面向自然语言处理的会话行为分类研究

Developing a Dialogue Acts Typology for Natural Language Processing

"会话行为"(dialogue act)指某个话语在对话中传达出来的说话人意图。近年来,国外把"dialogue ac"t的自动识别作为口语会话自动理解系统的关键,其中最基本的任务是建立会话行为的分类体系。本文借鉴国外相关领域内所建立的"会话行为"体系,建立汉语问句会话行为分类模型,为中文自然语言处理系统服务。本文详细说明了分类目的、标准和模型,描述了相关的语言特征,同时讨论和解释了其可靠性。

The term ＂dialogue act＂ refers to the speaker＇s intention in conversation. Over the last decade, there has been a growing interest in developing automatic systems for understanding spontaneous conversation. The task of building a typology for dialogue acts, therefore, has been recognized as one of the most essential stepping stones. In line with the above studies, the present paper proposes a typology for dialogue acts in news interviews, which aims to establish a frame;york of dialogue acts for natural language processing. We start our work by reviewing the concept of dialogue act and four existing tagsets. Then, based on the close study of spontaneous conversations, a new typology of dialogue acts is proposed. Meanwhile, the distinctive linguistic features of these dialogue acts are described and illustrated with the goal of examining what kinds of information are useful for automatic dialogue act tagging. The taxonomy adopted in the present scheme follows the general principles of other typologies but differs in the identification of some specific dialogue acts. Finally, the potential applications, as well as the reliability of the tagsets, are discussed.