Domain Adaptive Dialog Generation via Meta Learning

Proceedings of the 57th Annual Meeting of the Association for Computational Linguistics, pages 2639–2649 Florence, Italy, July 28 - August 2, 2019. c⃝2019 Association for Computational Linguistics 2639 Domain Adaptive Dialog Generation via Meta Learning Kun Qian Univeristy of California, Davis kunqian@ucdavis.edu Zhou Yu Univeristy of California, Davis joyu@ucdavis.edu Abstract Domain adaptation is an essential task in dia- log system building because there are so many new dialog tasks created for different needs every day. Collecting and annotating train- ing data for these new tasks is costly since it involves real user interactions. We propose a domain adaptive dialog generation method based on meta-learning (DAML). DAML is an end-to-end trainable dialog system model that learns from multiple rich-resource tasks and then adapts to new domains with mini- mal training samples. We train a dialog sys- tem model using multiple rich-resource single- domain dialog data by applying the model- agnostic meta-learning algorithm to dialog do- main. The model is capable of learning a competitive dialog system on a new domain with only a few training examples in an effi- cient manner. The two-step gradient updates in DAML enable the model to learn general features across multiple tasks. We evaluate our method on a simulated dialog dataset and achieve state-of-the-art performance, which is generalizable to new tasks. 1 Introduction Modern personal assistants, such as Alexa and Siri, are composed of thousands of single-domain task-oriented dialog systems. Every dialog task is different, due to the specific domain knowl- edge. An end-to-end trainable dialog system re- quires thousands of dialogs for training. How- ever, the availability of the training data is usu- ally limited as real users have to be involved to obtain the training dialogs. Therefore, adapting existing rich-resource data to new domains with limited resource is an essential task in dialog sys- tem research. Transfer learning (Caruana, 1997a; Bengio, 2012; Cohn et al., 1994; Mo et al., 2018), few-shot learning (Salakhutdinov et al., 2012; Li et al., 2006; Norouzi et al., 2013; Socher et al., 2013) and meta-learning (Finn et al., 2017) are introduced in solving such data scarcity problem in machine learning. Because every dialog do- main is very different from each other, general- ize information from rich-resource domains to an- other low resource domain is difficult. There- fore, only a few studies have tackled domain adap- tive end-to-end dialog training methods (Zhao and Esk´enazi, 2018). We propose DAML based on meta-learning to combine multiple dialog tasks in training, in order to learn general and transferable information that is applicable to new domains. Zhao and Esk´enazi (2018) introduces action matching, a learning framework that could re- alize zero-shot dialog generation (ZSDG), based on domain description, in the form of seed re- sponse. With limited knowledge of a new do- main, the model trained on several rich-resource domains achieves both impressive task comple- tion rate and natural generated response. Rather than action matching, we propose to use model- agnostic meta-learning (MAML) algorithm (Finn et al., 2017) to perform dialog domain adapta- tion. …

Content truncated to 500 words