RESEARCH ON A CHINESE LANGUAGE MODEL BASED ON MULTI KNOWLEDGE SOURCES AND ITS IMPLEMENTATION

A method of integrating statistical information and different kinds of rules for Chinese language modeling is presented, which represents the rule as figure, introduces the concept of syntactic and semantic rules matrix, and the embeds the phrase rules represented as CFG, the syntactic and semantic rules, and least segmentation principle into the N gram statistical Chinese language model by augmenting the word lattice and adjusting the N gram probabilities based on maximum likelihood. The technique is applied in Chinese Pinyin to character conversion and improves accuracy of the system.