RST Parsing from Scratch

Proceedings of the 2021 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies, pages 1613–1625 June 6–11, 2021. ©2021 Association for Computational Linguistics 1613 RST Parsing from Scratch Thanh-Tung Nguyen†¶, Xuan-Phi Nguyen†¶, Shafiq Joty¶§, Xiaoli Li†¶ ¶Nanyang Technological University §Salesforce Research Asia †Institute for Infocomm Research, A-STAR Singapore {ng0155ng@e.;nguyenxu002@e.;srjoty@}ntu.edu.sg xlli@i2r.a-star.edu.sg Abstract We introduce a novel top-down end-to-end formulation of document level discourse pars- ing in the Rhetorical Structure Theory (RST) framework. In this formulation, we consider discourse parsing as a sequence of splitting de- cisions at token boundaries and use a seq2seq network to model the splitting decisions. Our framework facilitates discourse parsing from scratch without requiring discourse segmen- tation as a prerequisite; rather, it yields seg- mentation as part of the parsing process. Our unified parsing model adopts a beam search to decode the best tree structure by searching through a space of high scoring trees. With extensive experiments on the standard English RST discourse treebank, we demonstrate that our parser outperforms existing methods by a good margin in both end-to-end parsing and parsing with gold segmentation. More impor- tantly, it does so without using any handcrafted features, making it faster and easily adaptable to new languages and domains. 1 Introduction In a document, the clauses, sentences and para- graphs are logically connected together to form a coherent discourse. The goal of discourse parsing is to uncover this underlying coherence structure, which has been shown to benefit numerous NLP applications including text classification (Ji and Smith, 2017), summarization (Gerani et al., 2014), sentiment analysis (Bhatia et al., 2015), machine translation evaluation (Joty et al., 2017) and con- versational machine reading (Gao et al., 2020). Rhetorical Structure Theory or RST (Mann and Thompson, 1988), one of the most influential the- ories of discourse, postulates a hierarchical dis- course structure called discourse tree (DT). The leaves of a DT are clause-like units, known as ele- mentary discourse units (EDUs). Adjacent EDUs and higher-order spans are connected hierarchically through coherence relations (e.g., Contrast, Expla- nation). Spans connected through a relation are categorized based on their relative importance — nucleus being the main part, with satellite being the subordinate one. Fig. 1 exemplifies a DT span- ning over two sentences and six EDUs. Finding discourse structure generally requires breaking the text into EDUs (discourse segmentation) and link- ing the EDUs into a DT (discourse parsing). Discourse parsers can be singled out by whether they apply a bottom-up or top-down procedure. Bottom-up parsers include transition-based mod- els (Feng and Hirst, 2014; Ji and Eisenstein, 2014; Braud et al., 2017; Wang et al., 2017) or globally optimized chart parsing models (Soricut and Marcu, 2003; Joty et al., 2013, 2015). The former con- structs a DT by a sequence of shift and reduce de- cisions, and can parse a text in asymptotic running time that is linear in number of EDUs. However, the transition-based parsers make greedy local deci- sions at each decoding step, which could propagate errors into future …

Content truncated to 500 words