First International Conference on Spoken Language Processing (ICSLP 90)

Kobe, Japan
November 18-22, 1990

Sentence Recognition Method Using Word Cooccurrence Probability and Its Evaluation

Isao Murase, Seiichi Nakagawa

Toyohashi University of Technology, Toyohashi, Japan

In this paper, we describe the sentence recognition method using word cooccurrence probability, and compare it with the method using CFG(Context-Free Grammar). Since the aim of this study is to investigate whether the word cooccurrence probability (bigram or trigram) is able to represent the corresponding context-free grammar or not, we calculated it from CFG, By comparing the results using the bigram method with those using the method of CFG, we realized that it is insufficiency to represent CFG with only word cooccurrence probabilities from both perplexity and sentence recognition accuracy. Therefore, to improve the bigram we have extended the bigram model to a quasi-trigram model (subclass-word-word). From the experimental results using this quasi-trigram we realized that the quasi-trigram model can represent the corresponding CFG better than the corresponding bigram model.

Full Paper

Bibliographic reference.  Murase, Isao / Nakagawa, Seiichi (1990): "Sentence recognition method using word cooccurrence probability and its evaluation", In ICSLP-1990, 1217-1220.