Semantics Recognition Based On Grammar Rules And Statistics Of Intelligent Speech Dialogue System

With the development of speech recognition technology, speech synthesis technology and natural language processing technology, Intelligent speech dialogue system, which integrates these three key technologies, has been applied to more and more industries. In the intelligent speech dialogue system, the interaction between people and system is in the form of dialogue, which makes people obtain information more anthropomorphic, more convenient and faster.Semantics recognition, which is the key technology to implement an intelligent speech dialogue system, makes the dialogue system understand the intention of a speaker and respond correctly by analyzing the result of speech recognition. This dissertation focuses on semantics recognition technology of the intelligent speech dialogue system. The main contributes includes:1. Realization of rule-based semantics recognition. This dissertation proposes a matching approach of semantic rules "Pattern", which represents the language commonly used in spoken dialogue. Pattern is used to match a part or full semantics of the alphabetic string from the result of speech recognition. The matching approach of pattern is based on exact string matching between a user query and the pattern. The dialogue system classifies the intention of a user query according to the matching result of pattern, and then extracts key information of the user query string through the label mechanism. According to the key information, the dialogue system inferences the correct intention of a user query, and makes a decision of the inference result. Finally, the dialogue system passes the user’s query intention to the host platform of the system in the form of parameters data.2. Realization of semantic binary classification based on statistic language models. With respect to the semantics recognition problem of some particular functional categories in the intelligent speech dialogue system, this paper presents a semantics decision-making approach through judging the maximum likelihood estimation of a user query, which is calculated by the n-gram statistic language model.3. Propose a semantics recognition approach which integrates the rule-based method and the statistic-based method. According to the recall rate of the statistic-based method and the high precision rate of the rule-based method, this paper presents a semantics recognition approach which integrates these two semantics recognition methods. This integrated approach makes a preliminary intention decision through judging the maximum likelihood estimation of a user query, and then inferences the user’s intention through parsing the key information from the result of speech recognition. This integrated method enhances the recall rate of some particular functional categories in the intelligent speech dialogue system, while it can ensure the high precision rate.