Mutual belief forming by robots based on the process of utterance comprehension

摘要

This paper describes a method of learning a mutual belief, necessary for natural language communication with people, in a process of utterance comprehension by robot. In this method, a system of mutual beliefs is represented by multiple beliefs, and the weights for the confidence that each of the belief is shared with a human. The beliefs dealt with in the method include phonemes, lexicon, grammar, influence of behavioral context, and other nonlinguistic belie* In the experiment, arm-robot with only a basic language knowledge interacts with human by using language and action. Through the interaction, robot learned the mutual beliefs and was eventually able to understand even fragmental and ambiguous utterances according to given situations, and act appropriately. The methods made it possible to integrate the language and cognitive processes, and thus introducing a new framework for natural communication.