The purpose of this study is to propose a speech meaning acquisition model, which can be applied for an adaptive interface system, from a perspective of human-human communication establishment process. The model was designed to discriminate the types of instructions based on salient prosodic features and to recognize the given instructions using a positive reward (given for its successful action) and a negative reward (from the utterance of a instructor which draws listener's attention). As a result of a test, this model could eventually learn to recognize the given instruction from an actual human instructor. It is expected that the constructed meaning acquisition model can be applied for an adaptive interface system, which provides a natural interaction environment for its user.
