Constructing Chinese Macro Discourse Tree via Multiple Views and Word Pair Similarity

Yi Zhou, Xiaomin Chu, Peifeng Li, Qiaoming Zhu

Abstract: Macro-discourse structure recognition is an important task in macro-discourse analysis. At present, the research on macro-discourse analysis mostly uses the manual features (e.g., the position features), and ignores the semantic information in topic level. In this paper, we first propose a multi-view neural network to construct Chinese macro discourse trees from three views, i.e., the word view, the context view and the topic view. Besides, we propose a novel word-pair similarity mechanism to capture the interaction among the discourse units and the topic. The experimental results on MCDTB, a Chinese discourse corpus, show that our model outperforms the baseline significantly.