物体操作发话理解确信度推定基发话动作生成-杉浦孔明.PDF

†,†† †,†† † () †† () Generation of Utterances and Motions Based on Confidence Measure Estimation for Utterance Understanding in Object Manipulation Tasks *Komei Sugiura†,†† Naoto Iwahashi†,†† †National Institute of Information and Communications Technology ††ATR Abstract— This paper proposes a method that generates motions and utterances in an object-manipulation task. Belief modules of speech, vision, motions are integrated in a probabilistic framework for understanding user’s utterances. Responses to the utterances are optimized based on a confidence measure function for the integrated belief modules. Fisher’s scoring algorithm is used for training the confidence measure function. Ex- perimental results reveals that the proposed method can improve the success rate in the object-manipulation task. Key Words: motion learning, confidence, logistic regression, spoken dialogue, language acquisition 1. (1) (2) (Fig. 1 ) ( ) (1) (2)