Automatic Utterance Generation by Keeping Track of the Conversation's Focus within the Utterance Window

摘要

The insufficiency in methods for generating utterances still remains as a critical issue unsolved in the community of non-task-oriented conversation. Previous studies provide various strategies to enrich the methods for generating utterances, thus making the conversation systems or agents appear more interesting. However, none of them could escape from the fact that they all generate utterances depending mainly on some particular kinds of templates or augmented templates. We propose here in this paper a thorough modification to a preceding work to address this problem. Specifically, we first introduce a concept Utterance Window to strengthen the association between continuous utterances, and then employ a Two-starting-word Markov connection to cope with the ease of losing focus of the current utterance. In addition, we try to keep tract of user's interests and reflecting them in the process of topic-word extraction and utterance generation as well. The experimental results show the effectiveness of our method.