Interactive Word Completion for Plains Cree

Proceedings of the 60th Annual Meeting of the Association for Computational Linguistics Volume 1: Long Papers, pages 3284 - 3294 May 22-27, 2022 c⃝2022 Association for Computational Linguistics Interactive Word Completion for Plains Cree William Lane Northern Institute Charles Darwin University Atticus Harrigan Alberta Language Technology Lab University of Alberta Antti Arppe Alberta Language Technology Lab University of Alberta Abstract The composition of richly-inflected words in morphologically complex languages can be a challenge for language learners developing lit- eracy. Accordingly, Lane and Bird (2020) pro- posed a finite state approach which maps pre- fixes in a language to a set of possible comple- tions up to the next morpheme boundary, for the incremental building of complex words. In this work, we develop an approach to morph-based auto-completion based on a finite state morpho- logical analyzer of Plains Cree (nêhiyawêwin), showing the portability of the concept to a much larger, more complete morphological transducer. Additionally, we propose and com- pare various novel ranking strategies on the morph auto-complete output. The best weight- ing scheme ranks the target completion in the top 10 results in 64.9% of queries, and in the top 50 in 73.9% of queries. 1 Introduction The ACL 2022 theme track asks how we can scale up current NLP technologies for the rich diversity of human languages. We contend that impactful work, intended to support local language commu- nity goals, does not necessarily focus on scale or current NLP technologies. Community values and motivations regarding language technology is as rich and varied as human language itself, and so- lutions which are well received in one community may not be adequate or appropriate in another. Con- text must be considered, from which novel tasks take shape. Lane and Bird (2020)’s re-imagining of word completion for morphologically-rich languages is an example of work born from a local con- text. Based in Kunwinjku-speaking communities in northern Australia, we wanted to support peo- ple’s desire to learn literacy and practice building the language’s long polysynthetic words. This led to the idea for a tool that helps users incrementally build complex words by suggesting completions Figure 1: Given some string prefix of a word in Plains Cree, Morph completion suggests continuations up to the next morpheme boundary (in bold), for interactive and incremental building of morphologically complex words. up to the next morph boundary (Figure 1). While Plains Cree and Kunwinjku speaking communi- ties are thousands of miles apart, the authors of this work noticed some similarities in their respec- tive contexts: both Plains Cree and Kunwinjku are polysynthetic languages, and both are working with communities to support language learning initia- tives. Other aspects of our situations differ. For ex- ample, the FST morphological analyzer for Kun- winjku can be described as a field tool, developed by a single researcher over the course of a couple years, while the Plains Cree morphological ana- lyzer has been in continuous development by a team for over 8 years. The Plains Cree model is much more extensive, …

Content truncated to 500 words