Thanh-Tung Nguyen, Xuan-Phi Nguyen, Shafiq Joty, XiaoLi Li
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.
| Task | Dataset | Metric | Value | Model |
|---|---|---|---|---|
| Discourse Parsing | RST-DT | RST-Parseval (Nuclearity) | 76 | End-to-end Top-down (XLNet) |
| Discourse Parsing | RST-DT | RST-Parseval (Relation) | 61.8 | End-to-end Top-down (XLNet) |
| Discourse Parsing | RST-DT | RST-Parseval (Span) | 87.6 | End-to-end Top-down (XLNet) |
| Discourse Parsing | RST-DT | Standard Parseval (Full) | 50.2 | End-to-end Top-down (XLNet) |
| Discourse Parsing | RST-DT | Standard Parseval (Nuclearity) | 64.3 | End-to-end Top-down (XLNet) |
| Discourse Parsing | RST-DT | Standard Parseval (Relation) | 51.6 | End-to-end Top-down (XLNet) |
| Discourse Parsing | RST-DT | Standard Parseval (Span) | 74.3 | End-to-end Top-down (XLNet) |
| Discourse Parsing | RST-DT | Standard Parseval (Full) | 46.8 | End-to-end Top-down (Glove) |
| Discourse Parsing | RST-DT | Standard Parseval (Nuclearity) | 59.6 | End-to-end Top-down (Glove) |
| Discourse Parsing | RST-DT | Standard Parseval (Relation) | 47.7 | End-to-end Top-down (Glove) |
| Discourse Parsing | RST-DT | Standard Parseval (Span) | 71.1 | End-to-end Top-down (Glove) |
| Discourse Parsing | RST-DT | Standard Parseval (Full) | 46.6 | Nguyen et al. (2021) |
| Discourse Parsing | RST-DT | Standard Parseval (Nuclearity) | 59.1 | Nguyen et al. (2021) |
| Discourse Parsing | RST-DT | Standard Parseval (Relation) | 47.8 | Nguyen et al. (2021) |
| Discourse Parsing | RST-DT | Standard Parseval (Span) | 68.4 | Nguyen et al. (2021) |