UC Santa Barbara

This dissertation is a comprehensive analysis of voice constructions in Kanakanavu, a critically endangered Formosan language (Austronesian language of Taiwan) spoken in the Namasia District of Kaohsiung, southern Taiwan. The data considered for the analysis are mainly drawn from a corpus of spontaneous speech produced by contemporary speakers, which are complemented by elicited data and a verb database constructed by the author.

The Kanakanavu voice system has been treated in various previous studies, but there is a general lack of consensus regarding several of its fundamental properties. The language has been analyzed as exhibiting a two-way, three-way or even four-way/Philippine-type voice distinction; it also remains unclear how voice interacts with tense-aspect-mood marking on the one hand, and transitivity and grammatical relations on the other. By investigating the morphosyntactic, semantic and discourse-pragmatic properties of voice, this dissertation (i) establishes a framework for describing and representing Kanakanavu natural discourse data, (ii) analyzes the morphological and syntactic foundations of Kanakanavu voice and (iii) explores the functional-typological implications of Kanakanavu voice constructions both within and beyond the Austronesian language family.

It is argued that Kanakanavu exhibits a binary voice distinction in its verbal-clause morphosyntax.The structural opposition --- between what are labelled "agent voice" and "patient voice" --- is evidenced in two specific patterns of interaction between voice marking and tense-aspect-mood marking: while agent-voice verbs always involve separate exponents for voice and tense-aspect-mood, patient-voice verbs are consistently marked by voice markers that also serve a tense-aspect-marking or mood-marking function. The finding regarding the number of voice distinctions is in line with those found in some recent analyses of Kanakanavu voice, but it is motivated by typologically informed analyses of verbal tense-aspect-mood marking in the language based on detailed examinations of how different verb forms are used in spontaneous speech.

At the syntactic level, voice alternation in Kanakanavu is argued to be essentially a transitivity-alternation phenomenon, leading to the analysis of the agent-voice construction as the intransitive construction (involving only one core argument), and the patient-voice construction as the transitive construction (involving two core arguments). The analysis is based on the observation from natural discourse that the patient-voice construction is always used for expressing transitive situations, while the agent-voice construction is the default construction for forming (i) basic intransitive clauses and (ii) clauses involving syntactically demoted and discourse-functionally backgrounded patients. The finding suggests that the Kanakanavu verbal clause exhibits ergativity, which is in line with many analyses of Formosan and Philippine languages where voice interacts closely with transitivity. It is, however, further shown that Kanakanavu exhibits differential (pronominal) agent marking in its patient-voice construction. This is a typologically and areally unusual phenomenon within the larger Western Austronesian context, which has been generally overlooked in previous studies on the language despite having been identified early on by Tsuchida (1976).