Tree Representations in Transition System for RST Parsing

Proceedings of the 28th International Conference on Computational Linguistics, pages 6746–6751 Barcelona, Spain (Online), December 8-13, 2020 6746 Tree Representations in Transition System for RST Parsing Jinfen Li College of Arts and Sciences, Syracuse University jli284@syr.edu Lu Xiao School of Information Studies, Syracuse University lxiao04@syr.edu Abstract The transition-based systems in the past studies propose a series of actions to build a right-heavy binarized tree for RST parsing. However, the nodes of the binary-nuclear relations (e.g., Contrast) have the same nuclear type with those of the multi-nuclear relations (e.g., Joint) in the binary tree structure. In addition, the reduce action only construct binary trees instead of multi-branch trees, which is the original RST tree structure. In our paper, we design a new nuclear type for the multi-nuclear relations, and a new action to construct a multi-branch tree. We enrich the feature set by extracting additional refined dependency feature of texts from the Bi-Affine model (Dozat and Manning, 2016). We also compare the performance of two approaches for RST parsing in the transition-based system: a joint action of reduce-shift and nuclear type (i.e., Reduce-SN) vs a separate one that applies Reduce action first and then assigns nuclear type. We find that the new devised nuclear type and action are more capable of capturing the multi-nuclear relation and the joint action is more suitable than the separate one. Our multi-branch tree structure obtains the state-of-the-art performance for all the 18 coarse relations. 1 Introduction It is expected that a document consists of text units that are logically connected within the context through discourse relations. Text-level discoursing parsing plays a significant role in many NLP downstream task, such as question-answering (Jansen et al., 2014), sentiment analysis (Mukherjee and Bhattacharyya, 2012) and abstractive summarization (Koto et al., 2019). The Rhetorical Structure Theory (RST)(Mann and Thompson, 1988) is one of the widely explored text-level discourse parsing theory, which parses a document into a hierarchical discourse tree. The discourse tree is built over its smallest parsing unit, namely, elementary discourse unit (EDU), placed in the leaf nodes. The tree’s non-terminal nodes bear the information of span, nuclearity and relation. The span represents the cover range of a sequence of EDUs; while the nuclearity status is the semantic role in a relation (i.e., nucleus or satellite). A text tagged by nucleus is more essential than the satellite. Conventionally, there are three nuclearity types including Nucleus-Satellite (NS), Satellite-Nucleus (SN) and Nucleus-Nucleus (NN). As for relation, there are mainly two types of relations: the mono-nuclear relation, such as “Attribution” or “Summary”, with the the nuclearity type being either “NS” or “SN”; and the multi-nuclear relation, such as “Contrast” or “Same-Unit”, with the nuclearity types of “NN”. Among all the parsing systems dedicated to RST parsing, the transition-based systems obtain com- parative performance with the CKY-like system. The transition-based algorithms are first explored in (Marcu, 1999), which adopt a sequence of shift-reduce action to construct rhetorical structures of texts. The study of Wang et al. (2017) achieves the best performance among …

Content truncated to 500 words