语言分析和知识库管理是口语理解与对话系统的两个重要组成部分 ,作者在这两方面提出了一些新的方法。一是提出并实现了词类概率模型 ,它具有较高的性能和较低的时间复杂度 ,是基于句法规则的语义分析和语言理解的基础。此外还提出了与数据无关的多叉树层次结构模型的知识表示方法 ,它具有很强的表达能力并易于扩展。在此基础上 ,实现了一个用以提供清华大学地理、办公、商业及其它一些相关信息检索、基于文本的口语对话系统 Easy Nav。实验表明 ,上述模型和方法具有很好的性能 Language analysis and knowledge base management are two important parts of the spoken language understanding and dialogue system. The author proposes some new methods in these two aspects. The first is to propose and implement the word probability model, which has high performance and low time complexity, and is the basis of semantic analysis and language comprehension based on syntactic rules. In addition, a knowledge representation method of multi-fork tree hierarchy model which has nothing to do with data is proposed. It has strong expression ability and is easy to be extended. On this basis, to achieve a to provide Tsinghua University, geography, office, business and other related information retrieval, text-based spoken dialogue system Easy Nav. Experiments show that the above models and methods have good performance