next up previous
次へ: Experiments for Sentence Speech 上へ: Automatic Acquisition on Stochastic 戻る: Analysis of a 16-state

Conclusion for Ergodic HMM as a language model

Although there are 114 different POSs, the experimental results show a high similarity between Ergodic HMM high output probability words and POSs. This means that the Ergodic HMM has an ability to automatically acquire both an SNG and the concept of POS simultaneously from (POS-untagged) training data.

We also observed that, as the number of states increases for an Ergodic HMM, the grouping of words starts from the conjunction and goes to the POS in Japanese.



Jin'ichi Murakami 平成13年1月19日