Extracting Semantic Classes and Morphosyntactic Features for English-Polish Machine Translation

This paper describes a procedure aimed at automatic extraction of certain noun and verb categories from Polish texts. The general goal is to construct a lexical database that should be incorporated into a system for machine translation and multilingual generation of summaries. High quality processing of inflectional languages like Polish requires quite elaborated lexical entries, it is therefore highly desirable to automate the process of lexicon construction, at least partially. However, purely statistical methods for languages with less elaborated inflectional systems do not perform especially well on Slavic languages. As primary cues for automatic subcategorization we used inflectional morphemes expressing the greatest number of semantico-syntactic functions. The crucial semantic category for noun classification was the degree of animacy. Morphosyntactically, this category is expressed by nominal suffixes and subject-verb agreement markers. The procedure for lexical extraction and classification was implemented in Delphi and the system was trained for extraction of so-called superanimate nouns, i.e. nouns denoting male human beings, or groups including both male and female humans. The usability of lexical extraction based on concurrence of morphological features rather than on concurrence of whole word forms is evaluated and discussed.