Towards A Robust Morphological Analyzer for Kunwinjku

Towards A Robust Morphological Analyzer for Kunwinjku William Lane and Steven Bird Charles Darwin University The Northern Institute Abstract Kunwinjku is a polysynthetic language spoken in northern Australia. Members of the commu- nity have expressed interest in co-developing language applications which could assist in the production of written language resources for education and language learning. Modelling Kunwinjku morphology is a step towards ac- complishing these goals. We discuss some of the modeling challenges presented by Kun- winjku verbal morphology, and in polysyn- thetic languages more generally. We show that a model using standard features of the Foma toolkit can account for much of the verb struc- ture. Our contributions include the first mor- phological analyzer for Kunwinjku, and a dis- cussion of polysynthetic language features and how they affect modelling decisions. Continu- ing challenges include robustness in the face of variation and unseen vocabulary, as well as how to handle complex reduplicative pro- cesses. 1 Introduction Kunwinjku is an Aboriginal language of the Gun- winyguan language family (ISO gup), spoken by about 2000 speakers in the West Arnhem region of northern Australia. Several Kunwinjku communi- ties have shown interest in leveraging technology to support the production of literacy materials and language learning applications (Bird, 2018). A major focus of our research group is to im- plement language technologies that have positive social impact, such as a morphologically-aware dictionary which lowers the barrier to entry for users who cannot reliably identify or spell cita- tion forms (Hunt et al., 2019; Arppe et al., 2016), or a tool that generates linguistic structures which could help language learners master conjugation and verb structure (Kazantseva et al., 2018). One thing that these applications have in com- mon is the need to decompose and manipulate text at the level of morphology. In order to accom- plish this, we must address polysynthesis, mor- phophonemic alternations, incorporation, redupli- cation, and long-distance dependencies. aspects of morphosyntax can we model?Which What are the limitations of computational approaches for modeling polysynthetic languages more gen- erally? In the sections that follow, we will first give an overview of those features of the language which affect how we approach the modelling task (sec 2). Next, we introduce our data sources and the met- rics we use to evaluate performance (sec 3). This is followed in section 4 by a detailed description of our implementation and how we addressed the linguistic features described in section 2. Finally, we report accuracy and coverage on both a devel- opment data set and a blind test set, provide an error analysis and discussion, and conclude with some thoughts on future directions. To our knowl- edge, this is the first morphological analyzer for Kunwinjku. 2 Features of Kunwinjku Verbs We model and evaluate the morphosyntax of Kun- winjku verbs according to Evans’ Pan-dialectal Grammar (Evans, 2003). In this section we de- scribe some of these features, and follow-up later with how we account for them in the model. 2.1 Polysynthesis and Agglutination Kunwinjku is a polysynthetic language, …

Content truncated to 500 words