Grammar Error Correction on Falko-Merlin German (test)
76F0.5 ScoreRothe et al.
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| Rothe et al.Backbone=gT5 xxl, Lang=Mono2026.06 | 76 | — | — | |
| Luhtaru et al.Backbone=NLLB, Lang=Multi2026.06 | 73.9 | — | — | |
| TransformerStage=finetuned2019.10 | 73.71 | 78.21 | 59.94 | |
| TransformerStage=finetuned base, Hardware=single GPU2019.10 | 73.4 | 78.11 | 59.13 | |
| Náplava and Straka2020.05 | 73.31 | 78.21 | 59.94 | |
| mBART-based2020.05 | 68.86 | 73.97 | 53.98 | |
| Li et al.Backbone=GPT-4o-mini2026.06 | 65.6 | — | — | |
| GERBackbone=Llama3.1 (8B)2026.06 | 63.7 | — | — | |
| Li et al.Backbone=Deepseek2.52026.06 | 63.4 | — | — | |
| Tang et al.Backbone=GPT-3.5-Turbo2026.06 | 58.8 | — | — | |
| Davis et al.Backbone=GPT-3.5-Turbo2026.06 | 57.2 | — | — | |
| TransformerStage=synthetic pretrain2019.10 | 51.41 | 67.45 | 26.35 | |
| Boyd (2018)2019.10 | 45.22 | 51.99 | 29.73 |