基于神经网络的汉语口语言语行为分析

DIALOG ACT ANALYSIS OF SPOKEN CHINESE BASED ON NEURAL NETWORKS

为正确理解口语对话、准确把握话者意图，除必要的语法和语义分析外，口语系统还需进行语用层面上的言语行为分析．文中提出一种基于精简循环网络的、综合使用语段级的微结构信息和语篇级的宏结构特征的汉语口语言语行为分析方法．针对会面安排领域口语语料库训练和测试。

In order to understand a spoken dialog properly and grasp the speakers' intention, a spoken language system must perform dialog act analysis on the pragmatic level in addition to syntactic and semantic analysis. This paper proposes a dialog act analysis method based on simple recurrent network, which makes use of both the micro structure information and the macro structure feature. Tested by a spoken language corpus for meeting schedule, the overall performance is satisfying.