Keywords:utterance intention understanding, spoken dialogue system, information transfer
We are developing a conversation system which efficiently transfers a massive amount of information like news articles by spoken dialogue. Here, "efficient" means that only the necessary information is transferred except unnecessary information for the user from target articles. In our system, feedbacks from the user are indispensable in order to realize high EoIT (Efficiency of Information Transfer). Therefore, we propose a utterance intention recognition method combining language information and prosodic information for the purpose of understanding diverse feedbacks from users. The feature of the proposed method is that it automatically extracts prosodic features with a high contribution ratio by using deep learning. We confirmed the effectiveness of the proposed method using a corpus with utterance intention tags designed based on dialogue data collected using our conversation system.