Grammatical Error Correction on RULEC-GEC Russian (test)
51.62F0.5 ScoregT5 xxl
Evaluation results
| Method | Links | |||
|---|---|---|---|---|
| gT5 xxl2021.06 | 51.62 | — | — | |
| Transformertraining_regime=finetuned2019.10 | 50.2 | 63.26 | 27.5 | |
| Náplava and Straka2020.05 | 50.2 | 63.26 | 27.5 | |
| SOTA2021.06 | 50.2 | — | — | |
| Transformertraining_regime=finetuned, hardware=base single GPU2019.10 | 47.15 | 59.13 | 26.05 | |
| mBART-basedUse of pseudo corpus=true, pseudo corpus size=10M2020.05 | 44.36 | 53.5 | 26.35 | |
| mT5 xxl#params=13B, Training Data=CLANG-82021.06 | 43.52 | — | — | |
| Transformertraining_regime=synthetic pretrain2019.10 | 40.96 | 47.76 | 26.08 | |
| mT5 xl#params=3.7B, Training Data=CLANG-82021.06 | 39.44 | — | — | |
| mT5 large#params=1.2B, Training Data=CLANG-82021.06 | 27.55 | — | — | |
| mT5 base#params=580M, Training Data=CLANG-82021.06 | 25.2 | — | — | |
| Rozovskaya and Roth2019.10 | 21 | 38 | 7.5 | |
| mT5 small#params=300M, Training Data=CLANG-82021.06 | 17.8 | — | — | |
| mBART-basedUse of pseudo corpus=false2020.05 | 15.38 | 32.13 | 4.99 |