Unsupervised Context Rewriting for Open Domain Conversation

Proceedings of the 2019 Conference on Empirical Methods in Natural Language Processing and the 9th International Joint Conference on Natural Language Processing, pages 1834–1844, Hong Kong, China, November 3–7, 2019. c⃝2019 Association for Computational Linguistics 1834 Unsupervised Context Rewriting for Open Domain Conversation Kun Zhou1∗, Kai Zhang2, Yu Wu3, Shujie Liu3, and Jingsong Yu1 1School of Software and Microelectronics, Peking University 2AI and Research Microsoft, Beijing, China 3Microsoft Research, Beijing, China franciszhou@pku.edu.cn {kaizh,Wu.Yu,shujliu}@microsoft.com yjs@ss.pku.edu.cn Abstract Context modeling has a pivotal role in open domain conversation. Existing works either use heuristic methods or jointly learn con- text modeling and response generation with an encoder-decoder framework. This paper proposes an explicit context rewriting method, which rewrites the last utterance by consid- ering context history. We leverage pseudo- parallel data and elaborate a context rewrit- ing network, which is built upon the Copy- Net with the reinforcement learning method. The rewritten utterance is beneficial to candi- date retrieval, explainable context modeling, as well as enabling to employ a single-turn framework to the multi-turn scenario. The empirical results show that our model outper- forms baselines in terms of the rewriting qual- ity, the multi-turn response generation, and the end-to-end retrieval-based chatbots. 1 Introduction Recent years have witnessed remarkable progress in open domain conversation (non-task oriented dialogue system) (Ji et al., 2014; Li et al., 2016a) due to the easy-accessible conversational data and the development of deep learning techniques (Bahdanau et al., 2014). One of the most difficult problems for open domain conversation is how to model the conversation context. A conversation context is composed of multi- ple utterances, which raises some challenges not existing in the sentence modeling, including: 1) topic transition; 2) plenty of coreferences (he, him, she, it, they); and 3) long term dependency. To tackle these problems, existing works either re- fine the context by appending keywords to the last turn utterance (Yan et al., 2016), or learn a vector representation with neural networks (Serban et al., ∗This work was done when the first author was an intern at Microsoft XiaoIce. Response: I think coffee is so bitter. Context Rewrite Model Response Generation/Retrieval Model Context: I hate drinking coffee. Last Utterance: Why? It`s tasty. Rewritten-Utterance: Why hate drinking coffee? It`s tasty. Figure 1: An Example of Context Rewriting. 2017b). However, these methods have drawbacks, for instance, correct keywords cannot be selected by heuristics rules, and a fix-length vector is not able to handle a long context. We propose a context rewriting method, which explicitly rewrites the last utterance by consider- ing the contextual information. Our goal is to gen- erate a self-contained utterance, which neither has coreferences nor depends on other utterances in history. By this means, we change the input of chatbots from an entire conversation session to a rewritten sentence, which significantly reduces the difficulty of response generation/selection since the rewritten sentence is shorter and does not has redundant information. Figure 1 gives an example to further illustrate our idea. The last utterance contains the word “it” which refers …

Content truncated to 500 words