In search of the recurrent units of translation

Abstract

Translation memory systems aim to reuse previously translated texts. Because the operational unit of the first-generation translation memory systems is the sentence, such systems are only useful for text types in which full-sentence repetition frequently occurs. Second-generation sub-sentential translation memory systems try to remedy this problem by providing additional translation suggestions for sub-sentential chunks. In this paper, we compare the performance of a sentence-based translation memory system (SDL Trados Translator's Workbench) with a sub-sentential translation memory system (Similis) on different text types. We demonstrate that some text types (in this case, journalistic texts) are not suited to be translated by means of a translation memory system. We show that Similis offers useful additional translation suggestions for terminology and frequent multiword expressions.