ResearchBenchmarksDialog response generation on FEWSHOTWOZ (test)Dialog response generation on FEWSHOTWOZ (test)Follow2.92InformativenessHuman2.26482.43492.6052.7751Feb 27, 2020Evaluation resultsMethodMethodLinksInformativenessNaturalnessHuman2020.022.922.72SC-GPTBackbone=GPT2-Medium,...Backbone=GPT2-Medium, Parameters=345M, Epochs=52020.022.642.47GPT-2Model size=Medium (345...Model size=Medium (345M parameters), Training=Fine-tuned on domain-specific labels2020.022.542.38SC-LSTM2020.022.292.13