Let's chat with a computer!

Conversational systems are actively investigated recently, since they are effective not only counseling or entertainment purpose, but also improvement of performance in task-oriented dialogues. Our aim is to develop conversational system that can respond to any kind of topics. To realize appropriateness of utterances and wide range of topic coverage simultaneously, we propose an approach that integrates a rule-based utterance generation method and a statistical method that leverages web sentences to create utterances. Our proposed approach estimates appropriate methods based on concordance rates between user utterances and rules, or word frequencies in web-corpus. Our approach generated more appropriate utterances rather than individual methods.

Please click the thumbnail image to open the full-size PDF file.