Fine-Grained Chinese Discourse Relation Labelling

Fine-Grained Chinese Discourse Relation Labelling Huan-Yuan Chen, Wan-Shan Liao, Hen-Hsen Huang, and Hsin-Hsi Chen Department of Computer Science and Information Engineering, National Taiwan University No. 1, Sec. 4, Roosevelt Road, Taipei, 10617 Taiwan E-mail: b00902057@ntu.edu.tw, wsliao@nlg.csie.ntu.edu.tw, hhhuang@nlg.csie.ntu.edu.tw, hhchen@ntu.edu.tw Abstract This paper explores several aspects together for a fine-grained Chinese discourse analysis. We deal with the issues of ambiguous discourse markers, ambiguous marker linkings, and more than one discourse marker. A universal feature representation is proposed. The pair-once postulation, cross-discourse-unit-first rule and word-pair-marker-first rule select a set of discourse markers from ambiguous linkings. Marker-Sum feature considers total contribution of markers and Marker-Preference feature captures the probability distribution of discourse functions of a representative marker by using preference rule. The HIT Chinese discourse relation treebank (HIT-CDTB) is used to evaluate the proposed models. The 25-way classifier achieves 0.57 micro-averaged F-score. Keywords: Discourse analysis; Discourse relation labelling; HIT Chinese discourse relation treebank. 1. Introduction Discourse relation labelling aims at predicting the most proper discourse relation between two discourse units such as clauses, sentences, and groups of sentences. The labelling task can be done at the intra-sentential and the inter-sentential levels depending on the analysis units. Several schemes have been proposed to define discourse relations to be analyzed. In the PDTB framework (Prasad et al., 2008), three levels of sense hierarchy are utilized. The four classes on the top level are Temporal, Contingency, Comparison, and Expansion. In this paper, we present Chinese discourse analysis on fine-grained discourse relations at intra-sentential level. HIT Chinese discourse relation treebank (Zhang et al., 2014) is used to investigate some specific issues. A discourse marker in Chinese may belong to various lexical categories including conjunction, adverb, noun, preposition, and verb. It may be a single word such as “但” (but) or a word pair such as “雖然...但” (although … but). A discourse marker may have more than one discourse function. Two discourse units may not always be connected with an explicit marker. Besides, a sentence does not always contain only one discourse marker. More than one discourse marker results in an ambiguous-linking issue. This paper is organized as follows. Section 2 surveys the past works on discourse relation labelling. Section 3 introduces the discourse treebank in this study. Section 4 presents the feature representation and relation labelling algorithm. Section 5 shows the experiments. Section 6 concludes the remarks. 2. Related Work Structure Theory Discourse Treebank (RST-DT) (Carlson et al., 2002) and the Penn Discourse TreeBank (PDTB) (Prasad et al., 2008) are two important resources to facilitate the researches of discourse relation recognition. Because most Chinese discourse corpora follow the PDTB scheme, this section surveys the related work from this direction. Pitler and Nenkova (2009) deal with two types of ambiguity of discourse connectives in English: non-discourse vs. discourse usage and unique vs. more discourse functions. They achieve 0.9419 F-score for the first issue and 0.9415 accuracy for the 4-way sense classification of explicits. Hernault et al. (2010, 2011) improve classifi-cation accuracy for the infrequent English discourse relation types. Because …

Content truncated to 500 words