Syntactic Typology: Studies in the Phenomenology of Language

< previous section | Jump to: next section >

Winfred P. Lehmann

7. Ergativity

Bernard Comrie

7.0. Introduction

7.0.1. Scope

Ergativity is a term used in traditional descriptive and typological linguistics to refer to a system of nominal case-marking where the subject of an intransitive verb has the same morphological marker as a direct object, and a different morphological marker from the subject of a transitive verb. The following Tongan sentences provide examples (Churchward 1953: 67, 68): intransitive subject and direct object (referred to collectively as absolute) have the preposition 'a, and transitive subject (ergative) has the preposition -e:

(1) Na'e tāmate'i 'e Tēvita 'a Kōlaiate.
  Past kill Erg. David Abs. Goliath
  'David killed Goliath.'
(2) Na'e lea 'a Tolu.
  Past speak Abs. Tolu
  'Tolu spoke.'

This chapter aims to examine constructions of this and related types in a variety of languages, in order both to give some idea of the kinds of manifestations of ergativity that are found in different languages and to try to establish generalizations concerning ergativity, both as a synchronic phenomenon and in its diachronic relations with nonergative systems. The chapter does not seek to put forward any one single theory of ergativity, but rather to illustrate the range of data that any such theory must be able to encompass, in the hope that future theoretical accounts of ergativity will take account of the full range of data provided by different subtypes of ergativity. The last few years have seen a significant increase in the amount of data available on ergative languages, in particular on the syntax of ergative languages, and I have taken account of this below.

7.0.2. S, A, P, and Case-Marking

Discussion of Tongan examples (1) and (2) above is in one sense premature: it presupposes that, despite the formal identity of morphological marking of 'a Kōlaiate and 'a Tolu and the different morphological marking of 'e Tēvita, it is possible to group 'e Tēvita and 'a Tolu together as subjects (transitive and intransitive) versus 'a Kōlaiate as direct object. Indeed, many linguists looking at ergative constructions like (1) have concluded that the term subject of the transitive verb in this construction should refer to 'a Kōlaiate, rather than to 'e Tēvita, or even that the term subject is completely inapplicable in the ergative construction. We return to the definition of subject in section 7.1.2; in the meantime, to avoid begging this question, I shall use the three symbols S, A, and P, rather than subject and direct object. S refers to the single argument of an intransitive verb, e.g., 'a Tolu in (2), or Tolu in its English translation; the symbol is clearly reminiscent of the word subject and generally in such single-argument sentences it is clear that the one argument is the subject. A refers to that argument of a transitive verb which would be its subject in a non-ergative language like English (e.g., 'e Tēvita in (1), David its English translation); and P refers to the argument that would be the direct object (e.g., 'a Kōlaiate in (1), Goliath in its English translation). A and P are reminiscent of the semantic terms agent and patient, but though there is a high correlation between the semantic opposition agent/patient and the syntactic opposition A/P, the two are not identical (see further section 7.2); for instance, in the English sentence John underwent an operation, John is A and an operation P, although John is not semantically an agent.

Given this tripartite distinction (S, A, P) there are five logically possible systems for assigning case to S, A, and P, as in Figure 1. V here refers to verb; the order of the constituents is not relevant to the argument, and verb-final has been chosen purely for expository convenience.

In type (a), the same morphological marker (here, as elsewhere, morphological markers may be null) is used for all three syntactic positions; this type is illustrated by English, with nonpronominal noun phrases:

(3) John came.
(4) John kissed Mary.

In type (b), S and A have the same morphological marker (nominative), while a different marker is used for P (accusative). This case-marking system may be referred to as nominative-accusative, and is illustrated by Latin, for instance:

(5) Puer venit.
  boy-Nom. came
  'The boy came.'
(6) Puer puellam amat.
  boy-Nom. girl-Acc. loves
  'The boy loves the girl.'

Figure 1. Case-Marking Systems for S, A, and P

In type (c), S and P have the same morphological marker (absolute), while a different marker is used for A (ergative); this is the ergative-absolute system, often referred to simply as the ergative system. It is illustrated by Tongan examples (1) and (2) above, as well as by the following Basque examples (adapted from Lafitte 1962). In Basque, the ergative ending is -(e)k; the absolute case has no ending:

(7) Martin ethorri da.
  Martin-Abs. came Aux.-3Sg.S
  'Martin came.'
(8) Martin-ek haurra igorri du.
  Martin-Erg. child-Abs. sent Aux.-3Sg.A-3Sg.P
  'Martin sent the child.'

(in Basque, the auxiliary codes the person and number of S, A, and P. Thus da is the appropriate form if there is a third person singular S (3Sg.S) and no P, and du the appropriate form if there is a third person singular A and a third person singular P.)

Type (d), with three different morphological markers, is relatively rare across the languages of the world. Some languages have this system of case-marking for a limited number of noun phrases: in the Australian language Dyirbal, for instance, the interrogative pronoun 'who?' has distinct forms for S (wanʸa), A (wanʸdʸu), and P (wanʸuna) (Dixon 1972: 53), but this is not true of noun phrases in general (see further section 7.4.2). Motu, an Austronesian language of New Guinea, might appear to be a type (d) language from the data cited by Capell (1969: 36, 43, 54), with the postposition na for S, ese for A, and no overt marker for P:

(9) Mero na e ginimu.
  boy S he stands
  'The boy is standing.'

(10) Mero ese aniani e heni-gu.
  boy A food he gave-me
  'The boy gave me food.'

Motu, however, is not a pure type (d) language, since the A postposition ese is in fact optional, and the conditions on the occurrence of the S postposition na are more complex (and vary from dialect to dialect), so that some instances of S also lack any overt marker, (I am grateful to A. J. Taylor for providing me with Motu material additional to that cited by Capell.)

Type (e), with the same marker for both A and P and a different marker for S, seems not to occur as an attested case-marking system; we return to this in section 7.4.1, where I try to give an explanation for the apparent absence of this logical possibility. For the purposes of the present chapter, I shall be concerned primarily with type (c), contrasting it in particular with type (b). Type (a) makes no distinctions at all between S, A, and P, so provides no interesting material for the present chapter; type (d) can be viewed typologically (perhaps also diachronically, though I leave this question open) as the intersection of types (b) and (c).

7.0.3. General Problems

So far I have spoken about ergativity solely in terms of case-marking of noun phrases. One might therefore ask whether ergativity is a purely superficial morphological phenomenon, bearing no relation to any other syntactic or semantic properties of the language in question; or whether it is rather a more deep-seated typological trait of languages with this case-marking system, permeating much more of the syntactic and semantic structure, and perhaps calling into question the whole theoretical apparatus of the subject/direct object distinction with respect to such languages. I discuss this problem in detail in section 7.1, showing that there is empirical evidence that can be brought to bear in answering this question, although the answer is by no means straightforward: rather, languages with morphological ergativity differ in the extent to which they have further repercussions of ergativity in their syntactic structure. In section 7.2 I discuss the attempt to define ergativity in semantic rather than syntactic terms, that is in terms of semantic agentivity; again, it turns out that languages vary in the extent to which this approach is applicable to them.

A second set of questions one can ask concerning ergativity relates to the motivation for this system; that is, why do languages exist with the ergative-absolute system, rather than all languages being nominative-accusative to the extent that they have case-differentiation of S, A, and P? One possible answer would be diachronic, that is, investigating how an ergative-absolute system can arise from a nonergative system: this possibility is examined in section 7.3, which also looks at relationships between ergativity and passive constructions. Another possible approach would be synchronic, asking whether the ergative-absolute system makes sense as a purely synchronic phenomenon, a question to which I turn in section 7.4. The questions in this paragraph have been posed from the viewpoint of a nominative-accusative system, as if such a system were the norm, and the ergative system to be regarded as some kind of aberration requiring special explanation. But, in sections 7.3 and 7.4 I adopt a more neutral stance, since one can equally ask, from the viewpoint of an ergative-absolute system, why the nominative-accusative type should exist. Although the ergative-absolute system is often regarded as an exotic linguistic type, this seems to stem largely from the fact that the most widely spoken European languages are not ergative, which in turn results from the historical accident of the spread of Indo-European languages to the exclusion of most other European languages (though Basque, surrounded by Indo-European, does have an ergative-absolute case-marking system). Indeed, apart from Europe, where ergativity is absent except for Basque and the Caucasus, and Africa, where there seem to be no ergative languages, ergative languages are to be found in nearly all parts of the world.

7.1. Morphological and Syntactic Ergativity

The five systems illustrated in Figure 1 have been discussed so far solely in terms of case-marking of noun phrases. However, there is no a priori reason for the differences among these five systems to be restricted solely to case-marking, or indeed solely to morphology, rather than being applied to a variety of syntactic phenomena. In other words, concentrating in particular on the nominative-accusative and ergative-absolute systems, we can ask whether there are syntactic phenomena in a given language which treat S and P alike in contrast to A, i.e., on the basis of the ergative-absolute system, where in other languages the same syntactic phenomenon might operate on the basis of S and A in contrast to P, i.e., on the basis of the nominative-accusative system. In this section, I examine both morphological and syntactic aspects of ergativity.

The general problem can be illustrated with examples from English. In English, as far as the vast majority of syntactic phenomena are concerned, the system is nominative-accusative. For instance, sentences (11) - (15) below show that in forming constructions of the type X wants Y to Z, irrespective of whether Z is transitive or intransitive, Y is always interpreted as the subject (S or A) of Z, and never as its object (P):

(11) The birds chirp.
(12) The wolf hunts the fox.
(13) I want the birds to chirp.
(14) I want the wolf to hunt the fox.
(15) *I want the fox the wolf to hunt.

However, even in English there are some fairly marginal constructions that operate rather on an ergative-absolute basis. For instance, in English it is possible to form compound nouns of the type N-V-ing, such as bird-chirping, foxhunting. However, the interpretation of such compounds depends on whether the verb is transitive or intransitive: with an intransitive verb like chirp, the noun is interpreted as its S, i.e., bird-chirping is related to birds chirp; whereas with a transitive verb like hunt the noun is interpreted as its P, i.e., fox-hunting is related to (someone) hunts foxes, not to foxes hunt (something). (It seems, incidentally, to be generally true across languages that where nouns can be incorporated into verbs, as in these compounds in English, P is easiest to incorporate, followed by S, with A being most resistant to incorporation.)

7.1.1. Morphological Ergativity: Case-Marking and Verb-Agreement

In section 7.0.2 I explained and illustrated the ergative-absolute case-marking system; there is then no need to repeat what was said there with regard to morphological ergativity of nominal case-marking. There is, however, another kind of morphological ergativity, namely ergativity in the verb-agreement system. Just as case-marking can operate in accordance with the five logically possible systems of Figure 1 (though, as already noted, (e) seems not to occur in practice), so too can verb-agreement. It is possible for verb-agreement to follow the nominative-accusative system, as for instance in the Bantu language Swahili, where the same agreement markers are used for S and A (preceding the tense morpheme), and a different set for P (following the tense morpheme):

(16) Hamisi a- li- fika.
  Hamisi he- Past- arrive
  'Hamisi arrived.'
(17) Hamisi a- li- mw- ona Juma.
  Hamisi he- Past- him- see Juma
  'Hamisi saw Juma.'

In some languages with an ergative-absolute case-marking system, the verb-agreement is determined equally on an ergative-absolute basis. This is so in Avar, a northeast Caucasian language, where verbs agree in noun class (this includes a male/female division for human nouns) with S and P, but have no overt agreement with A (Anderson 1976: 4):

(18) Vas v- ekerula.
  boy-Abs. Sg.Masc.Abs.- run
  'The boy runs.'
(19) Jas j- ekerula.
  girl-Abs. Sg.Fem.Abs.- run
  'The girl runs.'
(20) Vas-as: jas j- ec:ula.
  boy-Erg. girl-Abs. Sg.Fem.Abs.- praise
  'The boy praises the girl.'

As the glosses indicate, verbs agree with the absolute noun phrase.

There are even some languages that have verb-agreement on an ergative-absolute basis but have no overt case-marking of noun phrases; in fact, this type is not particularly rare, being found for instance in some of the Northwest Caucasian languages (Abkhaz, Abaza; see, for instance, Allen 1956), and quite generally in the Mayan languages of Mexico and Central America. The examples below are from Quiché, a Mayan language of Guatemala (Quiché examples here and below are from the Mayan Linguistics course given by Lyle Campbell at the 1976 Linguistic Institute).

(21) K- ox kam- ik.
  Asp.- 1Pl.Abs.- die- Ptc.
  'We die.'
(22) K- at- kam- ik.
  Asp.- 2Sg.Abs. die- Ptc.
  'You die.'
(23) K- at- ka- cuku- x.
  Asp.- 2Sg.Abs.- 1Pl.Erg.- seek- Act.
  'We seek you.'
(24) K- ox- a- cuku- x.
  Asp.- 1Pl.Abs.- 2Sg.Erg.- seek- Act.
  'You seek us.'

The independent pronouns are normally omitted, unless stressed. Like other noun phrases, they have no case-marking in Quiché; the usual word order is V-S or V-P-A. Verb-agreement is on an ergative-absolute basis: for the first person plural, ox- is the absolute prefix: S in (21), P in (24); and ka- the ergative prefix: A in (23). For the second person singular, at- is the absolute prefix: S in (22), P in (23); and a- the ergative prefix: A in (24). In Mayan linguistics, it is customary to refer to the two sets of verbal affixes as Set A (ergative, e.g., first person plural ka-) and Set B (absolute, e.g., first person plural ox-). Although arbitrary, this terminology has certain advantages — the affixes of Set A, for instance, are also used as noun prefixes indicating possession, e.g., Quiché ka-c'i:' 'our dog' — especially in dealing with the more complex distribution of the two sets in languages like Chol and Jacaltec, discussed in sections 7.1.3 and 7.3.2.

Assuming that neutral nominal case-marking is compatible with either a nominative-accusative or an ergative-absolute system, in that it does not make any distinction cutting across either of these two distinctions, we can say that the languages examined so far with regard to nominal case-marking and verb-agreement are either consistently nominative-accusative or consistently ergative-absolute. We might go on to ask whether there are any languages that combine the nominative-accusative morphological system with the ergative-absolute morphological system, having one for nominal case-marking and the other for verb-agreement. In fact, there are many languages which have an ergative-absolute system for nominal case-marking and a nominative-accusative system for verb-agreement (the inverse is rare or nonexistent). One such language is Walbiri, a Pama-Nyungan language of Central Australia (Hale 1973: 309, 328):

(25) ŋatʸu ka -ṇa puḷami.
  1-Abs. Tense -1Sg.Nom. shout
  'I shout.'
(26) ŋatʸuluḷu ka -ṇa -ŋku nʸuntu nʸanʸi.
  1-Erg. Tense -1Sg.Nom. 2Sg.Acc. you-Abs. see
  'I see you.'
(27) Nʸuntuluḷu ka -npa -tʸu ŋatʸu nʸanʸi.
  you-Erg. Tense -2Sg.Nom. -1Sg.Acc. 1-Abs. see
  'You see me.'

Inspection of these examples shows that the independent pronouns (which behave in this respect like other noun phrases) have case-marking on an ergative-absolute basis: for the first person singular, ŋatʸu is absolute: S in (25), P in (27); while ŋatʸuluḷu is ergative: A in (26). The forms for the second person singular are absolute nʸuntu, ergative nʸuntuluḷu. The verb-agreement affixes, however, which appear in sentence-second position together with the tense-marker ka, are assigned on a nominative-accusative basis, so that the first person "nominative" affix -ga occurs as S agreement affix in (25) and as A agreement affix in (26), while a separate form, -tʸu, is used as P agreement affix in (27); for the second person, the forms are nominative -npa, accusative -ŋku. In Walbiri, then, we see for the first time a discrepancy within the (core) grammatical structure; part of the morphology is based on the ergative-absolute distinction, and another part on the nominative-accusative distinction.

We may note in passing that one occasionally comes across a subvariety of type (d) of Figure 1 with verb-agreement, namely where portmanteau morphs are used to indicate certain combinations of A and P. Thus in Rembarunga, an Australian language of Arnhem Land, the agreement markers for Ss of the first and second persons singular are, respectively, ŋa and ŋinʸ, but the agreement marker for a second person singular A with a first person singular P is tan (although for most other combinations of A and P, verb-agreement is on a nominative-accusative basis, with relatively transparent morphophonemic variation in some of these instances) (McKay 1977: 501).

Even type (3) is occasionally found with verb-agreement, as in the Iranian dialect Dānesfāni (Yar-Shater 1969: 204), where the past participle agrees with S, but with no constituent of a transitive sentence:

(28) Hasan buma.
  Hasan-(Masc.) came-Masc.
  'Hasan came.'
(29) Zeynaba bumia.
  Zeynaba-(Fem.) came-Fem.
  'Zeynaba came.'
Hasan /Zeynaba šet -eš
Hasan-(Masc.) /Zeynaba-(Fem.) milk-(Masc.) -Aux.-3Sg.A
'Hasan/Zeynaba drank the milk.'
Hasan /Zeynaba āwa
Hasan-(Masc.) /Zeynaba-(Fem.) water-(Fem.) -Aux.-3Sg.A
'Hasan/Zeynaba drank the water.'

In all such instances known to me, however, the verb agrees with S, and shows no agreement whatsoever with either A or P; moreover, the examples with which I am familiar are all in languages where the ergative system is breaking down, being replaced by a nominative-accusative or neutral system. It seems likely that at an intermediate stage in the development from ergative-absolute to nominative-accusative, a situation can be reached where the conflict between moribund ergative-absolute morphology and nascent nominative-accusative morphology is resolved by simply omitting all morphological markers, giving rise to the type illustrated above from Dānesfāni: in the intransitive sentence there is no conflict, and the participle agrees with S; in the transitive construction, there is conflict as to whether the participle should agree with A or with P. The compromise reached is for it to agree with neither. Thus we expect type (e) to arise only as a result of conflict of this kind.

7.1.2. Syntactic Ergativity

In treating ergativity from a syntactic viewpoint, we are looking for syntactic phenomena in languages which treat S and P alike, and differently from A. Syntactic nominativity likewise means syntactic phenomena where S and A are treated alike, and differently from P. This distinction is connected with the general problem of subject identification: if in a language S and A are regularly identified, that is, if the language is consistently or overwhelmingly nominative-accusative, then we are justified in using the term subject to group together S and A; if in a language S and P are regularly identified (consistent or overwhelming ergative-absolute system), then we would be justified in using the term subject rather to refer to S and P, that is, in particular, to refer to P, rather than A, of the transitive construction. The fullest attempt to date to isolate syntactic subject properties is Keenan 1976b, which examines a large number of properties, including syntactic properties characteristic of subjects across a wide range of languages (in particular, nonergative languages). For present purposes, I shall examine a much more restricted range of properties, using these as illustrations of general trends.

In English and a number of other languages, certain verbal forms dependent on another verb may lack an overt subject if and only if the subject of the dependent verb is the same as the subject of the main verb. In English, the verbs want and can, for instance, behave in this way, with identity of subject of main and dependent verb:

(32) John wants to come.
(33) John wants to kiss Mary.
(34) John can come.

(35) John can kiss Mary.

Where the subject of the dependent verb is not the same as that of the main verb, as is possible with want, the subject of the dependent verb must be expressed overtly (in English, as surface direct object of want):

(36) John wants Bill to come.
(37) John wants Bill to kiss Mary.

I have stated this condition in terms of subjects (of the dependent verb), and the examples cited show that surface S and A are treated alike, in contrast to surface P. I now examine the same construction types in Khinalug, a Northeast Caucasian language with morphological ergativity in both case-marking and verb-agreement.

The morphologically ergative nature of Khinalug can be seen from the following (Kibrik et al 1972: 192-193):

(38) Lɨgɨld sacaχ-0 -q'iqomä.
  man-Abs. silent -Masc.
  'The man is silent.'
(39) Χińimk'ir sacaχ-z -q'iqomä.
  woman-Abs. silent -Fem.
  'The woman is silent.'
(40) Bɨj -i ši -0 -k′'i.
  father -Erg. son-Abs. awaken -Masc.  
  'The father awakened the son.'
(41) Bɨj -i riši -z -k′'i.
  father -Erg. daughter-Abs. awaken -Fem.  
  'The father awakened the daughter.'

The verb-agreement marker, which separates the two parts of a compound verb, agrees in noun class (including a male/female distinction for humans) with the absolute noun phrase. The verbs meaning 'want' and 'can' take a nonfinite dependent verbal form, the so-called dependent participle; where the subject of the dependent participle is not the same as that of the main verb, it appears in the same form (ergative or absolute) as it would with a finite form of that verb (Kibrik et al 1972: 194-195):

(42) Asɨr gada točkwi jukwathmä.
  I-Dat. boy-Abs. to-get-up want
  'I want the boy to get up.'
(43) As jukwath hini phšä q'izi.
  I-Dat. want she-Erg. bread to-bake
  'I want her to bake bread.'

(The verbs 'want' and 'can' take their subject in the dative case; as and asɨr are alternative dative forms of the first person singular pronoun.) Where, however the subject (S or A) of the dependent participle would be the same as the subject of the main verb, it must be omitted:

(44) As jukwešämä endžik -ondä.
  I-Dat. wanted to-descend -not
  'I wanted not to descend.'
(45) As uχur khheb läk'iri jet′:imä.
  I-Dat. to-you book-Abs. to-give non-want
  'I do not want to give the book to you.'
(46) Hinu lik′'uvri muxwižmä.
  she-Dat. to-sing can
  'She can sing.'
(47) Hinu phšä q'izi muxwižmä.
  she-Dat. bread-Abs. to-bake can
  'She can bake bread.'

From these examples, it is clear that this construction in Khinalug is controlled by the nominative-accusative opposition, since the dependent participles in (44) - (47) are both intransitive and transitive, and not by the ergative-absolute opposition. Thus despite morphological ergativity, Khinalug is syntactically a nominative-accusative language.

In the study of most languages that have morphological ergativity (nominal, verbal, or both), it seems to turn out that, apart from a few syntactic processes like incorporation that tend always to operate on an ergative-absolute basis rather than a nominative-accusative basis, the vast majority of syntactic phenomena operate on a nominative-accusative basis rather than on an ergative-absolute basis. That is, the majority of languages that are morphologically ergative are not syntactically ergative; in the majority of ergative languages ergativity seems to be a relatively superficial phenomenon, at least as far as morphology and syntax are concerned (I turn to semantics, especially agentivity, in section 7.2); for further exemplification of this point see Comrie 1973, Anderson 1976, Keenan 1976b. In such languages then we are justified in using the term subject to refer to S and A, and direct object for P, just as we are in English.

A different picture is provided, however, by Dyirbal, an Australian language of Northern Queensland. The illustrative construction chosen here is the formation of coordinate sentences where one of the components is a transitive sentence, the other an intransitive sentence, of the type The man came here and hit the woman, or The man hit the woman and came here. I present the material, for English and then for Dyirbal, by indicating first of all the relevant noncoordinate sentences (e.g., the man came here, the man hit the woman), then the possibilities for combining them, with omission of the common noun phrase from the second clause; the coordinate sentences will be written out with an indication of which simple sentences enter into their formation. The English examples are as follows:

(48) The man came here.

(49) The man hit the woman.
(50) The woman came here.
(51) The woman hit the man.
(52) The man came here and hit the woman. (48) + (49)
(53) The man hit the woman and came here. (49) + (48)
(54) The woman came here and hit the man. (50) + (51)
(55) The woman hit the man and came here. (51) + (50)

From the coordination possibilities — note, for instance, that (48) can be coordinated with (49), but not with (51) — it is clear that the relevant criterion for this kind of coordination is that the two simple sentences should have the same subject (S or A); the subject of the second clause in the coordinate construction may then be omitted. Since (48) and (49) have the same subject, they can be coordinated to give (52) or, with reverse order of the clauses, (53). It is not possible to coordinate (48) with (51), since although both have the noun phrase the man, this noun phrase is subject in (48) but direct object in (51). In other words, this kind of coordination in English works on a nominative-accusative basis.

Dyirbal has an ergative case-marking system (there is no verb-agreement): the ergative suffix has several allomorphs, including -ŋgu and -ṛu; the absolute case has no marker. In addition to these markers on nouns, a noun may be preceded by a nominal classifier indicating its noun class (for human nouns, this means a male/female distinction) and its case (including ergative versus absolute). Although word order is free, the usual order is P-A-V or (S)-V (Dixon 1972: 59, 130):

(56) Bayi yaṛa bani-nʸu.
  Masc.-Abs. man-Abs. come-Tense
  'The man came here.'
(57) Balan dʸugumbil baŋgul yaṛa- ŋgu balga- n.
  Fem.-Abs. woman-Abs. Masc.-Erg. man- Erg. hit- Tense
  'The man hit the woman.'
(58) Balan dʸugumbil bani-nʸu.
  Fem.-Abs. woman-Abs. come-Tense
  'The woman came here.'
(59) Bayi yaṛa baŋgun dʸugumbi- ṛu balga- n.
  Masc.-Abs. man-Abs. Fem.-Erg. woman- Erg. hit- Tense
  'The woman hit the man.'
(60) Bayi yaṛa baninʸu, baŋgun dʸugumbiṛu balgan. (56) + (59)
  'The man came here and the woman hit him.'
(61) Bayi yaṛa baŋgun dʸugumbiṛu balgan, baninʸu. (59) + (56)
  'The woman hit the man and he came here.'
(62) Balan dʸugumbil baninʸu, baŋgul yaṛaŋgu balgan. (58) + (57)
  'The woman came here and the man hit her.'
(63) Balan dʸugumbil baŋgul yaṛaŋgu balgan, baninʸu. (57) + (58)
  'The man hit the woman and she came here.'

Coordination in Dyirbal is indicated without any overt conjunction, simply by juxtaposing the two sentences, with optional omission of the appropriate noun phrase from the second clause. Inspection of the coordination possibilities shows that they are different from those for English, in fact strikingly so: the combinations that are possible in Dyirbal are impossible in English, while the combinations that are possible in English are impossible in Dyirbal. In Dyirbal, coordination operates on an ergative-absolute basis, that is, it is possible to treat as identical for purposes of coordination two absolute noun phrases even if one is an S and the other a P. But it is not possible to treat as identical for purposes of coordination an S and an A:

(64) *Bayi yaṛa baninʸu, balan dʸugumbil balgan. (56) + (57)
  'The man came here and hit the woman.'

In Dyirbal one finds, quite generally, that syntactic processes which in English operate on a nominative-accusative basis operate on an ergative-absolute basis. Given this situation, one is led to consider that perhaps, with regard to languages of this type, it is incorrect to identify A as the subject in transitive sentences: in a sentence like (57), nearly all the subject properties characterize the absolute noun phrase (P) balan dʸugumbil, rather than the ergative noun phrase (A) baŋgul yaṛaŋgu. If one were to consider P as the subject, and A as an agentive complement, then one is effectively claiming that such Dyirbal sentences have the syntactic structure of English passive sentences. We can illustrate this by considering again coordination possibilities, as in (48) — (55), but taking passive rather than active sentences with transitive verbs:

(65) The man came here.
(66) The woman was hit by the man.
(67) The woman came here.
(68) The man was hit by the woman.
(69) The man came here and was hit by the woman. (65) + (68)
(70) The man was hit by the woman and came here. (68) + (65)
(71) The woman came here and was hit by the man. (67) + (66)

(72) The woman was hit by the man and came here. (66) + (67)

(We return to the relation between ergative and passive for further discussion in section 7.3.) One factor that has to be taken into account is that languages do not divide neatly into those that are purely morphologically ergative (like Khinalug) and those that are consistently syntactically ergative (as well as morphologically ergative) (like Dyirbal); many languages occupy an intermediate position, with certain syntactic phenomena operating on a nominative-accusative basis, others on an ergative-absolute basis, so that the ergative languages of the world as a whole show a continuum from purely morphological ergativity to syntactic ergativity, though with a preponderance of languages at the purely morphologically ergative end of the spectrum. In languages where subject properties are roughly evenly distributed between A and P, the value of the term subject is seriously called into question. For a detailed discussion of a language with some syntactic phenomena on a nominative-accusative basis and others on an ergative-absolute basis, reference should be made to Dixon's (1977a) account of Yidinʸ, another Australian language of Northern Queensland, and for more general discussion, based on Australian material, to Blake 1976, and the articles on ergativity in Dixon, ed. 1976: 485-611.

7.1.3. Split Ergativity

One general conclusion that can be drawn from the discussion of section 7.1 so far is that it is rather misleading to speak of ergative languages, as opposed to nominative-accusative languages, since we have seen that it is possible for one phenomenon in a language to be controlled on an ergative-absolute basis while another phenomenon in the same language is controlled on a nominative-accusative basis. Thus one should ask rather to what extent a language is ergative-absolute or nominative-accusative, or, more specifically, which constructions in a particular language operate on the one basis and which on the other. In fact, the situation is even more complex than this, because we sometimes find the same phenomenon in the same language operating in some instances on a nominative-accusative basis, in others on an ergative-absolute basis. In the present section I shall examine some examples of this kind of "split ergativity"; most of them will be discussed further in sections 7.3 and 7.4, where explanations will be suggested for the various kinds of split ergativity.

One of the commonest ways in which languages manifest split ergativity is according to tense/aspect: in some tenses or aspects the language is nominative-accusative, in others it is ergative-absolute. I illustrate this by means of Georgian examples; languages with similar systems include Hindi and many other Indo-Iranian languages. In Georgian, a South Caucasian (Kartvelian) language, the aorist tense system has an ergative-absolute system, with the ending -i for the absolute and -ma for the ergative; the present tense system has a nominative-accusative system, with the ending -i for the nominative and -s for the accusative:

(73) Sṭudenṭ-i midis.
  student-Nom. goes
  'The student goes.'
(74) Sṭudenṭ-i c̣eril -s c̣ers.
  student-Nom. letter -Acc. writes
  'The student writes the letter.'
(75) Sṭudenṭ-i mivida.
  student-Abs. went
  'The student went.'

(76) Sṭudenṭ-ma c̣eril -i dac̣era.
  student-Erg. letter -Abs. wrote
  'The student wrote the letter.'

In these examples I continue to gloss the case-markers according to their role in the ergative-absolute or nominative-accusative system, although this means that Georgian -i, for instance, is glossed as nominative in (73) and (74), but as absolute in (75) and (76) — this to avoid the confusion that is inevitably engendered when traditional case labels (nominative, ergative, etc.) are used to refer to forms that occur in both ergative-absolute and nominative-accusative systems.

In Georgian (and likewise in the relevant Indo-Iranian languages), there is morphological identity between nominative of the nominative-accusative system and absolute of the ergative-absolute system, but this is by no means a necessary pattern in languages with split ergativity. In Chol, a Mayan language of Mexico, the tense/aspect basis of split ergativity is similar to that of Georgian (present tense nominative-accusative versus past tense ergative-absolute), but the distribution of the forms is different. (It will be recalled from section 7.1.1 that ergativity in Mayan languages is revealed primarily in the verb agreement. I am grateful to R. Freund for bringing the Chol data to my attention.)

(77) Ca čəmiy- on.
  Past die- 1Sg.Abs.
  'I died.'
(78) Ca čəmiy- et.
  Past die- 2Sg.Abs.
  'You died.'
(79) Ca h- k'eley- et.
  Past 1Sg.Erg.- see- 2Sg.Abs.
  'I saw you.'

(80) Mi k- čəmel.
  Pres. 1Sg.Nom.- die
  'I am dying.'
(81) Mi a- čəmel.
  Pres. 2Sg.Nom.- die
  'You are dying.'
(82) Mi h- k'el- et.
  Pres. 1Sg.-Nom.- see- 2Sg.Acc.
  'I see you.'

In Chol, the Set A affixes — first person singular k- (h-before a velar), second person singular a — function as ergative in the ergative-absolute system, but as nominative in the nominative-accusative system; the Set B affixes — first person singular -on, second person singular -et — function as absolute in the ergative-absolute system, but as accusative in the nominative-accusative system. In some Mayan languages, for example Jacaltec, a similar split occurs between main clauses (ergative-absolute) and subordinate clauses (nominative-accusative), with morphological identity of ergative and nominative; see section 7.3.2.

Another major type of split ergativity is where the conditioning factor is the status of the A and P noun phrases, that is, where certain noun phrases have a nominative-accusative case-marking system, but others an ergative-absolute case-marking system. I return to this kind of split ergativity, and further variations thereon, in section 7.4; for the present I simply give some examples from Dyirbal. As already noted, (56) - (59), most noun phrases in Dyirbal have an ergative-absolute case-marking system. Independent personal pronouns of the first and second persons, however, have rather a nominative-accusative case-marking system (Dixon 1972: 60):

(83) ŋadʸa bani-nʸu.
  I-Nom. come-Tense
  'I come.'
(84) ŋinda bani-nʸu.
  you-Nom. come-Tense
  'You come.'
(85) ŋadʸa ŋinuna balga- n.
  I-Nom. you-Acc. hit- Tense
  'I hit you.'
(86) ŋinda ŋayguna balga- n.
  you-Nom. I-Acc. hit- Tense
  'You hit me.'

As can be seen from these examples, ŋadʸa 'I' and ŋinda 'you' function as both S and A, whereas ŋayguna 'me' and ŋinuna 'you' function only as P. Pronominal and nonpronominal noun phrases may of course co-occur in the same sentence, and then the case-marking system for each operates independently:

(87) ŋadʸa bayi yaṛa balga- n.
  I-Nom. Masc.-Abs. man-Abs. hit- Tense
  'I hit the man.'
(88) ŋayguna baŋgul yaṛa- ŋgu balga- n.
  I-Acc. Masc.-Erg. man- Erg. hit- Tense
  'The man hit me.'

With this kind of split ergativity, not only can we not characterize a whole language as discretely ergative versus nonergative, we cannot even characterize the nominal case-marking of a whole sentence in these terms, since sentences like (87) and (88) contain elements of both the ergative-absolute and the nominative-accusative case-marking systems.

As a final observation on this split ergativity in Dyirbal, we should note that the nominative-accusative system found with certain pronouns is purely morphological; as far as syntax is concerned, for example coordination possibilities, such sentences still behave according to the ergative system (Dixon 1972: 131):

(89) ŋadʸa baninʸu, baŋgul yaṛaŋgu balgan. (83) (88)
  'I came here and the man hit me.'
(90) ŋayguna baŋgul yaṛaŋgu balgan, baninʸu. (88) (83)
  'The man hit me and I came here.'
(91) ŋadʸa bayi yaṛa balgan, baninʸu. (87) (56)
  'I hit the man and the man came here.'

In this respect, Dyirbal shows how different it is from most languages with ergativity, where the ergativity is essentially only morphological, not syntactic; in Dyirbal ergativity is syntactic and also characterizes most of the morphology, and the only exceptional nominative-accusative morphology does not affect the basic syntactic ergativity of the language.

Finally, with regard to split ergativity, we may note that some languages present idiosyncratic exceptions to the general system (ergative-absolute or nominative-accusative) that is used for a particular phenomenon. Thus Khinalug is basically ergative-absolute with regard to verb-agreement — cf. (38) - (41) — but the transitive verb č'aχi 'bring' takes nominative-accusative verb-agreement, although case-marking is still on an ergative-absolute basis (Kibrik et al. 1972: 193):

(92) Bɨj -i xu čhkha-0 -Ršämä.
  father Erg. water-Abs. bring-Masc.  
  'Father brought water.'
(93) Däd -i xu čhkha-zɨ -Ršämä.
  mother -Erg. water-Abs. bring-Fem.  
  'Mother brought water.'

7.2. Ergativity and Agentivity

In many discussions of ergativity, a key role is played by the notion of agentivity, and readers may be surprised that this notion has not been discussed so far in the present chapter. By agentivity I mean, roughly speaking, the degree of control which the referent of a noun phrase has over the situation described by the verb with which it is associated. Thus, in the following examples, the degree of agentivity of the subject noun phrase diminishes progressively through the list:

(94) John deliberately fell down.
(95) John, through his own negligence, fell down.
(96) John, through no fault of his own, fell down.

(94) assigns absolute control to the subject, that is, highest agentivity: his falling was due to a conscious effort on his part directed toward this goal. (95) assigns intermediate control: John could have avoided falling down, but he did not, that is, he had potential control over the situation, which he did not in fact exercise. (96) assigns no control at all; the fall was entirely outside John's control. Thus the claim that there is a close connection, perhaps even of identity, between ergativity and agentivity amounts to the claim that the function of the ergative case is not so much syntactic as semantic, to mark a noun phrase which has high agentivity — e.g., (94) versus (95) and (96), or (94) and (95) versus (96). The use of the ergative to mark A would then be an automatic consequence of the fact that in most transitive sentences A has high agentivity, at least relative to P. In this discussion I explicitly reject the identification of ergativity and agentivity, for reasons discussed in the present section; the line of argument will be to show that, despite some similarities between ergativity and agentivity, evidence from a wide range of ergative languages points against this identification.

The first piece of evidence against this identification is that in many languages ergative noun phrases do not have to be agentive, as in the following Basque examples (Lafitte 1962: 322):

(97) Herra -k z- erabiltza.
  hatred -Erg. you- move
  'Hatred inspires you.'
(98) Ur-handia -k d- erabilka eihara.
  river -Erg. it- move mill-Abs.
  'The river works the mill.'

Second, languages with split ergativity on a tense/aspect basis provide examples where the same noun phrase appears now in the ergative, now in the nominative, without any difference of agentivity, as in Georgian examples (74) and (76), reproduced below as (99) and (100):

(99) Sṭudenṭi (Nom.) c̣erils c̣ers.
  'The student writes the letter.'
(100) Sṭudenṭma (Erg.) c̣erili dac̣era.
  'The student wrote the letter.'

The main kind of evidence to be cited in this section is an extension of the second kind noted above, and consists in finding pairs of sentences where, for syntactic reasons or for semantic reasons unconnected with the agentivity of the given noun phrase, in one of the sentences it appears in the ergative, in the other in a nonergative case (in the examples cited here, usually the absolute case). The first set of examples concerns verbs that are basically transitive, but which can be used also without a P, although the same interpretation of agentivity is present in both, as in English:

(101) John is eating a fish.
(102) John is eating.

In some ergative languages, the subject of such an objectless transitive verb remains in the ergative case, thus providing perhaps some evidence of a link between ergativity and agentivity, as in Basque:

(103) Martin-ek jan du.
  Martin-Erg. ate Aux.-3Sg.A (-3Sg.P)
  'Martin ate.'

Martinek has the ergative ending -ek, and the auxiliary is the form used with a transitive verb having a third person singular A and a third person singular P; in fact, (103) can also mean 'Martin ate it.' As in many other languages, third person singular acts as an unmarked category, since Basque has no specific auxiliary forms to indicate the absence of a direct object with a transitive verb. In many other ergative languages, however, the subject of such a verb stands in the absolute case, for example, in Tongan:

(104) Na'e inu 'a e kava 'e Sione.
  Past drink Abs. the kava Erg. John
  'John drank the kava.'
(105) Na'e inu 'a Sione.
  Past drink Abs. John
  'John drank.'

In many languages with a case-marking system like Tongan in this respect, a marker is also required on the verb in such constructions to indicate that it is being used intransitively. In Dyirbal this is the reflexive suffix (Dixon 1972: 90):

(106) Balam wudʸu baŋgul yaṛa- ŋgu dʸaŋga- nʸu.
  Abs. fruit-Abs. Masc.-Erg. man- Erg. eat- Tense
  'The man eats fruit.'
(107) Bayi yaṛa dʸaŋgay- mari- nʸu.
  Masc.-Abs. man-Abs. eat- Refl.- Tense
  'The man eats.'

In Quiché, the corresponding suffix is -n:

(108) K- at- ka- kamisa- x.
  Asp. 2Sg.Abs.- 1Pl.Erg.- kill- Act.
  'We kill you.'
(109) K- ox- kamisa-n- ik.
  Asp. 1Pl.Abs.- kill- Ptc.
  'We kill.'

Similar pairs can be found where the same verb can be constructed either with a direct object or with some other kind of object (often with a difference in meaning between the two versions — I return to this below). Occasionally, the ergative is found in both members of such a pair, as in Walbiri, where the nondirect object appears in the dative (Hale 1973: 336):

(110) ŋatʸuluḷu ka -ṇa wawiri ḷuwaṇi.
  I-Erg. Tense -1Sg.Nom. kangaroo-Abs. shoot
  'I am shooting at the kangaroo.'
ŋatʸuluḷu ka -ṇa -ḷa -tʸinta wawiri
I-Erg. Tense -1Sg.Nom. -3Sg.Dat. -Intrans. kangaroo-
-ki ḷuwaṇi.
Dat. shoot
'I am shooting at the kangaroo (but may not hit it).'

More commonly, however, the difference in syntactic structure of the object brings about a difference in syntactic structure of the sentence, with the absolute case being used for the subject where the object is nondirect, as in the Bezhedukh dialect of West Circassian (Northwest Caucasian) (Anderson 1976: 21; the examples are attributed to John Colarusso):

(112) Č′'aaʎa- m č'əg°-ər yaź°a.
  boy- Erg. field-Abs. he-plows-it
  'The boy is plowing the field.'
(113) Č′'aaʎa- r č'əg°-əm yaź°a.
  boy- Abs. field-Nondir. he-plows-it
  'The boy is plowing away at the field (but may not complete it).'

Here the use of a nondirect object indicates noncompletion of the action. With some verbs, there is also a change in the form of the verb to indicate the change away from transitivity:

(114) P:śaśa- m chəy -ər yadə.
  girl- Erg. cherkesska -Abs. she-sews-it-Trans.
  'The girl is sewing the cherkesska.'
(115) P:śaśa- r chəy -əm yada.
  girl- Abs. cherkesska -Nondir. she-sews-it-Intrans.
  'The girl is sewing away at the cherkesska.'

In some other languages, the use of a special affix on the verb to indicate intransitivity is more widespread. In Dyirbal, for instance, either the reflexive suffix or the suffix -ŋay is used, and the object appears in the dative (or instrumental) (Dixon 1972: 90, 151-152):

(116) Balam wudʸu baŋgul yaṛa- ŋgu dʸaŋga- nʸu.
  Abs. fruit-Abs. Masc.-Erg. man- Erg. eat- Tense
  'The man eats fruit.'
Bayi yaṛa bagum wudʸu-gu dʸaŋga- mari-
Masc.-Abs. man-Abs. Dat. fruit-Dat. eat- Refl.-
'The man eats fruit.'
(118) Balan dʸugumbil baŋgul yaṛa- ŋgu balga- n.
  Fem.-Abs. woman-Abs. Masc.-Erg. man- Erg. hit- Tense
  'The man is hitting the woman.'
Bayi yaṛa bagun dʸugumbil- gu balgal-
Masc.-Abs. man-Abs. Fem.-Dat. woman- Dat. hit-
ŋa- nʸu.
'The man is hitting the woman.'

Similarly in Quiché, the form of the verb in -n is used in such instances. In Quiché the verbal affixes for first person singular and the polite second person happen not to have any overt distinction between Set A and Set B, so that the first person singular in- can be either ergative or absolute, as can polite second person singular la, so that a form like the following is potentially ambiguous:

(120) K- in- loq'o- x la.
  Asp. 1Sg.- love- Ptc. 2Sg.Polite
  'I love you' or 'You love me.'

The ambiguity can be resolved by constructing the object as a nondirect object and giving the verb the intransitivizing suffix -n:

(121) K- in- loq'o-n čeh la.
  Asp.- 1Sg.- love to you
  'I love you.'

Since the verbal form in (121) has no direct object, the first person singular prefix in- can only refer to its subject.

In recent work on ergativity, constructions where an otherwise transitive verb occurs with an oblique object — particularly where the verb has a special marker like Dyirbal -ŋay or Quiché -n, and where there is an absolute rather than an ergative noun phrase — are often referred to as the "anti-passive." Although this term can be rather misleading, for instance in that (unlike the situation with the passive) in the majority of languages with this construction there does not seem to be any difference in the subject status of the ergative and absolute noun phrases of the nonantipassive and antipassive respectively (with the possible exception of more consistent syntactically ergative languages like Dyirbal), it does provide a convenient nomenclature and one that has recently become widespread. The difference between nonantipassive and antipassive can be diagrammed as in Figure 2.

Figure 2. The Antipassive

It should also be noted that although the antipassive is, from a formal viewpoint (oblique object rather than P) a relatively uniform phenomenon across languages, its function can vary quite considerably from language to language. In Walbiri and Circassian, for instance, its main function is to express a difference in sentence-internal semantics, namely incompleteness of the situation described in so far as it affects the object: in (111) there is no suggestion that I actually hit the kangaroo; in (113) there is no suggestion that the boy will complete the plowing of the field. Many nonergative languages have similar pairs of sentences with direct and oblique objects expressing a similar difference in meaning: compare the English glosses with shoot versus shoot at, plow versus plow away at (Anderson 1971).

In Kala Lagau Langgus (Mabuiag, Western Torres Strait Language), an Australian language, another difference in sentence-internal semantics is carried by the antipassive. The antipassive is in complementary distribution with incorporation (used for body parts) and indicates complete affecting of P, which is a predetermined set (Bani and Klokeid 1976: 278):

(122) Ngath ngaungu ithab koei puil pathamadhin.
  I-Erg. myself these big tree-Abs.-Pl. cut-Pl.P-Past
  'I cut down these big trees.'
(123) Ngai ngaungu ithab koei puin pathaidhin.
  I-Abs. myself these big tree-Obl. cut-Past
  'I cut down all these big trees.'

(I am grateful to S. C. Dik for bringing these data to my attention.) We may note that incorporation, discussed below, also has different functions in different languages, although detailed discussion is beyond the scope of the present chapter.

In Dyirbal, on the other hand, the function of the anti-passive is primarily related to discourse structure: anti-passives are not usually found as the first clause of a discourse. But thereafter they can be found where the non-antipassive form is excluded because of the restrictions on coordination, and the like, of an absolute noun phrase with an ergative noun phrase. The following, for example, coordinate the intransitive sentences (56) and (58) with the antipassive versions of (57) and (59), and thus translate the English coordinate structures (52) and (54), which, as already noted, are impossible in Dyirbal using the nonanti-passive form (Dixon 1972: 130):

Bayi yaṛa bani-nʸu, bagun dʸugumbil-
Masc.-Abs. man-Abs. come-Tense Fem.-Dat. woman-
gu balgal-ŋa- nʸu.
Dat. hit- Tense
'The man came here and (he) hit the woman.'
Balan dʸugumbil bani-nʸu, bagul yaṛa- gu
Fem.-Abs. woman-Abs. came-Tense Masc.-Dat. man- Dat.
balgal-ŋa- nʸu.
hit- Tense
'The woman came here and (she) hit the man.'

Given that the prime function of the Dyirbal antipassive is in constructing discourses with all clauses having the same absolute noun phrase expressing the topic, it is not surprising that the antipassive should be paralleled by another derived verbal form allowing coordination of two sentences where A of the first is coreferential with S or P of the second, namely the -ŋura form, which allows the following coordination possibility (Dixon 1972: 77-78):

Bala yugu baŋgul yaṛa- ŋgu mada- n,
Abs. stick-Abs. Masc.-Erg. man- Erg. throw- Tense
'The man threw the stick and went uphill.'

A final set of pairs where absolute and ergative alternate without any necessary difference in degree of agentivity is where one member of each pair involves incorporation of P into the verb, that is, the formation of a complex intransitive verb from a construction with a verb and P. A particularly clear example is provided by Chukchee, a language of Eastern Siberia (Chukotka Peninsula):

(127) Tumg -e nantəwatən kupre- n.
  friend -Erg. they-set-it net- Abs.
  'The friends set the net.'
(128) Tumg -ət kopra- ntəwatg'at.
  friend -Abs. net- they-set
  'The friends set the net/engaged in net-setting.'

(127) is an ergative construction: P is a separate word in the absolute case, A is in the ergative case, and the verb agrees with both A and P. In (128), the object has been incorporated into the verb; it is part of the same word as the verb. In Chukchee this can be seen quite clearly because instead of the vocalism of the separate word kupre-, vowel harmony assimilates the vocalism to that of the verb, giving kopra- (in Chukchee, vowel harmony is coextensive with word boundaries). In the incorporated version (128), the construction is intransitive, since P no longer exists as a separate word, so the subject is in the absolute case; moreover, the verb is in the appropriate form for an intransitive verb with a third person plural subject (and, of course, no direct object). Not all languages have such clear phonological and morphological evidence of incorporation, but it is probably in this way that we should analyze the "direct objects" that occur in the following examples, (129-130) from Tongan (Churchward 1953: 77) and (131) from Quiché:

(129) Na'e inu 'a e kava 'e Sione.
  Past drink Abs. the kava Erg. John
  'John drank the kava.'
(130) Na'e inu kava 'a Sione.
  Past drink kava Abs. John
  'John drank kava/engaged in kava-drinking.'
K- e- q'oxoma- n le čirimiy le
Asp.- 3Pl.Abs.- play- Antipass. the chirimí the
ax q'oxom- ab'.
player- Pl.
'The players played the chirimí/engaged in chirimí-playing.'

In some languages, however, even incorporated Ps still require an ergative subject, as in Dargi (Abdullaev 1971: 82), a Northeast Caucasian language:

(132) Nu-ni q'as- barra.
  I-Erg. decision- made
  'I decided.'

More literally, (132) would be 'I made a decision', except that P is incorporated into the verb; nonetheless, the subject is in the ergative.

The ergative subjects in sentences like Basque (103), Walbiri (111), and Dargi (132) might provide some slight evidence for a link between ergativity and agentivity; yet these examples can equally, or even better, be accounted for in terms of intransitive derivatives of transitive verbs which retain some properties of the transitive from which they derive, in particular the ergative subject. But the weight of the examples discussed so far is strongly against identification of ergativity and agentivity.

We must now turn to some examples that suggest there is some connection, in some languages, between ergativity and agentivity, although this connection is by no means identity. It should be noted initially that there is bound to be a high correlation between ergative noun phrases and agentive noun phrases simply because As are typically high on the scale of agentivity; this is a factor quite independent of ergativity, however: as pointed out to me by Susumu Kuno, Japanese has a rather strong agentivity requirement on As, but no morphological or syntactic ergativity correlating with this. In some languages, however, one finds that A goes into the ergative case, and that S stands sometimes in the absolute case, sometimes in the ergative case, the ergative being used where S is more agentive, as in the following pair of Batsbi sentences (Batsbi is a Northeast Caucasian language) (Dešeriev 1953: 226):

(133) Tχo naizdraχ qitra.
  we-Abs. to-the-ground fell
  'We fell to the ground (unintentionally, not our fault).'

(134) Atχo naizdraχ qitra.
  we-Erg. to-the-ground fell
  'We fell to the ground (intentionally, through our own carelessness).'

Such examples suggest that languages may evince a connection between ergativity and agentivity, although there is no necessary connection. Diachronically, this could arise from the high correlation between As and agentive noun phrases, so that a marker of A status could easily be reinterpreted as a marker of agentivity, and vice versa. (Klimov 1973 argues that morphological ergativity almost invariably arises diachronically from agentivity; for a critical assessment of his arguments, see Comrie 1976b.) This process could be facilitated by the presence of objectless transitive verbs with ergative subjects (cf. Basque sentence 103), or by verbs with incorporated objects and ergative subjects (cf. Dargi example 132); note moreover that in Dargi, many such examples are probably to be analyzed synchronically as morphologically complex intransitive verbs, rather than as involving synchronically a derivation via incorporation. Lest it should be felt that this chapter is unjustly negative with regard to the connection between ergativity and agentivity, I should point out that the close relation between ergativity and agentivity in some languages is counterbalanced by a close relation between ergativity and nonagentivity in some others (see further section 7.4.2), e.g., Dalabon, an Australian language spoken in Arnhem Land, where the ergative suffix -yi occurs with all inanimate As, but not with all animate As, although the latter are more agentive than the former.

7.3. Ergative-Absolute and Nominative-Accusative Systems: Diachronic Relationships

When linguists, for the most part speakers of nonergative languages, first started investigating ergativity, one of the main questions that concerned them was: how can a language come to be ergative? There was thus a presupposition that ergativity is somehow aberrant and its deviation from the norm has to be explained as the result of some historical development from a more normal language-type. Had linguists developed first among speakers of ergative languages, the question might well have been posed in the opposite way: How could the aberrant nominative-accusative system arise from the normal ergative-absolute system? In this section, I shall try to show that both directions of development are possible and provide examples of both, although for the majority of ergative languages, our knowledge of their history is so limited that we can do little more than speculate on earlier stages.

7.3.1. Passive and Ergative

From the viewpoint of a nominative-accusative language, the ergative-absolute system is most reminiscent of passive constructions in a nominative-accusative language. In this part of the discussion, I shall continue to use A and P to refer to the two arguments of a transitive verb irrespective of which is subject (or which is treated like S for specific morphological or syntactic phenomena). Thus, in both active and passive sentences of English like Mary kissed John and John was kissed by Mary, the same noun phrase (here, Mary) will be A in both, and the same noun phrase (here, John) will be P in both. In nominative-accusative morphology, it is typical (though not quite universal) for the nominative to be less complex morphologically, abbreviated below as (-), than the accusative and oblique (+), whereas in ergative-absolute morphology it is typical for the absolute to be less complex morphologically (-) and the ergative and oblique more complex (+). The similarity between the passive construction in a nominative-accusative language and the ergative construction may be illustrated by the following structures:

(135) Nominative-accusative: active transitive
  A P V
  Nom. (-) Acc. (+)  
(136) Nominative-accusative: intransitive
  S V
  Nom. (-)  
(137) Nominative-accusative: passive
  P A V
  Nom. (-) Obl. (+)  
(138) Ergative-absolute: transitive
  A P V
  Erg. (+) Abs. (-)  
(139) Ergative-absolute: intransitive
  S V
Abs. (-)  

(The word order is not relevant to this argument, although the typical, but by no means universal, difference in the word order of passive and ergative constructions as illustrated by the actual order used in 137 and 138 is one argument against their identification.) In both the passive and ergative constructions, the A noun phrase is more complex morphologically than the P noun phrase, whereas the inverse is usually true in the active construction of the nominative-accusative system. In both the passive and the ergative, P is morphologically identical to S, whereas in the second nominative-accusative construction A is morphologically identical to S. The only difference would be that the passive contrasts with an active nominative-accusative construction, whereas the ergative does not enter into such a contrast; in other words, the ergative would be like an "obligatory" passive construction. From what has been said so far, it should be clear that, for the majority of languages with ergativity, this is not acceptable as a synchronic analysis: in most languages with ergativity, the majority of subject properties identify the A of the transitive verb as its subject, whereas in the passive construction the majority of subject properties identify the P as the subject of the sentence. Only for languages with consistent syntactic ergativity, such as Dyirbal, is the obligatory passive analysis a candidate for analysis of the ergative construction.

Another piece of evidence often cited against the obligatory passive analysis is that many languages have passive constructions contrasting with the active ergative, as in Basque:

(140) Haurra igorria da.
  child-Abs. sent-Pass. Aux.-3Sg.S
  'The child was sent.'

However, this is not an absolutely telling argument, since there are languages with more than one passive construction, so it could be the case that Basque, for instance, had two passives and no active. However, in the Basque ergative construction subject tests identify A as subject, whereas in the Basque passive they identify P as subject, i.e., the subject tests do not require the dubious reinforcement from the presence of a distinct passive form.

Diachronically, however, there is good evidence that some instances of morphological ergativity do arise from passive constructions, through loss of the corresponding actives. The best evidence comes from the development of ergativity in the Indo-Iranian languages (Pirejko 1968), especially in the Indic languages; these languages typically are ergative only in the past tense forms or, more restrictedly still, in the perfective past forms. In earlier stages of these languages, for instance Sanskrit, such tense/aspects had both active and passive forms. However, the passive forms increased in frequency relative to the active forms, until eventually the active forms became obsolete. Parallel with this development, the subject properties were transferred (no doubt gradually, although it may not be possible to reconstruct all the details) from the P noun phrase (of the passive construction) to the A noun phrase (of the ergative construction). In many modern Indo-Iranian languages, the only subject properties still characterizing P are its less complex morphology and verb-agreement with P rather than with the A, as in Hindi:

(141) Is laṛke ne pustak paṛhī.
  this boy Erg. book-(Fem.)-Abs. read-Fem.
  'This boy read a book.'

In many Indo-Iranian languages, even these last vestiges of ergativity (purely morphological) are being lost, often in stages which, synchronically, seem to make little sense, but can be explained in terms of the gradual loss of ergativity, such as constructions where the auxiliary clitic agrees in person and number with A, whereas the participial form of the verb agrees in gender (sometimes also in number) with P, as in the Iranian language Munjani (Grjunberg 1972: 431):

(142) Mən vow ž -əgh- əm.
  I-Erg. she-Pl. hit Fem. 1Sg.S
  'I hit her.'

In Modern Persian, ergativity has been lost completely, so that in the development from the oldest Persian to the modern

Figure 3. Development of Ergative from Passive

language one can see the rise and fall of ergativity. The development from passive to ergative can be represented schematically as in Figure 3. The further stage found in Modern Persian involves loss even of the morphological ergative system.

A similar development has been posited for the Polynesian languages (Hohepa 1969), though there it is a much more recent development than in Indo-Iranian. Some Polynesian languages are still almost completely nominative-accusative; some are well on the way to becoming ergative (the passive is very frequent relative to the active; subject properties are being transferred away from P in this construction to A); while others are completely ergative; in some, there has even been realignment of the morphology, in the loss of the passive suffix on verbs in the ergative construction, although the noun phrase morphology does not yet show any signs of loss, as it does in many Indo-Iranian languages (Polynesian languages do not typically have verb-agreement, so there is no possibility of ergativity being manifested there).

Further evidence for the passive origin of ergativity in some languages can be seen in the fact that certain correlations that hold between passive and other features hold equally between ergativity in these languages and these same features. For instance, passive tends to be more frequent with the perfect aspect (in particular the perfect of result) than with other verbal forms (Comrie 1976a: 84-86); in the development of the ergative in Indo-Iranian, the ergative is first found only in the perfect, and the later wider range of ergativity follows from the extension of these forms to oust all other past perfective forms (as in most Indic languages with ergativity), or even all other past forms (as in most Iranian languages with ergativity). Thus the current split ergativity along a tense/aspect basis in most Indo-Iranian languages with ergativity is a reflex of this correlation. Second, in many languages passives can only be formed, or can only be formed readily, or are formed much more readily, from dynamic, as opposed to stative, verbs. In Samoan, a Polynesian language, the ergative construction is used with more dynamic verbs, the nominative-accusative construction with more stative verbs (I am grateful to Dixie C. Samasoni for discussion of the Samoan examples):

(143) Ua sogi e le tama 0 le ufi.
  Tense cut Erg. the boy Abs. the yam
  'The boy cut the yam.'
(144) Ua alofa 0 le tama 'i le teine.
  Tense love Nom. the boy Acc. the girl
  'The boy loves the girl.'

The absolute and nominative are morphologically identical (in Samoan, with no preposition, contrasting with ergative e), as is expected from the diachronic development; both develop from subjects at an earlier stage of the language, the absolute from the subject of a passive construction, the nominative from the subject of an active construction. Synchronically, in Samoan, both (143) and (144) have passives with the same syntactic and morphological structure as each other, suggesting that the actives have the same syntactic structure as each other too:

(145) Ua sogi- ina 0 le ufi e le tama.
  Tense cut- Pass. Abs. the yam by the boy
  'The yam was cut by the boy.'
(146) Ua alofa- gia 0 le teine e le tama.
  Tense love- Pass. Abs. the girl by the boy
  'The girl is loved by the boy.'

Having observed the possible development from passive to ergative, we may note the opposite possibility whereby a nominative-accusative system could develop from the antipassive. Just as ergativity can develop through replacement of an active construction by a passive construction (and transference of subject properties to the passive agent), so a nominative-accusative system could develop through replacement of an ergative construction by an antipassive construction (though without the need for transference of subject properties). This has been suggested as a possible origin for split ergativity in Georgian, where the aorist tense forms seem to be earlier than the present tense forms; accordingly the ergative system for noun case-marking in the former might well be earlier than the nominative-accusative system in the latter. The nominative-accusative system in the present tense system would then represent widening of the aspectual value of the antipassive to replace all nonperfective (nonaorist) aspectual forms.

7.3.2. Nominalizations and Ergative

In many discussions of the origin of ergative constructions, nominalization is advanced as one possible source for ergativity. Often, the details of the development are not worked out; in particular, no explanation is offered for the crucial difference between the ergative-absolute and other systems, namely the differential treatment of A and S, the latter being treated in the same way as P. In many languages nominalization results in the replacement of all subjects (S and A) by the genitive, with direct objects being unaffected (cf. Comrie 1976c), and clearly this would not lead to an ergative system.

The same proviso actually applies also to the passive as a possible source of ergativity. The possibility of this development is fostered by the fact that in a large number of languages passives exist only of transitive sentences, so that reinterpretation of a passive as an (ergative) active can only apply to sentences with transitive verbs. In some languages, passives can be formed equally from intransitive sentences, and reinterpret ation of the passive agent as subject in such a language would not lead to ergativity. This development seems to have taken place in some North Russian dialects (Timberlake 1976).

In examining nominalization, one must therefore be careful to ask what feature of the nominalization construction is responsible for the ergative-absolute system. In English, which has both prenominal and postnominal genitives, both active and passive nominalizations of transitive verbs are possible:

(147) the enemy's arrival
(148) the enemy's destruction of the city
(149) the city's destruction by the enemy

In Russian, on the other hand, which has very much the same nominalization construction as English, but which does not have two different kinds of genitive, in effect only the passive nominalization is possible for transitive verbs.

(150) priezd vraga
  arrival enemy-of
  'the enemy's arrival'

(151) razrušenie goroda vragom
  destruction city-of enemy-by
  'the city's destruction by the enemy'

Russian has no direct equivalent of the English active nominal the enemy's destruction of the city; that is, as far as nominalizations are concerned, Russian has in effect an ergative system: S and P stand in the genitive (functioning as an absolute), while A stands in the instrumental (functioning as an ergative). The verbal system of Russian, however, does not share this kind of ergativity, although ergativity would be a logically possible development if finite verbal forms were to be replaced by nominalizations. I am not aware of any actual instances where ergativity in the verbal system arises from such an ergative nominalization construction.

In many ergative languages, the transitive sentence construction can be derived etymologically, either with certainty or with a high degree of plausibility, from a nominalization, whereas the intransitive sentence construction does not derive etymologically from a nominalization (Allen 1964). In one sense, this is ergativity via nominalization, but it should be noted that nominalization does not explain the differential treatment of transitives and intransitives in such instances — the explanation is required at an even earlier stage; namely, why are only transitive constructions replaced by nominalizations, and not intransitive constructions? The kind of situation discussed in this paragraph holds, for instance, in the development of ergativity in Iranian, from constructions like Old Persian manā kṛtam 'I did this, this was done by me', where manā is the genitive of the first person singular pronoun, and kṛtam is a past participle used as a noun phrase, i.e., more literally something like 'this is the my-having-been-done-thing'. Indeed, Cardona (1970) suggests that this nominalization was reinterpreted as a passive in the development of Iranian (interpretation of manā as passive agent), so that the ergative construction would derive most directly from a passive, and only indirectly from a nominalization. (A further factor in the development of ergativity in Indo-Iranian is that the past participle is already incipiently ergative in Proto-Indo-European, having active meaning when formed from an intransitive verb, 'having come', but passive meaning when formed from a transitive verb, 'having been killed'. Compare the discussion of ergativity in derivational morphology in section. 7.4.3.) In all instances known to me where ergativity is alleged to derive via nominalization, it seems at least a priori plausible that a development similar to that in Iranian occurred, with nominalization originally a device for forming passive constructions, and subsequent reinterpretation of these passives as ergatives.

Moreover, there are attested examples where nominalization is the prime factor in the destruction of ergativity, its replacement by the nominative-accusative system. This development can occur in languages where nominalization involves putting both S and A into the genitive, whereas P remains in the absolute case. The development is attested, for instance, in the verbal affixes of Chol. The Set A affixes are used in most Mayan languages as ergative affixes on verbs and as possessive affixes on nouns (including nominalizations, where the affixes refer to S and A alike — in this respect, the languages are syntactically nominative-accusative), the set B affixes as absolute affixes on verbs. In Chol (see examples 77-82), the past tense retains the earlier system; the present tense, however, has been replaced by a nominalization, so that all subjects have the Set A affixes (functionally, nominative), while direct objects, as usual unaffected by nominalization in Mayan languages, have the Set B affixes. Similarly in Jacaltec, many subordinate clauses derive from nominalizatlons, so that, whereas main clauses are ergative-absolute, these subordinate clauses have the nominative-accusative system deriving etymologically via nominalization (examples from Craig 1976: 102):

(152) Sikinax hač.
  tired 2Sg.Abs.
  'You are tired.'
(153) Šk- w- ila.
  Asp.- 2Sg.Abs. 1Sg.Erg.- see
  'I saw you.'
(154) Šk- to ku- saxčox.
  Asp.- 1Pl.Abs. go 1Pl.Nom.- play
  'We went to play.'
(155) Šk- to hač ku- kolo'.
  Asp.- 1Pl.Abs. go 2Sg.Acc. 1Pl.Nom.- help
  'We went to help you.'

In these Jacaltec examples, the Set A affixes are w- (first person singular) and ku- (first person plural); the set B affixes are (h) (second person singular) and (first person plural).

Since genitive noun phrases are typically more complex morphologically than absolute noun phrases, nominative-accusative systems that arise in the way just discussed would be expected to have a nominative (etymologically, genitive) more complex morphologically than the accusative (etymologically, absolute), unlike the general pattern in nominative-accusative languages where the nominative is less complex. There are, however, some nominative-accusative languages with the nominative more complex morphologically and the accusative less complex, such as some of the Yuman languages with nominative in -č (S and A) and accusative with no overt marker; nominalization is a possible diachronic explanation for such systems. (I am grateful to Pamela Munro for drawing the Yuman data to my attention.)

In conclusion to section 7.3: I have tried to show that the diachronic relations between ergative-absolute and nominative-accusative systems are rather more complex than is often assumed: there are indeed possibilities for a natural development from nominative-accusative to ergative-absolute, but equally there are possibilities for a natural development in the opposite direction.

7.4. Synchronic Function of Ergative Case-Marking

Having looked at some possible diachronic explanations for the ergative-absolute system, in this section I wish to examine some possible synchronic explanations for this kind of system, in particular with regard to noun-phrase case-marking. The question posed here is thus: is there any sense to the ergative-absolute case-marking system internal to this kind of case-marking system? We take as our starting point the notion that one of the functions of a case-marking system, in so far as it applies to S, A, and P, is to enable recovery from the form of the sentence of the syntactic relations contained in it, that is, to enable identification of the S, A, and P of the sentence. (There are, of course, other ways of coding the difference between S, A, and P, for example by verb-agreement or word order; I return to verb-agreement briefly at the end of this section, concentrating in the meantime on case-marking.) Various factors can operate to make identification of S, A, and P more difficult — for instance, in the absence of overt cues for S, A, or P status, a sentence containing both A and P will give more difficulty than one allowing only S, or only P. Therefore one might expect that utilization of overt markers of syntactic relations would be most likely to occur, across the languages of the world, in those instances where confusion of syntactic relations would be most likely in the absence of such overt markers. I shall try to illustrate this general principle, in its relation to ergativity, in the following sections.

7.4.1. Ergative and Antiergative

As already indicated, one factor complicating the identification of S, A, and P, in the absence of overt markers, is a syntactic construction allowing the presence of both A and P (in the same clause). Thus the transitive construction A-P-V presents more problems from this viewpoint than does the intransitive construction S-V, since in the former there is greater need to hit at least one of A and P to avoid misinterpretation (e.g., of John hit Mary as Mary hit John). If we refer back to Figure 1, we note that of the five systems presented there for marking of S, A, and P in transitive and intransitive sentences, two of these types are very highly motivated from the viewpoint of the discriminatory function of case-marking (discrimination of S, A, and P): type (b) (nominative-accusative) and type (c) (ergative-absolute) both distinguish A from P in the A-P-V (transitive) construction, in the one type (nominative-accusative) using a special form for P of the transitive construction and the same form for both A and S, in the other type (ergative-absolute) using a special form for A of the transitive construction and the same form for both S and P; both types have only two morphological categories, with which they make the relevant distinction among S, A, and P where it is most needed (clauses with both A and P), and from this viewpoint it is irrelevant whether S is identified morphologically with A or with P. In the majority of nominative-accusative languages the accusative is morphologically more complex than the nominative, and in nearly all ergative-absolute languages the ergative is morphologically more complex than the absolute, so that we can make the general claim: nominative-accusative languages distinguish A from P in the A-P-V construction by marking P; ergative languages distinguish A from P in the A-P-V construction by marking A. Of the other types given in Figure 1: type (a) does not use noun-phrase case-marking to distinguish S, A, and P (and typically languages with this system use some other means, e.g., verb-agreement or word order, to indicate this distinction); type (d) is, from the viewpoint of the discriminatory function of case-marking, unnecessarily explicit, but at least it does make the distinction between A and P of a transitive construction (type d appears also to be extremely rare among the languages of the world); type (3) seems to be nonexistent, and certainly the discussion of the present section would suggest that it should be at least extremely rare, if not totally excluded: A and P are not distinguished in the transitive construction, while a functionally unnecessary distinction between S on the one hand and A and P of the transitive construction on the other is maintained; that is, the distinction that it is most important to maintain is not in fact maintained, while an unnecessary distinction is made elsewhere. The absence or near-absence of type (e) is one piece of evidence in favor of this approach to the ergative case-marking system.

Another piece of evidence comes from the morphology of subjectless sentences. In some languages, subjectless sentences form a distinct sentence-type, with syntactic properties distinguishing them from sentences with subjects. Thus in addition to the constructions S-V and A-P-V, for such languages one should also consider the type P-V. (There is also the type V in such languages, lacking both subject and direct object, but since this leaves no possibility for case-marking, it is not relevant to the present discussion.) This last type, P-V, is another type where, parallel to S-V, there is no need to have an overt marker for the direct object, since there is no subject with which it can be confused in the clause, (in languages of this type, S-V and P-V are typically distinguished from one another by the fact that only certain verbs, or certain verbal forms, appear in P-V.) Therefore one might expect to find languages where a special marker is used for the direct object if and only if there is also a subject in the same clause, subjects and direct objects being otherwise unmarked. This would be the mirror-image of the ergative-absolute system, and might therefore be termed the 'antiergative-absolute' system, the antiergative being used for the P of A-P-V (compare the ergative for the A of A-P-V), and the absolute for the S of S-V, the A of A-P-V, and for the P of P-V (compare the absolute in the ergative-absolute system for the P of P-V, if this exists as a separate sentence-type, and of A-P-V and for the S of S-V), as in Figure 4.

Languages of this antiergative type do indeed exist, for instance Finnish, where, with singular pronominal noun phrases, the so-called accusative in -n is used for P where there is an A, but not where there is no A (e.g., in imperatives):

(156) Maija tuli.
  Maija-Abs. came
  'Maija came.'
(157) Maija söi kala-n.
  Maija-Abs. ate fish-Antierg.
  'Maija ate the fish.'

Figure 4. Ergative and Antiergative
(158) Syö kala!
  eat-Imperat. fish-Abs.
  'Eat the fish!'

Further discussion of the antiergative-absolute system with more exemplification, will be found in Comrie 1975; see also Bechert 1977, which became available to me only after completion of the body of this chapter. For present purposes, the important point to note is that the existence of the antiergative-absolute system provides further evidence in favor of the approach toward the synchronic explanation of ergativity advocated in this section.

7.4.2. Case-Marking and Interpretation Probability

So far, we have been assuming that case-marking will tend to be used to distinguish A from P, wherever both are present in the same clause, since otherwise confusion would be more likely. Clearly, however, the possibility of confusion also depends on the kinds of noun phrases used as A and P. Thus, given a transitive verb like eat and two noun phrases, one animate like man and the other inanimate like bread, even in the absence of any overt markers there would be a much greater likelihood of someone speaking about the man eating the bread than about the bread eating the man; similarly, the hearer would normally assign the former interpretation to an otherwise ambiguous sentence of this type. We might therefore expect to find languages where case-marking of A or P of the A-P-V construction is made only where the likelihood of confusion, allowing for the fact that certain situations are so unlikely in absolute or relative terms as to be able to be excluded from consideration, is still very high. Thus there would be no marking if we were to talk of the man eating the bread, but there would be marking if we were to talk of the man eating a lion (given that lions can readily eat men), and even more so if we wanted to talk about (metaphorical or science-fiction) bread eating the man. There are indeed languages which exhibit case-marking systems of this type. Since this chapter is on ergativity, I shall concentrate on such case-marking of A. (For a discussion of case-marking of direct objects from this viewpoint, see Comrie, forthcoming.)

Haiman (forthcoming) notes that in Hua, a language of the New Guinea Eastern Highlands, there is an ergative suffix -bamu', whereas the absolute case has no overt marker:

(159) Busa' rmie.
  Busa' he-went-down
  'Busa' went down.'
(160) Busa'-bamu' egbie.
  Busa'-Erg. he-hit-him
  'Busa' hit him.'
(161) Busa' egbie.
  Busa' he-hit-him
  'He hit Busa'.'

However, Haiman notes that although the ergative suffix can only be used with A (and never with S or P), it is not necessary to use this suffix with A, since it can he omitted where confusion would not otherwise result. Thus, of the following two sentences, (162) with -bamu', can only mean 'Who gave it to him?', whereas (l63), with no marking on kzo' 'who', can mean either 'Whom did he give it to?' (Hua does not distinguish morphologically between direct and indirect object) or, in the appropriate context, 'Who gave it to him?':

(162) Kzo'- bamu' mie?
  who Erg. he-gave-it-to-him
(163) Kzo' mie?
  who he-gave-it-to-him Ergativity and the Animacy Hierarchy

In many languages where case-marking of A and P is restricted to situations in which confusion is most likely, there is a more rigid specification of what class of situations is considered to be most likely to engender confusion, rather than the more general specification (where the speaker feels clarification is required) that seems to hold for Hua.

There seems to be a general supposition in human discourse that certain entities are inherently more agentive than others, and as such inherently more likely to appear as A of a transitive verb and less likely to appear as P of a transitive verb. The mainstay of this supposition is the animacy (agentivity) hierarchy, vhich claims basically that more animate entities will tend to act upon less animate entities rather than vice versa. The animacy hierarchy has first and second person pronouns at the top (i.e., active participants in the speech act are most agentive), followed by third person personal pronouns, followed by other human noun phrases, then other animate noun phrases, with inanimate noun phrases at the bottom. (Languages may make finer or less fine distinctions, and different languages vary with regard to some of the finer distinctions, but the general schema remains the same.) Given a transitive verb with A and P, there are three possibilities with regard to the animacy hierarchy: (a) A is higher on the hierarchy than P; (b) both A and P are of equal animacy; (c) P is higher on the hierarchy than A. In terms of the animacy hierarchy, (a) is expected, whereas (c) is unexpected, while no prediction is made with respect to (b). In some languages, one finds that no case-marking is used in (a), whereas case-marking is used in (c), and usually also in (b). This system obtains in Dalabon, an Australian language of the Northern Territory (Silverstein 1976: 129):

(164) Buluŋan ga'manbuniŋ.
  my-father he-made-it
  'My father made it.'

In (164), there is an animate noun phrase and an inanimate pronoun coded in the verb form; the expectation is that the animate noun phrase will be A and the inanimate noun phrase P, and as this expectation is borne out, there is no case-marking of A.

(165) buluŋan -yi wuduwud ga'nan.
  my-father -Erg. baby he-looks-at-him
  'My father is looking at the baby.'

In (165), both noun phrases are human, and, at least in Dalabon, are treated as being at the same level in the animacy hierarchy: the hierarchy thus makes no prediction as to which is more likely to be A and which P, so case-marking is required, and since Dalabon marks A in such instances, we have an ergative case, in -yi. As noted by Silverstein (1916: 129), inanimate As are invariably lower than or on the same level as their Ps, since inanimate is the lowest position on the hierarchy, and this accounts for the invariable marking of inanaimate As of transitive verbs with the ergative suffix -yi in Dalabon, contrary to the positive correlation noted in some other languages between ergativity and agentivity.

In some languages, this kind of case-marking is determined by the absolute position of a noun phrase on the hierarchy, rather than by the relative position of A and P on the hierarchy: noun phrases high on the hierarchy will not be marked when they are A, but will be marked (accusative) when they are P; noun phrases low in the hierarchy will not be marked when they are P, but will be marked (ergative) when they are A. We have already observed a language of this kind, namely Dyirbal (section 7.1.3), where first and second person pronouns and nonpronominal noun phrases have an ergative-absolute morphology, the latter being lower on the animacy hierarchy than the former. Further examples of this pattern, drawn mainly from Australian languages, are given by Silverstein (1976), in terms of a more detailed explanatory hypothesis similar to that advanced here. Silverstein also notes that in some languages there may be an area of overlap between the two systems, that is, some noun phrases that have both marked ergative and marked accusative cases, with an unmarked case used only as S, thus giving rise to a tripartite system as in Figure 1 (d); Dyirbal has this system for the human interrogative pronoun 'who?': wanʸa (S), wanʸdʸu (A), wanʸuna (P) (Dixon 1972: 53).

Note that this differs in terms of morphological marking from the Motu system given in (9) and (10) as an illustration of type (d) in Figure 1: in Motu, P, rather than S, is less complex morphologically. In Motu, moreover, this tripartite system is much more widespread than in languages with split ergativity involving limited overlap, so that all in all the tripartite case-marking system may well be a rather different kind of phenomenon in the two kinds of language (though involving some kind of intersection of ergative-absolute and nominative-accusative in both).

7.4.3. Synchronic Explanation of Other Manifestations of Ergativity

The discriminatory function of the ergative-absolute system as discussed in the present section relates primarily to morphological ergativity, in particular to ergativity in the case-marking system. It can also be extended to ergativity in verb-agreement, particularly in those languages (such as the Mayan languages) where there is no case-marking of S, A, and P, and verb-agreement is the prime indicator of S, A, and P (with word order playing a secondary role, more especially relevant when both A and P are third person). Indeed, in languages where verb-agreement does not play this role of distinguishing A from P, it might be more helpful to regard ergative-absolute verb-agreement as an instance of syntactic, rather than morphological, ergativity; the boundary is, in any event, rather fluid, and different grammatical traditions differ as to whether the use of morphological categories is properly part of morphology or rather part of syntax. Thus the major phenomenon still requiring an explanation — and one for which no explanation will be put forward here with any confidence — is syntactic ergativity, especially bearing in mind that syntactic ergativity is much less common, across the languages of the world, than is morphological ergativity. As noted in section 7.0.3, there is really no statistical basis for saying that morphological ergativity is abnormal, whereas there may be a statistical basis for arguing that syntactic ergativity is abnormal.

In section 7.1, we noted that one kind of syntactic ergativity is, however, quite widespread in languages, even in languages which are otherwise not ergative, namely ergativity in derivational morphology. I cited the example of formation of compound nouns by incorporation, where English allows incorporation of P of transitive verbs (e.g., fox-hunting) and of S of intransitive verbs (e.g., bird-chirping), that is, allows incorporation of absolute noun phrases. In looking at incorporation across languages, including languages where this is much more productive and much more a central syntactic process of the language than in English, one finds that P is easiest to incorporate, followed by S, with A most resistant to incorporation. This suggests that incorporation may originate with P, with subsequent spread to S, followed only very occasionally by spread to A. If we imagine an early stage where incorporation is restricted to P, then it follows that incorporation is restricted to transitive verbs; intransitive verbs will not be amenable to incorporation, which thus imposes a syntactic restriction on a morphological process, although there is no morphological reason for this restriction to exist, that is, incorporation applied to an intransitive verb is syntactically ill formed, but morphologically well formed. A logical development from this stage would be for incorporation to be extended to intransitive verbs; here, however, the only candidate for incorporation is S. As far as transitive verbs are concerned, the original situation is maintained, so that overall an ergative system is built up, although the motivation for the change would not be ergativity itself, but rather pressure against accidental morphological gaps. There is evidence for a development of this kind within Modern English derivational morphology. Although it is possible to incorporate both P and S to form compound nouns in English, the forms with incorporated S (e.g., bird-chirping) are much less natural than those with incorporated P (e.g., fox-hunting). English can form adjectives referring to a quality of the P of a verb with the suffix -able, e.g., washable 'able to be washed', not 'able to wash'; in addition, there are a few forms from intransitive verbs, and here the quality is attributed to S, e.g., perishable 'likely to perish'. The suffix -ee functions similarly, e.g., employee 'one who is employed', not 'one who employs', escapee 'one who has escaped', and, at least in some idiolects, standee 'one who stands', used of a standing passenger (though the validity of this last example is somewhat weakened by the use of the term dilutee in Britain during the Second World War to refer to an unskilled worker who diluted the existing work-force, rather than to those who were thus diluted). Thus this extension from P only to P and S seems to be an ongoing process in current English, and the explanation suggested here may be able to account for instances of syntactic ergativity involving derivational morphology, though extensions beyond this into more clearly syntactic phenomena are questionable, given the requirement of an initial stage where subjects (in most languages, S and A) are not amenable to the syntactic process in question. (Compare the claim in Keenan and Comrie 1977 that relative clause formation, and possibly many other syntactic processes, apply most readily to subjects, transitive or intransitive.)

In treating derivational morphology, we are approaching, if not entering, the field of the lexicon. In much recent vork on ergativity, or at least in much recent work using the term ergativity, the term has been extended to cover sets of sentences like the following English pair:

(166) John broke the window.
(167) The window broke.

The claim would be that John in (166) is ergative, while the window in both (166) and (167) would be absolute. This use of the term can be very misleading, in that ergativity does have a reasonably well defined traditional meaning in its application to case-marking and verb-agreement, and also in that this extended use of the term conceals the real differences between the morphology of languages like English and Basque, the real differences between the syntax of languages like English and Dyirbal; probably all languages have pairs of sentences like (164) and (165) in English, though usually with a morphological indicator of the difference between the transitive and intransitive verb, unlike English. Incidentally, ergative languages (morphological only or both morphological and syntactic) tend equally to have distinct transitive and intransitive forms in such pairs. But if we compare the above English pair with the following pair, there may be some sense in which we can use the term ergativity here:

(168) John ate a pie.
(169) John ate.

In (168) and (169), the S of the intransitive is identified, in terms of semantic roles, with the A of the transitive verb, whereas in (166) and (167) the S of the intransitive is identified with the P of the transitive. We might refer to the pattern of (166) and (167) as "lexical ergativity," and that of (168) and (169) as a "lexical nominative-accusative system." If we were to use this terminology, however, it would be essential to bear in mind that this is an extension of the traditional terminology, an extension that moreover seems to have little value in terms of typological classification — presumably all languages have a large amount of "lexical ergativity" — unlike morphological and syntactic ergativity, which provide useful means of grouping together a number of parameters along which languages may vary. Compare Dixon (1977b) on ergative-absolute and nominative-accusative patterns in the Dyirbal lexicon.

7.5. Conclusions

The discussion of section 7.4, moving from presentation of data to attempting explanations for the kind of data found, necessarily became much more speculative than that of the preceding sections. There is much in the area of the internal cohesion of ergative systems that still requires explanation; the formation and testing of such explanatory hypotheses is an important task for future research on ergativity.

However, much still remains to be done also in terms of gathering and processing data on ergative languages. By a combination of geographical, social, and political accidents, ergative languages have until recently been largely neglected (the major European languages have ergativity only in small parts of the derivational morphology), with information on the syntax of these languages being particularly scarce. One example will suffice to show how this scarcity of syntactic information can bias our general view on ergativity: the availability of a description of Dyirbal syntax, in the shape of Dixon's (1972) monograph, has revolutionized our view of ergativity, since for the first time it has become apparent that there is a language with near-consistent syntactic ergativity (though it has to be realized that not all ergative languages are like Dyirbal; the discovery of a continuum from purely morphological to near-consistent syntactic ergativity has also proved an important stage in the development of our ideas on ergativity). In this chapter I have tried to present in a systematic way the range of phenomena which, in my opinion, characterize and correlate with ergativity. It may be that further data from other languages will force us to widen this framework even further, but in the meantime it is essential that more theoretical, explanatory discussions of ergativity should take into account the range of phenomena which we at present know to be encompassed within ergativity. If this chapter does something to help bring this about, it will have served its purpose.


This chapter is a revised and expanded version of a lecture on "Ergativity" given originally as one of the American Council of Learned Societies Distinguished Scholar lectures at the 1976 Linguistic Institute of the Linguistic Society of America, Oswego, New York. This talk was also given at the Department of Modern Languages and Linguistics of Cornell University, the Institute of Linguistics of the Soviet Academy of Sciences, and the Leningrad Section of the Soviet Academy of Sciences. I am grateful to all those who participated in the ensuing formal and informal discussions.

For my initiation into ergativity in Mayan languages I am grateful to Lyle Campbell, and also to the other members of his Mayan Linguistics course at the 1976 Linguistic Institute. I must also acknowledge the debt I owe to the many people who have discussed ergativity with me, orally or by letter, or by exchange of unpublished material. In addition to those cited specifically in the text and in the bibliography, mention should be made of J. C. Catford, Talmy Givón, David E. Johnson, András Komlósy, Igor A. Melchuk, and G. K. Pullum.

My work on non-Slavic languages of the Soviet Union is supported by a grant from the Social Science Research Council, London.

< previous section | Jump to: next section >

  • Linguistics Research Center

    University of Texas at Austin
    PCL 5.556
    Mailcode S5490
    Austin, Texas 78712

  • For comments and inquiries, or to report issues, please contact the Web Master at