Lecture 8: Split-ergative and Inverse Systems
In this lecture we will examine various syntactic phenomena which reflect the deictic centrality of the speaker and addressee in the speech event. Deixis and its expression in various linguistic categories is an old and well-known concept in semantic analysis, but it is seldom invoked in syntactic analysis. After all, though the difference between the shifting reference of I and you and the fixed reference (within a given discourse) of other NP's is obvious, it has no obvious syntactic consequences--the 1st and 2nd person pronouns have the exactly same syntactic privileges as other NP's. But, as we will see, a great many languages manifest some kind of syntactic alternation which directly reflects the deictic status of the various core arguments. And, indeed, in the context of this demonstration, we will be able to see that the category is indistinctly but unmistakably reflected even in a language like English.
1.0 Split Ergative and Inverse Marking
Among the linguistic phenomena which pose long-standing problems for theories of grammatical relations are split ergative marking and direct/inverse marking in the verb. There are several types of ergative "split", in which case marking is sometimes according to an ergative pattern, and sometimes not; our interest here is in the nominal split pattern, in which the place of the A argument on a hierarchy of nominal types determines whether or not it will be marked as ergative. This, as we will see, is responsive to the same functional parameters as direct/inverse marking, where a transitive verb is marked to reflect whether the A or O is higher on the same nominal hierarchy.
1.1 Split-ergative Case Marking and Indexation
It has been generally recognized since Silverstein's seminal paper on the topic (1976) that one widely-attested pattern of split ergative marking reflects a hierarchy of nominal types. Dixon (1994:84ff) summarizes this "Nominal Hierarchy" of eligibility to be subject of a transitive verb as:
1st person 2nd person Demonstratives Proper Common
pronouns pronouns & 3rd person nouns nouns
pronouns
Following Silverstein, Dixon notes that:
Those participants at the left-hand end of the hierarchy are most likely to be agents, to be in A function, and those at the right-hand end are most likely to be patients, to be in O function. (Dixon 1994:85)
However, this interpretation of the facts, though standard, is somewhat misleading. In reality almost[1] all split ergative languages make the "split" between 1st and 2nd person pronouns (the Speech Act Participants, or SAP's), which do not distinguish A from S forms, and all other arguments, which do (DeLancey 1981a, cf. Dixon 1994:88).
This type of split is very common in Australian languages (Silverstein 1976, Heath 1976, Blake 1987), and attested in North America (e.g. Silverstein 1976, Mithun 1999:230-3, and below), Siberia (Comrie 1979a, b, 1980), and in a number of Tibeto-Burman languages (Bauman 1979). An example the last is Sunwar, a language of the Kiranti branch of Tibeto-Burman spoken in Nepal.[2] Lexical nouns and 3rd person pronouns (which in Sunwar are the demonstratives méko 'that' and mére 'yon') are unmarked in S function, and take ergative case (-Vm) as A's:
1) méko ?àl hí-t-a
DEM child come.down-PAST-3sg
'The child came down.'
2) méko ?àl-am tà-t-i
DEM child-ERG see-PAST-3sg61sg
'The child saw me.'
3) méko hí-t-a
DEM come.down-PAST-3sg
'He came down.'
4) méko-m tà-t-i
DEM-ERG see-PAST-3sg61sg
'He saw me.'
But there is no such alternation for 1st and 2nd person pronouns, which do not have ergative forms:
5) go hí-t-i
I come-down-PAST-1sgINTR
'I came down.'
6) go méko ?àl tá-t-a
I DEM child see-PAST-1sgTR
'I saw the child.'
There is also a head-marking version of the same phenomenon, with distinct ergative and absolutive verbal indices for 3rd, but not 1st or 2nd, person arguments, as in the Chinookan (Penutian) languages of the lower Columbia River. Consider these forms from the Kiksht[3] language (Dyk 1933, cf. Silverstein 1976):[4]
6) ni-n-i-waqw 'I killed him.'
8) a-tc-n-dwágwa 'He will kill me.'
(0-0) show the masculine singular 3rd person absolutive index i- (in italics) as A and as S. (0) shows the masculine singular 3rd person ergative index tc- as A. (0, 0, and 0) show the undifferentiated 1st person singular index n- (in boldface) as, respectively, A, O, and S. The structure of the agreement paradigm is the same as that of the Sunwar case paradigm: 3rd person forms distinguish ergative and absolutive, 1st and 2nd person forms do not.
The essential facts about split ergative marking are the special status of the SAP's, and the pattern of the split--it is not only that it is always the SAP's that get special treatment, but that the special treatment is always the same, with SAP A arguments unmarked where 3rd person A is a morphologically marked category.
1.2 Inverse systems
Another type of grammatical system which manifests exactly the same person hierarchy is found in inverse-marking languages.[5] In the usual sense of the term,[6] an inverse‑marking system is one in which there is a ranking of person in which SAP's outrank all 3rd persons (while ranking among SAP's is language‑specific, see DeLancey 1981a), and a transitive verb is marked to reflect whether or not the O argument outranks the A on the hierarchy. The configuration in which the O outranks the A is called inverse, and that in which the A outranks the O is direct.
Direct-inverse marking, like dative-subject marking, ergativity, and active-stative typology before it, is an "exotic" typological pattern which, once recognized, turns out to be far more common than anyone ever suspected. A generation ago it was considered, by those linguists who were aware of it at all, to be a strange idiosyncratic feature of Algonquian languages. As our database expanded, a handful of similar examples began to be pointed out (Comrie 1980, DeLancey 1980, 1981a, b, Whistler 1985, Grimes 1985). In the 1970's the phenomenon was of considerable theoretical interest to practitioners of Relational Grammar (e.g. LeSourd 1976, Jolley 1981), for the same reasons that is relevant to our present investigation--the fact that it involves different morphosyntactic indications of subjecthood being associated with different arguments. Recent years have seen a substantial number of analyses of inverse or inverse-like constructions in a range of languages (e.g. Ebert 1991, Payne 1994, Zavala 1994, 1996, Bickel 1995, Watkins 1996), and increasing interest in the topic in both formal and functional frameworks (Jelinek 1990, Arnold 1994, Givón 1994b, Payne 1994, Rhodes 1994).
1.2.1 Nocte
A maximally simple example of the system is found in Nocte (or Namsangia), a language of the Konyak branch of Tibeto-Burman spoken in Arunachal Pradesh (adapted from Weidert ms.; cf. Konow 1903, Das Gupta 1971, DeLancey 1981a, b, Weidert 1985):
10) õaa-mE @1te1-n@õ2 vaat-@õ1
I-ERG he-ACC beat-1sg
'I beat him.'
11) @1te1-mE õaa-n@õ2 vaat-h-@õ1
he-ERG I-ACC beat-INV-1sg
'He beat me.'
you-ERG he-ACC beat-2sg
'You beat him.'
13) @1te1-mE n@õ-n@õ2 vaat-h-o?
he-ERG you-ACC beat-2sg
'He beat you.'
The first thing to notice about this system is the fact that agreement is not always with the same grammatical role. The verbs in (0) and (0) both have 1st person agreement, although the 1st person participant is an ergative-marked A in (0) and an accusative-marked O in (0). (0) and (0) show the same pattern, with the 2nd person argument attracting agreement regardless of its grammatical role. The second interesting feature of the system is the ‑h- suffix found in some forms. These two phenomena are clearly related--we find the -h- suffix in just those forms where agreement is not with the subject.
These forms illustrate the basic structure of an inverse-marking system. Agreement is always with a SAP in preference to a 3rd person, regardless of grammatical role. When this results in the verb indexing a non-subject argument, a special inverse morpheme is added to the verb. Thus, although both 'I hit him' and 'he hit me' have 1st person agreement, the verb forms are not ambiguous, but distinguished by the presence or absence of the inverse -h-. The verb forms in (0) and (0), which lack the ‑h‑, are direct forms; in Nocte the direct category is unmarked.
In anticipation of discussion to come, note the behavior of Nocte verbs when both arguments are SAP's:
I-ERG you-ACC beat-162
'I beat you.'
15) n@õ-mE õaa-n@õ2 vaat-h-@õ1
you-ERG I-ACC beat-1sg
'You beat me.'
The inverse marker is absent in (0), with 1st person A and 2nd person O, and present in (0). The overall verbal indexing system is illustrated below (imperfective paradigm with singular participants, adapted from Weidert 1985:925):
O:
1st 2nd 3rd
A:
1st ‑E ‑aõ
2nd h‑aõ ‑o
3rd h‑aõ h‑o ‑a
The distribution of the inverse marker follows a simple formula: there is a person hierarchy in which 1st person outranks 2nd, and both outrank 3rd, which we will symbolize as 1 > 2 > 3. When an O argument outranks the A on this hierarchy, the verb is in its inverse form. We can almost capture the indexation pattern by simply saying that the verb indexes the argument which is highest on the person hierarchy, but this does not account for the anomalous agreement suffix in (0). By analogy with the rest of the paradigm we would expect 1st person indexation here; what we have instead is a suffix which occurs nowhere else in the singular paradigm--in fact, it is identical to the 1st person plural index.[7]
1.2.2 The classic direction system: Cree
In this section we will briefly examine a typically complex system from the Algonquian family, where the direction marking phenomenon was first recognized. The Algonquian systems are the most elaborate that I am aware of, most of them making all of the distinctions found in any other direction system; they represent a prototype in terms of which other systems are easily analyzable.[8] A straightforward example is Plains Cree (Wolfart 1973, cp. Dahlstrom 1986) which overtly marks four direction categories‑‑direct, inverse, and the two local (Hockett 1966) categories 162 and 261‑‑with morphemes from a single paradigm, and consistently indexes the principal participant in all configurations. The verb forms with both arguments singular in the independent order of the transitive animate paradigm (i.e. verbs with animate objects) can be schematized as follows (where V represents the verb stem):
O:
1st 2nd 3rd
A:
1st ki‑V‑i‑n ni‑V‑aa‑wa
2nd ki‑V‑eti‑n ki‑V‑aa‑wa
3rd ni‑V‑ekw ki‑V‑ekw V‑ekw / V‑ee‑wa
The prefixes and second position suffixes[9] are person indices: ki‑ '2nd', ni‑ '1st', ‑wa '3rd proximate', and ‑n '1st or 2nd'. The first position suffixes are direction markers: ‑ekw marks unambiguously inverse, and ‑aa unambiguously direct, configurations, while the two local categories each have their own direction marker, ‑i '162' and ‑eti '261'.
The distribution of the personal prefixes clearly reflects a 2nd > 1st > 3rd person hierarchy. Such a hierarchy should, as in Nocte, define every configuration except 363 as clearly either direct, i.e. with subject higher on the hierarchy than object, or inverse, with object higher than subject. In Cree, however, we find not a two‑ but a four‑term direction system; as I have argued at greater length in DeLancey 1981a, this reflects the fact that the language‑particular ranking among SAPs is of a different order from the universal SAP > 3rd ranking.
The other significant respect in which the Cree system differs from that of Nocte is in the subdivision of the 363 category according to the relative topicality of the two participants. The form ‑ee‑aw, which Wolfart glosses as 'direct‑3rd',[10] is used with proximate, i.e. more topical, subject and obviative, i.e. less topical, object, and the clearly inverse form ‑ekw with obviative subject and proximate object. Thus in Algonquian relative topicality can define the principal participant when hierarchical ranking fails to do so. This constitutes the major functional point of contact between direction and voice systems, as we will discuss below.
1.2.3 The direction-marking prototype
The deictic nature of these patterns is self-evident--in both Nocte and Cree, verbal morphology is obligatorily responsive to a fundamental distinction between the speech act participants and all other participants. But there are certain other distinctions which occur in either Cree or Nocte, but not both. In Cree (and all Algonquian languages, but cf. DeLancey 1981a:643) verbal indexation reflects a ranking of 2nd person above 1st. Direction marking, on the other hand, appears to treat them as equal; in any case it shows that the ranking of 2>1 is of a different order from the ranking SAP>3. Nocte explicitly ranks 1st above 2nd in direction marking, marking 261 but not 261 as inverse, but the odd personal index in the 162 form (and the fact that both SAP's, but not 3rd person, are indexed) suggest again that 1st person outranks 2nd by much less than both of them outrank 3rd. Cree treats both local categories as special direction categories, but both show normal hierarchical indexation. Nocte treats 261 as inverse and 162 as non-inverse, but the indexation paradigm treats 162 as a special category. Taken together, then, Nocte and Cree imply a universal schema in which SAP arguments are clearly distinguished from and ranked above all others, and there is no universal ranking of the two SAP's (since Cree shows one of the possible rankings, and Nocte the other).
2.0 Variations on a Theme
2.1 Hierarchical agreement
We are used to thinking of verb agreement as tied to grammatical relations: a common claim about the typology of verb agreement is that if a language has verb agreement it will index the subject; some languages index both subject and object, and a rare handful index only objects (e.g. Keenan 1976:316). However, there are languages in which indexation of arguments in the verb reflects not grammatical relations, but the person hierarchy. In these languages a verb will always agree with a SAP argument, regardless of its grammatical role. (Typically such languages have no 3rd person index). This is, of course, exactly the typical indexation pattern of a direct/inverse-marking language; in the languages which we will discuss in this section, however, we find the hierarchical indexation pattern without inverse marking on the verb. In earlier work I have described this pattern as a variation of split ergative marking (DeLancey 1980, 1981a, b), on the grounds that verb agreement and zero case marking serve the same function, of identifying one argument of the clause as the most topical or "starting point" (Delancey 1981a, see below). But this terminological extension is somewhat misleading; it is better to reserve the term "split ergative" for the Chinookan-type pattern. Nevertheless this pattern represents one more way of encoding exactly the same functional domain that we have discussed in the previous sections.
The hierarchical agreement pattern has not received as much theoretical or descriptive attention as split ergative or direction marking. I don't have a clear sense of how widespread it may be, but it is fairly common in Tibeto-Burman (Bauman 1979, DeLancey 1980, 1988, 1989, Sun 1983). One example which has been discussed elsewhere is Tangut (Kepping 1979, 1981, Comrie 1980, DeLancey 1981a, b). A slightly more complicated case is the Nungish (Tibeto-Burman) languages of Yunnan (Tarong (Dulung) data from Sun 1982, 1983:25-6; cp. Lo 1945):
O: 1st 2nd 3rd
A:
1st -õ -õ
2nd n@- -õ n@-
3rd n@- -õ n@- --
The transitive paradigm of Trung
There is no 3rd person index. The 1st person suffix -õ occurs on any verb with a 1st person argument. The n@- prefix occurs on intransitive 2nd person subject verbs, and in all transitive forms with a 2nd person argument except for the 162 form. (Recall that this form gets special marking also in Nocte). The synchronic identification of this prefix as a 2nd person index is complicated by its occurrence in the 361 form, which has no 2nd person argument, but this is demonstrably a secondary development involving the merger of a previously distinct prefix with the original 2nd person form.[11] Despite this complication the hierarchical nature of the indexation pattern is clear: any 1st person argument must be indexed; any 2nd person argument is indexed unless there is a 1st person actor.
2.2 Sahaptian
The Sahaptian languages Nez Perce and Sahaptin[12] show a fascinating combination of hierarchial indexation and a unique pattern of split-ergative case marking. In these languages pronominal clitics, ordinarily in sentence-second position, occur in a purely hierarchical indexation pattern. In Nez Perce these occur primarily in subordinate clauses; in Sahaptin they occur in main clauses (see exx. 0-0 below). The Nez Perce paradigm is (Aoki 1970, Rude 1985):
Intransitive 1st ‑x, 2nd ‑m, Inclusive ‑nm, 3rd 0
O:
1st 2nd 3rd
A:
1st ‑m‑ex ‑x
2nd ‑m ‑m
3rd ‑x ‑m ‑‑
Note that the 162 configuration is once again exceptional, in this case in having both arguments independently indexed. The same pattern occurs in Sahaptin (Jacobs 1931, Rude 1985, p.c.).
Along with this indexation pattern, the Sahaptian languages show interesting variations on the split ergative pattern (Rude 1991). Nez Perce has a typical pattern, with ergative case marking on 3rd person (0) but not SAP (0) A's:
16) hi-páayn-a haáma
3NOM-arrive-PAST man
'The man arrived.'
17) 'iin páayn-a
I arrive-PAST
'I arrived.'
18) wewúkiye-ne pée-'wi-ye háama-nm
elk-ACC 363-shoot-PAST man-ERG
'The man shot an elk.'
19) 'iin 'ew-'wii-ye wewúkiye-ne
I SAP63-shoot-PAST elk-OBJ
'I shot an elk.
Note, however, that O arguments are consistently case-marked (-ne in exx. 0-0). Sahaptin shows the same ergative split--3rd person A arguments take ergative marking, SAP A's do not (exx. from Rigsby and Rude 1996 and Rude 1991):
20) iwínš i-wínan-a
man 3NOM-go-PAST
'The man went.'
21) iwínš-in pá-tuXnana yáamaš-na
man-ERG1 363-shot mule.deer-ACC
'The man shot a deer.'
22) ín=aš[13] tuXnana yáamaš-na
I=1sg shot mule.deer-OBJ
'I shot a deer.'
But Sahaptin has two distinct ergative forms. Both occur only on 3rd person A arguments, but the -in morpheme seen in (0) occurs only when the O argument is also 3rd person. When the O is a SAP, there is a different ergative marker, -nm (which Rude, for obvious reasons, calls the "inverse ergative"):
23) áw=naš i-nák-wina k'waali-nm
now=1sg 3NOM-carry=go dangerous.one-INV.ERG
'Now the dangerous one has taken me along.'
man-INV.ERG=2sg 3NOM-see-IMPF
'The man sees you.'
Thus, where Nez Perce, in the typical split ergative pattern, distinguishes two transitive configurations:
SAP 6
3-ERG 6
Sahaptin distinguishes three:
SAP 6
3-ERG1 6 SAP
3-ERG2 6 3
And the additional distinction which Sahaptin makes once again reflects the SAP / 3rd division.
2.3 Inverse with non‑hierarchical agreement
Inverse marking languages came into theoretical prominence during the heyday of Relational Grammar, for which their peculiar use of verb agreement, which is normally thought of as a perquisite of subjecthood, and the formal similarity of inverse and passive constructions, formed a particularly intriguing puzzle, which still captures the attention formal theoreticians. Since I want to argue that inverse marking is a direct expression of a deictic category, it is significant that we find languages with very similar deictic systems where it is clearly distinct from grammatical relations.
2.3.1 Expansion of the Cislocative in Kuki-Chin
In several languages of the Kuki‑Chin branch of Tibeto‑Burman, spoken in western Burma and eastern India and Bangladesh, a simple inverse marking system has developed from the marking of deictic orientation on motion verbs (DeLancey 1980). In most of these languages a motion verb *hong 'come' has become partially or completely grammaticalized as a cislocative 'hither' prefix on motion verbs (see DeLancey 1985 for details). In the closely related Tiddim (Henderson 1965), Sizang (Stern 1963), and Paite (Konow 1904) dialects, this morpheme has developed the additional function of optionally marking some transitive or ditransitive configurations with SAP object.
In Tiddim and Sizang, 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):
25) k‑ong thûk kí:k lâ‑lê:u hî:
1st‑CIS reply again once more FIN
... I in turn reply to you.'
CIS beat ever INTER
'Do [they] ever beat you?'
27) hong sá:t lé: ká‑pe:ng tál do*ng ká‑ta:i tû:
CIS beat if 1st‑leg break until 1st‑flee FUT
'If [they] beat me I'll run till my legs break.'
The (h)ong in all of these examples occurs when there is a SAP goal or object, even when, as in (0), the subject is the other SAP. Its distribution in the transitive paradigm is thus almost identical to that of the Nocte inverse marker, except that it marks both local categories rather than only one:
object
1st 2nd 3rd
subject
1st hong ‑‑
2nd hong ‑‑
3rd hong hong ‑‑
This pattern suggests a natural category of marked direction which includes all configurations with SAP objects or goals, and provides further evidence for the non‑universality of any ranking of SAPs.
In Sizang‑Tiddim, unlike the languages that we have previously considered, personal indexation in the transitive verb, if present, is consistently with the subject rather than with the principal participant. (This is clear in ex. (0); in (0-0), with 3rd person agent, there is no subject index, but in most cases there would be a 3rd person prefix: a‑hong sa:t '3rd‑CIS beat' = 's/he beat you/me'; see Stern 1963:254‑5). Thus while inverse forms with SAP subject, such as (0), are unambiguous in isolation, 3rd person subject forms depend upon context for the identification of the SAP object.
2.3.2 The Dravidian "Special Base"
In two Dravidian languages, Kui (Winfield 1929) and Pengo (Burrow and Bhattacharya 1970), we find a similar system of inverse marking with consistent subject agreement, which appears to have the same cislocative origin as the Chin and Loloish inverse constructions. The morpheme in question is a suffix which forms what Burrow and Bhattacharya call the "special base" (glossed SB in the examples below) of the verb, after which are suffixed ordinary negative, tense/aspect, and personal index morphemes. It occurs "when the object, direct or indirect, is the first or second person" (Burrow and Bhattacharya p. 70), regardless of the person of the subject, as in:
28) huR‑d‑av‑at‑an
see‑SB‑NEG‑PAST‑3m.s.
'He did not see (me, us).'
29) huR‑d‑av‑at‑ang
‑1s
'I did not see you.'
As in Kuki-Chin, the subject is always indexed by the subject suffix. We can represent the distribution of the Pengo "special" morpheme /d/ and the personal indices as follows:
object 1st 2nd 3rd
subject
1st d‑1st -1st
2nd d‑2nd -2nd
3rd d‑3rd d‑3rd -3rd
Thus the distribution of Pengo ‑d is identical to that of Tiddim Chin ‑hong.
Emeneau (1945), on the basis of deictically‑specified verbs of motion and giving elsewhere in Dravidian, reconstructs essentially the Kui-Pengo inverse marking for Proto‑Dravidian, where it marked not only inverse transitive forms, but also, like the Chin *hong reflexes, motion verbs with SAP or deictic center as goal. Analogy with the Chin system, as well as the general tendency for historical development to proceed from more concrete to more