Machine Translation on WMT20 JA-EN (test)
21.5BLEUAT big
Evaluation results
| Method | Links | |
|---|---|---|
| AT bigReversible=false, Model size=big2021.05 | 21.5 | |
| REDER bigReversible=true, Model size=big, Beam search size (b)=20, AT-reranking=true2021.05 | 20.7 | |
| DOCmT5-25-LargeEvaluation Protocol=Without finetuning2021.12 | 19.17 | |
| DOCmT5-25Evaluation Protocol=Without finetuning2021.12 | 16.93 | |
| DOCmT5-5-LargeEvaluation Protocol=Without finetuning2021.12 | 1.16 | |
| mT5 w. DocNMTEvaluation Protocol=Without finetuning2021.12 | 0.92 | |
| DOCmT5-5Evaluation Protocol=Without finetuning2021.12 | 0.47 | |
| mT5 w. DocTLMEvaluation Protocol=Without finetuning2021.12 | 0.22 |