A Resource-light Approach to Russian Morphology: Tagging Russian using Czech resources

A Resource-light Approach to Russian Morphology: Tagging Russian using Czech resources Jiri Hana and Anna Feldman and Chris Brew Department of Linguistics Ohio State University Columbus, OH 43210 Abstract In this paper, we describe a resource-light system for the automatic morphological analysis and tag- ging of Russian. We eschew the use of extensive resources (particularly, large annotated corpora and lexicons), exploiting instead (i) pre-existing anno- tated corpora of Czech; (ii) an unannotated corpus of Russian. We show that our approach has benefits, and present what we believe to be one of the first full evaluations of a Russian tagger in the openly avail- able literature. 1 Introduction Morphological processing and part-of-speech tag- ging are essential for many NLP tasks, including machine translation, information retrieval and pars- ing. In this paper, we describe a resource-light ap- proach to the tagging of Russian. Because Russian is a highly inflected language with a high degree of morpheme homonymy (cf. Table 11) the tags in- volved are more numerous and elaborate than those typically used for English. This complicates the tag- ging task, although as has been previously noted (Elworthy, 1995), the increased complexity of the tags does not necessarily translate into a more de- manding tagging task. Because no large annotated corpora of Russian are available to us, we instead chose to use an annotated corpus of Czech. Czech is sufficiently similar to Russian that it is reasonable to suppose that information about Czech will be rel- evant in some way to the tagging of Russian. The languages share many linguistic properties (free word order and a rich morphology which plays a considerable role in determining agreement and argument relationships). We created a morpho- logical analyzer for Russian, combined the results with information derived from Czech and used the TnT (Brants, 2000) tagger in a number of differ- 1All Russian examples in this paper are transcribed in the Roman alphabet. Our system is able to analyze Russian texts in both Cyrillic and various transcriptions. krasiv-a beautiful (short adjective, feminine) muˇz-a husband (noun, masc., sing., genitive) husband (noun, masc., sing., accusative) okn-a window (noun, neuter, sing., genitive) window (noun, neuter, pl., nominative) window (noun, neuter, pl., accusative) knig-a book (noun, fem., sing., nominative) dom-a house (noun, masc., sing., genitive) house (noun, masc., pl., nominative) house (noun, masc., pl., accusative) skazal-a say (verb, fem., sing., past tense) dv-a two (numeral, masc., nominative) Table 1: Homonymy of the a ending ent ways, including a a committee-based approach, which turned out to give the best results. To eval- uate the results, we morphologically annotated (by hand) a small corpus of Russian: part of the transla- tion of Orwell’s “1984” from the MULTEXT-EAST project (V´eronis, 1996). 2 Why TnT? Readers may wonder why we chose to use TnT, which was not designed for Slavic languages. The short answer is that it is convenient and successful, but the following two sections address the issue in rather more detail. 2.1 The encoding of lexical information in TnT TnT records some …

Content truncated to 500 words