A Simple and Strong Baseline for End-to-End Neural RST-style Discourse Parsing
Naoki Kobayashi, Tsutomu Hirao, Hidetaka Kamigaito, Manabu Okumura, Masaaki Nagata
Code Available — Be the first to reproduce this paper.
ReproduceCode
- github.com/nttcslab-nlp/rstparser_emnlp22OfficialIn paperpytorch★ 23
Abstract
To promote and further develop RST-style discourse parsing models, we need a strong baseline that can be regarded as a reference for reporting reliable experimental results. This paper explores a strong baseline by integrating existing simple parsing strategies, top-down and bottom-up, with various transformer-based pre-trained language models. The experimental results obtained from two benchmark datasets demonstrate that the parsing performance strongly relies on the pretrained language models rather than the parsing strategies. In particular, the bottom-up parser achieves large performance gains compared to the current best parser when employing DeBERTa. We further reveal that language models with a span-masking scheme especially boost the parsing performance through our analysis within intra- and multi-sentential parsing, and nuclearity prediction.
Tasks
Benchmark Results
| Dataset | Model | Metric | Claimed | Verified | Status |
|---|---|---|---|---|---|
| Instructional-DT (Instr-DT) | Top-down (BERT) | Standard Parseval (Nuclearity) | 44.6 | — | Unverified |
| Instructional-DT (Instr-DT) | Bottom-up (DeBERTa) | Standard Parseval (Nuclearity) | 60 | — | Unverified |
| Instructional-DT (Instr-DT) | Top-down (DeBERTa) | Standard Parseval (Nuclearity) | 57.9 | — | Unverified |
| Instructional-DT (Instr-DT) | Bottom-up (XLNet) | Standard Parseval (Nuclearity) | 56.4 | — | Unverified |
| Instructional-DT (Instr-DT) | Top-down (RoBERTa) | Standard Parseval (Nuclearity) | 56.1 | — | Unverified |
| Instructional-DT (Instr-DT) | Bottom-up (RoBERTa) | Standard Parseval (Nuclearity) | 55.5 | — | Unverified |
| Instructional-DT (Instr-DT) | Top-down (XLNet) | Standard Parseval (Nuclearity) | 55.2 | — | Unverified |
| Instructional-DT (Instr-DT) | Top-down (SpanBERT) | Standard Parseval (Nuclearity) | 54.5 | — | Unverified |
| Instructional-DT (Instr-DT) | Bottom-up (SpanBERT) | Standard Parseval (Nuclearity) | 53.8 | — | Unverified |
| Instructional-DT (Instr-DT) | Bottom-up (BERT) | Standard Parseval (Nuclearity) | 46.3 | — | Unverified |
| RST-DT | Top-down (DeBERTa) | Standard Parseval (Full) | 54.4 | — | Unverified |
| RST-DT | Bottom-up (XLNet) | Standard Parseval (Full) | 54.2 | — | Unverified |
| RST-DT | Top-down (RoBERTa) | Standard Parseval (Full) | 53.8 | — | Unverified |
| RST-DT | Bottom-up (RoBERTa) | Standard Parseval (Full) | 53.7 | — | Unverified |
| RST-DT | Bottom-up (SpanBERT) | Standard Parseval (Full) | 52.7 | — | Unverified |
| RST-DT | Top-down (SpanBERT) | Standard Parseval (Full) | 52.2 | — | Unverified |
| RST-DT | Top-down (BERT) | Standard Parseval (Full) | 46.6 | — | Unverified |
| RST-DT | Bottom-up (BERT) | Standard Parseval (Full) | 46 | — | Unverified |
| RST-DT | Top-down (XLNet) | Standard Parseval (Full) | 54.8 | — | Unverified |