Dependency-based Discourse Parser for Single-Document Summarization

Proceedings of the 2014 Conference on Empirical Methods in Natural Language Processing (EMNLP), pages 1834–1839, October 25-29, 2014, Doha, Qatar. c⃝2014 Association for Computational Linguistics Dependency-based Discourse Parser for Single-Document Summarization Yasuhisa Yoshida, Jun Suzuki, Tsutomu Hirao, and Masaaki Nagata NTT Communication Science Laboratories, NTT Corporation 2-4 Hikaridai, Seika-cho, Soraku-gun, Kyoto, 619-0237 Japan {yoshida.y,suzuki.jun,hirao.tsutomu,nagata.masaaki}@lab.ntt.co.jp Abstract The current state-of-the-art single- document summarization method gen- erates a summary by solving a Tree Knapsack Problem (TKP), which is the problem of finding the optimal rooted sub- tree of the dependency-based discourse tree (DEP-DT) of a document. We can obtain a gold DEP-DT by transforming a gold Rhetorical Structure Theory-based discourse tree (RST-DT). However, there is still a large difference between the ROUGE scores of a system with a gold DEP-DT and a system with a DEP-DT obtained from an automatically parsed RST-DT. To improve the ROUGE score, we propose a novel discourse parser that directly generates the DEP-DT. The evaluation results showed that the TKP with our parser outperformed that with the state-of-the-art RST-DT parser, and achieved almost equivalent ROUGE scores to the TKP with the gold DEP-DT. 1 Introduction Discourse structures of documents are believed to be highly beneficial for generating informa- tive and coherent summaries. Several discourse- based summarization methods have been devel- oped, such as (Marcu, 1998; Daum´e III and Marcu, 2002; Hirao et al., 2013; Kikuchi et al., 2014). Moreover, the current best ROUGE score for the summarization benchmark data of the RST- discourse Treebank (Carlson et al., 2002) has been provided by (Hirao et al., 2013), whose method also utilizes discourse trees. Thus, the discourse- based summarization approach is one promising way to obtain high-quality summaries. One possible weakness of discourse-based sum- marization techniques is that they rely greatly on the accuracy of the discourse parser they use. For example, the above discourse-based summa- rization methods utilize discourse trees based on the Rhetorical Structure Theory (RST) (Mann and Thompson, 1988) for their discourse information. Unfortunately, the current state-of-the-art RST parser, as described in (Hernault et al., 2010), is insufficient as an off-the-shelf discourse parser. In fact, there is empirical evidence that the qual- ity (i.e., ROUGE score) of summaries from auto- parsed discourse trees is significantly degraded compared with those generated from gold dis- course trees (Marcu, 1998; Hirao et al., 2013). From this background, the goal of this paper is to develop an appropriate discourse parser for discourse-based summarization. We first focus on one of the best discourse-based single document summarization methods as proposed in (Hirao et al., 2013). Their method formulates a single doc- ument summarization problem as a Tree Knap- sack Problem (TKP) over a dependency-based dis- course tree (DEP-DT). In their method, DEP-DTs are automatically transformed from (auto-parsed) RST-discourse trees (RST-DTs) by heuristic rules. Instead, we develop a DEP-DT parser, that di- rectly provides DEP-DTs for their state-of-the-art discourse-based summarization method. We show that summaries generated by our parser improve the ROUGE scores to almost the same level as those generated by gold DEP-DTs. …

Content truncated to 500 words