Table-to-Text Generation on LOGICNLG (test)
53BLEU-1BART
Evaluation results
| Method | Links | |||||
|---|---|---|---|---|---|---|
| BART2022.10 | 53 | 32.9 | 19.2 | 50.1 | 83.7 | |
| T52022.10 | 52.6 | 32.6 | 19.3 | 48.2 | 80.4 | |
| REASTAP2022.10 | 52.5 | 32.5 | 18.9 | 54.8 | 89.2 | |
| R2D22022.10 | 51.8 | 32.4 | 18.6 | 50.8 | 85.6 | |
| PLOG2022.10 | 51.7 | 32.3 | 18.9 | 48.9 | 85.5 | |
| GPT2-TabGen2022.10 | 49.6 | 28.2 | 14.2 | 44.7 | 74.6 | |
| GPT-TabGenParadigm=Supervised, Pre-training=GPT-2-based2022.12 | 49.6 | 28.2 | 14.2 | — | — | |
| DCVED2022.10 | 49.5 | 28.6 | 15.3 | 43.9 | 76.9 | |
| DCVEDParadigm=Supervised, Samples=37k in-domain2022.12 | 49.5 | 28.6 | 15.3 | — | — | |
| BERT-TabGenParadigm=Supervised, Pre-training=BERT-based2022.12 | 49.1 | 27.7 | 13.5 | — | — | |
| GPT2-Coarse-to-Fine2022.10 | 49 | 28.3 | 14.6 | 45.3 | 76.4 | |
| GPT-Coarse-to-FineParadigm=Supervised, Pre-training=GPT-2-based2022.12 | 49 | 28.3 | 14.6 | — | — | |
| Field-InfusingParadigm=Supervised, Pre-training=Non-pretrained2022.12 | 43.7 | 20.9 | 8.4 | — | — | |
| MURMURParadigm=Few-shot, Demonstrations=1, Gold (table, path) pairs=2212022.12 | 39.8 | 22.2 | 11.2 | — | — | |
| MURMUR (without saliency)Paradigm=Few-shot, Demonstrations=1, Ablation=Random module selection2022.12 | 39.6 | 21.9 | 10.6 | — | — | |
| BART + SRParadigm=Supervised, Samples=37k in-domain, Type=Ablation (fine-tuned reasoning path generation)2022.12 | 39.2 | 20.6 | 9.5 | — | — | |
| Direct PromptingParadigm=Few-shot, Demonstrations=12022.12 | 37.2 | 18.8 | 8.6 | — | — | |
| CoT PromptingParadigm=Few-shot, Demonstrations=12022.12 | 35.6 | 18.6 | 8.8 | — | — |