Abstract
The field of open-domain conversation generation using deep neural networkshas attracted increasing attention from researchers for several years. However,traditional neural language models tend to generate safe, generic reply withpoor logic and no emotion. In this paper, an emotional conversation generationorientated syntactically constrained bidirectional-asynchronous frameworkcalled E-SCBA is proposed to generate meaningful (logical and emotional) reply.In E-SCBA, pre-generated emotion keyword and topic keyword are asynchronouslyintroduced into the reply during the generation, and the process of decoding ismuch different from the most existing methods that generates reply from thefirst word to the end. A newly designed bidirectional-asynchronous decoder withthe multi-stage strategy is proposed to support this idea, which ensures thefluency and grammaticality of reply by making full use of syntactic constraint.Through the experiments, the results show that our framework not only improvesthe diversity of replies, but gains a boost on both logic and emotion comparedwith baselines as well.