1. INTRODUCTION. (1) All of the Formosan languages, with the singular exception of Rukai, have a voice system that allows different core arguments to be placed in "subject" position, thereby marking them as identifiable, and that signals the presence of a particular semantic role associated with the subject. Two voice constructions are usually distinguished: actor voice and undergoer voice. The actor voice is basically intransitive, and the undergoer voice transitive.
However, the nature of the actor voice constructions has been a source of considerable debate (Chang 1997, 2004; Reid and Liao 2004; Ross and Teng 2005). Two questions that have often been raised are: What is the nature of the difference between the actor voice construction and the undergoer voice construction? Are there gradations of transitivity such that the actor voice construction in some languages may be transitive, though not necessarily as transitive as the undergoer voice construction in those languages?
Exploring these issues is beyond the scope of this paper, but is the focus of research in progress. In this study we investigate transitivity alternations and the connection between transitivity and case-marking in three Formosan languages, Kavalan, Squliq, and Tsou, especially the nature of the core/oblique distinction, based primarily on discourse-pragmatic evidence. We hope that the findings derived from the present study will help provide at least suggestive answers to these questions. Two subtypes of the intransitive construction are distinguished: the ordinary intransitive construction that takes no patient argument whatsoever, and the extended intransitive construction (EIC) that takes a patient-like argument. We will demonstrate below that these languages have grammaticized the core/oblique distinction in EICs, and that the patient arguments in EICs have little saliency, since they are rarely tracked in subsequent discourse.
Case, valency, and transitivity are central concepts in Formosan and Philippine linguistics, but they are also sources of considerable controversy. The controversy centers around the interpretation of the nature of the extended intransitive clauses (EICs). Most semantically transitive verbs in Formosan languages may occur in a grammatically transitive nonactor voice (NAV) construction, the canonical transitive clause, or they may occur in a grammatically intransitive EIC clause marked by the presence of an E, an oblique-marked NP (the term EIC is due to Dixon 1994). E refers to the second argument of a dyadic intransitive verb that is marked differently from S, A, and O. The exact marking is irrelevant; it may be oblique (as in the languages under examination here); dative, as in Dyirbal; partitive, as in Hungarian; or locative, as in Tongan (Dixon 1994:123). Subscripts can then be used to distinguish different types of E. For instance, [E.sub.o] is an E that is like an O, El is an E that is like a locative, and so forth.
It is commonly accepted that if a language has two distinct two-argument structures, one will be higher in transitivity than the other one. In high transitive clauses, objects tend to be more identifiable and the clauses tend to be perfective; in lower transitive clauses, the objects tend to be less identifiable, and the clauses tend to be imperfective. What is special about the two two-argument structures (EIC and NAV) in the languages examined here is that neither definiteness nor perfectivity can be shown to distinguish one clause type from the other, especially in Squliq and Kavalan. Rather, it is their difference with regard to participant tracking that is the hallmark of the distinction between the two clause types. If the same verbs and argument NPs can appear in two different constructions, EIC and NAV, then these constructions can be shown to differ in their discourse-pragmatic behavior, with the referents of the Es in EICs showing little continuity, while the referents of Os in NAVs are more salient and exthbit a much stronger propensity to be tracked. …