A Free/Open-Source Morphological Analyser and Generator for Sakha

Proceedings of the 13th Conference on Language Resources and Evaluation (LREC 2022), pages 5137–5142 Marseille, 20-25 June 2022 © European Language Resources Association (ELRA), licensed under CC-BY-NC-4.0 5137 A Free/Open-Source Morphological Analyser and Generator for Sakha Sardana Ivanova, Jonathan N. Washington, Francis M. Tyers Helsingin yliopisto, Swarthmore College, Indiana University Helsinki, 00014 Suomi, Swarthmore, PA 19081 USA, Bloomington, IN 47405 USA sardana.ivanova@helsinki.fi, jonathan.washington@swarthmore.edu, ftyers@iu.edu Abstract We present, to our knowledge, the first ever published morphological analyser and generator for Sakha, a marginalised language of Siberia. The transducer, developed using HFST, has coverage of solidly above 90%, and high precision. In the development of the analyser, we have expanded linguistic knowledge about Sakha, and developed strategies for complex grammatical patterns. The transducer is already being used in downstream tasks, including computer assisted language learning applications for linguistic maintenance and computational linguistic shared tasks. Keywords: morphology, Sakha, Turkic languages, FSTs, finite-state morphology, marginalised languages 1. Introduction This paper describes the development of a morphological transducer for Sakha using free/open-source tools available as part of Helsinki Finite State Technology (HFST). Sakha is a Turkic language, with around 450 000 native speaker (Eberhard et al., 2022), primarily residing in the Sakha Republic, where the language enjoys official status. The Sakha Republic is located in Northeast Asia, and is part of the Russian Far East. Sakha speakers are subject to increas- ing economic (Streletskiy et al., 2019) and cultural (Lavril- lier and Gabyshev, 2021) peril due to climate change, and the Sakha language faces social (Charter, 2022) and legal (Chevalier, 2017; Jankiewicz et al., 2020) marginalisation. The transducer described in this paper provides mor- phological analysis and generation for Sakha, is entirely hand-crafted, and is publicly available under the GPL v3 Free/Open Source licence.¹ Morphological transducers can be used in a wide range of language technology applications and “downstream tasks”; e.g., they may be repurposed as spell checkers and used in rule-based machine translation pipelines. The Sakha transducer described here is currently used in Revita, a language learning application designed to support individual efforts at language maintenance (Katin- skaia et al., 2018; Ivanova et al., 2019). Morphological transducers are an important technology for NLP, since they are linguistically informed and well understood, and require a single development cycle (Butt, 2020). This paper is structured as follows. Section 2 gives an overview of the Sakha language and related work. Sec- tion 3 describes the methodology for implementing the transducer, including details of how the morphotactics and morphophonology were dealt with. Section 4 presents an evaluation of the transducer, showing that it has decent cov- erage and precision and recall. Finally, future work and conclusions are presented in sections 5 and 6, respectively. ¹https://github.com/apertium/apertium-sah 2. Background and Methodology 2.1. Sakha The Sakha language (also known as Yakut), is the language of nearly half a million speakers, mainly in the Republic of Sakha in the Far East of Russia (Siberia), shown in Figure 1. Sakha belongs to the Lena group of the Turkic language family, and its lexicon …

Content truncated to 500 words