A morphological analyzer for Tamil

Abstract

This thesis deals with the designing and implementation of a morphological analyzer for the Tamil language. It also involves a comparative study of certain other models of morphological processing, in order to analyze the advantages of each, in terms of suitability for adaptation for a language like Tamil. This is primarily aimed at constructing a complete morphological module for Tamil that could be used in any Nl J5 application like a spell checker, POS tagger, or parser. Aspects of designing a computational model for morphological analysis include: 9*« Deciding a model based on psycholinguistic factors. 2*« Designing formal methods/ techniques that would enable converting theoretical descriptions into computational models. The analyzer under consideration relies on a theoretical blend of the LA and IP approaches to morphological decomposition. Wherever automatic phonological rules operate largelv, IP is incorporated. In areas where complex but non-automatic morphophonemics {sandbi) is involved, IA is the choice. Qualitative and quantitative methods in corpus linguistics were employed to extract frequency counts and collocations of words. All possible contexts of occurrence and usage of a word were studied. For every grammatical category of the language, an extracted list of the minimum number of word-forms required for a sufficient coverage had been prepared. Based on such attributes, and in consideration of the factors of coverage and effiaency for a morphological analyzer, an essential set of morphological paradigms for each word class in Tamil had been established. This served as a database comprising of different tables of inflectional forms of a word, for all tie words in the language.