Conversational Question Answering on CoQA without human rewrites v1.0 (test)
83.4Overall F1EXCORD (RoBERTa)
Evaluation results
| Method | Links | ||||||
|---|---|---|---|---|---|---|---|
| EXCORD (RoBERTa)QA model=RoBERTa, Approach=Ours2021.06 | 83.4 | 84.4 | 81.2 | 79.8 | 84.6 | 87 | |
| RoBERTa (End-to-End)QA model=RoBERTa, Approach=End-to-End2021.06 | 82.8 | 82.5 | 80.2 | 80.1 | 84.3 | 87 | |
| RoBERTa (Pipeline)QA model=RoBERTa, Approach=Pipeline2021.06 | 81.1 | 81.9 | 78.2 | 78.3 | 82.4 | 85.2 | |
| EXCORD (BERT)QA model=BERT, Approach=Ours2021.06 | 78.8 | 78.2 | 75.8 | 75.5 | 81.3 | 83.2 | |
| BERT (End-to-End)QA model=BERT, Approach=End-to-End2021.06 | 78.3 | 73.9 | 73.9 | 76.4 | 80.6 | 82.7 | |
| BERT (Pipeline)QA model=BERT, Approach=Pipeline2021.06 | 76.1 | 73.7 | 73.2 | 74.1 | 78 | 79.6 |