Discriminative Method for Japanese Kana-Kanji Input Method

The most popular type of input method in Japan is kana-kanji conversion, conversion from a string of kana to a mixed kanjikana string. However there is no study using discriminative methods like structured SVMs for kana-kanji conversion. One of the reasons is that learning a discriminative model from a large data set is often intractable. However, due to progress of recent researches, large scale learning of discriminative models become feasible in these days. In the present paper, we investigate whether discriminative methods such as structured SVMs can improve the accuracy of kana-kanji conversion. To the best of our knowledge, this is the first study comparing a generative model and a discriminative model for kana-kanji conversion. An experiment revealed that a discriminative method can improve the performance by approximately 3%.