Abstract
We propose a hierarchical model incorporating motion time series data, motion symbols and words. This paper describes a linguistic space which represents a network of the words linked to full body motions. The linguistic space is constructed by using the dissimilarity among words, which can be computed from the association probability of the words and motion symbols. In the linguistic space, words that are semantically similar are located close to one another and included in the same cluster. We validate our approach by constructing a motion symbol space based on the dissimilarity between words, which improves discrimination ability when compared to a motion space constructed based on dissimilarities between motions alone.
