Machine Translation on WMT German-Romansh 2024++ (test)
0.47FluencyHuman reference
Evaluation results
| Method | Links | ||
|---|---|---|---|
| Human referenceEvaluation protocol=Human evaluation2026.03 | 0.47 | 0.5 | |
| Fine-tuned NLLBData augmentation strategy=LR→HR augmentation, Prompting strategy=dictionary prompting, Evaluation protocol=Human evaluation2026.03 | 0.17 | 0.15 | |
| Fine-tuned NLLBData augmentation strategy=LR→HR augmentation, Evaluation protocol=Human evaluation2026.03 | 0.14 | 0.01 | |
| Gemini 3 Pro (preview)Evaluation protocol=Human evaluation2026.03 | -0.82 | -0.73 |