Data-to-text generation on WSQL (test)
58.71ROUGE-1CycleNLG
Evaluation results
| Method | Links | |||||||
|---|---|---|---|---|---|---|---|---|
| CycleNLGTraining setting=Low-resource cycle training2023.05 | 58.71 | 33.13 | 51.01 | 37.43 | 25.6 | 93.03 | 25.84 | |
| Fully-supervised fine-tuningTraining setting=Fully-supervised2023.05 | 58.27 | 32.77 | 48.4 | 37.95 | 22.97 | 93.18 | 24 | |
| Low-resource fine-tuningTraining setting=Low-resource2023.05 | 56.37 | 31.6 | 49.42 | 33.57 | 23.34 | 92.57 | 23.68 | |
| Low-resource fine-tuning + additional pre-trainingTraining setting=Low-resource, Additional pre-training=true2023.05 | 56.01 | 30.92 | 47 | 35.34 | 21.18 | 92.24 | 22.66 | |
| CycleNLGTraining setting=Unsupervised cycle training2023.05 | 42.24 | 15.17 | 33.52 | 29.45 | 4.03 | 85.37 | 14.63 |