RST Parsing from Scratch

We introduce a novel top-down end-to-end formulation of document-level discourse parsing in the Rhetorical Structure Theory (RST) framework. In this formulation, we consider discourse parsing as a sequence of splitting decisions at token boundaries and use a seq2seq network to model the splitting decisions. Our framework facilitates discourse parsing from scratch without requiring discourse segmentation as a prerequisite; rather, it yields segmentation 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 importantly, it does so without using any handcrafted features, making it faster and easily adaptable to new languages and domains.

PDF Abstract NAACL 2021 PDF NAACL 2021 Abstract

Datasets


Results from the Paper


Task Dataset Model Metric Name Metric Value Global Rank Result Benchmark
Discourse Parsing RST-DT End-to-end Top-down (XLNet) RST-Parseval (Span) 87.6 # 1
RST-Parseval (Nuclearity) 76.0 # 1
RST-Parseval (Relation) 61.8 # 3
Standard Parseval (Span) 74.3 # 6
Standard Parseval (Nuclearity) 64.3 # 9
Standard Parseval (Relation) 51.6 # 9
Standard Parseval (Full) 50.2 # 10
Discourse Parsing RST-DT End-to-end Top-down (Glove) Standard Parseval (Span) 71.1 # 12
Standard Parseval (Nuclearity) 59.6 # 16
Standard Parseval (Relation) 47.7 # 15
Standard Parseval (Full) 46.8 # 14

Methods