Machine Translation on WMT'16 En-Ro (test)
38BLEUCeMAT
Evaluation Results
| Method | Links | |
|---|---|---|
| CeMATTraining Size=597K, Pre-trained=true2022.03 | 38 | |
| mBARTTraining Size=597K, Pre-trained=true2022.03 | 37.7 | |
| mRASPTraining Size=597K, Pre-trained=true2022.03 | 37.6 | |
| DirectTraining Size=597K, Pre-trained=false2022.03 | 34.3 | |
| TransformerDecoding Strategy=Autoregressive2022.03 | 34.1 | |
| FullyNAT + CTC + GLATCategory=Alignment-Based, Inference mode=single-step2022.06 | 33.7 | |
| CeMATInitialization=CeMAT, Architecture=Mask-Predict, Decoding Strategy=Non-autoregressive2022.03 | 33.3 | |
| CMLMC + DiMSCategory=Proposed, Inference mode=single-step2022.06 | 33.2 | |
| AlignNARTCategory=Alignment-Based, Inference mode=single-step2022.06 | 32.5 | |
| OAXECategory=Alignment-Based, Inference mode=single-step2022.06 | 32.4 | |
| ImputerCategory=Alignment-Based, Inference mode=single-step2022.06 | 32.3 | |
| mRASPInitialization=mRASP, Architecture=Mask-Predict, Decoding Strategy=Non-autoregressive2022.03 | 32.2 | |
| Mask-PredictDecoding Strategy=Non-autoregressive2022.03 | 31.5 | |
| GLATCategory=XE-Based, Inference mode=single-step2022.06 | 31.2 | |
| AXECategory=Alignment-Based, Inference mode=single-step2022.06 | 30.8 | |
| Bag-of-ngramCategory=XE-Based, Inference mode=single-step2022.06 | 28.3 | |
| CMLMCCategory=XE-Based, Inference mode=single-step2022.06 | 28.2 | |
| CMLMCategory=XE-Based, Inference mode=single-step2022.06 | 27.3 |