

jaa

go


haibe.Pres.3.Sg

Shyams house is being sold off.

b. *Shyam=ko aspataal jaa pahuunchaa diy-aa rah-aa hai

c. *Shyam=ko aspataal rah-aa pahuunchaa diy-aa jaa hai

d. *Shyam=ko aspataal hai pahuunchaa diy-aa jaa rah-aa

e. *Shyam=ko aspataal jaa rah-aa pahuunchaa diy-aa hai

f. *Shyam=ko aspataal jaa rah-aa hai pahuunchaa diy-aa

To summarize, the standard construction can appear with the full range

of Hindi auxiliaries and they must occur in a strict order. In contrast,

the reverse construction is more restricted. In a reverse complex predicate

20


construction, the tense auxiliary ho (be) can optionally follow the two verbs,

just like in the standard construction. I illustrate this below in (26).

(26) a. Shyam=neShyam.M=Erg

kitaabbook.M.Sg

jor=seforce=Inst

degive:LV

phekh-aathrow-Perfv.M.Sg:MV

th-aa

be.Past-M.3.Sg

Shyam threw the book forcefully.

b. *Shyam=ne kitaab jor=se th-aa de phekh-aa

c. *Shyam=ne kitaab jor=se de th-aa phekh-aa

While the tense auxiliary can follow the reverse construction (26a), it

can neither precede (26b) nor appear within the reverse construction. The

above examples show, once again, that the auxiliary must appear last. With

respect to other auxiliaries, neither the progressive nor the passive auxiliary

can appear in a reverse construction. This is shown in (27a) and (27b)

respectively.

(27) a. *ShyamShyam.M

kitaabbook.M.Sg

jor=seforce=Inst

degive:LV

phekhthrow:MV

rah-aa

stay-Imperfv.M.Sg

th-aabe.Past-M.3.Sg

*Shyam threw the book forcefully.

b. *Kitaabbook.M.Sg

jor=seforce=Inst

degive:LV

phekh-aathrow:MV

ga-yaa

go-M.Sg

th-aabe.Past-M.3.Sg

*The book was thrown forcefully.

In (27a), the progressive rah (stay) cannot appear with the reverse con-

struction, and in (27b), the passive jaa cannot appear with the reverse

21


construction. One possible explanation as to why the progressive cannot

appear with a reverse complex predicate is that the reverse construction can

only be perfective and therefore cannot occur with the progressive or other

verbs in its position. Additionally, a reverse complex construction cannot

include the passive auxiliary jaa (go) as shown in (27b). I presently have

no explanation for this restriction. Nonetheless, the above data shows that

unlike the standard construction, the reverse construction appears with a

restricted set of auxiliaries.

3.3 Summary

The constituency tests show that the two verbs in the standard and the

reverse construction form a single constitent. The two differ in that the

reverse construction does not allow the insertion of any element between

the two verbs. However, as discussed above, there is a possible semantic

explanation for this fact and therefore it is insufficient to demonstrate that

the two constructions are phrase-structurally different. Despite being more

restricted internally (which main verbs and light verbs can combine) and

externally (which auxiliaries can take as argument the reverse construction),

the standard and reverse constructions involve the same phrase-structure

configuration. The main verb and the light verb in the standard and reverse

constructions have the following structure:6

6Butt (1994) and Mohanan (1994) suggest that Hindi has no VP (but a V and a flatstructure i.e., NPs and V s are direct daughters of S; I do not dispute this analysis.

22


(28)

a. V

MV LV

b. V

LV MV

As shown above, the standard construction in (28a) and the reverse

construction in (28b) have the same constituent structure. In the following

section, I present facts on case assignment to demonstrate that the two

constructions differ functionally.

23

4. CASE ASSIGNMENT

4 Case Assignment

The previous section has shown that the standard and the reverse complex

predicate constructions involve the same phrase structure configuration. I

now present data from case assignment to show that the two constructions

differ functionally.7 Case assignment on the subject is governed by the

light verb in the standard construction and by the main verb in the reverse

construction. I will also show that the same case assignment constraints

that are operative for other verbal constructions can model case assignment

facts for the standard and the reverse construction as well; case is assigned

by the last verb of the construction irrespective of its lightness.

Within the paradigm of Ergative-Absolutive and Nominative-Accusative

systems, Hindi is considered a split-ergative system; the ergative case is as-

pectually driven (Dixon, 1994).8 In addition to aspect, current research has

also shown that grammatical relations in Hindi can be marked with a num-

ber of different cases depending on the desired semantic interpretation of

the clause (Butt and Lahiri, 2002; de Hoop and Narasimhan, 2008). An in-

ventory of Hindi case clitics is provided in Table 4 (Kachru, 1980; Mohanan,

1994). I use the term clitic instead of suffix because Hindi case markers ex-

hibit phrasal scope. When two nominals are co-ordinated, which is possible

7The two constructions also differ functionally with respect to agreement. In a singlepredicate construction in Hindi, the verb agrees with the highest unmarked argument.In a complex predicate construction, the second verb agrees with the highest unmarkedargument: the light verb in the standard construction and the main verb in the reverseconstruction cf. (Butt, 1994; Mohanan, 1994).

8As the reader will see below, within the perfective, Hindi is split-intransitive, notergative. However, since the number of intransitive verbs that can take ergative case isvery small, and therefore I stick with the current categorization of Hindi as a split-ergativelanguage.

24

4. CASE ASSIGNMENT

only when they are identically case marked, the scope of a case marking

may extend over both nominals as in [madraas aur haiderabad ]=se from

Madras and Hyderabad (Mohanan, 1994, p. 60, brackets mine).

Case Clitic Function

Unmarked/Nominativene Ergativeko Dativese Instrumentalkaa/kii/ke Genitiveme Locativepar Locative (in, at)tak Locative (toward)

Table 2: Hindi case clitics

In this paper, I focus on the alternation between the unmarked and the

ergative case on the subject.9 My analysis of the ergative case in Hindi

follows closely the analysis outlined in Butt and Lahiri (2002). Where my

own analysis diverges from theirs, I will call attention to it. In this section,

I will show that the assignment of these two cases to the subject can be

captured by the following rules:

Rule 1: By default, the subject is unmarked.

Rule 2: If the verb is transitive and perfective, then the subject is

assigned ergative (erg) case.

Rule 3: If the verb is intransitive and perfective, denotes a bod-

ily/sound emission event, and the action is purposeful on the actors

9The unmarked case in Hindi is phonologically null and has been labeled as Nomina-tive by some scholars (Kachru, 1980; Mohanan, 1994; Butt, 1994; Butt and Lahiri, 2002;de Hoop and Narasimhan, 2008). However, both proto-agent and proto-patient roles canbe unmarked for case and we therefore call it unmarked.

25

4. CASE ASSIGNMENT

part, then the subject is assigned erg case.

Subj Case Example

Perfective ERG 29Imperfective 30

Table 3: Hindi case assignment: Transitive in Finite Clauses

The object case assignment in Hindi is straightforward. By default, the

object is unmarked for case but when the direct object of a transitive or

ditransitive verb is definite, the dative case is assigned.10 The pattern for

subject case assignment in transitive verbs is illustrated in Table 3. As pre-

viously stated, the selection of ergative case in Hindi is aspectually-driven.

The ergative case is restricted to subjects of transitive verbs in the perfec-

tive aspect (marked by adding the -(y)aa/ii suffix to the stem), making it

finite.11 The ergative subject is shown in example (29). In contrast, when

the clause is imperfective i.e. either in the habitual (30a) or the future (30b),

the subject cannot be ergative and is unmarked.


ghar=kohouse=Dat

banaa-yaamake-Perfv.M.Sg

Shyam made the house.

(30) a. ShyamShyam

ghar=kohouse=Dat

banaa-taamake-Impfv

haibe

Shyam makes the house.

10Butt and Lahiri (2002) for why the clitic ko- is labeled dative instead of accusative.11In infinitive clauses, the subject is typically assigned dative case. Butt and Lahiri

(2002) present data from the Lahori dialect of Urdu where the subject of infinitive clausesalternates between the ergative and dative case. As Butt admits, such an alternation isgenerally not found in dialects of Urdu/Hindi. Further, native Hindi speakers that I polledassociate it this phenomenon with Punjabi rather than Hindi. I therefore do not accountfor this alternation in this paper and assume the subject of infinitive clauses to be markedonly with the dative case.

26

4. CASE ASSIGNMENT

b. ShyamShyam

ghar=kohouse=Dat

banaa-yeg-aamake-Fut-M.Sg

Shyam will make the house.

As the examples in (30) show, transitive verbs do not select for an erga-

tive subject in the imperfective. This pattern for transitives is mirrored by

ditransitives as well. As the following examples show, the subject is assigned

ergative case in the perfective (31a) and is unmarked in the imperfective

(31b).

(31) a. Shyam=neShyam=Erg

Lee=koLee=Dat

kitaabbook

di-yaagive-Perfv.M.Sg

haibe

Shyam gave Lee a book.

b. ShyamShyam

Lee=koLee=Dat

kitaabbook

de-taagive-Impfv

haibe

Shyam gives Lee a book.

As Kachru (1980)[p. 52] points out, transitive verbs in Hindi can be ei-

ther volitional or non-volitional. Non-volitional verbs such as bhool (forget),

kho (lose), or jaan (know) also select for ergative subjects. Subject case

assignment in transitive verbs is thus purely aspect-driven. The pattern

in intransitive verbs, however, is motivated not only by aspect but also by

the semantics of the verb itself. The subject of most intransitive verbs are

unmarked for case, as shown by the verb fisal (slip).

(32) a. ShyamShyam.M

fisl-aaslip-M.Sg

Shyam slipped.

b. *Shyam=neShyam.M=Erg

fisl-aaslip-M.Sg

*Shyam slipped.

27

4. CASE ASSIGNMENT

Even verbs like bhaag (run), uchal (jump) or baith (sit), where the agent

must employ volition, take only an unmarked and not an ergative subject

as suggested below.


bhaag-aarun-Perfv.M.Sg

Shyam ran.

b. *Shyam=neShyam=Erg

bhaag-aarun-Perfv.M.Sg

*Shyam ran.

In addition to the intransitive verbs that take only an unmarked subject,

there are also some (called intransitive unergative verbs by Butt and Lahiri

(2002)) that can select either an ergative or an unmarked subject. Such

verbs are primarily bodily/sound emission verbs such as khaas (cough),

chiikh (sneeze), bhauk (bark), ciik (scream), cillaa (yell), muut (urinate),

and thuuk (spit) (de Hoop and Narasimhan, 2008). One such alternation is

shown below.

(34) a. ShyamShyam

khaans-aacough-Perfv.M.Sg

Shyam coughed (without meaning to).

b. Shyam=neShyam=Erg


Shyam coughed (purposefully).

The default subject for the verb khaans (cough) is (34a), which simply

states that the agent coughed. However, if the action was intentional, as

if to obtain attention, the subject is marked as ergative. This intention

28

4. CASE ASSIGNMENT

cannot be called volitionality, since, as shown in (33), verbs like bhaag (run)

that require a volitional agent selects only for an unmarked subject. In

fact, the ergative-unmarked alternation has previously been explained as

the conscious control that an agent is interpreted to have over the action

(Mohanan, 1994, p. 71). This notion has also been termed conscious choice

(Butt, 1994, p. 102). Both these analyses suggest that if the action was

performed under the control of the agent, the subject is assigned ergative

case and otherwise, it is unmarked. I argue that the subject case alternation

is based on an even narrower constraint and has to do with a kind of purpose

with which the agent performs the act. This can be illustrated with the

examples in (35).


Ram=parRam=Loc

thuukspit:MV


Shyam spit on Ram.

b. Shyam=neShyam.M

Ram=parRam=Loc

thuukspit


Shyam spit on Ram (on purpose e.g. with the intention ofinsulting him).

In both (35a) and (35b), Shyam spits on Ram. However, the subject is

assigned ergative case in (35b) to indicate that the act of spitting was done

purposefully by Shyam in order to insult Ram. Moreover, the contexts in

which the bodily/sound-emission verbs select for an ergative subject tend

to be negative as shown by (36).

(36) a. voPron.3.Sg

chillaa-yaayell-Perfv.M.Sg

aagfire

He yelled fire!

29

4. CASE ASSIGNMENT

b. #us=nePron.3.Sg=Erg


aagfire

#He yelled fire!

c. us=nePron.3.Sg=Erg

jaanbooj=karpurpose=do


aagfire

He yelled on purpose fire!

In (36), if the actor yelled that there is a fire, then the subject by default

must be unmarked (36a). Without any context, (36b) is infelicitous. If

the intention was to perform the act purposefully, for instance, if the actor

intended to frighten people on purpose or pretended that there was a fire

when there wasnt one, an adverbial modifier to that effect could be used

and then the subject would be assigned ergative case (36c).

The use of an adverbial modifier has no effect on transitive constructions.

In a transitive perfective example as the following (taken from the EMILLE

corpus), the transitive verb maar (kill, in this context) assigns ergative case

to the subject even though the act was unintentional.

(37) SriSri

LankaLanka

sena=nearmy=Erg

bhool=semistake=Inst

machuar-oo=kofisherman-Pl=Dat

maar-aakill-Perfv.M.Sg

The Sri Lankan army killed the fishermen by mistake.

The above example demonstrates that while the ergative case expresses

the on purpose notion of the agent in the intransitive examples, the on

purpose notion is not entailed by the ergative case.12 A summary of case

12(de Hoop and Narasimhan, 2008, p. 69) make the same argument with respect toergativity and volitionality.

30

4. CASE ASSIGNMENT

alternation on the subject of bodily emission verb depending on whether

or not the act was committed on purpose is shown in Table 4. If the act

performed by the agent was on purpose, then the subject is assigned ergative

case and otherwise it is unmarked.

Subj Case On purpose Example

34a, 36aerg yes 34b, 36c

Table 4: Hindi case assignment: Either (Bodily emission verbs)

In what follows, I show that the same case assignment rules apply for

complex predicate constructions as well.

Case Assignment in Complex Predicates

Previous research has shown that in the standard complex predicate con-

struction, irrespective of the (in)transitivity of the main verb, the light verb

assigns case to the subject (Butt, 1994). In this section, I show that while

this pattern holds true for the standard construction, it does not apply to

the reverse construction. In the reverse construction, it is the main verb

that assigns case to the subject.

In the standard construction, if the light verb is transitive, then the

subject must be ergative, if the light verb is intransitive, then the subject

must be unmarked. The transitive verb gaa (sing) can only take an erga-

tive subject in the perfective, similar to single predicate construction shown

previously in (29). However, in a complex predicate construction with main

verb gaa (sing), if the light verb is transitive e.g., daal (put), then the sub-

ject is assigned ergative case (38a) and if the light verb is intransitive e.g.,

31

4. CASE ASSIGNMENT

pad (fall), then the subject is unmarked (38b).


gaanaasong

gaasing:MV

daal-aaput-Perfv.M.Sg:LV

Shyam sang a song (because he had to).

b. ShyamShyam.M

gaanaasong

gaasing:MV

pad-aafall-Perfv.M.Sg:LV

Shyam sang a song (without wanting to).

Any intransitive main verb can co-occur with an intransitive light verb,

but what has not been noticed before is that within intransitive main verbs,

only bodily/sound emission verbs can co-occur with a transitive light verb.

The intransitive verb daud (run) can co-occur with an intransitive light verb

such as pad (fall) and take an unmarked subject, as in (39a). However, it

cannot co-occur with a transitive light verb such as daal (put).


daudrun:MV


Shyam ran.

b. *Shyam=neShyam.M=Erg

daudrun:MV


*Shyam ran.

In fact, only bodily/sound emission verbs can select for transitive light

verbs as the following examples illustrate. The subject of bodily/sound

emission verbs can be either unmarked (40a) or ergative (40b) depending

on the (in)transitivity of the light verb.


ciikhscream:MV


Shyam screamed suddenly.

32

4. CASE ASSIGNMENT

b. Shyam=neShyam.M=Erg

ciikhscream:MV


Shyam screamed violently.

The subject is unmarked if the light verb is intransitive (40a) and is

assigned ergative case if the light verb is transitive (40b), as posited by Rule

2 (if the verb is transitive and perfective, then the subject is assigned erg

case). Note that bodily/sound emission verbs do not appear in the reverse

construction.

As can be seen from examples (38-40), in the standard construction, case

assignment on the subject depends on the (in)transitivity of the light verb.

That is, case is assigned by Rule 2. As the following examples indicate,

irrespective of the transitivity of the main verb and whether or not the

action was performed on purpose, the subject is always assigned ergative

case if the light verb is transitive and perfective.


galti=semistake=Inst

chiinksneeze


He sneezed by mistake.


galti=semistake=Inst

paaniwater

piidrink

li-yaagive-Perfv.M.Sg

He drank water by mistake.

In (41a), the main verb chiink (sneeze) is intransitive and the light

verb de (give) is transitive. Even though the action was performed non-

volitionally, the subject bears the ergative case. Similarly, with a transitive

main verb and a transitive light verb in (41b), the subject remains erga-

tive. Thus, similar to the single predicate construction in (37), the subject

33

4. CASE ASSIGNMENT

in the complex predicate construction is assigned ergative case if the last

verb is transitive, irrespective of whether or not the action was performed

on purpose. The above examples solidify the argument that when a subject

is marked with ergative case, it does not entail volitionality.

I have thus shown that case assignment in the standard complex pred-

icate construction is always governed by Rule 1 and Rule 2 and it is the

transitivity/intransitivity of the light verb rather than its volitionality that

determines the subjects case. In what follows, I show that case assignment

in the reverse construction is governed by the main verb.

Recall that in the reverse construction, the light verb precedes the main

verb. Even though the light verb de (give) is transitive, the subject in the

following example is unmarked for case. This is because the main verb bhaag

(run) is intransitive.

(42) ShyamShyam.M

degive:LV

bhaag-aarun-Perfv.M.Sg:MV

Shyam ran (rapidly).

On the other hand, when the transitive light verb de (give) forms a

reverse complex predicate with a transitive main verb like maar (hit), the

subject is ergative.

(43) Shyam=neShyam.M=Erg

Leela=parLeela=Loc

kiitaabbook

degive:LV

maar-iihit-Perfv.F.Sg:MV

Shyam threw the book on Leela (forcefully).

Similarly, even with an intransitive light verb jaa (go), the transitive

main verb beech (sell) selects for an ergative subject.

34

4. CASE ASSIGNMENT

First verb Second verb Subject case

(Case-assigning verb)

Standard

TransMV IntransLV pad fall gaa sing TransLV dal put Erg

IntransMV IntransLV pad fall chiik scream TransLV dal put Erg

Reverse

TransLV IntransMV bhaag run de give TransMV maar hit Erg

IntransLV IntransMV baith sit jaa go TransMV beech sell Erg

Table 5: Subject case assignment in the standard and reverse complex pred-icate constructions


apnaaself

makaanhouse

jaago:LV

beech-aasell-Perfv.M.Sg:MV

Shyam sold his house.

As expected, when both verbs in the reverse construction are intransi-

tive, the case on the subject is unmarked. This is shown in (45) with the

intransitive main verb bhaag (run) and the light verb nikal (leave).

(45) ShyamShyam.M.Nom

kamre=seroom=Inst

nikalleave:LV


He ran from the room.

A summary of the case assignment pattern in the two constructions is

shown in Table 4. While the light verb assigns case to the subject in the

standard construction, it is the main verb that assigns case to the subject

in the reverse construction. Case assignment in complex predicate construc-

tions is therefore positional i.e. assigned by the last verb of the complex

35

4. CASE ASSIGNMENT

predicate. Crucially, the same rules model the assignment of subject case in

single as well as (standard and reverse) complex predicate constructions. In

the next section, I show that to model that generalization, case assignment

in Hindi must be done at the lexical level rather than at the phrasal level.

36

5. AN HPSG ANALYSIS

5 An HPSG Analysis

The previous section has shown that the subjects case is assigned by the

last verb, which is the light verb in the standard construction and the main

verb in the reverse construction. In the standard construction (46a), the

light verb is the semantic head i.e., the aspectual functor, as well as the

syntactic head, since it assigns case to the subject. In contrast, in the

reverse construction (46b), while the light verb is the semantic head, the

main verb is the syntactic head of the construction.

(46)

a. V

MV LV

semantic head

syntactic head

b. V

LV

semantic head

MV

syntactic head

In this section, I model Hindi case assignment rules in Head-driven

Phrase Structure Grammar (HPSG) and suggest that case assignment in

Hindi must be done at the lexical level rather than at the phrasal level.

I also propose that the light verb inherits its argument-structure from the

main verb in the standard, but not in the reverse complex predicate con-

struction.

5.1 Case Assignment Rules

HPSG is a constraint-based formalism with one level of representation,

where the different aspects of words or phrases, called signs (phonology,

37

5. AN HPSG ANALYSIS

morphology, syntax, sematics, information structure) are described in par-

allel. The signs are assigned types organized in a hierarchy. Each type is

associated with certain constraints and inherits the constraints of its super-

type(s), with the possibility of multiple inheritance i.e., a type can inherit

from more than one supertype. Every feature or constraint specified as

appropriate for the supertypes is also appropriate for the subtypes.

The Hindi case values are organized as shown in Figure 1. Case in Hindi

is either marked or unmarked; if it is marked, it can be assigned one of the

various cases such as ergative (erg), dative (dat), etc.

case

unmarked marked

erg dat gen loc inst

Figure 1: Hindi Case Ontology

I model the following three rules/constraints for ergative and unmarked

case on the subject, repeated below from Section 4, within HPSG.

Rule 1: By default, the subject is unmarked.

Rule 2: If the verb is transitive and perfective, then the subject is

assigned ergative (erg) case.

Rule 3: If the verb is intransitive and perfective, denotes a bod-

ily/sound emission event, and the action is purposeful on the actors

part, then the subject is assigned erg case.13

13Recall from Section 4 that this rule only applies to the single predicate constructions.

38

5. AN HPSG ANALYSIS

The feature case is specified as having a default value of unmarked

(Rule 1) as shown in (47).

(47) Rule 1[case /unmarked

]

The symbol / indicates the default value of the feature; this value is

defeasible and can be overriden. In other words, if there is no case specifica-

tion, then the NP is unmarked. The default in (47) is overriden by the other

two case assignment constraints. Importantly, case assignment constraints

must apply to a projections lexical head in Hindi. This is because any infor-

mational difference between the standard and the reverse complex predicate

constructions disappears at the constructions mother node, as illustrated

by the following two examples where the standard versus the reverse con-

struction have the same main and light verb. The subject in (48a) bears

ergative case because the second verb (light) is transitive; in contrast, the

subject in (48b) is unmarked because the second verb (main) is intransitive.


bhaagIntransrun:MV

di-yaaTransgive-Perfv.M.Sg:LV

Shyam ran.

b. ShyamShyam.M

deTransgive:LV

bhaag-aaIntransrun-Perfv.M.Sg:MV


The simplified trees for the standard and the reverse construction in (48)

is shown in (49a) and (49b) respectively.

(49)

39

5. AN HPSG ANALYSIS

S

1NP

Shyam=ne

phrasearg-st

1

bhaag di-yaa

b. S

1NP

Shyam

phrasearg-st

1

de bhaag-aa

A comparison between the standard (49a) and the reverse (49b) construc-

tions respective trees suggests that any informational difference between the

two constructions, such as argument-structure (and similarly, for other syn-

tactic features which are not shown above for space reasons) will disappear

at the phrasal level. That is, the mother nodes syntactic properties are

identical, and it is therefore not clear how one would model differences in

the constraints on the subjects case in the two constructions. But the syn-

tactic information associated with the right daughter will not be the same

if it is either the light verb or the main verb. By having case assignment

constraints apply at the lexical level, we can capitalize on these differences.

Rule 2 is modeled in (50). The left hand side of the rule preceding

the arrow lists the constraints that must be fulfilled in order for the right

hand side to be true of a structure. Note that these feature structures are

simplified and only show the attributes relevant to the present analysis.

(50) Rule 2tv-lxm

head

[asp

[perf

]]

arg-st

np

[case erg

], ...

The rule in (50) reads: If a verb is of type tv-lxm i.e., transitive, and its

aspectual value (asp) is perfective (perf), then the subject (the first NP

40

5. AN HPSG ANALYSIS

in the argument structure) is assigned ergative case. The aspectual value

of the verb is treated as a head feature since it affects verbal morphology.

The value of arg-st is a list of synsems, which are the syntactic arguments.

The following set of examples illustrate how the above rule is applied.

(51) a. Shyam=neShyam=Erg

ghar=kohouse=DAT

banaa-yaamake-Perfv.M.Sg

Shyam made the house.

b. ShyamShyam

ghar=kohouse=DAT

banaa-taamake-Impfv

haibe

Shyam makes the house.

c. *Shyam=neShyam

ghar=kohouse=Dat

banaa-taamake-Impfv

haibe

*Shyam makes the house.

The examples in (51) are single predicate constructions repeated from

(29-30). In (51a), the transitive verb banaa (make) is perfective and the

subject Shyam is assigned ergative case. When the transitive verb is imper-

fective, the subject is unmarked (51b) and cannot be assigned ergative case

(51c). Similarly, in complex predicate constructions, the (in)transitivity of

the second verb determines case assignment to the subject.


ciikhIntransscream:MV

daal-aaTransput-Perfv.M.Sg:LV


In (52), the intransitive main verb ciikh (scream) is non-finite and it

is the light verb daal (put) that is transitive and perfective. Therefore,

as predicted by (50), the subject is assigned ergative case. This example

41

5. AN HPSG ANALYSIS

also clarifies the motivation for selecting tv-lxm instead of simply using the

argument structure list to determine if the verb is transitive (2 arguments in

arg-st) or intransitive (only one argument in arg-st). Although we have a

transitive light verb (52), the arg-st list has only one NP because the main

verb is intransitive. Therefore, using the arg-st list on the left hand side

will not help determine the (in)transitivity of the light verb in the standard

complex predicate construction. I surmise that the types tv-lxm and iv-lxm

of light verbs are a synchronically arbitrary leftover of their main verb uses.

While the assignment of ergative case on transitive perfective verbs is

straightforward, recall that the assignment of ergative case to the subject

of an intransitive verb (Rule 3) is more constrained. According to Rule 3,

if the verb is intransitive and perfective, denotes a bodily/sound event, and

the action is purposeful on the actors part, then the subject is assigned

erg case. The relevant lexical constraint is shown in (53) using Minimal

Recursion Semantics (MRS) following Copestake et al. (2005).

The sem value encodes the central predication of a phrase as its key

and a list of relevant semantic relations rels. The general MRS approach

is neutral about what the inventory of relation features consists of, being

equally compatible with the use of generalized semantic (thematic) roles

such as actor and undergoer (e.g., Davis (2001)) or a semantically-bleached

nomenclature, such as arg1, arg2. In this paper, I use the latter.

42

5. AN HPSG ANALYSIS

(53) Rule 3

iv-lxm

head

[asp

[perf

]]

sem

key 3

bodily/sound-emission-rel

event 2

arg 1

rels 3

on-purpose-rel

event y

arg1 1

arg2 2

[arg-st

np 1

[case erg

]]

The intransitivity of the verb is shown by iv-lxm and its aspectual value

(asp) is indicated as perfective (perf). Tag identity of tags is used to

indicate that information is shared between parts of the structure; for ex-

ample, that the agent argument for both the bodily/sound-event-rel and

on-purpose-rel is the same. Semantically, (53) introduces a key relation of

type bodily/sound-emission-rel (supplying an event variable 2 ) with only

one additioanl argument. This argument is identified with the index of the

subject NP 1 . The semantic contribution of purposefulness is integrated in

the rels list of the semantics sem as an additional on-purpose-rel where

the two arguments are the subject 1 and the event 2 .14 The application of

(53) can be shown by comparing the following examples.

14Although technically what is on purpose here is the action and not the event, I do notdiscuss this issue here.

43

5. AN HPSG ANALYSIS

(54) a. ShyamShyam


Shyam coughed.

b. Shyam=neShyam=Erg


Shyam coughed purposefully.

In both examples, we have an intransitive bodily emission verb. However,

when there is no purposefulness to the action, then by default (Rule 1),

the subject is unmarked for case (54a). When the action is committed on

purpose, as in (54b), the subject is assigned erg case by Rule 3.

I have thus implemented the relevant lexical case assignment constraints

in HPSG. As discussed previously, the functional difference between the

standard and the reverse construction impacts which verb is the head. In the

following section, I propose an analysis for both constructions that involves

argument composition in the standard but not the reverse construction.

5.2 Clause Union Analysis

The two verbs in the standard and the reverse complex predicate construc-

tions form a clause-union (Aissen and Perlmutter, 1976, 1983), i.e., the two

verbs do not function as heads of independent clauses but form a complex

predicate of a single clause. Within HPSG, such constructions have been

analyzed as involving an operation of argument composition wherein the

light verb is considered an operator that subcategorizes for the main verb

and includes in its argument structure what its complement verb subcatego-

rizes for. Such analysis has been undertaken for various Romance complex

44

5. AN HPSG ANALYSIS

predicates (Abeille and Godard, 2002, 2007; Monachesi, 1993). An argu-

ment composition analysis is appropriate for the standard construction; as

both the syntactic and the semantic head of the standard construction, the

light verb in the standard construction subcategorizes for the main verb and

inherits its argument structure. This is shown in (55) with an abbreviated

phrase structure tree.

(55) Standard Construction (Argument composition)[head 1

]

MVcomp

2

[arg-st 3

...

] LVheadhead 1

arg-st

2

3

In (55), the light verb is the head of the construction as indicated by the

matched tag 1 at the lexical and phrasal level.15 The argument structure

of the light verb includes the main verb ( 2 ), indicating that the light verb

subcategorizes for it. The entire argument structure of the main verb i.e.,

the subject and its complements ( 3 ) is inherited by the light verb. This

is illustrated using (56), (repeated from (52)) whose simplified structure is

shown in (57).


ciikhscream:MV



15The labels MV (main verb) and LV (light verb) are purely mnemonic and providedto ease the readers understanding.

45

5. AN HPSG ANALYSIS

(57) phrase

head 1

spr comps

3

phrase

spr comps

head[case erg

]

Shyam=ne

phrase

head 1

spr3

comps

2

word

spr3

comps

arg-st 43

ciikh

word

head 1[verb

]spr comps

arg-st2

4

daal-aa

The tree representation in (57) only outlines the syntactic and the ar-

gument structure component of example (56).16 As shown in (50), ergative

case assignment on the subject in (57) is constrained by Rule 2. The sub-

ject ( 3 ) of the intransitive main verb ciikh (scream) ( 2 ) is placed on the

specifier list as well as the argument structure list of the main verb. The

light verb daal (put) subcategorizes for the main verb as indicated by 2 in

its arg-st. The argument structure ( 4 ) of the main verb, consisting in this

case of only the subject, is also inherited by the light verb. Therefore, the

arg-st list of the light verb daal (fall) is a concatenation of 4 (arg-st of

the main verb) and 2 (main verb). Since the verb daal is transitive, the

lexical constraint in (50) applies.

16The phrasal level of the concatenation of the two verbs is not saturated (i.e., the speclist is non-empty). As indicated previously, NPs are considered direct daughters of the Sin Hindi.

46

5. AN HPSG ANALYSIS

A typical argument composition analysis in line with complex predicates

in other languages such as Romance can account for the standard complex

predicate construction in Hindi. In the standard construction, the aspectual

operator (light verb) can combine with its argument (main verb) through an

a head-complement structure in HPSG. However, in the reverse construc-

tion, the main verb (which assigns case) is the syntactic head and the light

verb is the semantic head; the reverse construction thus shows a dissociation

between lightness and headedness. Additionally, the standard construction

shows that the argument selection in Hindi takes place from right to left

i.e., the light verb is expected to follow the main verb, but the reverse con-

struction violates this expectation. We therefore need another mechanism

to capture both these facts.

I analyze the light verb in the reverse construction as modifiers that take

what they modify as arguments. Modifiers in Hindi typically precede the

expressions that they modify. For instance, adjectives precede nouns and

adverbs precede verbs (Kachru, 1980). The modifier status of the light verb

in the reverse construction can be modeled using the mod feature in HPSG,

as outlined in (58).

(58) Reverse Construction (No argument composition)[head 1

]

LVmodhead

[mod 2

[asp perf

]]MVhead

2

[head 1

]

47

5. AN HPSG ANALYSIS

Crucially, the non-null value of the mod feature indicates that the light

verb cannot be the head of the construction. The light verb in the reverse

construction modifies the main verb but does not inherit its argument struc-

ture. Thus, unlike the standard construction, there is no argument compo-

sition in the reverse construction. The light verb selects for the main verb

2 which is always marked perf (perfective) for its aspectual (asp) value.

This is needed to account for the fact that complex predicate constructions

in the reverse construction are always perfective. The reverse construction

is illustrated using the following example, repeated from (42).

(59) ShyamShyam.M

degive:LV



(60) phrase

head 1

spr comps

3

phrase

spr comps

head[case unmarked

]

Shyam

phrase

head 1

spr3

comps

word

head

[mod 2

[asp perf

]]spr comps arg-st

de

2

word

head 1

spr3

comps

arg-st3

bhaag-aa

48

5. AN HPSG ANALYSIS

In (60), the light verb de (give) modifies the main verb and the head of

the phrase is the main verb bhaag (run). Notice that the subject Shyam ( 3 )

appears only on the specifier and argument-structure list of the main verb.

There is thus no argument composition in the reverse construction.

The Hindi complex predicate construction provides evidence for the

claim that syntactic and semantic structures are not mapped uniformly onto

each other. As we have seen, aspect operators in Hindi can combine with

their argument via either a head(operator)-complement(argument) struc-

ture i.e., through argument composition, or a modifier(operator)-modified

(argument) structure. A similar analysis has been presented for Thai by

Koenig and Muansuwan (2005), illustrating the structural and functional

differences between the various aspect markers in Thai. A framework like

HPSG that treats the various aspects of linguistic signs as independent levels

of representation can easily model such constructions.

49

6. CONCLUSION

6 Conclusion

This paper has looked at the standard and the reverse complex predicate

constructions in Hindi. The two verbs in both constructions form a single

unit and have the same phrase structure configuration. However, they differ

with respect to linear ordering. The standard construction shows that the

argument selection in Hindi takes place from right to left i.e., the light verb

is expected to follow the main verb, but the reverse construction violates

this expectation. The two constructions also differ functionally; the last

verb in the complex predicate construction, whether it is light or reverse,

assigns case to the subject.

The two constructions differ also in terms of which verb is the head: it is

the light verb in the standard and the main verb in the reverse. Furthermore,

the Hindi reverse construction shows a dissociation between semantic and

syntactic headedness: while the main verb assigns case, the light verb is the

aspectual functor. This distinction suggests that the mapping between

aspectual semantics and syntactic structure need not be uniform within

a language. A framework such as HPSG which uses independent levels of

representation, can easily account for such data. The Hindi data presented in

this paper also supports a more general claim, which is that case assignment

constraints, rather than being phrasal, are lexical.

Finally, with respect to restrictions on main verb-light verb selection,

this paper has noted that transitive light verbs can select only for those

intransitive main verbs whose semantics represent bodily/sound emission.

Further research is needed to determine the range of main verbs and light

50

6. CONCLUSION

verbs that can appear in the reverse construction. The particular semantics

of the reverse construction also remains to be investigated.

51

7. APPENDIX: LIST OF LIGHT VERBS

7 Appendix: List of Light Verbs

This section discusses how the light verbs discussed in this paper were se-

lected. A survey of previous works conducted by Hook (1974) indicates that

the list of verbs that were considered light ranges between eight to sixty.

Of course, both the terminology and the definition for what was considered

a light verb varies. All of these verbs are listed in Table 6.17 On closer

inspection of the cited examples, many of the verbs that were only cited by

one or two authors were actually examples of the reverse complex predicate

construction. For instance, as we can tell from the semantics of the sentence

(61a) below, the main verb is nikal (leave) and not aa (come) since the sen-

tence describes an event of leaving and not of arriving. Another example is

(61b) where the main verb is bhaag (run) and the light verb is nikaal (leave).

(61) a. us=kipron.3.Sg=Gen

aankh-ooeyes-F.Pl

meinin

aasuutears.F.Sg

aacome:LV

nikleeleave-F.Pl:MV

Tears left her eyes.

b. Meriimine

changul=secontrol=Inst

kiisiisome

tarahmanner

nikalleave:LV

bhaag-aarun-M.Sg:MV

(He) somehow ran away from my control.

The list in Table 6 was narrowed down in Hook (1977) to a more conser-

vative list of twenty-two light verbs, listed in Table 7. For the purposes of

this paper, I selected only monomorphic verbs so as to limit issues related to

compounding. This eliminated khara ho (stand up), chor de (leave), rakh de

17Hook used Devanagari alphabetical order in his papers and I have retained that orderfor the first two tables.

52


(keep), le aa (bring) and le jaa (take away). I also eliminated the modal verb

cuk (already) and the verbs chal (walk) and dhar (hold). For the remaining

verbs, I downloaded a random set of hundred sentences from the EMILLE

corpus. If they occurred in at least one sentence as either noun-light verb

or main verb-light verb, they were included in this study. Of the list pre-

sented in the table below, the only verb that was not found to be a light

verb in the random set of sentences extracted from the EMILLE corpus was

mar (die).

aa (come), 21a dat (be firm), 1 baahar kar (turn out), 1uth (rise), 23 daal (put), 23 baith (sit), 23uR (fly), 1 duub (sink), 1 bhaag (run), 1khaRa ho (stand up), 2 de (give), 23 bhir (attack), 1khap (be expended), 1 dikhaa (show), 1 bhej (send), 1khaa (eat), 2 dekh (see), 1 mang (order), 1gir (fall), 1 daur (run), 2 mar (hit), 4giraa (drop), 1 dhamak (scare), 4 maar (die), 8guzar (pass by), 1 dhar (hold), 2 mit (be obliterated), 1gher (surround), 1 dhaa (run), 2 mil (find), 2ghus (enter), 1 nikal (leave), 9 rakh (keep), 16cal (walk), 5 nikaal (remove), 2 rakh de (keep give i.e, place), 1cuk (be used up), 1 pakar (catch), 1 rah (stay) , 13chor (leave), 9 par (fall), 23 lag (touch), 1chor de (leave give), 1 pahunch (reach), 5 lipat (cling), 1jaa (go), 23 paa (find), 9 le (take), 23jhapat (grab), 1 phas (stuck), 1 le aa (take come i.e., bring), 1tapak (drop), 1 phek (throw), 1 le jaa (take go i.e., take away), 1tuut (break), 1 ban (become), 1 sak (able), 4thahr (stop), 1 bas (inhabit), 1 sunaa (listen), 1

Table 6: All light verbs in the literature, compiled by Hook (1974)

aAll numbers indicate the total number of works prior to Hook (1974) that consideredthe relevant verb light.

53


aa (come) jaa (go) baith (sit)uth (rise) dal (put) mar (die)khara ho (stand be i.e., stand) de (give) maar (hit)cal (walk) dhar (hold) rakh de (keep give i.e., put)cuk (able) nikal (leave) rah (stay)chor (leave) nikaal (remove) le (take)chor de (leave give i.e., leave) par (fall) le aa (take come i.e., bring)

le jaa (take go i.e., take away)

Table 7: H indi light verbs from Hook (1977)

54

Bibliography

Abbi, A. and D. Gopalakrishnan (1991). Semantics of Explicator CompoundVerbs in South Asian Languages. Language Sciences 13 (2), 161180.

Abeille, A. and D. Godard (2002). The Syntactic Structure of French Aux-iliaries. Language 78 (3), 404452.

Abeille, A. and D. Godard (2007). Complex Predicates in the RomanceLanguages. In Fundamental Issues in Romance Languages, pp. 107171.CSLI Publications.

Aissen, J. and D. Perlmutter (1976). Clause Reduction in Spanish. In Papersfrom the Second Annual Meeting of the Berkeley Linguistic Society, pp.130.

Aissen, J. and D. Perlmutter (1983). Clause Reduction in Spanish. InD. Perlmutter (Ed.), Studies in Relational Grammar. Chicago UniversityPress.

Butt, M. (2005). Light Verb Jungle. In Harvard Working Papers in Lin-guistics, Volume 19, pp. 149.

Butt, M. and A. Lahiri (2002). Historical Stability vs. Historical Change.Available at http://ling.uni-konstanz.de/pages/home/butt/stability.pdf.

Butt, M. J. (1994, June). The Structure of Complex Predicates in Urdu. Ph.D. thesis, Stanford University.

Copestake, A., D. Flickinger, C. Pollard, and I. Sag (2005). Minimal Re-cursion Semantics: An Introduction. Research on Language and Compu-tation 3, 281332.

Davis, A. (2001). Linking by Types in the Hierarchical Lexicon (First ed.).Stanford, CA: CSLI Publications.

55

BIBLIOGRAPHY BIBLIOGRAPHY

de Hoop, H. and B. Narasimhan (2008). Ergative case-marking in Hindi.In H. de Hoop and P. de Swart (Eds.), Differential subject marking, pp.6378. Springer.

Dixon, R. (1994). Ergativity. Cambridge, UK: Cambridge University Press.

Emonds, J. (1978). The Verbal Complex V-V in French. Linguistic In-quiry 9, 151175.

Grimshaw, J. and A. Mester (1988). Light verbs and Theta marking. Lin-guistic Inquiry 19, 205232.

Hook, P. (1974). Compound Verbs in Hindi. Ph. D. thesis, University ofMichigan.

Hook, P. (1977). The Hindi Compound Verb: What it is and What it does.In Readings in Hindi-Urdu Linguistics, pp. 129154. National PublishingHouse, New Delhi.

Hook, P. (2001). Where Do Compound Verbs Come From? (And WhereAre They Going?). In Yearbook of South Asian Languages and Linguistics,pp. 101130. Sage Publications, New Delhi.

Kachru, Y. (1980). Aspects of Hindi Grammar. Manohar Publishers andDistributors.

Koenig, J.-P. and N. Muansuwan (May 2005). The Syntax of Aspect inThai. Natural Language and Linguistic Theory 23, 335380(46).

Masica, C. P. (1991). The Indo-Aryan Languages (First ed.). Cambridge:Cambridge University Press.

Mohanan, T. (1994). Argument Structure in Hindi (First ed.). Stanford,CA: CSLI Publications.

Monachesi, P. (1993). Restructuring verbs in Italian HPSG Grammar. InPapers from the 29th Regional Meeting of the Chicago Linguistics Society,Volume 1, pp. 281295. Chicago Linguistics Society.

Rizzi, L. (1978). A Restructuring Rule in Italian Syntax. In Recent Trans-formational Studies in European Languages, pp. 115158. MIT Press.

Sharma, D. (1999). Nominal Clitics and Constructive Morphology in Hindi.In Proceedings of the LFG99 Conference. CSLI Publications: Stanford,CA.

56

Reverse Complex Predicate in Hindi

Documents