Machine Translation on ASPEC En-Ja
55.46BLEU ScoreKudo (2018)
Evaluation Results
| Method | Links | |
|---|---|---|
| Kudo (2018)Segmentation=Unigram Language Model2019.10 | 55.46 | |
| BPE-dropoutSegmentation=BPE-dropout2019.10 | 55 | |
| BPESegmentation=BPE2019.10 | 54.51 | |
| Compositional CodingM x K=64 x 16, #vectors=1024, vector size=3.50 MB, code len=256 bits, code size=2.44 MB, total size=5.94 MB2017.11 | 38.89 | |
| Iterative Pruningpruning ratio=90%, #vectors=80000, vector size=41 MB, total size=41 MB2017.11 | 38.56 | |
| Compositional CodingM x K=32 x 16, #vectors=512, vector size=1.75 MB, code len=128 bits, code size=1.22 MB, total size=2.97 MB2017.11 | 38.1 | |
| Baseline NMT#vectors=80000, vector size=274 MB, total size=274 MB2017.11 | 37.93 | |
| Iterative Pruningpruning ratio=98%, #vectors=80000, vector size=8.26 MB, total size=8.26 MB2017.11 | 37.09 |