next up previous
次へ: Second-order Ergodic HMM 上へ: Discussion 戻る: Discussion

Random Value

In this paper, word output probabilities are given random weights, when the number of states is increased (Technique 2). Otherwise, we use parts of speeches or categories using word dictionaries. For example, words belonging to the same category are given the same weight. This is a good way of trying to obtain better language models.



Jin'ichi Murakami 平成13年1月19日