Answer Span Extraction on CoQA (val)
63.65EMOracle
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| Oracle2023.05 | 63.65 | 74.08 | 58.1 | |
| SG-CQGBackbone=T5-base, Classifier=RoBERTa-large, Beam size=4, Batch size=42023.05 | 42.89 | 63.48 | 68.52 | |
| SG-CQG + w/o WTABackbone=T5-base, Ablation=without Winner-Takes-All2023.05 | 38.89 | 56.17 | 29.21 | |
| SG-CQG + w/o QTCBackbone=T5-base, Ablation=without Question Type Classifier2023.05 | 37.43 | 56.83 | 68.06 | |
| CoHS-CQG2023.05 | 35.14 | 52.08 | 30.02 | |
| T52023.05 | 34.29 | 48.67 | 23.33 | |
| BART2023.05 | 18.64 | 38.23 | 8.62 | |
| SG-CQG + w/o RFBackbone=T5-base, Ablation=without Relevance Filter2023.05 | 18.14 | 22.85 | 100 | |
| GPT-22023.05 | 17.28 | 30.22 | 34.5 |