Detection, Disambiguation and Argument Identification of Discourse Connectives in Chinese Discourse Parsing

Proceedings of COLING 2016, the 26th International Conference on Computational Linguistics: Technical Papers, pages 1891–1902, Osaka, Japan, December 11-17 2016. Detection, Disambiguation and Argument Identification of Discourse Connectives in Chinese Discourse Parsing Yong-Siang Shih and Hsin-Hsi Chen Department of Computer Science and Information Engineering National Taiwan University, Taipei, Taiwan {r02922036,hhchen}@ntu.edu.tw Abstract In this paper, we investigate four important issues together for explicit discourse relation label- ling in Chinese texts: (1) discourse connective extraction, (2) linking ambiguity resolution, (3) relation type disambiguation, and (4) argument boundary identification. In a pipelined Chinese discourse parser, we identify potential connective candidates by string matching, eliminate non-discourse usages from them with a binary classifier, resolve linking ambiguities among connective components by ranking, disambiguate relation types by a multiway classifier, and determine the argument boundaries by conditional random fields. The experiments on Chinese Discourse Treebank show that the F1 scores of 0.7506, 0.7693, 0.7458, and 0.3134 are achieved for discourse usage disambiguation, linking disambiguation, relation type disambigu- ation, and argument boundary identification, respectively, in a pipelined Chinese discourse parser. 1 Introduction Discourse relations represent how discourse units logically connect with each other. A discourse con- nective explicitly signals the presence of a discourse relation, and therefore it is an important clue for discourse analysis. There are several challenges in Chinese discourse parsing. Firstly, there are more discourse connectives in Chinese than in English and their parts of speech have more varieties (Huang et al., 2014). Therefore, it is likely to encounter words that have the same surface forms as real connectives but do not function as discourse connectives. Secondly, many Chinese connectives are parallel connectives that have multiple discontinuous components (Zhou and Xue, 2012). Each connective can be composed of one or more connective components. For example, ''雖然-但'' (although-but) consists of two connective components: ''雖然'' (although) and ''但'' (but). When multiple connectives are present in a paragraph, their components often link with each other in multiple possible ways. (S1) is an example. There are five possible con- nective candidates, including (1) ''除了...還'' (in addition to ... also), (2) ''還...也'' (also ... also), (3) ''除 了'' (in addition to), (4) ''還'' (also), and (5) ''也'' (also). Figure 1 illustrates the ambiguous linking. On- ly candidates (1) and (5) are correct. Moreover, when spurious component candidates exist in a para- graph, they form many more spurious connective candidates by linking together in different ways. Finding the correct linking between correct components is useful for discourse analysis because they provide clues to determine the positions of the relations. (S1) 除了投資環境優越，還在於這些企業所具有的產品優勢。(...) 對上海的產業優化也有很 大的帶動作用。(In addition to superior investment environment, it's also due to the product ad- vantages possessed by the enterprise. (...) It also has great effect in promoting the optimization of in- dustries in Shanghai.) This work is licensed under a Creative Commons Attribution 4.0 International Licence. Licence details: http://creativecommons.org/licenses/by/4.0/ 1891 Figure 1: Ambiguous linking between connective components. Thirdly, the sentence structures in Chinese texts are not clearly defined. Thus it is more challenging to detect the arguments for a …

Content truncated to 500 words