Machine Translation on IWSLT English-Vietnamese 2015 (tst2013)
29.12BLEUSeq2Seq + SACT
Evaluation results
| Method | Links | ||
|---|---|---|---|
| Seq2Seq + SACTSelf-Adaptive Control of Temperature=true2018.08 | 29.12 | — | |
| SAWR2019.05 | 29.09 | 0.8 | |
| Tree-Linearization2019.05 | 28.93 | 0.64 | |
| Tree-RNN2019.05 | 28.51 | 0.22 | |
| Baseline2019.05 | 28.29 | — | |
| NPMT+LMdecoding_strategy=Beam Search2017.06 | 28.07 | — | |
| NPMT2018.08 | 27.69 | — | |
| NPMTdecoding_strategy=Beam Search2017.06 | 27.69 | — | |
| CE+BTLoss function=Cross-Entropy, Data augmentation=BT2021.05 | 27.41 | — | |
| CE+OURSLoss function=Cross-Entropy, Data augmentation=OURS2021.05 | 27.12 | — | |
| CELoss function=Cross-Entropy, Data augmentation=None2021.05 | 27.09 | — | |
| Seq2Seqattention-based=true2018.08 | 26.93 | — | |
| NPMTdecoding_strategy=Greedy2017.06 | 26.91 | — | |
| WD+MLMLoss function=L2 Wasserstein distance, Data augmentation=MLM2021.05 | 26.88 | — | |
| WD+BTLoss function=L2 Wasserstein distance, Data augmentation=BT2021.05 | 26.74 | — | |
| WD+OURSLoss function=L2 Wasserstein distance, Data augmentation=OURS2021.05 | 26.73 | — | |
| WDLoss function=L2 Wasserstein distance, Data augmentation=None2021.05 | 26.69 | — | |
| CE+MLMLoss function=Cross-Entropy, Data augmentation=MLM2021.05 | 26.2 | — | |
| RNNSearch2018.08 | 26.1 | — | |
| Sequence-to-sequence model with attentiondecoding_strategy=Beam Search2017.06 | 26.1 | — | |
| Sequence-to-sequence model with attentiondecoding_strategy=Greedy2017.06 | 25.5 | — | |
| Luong & Manningdecoding_strategy=Beam Search2017.06 | 23.3 | — | |
| Hard monotonicdecoding_strategy=Greedy2017.06 | 23 | — |