Machine Translation on WMT newstest 2015 (test)
34.93BLEUCYCLE (REV)
Evaluation results
| Method | Links | ||||||||
|---|---|---|---|---|---|---|---|---|---|
| CYCLE (REV)#Params=343M, Training Data=Synthetic2021.04 | 34.93 | — | — | — | — | — | — | — | |
| CYCLE#Params=242M, Training Data=Genuine2021.04 | 34.04 | — | — | — | — | — | — | — | |
| CYCLE (REV)#Params=242M, Training Data=Genuine2021.04 | 33.98 | — | — | — | — | — | — | — | |
| SEQUENCE#Params=242M, Training Data=Genuine2021.04 | 33.84 | — | — | — | — | — | — | — | |
| Universal#Params=249M, Training Data=Genuine2021.04 | 33.53 | — | — | — | — | — | — | — | |
| Vanilla#Params=242M, Training Data=Genuine2021.04 | 33.52 | — | — | — | — | — | — | — | |
| NMTtraining_data=parallel + synthetic, ensemble=42015.11 | 31.6 | — | — | — | — | — | — | — | |
| B2T connectionEnc-Dec Layers=100L-100L2022.06 | 31.57 | — | — | — | — | — | — | — | |
| B2T connectionEnc-Dec layers=18L-18L2022.06 | 30.99 | — | — | — | — | — | — | — | |
| DeepNetEnc-Dec layers=18L-18L2022.06 | 30.6 | — | — | — | — | — | — | — | |
| Pre-LNEnc-Dec Layers=100L-100L2022.06 | 30.5 | — | — | — | — | — | — | — | |
| NMTtraining_data=parallel + synthetic2015.11 | 30.4 | — | — | — | — | — | — | — | |
| RealFormerEnc-Dec layers=18L-18L2022.06 | 30.36 | — | — | — | — | — | — | — | |
| AdminEnc-Dec layers=18L-18L2022.06 | 30.35 | — | — | — | — | — | — | — | |
| DLCLEnc-Dec layers=18L-18L2022.06 | 30.24 | — | — | — | — | — | — | — | |
| T-FixupEnc-Dec layers=18L-18L2022.06 | 30.13 | — | — | — | — | — | — | — | |
| Pre-LNEnc-Dec layers=18L-18L2022.06 | 29.74 | — | — | — | — | — | — | — | |
| DLCLEnc-Dec layers=6L-6L2022.06 | 29.71 | — | — | — | — | — | — | — | |
| Post-LNEnc-Dec layers=6L-6L2022.06 | 29.7 | — | — | — | — | — | — | — | |
| DeepNetEnc-Dec layers=6L-6L2022.06 | 29.62 | — | — | — | — | — | — | — | |
| AdminEnc-Dec layers=6L-6L2022.06 | 29.61 | — | — | — | — | — | — | — | |
| B2T connectionEnc-Dec layers=6L-6L2022.06 | 29.48 | — | — | — | — | — | — | — | |
| T-FixupEnc-Dec layers=6L-6L2022.06 | 29.45 | — | — | — | — | — | — | — | |
| RealFormerEnc-Dec layers=6L-6L2022.06 | 29.36 | — | — | — | — | — | — | — | |
| PBSMT2015.11 | 29.3 | — | — | — | — | — | — | — | |
| Pre-LNEnc-Dec layers=6L-6L2022.06 | 29.07 | — | — | — | — | — | — | — | |
| NMTtraining_data=parallel2015.11 | 26.7 | — | — | — | — | — | — | — | |
| ByteNetInputs=char, Outputs=char2016.10 | 26.26 | — | — | — | — | — | — | — | |
| Phrase Based MTInputs=phrases, Outputs=phrases2016.10 | 24 | — | — | — | — | — | — | — | |
| RNN Enc-Dec Att (Chung et al.)Inputs=BPE, Outputs=char2016.10 | 23.45 | — | — | — | — | — | — | — | |
| RNN Enc-Dec Att (Chung et al.)Inputs=BPE, Outputs=BPE2016.10 | 21.72 | — | — | — | — | — | — | — | |
| BPE-60ksegmentation=BPE, vocabulary (source)=60,000, vocabulary (target)=60,0002015.08 | — | 21.5 | 24.5 | 52 | 53.9 | 58.4 | 40.9 | 29.3 | |
| BPE-J90ksegmentation=BPE (joint), vocabulary (source)=90,000, vocabulary (target)=90,0002015.08 | — | 22.8 | 24.7 | 51.7 | 54.1 | 58.5 | 41.8 | 33.6 | |
| C2-50ksegmentation=char-bigram, shortlist=50,000, vocabulary (source)=60,000, vocabulary (target)=60,0002015.08 | — | 22.8 | 25.3 | 51.9 | 53.5 | 58.4 | 40.5 | 30.9 | |
| syntax-basedsegmentation=syntax-based2015.08 | — | 24.4 | — | 55.3 | — | 59.1 | 46 | 37.7 | |
| WDictvocabulary (source)=300,000, vocabulary (target)=500,0002015.08 | — | 22 | 24.2 | 50.5 | 52.4 | 58.1 | 36.8 | 36.8 | |
| WUnkvocabulary (source)=300,000, vocabulary (target)=500,0002015.08 | — | 20.6 | 22.8 | 47.2 | 48.9 | 56.7 | 20.4 | 0 |