The Blue Bird of Ergativity

Scott DELANCEY

University of Oregon

The ergativity industry in modern linguistics was created during the 1970's, taking off from Dixon's description of Dyirbal (1972), and established as a major area of research and theorizing by two survey articles at the end of the decade (Comrie 1978, Dixon 1979). During the 1970's the concept was primarily a plaything of the functional-typological school and of Relational Grammar, but for many years now the "problem of ergativity" has become a concern of syntacticians of all persuasions. My purpose here is to pose the question of whether there is actually a coherent, typologically and theoretically interesting phenomenon of "ergativity" which merits all of the attention given to it (for example in numerous conferences and workshops, such as this one, devoted to "ergativity in X"), and, if there might be, whether it can possibly subsume all of the different linguistic phenomena which are routinely adduced as examples of it.

The question is, when we have a workshop on "ergativity" in various languages, are we pursuing a coherent typological phenomenon, or just collecting specimens? Not that there is anything wrong with collecting and comparing specimens -- basic description of as wide as possible a range of languages and linguistic phenomena is the bedrock foundation of any worthwhile work in linguistics, and, as we all know, has been irresponsibly and shamefully neglected and undervalued in the field over the last generation. But when we limit a collection to certain kinds of specimens, there is a question whether a workshop on "ergativity" is analogous to an effort to collect, say, birds with talons -- an important taxonomic criterion --, birds that swim -- which is taxonomically only marginally relevant, but a very significant functional pattern --, or, say, birds that are blue, which will turn out to be pretty much a useless criterion for any biological purpose.

I would expect, for example, that a series of workshops on something like "subject" in Amazonian -- or Northwest North American, or South Asian, or what have you -- languages would be very likely to have theoretically interesting results. But if I think about what might result from a workshop on "ergativity" in Northwest North America, I suspect that we would end up with a kind of a natural history exhibit, with various strange and exotic specimens on display, but no systematic result issuing from it all. Now a workshop on ergativity in north India would be quite productive -- but that is because all the languages involved are genetically related and have some variant on the same kind of "ergativity", with the same origins. A workshop on ergativity in Himalayan languages might be of interest, but what we would find is several different patterns, with overlapping origins, each replicated with different variations in different languages. But what might come out of such a workshop would not be anything that could be usefully integrated with the outcome of the North Indian workshop, and might or might not show some patterns in common with what turned up in North America.

What do we mean by ergativity?

Dixon provides a version of the standard definition of "ergativity":

a grammatical pattern in which the subject of an intransitive clause is treated in the same way as the object of a transitive clause, and differently from transitive subject. (Dixon 1994:1)

which we might schematize as S = O A. Now, to start with, this phrasing is a bit out of touch with ordinary usage. Ergativity is generally talked of as being about marking, or marked treatment, of the transitive subject, not about the "treatment" of the intransitive subject. But, leaving aside the question of phrasing, the definition is still problematic. If this were actually the definition which we used, it might be more useful than the "ordinary language" use (or might not, that is an empirical question). But, although this definition is still often cited, I do not think anyone would seriously hold to it in dealing with data like these, from Tibetan:

1) kho-s blo=bzang-la gzhus-song

he-ERG Lobsang-LOC hit-PERF

He hit Lobsang.'

2) kho-s blo=bzang bsad-song

he-ERG Lobsang killed-PERF

He killed Lobsang.'

In Tibetan, the O arguments of certain transitive verbs, such as gzhus 'hit', are obligatorily marked with the locative (and dative) postposition -la, while other transitive verbs, such as bsad 'kill', do not allow such marking. Thus some, but not all, O's have the same case form as unmarked S's. Similarly, in some other ergative languages of the Himalayas and South Asia, including Tibeto-Burman languages such as Gurung or Kham and all (as far as I know) of the "ergative" Indic languages, we find some variant of typical "differential object" marking (Comrie 1979, Bossong 1985) in which O arguments are marked (with the dative postposition) if they are human, or animate, or highly topical. For example, in Hindi:

3) us-ne ek laD.kaa dekhaa

he-ERG a boy(NOM) see.sg/msc/perf

'He saw a boy.'

4) us-ne ek laD.ke-ko dekhaa

he-ERG a boy-DAT see.sg/msc/perf

'He saw the boy.'

This pattern, of course, occurs also in unmistakably nominative languages such as Spanish and Klamath, and, for that matter, in the non-ergative non-perfective constructions in Hindi:

5) wo ek laD.kaa dekhtaa hae

he(NOM) a boy(NOM) see.sg/msc/hab be.3sg.msc.pres

'He sees a boy.'

6) wo ek laD.ke-ko dekhtaa hae

he(NOM) a boy-DAT see.sg/msc/hab be.3sg.msc.pres

'He sees the boy.'

Again, this means that in languages such as Gurung, or in the "ergative" tense/aspects of Hindi, some but not all S's are (un)marked like O's. But no one hesitates to call these examples of "ergativity".

Thus, for the standard definition to actually single out the set of languages that we actually consider to be ergative, we need some specification that it is the unmarked O that the S behaves like. And then we are going to have a problem with what we mean by unmarked. I suppose we might agree that in the technical sense of "marked" -- i.e. where it does not necessarily refer to an actual mark -- the O that does not have an actual mark is the unmarked term in the differential object pattern (although according to Hopper and Thompson's (1980) concept of transitivity that would not be the case), but it would be a bit more difficult to make that argument for Tibetan, where the case marking or lack of it on an O argument directly marks underlying semantic role (DeLancey 2001). In fact what we would really want to mean is unmarked, in the sense of not having case marking, because that is actually what we mean when we say ergative -- that there is special marking on the A argument, which distinguishes it from S, which (almost always) has no overt marking. The notion of ergativity is the idea that all languages, or at least all constructions, which manifest this feature belong to a type. Obviously, strictly speaking this is true, in the sense that any feature defines a set. The question is, is the set of all languages which are "ergative" in this sense an interesting or typologically/theoretically useful one? In other words, do they have anything in common besides this superficial feature?

The concept of ergativity traces back to the observation (or "discovery", after the Columbus model) by Hugo Schuchardt (1895) of ergative patterns in Caucasian languages, and later in Basque. But its status as a major construct in typology and linguistic theory dates back only to the 1970's, when the phenomenon was brought to general attention by Dixon, Silverstein, Comrie and others. But from the outset of the ergativity explosion, it was evident that what had been brought to the public eye was not a phenomenon, but a cluster of phenomena which needed to be defined. However, this task of defining the object of study was never seriously undertaken -- I remember very little serious discussion about including the two or three recognized types of split ergativity under the general rubric of ergativity. (Notice how much is assumed by the locution "split ergativity" -- what is our basis for assuming that "split ergativity" is a single coherent phenomenon of which there are subtypes?) There was, however, considerable discussion about active-stative typology, which was widely (though, if I may say so, absurdly) listed as one more subtype of split ergativity well into the 1980's (and is still so treated in Dixon 1994). That is, the unity of several quite disparate and distinct linguistic patterns was simply assumed from the outset, and continues to be: when typologists talk about "alignments", there are typically only three listed: nominative-accusative, ergative-absolutive, and active-stative. Recent suggestions have proposed considering inverse-marking languages to represent a fourth "alignment", but I do not know of anyone who has suggested breaking up the "ergative" bloc -- although I demonstrated 25 years ago (at least to my own satisfaction) that the person-based "split ergative" pattern is probably better thought of as a variant of the inverse type than of the ergative type (DeLancey 1981). (One aim of that paper was to demonstrate an underlying functional/cognitive unity to the superficially disparate aspect-based and person-based ergative "splits", but in retrospect I would say that, if the paper still has any value today, it is not that).

Why we believe in ergativity

We have always, at some uncritical, pretheoretical level (and, most of the time, in a very explicit theoretical sense), believed in the universality of Subject, in at least in Aristotelian sense. And for the most part we have not clearly distinguished between that and believing in Subject and Object in the sense that they exist in Latin or English. We have always taken subject-formation, which we may as well label "nominativity", to be the natural behavior of languages, so that any language which does not immediately and obviously deviate from this pattern is automatically considered to belong to it. There are many languages which would still be generally analyzed as "nominative" which in fact seem to be organized on quite different principles -- especially what Li and Thompson (1976) have called "topic-prominent" languages like Chinese, and, even more, certain peculiar variations on that pattern such as Burmese. Chinese and similar languages have no surface indication of case at all, and therefore no immediately evident reason to consider them actively different from Latin or English; Burmese has distinct subject- and object-marking postpositions, but uses them primarily to mark discourse-pragmatically marked arguments (see e.g. Soe 1999).

Thus, when we encountered languages which are immediately obviously organized on different principles -- Basque, Georgian, Tibetan -- we had to recognize that there is another pattern, but, in good old-fashioned European dualistic thinking, that implies one other pattern, in opposition to the "natural" pattern that we are used to. As linguists slowly came to recognize the fundamental distinctness of the "active/stative" or "split-S" pattern of clause organization, we grudgingly allowed for one more type, although, as we have noted, back in the 1970's when these languages first started attracting the attention of typologists and theoreticians, there was considerable debate here and there about whether these should be considered a distinct type, or some sort of weird subtype of ergativity, and even now the distinctness of "split-S" typology is not universally recognized. But this was as far as anyone wanted to go; when we had to deal with the likes of Hindi and Kiranti, rather than recognizing further new types, these were clearly considered subtypes of ergative -- "split" ergative. Notice the presumption of relative markedness in this terminology -- some of us even at the time noted that a language like Hindi could just as well be called "split nominative" as "split ergative", but because it manifests, even in just one construction, the weird deviant pattern of "ergativity", it becomes "split ergative".

It is very telling that we talk about "ergativity" -- and have conferences and workshops on topics like "Ergativity in Amazonian Languages" -- but conferences or thematic volumes devoted to "accusativity" or "nominativity" (which I think is the preferable term) seem to be a less prominent feature of the linguistic landscape. I would suggest that we have this exactly backwards -- nominativity is an actual phenomenon, something which we could usefully talk about as a unified phenomenon. "Ergativity" is just bits and pieces of what is left over. Our tradition inclines us to think of "ergativity" as some special deviation from the norm; in fact it is a non-category, it is simply the absence of nominativity.

S, A, and O

Dixon's formulation of S, A and O, the "universal syntactic-semantic primitives", is based directly on the myth of ergativity as one of just two basic "alignments". This way of thinking of things would not occur to us if we only had nominative languages to look at; for those Subject and Object are sufficient. The distinction of A from O is a direct reflection of the idea (probably traceable to Sapir 1917) of ergativity as sort of a complement to nominativity, where there are exactly two possibilities -- the single argument of an intransitive can pattern with one or the other of the arguments of a transitive clause. Of course we know that the typological facts are not that simple; dative subject constructions and various kinds of "split accusativity" are awkward, the scheme leaves no place for "indirect objects", which are a salient part of the syntactic organization of many languages, and no sensible way of talking about the "primary object" (Dryer 1986) pattern. And most devastatingly, as Mithun and Chafe (1999) have pointed out, the pattern which is tendentiously called "split-S" by many scholars -- especially the version with 3-way splits, as in languages such as Choctaw -- pretty much vitiates the whole idea of S as a "primitive". (In any case, as Dixon describes A, S, and O, they are not semantic primitives, as they are based in semantic notions of control and activity). The reason why we still cling to S, A and O as though they had some meaning is because they derive neatly from the idea of European-style nominativity as the way languages are, with ergativity as its mirror image. Patterns which do not fit into this neat scheme are treated as derivative subcategories (note that Dixon still treats active-stative typology as a species of "split" ergativity (1994:70ff.)). I will return to the question of S, A and O below.

Some Ergativities

If we are debating the question of what "ergativity" might be, we have to start by looking at some ergative, or putatively ergative, languages to see what they might have in common. Let us begin with two Tibeto-Burman languages, Mizo (also known as Lushei), of the Kuki-Chin branch of Tibeto-Burman, spoken in Mizoram State in India and adjoining parts of Burma, and Tibetan, a language of the Bodic branch of the family.

Mizo (Chhangte 1993) has a case postposition in which occurs always and only on A arguments, in the classic ergative pattern:

7) ka-nw -tap

1sg-baby 3sg-cry

'My baby is crying.'

8) k-n-in -kow-c

1sg-mother-ERG 3sg-call-2obj

'My mother is calling you.'

9) y - zuang

dog 3sg - jump

'A dog is jumping.'

10) y-in m -se/

dog ERG man 3sg-bite

'A dog bit a man.'

Notice two essential facts. First, there is consistent ergative case marking -- all transitive A arguments, in all tenses and aspects, are marked with the postposition in. There is no case marking of S or, usually, O arguments. This appears to be as straightforward an ergative system as one could find. But notice also that that the agreement clitics which precede the verb follow a consistently nominative pattern: all of (7-10) 3rd person proclitic -, which always agrees with a subject argument. There is one deviation from this consistent pattern, however: first person objects are indexed (with a special form) in preference to second or third person subjects:

11) ui in mi = se/

dog ERG 1stOBJ = bite

'A dog bit me.'

In this language, I would argue, the case marking and agreement systems function purely to mark grammatical relations, with no particular cognitive motivation. The Mizo system nicely illustrates Givn's functional characterization of ergativity:

The ergative-absolutive system ... is, first and foremost, a system where case-marking does the syntactic distinction between transitive and intransitive clauses. (Givn 2001:208)

But maybe we do want to insist that in marks some category. In this language, that category is "A", pretty much just as Dixon defines it. We will return later to the question of what "A" might be.

Like Mizo, Tibetan has a case postposition which occurs with A arguments, and thus must be labelled "ergative". (The data discussed here represent the Lhasa dialect of Central Tibetan; it is not clear how much variation there may be across dialects in the details of the distribution of ergative marking). However, its distribution and function are entirely different from that of Mizo in. In the past (1984, 1985a, 1990) I have presented a simplified picture of Tibetan ergativity, as follows. Ergative marking is obligatory on the A argument of a perfective transitive clause, but optional in non-perfective tense/aspects:

12) nga-s stag bsad-pa yin

I-ERG tiger killed-PERF/CONJ

'I killed a tiger.'

13) nga(-s) stag gsod-kyi yod

'I am killing a tiger.'

14) nga(-s) stag gsod-kyi yin

'I will kill a tiger.'

It also occurs optionally on S arguments in perfective intransitive clauses which refer to intentional action, but never in non-perfective one-argument clauses or in non-control clauses:

15) nga(-s) bod-la phyin-ba yin

I(-ERG) Tibet-LOC went-PERF/CONJ

'I went to Tibet.'

16) nga(*-s) bod-la >gro-gyi yod

'I am going to Tibet.'

17) nga(*-s) bod-la >gro-gyi yin

'I will go to Tibet.'

In sum, Lhasa Tibetan presents a picture of an aspectually-split active-stative language, in which ergative marking is obligatory on the A arguments of perfective transitive clauses, optional on A arguments of non-perfective and S arguments of active intransitive imperfective clauses, and impossible on S arguments of non-perfective clauses. These facts are grossly correct, but this way of describing the situation obscures the fundamental pattern which we find in the language.

First, note that the facts described above can be neatly described in terms of the Hopper and Thompson notion of global transitivity. Besides the standard criterion of two rather than one argument, perfective vs. non-perfective aspect and active or volitional vs. stative or non-volitional are considered to be factors in the transitivity prototype. Then the pattern which we see in Lhasa ergative marking is that the closer a clause is to prototypical transitivity, the more likely it is to have an ergative-marked argument (DeLancey 1984). A two-argument clause can have an ergative argument even if it is not perfective or active:

18) nga(-s) de shes-kyi med

I-(ERG) DEM know-NEG/IMPF/CONJ

'I don't know that.'

But a formally intransitive clause must be both active and perfective to qualify for ergative marking. Various other facts about the language lend further support to this characterization, including "dative subject" marking with certain (though not all) experiencer predicates, and the fact that "light verb" constructions, while formally transitive, show ergative marking only with semantically volitional predicates:

19) nga-s zhabs=bro brgyab-pa yin

I-ERG dance throw-PERF/CONJUNCT

'I danced.'

20) nga(*-s) hab=brid brgyab-byung

I(*-ERG) sneeze throw-BYUNG

'I sneezed.'

But there is a basic problem which this characterization fails to address. Where I have characterized ergative marking as "optional", it serves a discourse-pragmatic function, serving to place the argument which it marks in focus (Tournadre 1991, 1996). In other words, while its distribution is sensitive to transitivity, in any sense of that word, it simply cannot be interpreted as serving to distinguish transitive from intransitive clauses, which in fact it does not consistently do. The primary function of ergative marking seems to be to mark a discourse-pragmatic, rather than a semantic or syntactic, category. This places the function of ergative marking in Tibetan in stark contrast to that in Mizo, where it is simply always present under appropriate syntactic conditions.

It is likely that the ergative constructions in Tibetan and Mizo have similar diachronic origins. But, synchronically, the functions of the ergative in the two languages have virtually nothing to do with one another -- one serves to indicate grammatical relations, the other to mark a particular discourse-pragmatic category. There is no point in considering them to be examples of the same linguistic type. In fact, Tibetan is much more usefully compared with "nominative" languages in which case marking serves a similar pragmatic function, such as Burmese (So 1999) or Meithei (Chelliah 1997).

Deictic categories: Person-based "split ergativity"

Besides languages like Hindi, which show ergative marking in some tense/aspect constructions and nominative patterns in others, there is another well-known "split" ergative pattern in which 3rd person A arguments take transitive marking, while 1st and 2nd A's do not. In DeLancey 1981 I argued that this kind of system is a direct grammatical reflection of the category of deixis, and is more appropriately analyzed as a version of inverse marking than as anything related to canonical ergativity. A few years after the publication of that paper, I was made aware (originally by Noel Rude) of a language in which the deictic basis for "split ergative" marking is even clearer and more explicit than in the languages I had examined previously. This is Sahaptin, which with Nez Perce is one of the two Sahaptian languages of Oregon, Washington, and Idaho. (Sahaptian is part of the Plateau Penutian family. For further information on the language see Rigsby and Rude 1996; see also Ziga 2003 for an analysis along the lines suggested here).

In (21) and (22) we see that Sahaptin has a suffix, -in, which marks transitive A but not intransitive S:

21) winsh i-win-na 'The man went'

man 3sg-go-PST

22) p-q'inu-sha wnsh-in tlaaki-n

3:3-see-PROG man-ERG1 woman-ACC

'The man sees the woman.'

Although the presence of object case marking on the O argument counts as a deviation from the standard definition of "ergativity", many linguists would not hesitate to describe the second construction as "ergative", given the occurrence of the case marker -in on the A argument, which is impossible on the S argument of (21).

Now, Sahaptin is a "split ergative" language, which means (in this case) that ergative case is marked only on 3rd person A arguments, never on 1st or 2nd. Thus there is no case marking on the pronominal 1st person A argument of (23):

23) n=ash -q'inu-sha tlaaki-n

I=1sg 3Obj-see-PROG woman-ACC

'I see the woman.'

This well-known and widespread pattern, attested in Australian and Tibeto-Burman as well as New World languages, has been variously interpreted, most popularly in terms of a hierarchy of NP types (Silverstein 1976) which is widely referred to as the "animacy hierarchy". In DeLancey 1981 I pointed out that the only "hierarchy" which is robustly attested in animacy splits is one in which 1st and 2nd person outrank 3rd. Since there is only one distinction that is relevant, the notion of a hierarchy is somewhat superfluous; what this form of split ergativity is really about is the special status of 1st and 2nd persons relative to all others. That is, what "ergativity" in a language of this kind is about is not agentivity, or transitivity, but about paying special grammatical attention to the deictic center. In Sahaptin the grammatical importance of the deictic center is evident in several aspects of the grammar besides the distribution of ergative marking.

To begin with, unlike other split ergative languages that I know of, Sahaptin has two distinct ergative markers. The -in which we saw in example (22) occurs only when the O argument, as well as the A, is 3rd person. When the O is 1st or 2nd person, a distinct inverse ergative marker, -nm, occurs:

24) iwnsh-nm=nash i-q'nu-sha (in-y)

man-ERG2=1sg 3Subj-see (I-OBJ)

'The man sees me.'

25) iwnsh-nm=nam i-q'nu-sha (ima-ny)

man-ERG2=2sg 3Subj-see-PROG (you-OBJ)

'The man sees you.'

This fact by itself points up the centrality of deixis to the system, and the fundamentally deictic basis of the system is further attested in the apparent origin of the Sahaptin inverse ergative morpheme in an old cislocative form (Rude 1991).

And the evidence for the grammatical prominence of the deictic center does not stop here. Phonologically independent 1st and 2nd person pronouns such as n and iny in the preceding examples occur only in emphatic or contrastive function. Ordinary 1st and 2nd person reference is accomplished by obligatory pronominal clitics, which occur cliticized to the first word of the sentence. These clitic pronouns are not marked for case:

26) -q'inu-sha=ash winsh-na

3Obj-see-PROG=1st man-OBJ

'I see the man.'

27) i-q'inu-sha=ash winsh-nm

3Subj-see-PROG=1st man-ERG

'The man sees me.'

They occur even in the presence of independent pronouns (see also ex. 23):

28) n=ash -tuXnana yamash-na

I-1sg 3Obj-shot deer-OBJ

'I shot a deer.'

And they occur indexing SAP possessors of arguments:

29) nish-pa=sh w kusikusi

house-in=1st be dog

'My dog is in the house.'

30) Xlk=nash w nsuX

lots=1sg be salmon

'I have lots of salmon.' (lit. something like 'Me, there's lots of salmon')

To summarize this picture, Sahaptin has invariable 1st and 2nd person pronominal forms which must be present in any clause which makes reference to a 1st or 2nd person. Put otherwise, the language has forms whose only function is obligatory reference to the participants at the deictic center. Combined with the remarkably deictic system of ergative marking, this shows the central place of the deictic center in Sahaptin morphosyntax.

Now, to bring this discussion back to our main thread of argument, we can see that the function of "ergativity" in Sahaptin is completely unrelated to that of what we called "ergativity" in either Tibetan or Mizo, and that in fact these phenomena have hardly anything in common structurally either -- which leaves us with the question of what use is a word, and more to the point, what use is a concept, which tries to subsume them both as examples of a single category?

Deictic marking in Kuki-Chin

In several languages of the Kuki‑Chin branch of Tibeto‑Burman (to which Mizo also belongs), a simple inverse marking system has developed from the marking of deictic orientation on motion verbs (DeLancey 1980). In these languages a motion verb *hong 'come' has become partially or completely grammaticalized as a cislocative 'hither' prefix on motion verbs (see DeLancey 1985b for details). In some languages this morpheme has developed the additional function of optionally marking some transitive or ditransitive configurations with 1st or 2nd person object. In Sizang (Stern 1963) we find the cislocative marker used at least optionally with any transitive or ditransitive verb with 1st or 2nd person object or goal, as in (exx. from Stern 1984:52, 56):

31) na‑l:i hong thk ka‑ng: a:

2nd‑letter CIS send 1st‑receive NF

'I having received your letter which [you] sent to me ...'

32) k‑ong thk k:k l‑l:u h:

1st‑CIS reply again once more FIN

'... I in turn reply to you.'

33) hong s:t th:i l:

CIS beat ever INTER

'Do [they] ever beat you?'

34) hong s:t l: k‑pe:ng tl dong k‑ta:i t:

CIS beat if 1st‑leg break until 1st‑flee FUT

'If [they] beat me I'll run till my legs break.'

(Note that (31) has an easily recoverable 2nd person subject which however is not indexed in the verb, and both (33) and (34) have a specific anaphoric third person plural subject, the second speaker's parents, which is explicitly mentioned in a full NP two clauses previous to these; these are thus not indefinite subject "agentless" passives). The (h)ong in all of these examples occurs when there is a 1st or 2nd person goal or object, even when, as in exx. 31 and 32, the subject is the other Speech Act Participant. In Sizang‑Tiddim, as in Mizo, personal indexation in the transitive verb is consistently with the subject rather than with the principal participant.

The point which I want to call attention to here is the parallel between the distribution of the verbal proclitic hong in Sizang and Tiddim and the deictic "ergative" suffix -nm in Sahaptin -- which is less than surprising given that both originate in a spatial cislocative. But since Sahaptin -nm suffixes to an A argument, we will inevitably call it "ergative", while hong, which is a verbal affix, would never be called any such thing. So the question is, what is the concept of "ergativity" doing for us here, if it simply obscures the strong functional parallel between the two constructions? Especially since Kuki-Chin languages are ergative, in the classic sense, as we have already seen in the Mizo data. So, if Sahaptin -nm marks the same functional domain as K-C hong , then it does not mark the same domain as the K-C ergative postposition in. So whatever "ergativity" is, they can't both be ergative.

And, just to avoid another possible confusion, let me point out that a system of deictic marking exactly like that in Sizang is also found in two Dravidian languages, Kui (Winfield 1929) and Pengo (Burrow and Bhattacharya 1970), which are uncompromisingly non-ergative in their morphology; and that we can see something quite similar in languages of the non-ergative Loloish branch of Tibeto-Burman (see Matisoff 1982:325-6).

What are the primes? S, A and O once more

As I have already pointed out, the notion of ergativity is usually defined in terms of S, A and O. Since the S/A/O schema derives directly from the concept of ergativity, and the idea that ergative and nominative between them essentially define the range of "alignment" possibilities, there is a certain circularity in such a definition, but the definition would be legitimized if we had an indepent definition of the "primes". But, if S, A and O are real categories of language, what exactly are they? The concept of ergativity more-or-less assumes that the A category has some sort of content, like Agent or Topic or (even) Subject. But what would that content be?

There are some categories, which at this level of discussion we can take as primitives, which we can define and specifiy the content of. There is general agreement on semantic roles (if not necessarily on a specific set of them), in particular Agent, which in some sense are certainly real, and pragmatic categories like topic and focus, which likewise are certainly real, even if we don't really understand yet what they are. Subject is trickier; the tradition over the last generation of seeing it as some kind of intersection or amalgam of Agent and Topic gives it some real content, but leaks terribly. I think, though, that, following research by Tomlin and others, we can define subject quite neatly in terms of the psychologically unassailable category of attentional focus. In spite of Dixon's admirable efforts in that direction, as far as I can see S, A and O cannot be given contentful definitions of this type.

S, A and O are useful etic concepts. Assuming that every language has intransitive clauses, it follows that S is certainly a descriptive universal. However, the abundant evidence from so-called "Split-S" languages makes it evident that it is no more than that -- there is no universal unitary S category. In fact, as far as I can see, there is no emic S category in any language. An emic category -- that is, a category which is actually part of the grammar of a language -- must be definable by either structural or functional criteria -- ideally both, of course, but let us agree that for now we will settle for one or the other. Of course, in a two-way split S language like Lakhota or Pomo, and even moreso in a three-way split S language like Choctaw-Chickasaw, there is no S category whatever. In nominative languages, S arguments are subjects, just as A arguments are, but in a language like English there is no S category apart from Subject -- that is, there are no structural or semantic/pragmatic criteria by which one can identify intransitive subjects as a distinct subcategory of subject. One might argue that in a prototypical ergative-absolutive language we could define an S category in structural terms as that argument which shares its morphological properties with O and its syntactic properties with A, and that might be enough to satisfy us that we have a distinct category -- but clearly not a universal one.

A and O are a bit more challenging. Assuming that all languages have transitive clauses, then in any language there will be two kinds of transitive argument. Referring to these by putatively universal labels like Subject and Object or A and O makes the claim -- explicitly asserted by Dixon (1994) -- that every language will identify the same two categories of argument within the transitive clause. The essential phenomenon which we need to deal with is that languages strongly tend to treat experiencers and agents as a natural class (DeLancey 1996). Even in a consistent dative-subject type language (e.g. Malayalam, Mohanan and Mohanan 1990), where we could argue that surface marking of arguments fails to recognize the A category, we typically find that syntactic phenomena treat agents and experiencers alike, as opposed to the "undergoer" roles that typically form O arguments. For Dixon, this can be explained in terms of the fact that typical experiencers are the same kinds of entities as typical agents, so that, of the two arguments of a transitive verb, the A will be "that semantic role which could exercise control, if anything could" (1994:31), that is, "that role which is most likely to be relevant to the success of the activity" (1994:52).

However, we do not actually have here evidence of the A category behaving as a natural class. In any language in which agent and experiencer arguments of transitive verbs behave alike with respect to syntactic processes, it is also the case that S arguments also show the same behavior. That is, once again we are looking at a subject category, with no particular reason to subdivide it into A and S primitives. For Dixon, Subject is a derived category, representing the union of A and S, and can be defined semantically, as both A and S "refer to functions that can be initiating/controlling agents" (1994:125). I think this is a hopeless enterprise -- the subject category, which unites agents, experiencers, possessors, located Themes and all other S arguments, cannot be defined semantically. It is a pragmatic category, best identified with the psychological notion of attentional focus (Tomlin 1997, Forrest 1999, DeLancey 2001). Attentional focus is indeed a prime, though it does not receive the same kind of grammatical attention in all languages.

Now, S, A and O are useful etic, which is to say, descriptive, categories, and in a way which reinforces what I have been saying. What does it mean to call them "descriptive"? For example, recall my thumbnail description of Tibetan above. In trying to quickly give another linguist an idea of what Tibetan is like, the terms A, S and O are very useful signposts. So, I could describe Tibetan as having differential marking on the A according to a complex mix of agentivity, aspect, and discourse-pragmatic factors, and differential marking on S according to a slightly different mix of the same set of factors, and marked or unmarked O according to a strict semantic-syntactic factor. Now, a believer in these "syntactic-semantic primes" might say, OK, notice that you describe A as being differentially marked according to a deictic hierarchy, and O as differentially marked according to the semantics of the verb -- but does not that description presuppose underlying A and O categories which can be differentially marked? No, in fact it does not. Tibetan has a few dative-subject predicates:

35 khong‑la snyu=gu cig dgo=gi

he-LOC pen a want-IMPF

'He needs/wants a pen.'

So ergative marking does not affect all A arguments, by Dixon's definition. Nor, as we have seen, does it affect only A arguments. Thus, it does not select Dixon's A as a class, and no other rule or pattern in the language does either. But problems with the A category are old news. More striking is the fact that, in Tibetan, there is no emic object, or O, category -- that is, there is no rule which affects, or construction which involves, or however you want to put it, all and only O arguments. Most dramatically, as we have seen, O arguments are differentially marked according to the type of verb. That is, Tibetan provides exactly the same kind of evidence for distinguishing Theme and Goal objects as Dyirbal does for distinguishing A and S.

What this means is that it is entirely possible to write a complete, comprehensive grammatical description of Tibetan and never mention the A or O category. In fact, this is objectively the best way to describe the language; any reference to an O or object category will only be for the purpose of making the facts of the language accessible to someone who is used to thinking in those terms. And, even if one should want to mention a category of transitive subject, or A, or something, that category does not correspond to Dixon's A, since it excludes some (but not all) experiencer subject clauses. And, since some intransitive arguments are zero-marked, others ergative marked, and a few dative-marked, there is no grammatically distinct S category in the language. These terms remain descriptively useful only because they provide a terminology for laying out these facts in a manner that will be accessible to someone who is used to thinking in terms of subject and object, but they have no reality whatever within the language itself.

What can we believe in??

The term ergative has a well-established and necessary place in linguistics, as the name for a surface case category. Many languages have a case form which occurs only or primarily with the subject-like argument of a transitive clause, and in describing such a language we need to call it something. It is also often useful to refer to ergative constructions, in languages where these contrast with nominative constructions. But the word ergativity implies something more, that languages with such a case form (or, perhaps, where such a case form occurs in the discourse-pragmatically least marked transitive clause pattern) have something more in common. I do not see that there is very much evidence for any such thing. I think that there is what we might call "nominativity", except that there are better ways to talk about it. That is, a great many languages are what Anderson (1979) calls subject-forming languages, and that is a very particular kind of grammatical organization. Now, it is not by any means obvious that all languages which lack any "ergative" marking are automatically subject-forming languages -- that may be true, but such a conclusion requires further research on inverse typology, Austronesian voice systems, Li and Thompson's topic-prominent typology, and on the odd variants of that (if that is what they are) in languages like Burmese and Meithei.

But as for the many languages in the world which do not have a clear subject category the way most European and many other languages do, we need to recognize that there are several different parameters which languages can grammaticalize in their system of marking core arguments. Transitivity is one of these, as Givn points out, and, independently (more or less) of that, the Agentivity complex -- though it remains an open question whether we can speak of Agentivity as a primitive in this way, or whether it must be broken down into components of eventivity, control, volition, and what all. In any case it probably has to be kept separate from general causation, which also has its effect in core argument marking. Another, which has been grossly neglected to date (though that is not MY fault) is deixis, which, when it is recognized at all, tends to be carelessly and incorrectly lumped together with animacy -- which may also be a relevant factor, and, of course, points back towards control, and hence toward Agentivity.

Now, languages grammaticalize many different combinations of these, in many different ways, and the grammaticalization of the same parameter may be accomplished in quite different ways, which may or may not involve case marking. For example, it might be useful to compare transitivity as reflected in Mizo ergative marking with transitivity as reflected in the detransitivizing so-called "reflexive" constructions of most European languages. And it is a rather obvious idea (as I pointed out many years ago) to compare the grammaticalization of deixis in "split ergative" languages and in inverse-type languages, which often show no morphological features that we would be inclined to call ergative.

My point, my final suggestion, is that being obsessed with the idea of "ergativity" actually pulls us away from these very promising lines of research, and leaves us comparing bluebirds and blue jays, for no better reason that they are sort of the same color.


References

ANDERSON, John. 1979. On being without a subject. Bloomington: Indiana University Linguistics Club.

BOSSONG, Georg. 1985. Empirische Universalienforschung: Differentielle Objektmarkierung in den neuiranischen Sprachen. Tbingen: Gunter Narr.

BURROW, T., and S. BHATTACHARYA. 1970. The Pengo language. London: Oxford.

CHELLIAH, Shobhana. 1997. A grammar of Meithei. Berlin: Mouton de Gruyter.

COMRIE, Bernard. 1978. Ergativity. Pp. 329-94 in Winfred Lehmann, ed., Syntactic Typology. Austin: University of Texas Press.

1979. Definite and animate direct objects: A natural class. Linguistica Silesiana 3.13-21.

DELANCEY, Scott. 1980. Deictic categories in the Tibeto‑Burman verb. Ph.D dissertation, Indiana University.

1981. An interpretation of split ergativity and related patterns. Language 57.626‑57.

1984. Transitivity and ergative case in Lhasa Tibetan. Proceedings of the Tenth Annual Meeting of the Berkeley Linguistics Society, 131-40.

1985a. On active typology and the nature of agentivity. Pp. 47-60 in Frans Plank, ed., Relational Typology. The Hague: Mouton.

1985b. The analysis‑synthesis‑lexis cycle in Tibeto‑Burman: A case study in motivated change. Pp. 367‑89 in John Haiman, ed., Iconicity in Syntax. Amsterdam and Philadelphia: Benjamins.

1990. Ergativity and the cognitive model of event structure in Lhasa Tibetan. Cognitive Linguistics 1.3:289-321.

1996. rev. of R.M.W. Dixon, Ergativity. Journal of Linguistics 32:173-177.

2001. The universal basis of case. Logos and Language 1.2:1-15.

DIXON, R.M.W. 1972. The Diyrbal language of North Queensland. Cambridge: Cambridge University Press.

1979. Ergativity. Language 55.59-138.

1994. Ergativity. Oxford: Oxford University Press.

DRYER, Matthew. 1986. Primary object, secondary object, and antidative. Language 62.808-45.

FORREST, Linda. 1999. The role of attention and activation in language production: A neurocognitive linguistic approach. Ph.D. dissertation, University of Oregon.

GIVN, T. 2001. Syntax: An introduction, Vol. 1. Amsterdam: Benjamins.

HOPPER, Paul, and S. A. THOMPSON. 1980. Transitivity in grammar and discourse. Language 56.251-99.

LI, Charles, and S. A. THOMPSON. 1976. Subject and topic: A new typology of language. Pp. 457-90 in Charles Li, ed., Subject and topic. New York: Academic Press.

MATISOFF, James. 1972. The grammar of Lahu. Berkeley and Los Angeles: University of California Press.

MITHUN, Marianne, and Wallace CHAFE. 1999. What are S, A and O? Studies in Language 23:3.569-606.

MOHANAN, K.P., and Tara MOHANAN. 1990. Dative subjects in Malayalam: Semantic information in syntax. pp. 43-57 in Manindra Verma and K.P. Mohanan, eds. Experiencer Subjects in South Asian Languages. Stanford: Center for the Study of Language and Information.

RIGSBY, Bruce, and Noel RUDE. 1996 Sketch of Sahaptin, a Sahaptian language. Pp. 666-692 in Ives Goddard, ed., Languages. (Handbook of North American Indians vol. 17). Washington: Smithsonian Institution.

RUDE, Noel. 1991. On the origin of the Nez Perce ergative NP suffix. IJAL 57:24‑50.

SAPIR, Edward. 1917. Review of Uhlenbeck, "Het passieve karakSter van het verbum transitivum of van het verbum actionis in talen van Noord-Amerika." IJAL 1.82-6.

SCHUCHARDT, Hugo. 1895. ber den passiven Charakter des Transitivs in den kaukasischen Sprachen. in: Sitzungsberichte der kais. Akademie der Wissenschaften, phil.-hist. Klasse, vol. 133. Vienna.

SILVERSTEIN, Michael. 1976. Hierarchy of features and ergativity. Pp. 112-71 in R.M.W. Dixon, ed., Grammatical categories of Australian languages. Canberra: Australian Institute of Aboriginal Studies.

SOE, Myint. 1999. A grammar of Burmese. Ph.D. dissertation, University of Oregon.

STERN, Theodore. 1963. A provisional sketch of Sizang Chin. Asia Major n.s. 10.222‑78.

1984. Sizang (Siyin) Chin texts. Linguistics of the Tibeto‑Burman Area 8.1:43‑58.

TOMLIN, Russell. 1997. Mapping conceptual representations into linguistic representations: The role of attention in grammar. in Jan Nuyts and Eric Pederson, eds., Language and conceptualization, pp. 162-89. Cambridge: Cambridge University Press.

TOURNADRE, Nicolas. 1991. The rhetorical use of the Tibetan ergative. Linguistics of the Tibeto‑Burman Area 14.93-108.

1996. L'Ergativite en tibetain: approche morphosyntaxique de la langue parlee. Paris: Peeters.

WINFIELD, W.W. 1929. A grammar of the Kui language. (Bibliotheca Indica no. 245, issue n.s. no. 1496). Calcutta: Asiatic Society of Bengal.

ZIGA, Fernando. 2003. Inverse systems in the indigenous languages of the Americas. Ph.D. dissertation, Universitt Zrich.