The Problem of Voice and Grammatical Functions
in Western Austronesian Languages

It is a characteristic of the Austronesian languages of Formosa, the Philippines, Madagascar, and (at least) western Indonesia (afterwards Western Austronesian or W.An) that many verbs can head more than one transitive clause type [1]. The form of the verb will be different for each possibility, and there may be changes in the marking or position of the noun phrases in the clause, but the verb root will be identifiably the same. Examples from Tagalog, Bahasa Indonesia and Balinese follow [2]:

 

1

Tagalog

a	Kumain k-um-ain	ng	isda	ang	bata
	um-eat		fish		child
	'The child ate (*the) fish'

b	Kinain k-in-ain	ng	bata	ang	isda
	-in-eat		child		fish
	'A child ate the fish'

 

2	Indonesian

a	Ali	membaca	buku	itu
	PN	meN-read	book	the
	'Ali read the book'

b	Buku	itu	dibaca	(oleh)	Ali
	book	the	di-read	(by)	PN
	'The book was read by Ali'

c	Buku	itu	saya	baca
	book	the	1SG	read
	'I read the book' or 'The book, I read'

3

Balinese

a	Nyoman	lempag	tiang
	PN	hit	1SG
	'I hit Nyoman'

b	Tiang	ng-lempag	Nyoman
	1SG	N-hit	PN
	'I hit Nyoman'

 

In each of these languages, the function of distinguishing the two arguments of the transitive verb is divided between verbal morphology and noun phrase coding properties. In Tagalog, for example, the infix in the verb stem in example 1a indicates that the noun marked by ang is the higher thematic argument of the verb, while the infix in example 1b indicates that the noun marked by ang is the lower thematic argument of the verb [3]. The verb prefixes of Indonesian and Balinese operate in similar fashion, indicating the thematic status of the leftmost argument of the clause [4]. Where the clauses are directly comparable (i.e. excluding example 2c), propositional meaning is not altered by the different codings (modulo constraints on the referentiality of various functions to be discussed later). Various syntactic tests establish that the Tagalog nominal marked with ang, the leftmost nominal in Indonesian and the pre-verbal nominal in Balinese have (at least some) subject properties, and that the other nominal has similar syntactic privileges in both clause types (Schachter 1976, Kroeger 1993, Artawa and Blake 1997). These characteristics pose serious problems for formal theories of grammar; they have been explored within the GB tradition with less than satisfactory results (Guilfoyle, Hung and Travis 1992, Voskuil 1993). Here we discuss the problems in terms of a lexicalist model with parallel representations (i.e. LFG), pointing out both the advantages of using such a model and also the points of strain when the current model is confronted with these data.

 We will assume that the semantic relations associated with basic lexical entries of verbs are not different in kind to those assumed for other languages. Thus, the Balinese verb exemplified above will have at least the following information associated with it in the lexicon:

 

lempag < Agent, Patient >

 We will also assume that it is possible to identify some notion of syntactic subject in W.An languages. Many scholars from Schachter (1976) onwards have noted that not all of the properties taken as typical of subjects in European languages are always associated with the subject so identified in W.An languages. For example, while the subject in these languages is the only argument accessible to relativization, it is only the antecedent for reflexives when it is also the highest thematic argument. When the subject and the highest thematic argument are not identical, antecedence of reflexives is a prpoerty which remains associated with the highest thematic argument. We suggest that this division of properties can be explained by assuming a division between properties which are universally semantically based, as argued by Dixon (1979, 1994) and properties which are associated with a syntactic notion of subject identified on a language- and construction-specific basis. The correct identification of this subject on the basis of syntactic properties has been investigated in detail by Schachter (1976) and Kroeger (1993) for Tagalog, Artawa and Blake (1997) for Balinese, and by various scholars for other languages (e.g. Chung 1976 for Indonesian and Schachter 1984 for Toba Batak). The division of properties between two nominals in the clause is handled naturally in a model allowing parallel representations such as LFG. The semantic properties are associated with the most prominent argument in a(rgument)-structure, and the syntactic properties are associated with the most prominent argument in f-structure, the grammatical function SUBJECT. Such a division of labour has been argued for by Manning (1996a,b) and profitably employed in other analyses (e.g. Wechsler and Arka, to appear). This approach is more constrained than an analysis such as that of Guilfoyle et al (1992), which allows two subject positions to be filled simultaneously.

 The discussion above made it clear that noun phrase coding properties are not consistent within the group of W.An languages we are considering. Balinese is clearly a configurational languages with a VP constituent. In the Balinese examples above, the post-verbal NP must be adjacent to the verb while the pre-verbal NP can appear following the VP. Similar patterns have been described for Toba Batak (Schachter 1984) and Pendao (Quick 1997). Indonesian has some clear configurational effects; for example, unaffixed verbs (example 2c) must be immediately preceded by their Agents, and di- verb forms (example 2b) only licence a bare NP Agent if it immediately follows the verb. Tagalog, on the other hand is non-configurational according to Kroeger (1993), and certainly has more word order possibilities than the other languages. Here again, a model such as LFG offers clear advantages for analysis. The interesting comparisons to be made between these languages are those at the level of predicate-argument relations and grammatical relations, and the discussion is greatly simplified when theory internal considerations do not require the varied surface coding of grammatical relations to be accounted for by configurations which are identical at some level (see also Austin & Bresnan 1996).

 We have established that W.An languages allow two codings for many transitive clauses and that a different nominal can be identified as syntactic subject in each of these clause types. Having understood this much, a problem arises immediately: can the second argument in clauses such as those above be treated as an object? The theory-neutral descriptive statement in our introductory paragraph would imply that there should be no doubt on this point; the clauses are transitive and have two arguments. Nevertheless, there has been a reluctance on the part of previous investigators to take a clear position on this issue. In part, this can be attributed to the wish to analyse the pattern seen in the examples as a voice alternation of a familiar type and therefore to treat one clause type as having an object, and the other as intransitive. We believe that the evidence provides no basis for any analysis which treats such pairs as differing in transitivity; it is not possible to analyse these pairs of clauses as active and passive (or active and antipassive). However, there is one view of ergative/absolutive systems which claims that there are no true transitive clauses in them (Bresnan and Kanerva 1989 n.32, cf. Manning 1996: 39 and references cited there) [5]. Therefore, on this view both clause types might be intransitive, no difference in transitivity would be assumed but there would also be no objects.

 The morphological evidence suggests that the form with the thematically lower argument as subject (afterwards INV(erse)) is basic. This is immediately evident in Balinese and Indonesian, where the unaffixed verb appears in INV clauses (but see footnote 2) and is also suggested by more complex considerations for Tagalog (De Guzman 1992). It is also the case that in all the languages which we will consider here, except Indonesian, the INV form is more common in discourse than the form in which the highest thematic argument is subject (afterwards DIR(ect)) [6]. This has led many scholars to take the INV forms as basic and to analyse these languages as having ergative/absolutive systems (e.g. Artawa & Blake 1997, Blake 1988, Gerdts 1988, De Guzman 1988). However, in each of the languages considered here, non-term arguments are clearly distinguished by the presence of prepositions, or in Tagalog, the oblique marker sa, and by syntactic tests (see for example Kroeger 1993: 40-48). This forces us to conclude that both INV and DIR clauses are true transitive clauses, and to reject the intransitive ergative analysis.

 Some doubt about the syntactic status of the non-subject argument of transitive clauses nevertheless remains, particularly for the Philippine languages where subject choice is not restricted to the term arguments of the verb. To take examples from Tagalog, the verb bigay 'give' takes only two term arguments in DIR and INV forms:

4	Nagbigay -in-mag-bigay	ng	korbata	sa	lalake	ang	babae
	PERF-DIR-give	TRM	tie	OBL	man	SUBJ	woman
	'The woman gave a/the man a tie'

 

Cross-linguistically, if second objects are possible, we would expect them to be possible with this verb. Therefore, Tagalog appears not to allow second objects. But when the same verb appears in a form which identifies the Beneficiary as SUBJECT, it has two arguments (the term arguments) marked in the same way as the non-subject argument of a transitive clause (cf. examples 1a and 1b):

5	Binigyan b-in-igay-an	ng	babae	ng	korbata	ang	lalake
	-PERF-give-OS(B)	TRM	woman	TRM	tie	SUBJ	man
	'A woman gave the man a tie'

 

This example might be taken to show that Tagalog does allow double object constructions [7]. Double objects also seem to be possible in clauses in recent perfective aspect, where no nominal appears with the marker ang, but two NPs can be marked as non-subject terms [8]:

6	Kakakain ka-ka-RD-in	ng	leon	ng	tigre
	RP-eat	TRM	lion	TRM	tiger
	'The lion ate the tiger' (*The tiger ate the lion)

 

This clause is not ambiguous, and it might therefore be argued that there is a dissociation of grammatical functions and surface marking in such clauses, with the subject being exceptionally marked by ng (this is essentially the position taken by Guilfoyle et al.). Kroeger (1993:53-54) shows that extraction, otherwise a property of syntactic subjects in Tagalog, is always possible for the Agent in these constructions, and sometimes also for the Patient, suggesting that there may not be a unique subject in such constructions. The interpretation of such clauses appears to be determined by linear order. Both the constructions discussed in this paragraph suggest that the status of the putative grammatical function object in Tagalog (the ng marked nominal in a transitive clause) is in some doubt, and that it is best treated as a marker of termhood, not as a marker of a specific grammatical function [9].

The typical situation which prevails in W.An languages, then, is that transitive verbs have two alternative lexical forms with the same semantic relations and with both arguments available to be linked to GFs. The linking of semantic roles to GFs is inverse (to borrow a term from Manning 1996) for the more common form, and direct for the other form. No theory of linking which is based on hierarchies of semantic roles and hierarchies of GFs can account for these patterns without stipulating the linking in one case. Treating a(rgument)-structure as an independent level which is the locus of valence-changing operations, as argued for by Manning (1996), does not seem to be any help with this problem. Manning argues that the information represented at a-structure is firstly, the division between terms and obliques, and secondly thematic obliqueness within each of these subdivisions. Thematic obliqueness is lexical information which cannot be manipulated in this scheme, therefore any change in linking patterns must be attributed to changes in which arguments are terms and which are not, as in the classical analysis of passive. We have argued above that the inverse and direct transitive clauses of W.An do not have different valencies, and therefore the existence of a separate level of a-structure does not solve this problem. Note that genuine valence-changing operations are available in these languages: Balinese has passive constructions (Arka and Simpson, this workshop), Indonesian has causatives and applicatives, and Tagalog has causatives (Carrier-Duncan 1985). Perhaps most strikingly, Toba Batak (Schachter 1984) has two possibilities for the coding of verbs with three logical arguments, such as 'give', essentially the same as dative shift in English. The beneficiary can either be introduced by a preposition, or with the same surface form as the other two arguments of the verb. Note that this valence-changing operation is independent of the marking of the verb for DIR or INV clause type, both prefixes allow either expression of the Recipient [10]:

7a	Mangalean	biang	si	Torus	tu	si	Ria
	DIR-give	dog	PROP	PN	to	PROP	PN
	'Torus is giving a dog to Ria'

b	Mangalean	si	Ria	si	Torus	biang
	DIR-give	PROP	PN	PROP	PN	dog
	'Torus is giving Ria a dog'

c	Dilean	si	Torus	biang	i	tu	si	Ria
	INV-give	PROP	PN	dog	the	to	PROP	PN
	'Torus gave the dog to Ria'

d	Dilean	si	Torus	si	Ria	biang	i
	INV-give	PROP	PN	PROP	PN	dog	the
	'Torus gave the dog to Ria'

Such evidence thus suggests that a-structure and subject choice are independent factors in W.An languages.

This possibility is further supported by evidence form Tagalog regarding clauses with two arguments marked by ng. Even when no subject seems to be selected (example 6), the a-structure status of arguments is still apparent. The same is true where a non-term subject is selected, as is possible in Tagalog, Malagasy and the Formosan languages. In all such cases in Tagalog, terms are marked by ng and their syntactic status is confirmed by the impossibility of fronting them (Kroeger 1993: 43-46). For example, a circumstantial oblique such as the location in the following clause can become a subject in these languages:

8a	Binili b-in-ili-Ø	ng	lalake	ang	isda	sa	tindahan
	-PERF-buy-INV	TRM	man	SUBJ	fish	OBL	store
	'The man bought the fish at the store'

b	Binilhan b-in-ili-an	ng	lalake	ng	isda	ang	tindahan
	-PERF-buy-OS(L)	TRM	man	TRM	fish	SUBJ	store
	'The man bought fish at the store'

 

As before, Agent and Theme both remain terms when another subject is selected. Assuming, as would seem plausible, that semantic relations such as Location are not part of the argument list in a verb's lexical entry, these considerations lead to a view of Philippine languages in which the a-structure projected by the lexical entries of verb roots is not the level at which subject selection is accomplished. It is the affixes which attach to the verb roots which complete the specification of syntactic relations. Further support for such a view comes from constraints on subject choice in W.An languages. Across these languages, such constraints operate and always depend on the referential nature of the NPs filling argument roles. Thus, Artawa and Blake note that the Balinese nasal construction is used regularly where the Patient is nonspecific or indefinite, and in imperatives the choice is obligatory (1997:488-9); Schachter cites work showing that in Toba Batak the primary determinant of verb type is the individuation of the Patient (1984:124); and where the Patient is definite in Tagalog, INV forms are effectively obligatory in free clauses [11],[12]. Valence-changing operations, which is to say a-structure operations, are not normally sensitive to such factors, which it would be more plausible to include in semantic representations.

The contributions of Foley and Kroeger to this workshop explore this view , that verb affixes are the locus of grammatical relations in W.An, in some detail and we would suggest that GB work such as that of Guilfoyle et al (1992) and Voskuil (1993) can be seen as similar in spirit. In those studies, verb affixes are interpreted as having specific case-assigning or case-absorbing properties and the NP movement then required to satisfy the Case Filter leads to the syntactic relations observed in surface strings. This is therefore an analogous treatment using the apparatus available within the GB theory. A similar view has also been advoicated by Sells (1995). However, the assignment of grammatical relations in W.An languages does share some characteristics with more familiar systems. Several languages restrict subject choices to terms (Balinese, Toba Batak, Indonesian), and the languages which allow other possibilities are still sensitive to the a-structure status of arguments. In Tagalog, the infix –um-, indicates that the highest thematic argument is subject, but it is semantically unrestricted. The dicussion of Carrier-Duncan (1985) makes this clear, giving examples of verbs such as b-um-uti ‘get better’, where the subject is certainly not an Agent. Similarly, the suffixes –in and –an indicate that the lower thematic term is subject. In this case, there is greater semantic specificity, but there is still some flexibility. But when non-terms are selected as subject, the verbal morphology is semantically specific; for example the prefix i- always indicates that a beneficiary is subject, and the suffix –an is always used for locational subjects. Another example of effects of this type can be seen in the Toba Batak dative shift construction (example 8). For the DIR verb form, the Agent is subject in both cases (example 8a and 8b), but the next highest non-oblique is included in the VP. So in example 8a, the Theme is adjacent to the verb, but in example 8b it is the non-oblique recipient. With the INV verb form, the second highest available thematic argument becomes subject; thus the Theme is subject in example 8c, and the Recipient in example 8d. The next available syntactic position within the VP is then occupied by the highest available argument , the Agent, in each case. The descriptive generalisation which Schachter makes on the basis of these examples is:

9	Interchange the Dative and the Patient (1984: ex.38)

 

and this indeed captures the a-structure effect of the change. When the Recipient is non-oblique, it is available for linking before the Theme. Toba Batak and Tagalog have alternate strategies for meeting functional needs here. Although Toba Batak allows only two verb forms, note that in the examples above it is possible for each of the three arguments to be subject, thus making it possible to relativize on each of the arguments [13].

The evidence discussed here leads to the view that subject selection in W.An languages is at least partially independent of the normal process of linking arguments to grammatical functions. When the effects of that independent process are discounted, however, it would seem that the linking principles used in these languages are very similar to those investigated in other languages. The two principles may even interact at times, as in the Toba Batak dative shift construction. Another interaction can be seen in the case of the Balinese passive. Passive is only compatible with the unaffixed (or zero-affixed) verb form and this is a predictable result on the account presented here. Passivization is an a-structure operation which makes the highest thematic argument unavailable for linking to a grammatical function. With only one term available for subject selection, only one verb form should be possible, and as the higher thematic argument has been made unavailable, the verb form which selects the lower thematic argument as subject (unaffixed / zero-affixed) is the only possibility.

The discussion above has, we hope, demonstrated that the issues to be faced in analysing W.An languages can be seen very clearly in a framework such as LFG. The correct answers do not fall out from the framework automatically, but it becomes clear that the solutions must be sought in the linkage between arguments in lexical entries and GF assignments in f-structure and the role that verb affixes play in mediating this linkage. Surface coding features of GFs are not relevant to this problem. Parallel representations also allow a simple account of the split in subject properties typical of these languages without resorting to devices such as assuming the existence of two subject positions in a single level of representation (Guilfoyle et al 1992). Serious questions remain to be answered however, which we hope will be explored in this forum:

• is it meaningful to speak of a voice alternation between two clauses which are similar in their valence? We would suggest that it is not, and this clearly has implications for other areas of syntax which LFG handles at f-structure. Arka and Simpson’s paper investigates this issue in relation to control in Balinese.
• if such possibilities exist in a language, does it make sense to treat one as basic and assign the system a place in the accepted typology of grammatical systems, or does it make more sense to augment the typology and argue for a distinct type of system? Foley’s paper is an extended argument for the second of these alternatives.
• are the possible clause types in W.An languages genuinely symmetrical, or are their markedness relations between them? This question can be posed at the level of morpohology, and is addressed briefly by Kroeger, and at the level of frequency of usage.
• are the additional subject choice possibilities of Malagasy and the Formosan and Philippine languages make these languages typologically distinct from languages such as Balinese and Toba Batak? Are the various oblique subject possibilities sub-types of the INV mapping, or are they separate?
• what sort of principles are needed to explain how GFs are assigned in these languages, and what do they tell us about the universal properties of language? Again, this is an issue which Foley nd Kroeger will explore in detail.
• why are there not other languages in the world which are organised along the same principles? Dukes’ paper raises some of the issues we have discussed here in an analysis of Tongan, a language from the Eastern (Oceanic) branch of Austronesian, and Foley suggests that some Native American languages are typologically similar to W.An.
• what sort of information is stored in lexical entries in these languages, and more specifically, are root words precategorial? Foley argues that this is the case and takes this property to be criterial for the existence of such grammatical systems.

Created: 8 June 1998
Last modified: 29 June 1998

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

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