Machine Translation on WMT newstest 2019
43.43BLEUCYCLE (REV)
Evaluation results
| Method | Links | |
|---|---|---|
| CYCLE (REV)#Params=242M, Training Data=Genuine2021.04 | 43.43 | |
| CYCLE#Params=242M, Training Data=Genuine2021.04 | 43.19 | |
| SEQUENCE#Params=242M, Training Data=Genuine2021.04 | 43.12 | |
| Universal#Params=249M, Training Data=Genuine2021.04 | 43.11 | |
| Vanilla#Params=242M, Training Data=Genuine2021.04 | 42.8 | |
| Kiyono et al. (2020)#Params=514M, Training Data=Synthetic2021.04 | 42.7 | |
| CYCLE (REV)#Params=343M, Training Data=Synthetic2021.04 | 42.18 | |
| Byte-nCFModel Type=ONE-HOT, Adaptation Mode=Zero-shot cross-domain adaptation2022.05 | 21.5 | |
| Byte-nCFModel Type=DENSE, Adaptation Mode=Zero-shot cross-domain adaptation2022.05 | 21.3 | |
| ByteModel Type=DENSE, Adaptation Mode=Zero-shot cross-domain adaptation2022.05 | 21.2 | |
| ByteModel Type=ONE-HOT, Adaptation Mode=Zero-shot cross-domain adaptation2022.05 | 21.1 | |
| SubwordModel Type=DENSE, Adaptation Mode=Zero-shot cross-domain adaptation2022.05 | 17.6 |