Machine Translation on WMT'15 German-English (test)
32.1BLEUall features
Evaluation results
| Method | Links | |||
|---|---|---|---|---|
| all featuressynthetic training data=true2016.06 | 32.1 | — | 57.5 | |
| baselinesynthetic training data=true2016.06 | 31.5 | — | 57 | |
| PBSMTsynthetic training data=true2016.06 | 29.9 | — | 56.2 | |
| SBSMTsynthetic training data=true2016.06 | 29.5 | — | 56 | |
| SOTA - phrase-based (Edinburgh)Methodology=Phrase-based2015.08 | 29.2 | — | — | |
| NMT + 5-gram rerank (MILA)Post-processing=5-gram rerank2015.08 | 27.6 | — | — | |
| Base (reverse) + global (dot) + drop + feed + unkSource Reversal=true, Attention=global, Alignment Function=dot, Dropout=true, Input Feeding=true, Unk Replacement=true2015.08 | 24.9 | 9.7 | — | |
| Base (reverse) + global (dot) + drop + feedSource Reversal=true, Attention=global, Alignment Function=dot, Dropout=true, Input Feeding=true2015.08 | 22.8 | — | — | |
| Base (reverse) + global (location) + feedSource Reversal=true, Attention=global, Alignment Function=location, Input Feeding=true2015.08 | 20.1 | 10.9 | — | |
| Base (reverse) + global (location)Source Reversal=true, Attention=global, Alignment Function=location2015.08 | 19.1 | 12.7 | — | |
| Base (reverse)Source Reversal=true2015.08 | 16.9 | 14.3 | — |