Predicting Co-Occurrence Restrictions By Using Semantic Classifications In The Lexicon

In this paper we investigate general principles of constructing semantic classifications that yield useful predictions combinatory options Several semantic Russian words are concernin~ of words. classes of discussed, implemented in an expert system named "Lex i co~raphe r", the "Lexicographer" is supposed to provide its users with all kind of information concerning some 15.000 most common Russian words. Alon~ with morphological, syntactic infor.~tion conventional system information and semantic usually stored in dictionaries, the should contain about referential characteristics of words and about restraints in combinability with other words in syntactic constructions of different types. In its final version "Lexicographer" should provide the users with all sorts of bibliographical information system beinff conceived as an aid both in the area of natural language processin~ arid in traditional lexicography. Semantic Features proposed regulate co-occurence of verbs with their non-obli?atory dependents-such as Modifiers of place or time; Instrumental and Benefactive objects and the like. (concern i n~ both i nd i v i dual words and semantic classes of words) and with concordances made on the bas i s of a suffi c i ent ly re presentat i ve corpus of Russian texts. One of the basic components of the system is its lexicon; the lexicon contains information not only about individual lexer~s, but also about sen~ntic and syntactic classes of lexemes. Thus, for nominal lexe~es such features are ~iven as: "NATURAL CLASS", "ARTEFACT", "MASS TERM", "SET", "BODY PART" and the like.