Word-level Quality Estimation on WMT23 Zh-En (test)
32.4MCCRubric-as-Experts (Ours-8B)
Evaluation Results
| Method | Links | ||||
|---|---|---|---|---|---|
| Rubric-as-Experts (Ours-8B)Backbone=Qwen3-8B2026.06 | 32.4 | 22.29 | 69.8 | 33.78 | |
| Rubric-as-Experts (Ours-4B)Backbone=Qwen3-4B2026.06 | 29.25 | 19.71 | 69.51 | 30.71 | |
| DCSQE2026.06 | 28.12 | — | — | 28.61 | |
| CometKiwi2026.06 | 26.9 | — | — | — | |
| Qwen3-14BBackbone=Qwen3-14B2026.06 | 24.52 | 21.94 | 44.78 | 29.45 | |
| Qwen3-8BBackbone=Qwen3-8B2026.06 | 23.36 | 18.22 | 53.55 | 27.18 | |
| Qwen3-4BBackbone=Qwen3-4B2026.06 | 19.7 | 17.04 | 45.79 | 24.84 | |
| GEMBA2026.06 | 18.21 | — | — | 16.11 |