Word-level Quality Estimation on WMT23 En-De (test)
29.57MCCRubric-as-Experts (Ours-8B)
Evaluation Results
| Method | Links | ||||
|---|---|---|---|---|---|
| Rubric-as-Experts (Ours-8B)Backbone=Qwen3-8B2026.06 | 29.57 | 22.68 | 60.16 | 32.94 | |
| DCSQE2026.06 | 27.11 | — | — | 30.61 | |
| Rubric-as-Experts (Ours-4B)Backbone=Qwen3-4B2026.06 | 26.6 | 31.75 | 31.4 | 31.57 | |
| CometKiwi2026.06 | 21.5 | — | — | — | |
| GEMBA2026.06 | 18.17 | — | — | 13.21 | |
| Qwen3-14BBackbone=Qwen3-14B2026.06 | 17.16 | 23.99 | 20.83 | 22.3 | |
| Qwen3-8BBackbone=Qwen3-8B2026.06 | 14.76 | 19.16 | 23.43 | 21.08 | |
| Qwen3-4BBackbone=Qwen3-4B2026.06 | 10.23 | 16.31 | 16.19 | 16.25 |