Towards a Lexicographic Approach to Lexical Transfer in Machine Translation (Illustrated by the German–Russian Language Pair)

Abstract

The translation of lexical items is still a formidable obstacle in the field of Machine Translation. The present article addresses this problem from the perspective of modern lexicography, putting forth detailed monolingual lexica which contain highly elaborated descriptions of lexical items. Leaving aside the resolution of homonymy on the source side and the selection of the appropriate synonym on the target side, we concentrate on the case of ``multiple lexical correspondences'', i.e., mismatches between monosemic lexical items. We propose to use the lexical transfer methodology. Our proposal exploits a fine-grained semantic decomposition of the meaning of both source and target lexical items as well as information on ``actantial valency'' (semantic restrictions) and on collocations. The lexical descriptions are formalized so as to allow for automatic matching thereof. We present a rule-based procedure that selects the most appropriate translation equivalent of a source lexical item by using matching. The proposal is illustrated drawing on three German and three Russian monosemic speech-act nouns; their lexical entries are given in an appendix. Two examples of lexical transfer between these lexemes are analyzed in detail.