Corpus-based Disambiguation for Machine Translation

This paper deals with problem of choosing a proper translation for polysemous words. We describe an original method for partial word sense disambiguation of such words using word sketches extracted from large-scale corpora and using simple English-Czech dictionary. Each word is translated from English to Czech and a word sketch for the word is compared with all word sketches of its appropriate Czech equivalents. These comparisons serve for choosing a proper translation of the word: given a context containing one of collocates from the English word sketch, result data can serve directly in the process of machine translation of the English word and at the same time it can be considered as a partial disambiguation of that word. Moreover, the results may be used for clustering word sketches according to distinct meanings of their headwords.