Error Span Detection on WMT Metrics Shared Task
86.4SPAMBR Distill
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| MBR Distilltraining_source=self-trained pseudo-labels2026.03 | 86.4 | 57.3 | 93.9 | |
| Base Modeladaptation=unadapted2026.03 | 85.1 | 51.8 | 87.4 | |
| Gold-SFTtraining_source=human-annotated data2026.03 | 82.6 | 57.1 | 91.5 |