Discourse Parsing on Molweni (test)
88.1Link (%)Multi-task
Evaluation results
| Method | Links | ||||
|---|---|---|---|---|---|
| Multi-taskBackbone=BERT-wwm2021.10 | 88.1 | 66.9 | — | — | |
| DP-onlyBackbone=BERT-wwm2021.10 | 86.6 | 64.9 | — | — | |
| Multi-taskBackbone=BERT-large2021.10 | 82.1 | 62 | — | — | |
| DP-onlyBackbone=BERT-large2021.10 | 81 | 61.5 | — | — | |
| Deep sequential2021.10 | 78.1 | 54.8 | — | — | |
| Multi-taskBackbone=BERT-base2021.10 | 75.9 | 56.2 | — | — | |
| DP-onlyBackbone=BERT-base2021.10 | 73.9 | 56.1 | — | — | |
| Deep Sequential ParserTraining data=Joint (STAC+Molweni)2021.10 | — | — | 77.4 | 54.3 | |
| Our Proposed Parser w/ language backboneTraining data=Joint (STAC+Molweni), Backbone=RoBERTa-base2021.10 | — | — | 80.2 | 56.9 |