In Search for a Linguistically Motivated Method for Matching in Translation Memories

Abstract: Translation memories are one of the most important tools used by professional translators to facilitate their work. Translation memories are essentially databases which store previous translations and retrieve them whenever the translator needs to translate a sentence which is similar to one translated before. Existing commercial systems rely on edit distance to determine similar translations. Edit distance is very fast, but it fails to work properly when two sentences express the same meaning using different words. This talk will present several similarity measures which attempt to address this drawback.