Machine Translation Evaluation on WMT17 (test)
0.653Kendall TauParaBLEU
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| ParaBLEUbackbone=RoBERTa-large, size=large, layers=24, hidden_units=1024, heads=16, pretraining_epochs=42021.07 | 0.653 | 0.843 | |
| BLEURT-largetype=learned neural metric2021.07 | 0.625 | 0.818 | |
| ParaBLEUbackbone=RoBERTa-base, size=base, layers=12, hidden_units=768, heads=12, pretraining_epochs=42021.07 | 0.589 | 0.785 | |
| BERTScorebackbone=DeBERTa-large, type=non-learned neural metric2021.07 | 0.58 | 0.773 | |
| BERTScorebackbone=RoBERTa-large, type=non-learned neural metric2021.07 | 0.567 | 0.759 | |
| BERTScorebackbone=T5-large, type=non-learned neural metric2021.07 | 0.536 | 0.738 | |
| chrF++type=character-level n-gram-matching2021.07 | 0.396 | 0.578 | |
| ROUGEtype=n-gram-based2021.07 | 0.354 | 0.518 | |
| TERtype=edit-distance-based2021.07 | 0.352 | 0.475 | |
| MoverScoretype=non-learned neural metric, algorithm=optimal transport2021.07 | 0.322 | 0.454 | |
| METEORtype=n-gram-based2021.07 | 0.301 | 0.443 | |
| BLEUtype=n-gram-based2021.07 | 0.292 | 0.423 |