Automatic Disambiguation of the Belarusian-Russian Legal Parallel Corpus in NooJ

The article deals with the problem of lexical ambiguity in the legislative texts of the Belarusian legal codes. Specialists of the Speech Synthesis and Recognition Laboratory in Minsk translate law codes from Russian into Belarusian to ensure equal access to them for speakers of both official languages of our country. First, the texts are translated automatically and checked and edited by translators, linguists, and lawyers. Over 20 of the 26 codes have already been prepared. The article examines types of homonyms, their semi-automatic extraction and classification using the NooJ development environment. There are two dominant classes of ambiguity resolution mechanisms: automatic, implying a completely computerized solution to this problem and interactive (semi-automatic), supposing a joint solution by a person and a computer. It means the user has a set of alternatives from which he should choose one option. About 17% of words in the experimental corpus have homonyms. The article examines their types, and presents several syntactic NooJ grammars, which allow for removing different types of homonymy.