Machine Translation on IWSLT Thai-English 2012/2013 2015 (test)
15.6BLEU (n=1)Baseline + Seq-Inter
Evaluation Results
| Method | Links | ||||||
|---|---|---|---|---|---|---|---|
| Baseline + Seq-InterArchitecture=2 x 500, Parameters=47m, Role=Teacher2016.06 | 15.6 | 1.3 | 16 | 0.3 | 55.1 | 6.8 | |
| BaselineArchitecture=2 x 500, Parameters=47m, Role=Teacher2016.06 | 14.3 | — | 15.7 | — | 22.9 | 2.3 | |
| Seq-KD + Seq-Inter + Word-KDArchitecture=2 x 100, Parameters=8m, Role=Student2016.06 | 14.2 | 3.6 | 14.4 | 1.7 | 117.4 | 3.2 | |
| Seq-KD + Word-KDArchitecture=2 x 100, Parameters=8m, Role=Student2016.06 | 13.7 | 3.1 | 14.2 | 1.5 | 67.4 | 3.1 | |
| Seq-KD + Seq-InterArchitecture=2 x 100, Parameters=8m, Role=Student2016.06 | 13.6 | 3 | 14 | 1.3 | 106.4 | 3.9 | |
| Word-KD + Seq-InterArchitecture=2 x 100, Parameters=8m, Role=Student2016.06 | 13 | 2.4 | 13.7 | 1 | 58.7 | 3.2 | |
| Baseline + Seq-InterArchitecture=2 x 100, Parameters=8m, Role=Student2016.06 | 12.9 | 2.3 | 13.1 | 0.4 | 52.8 | 2.5 | |
| Seq-KDArchitecture=2 x 100, Parameters=8m, Role=Student2016.06 | 12.8 | 2.2 | 13.4 | 0.7 | 125.4 | 6.9 | |
| Word-KDArchitecture=2 x 100, Parameters=8m, Role=Student2016.06 | 11.8 | 1.2 | 13.6 | 0.9 | 35.3 | 1.4 | |
| BaselineArchitecture=2 x 100, Parameters=8m, Role=Student2016.06 | 10.6 | — | 12.7 | — | 37 | 1.4 |