Non-productive machine transliteration

Abstract

This paper proposes a new framework of machine transliteration, called non-productive machine transliteration. In this framework, it is assumed that a large candidate list including the correct transliteration is given. Therefore, the transliteration problem is simplified into the selection problem of the correct entry from the large list. We have developed an efficient algorithm of this framework and applied it to English-Japanese transliteration of person names. Experimental results show that our algorithm is practical even if the size of the candidate list is over a million.