On the representation of Austronesian voice systems in Universal Dependencies

Universal Dependencies (Zeman et al., 2024) is a multilingual, cross-lingually consistently annotated corpus of dependency parsed sentences in many languages, assembled with the aim of enabling, among other things, corpus-based comparative and typological study at scale. A central desideratum of the UD annotation scheme is to be universal and consistent between languages, so that like-for-like comparisons can be made.

Despite constituting a large portion of the world’s documented languages, inclusion of Austronesian languages in Universal Dependencies is currently quite limited, with, as of version 2.15, only Indonesian, Cebuano, Javanese, and Tagalog represented.

Austronesian languages that exhibit symmetrical-voice systems have long been problematic for various syntactic theories because they do not seem to privilege any voice in a way that is characteristic either of accusative- or of ergative-aligned languages; that is to say, all voices are transitive with no demotion of argument (see Chen & McDonnell 2019). In some symmetrical-voice languages, an agent or patient argument can serve as a so-called “pivot” and thus trigger agreement, which is marked by verbal morphology; in other types, there can be even more voices, involving other kinds of arguments as pivots (e.g., location and instrument arguments). Normally, in the literature, the two-voice systems are characteristic of the so-called “Indonesian type”, and the four-voice systems are characteristic of the so-called “Philippine type”; some languages, such as Gorontalo (Sulawesi, Indonesia) can be found in the middle of the spectrum.

Representing all of these voice systems in Universal Dependencies presents a challenge. At present, the UD_Tagalog-TRG corpus is the only one to represent this voice system There is a need for expansion of UD to more Austronesian languages and for a survey of the kind of voice systems that we can expect to encounter when annotating these languages.

In this talk, we will make the following contributions: (1) A test-sized corpus of parsed Gorontalo sentences, comprising both elicited and naturally occurring examples, and showcasing the rich voice system in this language; (2) A preliminary survey of voice systems, including annotated examples from other Austronesian languages, showcasing the difference in parses and feature annotations between the languages; (3) Presentation of a syntactic relation and feature set that applies as widely and parsimoniously as possible to Austronesian languages; and (4) Recommendations for how to annotate Austronesian languages. We hope to lay the groundwork for future annotation efforts for Austronesian languages in UD and cross-linguistically consistent corpora for quantitative typological study of these languages.