Coreference Resolution on GENIA coreference
91.6Macro F1Fine-tuned Constrained Macaw-3B
Evaluation Results
| Method | Links | |
|---|---|---|
| Fine-tuned Constrained Macaw-3BBackbone=Macaw-3B, Evaluation Protocol=Fine-tuned, Structural constraints=True, Confidence elicitation strategy=multiple choice2025.08 | 91.6 | |
| Local Fine-tuned + Local Calibration (+ constr)Evaluation Protocol=Local Fine-tuned, Calibration strategy=Local, Structural constraints=True, Confidence elicitation strategy=multiple choice2025.08 | 89.1 | |
| Global Fine TuningEvaluation Protocol=Global Fine Tuning, Confidence elicitation strategy=multiple choice2025.08 | 89 | |
| Local Fine-tuned baseline (+ constr)Evaluation Protocol=Local Fine-tuned, Structural constraints=True, Confidence elicitation strategy=multiple choice2025.08 | 88.7 | |
| Local Fine-tuned baselineEvaluation Protocol=Local Fine-tuned, Structural constraints=False, Confidence elicitation strategy=multiple choice2025.08 | 88.3 | |
| Local Fine-tuned + Local CalibrationEvaluation Protocol=Local Fine-tuned, Calibration strategy=Local, Structural constraints=False, Confidence elicitation strategy=multiple choice2025.08 | 88.2 | |
| Local Fine-tuned + Global CalibrationEvaluation Protocol=Local Fine-tuned, Calibration strategy=Global, Confidence elicitation strategy=multiple choice2025.08 | 88.1 | |
| Pre-trained Global CalibrationEvaluation Protocol=Pre-trained, Calibration strategy=Global, Confidence elicitation strategy=multiple choice2025.08 | 80.5 | |
| Pre-trained Local Calibration (+ constr)Evaluation Protocol=Pre-trained, Calibration strategy=Local, Structural constraints=True, Confidence elicitation strategy=multiple choice2025.08 | 80 | |
| Pre-trained Local CalibrationEvaluation Protocol=Pre-trained, Calibration strategy=Local, Structural constraints=False, Confidence elicitation strategy=multiple choice2025.08 | 77.1 | |
| Pre-trained few-shot baseline (+ constr)Evaluation Protocol=Pre-trained few-shot, Structural constraints=True, Confidence elicitation strategy=multiple choice2025.08 | 75.9 | |
| Pre-trained few-shot baselineEvaluation Protocol=Pre-trained few-shot, Structural constraints=False, Confidence elicitation strategy=multiple choice2025.08 | 72.1 |