Unification of multi-word expression translations during collaborative compilation of bilingual lexicons

This paper addresses the collaborative compilation of a bilingual lexicon used as knowledge base for a natural language processing (NLP) system. A pilot experiment confirmed the inevitable variability of corpus-relevant translation equivalents offered by different translators for the same lexical units. The authors suggest the introduction of an additional step to unify translation equivalents for multi-word units in the lexicon compilation process to ensure terminological consistency of the NLP-system. A formalized procedure has been developed to perform this step. An experimental evaluation of the developed procedure based on a sample of target terms allowed the authors to conclude that it is possible to minimize the quantity of translation equivalents for components of multi-word expressions. Lexicon entry correction using the results of the suggested procedure enables unification of the entries’ lexical manifestations and, in the long term, can provide for higher quality NLP-systems.