Machine Translation Meta-evaluation on WMT EN-CS 2025
61.4Acc*EqMM (MLP)
Evaluation results
| Method | Links | ||
|---|---|---|---|
| MM (MLP)Pooling Strategy=Language-Separate, Model Variant=MM, Combiner=MLP2026.05 | 61.4 | 90.2 | |
| Gemini-2.5-Pro2026.05 | 61.3 | 89 | |
| DMM (MLP)Pooling Strategy=Language-Separate, Model Variant=DMM, Combiner=MLP, k=62026.05 | 61.3 | 89.6 | |
| MM (MLP)Pooling Strategy=Language-Pooled, Model Variant=MM, Combiner=MLP2026.05 | 61.2 | 89.8 | |
| DMM (MLP)Pooling Strategy=Language-Pooled, Model Variant=DMM, Combiner=MLP, k=62026.05 | 61 | 88.7 | |
| MM (GP)Pooling Strategy=Language-Pooled, Model Variant=MM, Combiner=GP2026.05 | 60.7 | 88.3 | |
| wmt22-comet-da2026.05 | 59.5 | 68.5 | |
| DMM (CL)Pooling Strategy=Language-Pooled, Model Variant=DMM, Combiner=CL, k=62026.05 | 59.1 | 70.9 | |
| DMM (CL)Pooling Strategy=Language-Separate, Model Variant=DMM, Combiner=CL, k=62026.05 | 58.5 | 70 | |
| MM (GP)Pooling Strategy=Language-Separate, Model Variant=MM, Combiner=GP2026.05 | 57.8 | 72.8 | |
| DMM (GP)Pooling Strategy=Language-Pooled, Model Variant=DMM, Combiner=GP, k=62026.05 | 57.8 | 78.1 | |
| DMM (GP)Pooling Strategy=Language-Separate, Model Variant=DMM, Combiner=GP, k=62026.05 | 57 | 70.2 | |
| MM (OLS)Pooling Strategy=Language-Separate, Model Variant=MM, Combiner=OLS2026.05 | 49.4 | 39 | |
| MM (OLS)Pooling Strategy=Language-Pooled, Model Variant=MM, Combiner=OLS2026.05 | 48.6 | 38 | |
| DMM (OLS)Pooling Strategy=Language-Pooled, Model Variant=DMM, Combiner=OLS, k=62026.05 | 47 | 34.5 | |
| DMM (OLS)Pooling Strategy=Language-Separate, Model Variant=DMM, Combiner=OLS, k=62026.05 | 46.4 | 33.4 | |
| gemba2026.05 | 36.9 | 81.4 |