A Research On The Dialogue-based Automatic Telephone Switching And Information Inquiry Technology

Spoken dialogue system, a hi-technology which can realize the man-machine dialogue, is hot in speech recognition field at present. There is a very good application prospect in traveling navigation, ticket service and telephone appointment, etc.This topic first analyze in Chinese the common grammar phenomena to conduct the research Chinese Natural Language Processing(NLP) based on the glossary dictionary. What is different to English is, Chinese noun, the verb and the adjective all have many kinds of grammatical functions, moreover the word expressions and the sentence expressions are basically same. All these phenomena bring great difficulty to Chinese Analysis. To overcome these disadvantageous phenomena which appears in Chinese, we design and realize a lexical analysis algorithm based on multi-HMM(Hiden Markov Model) and LR parsing algorithm.This thesis synthesize and improve a sentence similarity computation method. Through establishing the words corresponding relations after segmentation and tagging, sentence similarity is computed using words similarity. It even more conforms to the people's natural language understanding and enhances the system performance.We also design a new data structure to find the two sentences which have the most same or similar words. To reduce the search time and to realize timeliness, we need carry on off-line operation to the knowledge library to get a word-frequency library.Dialogue management model is the core partial in spoken dialogue system. To complete the function, we adopt the theme-trees structure, which can realize switching over freely between many themes. And it's organizational form is easily understood, easy to safeguard and expand. To optimize the memory of information, we introduce the information sharing.