Definition Generation on WordNet
47.9BLEULM-LEXICON-DENSE
Evaluation Results
| Method | Links | ||||
|---|---|---|---|---|---|
| LM-LEXICON-DENSEParameter Count=8B, Mode=BoN-Oracle2026.02 | 47.9 | — | — | 44.19 | |
| LM-LEXICON-MOEParameter Count=4x8B, Mode=BoN-Oracle2026.02 | 47.39 | — | — | 40.31 | |
| LM-LEXICON-MOEParameter Count=4x8B, Mode=BoN-ORM2026.02 | 40.33 | — | — | 40.69 | |
| LM-LEXICON-MOEParameter Count=4x8B, Mode=Zero-shot2026.02 | 40.09 | — | — | 40.51 | |
| LM-LEXICON-DENSEParameter Count=8B, Mode=BoN-ORM2026.02 | 37.73 | — | — | 37.94 | |
| LM-LEXICON-DENSEParameter Count=8B, Mode=Zero-shot2026.02 | 36.99 | — | — | 37.83 | |
| LlamaDictSupervised training=true2026.02 | 33.86 | — | — | 43.5 | |
| Flan-T5 XLTest=Soft domain shift2023.05 | 32.81 | 52.21 | 92.16 | — | |
| Huang et al. (2021)Test=Unknown2023.05 | 32.72 | — | — | — | |
| Flan-T5-DefSupervised training=true2026.02 | 31.96 | — | — | 40.45 | |
| MDM-T5Supervised training=true2026.02 | 31.18 | — | — | 32.55 | |
| GPT-4-TURBOIn-Context Learning=Random-ICL, shots=322026.02 | 30.95 | — | — | 32.61 | |
| Rerank-T5Supervised training=true2026.02 | 30.91 | — | — | 30.99 | |
| Contrast-T5Supervised training=true2026.02 | 30.81 | — | — | 26.27 | |
| Flan-T5 XLTest=Hard domain shift2023.05 | 29.55 | 48.17 | 91.39 | — | |
| SimpDefinerSupervised training=true2026.02 | 28.91 | — | — | 20.47 | |
| CLAUDE-3-OPUSIn-Context Learning=Random-ICL, shots=322026.02 | 28.63 | — | — | 27.84 | |
| GPT-4-TURBOIn-Context Learning=Retrieval-ICL, shots=322026.02 | 27.46 | — | — | 29.74 | |
| GEMINI-1.5-PROIn-Context Learning=Retrieval-ICL, shots=322026.02 | 25.24 | — | — | 27.88 | |
| GEMINI-1.5-PROIn-Context Learning=Random-ICL, shots=322026.02 | 23.42 | — | — | 26.27 | |
| CLAUDE-3-OPUSIn-Context Learning=Retrieval-ICL, shots=322026.02 | 18.57 | — | — | 21.76 | |
| Flan-T5 XLTest=In-distribution2023.05 | 11.49 | 28.96 | 88.9 | — | |
| Flan-T5 XLTest=Zero-shot (task shift)2023.05 | 2.7 | 12.72 | 86.72 | — |