Evidence for Grammatical Functions in Tongan

3.1 Evidence for the SEA.

Chung 1978 is the most complete defence of the surface ergative analysis of Polynesian ergativity, drawing primarily on data from Tongan and Samoan. Her arguments are of two general types. Firstly, she demonstrates quite clearly that in both Tongan and Samoan, there are important grammatical processes that treat absolutive NPs in absolutive (and middle) clauses as a natural class with ergative NPs in ergative clauses, arguing that this category should be identified with (nominative) 'subject'. Secondly she argues that there are processes which treat absolutive NPs in ergative clauses as a natural class with goal / locative NPs in middle clauses, suggesting that this class should be identified as (accusative) 'object'. This analysis also entails that middle clauses are classified as grammatically transitive.

The surface ergative analysis is appealing to many theoreticians because it underlies an account of the morphosyntax of such languages that is minimally different from familiar nominative / accusative [N/A] languages. To the extent that such reductions are empirically justifiable, the same grammatical machinery can be used to account for both types of language. Furthermore, this analysis provides the most straightforward account of linking between thematic relations and grammatical functions, matching thematic prominence directly to grammatical prominence. Nevertheless, there are good reasons for thinking that this reductionist account of Polynesian ergativity is on the wrong track and fails to account for important grammatical properties of these languages.

3.1.1 'Subject'-Referring Rules.

The evidence provided in Chung 1978 (chapters three and four) in support of the idea that ergatives and intransitive absolutives form a subject category in Tongan (and Samoan) is briefly summarized and discussed below. One argument, involving a raising process in Tongan, is omitted because the generalizations upon which the argument rests are incorrect (Dukes 1998). Examples and extended discussion are omitted in most cases for reasons of space. Additional discussion and exemplification can be found in Dukes 1998.

Preverbal Pronominalization. A rule of 'subject' pronominalization (found in all ergative Polynesian languages except Niuean) places a pronoun before the verbal complex, typically as an enclitic on the tense aspect marker or a conjunction. This rule can apply only to ergative pronouns and to absolutive pronouns in absolutive clauses; i.e. to the category identified as 'subject' under the SEA. This phenomenon provides clear support for the SEA.

(4)	Ne	ta:ta:puni	fakavave	leva	'a	e	fale	ifi,
	PAST	shut	quickly	then	ABS	DET	house
	ka	mau	hu:	ki	tu'a
	CONJ	1PL.EX	go	GOAL	outside
	'The we quickly shut the bandhouse and went outside' (FHJ)

(5)	Na'a	mau	fu'u	ongo'i	aupito	'a	e	momoko
	PAST	1PL.EX	very	feel	very	ABS	DET	cold
	'We were all really feeling the cold....' (MR)

Tongan equi. A rule labelled 'Tongan equi' deletes ergative pronouns and absolutive pronouns in absolutive clauses when they are embedded under a verb of motion that takes a complement clause marked with the conjunction / complementizer 'o .

(6)	Pea	'alu	'a	e	tokoua	'o	vakai	'a	e	vakatoka
	CONJ	go	ABS	DET	guy	CONJ	look.at	ABS	DET	ship
	"And then the guys wento to look at the ship' (MR)

To the extent that this phenomenon can be differentiated from a general discourse process of zero prominalization (not a trivial issue), the data can be taken as support for the SEA.

Samoan equi. An optional rule labelled 'Samoan equi' is claimed to be governed by a number of predicates with typical control-type semantics, including mana'o 'want', taumafai 'try' and fa'ama:losi 'encourage'. The targets of the process are claimed to be complement ergative NPs and absolutive arguments of absolutive predicates. To the extent that this process can be distinguished from zero-pronominalization (again, not a trivial issue) the data can be taken as support for the SEA.

Possessor Marking. The case marking of preverbal possessor pronouns in nominalized clauses follows an N/A pattern in both Tongan and Samoan. When a predicate is nominalized, an ergative or intransitive absolutive argument is replaced with a pronoun from the 'dominant' possessor series, while a transitive absolutive argument is replaced with a pronoun from the 'subordinate' series.

(7)a	he'eku	taki	b	hoku	taki	c	he'eku	mohe
	1SG.DOM	lead		1SG.SUB	lead		1SG.DOM	sleep
	'my leading'			my being led'			my sleeping

These basic facts are essentially parallel to those observed in the case of preverbal pronominalization in verbal clauses above and again point to a notion of subject consonant with the SEA.

Samoan raising. Chung describes a raising process in Samoan governed by the predicates mafai 'can', tatau 'be necessary', 'a:mata 'begin', fa'apea 'be done thus'. She shows that the targets of this raising process must be either ergative or intransitive absolutive NPs, i.e. subjects under the SEA. This data again provides support for the SEA.

It is important to note however that there are raising processes in several ergative Polynesian languages that appear to be quite unrestricted with respect to the target NP in the complement clause. Chung herself notes that raising with ma:sani in Samoan is unrestricted. Furthermore, raising predicates cognate with the Samoan ones are found in the ergative languages Tuvaluan (mafai, ttau, ka:mata, ma:sani; Besnier 1988) and Tokelauan (mafai, tatau, ka:mata, ma:hani; Hooper 1993). Besnier and Hooper demonstrate quite clearly that these predicates, amongst a number of others, can raise complement NPs of virtually any grammatical type, though they most commonly affect ergatives and absolutives. Thus while the facts involving some raising predicates in Samoan support the SEA, the facts of raising in most other languages don't point one way or the other. The only obvious exception is the case of raising in Niuean, which clearly points in favour of the ASH since raising is restricted to ergatives and absolutives in that language (Seiter 1980).

The Subject Referring Rules Summarized. The evidence summarized above provides a strong case for an N/A subject category in Tongan and Samoan or alternatively, for a semantically based notion of 'logical subject' or 'Actor' (Schachter 1976, 1977, Dixon 1979). It seems reasonable to assume therefore that some such notion is involved in the organization of Polynesian syntax. However, an additional crucial claim of the SEA hypothesis is that there is a corresponding notion of object subsuming the absolutive arguments of ergative clauses and the goal / locative arguments of middle clauses. We turn now to the status of this claim.

3.1.2 'Object'-Referring Rules.

Evidence provided for the idea that transitive absolutives and middle goal/locatives form an object category is considerably weaker than that for subjects and is summarized below. Only two arguments are provided in the relevant chapter of Chung's text for an object category in the ergative languages, but two others can be reconstituted from arguments in other chapters.

Object incorporation in Samoan. Chung notes the existence of a process of nominal incorporation in Samoan which is described as applying to some absolutive arguments in canonical transitive clauses and to some goal / locative arguments in middle clauses. The relevant NP is stripped of any case marking and is placed to the right of the verb, forming an intransitive compound verb. To the extent that the rule is restricted to the categories described above it lends support to the SEA notion of objecthood. Such a rule is also observed in Tongan where does not apply with any great frequency. However it does not apply solely to objects, as Churchward (1953:77) points out. Chung (1978:187) also notes that incorporation in Samoan can apply to affected locatives.

A more general problem with this argument is that it is far from clear whether incorporation is in general constrained directly by grammatical relations or by some more semantically based notion of argument roles, as is widely supposed in the morphological and lexicalist literature (e.g. Mithun 1984). Indeed, facts from Niuean, the language most closely related to Tongan, appear to support this contention. Seiter 1980 shows that, by all available tests, middle objects are clearly obliques in Niuean. But there is one curious exception to the generalization. Noun incorporation can affect middle objects, suggesting that the targets for incorporation are not grammatical objects but a more general, semantically based category which includes logical objects. Such facts call into the question the relevance of the noun incorporation data for the SEA / ASH debate.

Quantifier Float in Tongan. Chung describes a process of Quantifier Float [QF] involving the quantifier kotoa 'all, every, whole' in Tongan which is claimed to affect ergative NPs, absolutive NPs and goal / locative arguments of middle predicates. This set of NPs corresponds to the class of subjects and objects under the SEA. The quantifier normally appears as the terminal element of an NP. When QF applies it appears to the immediate right of the governing verb.

(8)	Na'a	na	fai	pehe:	he	'aho	kotoa	'o
	PAST	3DL	do	thus	DET	day	all	CONJ
	na	a'u	ki	he	hela'ia
	3DL	came	to	DET	tiredness
	'So they did this all day and became tired' (PF) (No QF)

 

(9)

Na'e

hanga

leva

'e

he

tamasi'i

'o

ta:naki

kotoa

'a

e

kie

PAST

turn

then

ERG

DET

boy

CONJ

collect

all

ABS

DET

leaf

'And so then the boy collected up all the (pandanus) leaves....' (PF) (QF applies)

I have recently examined the behaviour of kotoa in Tongan, looking for textual occurrences of the word in a (so far) small corpus (approx. 35,000 words from five different sources) and eliciting examples from native speakers. My own conclusions regarding the restrictions on the application of QF are quite different from those described above. The corpus contains 52 occurrences of kotoa, of which 14 cases can be plausibly described as involving QF. All the cases in the texts involve QF from an absolutive argument, 12 of them in absolutive clauses, 2 in ergative clauses. In both the latter cases, the ergative coargument is not overt in the clause. These facts suggest that QF is restricted to absolutives in Tongan, perhaps even to absolutives that are the sole overt argument in a clause.

Elicitation work with native speakers leads similarly to the conclusion that QF is considerably more restricted than was earlier thought. Only my most liberal consultant accepted QF with anything other than absolutives, though she never volunteered such examples herself.

Similar problems in pinning down the precise properties of QF can be noted in the literature from other languages, including Samoan (Cook 1988, 1991). Again, however, absolutives appear to have a distinguished status. Cook 1991 notes that Samoan absolutives may freely float the quantifier 'uma from any position following the verb, while ergatives and locatives may only float 'uma when they are immediately adjacent to the verb. Preverbal ergative arguments cannot be associated with 'uma . Mosel & Hovdhaugen 1992 also note that 'uma is usually associated with the absolutive argument of an ergative predicate and can only be associated with the ergative NP if it is immediately postverbal. They also state that 'uma must be associated with the absolutive argument of an absolutive predicate. Thus, asappears to be the case in Tongan, Samoan QF is largely associated with absolutive NPs. Indeed, the Samoan data suggests that ergative arguments appear to pattern most closely with obliques with respect to QF rather than with absolutives.

Possessor Marking in middle clauses. Chung (1978: 305-7) claims that 'middle objects' (i.e. semantically selected goal/locative arguments in middle clauses) can undergo the same 'possessive marking' process as absolutive arguments of ergative predicates in Samoan. Mosel & Hovdhaugen (1992: 544, 552) on the other hand claim that middle objects cannot be possessive marked in Samoan, nor can they be represented by prenominal pronouns. Such structures are rejected by all their consultants. Thus contrary to Chung's claim, middle objects in Samoan do not undergo the same possessive marking processes as absolutive arguments in ergative clauses. Nor can middle objects in Tongan, as far as I can determine. Native speaker consultants find the suggested structures rather bizarre. Thus the facts of possessive marking cannot be used to support the SEA notion of objecthood.

Relativization in middle clauses. Chung 1978 and Seiter 1978 discuss a possible argument for the category 'object' in Tongan, though Seiter notes that the evidence is 'somewhat equivocal'. They claim that a third person subject of a middle predicate can relativize either by deletion or by pronominalization, as is the case with absolutive arguments of ergative predicates. Absolutive arguments of intransitive predicates, by contrast, must relativize by deletion. This data suggests that middle predicates are transitive. However, both authors note that middle absolutives delete much more often that they pronominalize, thus usually patterning with intransitive subjects. Furthermore, middle objects themselves clearly pattern with obliques with respect to relativization, leaving behind a resumptive oblique trace ai. Thus the facts of relativization seem to point at least as strongly towards the treatment of middle objects as grammatical obliques as they do towards their objecthood.

The Object Referring Rules Summarized. The case presented for an object category in the ergative Polynesian languages is very weak. Additional problems for the hypothesis are found in Seiter 1977, 1980 who provides a number of tests showing that 'middle objects' are not objects in Niuean but obliques. The Niuean data, taken together with the QF and nominalization data from Tongan and Samoan suggest that the goal / locative arguments of middle predicates are not grammatical objects. The incorporation data is arguably irrelevant to the issue.

3.2 Evidence for the ASH.

We have just seen in the preceding section that there is good evidence for a notion of 'subject' consisting of ergative arguments of ergative clauses and absolutive arguments of absolutive clauses but scant evidence for a correponding notion of 'object', putatively consisting of middle goal / locatives and absolutive arguments of ergative predicates. Indeed, some of the data clearly pointed to absolutive arguments in general being treated as a class distinct from all others. Ergative arguments were also seen to pattern with obliques with respect to some phenomena.

There are several robust and fairly theory-independent pieces of evidence supporting the ASH, apart from the ones noted above in the discussion of the SEA(Fn2).

Case Marking. The first and most obvious advantage of the ASH is that it provides a clear syntactic motivation for the case marking facts in a way that the SEA fails to do. The ASH allows absolutive case marking to be interpreted directly as information picking out the 'subject' (i.e. the grammatical function SUBJ). Given an appropriate formalism, it should be possible to state a single constraint governing the appearance of absolutive case in all Tongan clauses, something that is out of the question under the SEA, which leaves the appearance of the same case marking on two putatively distinct classes of NPs as something of a morphological accident. Under this approach it is appropriate to relabel absolutive arguments as 'nominative', but I will maintain the labels used up to this point for the sake of consistency.

Argument Saturation. The treatment of absolutives as grammatical subjects provides a natural account of why it is that all Tongan predicates requiring at least one argument must obligatorily have an absolutive argument (rather than, say, an ergative argument). Under the widely held view that the subject is universally the 'least oblique' or 'final' argument to combine with an almost saturated predicate (Keenan 1974, Dowty 1982), the obligatoriness of the absolutive follows from its subjecthood. This simple datum is also inexplicable under the SEA.

Argument Obligatoriness. Biggs 1974, discussing the facts of East Futunan (another ergative language of Polynesia) and Tchekhoff 1981, discussing Tongan, point out that there is an additional sense of obligatoriness applying to the absolutive argument. They claim that an absolutive argument cannot be omitted from a sentence with the same freedom as an ergative. An omitted ergative argument need not presuppose any particular referent in the discourse and may in fact be interpreted existentially in much the same way an omitted agent in an English passive is interpreted. When an absolutive is omitted however, it must be interpreted referentially as a null pronoun picking out some previously mentioned individual. As Biggs puts it, native speakers of these languages consider sentences which are missing an absolutive to be 'incomplete', whereas sentences missing ergatives are not.

The requirement that an NP must have a presupposed referent is widely held to be a diagnostic property of subjects, particularly in Austronesian languages (Keenan 1976, Schachter 1977, etc). My own examination of Tongan texts suggests that, as Biggs and Tchekhoff point out, there is a clear distinction between the behaviour of ergative arguments and absolutive arguments with respect to referentiality under deletion. Ergative arguments, including preverbal pronouns tend to be omitted at least as often as not, whether or not a plausible potential referent has been identified in the discourse. Absolutives, on the other hand, tend to delete only under more specific circumstances when an intended referent has been previously identified. Nevertheless since third person singular absolutive pronouns are standardly deleted in absolutive clauses there is also a clear distinction between the behaviour of absolutive NPs in ergative as opposed to absolutive clauses.

The Two Constituent Bias. The differential discourse status of ergative and absolutive NPs in Samoan has been the subject of investigation by Duranti & Ochs 1990 and Mosel & Hovdhaugen 1992. Duranti & Ochs describe a general tendency in Samoan discourse to omit ergative arguments whenever possible, particularly when an absolutive argument of a (semantically) transitive predicate is also mentioned overtly in a sentence. They dub this tendency the 'Two Constituent Bias', since it leads to a high proportion of sentences consisting of a verb followed by an absolutive argument. This desired sentence structure can be achieved by several strategies. Under one strategy, the ergative argument is simply omitted from the sentence, whether or not an agent is understood from the discourse context. Alternatively, the ergative argument can be 'replaced' by a genitive modifier appearing inside the absolutive argument of the verb. A third strategy for creating a 'two constituent clause' involves the use of an absolutive predicate in place of an ergative one, thus eliminating even the possibility of an ergative argument appearing despite the fact that the identity of the agentive party is often easily recoverable from context. While I have yet to carry out any statistical examination of a large body of Tongan texts to confirm thishypothesis, exactly the same strategies are widely observed in Tongan discourse as a means of removing the ergative argument. Thus the same detransitivizing strategies also seem to play an important role in the structuring the Tongan discourse, lending particular prominence to the role of the absolutive argument irrespective of whether the predicate is itself semantically transitive or intransitive. Clauses containing semantically transitive predicates thus appear to have the constituent structure of clauses containing semantically intransitive predicates (Fn3).

Perceived Disadvantages of the ASH. The primary perceived disadvantage of the ASH is one which also afflicts the theoretical analysis of 'deep ergative' languages and explains the desire of theoreticians to reduce ergative languages to accusative ones: why should semantically transitive predicates in these languages treat their most patient-like argument as grammatically the most prominent (i.e. as the 'pivot' to use the terminology of Dixon 1979) in preference to the most agentive argument? I believe that the answer to this question does not directly impinge on the empirical issue of whether some set of the world's languages actually do have ergative / absolutive organization. If there is robust grammatical evidence in some language that the grammar is organized in such a way then grammatical theory simply must accommodate that evidence. The inability of a theory to deal with such facts should not be held as a valid motivation for rejecting an ergative analysis of the language concerned.

Another apparent disadvantage of the ASH is that it fails to account for some of the natural classes identified by Chung 1978 and others, in particular the class of subjects under the SEA definition. In the absence of a particular theory of grammatical relations, this is a valid objection to the proposal. Indeed, most of the advocates of the ASH have done so primarily for descriptive reasons and have not articulated a detailed theory-based account of grammatical relations in Polynesian. But as we will see below, current grammatical theory can accommodate the apparently conflicting sets of facts supporting the ASH and SEA relatively straightforwardly.

Created: 3 June 1998
Last modified: 12 June 1998

Authorised by: P. Austin, Professor and HOD, Linguistics and Applied Linguistics

Access: International
Copyright © 1998, The University of Melbourne.
Maintainer: Simon Musgrave