Morphological Analysis of the Slovak National Corpus

1. Basis of a morphological analysis of the Slovak National Corpus A question of morphological (or morphosyntactic) analysis has been a key problem for natural language processing (NLP) for several years. Automatic morphological annotation is a useful tool especially with regard to the corpus data processing. In this respect morphological annotation has been considered also during the development of the Slovak National Corpus (SNC). Theoretical aspects of morphological analysis and its application in corpus tagging associated with the morphological tagset preparation for the manual tagging of SNC were outlined by M. Forróová and A. Horák (2003, in press). Annotation, generally understood as the process of adding some information to texts, is undoubtedly a convenient tool – in spite of different views – in verification of linguistic theories, but also in carrying out various lexicographic projects. Morphological analysis is generally understood as an assignment of a base form (lemmatization), classification of words into grammatical and semantic classes and assignment of grammatical categories to words in texts (in a form of tags). Generally, for a language-competent person this kind of analysis is not difficult, on the other hand for computer processing it is a hardnut task (Foróová – Horák, 2003). At the same time we take into account the problem of formal homonymy resulting from automatic morphological annotation; this problem requires a subsequent disambiguation. From the very beginning we were aware of the fact that a set of morphological tags should have represented language properties of a text, in other words, somehow it should have interpreted a text. We should decide whether the proposed tagset would result in a new formal description of language, or if we reflect to the current existing linguistic descriptions, and tried to formalise them. Foróová and Horák (2003) point to 7 maxims proposed by G. Leech, providing regularity of annotation, and guarantee that annotation wouldn’t result in misinterpretations of corpus data. We would like to emphasize a maxim of accessibility and a maxim of consensus of academic theories, theoretical “neutrality” which was predominantly taken into consideration when preparing a tagset. The central issue can be formulated as follows: to which extent we can refer to the traditional grammatical descriptions of Slovak morphology when preparing a lemmatization and a tagset? We considered relevant to take into account a systemic description made by academic Morfológia slovenského jazyka (lit. Morphology of the Slovak language, MSJ, 1966), eventually some other works dealing with morphology (Oravec – Bajzíková – Furdík, 1984; Dvonč, 1984). The conflict between representatives of traditional grammatical categories and a possibility of automatic language processing is reflected also in approach of morphological tagset for SNC. M.