Evaluating a Morphological Analyser of Inuktitut

2012 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies, pages 372–376, Montr´eal, Canada, June 3-8, 2012. c⃝2012 Association for Computational Linguistics Evaluating a Morphological Analyser of Inuktitut Jeremy Nicholson†, Trevor Cohn‡ and Timothy Baldwin† †Department of Computing and Information Systems, The University of Melbourne, Australia ‡Department of Computer Science, The University of Sheffield, UK jeremymn@csse.unimelb.edu.au, tcohn@dcs.shef.ac.uk, tb@ldwin.net Abstract We evaluate the performance of an morpho- logical analyser for Inuktitut across a medium- sized corpus, where it produces a useful anal- ysis for two out of every three types. We then compare its segmentation to that of sim- pler approaches to morphology, and use these as a pre-processing step to a word alignment task. Our observations show that the richer ap- proaches provide little as compared to simply finding the head, which is more in line with the particularities of the task. 1 Introduction In this work, we evaluate a morphological analyser of Inuktitut, whose polysynthetic morphosyntax can cause particular problems for natural language pro- cessing; but our observations are also relevant to other languages with rich morphological systems. The existing NLP task for Inuktitut is that of word alignment (Martin et al., 2005), where Inuktitut to- kens align to entire English clauses. While Langlais et al. (2005) theorises that a morphological analyser could aid in this task, we observed little to no im- provement over a baseline model by making use of its segmentation. Nonetheless, morphological anal- ysis does provide a great deal of information, but the task structure tends to disprefer its contribution. 2 Background 2.1 Inuktitut Inuktitut is a macrolanguage of many more-or-less mutually intelligible dialects (Gordon, 2005). The morphosyntax of Inuktitut is particularly marked by a rich polysynthetic suffixing morphology, including incorporation of arguments into verbal tokens, as in natsiviniqtulauqsimavilli in (1). This phenomenon causes an individual token in Inuktitut to be approx- imately equivalent to an entire clause in English. (1) natsiq- seal -viniq- meat -tuq- eat -lauq- before -sima- ever -vit INT-2s -li but “But have you ever eaten seal meat before?” Lowe (1996) analyses the morphology as a four- place relationship: one head morpheme, zero or more lexical morphemes, one or more grammatical morphemes, and an optional enclitic. The morpho- tactics causes, amongst other phenomena, the final consonant of a morpheme to assimilate the manner of the initial consonant of the following morpheme (as in -villi), or to be dropped (as in natsiviniq-). Consequently, morphemes are not readily accessible from the realised surface form, thereby motivating the use of a morphological analyser. 2.2 Morphological analysis For many languages with a less rich morphol- ogy than Inuktitut, an inflectional lexicon is of- ten adequate for morphological analysis (for exam- ple, CELEX for English (Burnage, 1990), Lefff for French (Sagot et al., 2006) or Adolphs (2008) for German). Another typical approach is to perform morphological analysis at the same time as POS tag- ging (as in Hajiˇc and Hladk´a (1998) for the fusional morphology in Czech), as it is …

Content truncated to 500 words