Development of a translation memory system for Turkish to English

Abstract

This thesis studies translation memory systems and development of a translation memory system from Turkish to English. A translation memory system is a tool that is designed for helping human translators during translation. It uses a database that consists of text parts such as blocks, paragraphs, sentences, or phrases in one language that is called source language and their translations in another language called target language. The translation memory system searches its database to nd the closest sentence for the source sentence that has to be translated. Translation memory systems that are used by human translators o er the best matches for the review of the translator. The translator either accepts it, or edits it, or rejects it and asks for the second best result. Translation memory system concept is not popular for Turkish. To the best of our knowledge, there is no translation memory system developed for Turkish to English. This is the rst translation memory system from Turkish to English, noting the existence of a previous example based machine translation system for Turkish to English, a similar but di erent concept. The proposed system uses a sentence level memory but exploits words and various linguistic features of the language. The similarity search algorithm takes advantage of highly agglutinative word structures, in ectional and derivational a xes of Turkish. The presented framework considers orthographic, morphologic, lexical, semantic and syntactical features. It gains good success rate on Turkish (about 0.40 BLEU score) and is expected to be helpful on the translation of languages that have similar linguistic structure such as Hungarian or Finnish.