Joint Modeling of Structure Identification and Nuclearity Recognition in Macro Chinese Discourse Treebank

Proceedings of the 27th International Conference on Computational Linguistics, pages 536–546 Santa Fe, New Mexico, USA, August 20-26, 2018. 536 Joint Modeling of Structure Identification and Nuclearity Recognition in Macro Chinese Discourse Treebank Xiaomin Chu1, Feng Jiang1, Yi Zhou1, Guodong Zhou1, Qiaoming Zhu1,2,† 1School of Computer Science and Technology, Soochow University, China 2Institute of Artificial Intelligence, Soochow University, China {xmchu,fjiang,yzhou}@stu.suda.edu.cn; {gdzhou,qmzhu}@suda.edu.cn Abstract Discourse parsing is a challenging task and plays a critical role in discourse analysis. This paper focus on the macro level discourse structure analysis, which has been less studied in the previous researches. We explore a macro discourse structure presentation schema to present the macro level discourse structure, and propose a corresponding corpus, named Macro Chinese Discourse Treebank. On these bases, we concentrate on two tasks of macro discourse structure analysis, including structure identification and nuclearity recognition. In order to reduce the error trans- mission between the associated tasks, we adopt a joint model of the two tasks, and an Integer Linear Programming approach is proposed to achieve global optimization with various kinds of constraints. 1 Introduction A typical document is usually organized in a coherent way that each discourse unit is relevant to its con- text and plays a role in the entire semantics. Discourse structure analysis not only helps to understand the discourse structure and semantics, but also can benefit variety of downstream applications includ- ing question answering (Sadek and Meziane, 2016), machine translation (Guzm´an et al., 2014), text summarization (Ferreira et al., 2014; Cohan and Goharian, 2017), and so forth. There exist two hierarchical levels of discourse structures: micro level and macro level. The micro level structure refers to the structure and relation among the discourse units in a sentence, or consecutive sentences, or sentences groups. The macro level structure refers to the structure and relation among paragraphs, or chapters, or discourses. Corresponding to related research based on Rhetorical Structure Theory Discourse Treebank (RST-DT), the micro level is similar to the sentence-level discourse parsing, and the macro level is similar to the document-level discourse parsing. To make a clearer explanation of the macro discourse structure, take the chtb 0019 as an example, which is a typical news article from Chinese Treebank 8.0 (Xue et al., 2013). The macro discourse structure of this article is shown as Figure 1. There are five paragraphs (P1, P2, P3, P4 and P5) in the news “Significant achievements in the construction of Ningbo Bonded Area”. The paragraphs are connected by discourse relations (Elaboration, Background, Joint). In this article, paragraph P1 points out the theme of the overall article. Depending on the direction of arrows from the root node to the leaf node in the discourse structure tree, the most important part can be quickly located. Limited to the length of this paper, the full discourse text of this example is not included, please refer to the corpus. The main contents of the five paragraphs respectively are: P1) Ningbo Bonded Area achieved fruitful results after three years of construction; P2) the …

Content truncated to 500 words