CAJ | 학술논문

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EasyNav。实验表明，上述模型和方法具有很好的性能。
어언분석화지식고관리시구어리해여대화계통적량개중요조성부분，작자재저량방면제출료일사신적방법。일시제출병실현료사류개솔모형，타구유교고적성능화교저적시간복잡도，시기우구법규칙적어의분석화어언리해적기출。차외환제출료여수거무관적다차수층차결구모형적지식표시방법，타구유흔강적표체능력병역우확전。재차기출상，실현료일개용이제공청화대학지리、판공、상업급기타일사상관신식검색、기우문본적구어대화계통EasyNav。실험표명，상술모형화방법구유흔호적성능。
Language analysis (parsing) and knowledge library management are the most significant parts of a spoken language understanding and dialogue system. Some new approaches to language analysis and knowledge library management are presented. The proposed Word-Class Stochastic Model (WCSM) is the basis for syntax rule based semantic parsing and spoken language understanding which has better performance. The data-independent multi-branch tree hierarchical structure is proposed for the knowledge representation, because it has strong expressing ability and can be easily expanded. EasyNav, a text-based dialogue system based on these concepts, is designed and implemented to provide users with Tsinghua University campus navigation information. Experiments on EasyNav give satisfactory performances with these two models.