Philippine subjects in a monostratal framework

PIV

The other "subject" function, which, following Dixon, we will call PIV(OT), is harder to pin down. Schachter associates it with "reference-related properties," such as definiteness, relativizability, and the ability to float quantifiers. Following traditional Philippinist usage, he calls it the topic, but makes it clear that this is a different use of the term from the usual discourse notion. Cooreman, Fox & Givón (1988) show that the PIV in Tagalog does not have the discourse-continuity properties (measured in terms of referential distance and topical persistence) that a discourse topic would exhibit.

More useful is Dixon's (1979: 101) characterization of PIV as "the pivot for operations of coördination and subordination." The PIV is involved primarily in subordination and coordination contexts, and relates more specifically to coreferentiality of arguments in different clauses. That is to say, it provides continuity between different clauses within the same sentence. In the words of Foley & Van Valin (1984: 128-9), "[the] most notable function [of pivots] is in cross-clausal syntax, where they are important as the controllers and targets of NP ellipsis." In this respect, it is similar to the discourse topic, which provides continuity between sentences in the discourse. Unlike £ F, PIV is not an a-function, and thus is not part of the a-structure of the predicate. It formally has the status of a d(iscourse)-function, and by LFG's Extended Coherence Condition (Bresnan ms) it must be defined as identical to some a-function. In some languages (those often called syntactically nominative-accusative) PIV is identified with ¤ F, while in others (syntactically ergative) it is identified with OBJ (in a transitive clause). In Philippine-type languages, a third approach is in evidence. The identity of the pivot is not inherently determined by the syntax of the language; instead, it is determined by the morphological shape of the verb; specifically, by the so-called "voice" affixes. Within a constraint-based lexicalist framework like LFG, these affixes would be associated with the appropriate specifications for unification of the features of the PIV function with some a-function. In the LFG formalism, this is achieved very naturally through functional equations: 2 1. "Active voice": ( "Indirect object/locative voice": ( Note the following examples (Kroeger 1993 (3.13)) and their f-structures. 3 2. a. B-um-ili ang lalake ng isda sa tindahan. PERF.ACT-buy NOM man ACC fish DAT store 'The man bought fish at the store.' b.

3. a. B-in-ili-0 ng lalake ang isda sa tindahan. PERF-buy-DO ERG man NOM fish DAT store 'The man bought the fish at the store.' b.

4. a. B-in-ilh-an ng lalake ng isda ang tindahan.

PERF-buy-IO ERG man ACC fish NOM store 'The man bought fish at the store.' b.

Predictions

Our descriptions of the grammatical functions F and PIV provide a basis for us to predict which subject function will be implicated in various constructions.

Anaphora

One construction that is sensitive to subjecthood is anaphoric binding. According to Jackendoff (1990Jackendoff ( , 1992, anaphoric binding is the syntactic expression of argument binding in Lexical Conceptual Structures. If Jackendoff is right, we would expect that binding in the syntax would be expressed in terms of syntactic concepts related to Conceptual Structure: a-functions, specifically F. That binding is sensitive to a-functions is a well-established fact. In many languages, reflexives must be bound by "subjects." In Philippine-type languages, as noted by Schachter and others, it is the "actor" (i.e. F) that antecedes reflexives. 5. (Schachter 1976 (21- The syntactic side of imperatives is the lexical assignment by the imperative verb of addressee features (such as second person) to one of its arguments. This is done through a-functions, and in particular the a-function generally assigned to Agents.

(

This also falls out naturally under the present theory. (There is also clearly semantic motivation for Agents to be imperative addressees.)

The controller in either kind of control is determined semantically, as shown by the work of Jackendoff (1990), Sag & Pollard (1991), and others. The present theory suggests, but does not predict, that the functional controller should be a PIV. It suggests it because control is related to interclausal continuity, which is what the PIV function exists for. It does not predict it, however, because an equation associated with a control verb can reference ( 9 @ F) (or ( A OBJ)) without violating whatever locality principle there might be on a-functions. One might expect, then, that languages could differ on this: some languages (as in Bresnan 1982) specifying the controller as some core function (@ F, OBJ, OBJB ) and others specifying it as PIV. In anaphoric control, the controller has no special syntactic status, so there is no reason to expect any limitation on what function it can have.

The choice of controllee is more limited than the choice of controller. Here, too, semantics plays a role, in that the controllee must be the appropriate kind of participant (generally one able to control the action of the subordinate clause). However, there are also syntactic constraints. In anaphoric control, the controllee argument must be lexically specified by the verb of which it is an argument as a null pronoun with whatever features are necessary for control. 4

14. ( C AF PRED) = 'PRO' Such lexical specification follows the relational hierarchy of a-functions, so if it is limited to one function that one will be D F.

The controllee in functional control is a completely different matter. With Raising, there are no semantic restrictions on the controllee, since there is no semantic relationship involved. (For Equi constructions, semantics does still play a role, of course.) However, the controllee is specified syntactically not by the verb of which it is an argument, but rather by the higher verb. The requirement that a-functions only be specified by the lexical entries of the heads of which they are arguments prohibits the higher verb from designating an a-function of its complement. It can either indicate an unspecified GF or specify PIV. That is to say, choice of functional controllee should either be free or limited to PIV. We summarize our syntactic predictions below.

15. Anaphoric control: controller-no syntactic restriction controllee-possibly restricted to D F Functional control: controller-core function or PIV controllee-unlimited or PIV In Tagalog, Kroeger (1993) distinguishes between anaphoric and functional control, both of which he claims exist in the language. In the anaphoric control construction, the controller is determined by semantics and the controllee is generally D F, regardless of PIV status. 5 16. (Kroeger 4.39) Nag-atubili si Maria =ng hiram-in ang pera. PERF.ACT-hesitate NOM Maria COMP borrow-DO NOM money 'Maria hesitated to borrow the money.'

In functional control constructions (including Raising), both controller and controllee must be PIV. Tagalog thus conforms to our predictions.

Chaining

In coordinate structures in many languages, one constituent can be shared among all the conjuncts. This shared constituent can be called the chained constituent, and (following Dixon 1979) the construction can be called Chaining, or Topic Chaining. From the perspective of the present theory, Chaining involves continuity between clauses; it is not clause-internal or a-structure related. Therefore, if it is GFbased, it should be sensitive to PIVhood, not $ Fhood. This seems to be correct, based on the few languages for which there is evidence. The best known case is the ergative language Dyirbal. Kroeger (1993) shows that the same pattern holds in Tagalog

Long distance dependencies

As first observed by Keenan & Comrie (1977), relativization and other long distance dependency (wh movement) constructions are often limited to subjects. The framework developed here, combined with the LFG formalization of wh movement constructions in terms of functional uncertainty, predicts that in a language with such a restriction the relevant notion of subject is PIV. Since PIV is the function of interclausal continuity, dependencies that can extend over several clauses should involve PIV. If, as is plausible, a-functions can only be referenced in the lexical entries of the heads of which they are arguments, a functional uncertainty equation of the form (11a) would be allowed but one of the form (11b) would be disallowed.

a. (

The evidence from languages in which PIV and & F do not coincide is that this prediction is correct. In Philippine languages, only PIV can be extracted.

Control constructions

We turn now to a more difficult family of constructions: those involving control. Control, in its broadest sense, refers to constructions in which an element (usually the subject) of a (usually nonfinite) subordinate clause is unexpressed, and is interpreted either as being identical to an element of the main clause or as arbitrary (generic). The main clause element can be called the controller and the unexpressed position in the lower clause the controllee.

The two major kinds of control constructions in LFG are functional control and anaphoric control (Bresnan 1982), formally very different constructions. In functional control, the controller and controllee have the same value. This kind of control is a lexical property of the governing verb, which has the following equation in its lexical entry (with some core function specified for GF). In anaphoric control, on the other hand, the controllee is an unexpressed pronoun and the control relation is one of anaphoric binding. The lexicon of the language allows an empty pronominal element to be the value of some argument, and an anaphoric link is established.

Discourse prominence

In some languages, it has been shown that the PIV has special discourse prominence, as discussed by Foley & Van Valin (1984) and Manning (1996). This prominence is difficult to pin down, but as Manning shows for Inuit it relates somehow to definiteness, specificity, and/or wide scope. For Tagalog, it has generally been identified as definiteness. Since PIV is a d-function, it is to be expected that being PIV would have discourse related consequences.

We also suggest, with a little trepidation, that the ability to launch floating quantifiers may be a discourse related property. In the Philippine languages, this is a property that is unique to the PIV. However, there are other languages where the ability to float quantifiers seems to be subject to the relational hierarchy of a-functions.

Comparison with structural approaches

Approaches very similar to the one argued for here, but in a purely c-structural framework, have been proposed by Guilfoyle, Hung & Travis (1992) and Bittner & Hale (1996). We will very briefly consider the Guilfoyle et al. analysis here.

Based on data from Malagasy, Tagalog, Cebuano, and Malay/Indonesian, Guilfoyle et al. argue for a GB analysis in which there are two "subject" positions: SpecIP and SpecVP. Building on the VP-internal Subject Hypothesis, they suggest that in some languages both positions can be filled at S-structure.

Given the assumptions of GB theory and the VP-internal Subject Hypothesis, SpecVP is a position while SpecIP is not. Thus, SpecVP is filled at D-structure (by the Agent argument of the verb), while SpecIP is empty at D-structure and gets filled by the movement of an independently present nominal. From the perspective of the approach developed here, SpecIP is Guilfoyle et al.'s structural equivalent of the function PIV, while SpecVP is F F. Strangely, they refer to both positions as subject positions and explicitly refer to the SpecIP (i.e. PIV) as an argument position. Guilfoyle et al.'s discussion about the division of labor between PIV and F F corresponds closely to ours. Their ability to actually predict these properties is less clear. As in our analysis, F F is related to argument structure and thus has argumentrelated subject properties. PIV, on the other hand, has properties relating to factors other than argument structure. Among the properties of PIV that they enumerate are accessibility to extraction and the ability to launch floating quantifiers. The latter they link to the structural position of floating quantifiers (adjoined to Infl), while they don't actually explain the former. Among the properties of F F is antecedence of reflexives, explained by the G sensitivity of binding. Ability to be controlled seems to be shared by both. The reasons for this are not entirely clear: on the one hand they endorse the classical GB view that PRO is ungoverned, and thus restricted to SpecIP of nonfinite Infl, on the other hand they suggest an unspecified relationship to binding theory and optionality of government of the SpecVP position to explain the ability of PRO to appear in SpecVP. It is thus not entirely clear that the structural account explains the facts.

Within the framework that they assume, a nominal becomes PIV (moves to SpecIP) by virtue of not being assigned Case. They work out the analysis in detail for Malagasy, where the prefix anappears on actor voice verbs, the suffix -na on direct object voice verbs, and both appear on the verb if the pivot is something else specified in the lexical entry of the verb.

19. (Guilfoyle et al. (4,6)) a. M-an-sasa (manasa) ny lamba amin' ny savony ny zazavavy. TNS-ACT-wash the clothes with the soap the girl b. Sasa-na (sasan') ny zazavavy amin' ny savony ny lamba.

wash-DO the girl with the soap the clothes c. An-sasa-na (anasan') ny zazavavy ny lamba ny savony.

ACT-wash-DO the girl the clothes the soap 'The girl washes the clothes with soap.'

The analysis is that the prefix is part of the verb and assigns Case to the Patient, while the suffix is part of Infl and assigns Case to the Agent in SpecVP. The one that is not assigned Case moves to SpecIP, where it can be marked nominative. The attractiveness of this proposal comes from the combination of prefix and suffix: in such a case, both Agent and Patient are assigned Case and something else must move instead. An alternative analysis must treat the circumfix an-…-na as a third morphological element, unrelated to the active voice prefix and direct object voice suffix. Furthermore, constituent order facts in Malagasy support this analysis: the trace of the verb (which moves to Infl) intervenes between the Agent and the Patient, so the Patient is in a position adjacent to the verb if it is Case marked.

Attractive as it is at first glance, the Case-theoretic account faces some problems. Some of these problems are apparent in the analysis of Malagasy. In the first place, it requires an approach under which transitive verbs do not have the inherent ability to assign Case. This contradicts most approaches to Case in the GB tradition. Second, the fact that both Agent and Patient are assigned Case if the circumfix appears on the verb is not enough to explain the movement of another argument to SpecIP. As Guilfoyle et al. observe (fn 7), the preposition must be assumed to incorporate into the verb as well. If it did not, it would surface and assign Case to the nominal. This incorporation must be stipulated, and though Guilfoyle et al. claim that it is similar to an applicative construction, in the Malagasy case there is no morphological indication of the incorporation.

Further problems emerge when the analysis is extended to other Austronesian languages. Unlike Malagasy, languages like Tagalog do not combine the active voice affix and the direct object voice affix if an oblique argument becomes the pivot. This may indicate that, messy though it may be for Malagasy, the correct treatment of oblique-pivot affixes is simply as separate affixes. The word order facts are also less cooperative in other languages, as Guilfoyle et al. point out and discussed in detail for Tagalog by Kroeger (1993). Finally, it is unclear how the Guilfoyle et al. analysis would extend to syntactically ergative languages.

Final comments

It has been claimed (e.g. by Marantz 1984) that theories of syntax in which grammatical functions are not defined in terms of structural configurations are inherently less explanatory than GB-style theories. In fact, it is often the opposite that is the case. Hiding grammatical functions behind an array of structural constraints often obscures their nature as functions.

In this study, we have attempted to motivate the properties normally attributed to subjects in terms of two more basic functions. We have shown that hypothesizing the functions H F and PIV, and treating them as functions, can explain the properties they exhibit, particularly the array of properties in Philippine-type languages. The resulting system is more explanatory than related c-structural approaches, in which the surface system is a coincidence instead of the direct result of the need for interclausal continuity. It is also superior to a bistratal GF-based approach, in which the function name SUBJ is arbitrarily used to refer to distinct functions at the different strata. * I would like to thank Joan Bresnan for comments on much of the material in this paper. All the usual disclaimers apply.

1 Both Schachter and Dixon make additional claims, more inimical to the LFG conception of syntax, which I do not address here and do not subscribe to. Schachter, at least by implication, questions the universality of grammatical functions in general. It seems to me that that step is not justified by the facts of Philippine-type languages and ergative languages. Dixon claims that his subject and pivot are defined in terms of the more primitive functions S, A and O. Here again, I do not believe that a valid case has been made.

2 There are some interesting complications concerning objects, obliques, and adjuncts. In the first place, the "direct object voice" suffix is -in and the "indirect object voice" suffix is -an, but with some verbs -in is used when a normally dative-marked nominal becomes pivot, and with other verbs -an is used when the pivot is what would normally be marked accusative (Schachter 1987). I assume without argument that the a-function of the pivot is accurately reflected by the voice affix, and not by the Case it would normally be marked with.

A second problem is the ability of certain adjunct-like elements to be pivot. I suspect that this is yet another piece of evidence that the complement/adjunct distinction is more fluid than is usually thought, as suggested by Jackendoff (1990) and Alsina (1996). Foley and Van Valin (1984) remark on the general ability of Philippine languages to have pivots that are not core arguments; in the present context, however, I don't see that as a problem. The fact that these languages have a morphological indication of which argument has been chosen as pivot makes the wider options for pivothood natural. 3 A note on the glosses. The voice affixes are glossed ACT for active, DO for direct object, and IO for indirect object. Case clitics are glossed ERG (for the Case marking the Actor), ACC, DAT, OBL. Finally, I am assuming that the locative phrase in this set of sentences is a secondary object, and thus is marked as pivot by the IO affix. 4 In Bresnan (1982), this is indicated informally with the feature [U +]. 5 More precisely, if the subordinate verb is in the volitive mood the controllee must be I F for semantic reasons. In the nonvolitive mood, in which I F is semantically marked as not in control of the action, I F can be unexpressed only as an arbitrary controllee. According to Kroeger, in nonvolitive complements the controllee must be PIV. The restriction to PIV is not explained by my account (nor his), but if essentially correct it may be due to the function of PIV. The fact that I F can be specified as an unexpressed pronoun in the nonvolitive even though it cannot be controlled shows that there is a syntactic aspect to the choice of I F as anaphoric controllee. Schachter, P. (1987). Tagalog. In The world's major languages, edited by B. Comrie, 936-958. New York: Oxford University Press.