e diachrony of light and auxiliary verbs in Indo-Aryan*

Benjamin Slade[accepted for publication in Diachronica]

Abstract

e diachronic properties of light verbs have been the topic of recent debate, which have raised various questionsregarding the diachronic stability of light verbs as well as their historical relationship to auxiliaries. is papercontributes to this discussion through an examination of the historical development of light verbs and auxiliariesin Indo-Aryan. I investigate the origins of the modern Indo-Aryan compound verb construction, and comparethis construction with other light verb constructions in Indo-Aryan. Examination of the antecedents of the Indo-Aryan compound verb construction alongside other Indo-Aryan light verb constructions, combined with analysis oflexical and morphosyntactic differences between the compound verb systems of two Indo-Aryan languages (Hindiand Nepali), reveals that light verbs are not a stable or unchanging part of grammar. Finally, Indo-Aryan providesclear cases of light verbs reanalysed as tense/aspect auxiliaries.

Keywords: syntax, morphology, compound verbs, light verbs, auxiliary verbs, reanalysis, grammaticalisation, chainshis, Indo-Aryan, Hindi, Nepali, Sanskrit

Introduction: Indo-Aryan compound verbsCompound verb [CV] structures are a conspicuous aspect of the syntax of Hindi and other modern Indo-Aryan[IA] languages. Masica (: ), in his survey of Indo-Aryan languages, notes that “[i]t will be found that anydescriptive grammar of a [modern IA] language has a section (usually inadequate) devoted to the phenomenonmost commonly known as the compound verb.” e prototypical structure of a CV in the modern IA languagesconsists of an form of the verb (the main verb, oen referred to in the Indological tradition as a )collocated with another verb (a light verb, traditionally termed a ), the laer acting as a modifier of the mainverb, contributing Aktionsart (usually a perfective-completive sense), aitudinal information, and other semanticnuances. See the examples from Hindi in () and () below: the (a) examples are sentences using simplex verbalforms; the (b) and (c) examples show corresponding CV constructions.

*Special thanks to Hans Henrich Hock for extensive discussion of the issues tackled here, and for comments on various early dras of thispaper over a number of years. anks also to John Peterson, Jim Gair, Shakthi Poornima, Claire Bowern, and the participants of the th All IndiaConference of Linguists (Shillong, Meghalaya, India), the th South Asian Linguistic Analysis conference (Urbana, Illinois), the th South AsianLinguistic Analysis conference (Madison, Wisconsin), and the rd Linguistic Society of America conference (San Francisco, California) for theirfeedback on early versions of sections of this paper. anks also to my Nepali consultants: Sonia Cheri, Manish Shrestha, and Komal PrasadPhuyal; and my Hindi/Urdu consultants: Archna Bhatia, Vandana Puri, and Aamir Wali. e usual caveats apply. Translations into English aremine, unless otherwise indicated.

Abbreviations used are: , first person; , second person; , third person; , absolutive; , accusative; , agentive; , causative;, converb; CV, compound verb; , emphatic; , feminine; , genitive; , gerund; , honorific; IA, Indo-Aryan; , imperative;, imperfect; _, imperfect participle; , infinitive; , locative; , masculine; MH, middle-grade honorific; , neuter;, nominative; , oblique; , optative; , past; _, perfect participle; , plural; , present; , progressive; _,relative pronoun; , singular; _, ta-participle; , vocative.

CV constructions are found throughout the modern IA languages, as well as in Indian members of other language families, that is, Dravidian,Tibeto-Burman, and Munda, as well as in Central Asian languages (Masica : ch. ). is paper focusses, however, on the development of CVsand other light verb constructions within Indo-Aryan, though this should not be taken to imply that I dismiss the possibility that the IA CV systemreflects at least partial convergence with other language families. For general discussion of compound verbs in Indo-Aryan (and, more broadly,in South Asian languages), see Masica ; Kachru and Pandharipande ; Singh et al. ; Abbi and Gopalakrishnan ; and many of thepapers in Verma ; on Hindi CVs, see, amongst others, Hook ; Nespital ; on Nepali CVs, see Sharma ; Pokharel ; Peterson.

e absolutive is a fixed, indeclinable verbal form.Note additionally in () and () that the use of an intransitive vector blocks the appearance of the agentive marker on the subject; see further

Section ...

() a. us-nehe/she.-

khānāfood..

khāy-āeat._.

“He/she ate the food.”

b. vahhe/she..

khānāfood..

khāeat.

gay-āgo._.

“He ate up the food.” [perfective]

() a. us-nehe/she-

kitābbook..

paṛh-īread._.

“He/she read the book.”

b. us-nehe/she-

kitābbook..

paṛhread.

dīgive._.

“He/she read the book aloud (i.e. for someone else’s benefit).” [other-directed action]

c. us-nehe/she-

kitābbook..

paṛhread.

lītake._.

“He/she read the book silently (i.e. for his/her own benefit).” [self-directed action]() a. maiṁ-ne

I.-kyāwhat

kiy-ā?do._.

“What did I do?”b. maiṁ

I.kyāwhat

kardo.

baiṭh-ā?sit._.

“Oh what did I do?” [regret]

Indo-Aryan light verbs employed in compound verb constructions are usually form-identical with a full/main lexicalverb. For example, the light verb used with a general-purpose completive sense in (-b), gayā (< jānā “to go”), whenused as a main verb means simply “went”; likewise, dī as a main verb means “gave” and lī “took”. e examples in() and () show that, in addition to perfective semantics, light verbs may also contribute other information: in (-b)dī “gave” contributes a sense of ‘other-directedness’, in (-c) lī “took” contributes a sense of ‘self-directedness’, andin (-b), baiṭhā “sat” contributes a sense of ‘regret’.

CV constructions historically derive from collocations involving , where a converb is a non-finite verbalform which, in contrast to CVs, denotes an independent event or state from that expressed by the finite verb of theclause, but usually shares the subject/agent of the finite verb. Converbs, which persist into modern IA, typicallydenotes actions or states which are completed before the beginning of the action or state expressed by the finite verb,as in the Hindi example (); though sometimes they indicate simultaneous action, as in the Hindi example ().

() vahhe/she-

khānāfood..

khā-keeat-

gharhome

gayāgo._.

“(Aer) having eaten food, he went home.”() vah

he.shehaṁs-kelaugh-

bol-īsay._.

‘hāṁ’‘yes’

“Laughing(ly) she said ‘yes’.”

In some modern IA languages the absolutive form of the verb used in CV constructions is identical in form tothe converb. For example, though in Hindi converbs are usually followed by kar or ke, they may also appear witha zero-ending, thus being identical in form to an absolutive, resulting in potential ambiguity, as shown by examples(-a), (-b).

With khā “eat”, lenā “take” is perhaps the more commonly used light verb; e.g. us-ne khānā khā liyā “He ate up the food”.In general, in both early and modern Indo-Aryan, converbs are oen used where English would typically use a VP and VP construction, e.g.

() might be rendered as “He ate food and went home”.On the translation of haṁs-ke, cp. Section ..

() usne mujhe ek thamz-ap kharīd diyā (cp. Hook : ,h)a. us-ne

he/she-mujheme

ekone

thamz-apums Up

kharīdbuy.

diy-āgive._-.

“He/she bought a ums Up (soda) and gave it to me.”

b. us-nehe/she-

mujheme

ekone

thamz-apums Up

kharīdbuy.

diy-āgive._-.

“He/she bought me a ums Up (soda) (for my benefit).”

emorphological similarity or identity of IA converbs and absolutives is not accidental, as these both ultimatelyderive from the same morphological source: the Old Indo-Aryan converb. Since I will be considering early IAexamples which are at least potentially ambiguous between converb and CV readings, for Sanskrit and Pāli examplesI utilise the term , a term taken from the western Sanskritist tradition, simply to indicate the morphologicalform while remaining neutral on how it is to be interpreted.

In addition to CVs and converbs, modern IA languages employ other verb-verb collocations (discussed in Section), where the first member appears as a past or present participle, rather than an absolutive; these collocationstypically exhibit continuative semantics, rather than the perfective semantics usual of CV constructions.

IA, given the length of its linguistic record, present an ideal opportunity to evaluate diachronic properties of lightverbs. Bu and Lahiri () and Bu (), on the basis of IA evidence, argue that light verbs constitute a specialclass of elements diachronically, in that they are historically stable and “do not enter the grammaticalization cline,i.e., they are not main verbs which have been reanalyzed as light verbs and which are now prone to further analysis”(Bu : ). Bowern (), in a study of light verbs crosslinguistically—while observing that light verbs canundergo historical change—agrees with Bu’s suggestion that light verbs do not become auxiliaries. In support ofthe claim that light verbs are historical stable, Bu and Lahiri () and Bu () examine CV structures in Indo-Aryan, and suggest that CV constructions are found through the history of Indo-Aryan, as far back as Vedic Sanskrit(ca. B.C.). In this paper, I examine the evidence for both the claim of the antiquity of IA CV constructions, andthe related claim regarding the stability of light verbs.

e following section, Section , provides a brief discussion of light verbs and auxiliaries and potential differencesbetween them. In this section I establish some rough criteria for distinguishing between light verbs and auxiliaries,though I suggest that establishing a clear distinction between the two theoretical categories is difficult, particularlyin any crosslinguistically robust fashion.

In Section , I investigate the history of verb-verb collocations in IA and evaluate Bu & Lahiri’s () positionon the diachrony of (light) vector verbs in IA, concluding that the evolution of vector verbs does, in fact, representa development in the history of IA. is is a conclusion which accords with previous research; for example, in hisstudy cited from the beginning of this section, Masica (: ) goes on to remark that “[compound verbs are] oneof the true innovations of [modern IA], unknown to Sanskrit.”

An examination of lexical and morphosyntactic properties of CV constructions in two modern IA languages, towit, Hindi and Nepali, in Section further bolsters the position that light verbs are not exempt from diachronicchange, revealing numerous differences in the CV systems of Hindi and Nepali.

In her examination of light verbs crosslinguistically, Bowern (: ) remarks that she “[has] found no evid-ence for light verbs being reanalysed as auxiliaries”. In Section I provide evidence for such changes as takingplace in IA, involving especially the reanalysis of light verbs as tense/aspect auxiliaries. Additionally, this sectiondiscusses the fact that some of the developmental processes responsible for reanalysis of light verbs as auxiliaries donot involve evolution along grammaticalisation clines, but rather take place as part of larger structurally-motivatedchanges, exhibiting changes which involve morphosyntactic chain shis.

Finally, in Section , I provide a summary and conclusions, along with discussion of avenues for future research.

Light verbs and auxiliariesLanguages employ a number of verbal or verb-like functional elements, most prominent being , includingtense marking auxiliaries like English have and be, as well as ‘dummy’ auxiliaries like English periphrastic do.

Specifically from the form the Sanskrit converb appears in when formed from a prefixed verb, i.e. with the suffix -(t)ya/(t)yā (and not fromthe unprefixed form of the converb, which takes the suffix -tvā).

Another category of functional elements is what has been termed (Jespersen ; Caell ;Grimshaw and Mester ) constituting verbs which typically have (homophonous) full lexical verb counterparts,but which enter the derivation structurally deficient in some respect. is structurally deficiency may include re-duced or null semantic values, or the lack of θ-roles (i.e., failing to assign a semantic role to one or more of theirarguments).

Prototypical light verbs occur in noun-verb complex predicates, as in English do dishes, do windows etc. Since doin do dishes is essentially semanticallly inert, the question arises of how to distinguish this light verb use of do fromthe auxiliary verb do. e difference, simply put, is that the two uses of do have different syntactic properties, asshown by the fact that the periphrastic auxiliary do is still required in () despite the presence of the light verb do.

() a. I don’t do dishes.b. *I don’t dishes.c. *I not do dishes.

Modern Indo-Aryan languages form noun-verb and adjective-verb complexes in a similar way, using light verbswhose full verb meanings correspond to “be” or “do/make”, as shown in the Hindi examples in ()

() a. mujheme.

yādmemory...

hai.be..

“I remember.”b. maiṁ-ne

I-uskāhe/she.

intazārwaiting...

kiyāmake._..

“I waited for him.”

Verb-verb complexes such as Indo-Aryan compound verbs have also been analysed as involving light verbs, includingBu and Lahiri’s () analysis of Hindi and Bengali verb-verb constructions. In contrast to the light verbs involvedin noun-verb complexes, however, the light verbs of verb-verb complexes do make a semantic contribution to thepredicate, as shown above in Section .

Many historical treatments do not distinguish between light verbs and auxiliaries, as noted Bu and Lahiri (:), who argue that such a distinction in fact should be made. Indeed, just as light verb do and the periphrasticauxiliary do have different properties, so too Indo-Aryan light verbs employed in compound verb constructions arestructurally distinct from more prototypical auxiliaries like tense-marking honā “be”; they do not occur in the samesyntactic contexts nor do they contribute the same sort of semantics.

However, in practice it is difficult to maintain a binary distinction between light and auxiliary verbs: on the onehand, all of these functional elements share a number of properties, making it difficult in some cases to distinguishlight verbs from auxiliary verbs in a non-arbitrary manner; on the other, even where verbal functional elements showclear differences in their formal properties, these differences require a manifold rather than a binary division—e.g.the light verbs of Hindi CV constructions bear formal properties different from other light verb constructions foundin IA, see Sections and . below. In other words, some light verbs are more auxiliary-like than others—a situationwhich makes sense if one accepts that light verbs can in fact become auxiliaries.

Such difficulties are particularly obvious in the case of trying to formulate crosslinguistically robust criteriafor distinguishing between the two notional categories. As one anonymous reviewer points out, in languages likemodern English auxiliaries might reasonably be distinguished with respect to the special morphosyntactic propertiesthat they bear (e.g. with respect to negation, inversion, and ellipsis), while in languages like modern Greek elementsmeaning “have” or “be” form periphrastic verbal combinations and thus appear comparable to similar elements inEnglish, but unlike in English, these elements in Greek do not exhibit special morphosyntactic properties.

Given such difficulties, here I focus more on the evaluation of distinguishing criteria relevant to IA. Bu (: )suggests that one notable difference between light verbs and auxiliaries is that the former generally involve some sortof semantic contribution beyond that of purely functional tense/aspect information. us, while light verbs oendo signal information regarding telicity, causation, or temporal boundedness, they usually also involve semanticcontributions of other sorts (including forcefulness, surprise, regret, benefaction, volitionality etc.). Additionally,Bu and Lahiri () and Bu () suggest that light verbs can be distinguished from auxiliaries in that theformer are always form-identical to some (semantically contentful) full/main verb, and cannot undergo changes

ough Bu (: ) points to noun-light verb constructions like give an answer as being distinct from the simplex verb answer, suggestingthe former signals deliberateness of the action, citing Traugo ().

(phonological or otherwise) which do not simultaneously affect their main verb counterparts. A difficulty ariseswith respect to this criterion due to the fact that there do exist sporadic examples of light verbs without full verbcounterparts.

While recognising difficulties in distinguishing consistently (particularly in any crosslinguistically robust fash-ion) between light verbs and auxiliaries, the best distinguishing criterion, at least for IA languages, is whether theelement in question contributes non-functional semantic information or is a more purely functional element requiredby the grammar in order to express some sort of paradigmatic information (tense, aspect, number etc.). Section .provides further discussion of the light verb-auxiliary distinction.

In the next section I turn to an examination of light verbs in various stages of IA.

e origin of Indo-Aryan compound verb constructionsAs discussed in Section , the morphological form of the main verb (“pole”) in modern IA CV constructions derivesfrom the so-called Old Indo-Aryan [OIA] past gerund of prefixed verbs in -(t)ya/(t)yā (Chaerjee ; Hendriksen; Tikkanen ), an element which has the “virtual value of an indeclinable participle” (Whitney /:§); which in general functions as a converb, meaning something like “having X-ed” (though it does not alwayshave a past tense value; see further below). It is clear that the origin of modern IA CV structures ultimately lies ina reanalysis of structures involving a gerund combined with another verb [V2], so that the Sanskrit example in ()can in a certain sense be seen as the precursor of Hindi (-b), repeated below as ().

() annamfood.

sam-khād-yatogether-eat-

gataḥgo._..

astibe..

“He ate up the food and le” (Lit., “Having eaten up the food, (he) le.”)() vah

he/shekhānāfood..

khāeat.

gay-āgo._.

“He ate up the food.”

e development from the gerund+V2 construction of type exemplified by () to the CV construction of thetype exemplified by () took place via the reanalysis of gerund+V2 as a single predicate, in which the semanticcontribution of the gerund was taken to be primary, and the V2 as a modifying element (i.e. a vector or light verb).is only happened with a subset of verbs occupying the V2 position, specifically with verbs with broad semanticvalues, e.g. verbs meaning “go”, “give”, “take” etc. Further, converb constructions did not disappear from modern IA(see the converb examples from Hindi in Section above); rather a sort of morphosyntactic split took place, withsome gerund+V2 constructions being reinterpretable/reinterpreted as CVs.

. Verb-Verb collocations in Vedic & Classical Sanskrit

Bu and Lahiri () suggest that in fact the modern IA system of CVs does not represent an innovation but rathercontinues a system of light verb constructions found throughout all stages of IA, and argue that examples like (),taken from Rgvedic Sanskrit (the earliest aested form of IA), may represent CV structures comparable to thosefound in modern IA.

See Section ...In some languages, such as Nepali and Braj, the -ya survives as -i. In Hindi, the ending has been completely lost, so that Hindi absolutives

have the appearance of bare stems. In most cases, modern IA absolutives are form-identical with converbs. Some languages show secondarydifferentiation of the converb forms, e.g. Nepali converbs end in -ī, -īkana, and -era (the first of these is actually only orthographically distinct inNepali, since in spoken Nepali there is no long/short distinction for high vowels, and thus absolutive -i is indistinguishable from converb -ī ). InHindi, converbs are usually marked by the addition of -kar or -ke (though this is not obligatory). e Hindi forms in -kar, -ke appear to derivefrom an earlier pleonastic addition of a converb form of kar ‘do’—cp. the various possible converb forms in closely related Braj: mār-i, mār-i-kai,mār-i-kar-i “having struck” (Kellogg )—thus Hindimār kar “having struck”< *māri kari, lit. “having struck, having done”> “having struck”.

Example () and all subsequent examples from Sanskrit are shown in unsandhied form.

() iméhere

teyours

indraIndra.

té_

vayámwe.

puruṣṭutapraised-by-many.

yé_

tvā+ārábhyayou.+grasp.

cárāmasimove..

prabhūvasorich-one.

a. “We here, O Indra, are yours, O one praised by many, [we] who keep holding on to you, O richone.” [complex predicate reading]

b. “We here, O Indra, are yours, O one praised by many, [we] who having taken hold of you, move(around), O rich one.” [literal converb reading](Rgveda I.,; cf. Graßmann : , Delbrück : , Tikkanen : )

Bu and Lahiri () take examples like () as evidence for the existence of CV structures in Old IA, becauseof the etymological connection between absolutives and converbs discussed earlier.

Now it is true, as observed by Delbrück (: –), that car- “move” does seem to be able to bear an idiomaticsense when used with a gerund in examples like (). But, in fact, gerund+car- is part of a larger set of constructionsused in Sanskrit to indicate continuing action, a set which includes collocations involving not only gerunds, but alsopresent participles. () provides constructed examples (with cal- “move” as the main verb) of a number of thesepossible collocations, indicating the periods for which they are aested. Actual aested examples of type (-a) aregiven in () and (); of type (-b) in (); and of type (-c) in () and ().

() a. calitvāgo.

caratimove..

//etimove..

//tiṣṭhatistand..

“keeps on going” (Rig-Veda onwards)

b. calango..

āstesit..

//tiṣṭhatistand..

//caratimove..

“keeps on going” (Vedic onwards)

c. calango..

bhavatibe..

//astibe..

“keeps on going” (Classical Sanskrit)() eka

oneeva

_asyahis

doṣofault..

hi

guṇānvirtue..

ākramyaoverpower.

tiṣṭhatistand..

a. “[He has] one fault which surpasses [∼ keeps overpowering] [his] virtues”. [complex predicatereading]

b. “[He has] one fault which stands having overpowered [his] virtures.” [literal converb reading](Mahābhārata .,ab; cf. Tikkanen : )

() yathāas

sūcyāneedle..

vāsaḥclothing..

saṁdadhadmend._...

iyādgo...

evamthus

evāi

’tābhirthese..

yajñasyasacrifice..

chidraṁdefect..

saṁdadhadmend._...

etigo..

a. “Just as onewould [habitually]mend a garmentwith a needle, sowith these one [habitually]mends[any] defect of the sacrifice.” [complex predicate reading]

b. #“Just as one would move mending a garment with a needle, so with these one moves mending [anydefect o] the sacrifice.” [literal present participle reading](Aitareya Brāhmaṇa ,,; cf. Whitney /: §a)

() tasyahis..

_ahaṁI..

tapasoasceticism..

vīryaṁpower..

jānannknow._...

āsambe..

a. “I kept being aware of the power of his asceticism’. [complex predicate reading]b. #“I was one knowing the power of his asceticism.’ [literal present participle reading]

(Mahābhārata .,; cf. Speijer : §,iv)Example () is shown in unsandhied form.

() mānot

mrtaṁdead one..

rudatīweep._...

bhavabe..

a. “Do not keep weeping for the dead one”.b. #“Do not be one weeping for the dead one.”

(Rāmāyaṇa .,; cf. Speijer : §,iv)

ere are several differences between modern IA CVs and the light verb constructions of ()–(). Firstly, unlikemodern IA CVs, the constructions in () are grammatically peripheral, in the sense that they are not part of thecentral tense/aspect system of Vedic or Classical Sanskrit. Secondly, unlike modern IA CVs, which overwhelminglyimpart a ‘perfective’ sense, the constructions in () all contribute a durative/continuitive sense.

If, as Bu and Lahiri () maintain, light verbs, a category which they take to include the light “vector” verbs ofmodern IA CV constructions, are historically stable, and further, unlike auxiliaries, do not have lexical entries separ-ate from those of their full verb counterparts, then the differences between the semantic contributions of modern IAlight CV-type verbs and the Sanskrit light verbs of () are doubly surprising. Bu & Lahiri’s claim that light verbsdo not have separate lexical entries entails that, for instance, the semantics of full verb GO and light verb GO canbe somehow derived from a single (underspecified) lexical entry. us the perfective sense of the Hindi light verbjānā is supposed to be derivable from the same lexical entry from which the full verb sense “go” also derives. Evengranting that both senses could be in fact somehow derived from the same underspecified lexical entry, this claim isproblematic given that in Vedic Sanskrit i- “go” contributes a durative/continuative rather than the perfective senseof Hindi jā- “go”. Table summarises this and other discrepancies between the semantics of comparable Sanskritgerund+light verb and Hindi absolutive (< Skt. gerund)+light verb collocations.

Sanskrit Sanskrit Full verb Hindi Hindilight verb stem meaning light (vector) verb stem

semantic contribution semantic contributiondurative/continuative i-, “go, move” jā- perfective

car-durative/continuative ās- “sit” baiṭh- perfective

(‘regret’)durative/continuative sthā- “stand” khaṛā ho- perfective

(‘deliberate’ or‘immediate’)

Table : Skt. +light verb vs. Hindi + light (vector) verbs collocations

us the light verb constructions of Sanskrit differ significantly from the light (vector) verb constructions ofmodern IA, both in terms of their integration into the grammar and in terms of their semantic contribution. Whatthe light verb constructions of Sanskrit do resemble are the present participle+light verb constructions of modernIA, as exemplified by the Hindi examples in () and (), and the Nepali examples in () and (). Note that theseconstructions are morphologically and semantically distinct from CV constructions.

An example of khaṛā honā used as a light vector verb is given below.

(i) pulis-kopolice-

ātācome._..

dekhsee.

guṇḍevillain...

bhāgflee.

khaṛestand...

huebe...

“Having seen the police coming, the villains took to their heels.” (cited from Nespital : )

Note that this light verb is itself composite, consisting of an adjectival form and a (non-CV-type) light verb.

With certain verbs whose perfective participles effectively express present states, we find these sorts of constructions formed with perfectiveparticiples, e.g. (from McGregor (: )):

(i) laṛkīgirl

ekone

ghaṇṭāhour

vahāṁthere

baiṭhīsit._..

rahīremain._..

“e girl stayed siing there for an hour.”

We also find collocations of perfect participles (with invariant default masculine singular agreement morphology) with karnā “to do”, which serveto indicate the habitual (rather than the habitual-continuative of the examples in (), (), (), ()) nature of the action, e.g. (from McGregor

() vahhe/she

dinday

bharfull

so-tāsleep._..

gay-āgo._.

“He kept on sleeping all day.”() vah

he/shedinday

bharfull

so-tāsleep._..

rah-ācontinue._.

“He kept on sleeping all day.”() ma

I.bhan-daispeak-_

jān-u,go.,

taṁyou.

lekh-taiwrite-_

jāgo.

“I will keep dictating, and you keep writing.” [cited from Pokharel : ]() āun-dai

come-_garnudo.

“Keep on coming (from time to time).” [cited from Ibid.]

Not only are modern IA constructions in ()–(), like the Sanskrit constructions in ()–() above, grammat-icality peripheral, in that they are not part of the core tense/aspect system of Hindi or Nepali, but these modern IAlight verbs also make the same sort of semantic contribution as do those of Sanskrit, namely durative/continuative.Further, as noted above in (), the Sanskrit light verb constructions include not only gerund+light verb, but also—asin Hindi and Nepali—present participle+light verb, both of which contribute durative/continuative senses. erefore,examples suggested by Bu & Lahiri as being early CV constructions in Sanskrit are actually much more comparableto the grammatically-peripheral present participle+light verb constructions of ()–().

Both in Old IA and modern IA, there are a variety of different V-V collocations, involving the light verb of theconstruction selecting for morphologically different types of V1s. In Old IA, both the collocations involving gerundV1s and those involving present participle V1s are grammaticality-peripheral and involve continuative semantics.us both are actually more comparable to the Modern IA collocations involving present participle V1s than to themodern IA CV construction.

Bu and Lahiri (: ) cite an example of a Skt. gerund+light verb construction, (), where they take thelight verb as contributing a perfective sense comparable to that of modern IA light (vector) verbs.

() tataḥthen

makṣikā+fly

uḍḍīyafly-up.

gatāgo._..

a. “en the fly flew off.” [complex predicate reading]b. “en the fly, having alit (=flown up), le.” [literal converb reading]

(Pañcatantra ., cited Tikkanen : )

ey claim that here “the verb ‘go’ cannot be seen as the main predication of the sentence. . .[r]ather it modifiesthe event semantics of the participle ‘fly’ ”, comparing it to modern IA examples like Hindi ().

() makkhīfly...

uṛfly.

gayīgo._..

“e fly flew off.”

Although () looks superficially like (), it is perfectly possible, despite Bu & Lahiri’s assertions to the contrary, tointerpret the gerund literally, as indicated by the translation in (-b). In fact, the overall absence of examples of this

(: )):

(ii) pahlebefore

meremy

yahāṁhere

āyācome._..

kartīdo._..

thī,be...,

abnow

nahīṁnot

ātīcome._..

“She used to frequently come to my place, but now she doesn’t.”

Note that in this construction, the main verb, occurring as a perfect participle, always takes the default masculine singular form, regardless of thenumber/gender of the subject.

e peripherality of such constructions is such that a speaker could still sound idiomatic and nativelike without making use of these con-structions. e same would not be true of a speaker who failed to use CV constructions, on which see further Hook (), who notes the factthat CVs are nearly obligatory in realis contexts (at least in Hindi; this is less true for certain other IA languages, such as Nepali).

Example () is shown in unsandhied form.

sort from Sanskrit suggests that positing a “perfective” light verb GO in Sanskrit would require special pleading.More generally, the close examination of the putative examples of early IA CV constructions cited by Bu and

Lahiri () has revealed that none of the instances they cite are truly comparable to CV constructions of modernIA. Some of these examples were shown to be perfectly acceptable with their literal converb readings, while the non-spurious examples of light verb constructions in Sanskrit were shown to most closely resemble the grammatically-peripheral light verb constructions in modern IA involving imperfect or perfective participles. In sum, there is noevidence which would suggest that we aribute CVs to Old IA.

. Excursus: Periphery vs. Core

In Section . above, I distinguished between peripheral light verb constructions, like the continuative constructionsof Sanskrit () and Hindi (), () (and Nepali (), ()), and core grammar constructions, like Hindi CV con-structions. In this section I consider examples of core vs. peripheral constructions in English in order to make thisdistinction more concrete.

In general, core grammar constructions are less specific in their functions than more peripheral constructions,and are involved in the expression of more basic functions. us in () below, will X is the basic construction forexpressing future tense in English, whereas the constructions to the le of the doed line on the periphery-corecontinuum can be considered to be more peripheral, non-basic constructions associated with future time.

() English future-oriented constructions: (X = bare infinitive verb)

..P

.C

.

BE

. under oath to X.

BE

. sworn to X.

BE

. bound to X. will X

e core-periphery distinction is at least somewhat gradient in nature, as shown by the fact that the constructionBE bound to X can have a much less specific sense than either BE sworn to X or BE under oath to X, namely “have alogical necessity to X”. In origin, BE bound to X is nearly equivalent to the other two constructions, namely bearingthe sense “having entered into a contract binding to service; under legal or moral obligation” (OED); showing thatBE bound to X has undergone grammaticalisation, which has led to the construction bearing a more general future-oriented sense.

Turning to an English example which is semantically closer to the sense of the peripheral constructions con-sidered above, namely Sanskrit () and Hindi (), (), consider the English continuative constructions in ().

() English continuative constructions: (X = bare infinitive verb)

..P

.C

.

BE occupied

. in X-ing.

BE engaged

. in X-ing.

BE in the process

. of X-ing.

keep on

. X-ing.

keep

. X-ing.

BE

. X-ing

Here the construction BE X-ing is the most basic expression of continuative action in English, a core grammar con-struction. e constructions to the le of the doed line are more peripheral constructions with less basic continu-ative senses. Again, as in the case of BE bound to X, some of these peripheral constructions, namely keep X-ing andkeep on X-ing, have undergone grammaticalisation and become more general—but still not as general as the core

e only other example involving an apparent light verb meaning “go” I have seen presented is also from Tikkanen (: ):

(i) tatothen

yāvatwhen

sāshe..

tāṁthat..

sārikāṁSarika-bird..

galamoṭanapūrvaṁbefore wringing the neck

vināśayaticause the destruction of..

tāvadthen

uḍḍīyafly-up.

yayaugo._.a. “en, before she had time to strangle that Sarika-bird to death, it flew away.” [complex predicate reading]b. “en, before she had time to strangle that Sarika-bird to death, it alit and le.” [literal converb reading]

for which the same concerns may be raised as for ().

construction BE X-ing, as shown by the contrast between (-a) and (-b) below.

() a. John was working all day.b. John kept on working all day.

e core-periphery distinction between BE X-ing and keep (on) X-ing is neatly paralleled by the distinction in Hindibetween X-tā rahnā “keep on X-ing” and X rahā HONĀ “is X-ing”, see ().

() Hindi continuative constructions: (X = bare verb stem)

..P

.C

.

X-tā rahnā

. “keep on X-ing”. CVs.

X rahā HONĀ

. “is X-ing”.

X-tā HONĀ

. “X-es”

CV constructions—in contrast to X-tā rahnā—can be considered part of the core grammar in view of the fact thatthey are nearly obligatory even in very basic expressions. For example, the simple statement he died is much moreidiomatically expressed with a CV (-a) than with a simplex verb (-b).

() a. vahhe/she

mardie.

gayā.go._..

“He died.”b. vah

he/shemarā.die._..

“He died.”

. Verb-verb collocations in Middle Indo-Aryan

Bu and Lahiri () also point to possible examples of early CV constructions in Pāli texts from Sri Lanka (Tikkanen; Hook ), with the most convincing examples involving gerund+“give”, as in (), ().

() ath’then

assahim

sahāmaster

udakasāṭakaṁwater-cloak

khipitvāthrow.

adāsigive..

a. “en the master threw a bathrobe to him.” [complex predicate reading]b. “en the master gave him a bathrobe, throwing (it).” [literal converb reading]

(Dhammapad-Aṭṭhakathā II,,; cited Hendriksen : )() so

hetassāher

saddaṁcry.

sutvāhear.

. . .

. . .assamapadaṁhermitage.

ānetvābring.

aggiṁfire.

katvāmake.

adāsigive..

a. “He, having heard her cry, having brought her to his hermitage, made a fire (for her).” [complexpredicate reading]

b. “He, having heard her cry, having brought her to his hermitage, having made a fire, offered it (toher).” [literal converb reading](Jātaka I,,; cited Hendriksen : )

e examples in () and () look much more promising than any of the other supposed early examples of CVconstructions, however even here (as indicated by the two possible translations) it is not necessary that we interpretsuch examples as involving complex predicates.

In example () a literal converb reading is possible, as indicated by the translation, and thus again there is noevidence which forces us to interpret “give” here as a light (vector) verb—in fact, given that () involves a string ofgerunds (“having heard her cry, having brought her to his hermitage, . . .”), a complex predicate interpretation seemseven more unlikely.

Example (), at first blush, looks much less likely to be interpretable as involving a literal converb reading.However, gerunds sometimes bear ‘non-past’ interpretations, see the Sanskrit examples () and () below.

Constructions similar to, though not completely identical with, keep on X-ing include get on X-ing, carry on X-ing, press on X-ing. See Cappelle() for further discussion, including the semantic distinctions and historical differences between keep X-ing and keep on X-ing.

On ‘non-past’ readings of Sanskrit gerunds, see Hock ().

() sītāSita...

mad-vacanātmy-command.

vācyāspeak.

samāśvāsyaconsole.

prasādyacalm.

caand

a. #“By my command Sita is to be spoken to, having consoled and calmed (her).” [past-tense converbreading]

b. “By my command Sita is to be spoken to consolingly and calmingly.” [non-past converb reading](Mahābhārata .,, cited Tikkanen : )

() vikramārkoVikramārka.

nītimrules-of-ethics.

ullaṅghyatransgress.

rāyjaṁrule.

nanot

karotido..

a. #“Vikramarka, having transgressed the rules of ethics, does not rule (his kingdom).” [past-tenseconverb reading]

b. “Vikramarka does not rule (his kingdom), transgressing the rules of ethics.” [non-past converbreading](Vikramacarita ., cited Tikkanen : )

As indicated by the translations, in these examples the past-tense converb reading is unavailable. In Pāli as wellwe find examples where gerunds must be interpreted with a non-past reading, as in example () below.

() . . .. . .

athathus

sohe.

. . .

. . .bhatiṁwages.

katvādo.

jīvatilive..

a. #“. . . and so he, having worked as a day-labourer, lives.” [past-tense converb reading]b. “. . . and so he lives by working as a day-labourer.” [non-past converb reading]

(Jātaka , cited Tikkanen : )

us a literal non-past converb reading of example () is perfectly plausible, i.e. “then the master gave him abathrobe, (by) throwing (it)”.

e Pāli examples in () and () thus need not be interpreted as involving complex predication. Such examplesdo look like excellent starting points for the reanalysis of gerund constructions as complex predicates, but there is noreason to assume that such a reanalysis has already occurred by the stage of Pāli since in all cases the literal converbreadings are perfectly possible.

. Verb-verb collocations in late Middle Indo-Aryan and early Modern Indo-Aryan

It has been suggested that CV constructions are found in Apabhraṁśa (Hook ; Bubenik ). However, mostif not all of Bubenik’s examples seem to be more felicitously interpreted in other ways, as suggested to me by EvaDe Clercq (p.c., Jan ). Singh (: –) lists, without context, a number of examples of potential CVconstructions in Apabhraṁśa. Given the uncertainties of interpretation of the Apabhraṁśa data then, I leave themout of consideration. is is an area requiring further investigation.

e first unequivocal examples of CV constructions do not appear until the early modern IA period (th–thcenturies). See the examples from early Braj Bhāṣā (a close relative of the ancestor of modern Hindi) in (), (),and the Nepali examples in (), ().

() nātaruotherwise

jauif

vacchācalf.

maridie.

jāi,go..,

tauthen

gāicow.

cchiṃḍāītake_away._..

jāigo..

a. “Otherwise, if the calf dies, then the cow is taken away.” [complex predicate reading]b. #“Otherwise, if the calf having died goes, then the cow is taken away.” [literal converb reading]

(ca. A.D., Indrajit of Orchā’s commentary on the Nītiśataka of Bhartrhari, f.b, cited fromMcGregor : )

() bhayabhītawithout fear

hvai-karibe-

samudraocean

kauof

mathivauchurning

nanot

cāḍ-ileave-

dayaugive._..

Even if one were to assume that examples () and () represent nascent CV constructions, since these texts are from Sri Lanka, it is not clearwhat bearing they have on the development of compound verb constructions in mainland IA, especially given that they may reflect the influenceof Dravidian Tamil or Sinhala (Sinhala shows a well-developed CV system as early as the Sigiri graffiti texts of the th–th c., see Paranavitana: .§).

Again, I note that Masica (: ) concurs on this point.

a. “Having become fearless, (he) did not leave off the churning of the ocean.” [complex predicatereading]

b. #“Having become fearless, (he), having not le the churning of the ocean, gave.” [literal converbreading](ca. A.D., Indrajit of Orchā’s commentary on the Nītiśataka of Bhartrhari, f.a, cited fromMcGregor : )

() tahāṁthere

kājikanaKazi.

vujhāiunderstanding

samjhāiunderstand..

ñahāhere

paṭhā-i-send-

dinu havasgive.

a. “Persuade the Kazi there, and send him here.” [complex predicate reading]b. #“Persuade the Kazi there, and having sent him here, give (him).” [literal converb reading]

(/ A.D. ∼ V.S., Prithvinarayan Shah, leer to Paṇḍit Rājīvalocana, cited from Pokharel: )

() tyothat(thing)

cāṁḍoquickly

māgerarequest.

paṭhā-i-send-

deugive.

a. “ickly request that thing and send it (to me).” [complex predicate reading]b. #“ickly request that thing and having sent it, give (it) (to me).” [literal converb reading]

(/ A.D.∼ V.S., Prithvinarayan Shah, leer to Haripaṇḍit, cited from Pokharel : )

As indicated by the translations, in these examples a literal converb reading is unavailable, demonstrating that bythe early modern period of IA, CVs have made their appearance.

. Conclusions

Sound evidence for CV constructions in IA thus does not appear until the early modern period (pace Bu & Lahiri).Verb-verb collocations apparently involving light verbs do appear early on in IA, but these are grammatically-peripheral constructions which are moreover distinct in their semantics from CV constructions, resembling insteadthe (likewise peripheral) present participle+light verb constuctions of modern IA.

Morphosyntactic and lexical ange in IA CVsIf light verbswere, as claimed by Bu and Lahiri (), “diachronically stable”, we should expect themorphosyntacticand lexical properties of the CV systems of the various modern IA languages to be very similar. is situationhowever does not obtain. Hook and Pardeshi () point out a number of light verb constructions in IA which showevidence of historical change, and herein I provide a number of additional instances of historical changes involvinglight verbs in IA. In this section I examine particularly two closely related IA languages, Hindi and Nepali, andshow that, both in terms of morphosyntax and in terms of the lexicon, the CV systems of these two languages differsignificantly. ese differences indicate that change has taken place in the CV systems of one or both of theselanguages.

. Morphosyntactic ange in IA CVs

I examine four differences between the morphosyntax of Hindi and Nepali CV constructions. ese properties Irefer to as , the ability of elements to intervene between the main verb and the light verb of theCV construction; , whether a main verb may be modified by more than one light verb; , i.e. which syntactic constructions light (vector-type) verbs can occur in; and ,whether agentive marking of subject of transitive verbs in perfective tenses is controlled by the transitivity of themain verb or the transitivity of the light verb.

Nepali and Hindi are fairly closely related languages: as Mahews (: i) remarks, “[Nepali and Hindi] are in fact so close to each otherthat early Western grammarians regarded Nepali merely as a dialect of Hindi”. e relationship between the two languages is roughly paralleledby that between Italian and Spanish.

ese properties represent a subset of the properties derived in a boom-up fashion based on consideration of the major differences in Hindiand Nepali CV constructions as found in my data. ese data were drawn from previous accounts of compound verbs in Hindi and Nepali (seefn. ), from grammars (Hindi: Kellogg , McGregor ; Nepali: Verma and Sharma , Mahews ), and from consultation with nativespeakers. Early Nepali examples are taken from texts contained in Pokharel .

.. Interruptibility

Hindi CVs are interruptible; that is, other words may occur in-between the main verb and the light verb componentsof the CV, as shown in ().

() mai-neI-

khānāfood

khāeat.

to

liy-ā,take...

lekinbut

phirthen

ulṭīvomit

bhīalso

ā-īcome...

“I did eat food, but then I also vomited.”

Nepali CVs are not interruptible, even by particles like ta (equivalent to Hindi to):

() *mai-leI-

bhātfood

khā-i-eat--

ta

sak-em,finish-..

. . .

“I did eat food, . . .”

is difference is possibly reflected in writing, as Hindi CVs are wrien as separate words, but Nepali CVs as asingle word.

e inseparability of Nepali CVs suggests that they are either formed in the lexicon, or else (e.g. in a DistributedMorphology framework, cf. Halle and Marantz ) composed at a much lower level of syntax than Hindi CVs. Inother words, Nepali vectors appear to be more affixal in nature than are Hindi vectors.

.. Recursion

Hindi allows for only one light verb per main verb, as shown in ().

() a. us-nehe/she-

kicankitchen

sāfclean

kardo.

dīgive....

“He/she cleaned the kitchen for me.”

b. us-nehe/she-

kicankitchen

sāfclean

kardo.

ḍāl-īput...

“He/she cleaned the kitchen straightaway.”

c. *us-nehe/she-

kicankitchen

sāfclean

kardo.

degive.

ḍāl-īput...

“He/she cleaned the kitchen for me straightaway.”

In Nepali, on the other hand, CVs may involve up to two light verbs, as in (), where the main verb, gari, ismodified by both dii (itself in absolutive form) and hālin.

() un-lehe/she.MH

kicankitchen

saphāclean

gar-i-do--

di-i-give--

hāl-input-..

All Hindi speakers consulted allowed for particles like to to intervene between main verb and light verb, as in (); most speakers also allowpronouns to intervene, as in (i-a); some also allow for full NPs, as in (i-b), (i-c).

(i) a. ācome.

to

vahhe

gay-āgo...

hai,be...

lekinbut

sabzivegetables

nahīṁnot

lāy-ābring...

“He has indeed come, but he didn’t bring the vegetables.”

b. %khāeat.

to

khānāfood

liy-ātake...

“(I/he/she) did eat food.”

c. %liy-ātake...

to

khānāfood

khāeat.

“(I/he/she) did (in fact) have food.”

Note that this is unlikely to be an artefact of the script, as both languages employ the same writing system (devanāgarī ). An anonymousreviewer points that that Nepali CVs are not invariably wrien as if single words, though this is the majority paern.

“She cleaned the kitchen for me straightaway.”[Example from Peterson (: )]

While it is not entirely clear which of these paerns represents the more innovative andwhich themore conservative,it is obvious that the CV system of at least one of the two languages has undergone historical change.

.. Construction-specific restrictions

Nepali CVs can occur in conjunctive participles. us, for example, a converb in Nepali may be composed from aCV, as in ().

() bhātfood

khā-i-eat--

sak-erafinish-

uhe/she

sut-nasleep-.

gay-ogo.

“Having finished dinner, he went to sleep.” [Example from Peterson (: )]

In contrast, Hindi converbs can only be formed from simplex verbs. erefore while (-b) is grammatical, (-a)is not.

() a. *khānāfood

khāeat.

letake

kar

vahhe/she

gay-āgo._.

“Having eaten up the food, he le.”b. khānā

foodkhāeat.

kar

vahhe/she

gay-āgo._.

“Having eaten the food, he le.”

With respect to this property, Hindi appears to have innovated: what is now the overt converb marker in Hindi, karor ke, derives ultimately from the pleonastic addition of a converb form of kar “do” (see footnote above), pointingto the possibility of forming converbs of CV collocations in earlier Hindi, and suggesting that with respect to thisproperty Nepali is conservative.

.. Dominance and transitivity issues

In both Hindi and Nepali, the question arises of how to reconcile clashes in transitivity between main verb andlight verb in CV combinations. Before going into the details, it is important to understand the basic paerns ofcase-marking morphology in these two languages.

Both Hindi and Nepali (more or less) employ a special marking for agents of transitive verbs in perfectivetenses, -ne in Hindi, -le in Nepali. In Hindi, the marking of agents correlates with verbal agreement: agentivecase-marked nouns do not control predicate agreement. In Nepali, (non-experiencer) subjects control predicateagreement regardless of whether they are agentive case-marked or not.

More relevant, for the purposes of CVs, is the fact that in Hindi it is the light verb member of the CV whichdetermines whether the entire CV is treated as transitive or intransitive (and thus whether the subject receivesagentive casemarking)—except in the case of intransitive main verb combined with transitive light verb (where wefind variation in the assignment of transitivity to the compound for the purposes of case-assignment); in Nepali, itis always the main verb which determines the transitivity of the entire verbal compound.

Examples like (-a) are possible in Dakkhini Hindi (spoken in Hyderabad), which has undergone convergence with Dravidian Telugu, seeArora ().

Nepali also sometimes displays agentive case-marking in non-perfective contexts; the exact conditions on the use of -le in such cases is notentirely clear: it seems to be more frequent with third-person subjects. Li () suggests that it is obligatory with inanimates, while Poudel() suggests that -le may distinguish individual-level predicates (a property that holds for a slice of the referent’s spatio-temporal existence)from stage-level predicates (denoting a lasting/inherent property of the referent), whose subjects are Ø-marked. Some dialects of Nepali havegeneralised agent marking to all transitive verbs regardless of tense/aspect, predicate-type, animacy—this probably reflects convergence withTibeto-Burman.

e unmarked object usually controls verb agreement; if this is blocked by the oblique postposition ko, the verb takes default masculine,singular, third-person agreement.

Combinations of intransitive main verb and transitive light verb are very rare in Hindi, apparently largely avoided by Hindi speakers. I havefound only two examples from Nespital (: -) where the case-assignment is clear (i.e. where the finite verb appears as a perfect participle).

(i) śatruenemy

senāarmy

acānaksuddenly

hī

hamārīour

senāarmy

paron

ācome.

lītake._..

“e enemy army suddenly pounced on our army.”

() H CVa. Intransitive main verb + intransitive light verb = Intransitive:

vahhe

ācome.

gay-āgo...

“He came.”

b. Transitive main verb + intransitive light verb = Intransitive:vahhe

khānāfood

khāeat.

gay-āgo...

“He ate up the food.”

c. Transitive main verb + transitive light verb = Transitive:us-nehe-

khānāfood

khāeat.

liy-ātake...

“He ate up the food.”

d. Intransitive main verb + transitive light verb = (variation) (see fn. )vah/us-nehe./he-.

baiṭhsit.

liy-ātake._.

“He sat down.”() N CV

a. Intransitive main verb + intransitive light verb = Intransitive:uhe

mandir-mātemple-

ga-i-go--

ā-ekocome-...

chabe..

“He kept going to the temple.”

b. Transitive main verb + intransitive light verb = Transitive:us-lehe-

yothis

kāmwork

gar-i-do--

ā-ekocome-...

chabe..

“He has continued to do this work.”

c. Transitive main verb + transitive light verb = Transitive:us-lehe-

bhātfood

khā-i-eat--

di-yogive-..

“He ate up the food.”

d. Intransitive main verb + transitive light verb = Intransitive:

(ii) jabwhen

satīśSatish

apnehis_own

gharhouse

sefrom

nikl-āexit-_..

tothen

uskāhis

kuādog

bhīalso

uskehis

pīchebehind

hobe.

liy-ātake-_..

“When Satish le his house, then his dog also followed him.”

In both of these cases, the subject is not marked with an agentive marker, despite the light verb member being transitive. Remarkably, in both(i) and (ii), not only is the light verb transitive, but the CV itself appears to be semantically transitive (in the sense that both pounce and followrequire complements).However, more generally combinations of intransitive main verb and transitive light verb seems to result in variation in Hindi (probably as

a result of the rarity of such combinations). at is, some speakers prefer (iii), others prefer (iv) (yet others prefer to avoid such combinationsaltogether).

(iii) vahhe

baiṭhsit.

liy-ātake...

“He sat down.”

(iv) us-nehe-

baiṭhsit.

liy-ātake...

“He sat down.”

Speakers show similar variation with respect to other combinations, such as dikh diyā “appeared”, and ghum liyā “roamed”.

uhe

rāmRam

ko-lāgifor-sake-of

bajārmarket

ga-i-go--

diy-ogive-..

“He went to the market for Ram’s sake/in place of Ram.”

us, in Hindi it is the light verb which usually determines the transitivity of the entire compound, for purposes ofassignment (or non-assignment) of agentive marking to the subject. In Nepali, it is the main verb which determinesthe transitivity. Again, though it is unclear which of these represents the more conservative system, it is clear thatchange has taken place in the CV system of one or both of these languages.

. Lexical ange in IA CVs

While there is much commonality in the sets of light verbs employed by the different modern IA languages in CVconstructions, these sets are not identical. us, in some cases the same light verb sense is expressed using verbswith the same full verb meaning, e.g. Hindi ḍāl- and Nepali hāl, Hindi de- and Nepali di- (etymological cognates, inthe laer case); some light verb senses are expressed using verbs with distinct full verb senses, e.g. Hindi baiṭh- andNepali paṭhau-; in some cases a light verb sense is expressable only in one of the languages, e.g. Hindi le-. See Table.

H H L N N

ḍāl throw immediacy throw hālde give other-directed give dile take self-directed —cuk finish completive complete / be able to sakbaiṭh sit regret send paṭhāu

Table : Selected Hindi and Nepali light verbs compared

ese various differences are the result of different historical developments in the CV systems of Hindi andNepali.

.. Against shared full verb/light verb entries : vectors of regret

In their analysis of IA light verbs, Bu and Lahiri (: –) suggest that full verbs and light verbs of IA CVconstructions can be derived from a single underspecified entry, e.g. that both the full and light verb uses of Hindidenā “give” derive from a single lexical entry, as in ().

() ..“give”. Underspecified Entry.

Full Verb

.

Light Verb

Bu and Lahiri () adopt a semantics (see further, Bu and Ramchand ; Bu and Geuder , ; Bu, ) involving subevents, where the light verb use of “give” can be “taken to modify the superordinate eventby contributing lexical semantic information which is loosely based on the predicational force of the main verb ‘give”’(). More specifically they posit that this information consists of an adverbial event modification where “give” con-tributes to the main verb semantics the fact that the event (a) involves force emission and (b) is agentive/purposeful.

It is far from clear what the nature of the underspecified entry of () would be, such that both the full verb andlight verb uses of “give” could be systematically derived from it. However, even if we grant that this theory could bemaintained for light verbs like GIVE, it still requires that the system of derivation of full and light verbs from a singleunderspecified lexical entry be crosslinguistically robust, in the sense that the connections between the full verb andlight verb senses should be predictable. is might seem reasonable in the case of light verbs like GIVE—which do

at least tend crosslinguistically to involve similar light verb semantic contributions—but even within IA, languagesvary greatly in the association between full verb and light verb meanings.

For example, both Hindi and Nepali have light verbs used when the speaker wishes to express regret; however,Hindi and Nepali use light verbs whose full verb counterparts bear completely different senses: the Hindi light verbindicating regret is baiṭhnā “to sit” (see ()), the Nepali light verb indicating regret is paṭhāunu “to send” (see ()).

() maiṁI.

kyāwhat

kardo.

baiṭh-ā?sit._.

“Oh what have I done?”() mai-le

I-kewhat

gar-i-do--

paṭhā-eṁ?send._

“Oh what have I done?”

Not only is it unclear what sort of single underspecified lexical entry could underlie the full and light verb sensesof Hindi baiṭhnā or Nepali paṭhāunu, but the very fact that these languages, for the same light (vector) verb semanticcontribution, employ verbs with completely different full verb senses strongly suggests that, even where there ishomophony between full and light verbs, light verbs have their own lexical entries, separate from those of their fullverb counterparts (where counterparts even exist).

More relevantly for my current purposes, the lexical differences in the light verb inventories of Hindi and Nepaliare another indication of historical change affecting light verbs independently of their full verb counterparts.

.. Against shared full verb/light verb verb entries : independent phonological ange

e Nepali vector baksinu provides another example of change affecting a light verb independently of its full verbcounterpart.

Nepali baksinu is a light verb employed in CV constructions as an honorific when referring to Nepali royalty andother persons due great respect, e.g.

() mausuph-leHis Majesty-

yothis

kurothing

ghosaṇāproclamation

gar-i-do--

baksi-yobestow..-

“His Majesty made this proclamation.” (Sharma : )

A reduced form of baksinu is also employed in upper-class Kathmandu families (e.g. by children to parents, wife tohusband etc.), e.g.

() buwā-ledad-

bhujārice

khā-i-eat--

s-yo-

“Dad ate rice.” (Sharma : )

While baksinu is employed as a main verb, (), -s- is not, ().

() mahārāni-lequeen-

ma-lāiI-

takmāmedal

baksi-yobestow-.

“e queen bestowed a medal upon me.” (Ibid.)() *mahārāni-le

queen-ma-lāiI-

takmāmedal

s-yohon-.

at baksinu as a light verb, but not as a full verb, can appear in the reduced form -s- is problematic for Bu & Lahiri’sclaim that, since light verbs share a single lexical entry with their full verb counterparts, they cannot undergo anychange not also experienced by their full verb counterparts.

See, for instance, Masica (: -), who notes, beside the ubiquity of GIVE in Indo-Aryan, similar functions for light verb uses of GIVEin Burmese, Tajik, and Uzbek. Note however, that even in other languages employing light verb type constructions, for instance Japanese andTurkish, GIVE appears with rather different functions (as an honorific in Japanese, as a light verb indicating “rapidity, ease, suddeness” in Turkish).Note further that even in Indo-Aryan GIVE does not always contribute a sense of “other-directedness”; in Nepali, alongside of this function, italso can simply mark general perfectivity of the action.

Nepali baksinu is ultimately a loanword from Persian bakhś “to give”, and obviously was borrowed as a full verb, not as a light verb, since, asshown by example (), it can still be used as a full verb.

.. Light verbs laing full verb counterparts

ough Bu and Lahiri () maintain that light verbs always have full verb counterparts, it is not clear that this isthe case.

Nepali employs ṭopalnu as a light verb indicating pretense, as in example () below.

() uhe/she

gā-i-sing--

ṭopal-dai-pretend-_-

chabe..

“He is pretending to sing.” (cited from Pokharel : )

Turner (: ) suggests that a full verb counterpart may have once existed, possibly meaning “to cover” (cp.Nepali ṭopi “helmet”, cf. Hindi ṭopī ‘hat’, Hindi ṭopnā “to cover”). However, synchronically there is no full verbcounterpart of this Nepali light verb, cf. ().

() *uhe/she

ṭopal-dai-pretend-_-

chabe..

“He is pretending.”

e lack of a full verb counterpart for ṭopalnu demonstrates that Bu & Lahiri’s claim that light verbs always sharea single lexical entry with their full verb counterparts cannot be maintained, if we assume that ṭopalnu is a lightverb. However, given that ṭopalnu makes a rather more substantial semantic contribution than do most light verbs,it might be tempting to treat it as either a (modal-like) auxiliary, or else as a full lexical verb which obligatorily takesan absolutive as its complement.

A similar case is found in Hindi, where the element saknā “to be able to” selects for an absolutive complement,just as do light verbs in Hindi. Like Nepali ṭopalnu, Hindi saknā cannot appear as a main verb, in the sense that itobligatorily takes an absolutive complement.

() a. vahhe/she.

gāsing.

sakt-ābe_able._.

haibe..

“He can sing.”b. *vah

he/she.sakt-ābe_able._.

haibe..

“He can.”

Not only does saknā in Hindi select for an absolutive complement, but it also behaves like a light verb in that itfollows the Hindi rule of a maximum of one light (vector) verb per main verb (see above, Section ..), i.e. saknācannot co-occur with a(nother) light verb (see Nespital : ).

() a. vahhe/she.

bolspeak.

sakt-ābe_able._.

haibe..

“He can speak.”b. *vah

he/she.bolspeak.

degive.

sakt-ābe_able._.

haibe..

“He can speak (out).”c. *vah

he/she.bolspeak.

sakbe_able.

det-āgive._.

haibe..

“He can speak (out).”

us both Nepali ṭopalnu and Hindi saknā are ungrammatical as main verbs, and behave morphosyntacticallylike light verbs. e analysis of these elements in terms of categorical distinctions between lexical verb vs. light verbvs. auxiliary is therefore unclear, pointing again to inherent difficulties in distinguishing between different functionalcategories.

. Conclusions

In summary, we have seen that the CV systems of Hindi and Nepali differ not only in their vector inventories, butalso in various aspects of morphosyntax.

In section ., it was shown that Hindi and Nepali light verbs in CV constructions show only partial overlap interms of light verb meanings; and in some cases, the same light verb meaning is realised but with each languageusing verbs with very different full verb senses (SIT in Hindi vs. SEND in Nepali for the vector sense of ‘regret’). Wealso found that not all light verbs have full verb counterparts—notably, the Nepali vector ṭopal- has no (synchronic)full verb counterpart.

Turning to morphosyntactic differences, section .. demonstrated that while in Hindi certain elements externalto the CV can intervene between sequences of main verb-light verb, in Nepali no element may intervene betweenthe main verb and the light verb(s). is is the first piece of data suggesting that Nepali vectors are more ‘affixal’ innature.

Section .. presented data which show that Hindi allows for one light verb per main verb, while Nepali allowsfor at least two light verb to occur with a single main verb. Here again, Nepali light verbs appear more affixal, sincea Nepali CV behaves like a simplex verb in that it can combine with another light verb.

With respect to construction-specific conditions on light verbs, as discussed in section .., Nepali CVs againbehave like simplex verbs, in that they can appear in the same set of environments in which simplex verbs canappear—once again pointing to the affixal nature of Nepali light verbs.

Finally, with respect to dominance and transitivity marking, section .., Nepali light verbs again are found toact more like affixes than do Hindi light verbs, since the transitivity of the entire CV compound is always based solelyupon the transitivity of the main verb in Nepali. In Hindi, conversely, it is usually the light verb which determineswhether the CV is treated as transitive or intransitive for purposes of case-marking.

us, with respect to all four of these properties, Nepali light verbs in CV constructions appear more affix-likethan do Hindi light verbs—whether this means that Nepali light verbs are ‘more grammaticalised’ than Hindi vectorswill be considered in Section ..

e variation in morphosyntactic and lexical properties of light verbs in Nepali and Hindi is not characteristic ofa ‘stable system’, contra Bu and Lahiri (), Bu () etc. Change has taken place in the CV systems of one, ormore likely, both of these languages.

Evolution from light verb to auxiliaryBowern (: ), while recognising that light verbs can undergo change (contra Bu and Lahiri ), says shehas found “no evidence for light verbs being reanalysed as auxiliaries”, and suggests that “they have different originsand do not readily change into one another. . .therefore representing them on the same cline is misleading” (Bowern: ). Bowern refers to the clines of Grammaticalisation eory (e.g. Hopper and Traugo ), which aresaid to represent ‘universal pathways of change’. e type of cline Bowern takes exception to is one which positslight verb as the stage preceding auxiliary verb, as in the cline represented in ().

() (a) parataxis > (b) hypotaxis > (c) serialisation > (d) light verb > (e) auxiliary verb > () univerbatedaffix (Andersen )

Bowern () suggests that light verbs and auxiliaries should instead be understood to follow different ‘grammatic-alisation’ clines, in the apparent absence of evidence of light verbs becoming auxiliaries.

Indo-Aryan, however, provides us with instances of light verbs becoming auxiliaries, specifically becomingtense/aspect auxiliaries, as discussed in the remainder of this section. In modern Hindi, rahnā, which functionedas a light verb in th-century Hindi, has become a grammatically-central auxiliary. e case of Nepali rahanu, dis-cussed in . is a particularly interesting example, as it is apparently a light verb which is currently in the process ofbecoming an auxiliary.

Having found no evidence for light verbs becoming auxiliaries, Bowern (: ) posits that “light verbs andauxiliaries have. . .distinct diachronies”. e Indo-Aryan evidence presented here demonstrates that light verbs canin fact become auxiliaries. Further, these changes, involving a grammatically-peripheral light verb contributing adurative/continuative sense being reanalysed as a core continuative aspect auxiliary represent a repetition of history:the auxiliary honā of themodernHindi simple present (whichwas in earlier Hindi a present progressive construction)derives from an early IA use of “be” as a grammatically-peripheral durative/continuative light verb.

It is interesting to note that these changes do not appear to take place via gradual grammaticalisation along acline, but rather as part of a larger morphosyntactic chain shi, triggered by the reanalysis of old simple presentforms as modal.

. e development of the Hindi auxiliary rahnā

e Hindi auxiliary rahnā itself originated as a light verb. In modern Hindi, the simple present and the presentcontinuous are clearly distinguished, as shown in ().

() a. maiṁI

skūlschool

jātāgo._..

hūṁbe..

“I (habitually) go to school.”

b. maiṁI

skūlschool

jāgo.

rahāremain._..

hūṁbe..

“I am going to school (just now).”

But Kellogg (: §), in the late th century, cites forms like maiṁ jātā hūṁ as meaning either “I go” or “I amgoing”. Further, he categories rahnā as a light verb (Kellogg : §) rather than an auxiliary, suggesting that—as in the case of modern Nepali rahanu—the use of rahnā as indicating continuous action in th century Hindi wasmore peripheral and not yet integrated as part of the core grammar as a clear aspectual auxiliary; see the examplesin () and (), taken from Kellogg (: §), retaining his translations.

() a. donoṁboth

laṛkeboy.

khelteplay._..

thebe..

“e two children were playing.”b. donoṁ

bothlaṛkeboy.

khelplay.

raheremain._..

thebe..

“e two children were engaged in play.”() a. vah

hesuntāhear._..

haibe.

“He hears.” / “He is hearing.”b. vah

hesunhear.

rahāremain._..

haibe.

“He is occupied in hearing.”

e later ‘auxiliary’ nature of Hindi rahnā represents a reanalysis which is part of larger reconstructing of the Hindiverbal system, see further below in Section ., especially Table .

. Nepali rahanu: light verb > aspectual auxiliary

Like baksinu (see above Section ..), Nepali rahanu ‘remain, stay’ behaves morphologically like other light verbs,selecting for an absolutive participle in -i. Rahanu is also identical to other light verbs with respect to restrictionson recursion and construction-specific restrictions (see above sections .. and .., respectively).

However, unlike other Nepali CV constructions, CVs involving rahanu (or rākhnu ‘keep’) never require agentive-marking on the subject, regardless of the transitivity of the main verb. In this, the construction including rahanu inits light verb function appears to be a periphrastic counterpart of the Nepali synthetic imperfect construction, verb-stem+dai+inflected form of “to be”. In other words, rahanu as a light verb, ()—with respect to agentive casemarkingassignment—paernswith the synthetic imperfect in -dai, (), not with other CVs (such as the light verb di- as shownin ()).

() a. maI

mandir-mātemple-

ga-i-go--

di-ekogive-_..

chube..

“I have gone to the temple (for someone).”

b. mai-leI-

yothis

kāmwork

gar-i-do--

di-ekogive-_..

chube..

“I have done this work (for someone).”A situation which persists marginally in modern Hindi, in much the same way as the modern English simple present can be employed with

a progressive sense in certain contexts, e.g. “I am aaching a document to this email. . .” vs. “I aach a document to this email. . .”

() a. maI

mandir-mātemple-

ga-i-go--

rah-ekoremain-_..

chube..

“I am going to the temple.” / “I have been going to the temple.”

b. maI

yothis

kāmwork

gar-i-do--

rah-ekoremain-_..

chube..

“I am doing this work.” / “I have been doing this work.”

() a. maI

mandir-mātemple-

jāṁ-daigo-_

chube..

“I am going to the temple.”

b. maI

yothis

kāmwork

gar-daido-_

chube..

“I am doing this work.”

In contrast to the case of normal CV constructions (like those shown in ()), constructions inwhich rahanu functionsas a light verb do not require agentive marking on the subject (even when the main verb is transitive), see (). Inthis, the CV construction with rahanu closely resembles the synthetic imperfect construction, as shown in ().

e Nepali light verb rahanu appears thus to be in the process of becoming an auxiliary like Hindi rahnā:— seethe Hindi examples in various tenses in ().

() a. maiṁI

calmove.

rahāremain._..

hūṁbe..

“I am going.”

b. maiṁI

calmove.

rahāremain._..

thābe..

“I was going.”

Likewise, in Nepali the construction absolutive particle+rahanu can be used in various tenses, a subset of whichare shown in ().

() a. maI

cal-i-move--

rahekoremain._..

chube..

“I am going.”

b. maI

cal-i-move--

rahekoremain._..

thieṁbe..

“I was going.”

c. maI

cal-i-move--

rahekoremain._..

huṁlābe..

“I (probably) shall be going.”

However, there are differences between the Hindi rahnā and Nepali rahanu constructions. Hindi rahnā can beused as an auxiliary in the present tense with future connotations, as in ().

() kaltomorrow

maiṁI

gharhome

jāgo.

rahāremain._..

hūṁbe..

“I’m going home tomorrow.”

In Nepali this is not the case. While the synthetic imperfective construction can be used in the present tense withough, as an anonymous reviewer points out, at this point Nepali rahanu may be best analysed as an atypical light verb.See Mahews (: -) for more examples.

future reference, as in (-a), the periphrastic absolutive+rahanu—which is formally similar to the Hindi constructionin ()—cannot, see (-b).

() a. bholitomorrow

maI

gharahome

jāṁ-daigo-.

chube..

“I am going home tomorrow.”

b. *bholitomorrow

maI

gharahome

ga-i-go--

rahekoremain-...

chube..

us Nepali rahanu has not acquired all of the properties of an auxiliary like Hindi rahnā since it cannot be used infuture contexts, as can imperfects in -dai (Sharma : –). Further, it has not so far supplanted the synthetic-dai imperfective construction. Finally, rahanu exhibits some properties prototypical of light verbs used in CV con-structions, some prototypical of auxiliaries—and therefore appears to represent a stage intermediate between lightverb and auxiliary.

Hindi rahnā combined with bare verb stems (morphologically identical to a CV construction) clearly functionsas an auxiliary in present day Hindi (see () above), but just as clearly acted more as a light verb as late as theth century, as discussed above. Present day Nepali rahanu, in its use in structures like those shown in () above,represents a stage intermediate between light verb and auxiliary.

. Development of Hindi simple present

e Hindi auxiliary rahnā was shown to have evolved from a light verb, and the Nepali light verbs rahanu/rakhnuappear to be in the process of becoming auxiliaries. In fact, even the auxiliary honā “to be” as it appears in themodern Hindi simple present, e.g. (), can be shown to have evolved in a fashion parallel to that of the auxiliaryrahnā. Both of these changes appear to have involved a morphosyntactic chain shi (discussed further below)—thusrepresenting not a gradual clinelike grammaticalisation process, but rather a structurally-induced change.

() jā-tāgo-_..

haibe.

“goes”

Hindi honā “to be” and its finite forms including hai “is” derive from earlier Sanskrit bhū- “become/be/exist” (e.g.bhavati “becomes”). One might suppose that this auxiliary use of honā developed directly from the full verb honā“to be”. But even in this case we find that hai in the simple present construction, (), derives from an earlier lightverb construction.

In section . above, it was noted that forms like () were earlier (into the th-c.) present progressives, and thatthe modern Hindi present progressives, (), were recruited from a peripheral construction involving a light verb useof rahnā. e construction in (), with its earlier present progressive value, derives from a structure represented inClassical Sanskrit by the type shown in () (cp. the simple present in ()), as discussed above in Section ..

() calatimove..“goes; is going”

() calanmove..

bhavatibe/become..

//astibe/become..

“keeps on going”

However, recall from above (see example ()), that in Sanskrit, bhū- (e.g. sg bhavati) was only one of a numberof verbs which could appear in this construction. Other verbs which could appear in this peripheral constructioninclude ās- “sit” and sthā- “stand”—whose (full verb) meanings are those typical of light verbs.

In later IA (Apabhraṁśa), the construction present participle + “be” (cp. ()) became grammatialised as a pro-gressive present, with the old simple present tending to be used as a non-continuative (though it is still possible incontinuative contexts as well), as shown in () (see Singh : ).

() a. calaïgo..“goes; (is going)”

b. calantugo._..

acchaïbe..

“is going”

In Hindi, the value of the earlier simple present (Skt. (), Apa. (-a)) shis from present to modal, leaving theperiphrastic progressive present of Apabhraṁśa (-b)—itself a grammaticalisation of an earlier peripheral continu-ative/durative construction, see examples like ()—as the new unmarked present indicative. is is the situationin th-century Hindi. By the th century, the Apabhraṁśa shi repeats, with the grammaticalisation of the peri-pheral continuative absolutive + rah- construction as the new progressive present. ese changes are summarisedin Table .

C G PModal Simple Present Present Progressive Continuative

Stage A calan tiṣṭhati(Sanskrit) calati calati calan bhavati

↓ ↙ etc.Stage I cale caltā hai caltā rahtā hai(early Hindi) ↙ ↙ ↓ cal rahā haiStage II caltā rahtā hai(th-c. Hindi) cale caltā hai caltā hai cal rahā hai

↓ ↙Stage III cale caltā hai cal rahā hai caltā rahtā hai(modern Hindi)

Table : Morphosyntactic Chain Shi in Hindi (adapted from Hock )

is ‘morphosyntactic chain shi’ resulted, ultimately, from the old simple present (cale-type) taking on modalfunctions, leaving a ‘gap’ in the paradigm which was filled by the present progressive (caltā hai-type)—which thenbore two values (simple present and present progressive). is ambiguity is resolved with the reanalysis of the oldperipheral continuative construction of the type cal rahā hai as the (core-grammar) present progressive, as describedabove. For more on the this morphological chain shi in Hindi, see Hock (with reference to Bloch ).

As diagrammed in Table , the change between Stages II and III is essentially a repetition of the change whichoccurred between Stages A and I. In both cases, a peripheral construction, involving a light verb, is recruited intothe core grammar, at which point the light verb is reanalysed as an auxiliary.

. Conclusions

I have noted several examples of light verbs becoming auxiliaries in IA: (a) the reanalysis of the light verb (Skt. as-/bhū-; Apa. acch- “be”) of the durative/continuitive present participle+light verb construction as the auxiliary usedin the progressive tenses; (b) the reanalysis of the early Hindi light verb rah- in CV constructions as a progressive-marking auxiliary; (c) the ongoing reanalysis of the Nepali light verbs raha-, rakh- as progressive auxiliaries.

An anonymous reviewer forDiachronica objects that these changes could simply involve the light verbs undergoing a lexical change, whereasBowern’s () claim applies specifically to constructions. However, it is clear that these changes did in fact involve reanalysis of the light verbsin the context of particular constructions, as discussed in detail above.e same reviewer also proposes that the Nepali change (c) might be due to the influence of Hindi and thus be exempt from Bowern’s general-

isation. ere is no evidence which would suggest that ongoing changes raha- is experiencing are motivated by contact with Hindi; furthermore,the development of rakh- towards becoming a progressive auxiliary in unparalleled in Hindi. Hindi does possess a homophonous verb rakh-,which, in its main verb usages, is largely synonymous with the main verb uses of Nepali rakh-. Hindi rakh- is in fact also employed as a light verbin CV constructions (mostly contributing a sense that the action was completed some time ago, oen with a certain purpose in mind relating toa later event; see Nespital : –), but is not employed in any progressive-auxiliary-like function.

Summary of findings and implications

. Grammaticalisation and the light verb/auxiliary contrast

Bu and Lahiri (: -) use three major criteria to distinguish between light verbs and auxiliaries in IA: ()(non)-interchangeability, () interruptibility, () ability to undergo reduplication.

With regard to the first criterion, they mean simply that light (vector) verbs like jā- and auxiliaries like rah- arenot grammatical in the same set of syntactic contexts. is to me seems a poor measure, since even in the case ofelements which are clearly auxiliaries, different auxiliaries are grammatical in different syntactic contexts, e.g. theEnglish auxiliaries have and be have distinct distributions.

e property of interruptibility refers to whether other elements may intervene in a V1+V2 collocation, see aboveSection ... Bu and Lahiri () observe that, in Hindi, a main verb can be separated from a light verb, but theyclaim that main verb may not be separated from its accompanying auxiliary and therefore that sentences like ()are ungrammatical. However, my Hindi consultants in fact accept examples like (-b) as grammatical—this possiblyreflects dialectal differences between the grammars of my consultants and Bu’s Hindi/Urdu.

() a. sosleep.

to

baccāchild

gayāgo._..

haibe..

“e child has gone to sleep.”

b. %sosleep.

to

baccāchild

rahā..

haibe..

“e child is sleeping.”

Finally, they observe that light (vector) verbs, but not auxiliaries can undergo reduplication (see Bu and Lahiri : for details).

If we try to apply these tests to the whole range of ‘light verbs’ constructions in Hindi, the distinction betweenlight verb and auxiliary becomes even murkier. Recall that in Hindi we also find various peripheral light verb con-structions, including imperfect participle+rah-/jā- and perfect participle+kar- (see above, Hindi (), (), (), ();and (ii) in fn., respectively). ese elements are interruptible, as shown by ():

() a. so-tāsleep-_..

to

baccāchild

rahāremain._..

“e child kept sleeping.”

b. āy-ācome-_..

to

laṛkīgirl

kar-tīdo-_..

thībe...

. . .

. . .

“e girl used to come frequently . . .”

But the light verb construction perfect participle+kar- does not behave as we would expect it to with respect tothe property of dominance (see Section .. above). at is, we would expect that, like other transitive light verbs(i.e. like Hindi vector verbs), it should require agentive marking on the subject in perfective tenses—but it does not,see example (). Rather, like the Hindi progressive present (absolutive+rahā), this construction always behaves asif the verb were in a non-perfective tense, for purposes of case-assignment.

() a. tereyour

pūrvajancestors..

ghāsgrass

īl-āscrape._..

ki-edo._.

“Your ancestors scraped grass! (i.e. did menial work)” (from McGregor : )

b. *tereyour

pūrvajoṁancestors...

-ne-

ghāsgrass..

īl-āscrape-_..

kīdo._.

e situation becomes even more complicated once we include in our consideration Nepali CVs and other lightverb constructions. Table summarises the relevant properties of seven types of ‘auxiliary’ and ‘light verb’ construc-

One of my anonymous reviewers confirms that (-b) is grammatical.

tions in Hindi and Nepali. emorphosyntactic structure of each type is given, followed by an example, the semanticcontribution of the auxiliary/light verb, and two further properties. ese properties are interruptibility—whetheror not other elements may intervene between the V1 and V2—and dominance—which member of the collocationdetermines the overall transitivity for the purposes of case-marking (see Sections .. and .., respectively). epossible values for this final property are: V1 (=transitivity is determined based on the transitivity of V1), V2 (=trans-itivity is determined based on the transitivity of V2), and ‘—’ (=construction is always treated as a non-perfectivetense).,

Type Example Semantics Interrupt. DominanceIa. (Hindi) cal rahā hai progressive yes — + _ of rah- “is going”Ib. (Bu’s Hindi/Urdu) cal rahā hai progressive no — + _ of rah- “is going”II. (Nepali) cali-raheko cha progressive no — + _ of raha- “is going”/

“has been going”III. (Hindi) caltā rahtā hai continuative yes —_ + rah- “keeps going” (durative)IV. (Nepali) cali-rahancha continuative no — + finite form of raha- “keeps going” (durative)V. (Hindi) calā kartā hai continuative yes —_ + kar- “usually goes” (habitual)VI. (Nepali) cali-diyo perfective/ no V1

+ vector “went” subjectiveVII. (Hindi) cal gayā perfective/ yes V2

+ vector “succeeded”; subjective“went off (of a gun)”

Table : Classification of selected Hindi and Nepali ‘light verbs’ & ‘auxiliaries’

Based on the data summarised in Table , how are we to classify these types as light verbs or auxiliaries? Bu andLahiri () consider only the contrast between types I and VII, categories which are perhaps disjoint in features (ifwe accept Bu’s judgement that such sequences are non-interruptible). Considering a fuller range of constructions,we find considerable intertype overlap of properties.

We cannot distinguish these grammatical elements on the basis of the morphological category of the V1, sincetypes both at the top and boom of the table use an absolutive as the V1—and, in fact, all of the Nepali constructionsinvolve an absolutive as V1.

Interruptibility certainly does not serve to distinguish ‘light verb’ from ‘auxiliary’, since none of the Nepaliconstructions are interruptible, and Bu’s claimed non-interruptibility of type I elements is questionable (or, at least,subject to dialectal variation). Dominance fares no beer, since it is irrelevant for types I–V (as they are treated asnon-perfective), and types VI and VII—which we would like to classify together, as they are both CV constructions—differ on which member controls transitivity.

In sum, I can see no single formal property or set of properties which suffices to distinguish ‘light verbs’ from‘auxiliaries’ in both Hindi and Nepali, despite the great overall grammatical and lexical similarities of these two

Recall that in both Hindi and Nepali agentive case-marking on the agent of a transitive verb is only required in perfective tenses.‘Subjective’ refers to Tikkanen’s () ‘subjective aspect’, Abbi & Gopalakrishnan’s () ‘aitudinal meaning’, and includes the addition

of shades of meaning referring to the role or perspective of the speaker, e.g. Hindi light verbs de- “other-directedness”, baiṭh- “regret”.I label this type as perfective/subjective for the sake of precision, for—though CVs usually have a perfective sense as shown by Hook (,

)—Bu and Lahiri (: ) observe that CVs can occur where perfectivity is excluded, as in (i) below.

(i) marīamMiriam.

īmelemail.

likhwrite.

mārhit

rahī..

thībe...

jabwhen

vīlīWilli.

kamreroom..

meṁin

āy-ācome._.

“Miriam was dashing off an e-mail when Willi came into the room.”

But see Section .. above for complications involving transitive V1 and intransitive V2.

languages. At best we might divide the types of Table into three sets: () types I-II, progressives; () typesIII-V, continuatives; () types VI-VII, CV constructions, with perfective/subjective senses—but note that this is asemantic/pragmatic classification, not a formal one. And even this classification involves a degree of overlap: typesIII-V include different types of continuatives (durative vs. habitual), and type II, unlike type I, cannot be used withfuture reference (see examples (), () above).

ough it may be tempting to view the data in Table as evidence for a grammaticalisation continuum—with typeI being themost auxiliary-like and fully grammaticalised and typeVII themost light verb-like and least grammaticalised—themultiply overlappingmorphosyntactic properties and semantic/pragmatic functions suggests tome that the typesin Table simply represent grammaticalisations. Nepali CVs (type VI) developed in such a way that theyare non-interruptible, unlike Hindi CVs (type VII); while non-interruptibility certainly seems to be a property associ-ated with more grammaticalised elements, I would not want to claim that Nepali CVs are more grammaticalised than,say, the Hindi durative construction caltā rahtā hai (type II), which is interruptible. With respect to semantic bleach-ing and obligatoriness, Nepali CVs have the same status as Hindi CVs, though they display different morphosyntacticproperties.

Grammaticalisation appears then not to be a one-dimensional process—rather linguistic elements may undergogrammaticalisation along various axes, and an element grammaticalised with respect to certain features is not neces-sarily grammaticalised with respect to others. us, a one-dimensional representation of grammaticalisation, likethe cline shown above in (), repeated below as (), may be misleading.

() (a) parataxis > (b) hypotaxis > (c) serialisation > (d) light verb > (e) auxiliary verb > () univerbatedaffix (Andersen )

For instance, based on certain criteria (i.e. the four properties discussed in Section ., including non-interruptibility),Nepali vector verbs appear to be univerbated affixes; but semantically and pragmatically they behave like Hindivectors (which based on the same criteria, are definitely not univerbated affixes), and therefore do not have thesame status as, for instance, an obligatory progressive aspect affix (e.g. type I). Another ‘grammaticalisation axis’ wemight consider is the continuum between core-grammar and peripheral constructions (see Section .), a distinctionreferred to throughout this study. As mentioned earlier, types II-V represent peripheral constructions, and typesI, VI, VII represent core-grammar constructions—a classification which cuts across many of the other distinctionsdiscussed above. Peripheral constructions can be considered less grammaticalised in the sense that they are less fullyintegrated into the grammar. us, from this perspective, types I, VI, VII are more fully grammaticalised than typesII-V, despite the fact that these two groupings are heterogeneous with respect to other properties considered.

Viewing grammaticalisation in this multidimensional way offers an explanation of why it is difficult to dividethe types in Table into two discrete categories of ‘light verb’ and ‘auxiliary’. It might be useful in a synchronicdescription—of a particular language—to distinguish between ‘light verbs’ and ‘auxiliaries’, so long as these areunderstood to be prototypical categories and not universal primitives. However, it may be more useful to considerthese types of grammaticalised verbal elements in terms of clusters of properties—clusters which turn out, at least inthe case of Nepali and Hindi, to involve a large degree of (multiply) overlapping properties, as diagrammed in Table.

In terms of such clusters of properties, the Nepali grammaticalised verbal elements are less distinguished fromone another in terms of formal properties—all are constructed from an absolutive V1 and none are interruptible;whereas the Hindi grammaticalised verbal elements exhibit less overlap of properties, showing variation both in themorphology of the V1 and with respect to interruptibility, and thus can be more easily divided into notional discretecategories.

. Conclusion

In this paper I have investigated the origin and development of the IA CV construction. ough Bu and Lahiri() have pointed to the existence of grammaticalised verb-verb constructions in Old IA, a closer examination of

e other two properties discussed in Section ., namely restrictions on recursion and construction-specific restrictions, were we to extendthem to other light verb/auxiliary constructions, would only serve to distinguish type VI (which can occur in conjunctive participles and canundergo one level of recursion) from all other types.

Other misleading aspects of the representation in () include the implicit representation of unidirectionality; for critical discussion of thisand other aspects of grammaticalisation theory, see van Gelderen , Newmeyer , Lass , Janda , Joseph , Lightfoot , ,Joseph , .

these constructions reveals that they are formally and functionally distinct from the CV constructions of modern IA.e earliest unequivocal examples of CVs are found in early Modern IA (th-th c. A.D.). An examination of twoclosely related modern IA languages, Hindi and Nepali, reveals that numerous morphosyntactic and lexical changeshave taken place in these languages, suggesting that light verbs, like all other elements, are not historically stablebut rather are subject to reanalysis and change (pace Bu and Lahiri () and Bu ()), including reanalysis asauxiliaries. Finally, Indo-Aryan fills in the data gap pointed to in Bowern (), as such changes include reanalysisof light verbs as auxiliaries.

e reanalysis of the descendant of the Sanskrit light verb bhu- “be” as an aspectual auxiliary in early Hindi, aswell as the parallel reanalysis of the Hindi light verb raha- as an aspectual auxiliary are both noteworthy from thestandpoint that they do not take place via grammaticalisation along a cline, but instead as part of larger, structurally-motivated morphosyntactic chain shis. Such changes highlight the importance of looking beyond the grammatic-alisationist paradigm when evaluating historical changes.

Much more work remains to be done on IA CVs. e most complete collections of data and examinations ofthe properties of IA CVs are those of Hindi (e.g. Nespital’s () dictionary, Hook’s () monograph). A morecomplete picture requires much more collection and analysis of data from other IA languages.

ReferencesAbbi, Anvita, and Devi Gopalakrishnan. . Semantics of explicator compound verbs in South Asian languages.Language Sciences :–.

Andersen, Gregory. . Auxiliary verb constructions. Oxford: Oxford University Press.

Arora, Harbir K. . Syntactic convergence: e case of Dakkhini Hindi-Urdu. Delhi: Publications Division, Univer-sity of Delhi.

Bandhu, Vishva, ed. -. Ṛgveda with the padapātha and commentaries of Skandasvāmin, Udgītha, Veṅkatamād-hava and Mudgala. vols. Hoshiarpur (Panjab, India): Vishveshvaranand Vedic Research Institute.

Bloch, Jules. . La formation de la langue marathe. Paris: Honoré Champion.

Bowern, Claire. . e diachrony of complex predicates. Diachronica :–.

Bubenik, Vit. . A historical syntax of late Middle Indo-Aryan. Amsterdam: John Benjamins.

Bu, Miriam. . e light verb jungle. Harvard Working Papers in Linguistics :–.

Bu, Miriam. . e light verb jungle: still hacking away. In Complex predicates: cross-linguistic perspectives onevent structure, ed. Mengistu Amberber, Bre Baker, and Mark Harvey, –. Cambridge: Cambridge UniversityPress.

Bu, Miriam, and Wilhelm Geuder. . On the (semi)lexical status of light verbs. In Semi-lexical categories: On thecontent of function words and the function of content words, ed. Norbert Corver and Henk van Riemsdijk, –.Berlin: Mouton de Gruyter.

Bu, Miriam, and Wilhelm Geuder. . Light verbs in Urdu and grammaticalization. InWords in time: diachronicsemantics from different points of view, ed. Klaus von Heusinger, Regine Eckardt, and Christoph Schwarze, –.Berlin: Mouton de Gruyter.

Bu, Miriam, and Aditi Lahiri. . Historical stability vs. historical change. Ms., Universität Konstanz.Hp://ling.uni-konstanz.de/pages/home/bu/stability.pdf .

Bu, Miriam, and Gillian Ramchand. . Complex aspectual structure in Hindi/Urdu. Oxford Working Papers inLinguistics, Philology and Phonetics :–.

Cappelle, Bert. . Keep and keep on compared. Leuvense Bijdragen :–.A careful consideration of Apabhraṁśa data is also highly desirable, given the possible examples of Apabhraṁśa CVs cited by Singh (),

Bubenik (), and others.

Caell, Ray. . Composite predicates in English. North Ryde, New South Wales: Academic Press Australia.

Chaerjee, Suniti Kumar. . e origin and development of the Bengali language. Calcua: Calcua UniversityPress. [revised, ; reprinted, New Delhi: Rupa, ].

Delbrück, Berthold. . Altindische Syntax. Halle: Verlag der Buchhandlung des Waisenhauses.

van Gelderen, Elly. . Verbal agreement and the grammar behind its ‘breakdown’: Minimalist feature checking.Tübingen: Max Niemeyer.

Graßmann, Hermann. . Wörterbuch zum Rig-Veda. Wiesbaden: Harrassowitz.

Grimshaw, Jane, and Armin Mester. . Light verbs and θ-marking. Linguistic Inquiry :–.

Halle, Morris, and Alec Marantz. . Distributed morphology and the pieces of inflection. Ine view from Building: Essays in linguistics in honor of Sylvain Bromberger, ed. K. Hale and S. Keyser, –. Cambridge, MA: MITPress.

Hendriksen, Hans. . Syntax of the infinite verb forms of Pāli. Copenhagen: Einar Munksgaard.

Hock, Hans Henrich. . Review of Tikkanen (): e Sanskrit gerund: a synchronic, diachronic, and typolo-gical analysis. Kratylos :–.

Hock, Hans Henrich. . A morphosyntactic chain shi in Hindi. Paper given at the South Asian LinguisticAnalysis (SALA)- Conference, University of Wisconsin, Madison.

Hook, Peter Edwin. . e compound verb in Hindi. Ann Arbor: Center for South and Southeast Asian Studies,University of Michigan.

Hook, Peter Edwin. . e distribution of the compound verb in the languages of North India and the questionof its origin. International Journal of Dravidian Languages :–.

Hook, Peter Edwin. . e emergence of perfective aspect in Indo-Aryan languages. In Approaches to grammat-icalization, ed. Elizabeth Traugo and B. Heine, –. Amsterdam: John Benjamins.

Hook, Peter Edwin. . Aspectogenesis and the compound verb in Indo-Aryan. In Complex predicates in SouthAsian languages, ed. Manindra K. Verma, –. New Delhi: Manohar.

Hook, Peter Edwin, and Prashant Pardeshi. . Are vector verbs eternal? Paper given at the South Asian LinguisticAnalysis (SALA)- Conference, University of Illinois, Urbana/Champaign, - September, .

Hopper, Paul, and Elizabeth C. Traugo. . Grammaticalization. Cambridge: Cambridge University Press.

Janda, Richard D. . Beyond ‘pathways’ and ‘unidirectionality’. On the discontinuity of transmission and thecounterability of grammaticalization. Language Sciences :–.

Jespersen, Oo. . A modern English grammar on historical principles. London and Copenhagen: George Allen &Unwin and Ejnar Munksgaard.

Joseph, Brian D. . Is there such a thing as grammaticalization? Language Sciences :–.

Joseph, Brian D. . Rescuing traditional (historical) linguistics from Grammaticalization ‘eory’. InUp and downthe cline, ed. Olga Fischer, Muriel Norde, and Harry Perridon, –. Amsterdam/Philadelphia: John Benjamins.

Joseph, Brian D. . How accomodating of change in grammaticalization? e case of ‘lateral shis’. Logos andLanguage :–.

Kachru, Yamuna, and Rajeshwari Pandharipande. . Toward a typology of compound verbs in South Asianlanguages. Studies in the Linguistic Sciences :–.

Kellogg, S.H. . A grammar of the Hindi language, in which are treated the High Hindi, Braj, and the Eastern Hindiof the Ramayana of Tulsidas, also the colloquial, dialects of Rajputana, Kumaon, Avadh, Riwa, Bhojpur, Magadha,Maithila, etc. London: Paul, Trench, Trubner & Co., nd edition. [reprinted, Delhi: Munshiram Manoharlal, ].

Lass, Roger. . Remarks on (uni)directionality. In Pathways of change, ed. Olga Fischer, Anee Rosenbach, andDieter Stein, –. Amsterdam/Philadelphia: John Benjamins.

Li, Chao. . Split ergativity and split intransitivity in Nepali. Lingua :–.

Lightfoot, David W. . Myths and the prehistory of grammars. Journal of Linguistics :–.

Lightfoot, DavidW. . Grammaticalisation: Cause or effect? InMotives for language change, ed. Raymond Hickey,–. Cambridge: Cambridge University Press.

Masica, Colin P. . Defining a linguistic area: South Asia. Chicago: University of Chicago Press.

Masica, Colin P. . e Indo-Aryan languages. Cambridge: Cambridge University Press.

Mahews, David. . A course in Nepali. London: Curzon.

McGregor, R.S. . e language of Indrajit of Orchā: A study of early Braj Bhāṣā prose. Cambridge: CambridgeUniversity Press.

McGregor, R.S. . Outline of Hindi grammar . New Delhi: Oxford University Press, rd Indian edition.

Nespital, Helmut. . Hindī kriyā-koś/Dictionary of Hindi verbs. Allahabad: Lokbharati.

Newmeyer, Frederick. . Language form and language function. Cambridge, MA: MIT Press.

Paranavitana, Senarat. . Sigiri graffiti, Sinhalese verses of the eighth, ninth and tenth centuries. London: OxfordUniversity Press.

Peterson, John. . e Nepali converbs: A holistic approach. In Yearbook of South Asian languages and linguistics,ed. Rajendra Singh, –. New Delhi/ousand Oaks/London: Sage Publications.

Pokharel, Balkrishna, ed. . Pañc śaya varṣa. Kathmandu: Sajha Prakashan.

Pokharel, Madhav P. . Compound verbs in Nepali. Contributions to Nepalese Studies :–. [reprinted inTopics in Nepalese Linguistics, ed. Yogendra P. Yadava and Warren W. Glovers, -. Kathmandu: Royal NepalAcademy, ].

Poudel, Tikaram. . Tense, aspect and modality in Nepali and Manipuri. Doctoral Dissertation, Tribhuvan Uni-versity, Kathmandu.

Sharma, Tara Nath. . e auxiliary in Nepali. Doctoral Dissertation, University of Wisconsin, Madison.

Singh, Ram Adhar. . Syntax of Apabhraṁśa. Calcua: Simant Publications India.

Singh, U.N., K.V. Subbarao, and S.K. Bandyopadhyay. . Classification of polar verbs in selected South Asianlanguages. In South Asian languages: Structure, convergence and diglossia, ed. Bh. Krishnamurti, –. Delhi:Motilal Banarsidass.

Speijer, J.S. . Sanskrit syntax. Leiden: E.J. Brill. [Reprinted, Delhi: Motilal Banarsidass, ].

Tikkanen, Bertil. . e Sanskrit gerund: A synchronic, diachronic and typological analysis. Helsinki: FinnishOriental Society.

Traugo, Elizabeth Closs. . A historical overview of complex predicates. In Collocational and idiomatic aspectsof composite predicates in the history of English, ed. Laurel J. Brinton and Minoji Akimoto, –. Amsterdam:John Benjamins.

Turner, Ralph Lilley. . A comparative and etymological dictionary of the Nepali language. London: Routledge andKegan Paul, Ltd. [reprinted, New Delhi: Allied Publishers, Ltd., ].

Verma, Manindra K., ed. . Complex predicates in South Asian languages. New Delhi: Manohar.

Verma, Manindra K., and Tara Nath Sharma. . Intermediate Nepali structure. New Delhi: Manohar.

Whitney, William Dwight. /. Sanskrit grammar, including both the classical language and the older dialectsof Veda and Brahmana. Leipzig: Breitkopf and Härtel, st/nd edition. [reprinted, Delhi: DK Publications, ].

Last modified: -- :: - — Git Hash: efcdcfaaedab