The University of Queensland, Brisbane

Mirriam Butt and Tracy Holloway King (Editors)

1998

CSLI Publications

http://www-csli.stanford.edu/publications

This paper deals with the voice system of Indonesian, and argues that certain of the constructions traditionally analysed as passives, should be given a different treatment, parallel to arguments by Kroeger (1993) for Tagalog. We examine the role of different conceptions of subject and their place in binding. We show that, unlike other Western Austronesian languages, the logical subject – l-subject for short – (i.e., the semantically most prominent argument) plays little role in binding: being a logical-subject alone does not make an argument a binder. Syntactic prominence is crucial, and in particular the data on binding in Indonesian presented here further confirms the notion of syntacticised argument structured (a-str) first proposed in Manning (1994, 1996b) and also adopted in Arka (1998) wherein a central role is given to the notion of a-subject. Like other Austronesian languages, the (surface) grammatical subject (i.e., the SUBJ in the f-structure or gr-subject for short) plays little role, especially in the binding of morphologically complex reflexives. The data from binding is supported by other syntactic tests such as topicalisation with pronominal copy.

Indonesian transitive verbs can appear prefixed with meN- or di- or without a prefix. There is evidence that the Agent/l-subject Amir appearing with meN- (henceforth Agentive voice or AV) verbs in Indonesian as in (1) is syntactically the surface grammatical subject.

Among the important properties of the gr-subject in Indonesian are: (a) appears canonically in a preverbal position, (b) the only function that can be questioned by a clefted question word, relativised on or clefted, and (c) the only function that can be controlled, either as an equi-target of certain verbs or as the gapped function in controlled adverbial clauses. An additional test of a morphosyntactic character is that the 3sg pronoun can optionally be just ia rather than the usual dia when it is functioning as the gr-subject of a clause (adding adverbs etc. shows that this form is indeed grammatically not phonologically conditioned).

Thus, Amir in (1a) is the gr-subject because it comes preverbally, it can be relativised in a cleft sentence (to give a slightly different pragmatic implication):

It can be an equi-target:

It can be replaced by ia:

It is also widely agreed that the Agent of di- verbs expressed by a PP, as in (2), is an Oblique, while the Theme has grammatical subject properties. The grammatical relations in (2) thus mirror an English passive, and one might presume that di- is a passive marker, and we gloss it that way in (2). But actually the situation is a little more complicated, as we discuss below.

The situation is less clear in other constructions where the l-subject is not the gr-subject, namely when it is expressed by the pronominals saya/kamu/dia or the clitics ku-/kau-/-nya, in the sentences shown in (3). All these sentences have the Undergoer as gr-subject, and the verbs lack the meN- (i.e., the AV) marker, being either bare, or prefixed with di-. Many studies in Indonesian syntax are unclear as to quite what syntactic status to give such sentences. As suggested by the glosses, such sentences are normally appropriately translated into English with active sentences, but syntactically they have been analysed as passives by previous studies , apparently due to the clear property that the Undergoer is the gr-subject and comes sentence initially.

Kana (1986) explicitly claims that the l-subject or the initial subject (i.e., the initial-1 in RG terminology) in a sentence of the type in (3) is a final 1-Chomeur (i.e. a non-core argument, basically an Oblique). However, in what follows we show evidence that the pronominal immediately preceding the verb (3a), the proclitic (3b), and enclitic (3c) are still Term/core arguments. The evidence is mainly from binding, with some supporting evidence from a pronominal copy test, control, and discourse properties. Hence, in our view, the passive analysis for (3) is untenable.

Unlike -nya '3' (3c) (see also 2.3.3), the other enclitics -ku '1', -mu '2' and -kau '2' cannot be understood as l-subjects (4a), but only as undergoers (4b):

LFG has proposed a model of parallel representations, and in general prominence can be defined on any level. Accounts such as Dalrymple and T. Mohanan have made use of this to propose that some parts of binding theory may be sensitive to one level, and other parts to another level. In contrast, Manning has argued that the principal constraints of binding theory can be defined on a level of syntacticised a-str, while admitting that some anaphors may require additional constraints, such as also requiring the binder to be a gr-subject. Within this theory, term arguments outrank obliques in a-str, and within each of those groupings, prominence is based on thematic or Lexical Conceptual Structure prominence (following Hellan 1988).

In the AV constructions marked by meN-, the l-subject, a-subject and gr-subject are identical. For example, the relativization test shows that the agent saya in (5) is the gr-subject (5b). By way of contrast, the reflexive object cannot be relativised (5c). It is also the a-subject, a-commanding the reflexive theme diri saya (i.e. the object) in (5a), which we assume to have an argument structure as in (5d) – where the vertical bar is used to separate core or term arguments from obliques. By way of contrast, an attempt to make the gr-subject an anaphor fails as is shown in (5e).

In short, the l-subject/Agent in the AV construction is an a-subject (and also a gr-subject). Binding in (5) is straightforward and exactly as one would expect from well-known accusative type languages. The data thus far does not serve to isolate any particular analysis.

Binding properties show that non AV verbs cannot be lumped together as a homogenous class, traditionally simply called passives. In what follows, we discuss a variety of non AV verbs and examine the syntactic status of their l-subjects based on evidence from reflexive binding.

As shown in (6), an l-subject appearing as an Oblique PP cannot bind a reflexive functioning as a gr-subject. This is consistent with a passive analysis of this construction, with the l-subject being an oblique.

The same is true for pronominal agents which can appear either as an enclitic to the preposition oleh-nya or as a prepositional object oleh dia. They cannot bind a reflexive gr-subject.as shown by the contrast in (7):

The failure of binding in (6)-(7) shows that semantic binding does not apply in Indonesian. It is not the case that there is 'semantic' binding and all l-subjects are possible binders and can bind thematically lower arguments within their clause. Rather, it seems to be the case that, although the passive agent is an l-subject, the crucial fact is that it does not a-command the reflexive, since the reflexive gr-subject is higher in the a-str, because it is promoted in the passive .

It is not that the passive agent is inert with respect to the binding theory, however. Examine the following sentences with a three place predicate 'ask':

As expected according to the theory of Manning , the Agent oblique can bind other oblique arguments, such as the oblique theme in (8b), because it a-commands such arguments. The argument structure of (8b) would be as in (8c):

Di- verbs cannot appear with a non-third-person Agent:

Backgrounding of non-third persons is not possible with the di- passive, but it is possible with an otherwise similar construction: the ter- verb prefix. The prefix ter- has various functions such as expressing a sense of ability or possibility, which generally appears in negative sentences, as in (10a), or an accidental event with a non-volitional doer as in (10b).

In all these cases the Agent is backgrounded and can be expressed in a PP. In these constructions, the Agent again appears to be an oblique, as is shown by the inability to form (11b):

The l-subject of the type exemplified in (3a-b) can be of any person, but it must be a pronominal. A common noun cannot appear in this construction (unless it is used vocatively):

There are two forms for 1sg and 2sg, and the orthography writes the shorter, perhaps reduced ones as attached clitics, but all of them must appear immediately preceding the verb. (We are unsure at this point whether there is good phonological evidence for regarding any of them as phonologically attached.) Nothing can intervene in between: Sentence (13b) is bad because the auxiliary akan intervenes; (13c) is bad because an adverb intervenes. This suggests these words occupy a position at the left edge of the VP reserved for pronouns or pronominal clitics.

If the bare form of the verb is used, as in these examples, then the pronominal form cannot be omitted:

The fact that the Agent must be present might be an indication that it is a Term, rather than an Oblique. Note that sentence (14) is not acceptable in any interpretation, e.g., it cannot be interpreted as having a first or second person l-subject.

Also, there is good evidence that the sentence-initial NP in these examples is the gr-subject, rather than just some kind of preposed topic. For instance, it is the NP that is the equi-target (Chung 1976b):

Evidence from binding further shows that the Agent pronominal in this construction has a very different status to a passive agent. Indeed, we argue that it is really a Term, hence an a-subject. In the following sentences, the reflexive gr-subjects can be bound by the preverbal pronominals (16). Evidence that the reflexives are gr-subjects comes from their appearance in the canonical subject position and the possibility of cleft formation by yang (17). Attempts to cleft a non-subject reflexive (i.e. by making the verbs appear in AV) fail (18)-(19).

Crucially, this binding behaviour differs from that of the oblique agent appearing in a PP headed by oleh, of the type that was shown in (6). This suggests that the syntactic status of the l-subject appearing as a preverbal pronominal in a non AV verb exemplified in (16)-(17) differs from that of an l-subject appearing in the PP with the di- verb in (6). The di-verb with the PP agent is a passive construction with the Agent/l-subject being an oblique. The verb without meN- with a preverbal pronominal is not a passive verb. The l-subject is a Term, hence an a-subject. This corresponds to the idea that the sentence feels semantically 'active' (usually translated as an active), despite the fact that the non-agent argument is the surface gr-subject.

A construction with a cross mapping where an Agent a-subject is not a gr-subject, but still a term, and the gr-subject is a non-Agent core argument is an ergative construction . Following the terminology for Tagalog from Kroeger and Balinese by Arka and Wechsler and Arka , the Indonesian verbs with cross-mapping exemplified in (16)-(17) can be labelled as Objective Voice (OV) verbs. But this 'voice' should really be interpreted as an ergative construction within the language. Given the pervasive evidence from binding cross-linguistically , it is misleading to collapse OV/ergative constructions with passives, or indeed any of the traditional 'voices'.

Additional evidence for a preverbal pronoun being a term comes from control of complex arguments. It has been observed that the functional controller of a complex argument is restricted to a term (Bresnan 1982, Arka and Simpson 1998). For example, the sentence *To go there was asked of John by me is unacceptable because we cannot express the controller (of John) as a term argument, as an NP. Like Balinese (Arka and Simpson 1998), Indonesian shows a possible control into a complex argument acting as gr-subject, and crucially the controller must be a term:

Coba 'try' semantically has two arguments: a trier (a simple argument) and the thing tried (a proposition, a complex argument). It is a commitment type of verb, characterised by having a committer (i.e. the trier) as a controller. (20a) shows the AV construction with the controller as gr-subject (acceptable), (20b) shows the OV construction with the controller as a preverbal pronoun (acceptable), (20c) shows the controller as a non-term (oblique) and, crucially, the sentence is then unacceptable. This test again shows the preverbal pronoun grouping with other terms as opposed to obliques.

A little further evidence for the pronouns before the verb being term arguments can be derived from examining the construction where an NP becomes an external topic at the left margin of the clause, and then is repeated by a pronoun within the clause. This is possible when the pronoun is a term argument, as in (21a), but it is not possible with clear obliques such as the objects of prepositions, see (21b):

Note now that topicalization with pronominal copy is possible with the pronominal arguments that precede the verb, supporting our regarding them as term arguments:

The enclitic -nya attached to a head verb always expresses a core argument that is not the gr-subject (what we might term an OBJ or a term-complement). It can express an l-subject/agent as in (3c) or a patient functioning as an Object as in (23a). -nya cannot be the gr-subject (23b). That is, the structure in (23b) is forced to be an OV construction by dropping meN- making the preverbal pronominal agent dia a non gr-subject. We attempt to force the enlitic patient -nya to act as the gr-subject instead. It fails. In other words, although both the agent and patient arguments of the transitive verb are present in sentence (23b), the sentence is bad because it lacks a gr-subject; neither argument can act as the gr-subject. Note that a normal pronominal gr-subject can come post verbally (23c). The point is that -nya can appear attached to the verb only when another argument is the gr-subject: in the AV verb (marked by meN-, as in (23a)) where -nya is the undergoer, or else in the di- verb as in (3c) where -nya is the actor.

The enclitic -nya can also appear attached to the preposition expressing an Oblique agent (24a). As noted previously, the pronominal dia is also possible. These forms must again appear with a di- verb, hence the unacceptability of (24b). The enclitic -nya cannot be doubled with the appearance of the preverbal pronominal dia (24c-d). (Thus, the contrast between (24a) and (24c-d) suggests that di- is not really a pronominal, pace Kana (1986) who suggests that di- is a shortened form of dia.)

This behaviour contrasts strongly with the binding behaviour of the enclitic -nya attached to the head verb. Consider:

It can be concluded that the third person agent appearing in PP is an Oblique, whereas the pronominal clitic hosted by the head verb is not, but rather a term complement in an ergative construction. It is still a term and an a-subject and so can bind the reflexive gr-subject.. This perhaps in part motivates its interesting discourse function briefly mentioned below.

The binding evidence supporting regarding -nya as a term in a transitive clause is again backed up by evidence for the possibility of topicalization with a pronominal copy, which as we have seen is only possible with term arguments:

There is one final complication in the discussion of verbs with a di- prefix. Until now, we have shown examples with the agent expressed within a PP. But, somewhat surprisingly, di- verbs can also take a postverbal NP agent as in (27). Indeed, when the NP is indefinite as in (27a), the Agent NP is preferred to the PP. Some accounts suggest that this is possible because the preposition is in some sense optional, but this does not seem to be correct as a postverbal Agent NP is only possible when it is adjacent to the verb (Myhill 1988). Hence the acceptability contrast in (28a-d). This suggests that the agent NP occupies a different phrase structure position to the agent PP.

The question is what is the status of the postverbal NP agent. The fact that it occurs without a preposition suggests that it is a term argument. On the other hand, it cannot bind the gr-subject reflexive:

Furthermore, it cannot bind the theme object:

If the Agent NP ayah is a term, it should be a possible binder for dirinya because it is thematically the most prominent item. This suggests that it should be regarded as an oblique.

Further evidence comes from (possessor) topicalisation with a pronominal copy. Consider the possessor topicalisation of the subject (30a), of the object (30b) and of the postverbal agent NP (30c). Only the first two are acceptable. (Sentence (30d) shows the non topicalised version of (30c).)

Myhill (1988) in fact argues that the agent noun is here incorporated. It is unclear to us whether we would want to say that – Myhill is basing this analysis on the loose definition of incorporation from Mithun (1984:849) which covers cases where 'a verb and its direct object are simply juxtaposed to form an especially tight bond. The verb and noun remain separate words phonologically, but Ö the N loses its syntactic status as an argument of the sentence, and the VN unit functions as an intransitive predicate.' But it is interesting to note that in this case there is evidence of the agent noun preceding enclitic particles which are semantically modifying the verb. For example, Myhill gives the example in (31), where the particle -lah is giving emphasis to the temporal sequencing of the verb, and not to the agent noun.

On the other hand, this construction is definitely not the canonical case of noun incorporation widely discussed in the syntactic literature, since, as Myhill discusses, multiword agent NPs can appear in this construction. At any rate, all the available evidence suggests that the postverbal NP agent is not a term but an oblique, and so we will analyse it thus.

To summarize the discussion so far, binding suggests that an Agent/l-subject can have the syntactic expressions shown in Table 1. Given the a-str based binding theory, only the l-subject in AV (a), OV (b), and di-verb-nya (c.i) is a possible binder of term arguments within the same clause in Indonesian.

Table 1.

Note now that the presence of di- in conjunction with -nya (i.e., c.i. in Table 1, example (25)) argues that di- is not really a passive marker, because clauses with di- and -nya represent an ergative construction, which is still transitive. Rather, di- seems to be best analysed as simply encoding the mapping of an Undergoer Term to SUBJ (which is only part of what a passive marker does). Di- leaves the status of the l-subject unspecified, allowing other specifications such as the information structure to determine the exact syntactic expression of the l-subject. In the examples we have seen, the l-subject can be expressed in any of five ways: (i) as a preverbal pronoun (ii) as a pronominal prefix, (iii) as -nya as a suffix to the verb, (iv) as a prepositional phrase headed by oleh (and involving either a noun/pronoun or -nya again), or (v) as a postverbal oblique NP.

While there are various other possibilities, such as gr-subject postposing, it seems that the basic phrase structure that we have to work with is the following:

All the verbal clitic positions, including the preceding full pronouns, must be immediately adjacent to the verb and are reserved for words with pronominal meaning that express the OBJECT/Term-complement of the clause. These are used when the verb remains transitive. As shown by (32), the preverbal positions are positions for agent term complements only. When these positions are occupied, the clause is in the objective voice. The post verbal clitic position is not restricted to an agent. We have observed that -nya appearing in this position can be an agent (example (3c)) or a non agent (example (23a)). However, we note that -nya is somewhat exceptional in this regard, since the other enclitics that appear in this slot only express an undergoer term complement. When the verb has been passivized, these slots cannot be used, but the agent can be realized as an oblique, either as an NP oblique, which again must be adjacent to the verb, or as a PP oblique, which need not be.

These informal remarks raise some questions about how to treat and constrain the alternation between a syntactically active and ergative transitive construction and the passive construction. The most revealing approach to us appears to be to say that various morphemes serve merely to place constraints on the mapping between argument structure and grammatical relations. Using the terms Actor and Undergoer as convenient if informal shortcuts for the first two term arguments in the argument structure, and a as a correspondence function picking out the argument structure, we could then suggest the following constraints:

These constraints, together with a constraint on mapping to the effect that there must be a gr-subject (Bresnan and Kanerva 1989) are sufficient to ensure that only the observed patterns of linking actually occur. The constraint (33f) is somewhat unsatisfactory but reflects that Indonesian does not allow the free appearance of NP Actor Term-complements. In this respect it is like English, and unlike, say, Balinese (Arka 1988). However, Actor Term-complements can be expressed by the various pronouns and verbal clitics. To work through some of the possibilities, in turn:

1. If meN- is prefixed to the verb, then the Actor must be the Subject. The other term in a transitive argument structure must become the object and can be expressed either as an object NP or via a enclitic suffix on the verb.
2. If di- is prefixed to the verb, then the Subject is the Undergoer. This could be either because the verb is passivized or because the ergative construction is being used.
1. If the verb is passivized, then the optional agentive oblique can be expressed either as a PP headed by oleh or in the immediately postverbal position for realization of agentive obliques that we discussed in 2.3.6.
2. If the verb is not passivized, then the actor remains a term argument, and must be expressed in the sentence. Since the pre-verbal slot for expression is already taken, and an NP inside the VP and most of the enclitics cannot express an Actor, the only possibility is when the agent is realized by -nya.
3. If saya/kamu/dia immediately precedes the verb or ku-/kau- appear as proclitics on the verb, then they express the agent, but as a object/term complement. Therefore, the Undergoer must fill the subject slot. However, these are pronominal clitics, and therefore they cannot co-occur with another expression of the agent, such as a PP headed by oleh.

Before closing, we will briefly touch on one of the interesting consequences of this analysis for a theory of information structuring, and in particular how it challenges even the simple theory of information structure that is commonly accepted in LFG. This section is largely based on material from McCune (1979). As has already been noted, ergative sentences with -nya are naturally translated with actives in English. For example, consider the following text:

The verbs in italics in the free translation are ergative clauses with -nya in the original. Note firstly that a passive translation of either of these sentences is implausible. But then note further that the discourse structure here thus goes against what is commonly assumed. Bresnan (1995) suggests that the grammatical subject is universally optionally identified as the default topic of the clause. But in Indonesian narratives, of which this one is quite typical, the subject is not used as a default topic. Rather, after the first sentence, the topic of this excerpt is young Mr Cockatoo, and he is consistently referred to by the term-complement enclitic -nya. The subject actually expresses new information, a pattern that is common in Indonesian (and Balinese). In both cases the verb appears before the subject. This option is generally available in Indonesian, and taking it here fits with the general tendency for new information to appear later in the sentence.

It is somewhat unclear whether to view this alternative as subject postposing or verb preposing, but we are tempted to analyse it as the latter because the verb receives some kind of pragmatic prominence in such sentences. On such an analysis, we might propose the structure in (35) for a simplified version of the last sentence in (34):

At any rate, this use of the subject position to express new information, which is not the theme of the narrative challenges most existing theories of information structuring, including that of LFG. Recognising that these sentences with -nya are not passives seems part of the solution of the problem, in that we would expect a term argument to have greater discourse prominence than an oblique agent, but clearly much more work in this area needs to be done.

We have shown that the Indonesian 'passive' should be divided between constructions that are genuinely comparable to an English passive, and ergative constructions that are not. For both these kinds of constructions, we have found strong support for an a-str based theory of binding. In the ergative construction, we find that term-complement a-subjects can bind gr-subjects, as for Toba Batak and Balinese (Arka 1998). Within the passive constructions, although the oblique agent cannot bind any of the term arguments, it remains an a-subject and can bind other obliques. The Indonesian data presented here thus provide further congruent evidence in support of an a-str based theory of binding and mixed mappings between argument structure and grammatical relations in Western Austronesian languages. Finally, the ergative analysis of clauses with di-V-nya verbs sheds some light on their use in narrative texts, but raises new challenges for information packaging.