Ergativity in Indo-Aryan

Miriam Butt and Ashwini Deo
Universität Konstanz, Stanford University


Many of the Indo-Aryan languages are characterized by morphological ergativity. This contribution briefly surveys the types of ergative patterns and the particular case markers that are employed in the Indo-Aryan languages of South Asia. The synchronic discussion is followed by a look at what has been proposed for the historical development of ergativity in these languages.  

1. Ergative Patterns in Indo-Aryan

The ergative marker was first named as a special marker for subjects with reference to Caucausian languages such as Georgian (Dirr 1928). The same type of case marker had been noted for languages such as Basque and Greenlandic (Pott 1873), but was generally referred to as an ``agentive nominative'' in opposition to a ``neutral nominative'', i.e. what we simply call nominative or absolutive today. The semantic parameter of agentivity that had been noted consistently by the linguists of the last century in connection with the ergative has been replaced by a purely structural division in this century.

The standard formulation of the conception of ergativity goes back to Fillmore (1968). Plank (1979:4) concisely summarizes the basic idea as follows:

a. A grammatical pattern or process shows ergative alignment if it identifies intransitive subjects (Si) and transitive direct objects (dO) as opposed to transitive subjects (St).

b.It shows accusative alignment if it identifies Si and St as opposed to dO.

According to this idea, languages can be grouped into several types, based on the case marking displayed by subjects and objects. This is illustrated by the following table.

Clause TypeLanguage Type
 ErgativeAccusative Active
Transitive Erg-NomNom-Acc Erg-Nom
Intransitive (Unaccusative) Nom Nom Nom
Intransitive (Unergative) Nom Nom Erg

As far as we know, all the Indo-Aryan languages which have an ergative case fall under the "active" type of language. We exemplify this on the basis of Urdu/Hindi. (Hindi is spoken in the northern part of India and is written with the Devanagari script. Urdu is spoken in Pakistan and the northern part of India and is written with the Arabic script. Despite these and some differences in vocabulary, the two languages are structurally identical.) (1) shows a transitive clause with an ergative, (2) an unaccusative intransitive where no ergative is possible, and (3) shows an unergative intransitive where the ergative is optional and generally signals a degree of agentivity.
(1)ram=ne gari cala-yi (hai)
 Ram.M.Sg=Erg car.F.Sg.Nom drive-Perf.F.Sg be.Pres.3.Sg
 `Ram has driven a/the car.'(Urdu/Hindi)
(2)ram/*ram=ne ga-ya
 Ram.M.Nom/Ram.M=Erg go-Perf.M.Sg
 `Ram went.'(Urdu/Hindi)
(3a)ram kHaNs-a
 Ram.M.Nom cough-Perf.M.Sg
 `Ram coughed.' (Urdu/Hindi)
(3b)ram=ne kHaNs-a
 Ram.M=Erg cough-Perf.M.Sg
 `Ram coughed (purposefully).' (Urdu/Hindi)

None of the Indo-Aryan languages exhibit "syntactic ergativity". Syntactic ergativity is found in some languages, such as the Australian language Dyirbal (Dixon 1994). Syntactically ergative languages encode the ergative pattern purely structurally. However, most languages are morphologically ergative in that pieces of the morphology serve to mark the ergative or active pattern. The Indo-Aryan languages fall under this class.

Agreement patterns vary from language to language. In Urdu/Hindi the verb can only agree with NPs which are direct arguments and which do not bear overt case marking. While most Indo-Aryan languages appear to follow this pattern by disallowing agreement with a non-nominative argument, some languages allow it. Nepali, for instance allows agreement with an ergative subject, while in Gujarati, the verb agrees with the direct object, whether accusative marked or unmarked, when the subject of the clause is ergative. Examples (4) and (5) illustrate the Nepali and Gujarati patterns, respectively.
(4)mai-le mero lugga dho-en
 I=Erg I.M.Sg.Gen clothes.M.Pl.Nom wash-Perf.1.Sg
 `I washed my clothes.'(Nepali)
(5)ram-e gadi=ne jo-yi
 Ram.M.Sg=Erg car.F.Sg=Acc see-Perf.F.Sg
 `Ram has seen a/the car.'(Gujarati)

Ergative morphology in most Indo-Aryan languages furthermore shows a split along the lines of tense/aspect. We again illustrate this with an example from Urdu/Hindi. Here the ergative case marker ne is required by perfect verb morphology. The association of ergativity with perfect morphology is crosslinguistically well-established and is one of the factors that has contributed to the idea that ergative structures must arise out of passive constructions (see section below). However, not all Indo-Aryan languages are split-ergative. Assamese, for example, does not seem to exhibit such a split (Devi 1986).
(6)ram gari cala-ta (hai)
 Ram.M.Sg.Nom car.F.Sg.Nom drive-Impf.M.Sg be.Pres.3.Sg
 `Ram drives a car.'(Urdu/Hindi)
(7)ram=ne gari cala-yi (hai)
 Ram.M.Sg=Erg car.F.Sg.Nom drive-Perf.F.Sg be.Pres.3.Sg
 `Ram has driven a/the car.'(Urdu/Hindi)

Another very common split crosslinguistically is the so-called NP-split, whereby only a subset of the nominals may display ergative morphology. Urdu/Hindi does not display this kind of a split, but it can be found in the closely related language of Punjabi, for example. In Punjabi, the first and second person pronouns are not marked for ergativity, whereas third persons are (pronouns as well as nominals).
 `I did some/the work.'(Punjabi, based on Bhatia 1993)
 `He/She did some/the work.'(Punjabi, based on Bhatia 1993)

Not all Indo-Aryan languages show ergative patterning. Historically, the construction which the ergative pattern is based on was inherited by all the modern languages from the parent language, Sanskrit. However, not all the modern languages have retained this pattern. Notably, a difference may be observed in the Western and the Eastern vernaculars, where the Western subgroup consisting of languages such as Urdu/Hindi, Marathi and Gujarati retain the ergative marking pattern, while the Eastern subgroup, consisting of languages such as Bengali, Oriya and some dialects of Eastern Hindi have lost this pattern and display a uniform accusative pattern in all tenses and aspects. Older variants of these languages, Old Bengali, for instance, still show ergative patterning in the perfect aspect. This pattern is lost in the modern language, as illustrated in example (9), which documents the absence of ergative morphology in the modern Bengali perfect aspect.
(9)ami sita=ke dekh-lam
 I-Sg.Nom Sita.F.Sg.Acc see-Perf.1.Sg
 `I saw Sita.'(Bengali)

2. The Ergative Case Markers

Ergative case markers in Indo-Aryan take a variety of forms. Here we show a selection and compare the ergative markers to dative markers of the same languages in order to show that some of the forms appear to have a marked similarity (this preshadows some of the discussion in the next section).

  Dative Ergative
  (subjects and objects) (subjects only)
Hindi/Urdu ko ne
Punjabi nuN ne
Gujarati ne -e
Marathi la ne/ni
Bengali ke NONE
Oriya ku NONE
Assamese ko/no -e
Nepali lai le
Some of the ergative markers are inflectional. These are prefixed with an "-". All the others presumably function as clitics, as in Urdu/Hindi (Butt and King 2001, Sharma 1999).  

3. The Origin of Ergative Case

There are several theories as to the origin of the ergative in Indo-Aryan. Most of the discussions take Urdu/Hindi as a representative language. In this section, we give a short overview of the proposals, along with some commentary.

Ergative as Passive The early (Western) linguistic literature on South-Asian languages (18th-19th century) refers to the ergative alternatively as an agentive or instrumental. Because the ergative in many languages has connotations of agency and shares features with an instrumental, the ergative construction was first analyzed as a passive in many languages (see Trask 1979:390 for some discussion). However, this view soon became a minority view due to detailed language-specific work, which showed that more often than not, ergatives were subjects of active sentences.

Passive/Participle to Ergative With respect to language change, the connection to a passive forms the basis for a hypothesis that ergative constructions arise from former passive constructions via a reanalysis of the following type:

NPinstr NPnom V > NPerg NPnom V (adapted from Garrett 1990:265)

The precise morphology involved on the verb was a -ta participle in Sanskrit which has either been lost or retained as a glide or an -e in most of the modern Indo-Aryan languages. The Sanskrit -ta participle finds its origin in the Proto Indo-European deverbal adjective in *-to-. In Sanskrit, the -ta formed a deverbal participle which agreed with a noun. This participle had passive interpretation with transitive verbs but active interpretation with intransitives and verbs of motion (Garrett 1990:263, Speijer 1886:280). It is indubitably the case that the modern ergative patterns occur primarily in conjunction with the verbal morphology descended from the original Sanskrit -ta. However, the precise nature of the original Sanskrit participle and its modern descendents remain the subject of debate.

Despite the possible active interpretations of the participle, the dominant idea for the for the development of ergativity in modern Indo-Aryan languages is one which sees a passive construction as being reanalyzed as an ergative. It is this proposal which has become accepted as common wisdom, despite many dissenting voices (e.g., Beames 1872, Kellogg 1893, Klaiman 1978, Zakharyin 1979, Andersen 1986, Hock 1986). Consider, for example, the quote from Dixon (1994) where this hypothesis is presented as textbook knowledge (also see Harris and Campbell (1995:263)).

We might thus expect a split ergative system conditioned by aspect or tense, where the ergative is found in perfect aspect or past tense, to be likely to have a passive origin.

This is precisely what happened in the Indic and Iranian branches of Indo-European (for which we do have written records and can be fairly certain about what happened, although different scholars have suggested diverse interpretations). [Dixon 1994:190]

The dissenting voices mentioned by Dixon range from an argument that Sanskrit as well as modern Hindi were basically ``patient-oriented'' and thus should both be considered ergative (Hock 1986), to the interesting observation that stativity may be the relevant factor that is coming into play (Trask 1979:397) in the sense that a deverbal stative predicate is made active via an integration into the inflectional paradigm into the language (see Deo 2001a for a case study of Marathi).
Direct Descendent of the Sanskrit Instrumental -ina Both the proponents of the passive-to-ergative view and the dissenters (with the exception of Deo) assume that the modern Urdu/Hindi ergative ne is a direct descendant of the original Sanskrit inflectional instrumental -ina (and allomorphs thereof). However, the historical facts clearly speak against such an analysis.

Researchers of the last century such as Beames (1872-79) and Kellogg (1893), are very clear on the idea that the modern Urdu/Hindi ne could not possibly be a descendant of the Sanskrit instrumental -ina. Kellogg essentially lists three main problems with the hypothesis that the ergative ne be descended from the Sanskrit -ina: erosion, timing, and usage.

The highly inflected case system of Sanskrit underwent a general collapse over the ages and the case endings eroded and fell together. According to Sen (1973:68), the instrumental -ina/-ena eroded to eN by Middle Indo-Aryan and fell together with what was left of the dative: e. It is generally agreed (e.g., Sen 1973, Beames 1872-79, Kellogg 1893) that this eN/e furnished the current oblique marker of Urdu/Hindi. We may further assume that this morpheme is the ancestor of the inflectional ergative morphology in Assamese and Gujarati (see the table above). However, this morpheme could not have been the ancestor of the other ergative forms in the table.

To take modern Urdu/Hindi ne as an example again: this form is often described as a postposition in the literature (e.g., Davison 2000, Mahajan 1990). We follow Mohanan (1994) in treating it as a clitic (see Butt and King 2001 for a detailed discussion). Furthermore, as Kellogg also points out, in synchronic terms the ergative ne behaves much like other postpositions (or clitics) which are known to have developed from nouns: meN `in' and par `on'. The synchronic and diachronic data therefore seem to point to a relatively normal path of development: the instrumental -ina eroded away and the ergative ne came into the language as a grammaticalized form of a noun. On the other hand, the commonly assumed development from the inflectional morpheme -ina to a clitic (or postposition) ne would need to involve degrammaticalization, a highly unusual form of historical development and one which cannot be substantiated by the known synchronic and diachronic facts.

Another problem with the -ina to ne hypothesis is the relatively late appearance of the ergative in High Hindi. Beames (1872-79:267-271) surveys Old-Hindi writers like Chand, Kabir, Tulsi Das and Behari Lal and finds that he cannot trace the ergative ne back to more than 200-300 years ago (1600-1700). The writers he surveyed tend to use the oblique form -e (the old instrumental) of nouns/pronouns in constructions that today would be termed ``ergative''. The question then arises, if an ``ergative'' pattern based on the old instrumental was already in place, why then introduce a new marker into the language?

Beames (1872-79:270) traces the modern ergative ne to a dative form neN that was used in a dialect of Hindi spoken in provinces adjacent to the Moghul court during the reign of the Moghul Emperor Shah Jehan (1627-1658). Beames sees this time period as a very likely one because a change in administrative policies led to an influx of Hindu administrators, who might have put their stamp on the language of the court. Beames does not say which dialect the dative ne could have been borrowed from.

The precise origin of the non-inflectional forms of the ergative in the table above thus remains to be determined.
Language Contact Zakharyin (1979) ascribes the Urdu/Hindi ergative form ne to language contact with Tibeto-Burman, who use an ergative form na. However, this hypothesis does not explain why Nepali, a language which is geographically very close to the Tibeto-Burman languages, would employ le as an ergative marker (Devyani Sharma, p.c., August 2000).
Historical Stability Butt 2001, on the other hand, proposes that ne might be a reduced form of the Sanskrit locative janiye `for the sake of, because of, caused by' (based on a suggestion by Aditi Lahiri, p.c., December 1999). The semantics of this form are compatible with both agentive and dative (goal/benefactive) uses and as such this noun may have given rise to both dative and ergative forms in the table shown above. Butt further proposes that the ergative pattern in modern Urdu/Hindi is an instance of historical stability rather than an example of a radical accusative-to-ergative shift. This is based on the idea that the original Sanskrit -ta participle already formed an "ergative" pattern in the sense that the logical subject was marked nominative with intransitives, inceptives and verbs of motion, but instrumental with all others. This pattern has simply been retained in the modern language, though instantiated through new case morphology. This idea is consonant with Hock's 1986 claim that both Sanskrit and Hindi were essentially patient oriented and that in terms of this basic property, no historical change has taken place.
Ergativity as Licensed by the Development of IP In contrast, Deo 2001a argues that the historical development of ergativity in modern Indo-Aryan languages can be explained by a cross-linguistically attested historical shift: the development of a more articulated phrase structure in the form of an IP. This study looks at syntactic and morphological data from Old Indo-Aryan, Middle Indo-Aryan and Marathi, one of the modern Indo-Aryan languages.

The analysis rests on the crucial assumption that loss of morphology in a language (or language family) triggers phrase structural changes. The introduction of functional categories in compensation for the reduced functionality of inflectional morphology has been argued to be a general tendency in Indo-European syntactic change and this has been discussed in detail for Middle English and Greek (Kiparsky 2000, Kiparsky 1996). The hypothesis that Indo-Aryan also confirms to a similar trajectory is supported by the following syntactic and morphological changes in the language family, exemplified by a representative language, Marathi.