Efficient Grapheme-phoneme Alignment for Japanese
Lars Yencken, Timothy J. Baldwin · 2005
Current approaches to the grapheme-phoneme alignment problem for Japanese achieve good accuracy, but are extremely computationally expensive. In this paper we evaluate various modifications to previous algorithms for both the alignment and okurigana detection subtasks. The best algorithm achieved accuracy of 96.2% for the combined task on a limited data set, and was significantly more efficient than previous approaches. 1