A Finite-state Morphological Analyser for Tuvan

A Finite-State Morphological Analyser for Tuvan Francis M. Tyers, Jonathan North Washington, Aziyana Bayyr-ool, Aelita Salchak HSL-fakultehta, Depts. of Linguistics and Central Eurasian Studies, Institute of Philology, Dept. of Tuvan Philology UiT Norgga árktalaš universitehta, Indiana University, Russian Academy of Sciences, Tuvan State University N-9019 Romsa; Bloomington, IN 47405; Novosibirsk; Kyzyl francis.tyers@uit.no, jonwashi@indiana.edu, azikoa@mail.ru, aelita_74@mail.ru Abstract This paper describes the development of free/open-source finite-state morphological transducers for Tuvan, a Turkic language spoken in and around the Tuvan Republic in Russia. The finite-state toolkit used for the work is the Helsinki Finite-State Toolkit (HFST), we use the lexc formalism for modelling the morphotactics and twol formalism for modelling morphophonological alternations. We present a novel description of the morphological combinatorics of pseudo-derivational morphemes in Tuvan. An evaluation is presented which shows that the transducer has a reasonable coverage—around 93%—on freely-available corpora of the languages, and high precision—over 99%—on a manually verified test set. Keywords: morphological analysis, finite-state transducers, Tuvan 1. Introduction This paper describes the development of a morphological transducer for Tuvan. The paper is laid out as follows: §2. gives a short introduction to Tuvan and §3. describes some prior work on computational linguistics for Tuvan. Then §4. documents how a number of issues related to mor- photactics (§4.2.) and morphophonology (§4.3.) were dealt with. An evaluation of the transducer is provided in §5., and §6. outlines future work related to the transducer. 2. Language Tuvan (demonym [tɯβɑ]) is the largest member of the Sayan branch of Turkic languages. It is an official lan- guage of the Tuva Republic (in Southern Siberia, within the Russian Federation, see figure 1), and is also spoken in the surrounding areas. Russia’s 2010 census (Росстат, 2011) recorded over 250,000 Tuvan speakers, and Lewis et al. (2015) report about 27,000 speakers in Mongolia and about 2,400 in China. Many Tuvan speakers also know Russian, Mongolian, or Chinese, depending on which country they are from. Like other Turkic languages, Tuvan exhibits a rich system of agglutinating morphology, repleat with productive and idiosyncratic morphotactics and morphophonology. There have been a number of grammars written for Tuvan, in- cluding a large academy grammar in Russian (Исхаков and Пальмбах, 1961), and a grammar sketch in English (An- derson and Harrison, 1999). 3. Prior work Very little work has been done on computational linguis- tics for Tuvan, even basic resources are lacking. Of the two publications on computational linguistics, we find one pa- per on proposing a tagset for the Tuvan National Corpus (Bayyr-ool and Voinov, 2012), and one Bachelor’s thesis on Tuvan–English statistical machine translation (Killackey, 2013). The analyser presented in this paper does not follow the tagset designed by Bayyr-ool and Voinov (2012), and instead uses a pan-Turkic tagset being adopted by the Aper- Figure 1: Location of the Tuva Republic tium project.1 It is worth noting however that our tagset is a superset of the tagset of Bayyr-ool and Voinov (2012), that is it makes more distinctions rather than fewer distinctions, and as such conversion from our tagset …

Content truncated to 500 words