Abstract
A motion capture system is a useful tool to realize humanoid robots or CG characters that can perform human-looking behaviors. Although the amount of captured data increases, we can't necessarily make efficient use of the capture data corpus. We have studied the imitative learning model based on motion recognition/generation and communication model. Especially, language is important factor for high level communication ability. In this paper, we aim to intelligent information processing which is capable of acquiring elementary language, and apply it to construction of capture database. Language model consists of relation and alignment between words and motion representation. These two phases lead to semantic theory and grammatical structure of language.
