Phrase-based Machine Transliteration

This paper presents a technique for transliteration based directly on techniques developed for phrase-based statistical machine translation. The focus of our work is in providing a transliteration system that could be used to translate unknown words in a speech-to-speech machine translation system. Therefore the system must be able to generate arbitrary sequence of characters in the target language, rather than words chosen from a predetermined vocabulary. We evalauted our method automatically relative to a set of human-annotated reference transliterations as well as by assessing it for correctness using human evaluators. Our experimental results demonstrate that for both transliteration and back-transliteration the system is able to produce correct, or phonetically e q u i v a l e n t t o c o r r e c t o u t p u t i n approximately 80% of cases.