Text Generation Evaluation Correlation on WebText (test)
0.643Perplexity (PPL)Bradley-Terry Score (Interesting)
Evaluation results
| Method | Links | ||||||
|---|---|---|---|---|---|---|---|
| Bradley-Terry Score (Interesting)Human Evaluation Criterion=Interesting2021.02 | 0.643 | 0.524 | -0.143 | 52.4 | 40.5 | 0.81 | |
| Bradley-Terry Score (Sensible)Human Evaluation Criterion=Sensible2021.02 | 0.738 | 0.69 | -0.071 | 59.5 | 52.4 | 0.857 | |
| Bradley-Terry Score (Human-like)Human Evaluation Criterion=Human-like2021.02 | 0.81 | 0.833 | -0.167 | 73.8 | 59.5 | 0.952 |