Boosting Dialog Response Generation

Proceedings of the 57th Annual Meeting of the Association for Computational Linguistics, pages 38–43 Florence, Italy, July 28 - August 2, 2019. c⃝2019 Association for Computational Linguistics 38 Boosting Dialog Response Generation Wenchao Du Language Technologies Institute Carnegie Mellon University wenchaod@cs.cmu.edu Alan W Black Language Technologies Institute Carnegie Mellon University awb@cs.cmu.edu Abstract Neural models have become one of the most important approaches to dialog response gen- eration. However, they still tend to generate the most common and generic responses in the corpus all the time. To address this prob- lem, we designed an iterative training process and ensemble method based on boosting. We combined our method with different training and decoding paradigms as the base model, including mutual-information-based decoding and reward-augmented maximum likelihood learning. Empirical results show that our ap- proach can significantly improve the diversity and relevance of the responses generated by all base models, backed by objective measure- ments and human evaluation. 1 Introduction Sequence-to-sequence models (Sutskever et al., 2014) has become one of the most popular ap- proaches to dialog systems, for it provides a high degree of automation and flexibility. On the other hand, they are known to suffer from the “dull- response” problem (Li et al., 2015). Various re- search attempts have been made to improve the diversity of responses generated by sequence-to- sequence models. One line of research investigate alternatives to maximum likelihood learning and decoding, which is believed to be the main cause of monotonicity. (Li et al., 2015) employed a de- coding objective based on mutual information be- tween contexts and responses; (Li et al., 2017a) used reinforcement learning techniques for train- ing the decoder to generate responses that max- imize pre-defined rewards instead of perplexities; (Li et al., 2017b; Xu et al., 2017) adopted adversar- ial learning, in which a generator is trained to de- ceive a discriminator that tries to differentiate be- tween generated responses and human responses. Beside changing training and decoding objectives, (Liu et al., 2018; Lison and Bibauw, 2017) consid- ered reweighting data points by penalizing those with overly frequent responses or by emphasizing high-quality responses. (Serban et al., 2017; Zhao et al., 2017) introduced stochastic latent variables into their models to capture discourse information on an inter-utterance level. (Shao et al., 2017) ex- perimented with a novel segment-based training and decoding paradigm to help mitigate the prob- lem of redundancy and contradiction. Yet another type of approach has not been in- vestigated in the literature in the context of re- sponse generation – boosting and ensembling, de- spite having been studied for machine translation (Xiao et al., 2010; Zhang et al., 2017). Being a long established machine learning method (Freund and Schapire, 1997), the process typically involves iteratively training multiple models on reweighted instances according to the error of the previous models and combining these models. The idea has been recently revived and extended to generative models and image generation, which also suffers from diversity problem (Tolstikhin et al., 2017; Grover and Ermon, 2018). In computer …

Content truncated to 500 words