Validation timing detection on M-EDESConv Japanese
61.2Overall Macro-F1MEGUMI
Evaluation results
| Method | Links | ||||
|---|---|---|---|---|---|
| MEGUMI2026.06 | 61.2 | 48.83 | 75.87 | 59.42 | |
| XLM-RoBERTa2026.06 | 60.76 | 47.52 | 69.73 | 56.56 | |
| BERT2026.06 | 57.35 | 44.83 | 76.57 | 56.61 | |
| ModernBERT2026.06 | 51.15 | 42.02 | 86.48 | 56.67 | |
| Random Baseline2026.06 | 49.01 | 37.23 | 50.09 | 42.71 | |
| GPT 4o NanoEvaluation Protocol=Zero-shot2026.06 | 47.03 | 40.28 | 81.51 | 53.91 | |
| GPT 4o NanoEvaluation Protocol=CoT2026.06 | 40.02 | 39.47 | 93.43 | 55.5 | |
| GPT 4o NanoEvaluation Protocol=3-shot2026.06 | 39.67 | 39.28 | 99.94 | 55.22 | |
| Llama 3.1 8bEvaluation Protocol=LoRA2026.06 | 38.71 | 34.02 | 85.71 | 48.71 | |
| Llama 3.1 8bEvaluation Protocol=Zero-shot2026.06 | 30.98 | 37.79 | 97.82 | 54.51 | |
| Llama 3.1 8bEvaluation Protocol=3-shot2026.06 | 29.22 | 37.69 | 99.44 | 54.67 |