关于人机对话系统的思考

[摘要] 提出了一系列非常重要、影响人机对话质量的问题，包括：如何面向自然语言理解（NLU）构建对话任务分析、深度推理，如何利用语言学尤其是互动语言学研究成果构建对话管理（DM），如何有效建模人机对话中不同任务间的关联约束来发展联合模型等。认为尽管人机对话系统的基础模型已取得了长足进步，但如果不能有效地解决上述问题，就不可能获得高质量的人机对话系统，自然语言处理的水平也就难以得到实质性提升。

[关键词] 人机对话系统；NLU；DM；自然语言生成（NLG）

[Abstract] In this paper, a series important problems which have significant influence on the quality of human-computer dialogue are proposed, including how to normalize task analysis and introduce deep reasoning for natural language understanding (NLU), how to build linguistically, especially interact linguistically reasonable dialogue management (DM), and how to develop joint method for modeling correlations between different subtasks in dialogues. There has been lots of progresses on human-computer dialogue, but if the problems mentioned above cannot be solved, it is difficult to achieve high quality dialogue systems and important improvements in natural language processing.

[Keywords] human-computer dialogue system; NLU; DM; natural language generative(NLG)