Segment-level Speech Translation Quality Estimation on IWSLT Metrics Shared Task 2026 (dev)
32.7Kendall's Tau (en-de)HydraQE (xCOMET head)
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| HydraQE (xCOMET head)2026.06 | 32.7 | 37.1 | 34.9 | |
| COMET2026.06 | 32.6 | 36.5 | 34.6 | |
| HydraQE (DA head)2026.06 | 32.5 | 39.9 | 36.2 | |
| HydraQE (primary)2026.06 | 32.1 | 39.8 | 36 | |
| HydraQE (all heads avg.)2026.06 | 32 | 38.7 | 35.4 | |
| HydraQE (MetricX head)2026.06 | 31.8 | 39.3 | 35.6 | |
| CometKiwivariant=XXL2026.06 | 31.4 | 35.9 | 33.7 | |
| xCOMET-XXL QE2026.06 | 30.4 | 35.5 | 33 | |
| MetricX-24-XXL QE2026.06 | 29.7 | 31.2 | 30.5 | |
| SpeechQE2026.06 | 26.6 | 31.8 | 29.2 | |
| BLASER 2.0 QE2026.06 | 22 | 26.8 | 24.4 | |
| COMET (partial)2026.06 | 11.3 | 12 | 11.6 |