Learning Discourse-level Diversity for Neural Dialog Models using Conditional Variational Autoencoders

Proceedings of the 55th Annual Meeting of the Association for Computational Linguistics, pages 654–664 Vancouver, Canada, July 30 - August 4, 2017. c⃝2017 Association for Computational Linguistics https://doi.org/10.18653/v1/P17-1061 Proceedings of the 55th Annual Meeting of the Association for Computational Linguistics, pages 654–664 Vancouver, Canada, July 30 - August 4, 2017. c⃝2017 Association for Computational Linguistics https://doi.org/10.18653/v1/P17-1061 Learning Discourse-level Diversity for Neural Dialog Models using Conditional Variational Autoencoders Tiancheng Zhao, Ran Zhao and Maxine Eskenazi Language Technologies Institute Carnegie Mellon University Pittsburgh, Pennsylvania, USA {tianchez,ranzhao1,max+}@cs.cmu.edu Abstract While recent neural encoder-decoder models have shown great promise in mod- eling open-domain conversations, they often generate dull and generic responses. Unlike past work that has focused on diversifying the output of the decoder at word-level to alleviate this problem, we present a novel framework based on conditional variational autoencoders that captures the discourse-level diversity in the encoder. Our model uses latent vari- ables to learn a distribution over potential conversational intents and generates diverse responses using only greedy de- coders. We have further developed a novel variant that is integrated with linguistic prior knowledge for better performance. Finally, the training procedure is improved by introducing a bag-of-word loss. Our proposed models have been validated to generate significantly more diverse responses than baseline approaches and exhibit competence in discourse-level decision-making. 1 Introduction The dialog manager is one of the key components of dialog systems, which is responsible for mod- eling the decision-making process. Specifically, it typically takes a new utterance and the dialog con- text as input, and generates discourse-level deci- sions (Bohus and Rudnicky, 2003; Williams and Young, 2007). Advanced dialog managers usu- ally have a list of potential actions that enable them to have diverse behavior during a conver- sation, e.g. different strategies to recover from non-understanding (Yu et al., 2016). However, the conventional approach of designing a dialog manager (Williams and Young, 2007) does not scale well to open-domain conversation models because of the vast quantity of possible decisions. Thus, there has been a growing interest in applying encoder-decoder models (Sutskever et al., 2014) for modeling open-domain conversation (Vinyals and Le, 2015; Serban et al., 2016a). The basic ap- proach treats a conversation as a transduction task, in which the dialog history is the source sequence and the next response is the target sequence. The model is then trained end-to-end on large conver- sation corpora using the maximum-likelihood esti- mation (MLE) objective without the need for man- ual crafting. However recent research has found that encoder-decoder models tend to generate generic and dull responses (e.g., I don’t know), rather than meaningful and specific answers (Li et al., 2015; Serban et al., 2016b). There have been many attempts to explain and solve this limita- tion, and they can be broadly divided into two cat- egories (see Section 2 for details): (1) the first cat- egory argues that the dialog history is only one of the factors that decide the next response. Other features should be extracted and provided to the models as conditionals …

Content truncated to 500 words