An improved MAP method for language model adaptation

This paper presents an improved MAP method for language model adaptation. The traditional MAP method mixes the task independent corpus and task dependent corpus using a fixed weight. In the method presented in this paper, we replaced the fixed weight with a function of history word. Another work in this paper is that a fuzzy controller was introduced in adaptation process, and three factors were used to be the input of the controller, they are: 1) the confidence of the estimation value, 2) the importance of the word, 3) the difference between the estimation value from the general corpus and from the adaptive corpus. The experiments showed that the improved method has the better performance than traditional model.