Research On Chinese Spoken Language Processing In Dialogue System

With the high-speed development of information technology, how to receive and send information efficiently has become an urgent requirement. The research on human-machine dialogue system is just focused on such question. The aim of the research is to make the communication between human and machine easier.In this paper, a Chinese spoken dialogue system between human and machine in specialized domain is realized based on the deep research of the necessary and possible technique in a dialogue system. The emphases of the research are the architecture of dialogue system, the application of the technique dealing with language and the choice of the method of dialogue management.Spoken language processing is tied up with traditional natural language understanding technology, but has its own characters and difficulties. The problems possibly encountered in spoken language parsing and corresponding solving approaches are discussed in this paper. Some successful spoken language dialogue systems all over the world are introduced and some typical parsing methods adopted are analyzed in detail. The conventional processing method of nature language cannot fit the phenomenon of spoken language which contains lots of repetition, ellipsis, disordered components and liberty of mood. In order to solve those problems, an improved Context Free Grammar(CFG) is put forward based on the idea that enhancing the description power of the rules. Meanwhile, the conventional Tomita algorithm is modified to accept those rules and grammar. And a spoken language parser based on the concept recognition is designed. After that, the method of discourse analysis in dialogue management is discussed, a knowledge representation method using the topic tree is proposed. This method uses a topic tree to describe the information in the dialogue and combines the dialogue history with the concept in domain. And then the knowledge database is built up and the answering strategy is given.Combined with analysis of different knowledge sources, such as Chinese syntax and semantics, a Chinese spoken dialogue system for makeup information consultation is implemented. The spoken words in this field are collected and the key concept words are picked up. The parser unifying the analysis on grammar and semantics is established, too. Experiment shows that the dialogue has good robustness and achieves good performance in relevant domain. In the last part of the thesis, a summary was given to all of the research work and an expectation was also taken for what would be done to improve the performance of the spoken language dialogue in the future.