Fajri Koto, Jey Han Lau, Timothy Baldwin
We introduce a top-down approach to discourse parsing that is conceptually simpler than its predecessors (Kobayashi et al., 2020; Zhang et al., 2020). By framing the task as a sequence labelling problem where the goal is to iteratively segment a document into individual discourse units, we are able to eliminate the decoder and reduce the search space for splitting points. We explore both traditional recurrent models and modern pre-trained transformer models for the task, and additionally introduce a novel dynamic oracle for top-down parsing. Based on the Full metric, our proposed LSTM model sets a new state-of-the-art for RST parsing.
| Task | Dataset | Metric | Value | Model |
|---|---|---|---|---|
| Discourse Parsing | RST-DT | Standard Parseval (Full) | 50.3 | LSTM Dynamic |
| Discourse Parsing | RST-DT | Standard Parseval (Nuclearity) | 62.3 | LSTM Dynamic |
| Discourse Parsing | RST-DT | Standard Parseval (Relation) | 51.5 | LSTM Dynamic |
| Discourse Parsing | RST-DT | Standard Parseval (Span) | 73.1 | LSTM Dynamic |
| Discourse Parsing | RST-DT | Standard Parseval (Full) | 49.4 | LSTM Static |
| Discourse Parsing | RST-DT | Standard Parseval (Nuclearity) | 61.7 | LSTM Static |
| Discourse Parsing | RST-DT | Standard Parseval (Relation) | 50.5 | LSTM Static |
| Discourse Parsing | RST-DT | Standard Parseval (Span) | 72.7 | LSTM Static |
| Discourse Parsing | RST-DT | Standard Parseval (Full) | 49.2 | Transformer (dynamic) |
| Discourse Parsing | RST-DT | Standard Parseval (Nuclearity) | 60.1 | Transformer (dynamic) |
| Discourse Parsing | RST-DT | Standard Parseval (Span) | 70.2 | Transformer (dynamic) |
| Discourse Parsing | RST-DT | Standard Parseval (Full) | 49 | Transformer (static) |
| Discourse Parsing | RST-DT | Standard Parseval (Nuclearity) | 59.9 | Transformer (static) |
| Discourse Parsing | RST-DT | Standard Parseval (Relation) | 50.6 | Transformer (static) |
| Discourse Parsing | RST-DT | Standard Parseval (Span) | 70.6 | Transformer (static) |