A Mapudüngun FST Morphological Analyser and its Web Interface

Proceedings of the 13th Conference on Language Resources and Evaluation (LREC 2022), pages 6540–6547 Marseille, 20-25 June 2022 © European Language Resources Association (ELRA), licensed under CC-BY-NC-4.0 6540 A Mapud¨ungun FST Morphological Analyser and its Web Interface Andr´es Chand´ıa Universitat Pompeu Fabra Translation and Language Sciences Department andres@chandia.net / andres.chandia@upf.edu Abstract This paper describes the development and evaluation of an FST analyser-generator for the Mapuche language, which is publicly available through a web interface. As far as we know it is the first system of this kind for Mapud¨ungun. Following the Mapuche grammar by Smeets, we have developed a machine including the main morphological and phonological aspects of Mapud¨ungun. Through this computational approach we have produced an FST analyser-generator capable of classifying and appropriately tagging all the components (roots and suffixes) that interact in a Mapuche word-form. A double evaluation has been carried out showing a good level of reliability. In order to face the lack of standardization of the language, additional components (an enhanced analyser, a spelling unifier and a root guesser) have been integrated into the tool. The generated corpora, the lexicons and the FST grammars are available for further development and comparison of results. The Mapuche people inhabits southern Chile and Argentina and they speak Mapud¨ungun. Keywords: Mapud¨ungun, Mapuche, Morphology, Morphological Analyser, Finite-State Technology, FST, FOMA 1. Introduction We have defined and developed an FST analyser and generator for Mapud¨ungun using the finite state tech- nology. After the development and compilation of the transducer, the system has been evaluated and imple- mented as a web service for public free access. This paper introduces the basics of the system devel- opment, the evaluation and the codes and files com- prised in the machine, as well as a brief description of the web interface. A long description of the grammar is FST Morphological Analyser and Generator for Ma- pud¨ungun (Chand´ıa, 2021) that may be consulted for details. D¨ungupeyem is a computational tool (a rule-based mor- phological analyser and generator) for Mapud¨ungun, built with FST using foma (Hulden, 2009), an open source finite state compiler. Foma language descrip- tions exist for several languages including morpholog- ically complex ones as Quechua (R´ıos, 2015), Basque (Alegr´ıa et al., 2009) and others. The key component is a set of regular expressions en- coding the grammar. The main taking parts are de- clared in several scripts, lexicon (roots, suffixes, par- ticles, etc.) phophonology.and rules for morphotactics and mor- Regarding other computational projects on Ma- pud¨ungun, there are really few; the most remarkable one is a data collection and an attempt to develop a morphological analyser and a spellchecker (Levin et al., 2002) at the Carnegie Mellon University. There is also a basic Spanish-Ches¨ungun1 translator (Ru- mian, 2011). Our own related projects include COR- LEXIM (Chand´ıa, 2014a), a collection of 8 digitized dictionaries of Mapud¨ungun. A spellchecker prototype 1A Williche (southern) variant of Mapud¨ungun. (Chand´ıa, 2016), online and as a LibreOffice plug-in. IECMap (Chand´ıa, 2017), a morphologically annotated corpus and interface to exploit it. And …

Content truncated to 500 words