Top-Down RST Parsing Utilizing Granularity Levels in Documents

Abstract

Some downstream NLP tasks exploit discourse dependency trees converted from RST trees. To obtain better discourse dependency trees, we need to improve the accuracy of RST trees at the upper parts of the structures. Thus, we propose a novel neural top-down RST parsing method. Then, we exploit three levels of granularity in a document, paragraphs, sentences and Elementary Discourse Units (EDUs), to parse a document accurately and efficiently. The parsing is done in a top-down manner for each granularity level, by recursively splitting a larger text span into two smaller ones while predicting nuclearity and relation labels for the divided spans. The results on the RST-DT corpus show that our method achieved the state-of-the-art results, 87.0 unlabeled span score, 74.6 nuclearity labeled span score, and the comparable result with the state-of-the-art, 60.0 relation labeled span score. Furthermore, discourse dependency trees converted from our RST trees also achieved the state-of-the-art results, 64.9 unlabeled attachment score and 48.5 labeled attachment score.

RST Discourse Parser for Russian: An Experimental Study of Deep Learning Models

“…In [16], the segmenter and the sentence-level parser are trained jointly as parts of the unified encoder-decoder architecture, achieving superior results in the parsing performance, as well as in the parsing speed compared to previous end-to-end sentence-level bottom-up discourse parsers, namely SPADE [26] and DCRF [13]. Kobayashi et al [15] has recently proposed a top-down method that takes into account granularity levels of spans, namely document, paragraph, and sentence. The authors show that the granularity levels are important features for discourse parsing and achieve 60% micro F score on the RST-DT corpus.…”

Chinese Paragraph-level Discourse Parsing with Global Backward and Local Reverse Reading

“…In English, RST-DT (Carlson et al, 2003) is one of the popular discourse corpora (Subba and Di Eugenio, 2009;Zeldes, 2017;Kolhatkar and Taboada, 2017), which annotates the discourse structure, nuclearity, and relationship of a document. Most previous studies have focused on complete discourse parsing and can be mainly categorized into the shift-reduce algorithm (Ji and Eisenstein, 2014;Wang et al, 2017;Yu et al, 2018;Jia et al, 2018), the probabilistic CKY-like algorithm (Joty et al, 2013;Li et al, 2014a;Li et al, 2016), and the bottom-up algorithm (Hernault et al, 2010;Feng and Hirst, 2014;Kobayashi et al, 2019;Kobayashi et al, 2020). Recently, the generative algorithm (Mabona et al, 2019) and the top-down algorithm (Liu et al, 2019;Lin et al, 2019) tried out discourse parsing.…”

“…In RST-style discourse parsing, the parser first identifies whether there is a rhetorical relationship between discourse units to construct a naked tree and then recognizes the nuclearity and relation labels for each relationship, as shown in Figure 1. According to the granularity of the leaf nodes, the discourse tree is divided into three levels: clause level, sentence level and paragraph level (Kobayashi et al, 2020). This paper focuses on constructing paragraph-level Chinese discourse trees where the leaf node is a paragraph.…”

Improving Neural RST Parsing Model with Silver Agreement Subtrees

“…In most cases, RST parsers have been developed on the basis of supervised learning algorithms (Wang et al, 2017b;Yu et al, 2018;Kobayashi et al, 2020;Lin et al, 2019;Zhang et al, 2020), which require a high-quality annotated corpus of sufficient size. Generally, they train the following three components of the RST parsing: (1) structure prediction by splitting a text span consisting of contiguous EDUs into two smaller ones or merging two adjacent spans into a larger one, (2) nuclearity status prediction for two adjacent spans by solving a 3-class classification problem, and (3) relation label prediction for two adjacent spans by solving an 18-class classification problem (see Section 3.3 for details).…”