Extending Association from Human Motion Symbolsby Using Latent Semantics of Digital Book Library

10.6567/IFToMM.14TH.WC.OS13.128

Ikuo Kusajima；Wataru Takano；Yoshihiko Nakamura

Natural Language ； Motion Symbol ； Concept Acquisition ； Motion Language Association ； Parallel Computing

Proceedings of the 14th IFToMM World Congress

14th-4（2015 / 11 / 04）

400 - 405

英文

This paper proposes the methods to associate various sentences with motions using stochastic models. The motion document language model, which associates the words with the motions through the latent concepts extracted from motions and documents stochastically. The Ngram language model represents the natural sequence of words. Sentences can be associated with motion symbols using the motion document language model and N-gram language model by rearranging words generated from motion. The sentence association algorithm is parallelized with MPI. Our method of sentence association was tested on captured human motion and a large corpus of digital books, and its validity was demonstrated. And the parallel computation is executed on cloud server of Amazon EC2.