Argument Component Detection on Persuasive Essays (PE) (test)
88.6Macro F1Human Upper Bound
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| Human Upper BoundDataset=PE2026.03 | 88.6 | — | |
| Llama-3-8BDataset=PE, Learning Protocol=Instruction-tuned2026.03 | 87.78 | 90.04 | |
| CRF with featuresDataset=PE2026.03 | 86.7 | — | |
| GPT-2-1.5BDataset=PE, Learning Protocol=Instruction-tuned2026.03 | 85.21 | 88.04 | |
| OPT-6.7BDataset=PE, Learning Protocol=Instruction-tuned2026.03 | 85.18 | 88.56 | |
| MT-allDataset=PE2026.03 | 75.66 | — | |
| DeBERTa-v3Dataset=PE, Learning Protocol=Fine-tuned, Max input length=64, Epochs=10, Learning rate=1e-4, BIO scheme=True2026.03 | 71.12 | — | |
| RoBERTaDataset=PE, Learning Protocol=Fine-tuned, Max input length=64, Epochs=10, Learning rate=1e-4, BIO scheme=True2026.03 | 69.33 | — | |
| Heuristic BaselineDataset=PE2026.03 | 64.2 | — |