An Expanded Finite-State Transducer for Tsuut’ina Verbs

Proceedings of the 13th Conference on Language Resources and Evaluation (LREC 2022), pages 5143–5152 Marseille, 20-25 June 2022 © European Language Resources Association (ELRA), licensed under CC-BY-NC-4.0 5143 An Expanded Finite-State Transducer for Tsuut’ina Verbs Joshua Holden1, Christopher Cox2, Antti Arppe1 University of Alberta1 Carleton University2 holden1@ualberta.ca, christopher.cox@carleton.ca, arppe@ualberta.ca Abstract This paper describes the expansion of a finite state transducer (FST) for the transitive verb system of Tsuut’ina (ISO 639-3: srs), a Dene (Athabaskan) language spoken in Alberta, Canada. Dene languages have unique templatic morphology, in which lexical, inflectional and derivational tiers are interlaced. Drawing on data from close to 9,000 verbal wordforms, the expanded model can handle a great range of common and rare argument structure types, including ditransitive and uniquely Dene object experiencer verbs. While challenges of speed remain, this expansion shows the ability of FST modelling to handle morphology of this type, and the expanded FST shows great promise for community language applications such as a morphologically informed online dictionary and word predictor, and for further FST development. Keywords: Na-Dene, Indigenous languages, finite state transducers, morphology 1. Introduction This paper describes the development of a finite state transducer (FST) for the inflectional system of verbs in Tsuut’ina (ISO 639-3: srs), a highly endangered Dene (Athabaskan) language spoken near Calgary, Alberta, Canada. Arppe et al. (2017) outlined the creation of an prototype FST for intransitive verbs in Tsuut’ina, with only a few exemplary lexemes. The current pa- per describes the expansion of this FST architecture with a larger set of intransitive verbs as well as to encompass transitive verbs of various argument struc- tures, which vastly increases the range and complex- ity of the morphology that the FST needs to repre- sent. Full verb conjugations are needed for language learning resources, but manually creating paradigm ta- bles would be impractical. The FST automates this and responds to the Tsuut’ina community’s desire for computer-based language tools that are typical FST ap- plications, such as a morphologically enhanced online dictionary, a spellchecker, and predictive word sugges- tions on mobile devices. Through the integration of this FST with the GiellaLT infrastructure, all of these appli- cations are readily derivable from the FST described in this paper.1 FSTs (Beesley and Karttunen, 2003) use bidirectional rule-based mapping between sets of items, e.g. a string of characters as input, and a morphological analysis as output, or vice versa. FSTs are ideal for this work: they have open source implementations of compilers such as Foma (Hulden, 2009) (used in this project); they are compatible with most operating systems; their compu- tational properties and end-user applications are well established (see Moshagen et al. (2013)) and Trosterud (2004; Trosterud (2006), Arppe et al. (2016), Anton- sen et al. (2013), Johnson et al. (2013) for discus- sions of similar language-learning applications.). As data structures they are efficient for generating rule- 1https://giellalt.uit.no/infra/ based verb paradigms, which is advantageous given the limited data and corpora that are available for Tsuut’ina. However, caution is needed: rule-based paradigm gen- eration is only …

Content truncated to 500 words