Dialogue Response Generation on Dialogue Dataset (test)
37.2Adversarial SuccessSampling
Evaluation results
| Method | Links | ||
|---|---|---|---|
| SamplingTraining Strategy=Maximum Likelihood Estimation, Decoding Strategy=Sampling2017.01 | 37.2 | 0.679 | |
| Adver-REGSTraining Strategy=Adversarial Reinforcement Learning with Reward for Every Generation Step, Decoding Strategy=Greedy decoding2017.01 | 9.8 | 0.952 | |
| MMI-p(t)Training Strategy=Mutual Information Reranking, Decoding Strategy=Penalized language model probability2017.01 | 9 | 0.88 | |
| Adver-ReinforceTraining Strategy=Adversarial Reinforcement Learning, Decoding Strategy=Greedy decoding2017.01 | 8 | 0.945 | |
| MMI+p(t|s)Training Strategy=Mutual Information Reranking, Decoding Strategy=Reranking with backward probability p(s|t)2017.01 | 7.3 | 0.953 | |
| MLE-GreedyTraining Strategy=Maximum Likelihood Estimation, Decoding Strategy=Greedy decoding2017.01 | 4.9 | 0.945 | |
| MLE-BSTraining Strategy=Maximum Likelihood Estimation, Decoding Strategy=Beam Search2017.01 | 3.7 | 0.942 |