Using a Third Language to Improve Extraction of Bilingual Term Correspondences

This paper describes a method for improving the quality of the extraction of bilingual term correspondencies. The basic intuition is that using a parallel text consisting of three (or more) languages, the third language will provide additional information when extracting term correspondencies between the original source and target languages. First experiments show encouraging results, with overall accuracy increasing about two percent, varying slightly with the direction of translation and method of evalutation. Further exten-sions of the method are suggested, which are hoped to be effective in cases where the current method is unable to improve the preformance. 1