RDF-to-text generation on OpenDialKG (test)
98.5GrammaticalityLlama 3.3
Evaluation Results
| Method | Links | |||||||
|---|---|---|---|---|---|---|---|---|
| Llama 3.3Model Category=Neural models2025.12 | 98.5 | 3 | 6.3 | — | — | — | — | |
| Rule-based NLGTrained by=GPT 4.12025.12 | 92.3 | 1.3 | 2.2 | — | — | — | — | |
| Rule-based NLGTrained by=Qwen 3 235B2025.12 | 84 | 1.8 | 10.6 | — | — | — | — | |
| Rule-based NLGTrained by=Qwen 2.5 72B2025.12 | 66.6 | 3.3 | 14.6 | — | — | — | — | |
| Rule-based NLGTrained by=Llama 3.3 70B2025.12 | 59.8 | 5.6 | 7.5 | — | — | — | — | |
| BARTModel Category=Neural models2025.12 | 50.2 | 5.2 | 13.9 | — | — | — | — | |
| BARTtype=Neural model, training=Fine-tuned2025.12 | — | — | — | 93.72 | 98.49 | 99.73 | 93.4 | |
| Llama 3.3 70btype=Neural model, training=Prompted2025.12 | — | — | — | 20.4 | 62.89 | 91.04 | 13.48 | |
| Rule-based NLGtrained by=GPT 4.12025.12 | — | — | — | 31.44 | 73.13 | 92.72 | 32.47 | |
| Rule-based NLGtrained by=Qwen 3 235b2025.12 | — | — | — | 34.72 | 69.47 | 92.39 | 10.47 | |
| Rule-based NLGtrained by=Qwen 2.5 72b2025.12 | — | — | — | 34.13 | 70.3 | 92.65 | 23 | |
| Rule-based NLGtrained by=Llama 3.3 70B2025.12 | — | — | — | 31.2 | 65.17 | 92.16 | 18.82 |