Split Transitivity and Coreference in Katukina-Kanamari

Francesc QUEIXALÓS

CELIA (CNRS-IRD) / LALI (Universidade de Brasília)

After Scott DeLancey has, in the opening of this volume, straightened out some of the more frequent conceptual imbroglios concerning ergativity, I feel compelled to say a word on what this notion refers to in my own work. An ergative pattern is one in which core arguments of the basic transitive construction display a mapping between their semantic roles and their morphosyntactic properties such that the patient is formally ranked above the agent. "Basic" is to be understood in terms of semantic prototypicality, simpler formal definitory features, and higher frequency in discourse. "Formally ranked above" has two readings: 1) an argument is non-marked in terms of coding devices, i.e. it is coded in the same way as the sole argument of the basic intransitive construction; 2) an argument is privileged for accessing syntactic phenomena sensitive to some hierarchy of arguments, in much the same way as the sole argument of the basic intransitive construction. Thus, collapsing ergativity with passive is discarded because passive is neither basic nor transitive; with inverse because inverse is not basic; with (classical) split intransitivity because there is, in its case, no intransitive construction to be called more basic than the other. Whether or not one and the same argument captures properties 1) and 2) yields, respectively, homogenous ergativity (morphology and syntax) or heterogenous ergativity (only morphology). Notice that by these definitions, the difference between the nominative-accusative pattern and the absolutive-ergative pattern rests solely on how semantic roles map onto morphosyntactic entities. (A clear parallel to this is the direct-indirect object vs. primary-secondary object distinction.) If we were to neglect the existence of semantic roles — or, rather, their mapping onto formal properties — there would be no difference to be made between the nominative-accusative pattern and the absolutive-ergative pattern (nor between the direct-indirect object pattern and the primary-secondary object pattern). Now, it is also true that if we were to do such a thing, we would probably deprive ourselves of any possibility of understanding the fundamental asymetry between both kinds of patterns, namely, the fact that cross-linguistically ergativity is predominantly heterogenous while accusativity is massively homogenous.

Katukina is a rather isolating language^[1], with a parcimonious morphology mainly concerning person, direction, and aspect affixes. Person prefixes distinguish three persons and two numbers. TAM particles are pervasive in discourse in certain dialects (Itaquai), and scarce in others (Bia). The language displays ergative patterning at morphological and syntactic levels. I will first present a split in transitive constructions, showing the extent to which each pattern, accusative vs. ergative, is internally consistent. As far as coreference is concerned, the accusative construction seems more consistent than the ergative construction. I will take this as the basis for a hypothesis concerning their asymmetry in terms of diachrony.

1 Split

Two transitive construction types.

1.1 Ergative

Two clausal constituents. Order: verb phrase - noun phrase. Case morphology on the noun internal to verb phrase, the verber. No morphology on the external noun, the verbee^[2].

(1)

verber VRB verbee

(2)^ITQ?	[mapiri-na	duni]	takara
	Snake-MkCase	Catch	Hen
	'snake cought hen'

1.2 Accusative^[3]

Two clausal constituents. Order: verb phrase - noun phrase. No case morphology either on the noun internal to verb phrase, the verbee, or on the external noun, the verber.

(3)

verbee VRB verber

(4)^ITQ?	[takara	duni]	mapiri
	Hen	Catch	Snake
	'snake cought hen'

1.3 Motivation for the split

The accusative construction is about ten times less frequent in texts than the ergative construction.

Motivations are unclear. Maybe genericity of verbee: in the accusative construction the latter is often generic; first and second person cannot occur in verbee position in the accusative construction^[4]; sometimes the accusative construction is translated by the informant as a cleft focussing on the verber "it was X who did such and such things to Y".

2 Alignments

Here I justify "ergative", "acusative" and "marked case".

As far as intransitive is concerned, there is only one construction type: two clausal constituents; order verb phrase - noun phrase; no pronominal morphology either on verb or on noun.

(5) VRB unic

(6)	hoki	Kariwa
	Talk	NonIndian
	'the non-Indian talked'

2.1 Morphology

Morphological alignment appears only in ergative construction

- marked case on verber noun phrase / no mark either on verbee or on unic

- pronominal prefix on transitive verb for verber / no affix for either verbee or unic

(7)^ITQ?	a-duni	takara
	3ºSingular-Catch	Hen
	'it cought hen'

In the accusative construction we see neither case marker on noun nor pronominal affix on verb.

This identifies a marked case, present on the verber of the ergative construction (but also, with the same suffix, on genitive noun and object of postposition; phonologically, the case marker appears as procliticized onto the phrase head; it is "homophonous" [of course, this is plausibly no homophony] with an oblique allative suffix [present on the Bia but not on the Itaquai], used for spatial target and recipient [possibly oblique] of "give").

2.2 Syntax

2.2.1 Constituency

External to verb phrase and postposed to it: unic, and 1) verbee in ergative construction, 2) verber in accusative construction. Internal to verb phrase: verber in ergative construction, verbee in accusative construction; no word (particle, adverb) can intervene between the internal argument and the verb.

2.2.2 Movement

In both constructions, the external argument can be moved, whereas the internal argument cannot move out of the verb phrase without changing or loosing its argument status^[5].

2.2.3 Nominalization, relativization, focalization, pronominalization by demonstrative, pronominalization by zero, show an alignment between verbee and unic in ergative construction (see Queixalós 2003). Most of these constructions have yet to be fully tested on the accusative examples, if occurring. But, concerning the accusative construction, I can already assess that, differently from verbee, verber is accessible to omission:

(8)^ITQ	[tukuna	makoniok]	Ø
	People	GiveAdvise
	'He gave advise to the people'

and to focus:

(9)^ITQ?	adu	na	[wiri	hak]
	1º	Focus	Pig	Arrow
	'I arrowed a pig'

3 Coreference

Patterns of coreference seem to be more stable and consistent in accusative constructions than in ergative constructions.

As far as I can see, there is no difference in Katukina between conjuntion reduction (he₁ went out and Ø₁~ he₁ coughed)) and zero pronominalization (each time John₁ went out Ø₁ coughed, impossible in English). The possibility of a choice between zero and an explicit full pronoun [in fact between Ø/prefix versus a pronominal noun phrase] allows the latter to signal that a switch in reference is taking place (disjoint reference)^[6]. We will see an example of that below, in (17).

3.1 Accusative construction

Differently from Dyirbal (Dixon 1994 : 162), Katukina displays an accusative coreference pivot on the accusative construction.^[7]

An example of coordination in which a verber controls^[8] a subsequent unic is:

(10)^ITQ	[tukuna₁	buhuk]	Tamakori₂	tona	niama Ø₂^[9]
	People	Make	Tamakori	Leave	Then
	'Tamakori₂ created the people and then left₂'

Conversely, a unic in a controlling a subsequent verber in b:

(11)^BIA a		da/adik	Tamakori₁		hak-dik ...
		GoOut	Tamakori		House-Locative
		'Tamakori₁ left his house ...'
b	... [wanadakbii			ha/ori₂		buhuk]	niama	Ø₁
	PalmSp.			Rope		Make	Then
	'... and then he₁ made a palm sp. rope'^[10]

We can even observe a coreference pivot between a verbed in ergative construction and a verber in accusative construction, but in what should maybe be seen, due to the absence of niama, as a continuous sequence of two syntactically unconnected sentences:

(12)^BIAa	[piida₁-na	homan]	oon₂...
	Jaguar-MkCase	Call	Toad
	'the jaguar₁ called the toad₂...'
b	... a₁-pata	[kori/on₃	waikman]	Ø₂
	3º-Onto	Vine	Throw
	'... he₂ threw a vine down to him₁'^[11]

These examples lend some support to the notion of a non marked case subsuming those of absolutive in the ergative pattern and nominative in the accusative pattern: both arguments are external to the verb phrase.

3.2. Ergative construction

It is fairly mixed as far as coreference patterns are concerned, but displays some typical facts of ergative alignments, the latter seeming impressionistically predominant in texts. Here I must mention the preliminar counts on referential distance and topic persistence that Spike Gildea made on one of my texts a few years ago, which seem to show a higher discourse topicality of verber in relation to verbee^[12].

Each time it is possible, I will try to illustrate the patterns by identifying the argument status of controller referents (antecedents) and controlled referents (pronominal forms, including zero). Two factors have to be considered concerning the controller-controlled relation: linear order and mutual structural position (the latter being rendered in terms of c-command in the generative framework).

3.2.1 Intraclausal

3.2.1.1 Core participants

The issue concerning control of reflexive (I do not include possessive in "reflexive") does not arise, since the reflexive construction is equivalent to an intransitivization of the verb.

(13)^ITQ?	i-tohik	Owi
	1ºSingular-See	Owi
	'I saw Owi'
(14)^ITQ?	tohik-i		Owi
	See-Intransitivizer		Owi
	'Owi saw herself'

The -i sufix is specialized in intransitivization for reflexivity purposes. No reflexive pronominal form occurs.

Now, the possessive proper can be controlled by either argument. By the verbee, as in:

(15)^ITQ?	[a₁-obatyawa-na	todiuk]	Mayon₁
	3ºSingular-Wife-MkCase	Hate	Mayon
	'Mayon₁'s wife hates him₁' (litt.: His₁ wife hates Mayon₁')

Here, the antecedent does not precede the anaphoric expression, but it c-commands it, see (1).

Or by the verber, as in:

(16)^ITQ?	[Dahwi₁-na	bobo]	a₁-batsawa
	Dahwi-MkCase	Beat	3ºSingular-Wife
	'Dahwi₁ beat his₁ wife'

Here, the antecedent does precede the anaphoric expression, but it does not c-command it.

Verber control of possessive on verbee, as in (16), seems to be obligatory when verbee comes after, since, in that order, disjoint reference (i.e. non expected) must be signalled by insertion of a noun phrase, for example a third person pronoun:

(17)^ITQ?	[Dahwi₁-na	ti]	anyan₂-na	wa	takara
	Dahwi-MkCase	Kill	3º-MkCase	RGN^[13]	Hen
	'Dahwi₁ killed his₂ hen'

The verber can control the possessive on the verbee even though the verbee comes first:

(18)^ITQ	ma₁-wa	baohnin	[paiko	hinuk₁-na	manamana-nin]	tyaninhan
	3ºPlural-RGN	Garden	Grand-Father	Collective-MkCase	Make-Progressive	LongAgo
	'our grand-fathers₁ were making their₁ gardens long ago '

Here, none of both conditions (order, structural position) is met.

But in this circumstance control ceases to be obligatory:

(19)^ITQ	ma₂-obatyawa	[kotyia₁-na	dyo/oro]
	3ºPlural-Wife	Otter	Sodomize
	'otters₁ sodomized their₂ wives'

3.2.1.2 Non-core participants

There is some asymetry between both arguments concerning their ability to control possessive on non-core noun phrases, in that verbee aligns more clearly with unic. Respectively:

(20)^ITQ?	[Dahwi₁-na	bobo]	ityaro₂		a₂-wa	hak-naki
	Dahwi-MkCase	Beat	Woman		3ºSingular-RGN	House-Locative
	'Dahwi₁ beat the woman₂ in her₂ house'
(21)^ITQ?	horon	Dahwi-na		obatyawa₁	a₁-wa	panera-katu
	GetBurnt	Dahwi-MkCase		Wife	3º-RGN	Pot-With
	'Dahwi's wife₁ got burnt with her₁ pot'

Control by verber seems to be less natural. We can get

(22)^ITQ?	[Poroya₁-na	tohik]	Dahwi₂	a₁-wa	hak-naki
	Poroya-MkCase	See	Dahwi	3ºSingular-RGN	House-Locative
	'Poroya₁ saw Dahwi₂ in his₁ house'

but most of the time the informant changes the scene so as to have the verbee in his own house, or shifts to a construction resuming the verber noun phrase.

(23)^ITQ?	[Poroya-na	bobo]	Dahwi	Poroya-na	wa	hak-to
	Poroya-MkCase	Beat	Dahwi	Poroya-MkCase	RGN	House-Locative
	'Poroya beat Dahwi in Poroya's house'

which is formally equivalent to introducing a new participant in the owner slot, as in

(24)^ITQ?	horon	Dahwi-na	obatyawa₁	anyan₂-na	wa	panera-katu
	GetBurnt	Dahwi-MkCase	Wife	3º-MkCase	RGN	Pot-With
	'Dahwi's wife₁ got burnt with his/her₂ pot'

To be honest, I have also one single example where the informant resumes the verbee noun phrase.

(25)^ITQ?	[Poroya-na	bobo]	Dahwi	Dahwi-na	wa	hak-to
	Poroya-MkCase	Beat	Dahwi	Dahwi-MkCase	RGN	House-Locative
	'Poroya beat Dahwi in Dahwi's house'

3.2.2 Interclausal

We will not be able to get a complete picture of the situation here, since many relevant data are lacking.

3.2.2.1 Coordination

There is no coordination marker. What we have are paratactic sequences (unless intonation give clues to coordination, but this aspect has not not been studied yet). But the discourse connecting particle niama makes two sentences more tightly linked to each other than the lack of the particle. Here again I distinguish controller and controlled in the sense mentioned above. As far as controlling is concerned, once again verbee and unic show more affinities to each other than any of both to verber.

Unic controls a subsequent unic,

(26)^ITQ?	waokdyi	opatyin₁	dyadyi	niama	Ø₁
	Arrive	Child	Sing	Then
	'The child₁ arrived and then he₁ sang'

as it controls a subsequent verbee:

(27)^ITQ?a	waokdyi	Nodia₁,	[Yowai₂-na	tohman]	niama	Ø₁	...
	Arrive	Nodia	Yowai-MkCase	Kill	Then
	'Nodia₁ arrived, and then Yowai killed him₁ ...'

(The sentence ends with

b	...	dadyoran-nin	Ø₁	ton
		Enter-Progressive		Locative
	'... as he₁ came in').

Verbee controls a subsequent unic:

(28)^ITQ	[dyoori₁-na	man]	wa	hinuk₂,	dadohan	niama	Ø₂
	Termite-MkCase	Do	Woman	Collective	ClimbUp	Then
	'the termite advised the women, and then the latter climbed up (a tree)'

Verbee's control of subsequent verbee will be shown through what I call homothetic coreference (verbee controls verbee, verber controls verber). Compare a and c:

(29)^ITQa

i-tyanhwan,

tyiktyikna

adik,

[a₁-daman]

niama

wa₂...

1º-SisterInLaw

Urinate

3º-Say

Then

Woman

'hey, sister-in-law, let's go urinate! she then said to the woman...'

... [a₁-dahu-nin]

Ø₂...

3º-Take-Progressive

'...taking her (down to the river), ...'

... [a₁-ti]

niama

Ø₂

wah-tata

3º-Kill

Then

River-Locative

'...and then she killed her in the river'

But homothetic coreference does not test any kind of alignment, since it shows no asymetry between both arguments.

If we compare the following example to (28), we get a sort of minimal pair showing that control of a subsequent unic is possible for verbee (28) and for verber as well (30), thus revealing the lack or weakness of a coreferential pivot. Both are extracted from the same episode of one single text.

(30)^ITQ	[...] [wa₁-na	da-man]	dyoori₂,	da/an	niama	Ø₁
	Woman-MkCase	WhileLeaving-Do	Termite	Go	Then
	'[...] the women said to the termite, and then they went away'

3.2.2.1 Subordination

Now let us turn to subordination. Subordination is not easy to identify, since 1) independent clauses can appear without any TAM particle while keeping their grammaticality; 2) the person prefix paradigm is retained in all transitive clauses built along the ergative pattern, be they independent or dependent; 3) morphemes which seem to be used as subordinators display other — probably more basic — functions in independent clauses. For example -nin is a gerund marker, as well as a progressive aspect marker in independent clause; niama is a purpose subordinator, and a discourse connector 'then' on independent clause; ninwu is a purpose subordinator, and also a volition auxiliary in independent clause^[14]. The following fragment shows both functions of niama, first syntactic, then discoursive^[15]:

(31)^ITQa	[pida₁-na	odantabu]				waadyo₂...
	Jaguar-MkCase	RunAfter				Monkey
	'jaguar₁ ran after monkey₂...'
b	... [a₁-ti]			niama			Ø₂, ...
	3ºSingular-Kill			Purpose
	'...to kill₁ it₂, ...'
c	... dadyoran		niama		Ø₂
	Enter		Then
	'... and then it₂ entered (into a hole)'

b cannot be interpreted as coordinated — niama's discoursive function — meaning 'then it_jaguar killed it_monkey' since the monkey has got to stay alive in order to enter the hole in c. So, for the time being, I will have to rely on semantics — and other expedients — more than I would like to for sorting out what I will be calling subordinate clauses^[16]. Note the homothetic coreference between a and b, and also the verbee as controller in a and b of a subsequent unic in c.

The progressive morpheme as a gerund mark we have already seen in (27)b, and perhaps in (29)b. Another example is^[17]:

(32)^ITQ?	[o/-na	tohik]	Wura₁	totatomahik-nin	Ø₁
	SomeoneElse-MkCase	See	Wura	Die-Gerund
	'someone else saw Wura₁ as he₁ died'

(The exact nature of Ø here has not been tested, i.e. whether it should be seen as a zero pronoun or the trace of Wura's raising. The test is easier for the subordinate ergative verber, due to its explicit case mark, which it looses when raised to the matrix verbed position.) The informant rejects the sentence without -nin, behaviour which I interpret as a clue to the subordinate status of the second clause. In one reading of Ø we would have again a verbee as controller of a subsequent unic. Now we see a gerund with a controlled verbee:

(33)^ITQ?	horon	Dahwi-na	obatyawa₁	[a-wa	panera-na	bok-nin] Ø₁
	GetBurnt	Dahwi-MkCase	Wife	3ºSingular-RGN	Pot-MkCase	Roast-Gerund
	'Dahwi's wife got burnt with her pot' (litt.: 'Dahwi's wife₁ got burnt, her pot roasting her₁')

My use of the term gerund does not imply any "same subjet" kind of constraint on coreference, no matter what should or could be called "subject" in this language. In both (33)-(34), one of the arguments of the gerund clause finds its antecedent in the main clause. In (33) a verbee is controlled by a unic — if verbee were subject, we would have a "same subjet" gerund. But in (34) it is a verber (a₂-) which is controlled by a verbee (Ø₂) in a sequence of two transitive clauses.

(34)^ITQ	[ma₁-man-na]	Ø₂	wiri	[a₂-man-nin]
	3ºPlural-Do-Directional		WildPig	3ºSingular-Do-Gerund
	'they₁ sent him₂ get₂ wild pigs'

An example of homothetic coreference in subordination appears in

(35)^ITQ	[paiko	hinuk₁-na	todiuk-nin]	Ø₂	[ma₁-hakhak]	ninwu	Ø₂
	Elder	Collective-MkCase	Fight-Progressive		3ºPlural-Stab	Purpose
	'the elders₁ were fighting them₂ planning to stab₁ them₂'

So, if we could prove that what we have in (31)b, (33) and (35) are non finite subordinate clauses, we would be facing what in Kakutina would look most like a PRO^[18], parallel to Dyirbal's (Dixon (1994:168)

(36)	yabu₁-Ø	[Numa₂-Ngu	giga-n]	Ø₁	[gubi₃-Ngu	mawa-li]
	Mother-Absolutive	Father-Ergative	Send-NonFuture		Doctor-Ergative	Examine-Purpose
	'Father₂ sent mother₁ for the doctor₃ to examine her₁'

The interesting fact, of course, is that in both languages the putative PRO is a verbee.

4 Split, coreference and diachrony

We have seen 1) a consistent accusative pattern of coreference in the accusative construction — to be confirmed by supplementary data, of course —; 2) a weak ergative pattern of coreference in the ergative construction. The overall picture is obscured because in several occasions I am unable to state, for the phenomenon currently described, two basic conditions concerning the grammatical status of phonologically null noun phrases: 1) obligatoriness of zero realization, and 2) obligatoriness of coreference. For sure, the existence of coreference pivots is no obligation for a language^[19]. But Katukina, with its poor morphology, seems to be the kind of language where reference tracking devices should operate on the basis of invisible rules sensitive to the arguments hierarchy as observed through constituency and accessibility.

I would like to put forth the idea, based on a few preliminary and non systematic observations of languages other than Katukina, that 1) each time a so-called ergative language displays various kinds of splits in transitivity, coreference is among the phenomena that work on an accusative basis (Pano could be an example among others^[20]; Dyirbal could be a counter-example, with its coding devices split between nouns and pronouns, and its homogeneous ergative coreference pattern^[21]); 2) if a so-called ergative language displays a single split, the accusative facet of the split is about coreference, e.g. yanomami (Ramirez 1994), karao (Brainard 1997).

The diachronic reading of this idea I include within a hypothesis which says not only that ergativity is precarious in nature (Givon's "self defeating" pattern; see also Nichols 1993), but that 1) its fate is to gradually shift to accusativity; 2) this shift takes place along a relatively fixed path, beginning with syntax then moving on into morphology, and within syntax, beginning with coreference then continuing with accessibility and constituency phenomena.

(37) The accusativization path:

Syntax Morphology

... ...

Coreference Other

First line of (37) shows Givón's idea (2001 : 219) that behaviour and control properties undergo diachronic change before morphological properties, because of their greater exposure to discourse pressures (topicality, in the case of the accusativization path). The second line represents Givón's idea (1997a) that in karao the accusative pattern for coreference is a historical retention of accusative properties.

Turning back to Katukina, if the outline in (37) is correct, the weak ergative pattern of coreference observed in the data of 3.2 is the result of a diachronic erosion, corresponding to the first steps of the language on the path of accusativization. This would imply that at former diachronic stages, coreference in Katukina was more strongly anchored in the ergative pattern. A fact that could be reminiscent of such a former stage: the use of antipassive, a typical feature of ergative patterning of constructions, for coreference purposes, when a verber has to be controlled by a unic or by a verbee. We saw in (34) a very simple way to achieve this, in fact a way consistent with the lack of pivot or, maybe, a split pivot with control verbs proper. The other way to make this available, now consistent with the existence of a strong ergative-type pivot, is turning the second clause to an antipassive, so as to get the verber in the external argument's position:

(38)^ITQ?	[Nodia₁-na	pikan]	Owi₂	wa-tohik	tu	niama Ø₂
	Nodia-MkCase	Hear	Owi	Antipassive-See	Negation	Then
	'Nodia₁ heard Owi₂, but she₂ did not see (him₁)'

A more natural example, but less neat^[22], is

(39)^ITQ	[i-tohman]	anyan₁	tya	bo,	wa-biwik-nin	Ø₁	kotu	da
	1º-Shoot	3º	Future	Exclamatory(?)	Antipassive-Smoke-Gerund		Particle	Again
	'I'll shoot that one₁ if he₁ smokes^[23] again'

In Dyirbal — where coreference splits are unknown — the antipassive is also obligatory in this case:

(40)	yara₁	[gubi₂-nggu	balga-n],	jugumbil₃-gu	Ø₁	jilwal-ng-nyu
	ManAbsolutive	Doctor-Ergative	Hit-NonFuture	Woman-Dative		Kick-Antipassive-NonFuture
	'the doctor₂ hit de man₁ and he₁ kicked the woman₃'^[24]

Within my hypothesis for Katukina, (38)-(39) would be conservative, reflecting a stage similar to Dyirbal's, whereas (34) would be innovative.

Now the issue arises of the position of the accusative alignment within this diachronic picture. Let us assume that the idea of coreference being the very first ergative pattern eroded by accusativization of the morphosyntax is right. And let me remind that in Dyribal the coreference pattern within the accusative construction is all the way ergative. How could the Katukina pattern of coreference be so consistent within the accusative construction, while being so seriously damaged within the ergative construction? Only if the accusative construction in Katukina were no innovation towards accusativity — as plausibly are many accusative constructions in transitive splits — but a testimony of what was, in pre-ergative times, the form of the basic active transitive construction.

Of course this suggests a historical scenario for the origin of ergativity in Katukina — I am not speaking of motivation^[25]. We had once an accusative language with constituency and order of elements as

(41) [verbee VRB] verber

(42) [VRB] unic

Eventually there appears a formally intransitive construction where the verb whether receives a genitive person prefix or heads a noun phrase with verber as adnominal complement, duly marked for genitive case. The verbee is the subject of the nominal predicate. Now, the construction type (41) needs not be eliminated, it just ceases to be the most basic one.

(43) [pp-VRB] verbee = Pig is [his killing]

(44) [verber-genitive VRB] verbee = Pig is [John's killing]

What makes the clause transitive, and hence ergative, is the recovery of finite verb properties by the former nominalized verb. This is something I have no clear idea of how it has been achieved in Katukina. Plausibly through the TAM particles, so discoursively prolific in certain dialects and so scarce in others. Whether this difference, at first sight in style rather than in grammar, is in fact imputable to the mentioned process of recovery will probably remain to be investigated until access to comparative data is granted. That will only have a real chance to happen if we manage to discover the location of the Katawixi indians, supposing they are not an isolated or extinct society.

References

BATH, D. N. S. 1991. Grammatical relations. The evidence against their necessity and universality, London & New York, Routledge

BIGGS, B. (1974. "Some problems of Polynesian grammar" Journal of the Polynesian Society, 83, Auckland

BITTNER, M. & HALE, K. 1996. "Ergativity: Toward a Theory of a Heterogeneous Class", Linguistic Inquiry 27.4, 531-604

BRAINARD, S. 1997. "Ergativity and Grammatical Relations in Karao", Givón, T. 1997b.

GIVON, T. 1983. "Topic continuity in discourse: the functional domain of switch reference" Haiman, J. & Munro, P. (eds.) Switch-Reference and Universal Grammar, John Benjamins, Amsterdam/Philadelphia

1997a "Introduction" Givón, T. 1997b.

1997b. (ed.) Grammatical Relations. A Functionalist Perspective, Typological studies in language, 35, Amsterdam/Philadelphia, John Benjamins

LEVINSON, S. 1987. "Pragmatics and the grammar of anaphora" Journal of Linguistics, 23, pp. 379-434

MEL’CUK, I. 1983. “Grammatical subject and the problem of the ergative construction in Lezgian” Haronson, H. & Darden, B. (eds) Proceedings of the second conference on the non-Slavic languages of the USSR, 5, pp. 246-293

MITHUN, M. 1991. "The role of motivation in the emergence of grammatical categories: the grammaticalization of subjects" Approaches to grammaticalization 02, Traugott, E. & Heine, B. (eds.), Amsterdam, John Benjamins, pp. 160-184

NICHOLS, J. 1993. "Ergativity and linguistic geography" Australian Journal of Linguistics 13: 39-89

PAYNE, Th. 1982. "Role and reference related subject properties and ergativity in Yupi’k Eskimo and Tagalog", Studies in Language, 6, pp. 75-106

RAMIREZ, H. 1994. Le Parler Yanomamö des Xamatauteri. Doctoral Dissertation, Aix-en-Provence, Université de Provence

[1] Family Katukina, state of Amazonas, Brazil. itq heads data from Itaquai river (also known as the Kanamari "language") . bia heads data from Bia river. Question mark: elicited data. Square brackets: verb phrase in transitive clause. MkCase means "marked case".

[2] See see Section 2.1 for verbal morphology. See Queixalós (2003) for justification of the "semantic" labels verber and verbee.

[3] An accusative alignment which I will not comment on here concerns the imperative construction: the only allowed noun phrase is that of the verbed of transitives.

[4] But there are examples of generic verbee in the ergative construction and examples of specific verbee in the accusative construction. Examples of the latter include

1) a determined noun:

(a)ITQ?	[itian	koya	buhuk]	idik
	This	Beverage	Make	You
	'you made this beverage'

(other informants (Bia) reject this construction, as they reject it with a possessive instead of demonstrative, switching to ergative construction), or

2) a third person pronoun:

(b) ITQ?	anyan	duni	mapiri
	3º	Catch	Snake
	'snake cought him'

(other informants (Bia) reject this construction), or

3) a proper noun:

(c)BIA?	[Antonio	tohikman]	Ayobi
	Antonio	LookAt	Ayobi
	'Ayobi looked at Antonio'

or even

4) a bare noun whose referent is pointed to within the conversation scene:

(d) BIA	[oman	toki]	adu
	Pole	Put	1º
	'I put that pole'

Examples (a)-(c), if confirmed at least for ITQ, would show the difference between accusative constructions and what could be read as noun incorporation. The test is not always available for distinguishing between both types of structures on formal grounds, but it seems that noun incorporation is always redistributive (i.e. it preserves the grammatical relations slots), keeping the ergative pattern.

[6] Levinson (1987: 384), among others: "the more 'minimal' a form, the stronger the preference for a co-referential reading". In terms of testing constructions for antecedents, I('ll) will make no difference between zero and verb prefix (Givón 1982: both "correspond to the coding points of highest topic continuity").

[7] Katukina is more like Yidiny, in that coreference alignments parallel coding alignments (Dixon 1994: 175).

[8] I('ll)will be using 'control' for the relation between an anaphoric expression — including zero — and its antecedent, in a broader sense than that which appears in so-called control constructions/verbs (promise, persuade, etc.).

[9] Ø is noted for the sake of clarity in reading the examples. Its order relative to niama is inferred from examples like (31)a below.

[12] Quoting Spike's comments: "The agents of transitive clauses are definitely topical, and they are the primary topic a substantial percent of the time." Notwithstanding, "topic persistance counts for O [my verbed, FQ] are surprisingly high". There is here an issue that desserves attention from our part: what is the kind of consistence that we can expect to observe between statistical results of topicality counts and grammatical constraints on coreference? Particularly in a context of mixed coreference, it seems to me that it is quite possible to see in the same piece of discourse predominant grammatical coreference patterns of the ergative type and an overall more topical verber in discourse. For example: a series of ten transitive clauses with same verber, say he1 did such and such things, among which are three clauses of the type

1) (...and) he1 met Paul2 in his2 house

and one clause of the type

2) (...and) he1 met Peter2 in his1 house

This would lend in discourse a ratio of 11/1 anaphoric mentions in favor of verber, but a ratio of 3/1 instances of intraclause control of possessive in favor of the verbee.

[13] Relational generic noun (so-called "possessive classifiers"); see Queixalós (2005).

[15] I ('ll) will tentatively give a different gloss for each function.

[16] Levinson (1987), elaborating on Haviland's description, notes that "nearly all the so-called subordinate constructions in Guugu Yimidhirr, including the purposive, can occur as main clauses".

[18] The hallmark for subjectness in certain framewoks (Bittner & Hale 1996).

[19] See Mithun (1991) and Bath (1991), on morphologically rich languages.

[20] If we discard alignment in relativization as a kind of coreference alignment: in Shipibo-Konibo (Valenzuela 2002), relativization is ergative, but coreference is accusative.

[21] It is unclear whether yupik should be held as another counter-example (Payne 1982).

[22] "Natural" because it was spontaneously uttered by the informant in the middle of an elicitation session. Less neat because antipassive here could have a semantic motivation.

[24] Taken from Comrie (2003), where probably ng = N, and -Ø absolutive is not markerd.

[25] For syntactically ergative languages, Givón assumes an accusative ® inverse ® ergative path instead of the more general passive origin assumption. Another possible non passive origin would be what I call the intransitive languages (Mel'cuk's (1984) Lezgh, Biggs' (1973) Polynesian).