RDF-to-text generation on WebNLG+ 2020 3.0 (testB)
0.563BLEUReGen
Evaluation Results
| Method | Links | ||||
|---|---|---|---|---|---|
| ReGenBackbone=t5-large, Training Protocol=Reinforcement Learning (RL)2021.08 | 0.563 | 0.559 | 0.425 | 0.706 | |
| ReGenBackbone=t5-large, Training Protocol=Cross-Entropy (CE)2021.08 | 0.553 | 0.549 | 0.418 | 0.694 | |
| Amazon AI (Shanghai)2021.08 | 0.54 | 0.535 | 0.417 | 0.69 | |
| OSU Neural NLG2021.08 | 0.535 | 0.532 | 0.414 | 0.688 | |
| ReGenBackbone=t5-base, Training Protocol=Reinforcement Learning (RL), Selection Strategy=early CE (Early Stopping)2021.08 | 0.531 | 0.527 | 0.41 | 0.686 | |
| FBConvAI2021.08 | 0.527 | 0.523 | 0.413 | 0.686 | |
| ReGenBackbone=t5-base, Training Protocol=Reinforcement Learning (RL), Selection Strategy=best CE2021.08 | 0.527 | 0.523 | 0.408 | 0.681 | |
| ReGenBackbone=t5-base, Training Protocol=Cross-Entropy (CE), Selection Strategy=best CE2021.08 | 0.524 | 0.52 | 0.404 | 0.677 | |
| ReGenBackbone=t5-base, Training Protocol=Cross-Entropy (CE), Selection Strategy=early CE (Early Stopping)2021.08 | 0.522 | 0.518 | 0.404 | 0.675 | |
| bt52021.08 | 0.517 | 0.517 | 0.411 | 0.679 |