Automatic generation and inferring semantic structure of verbal instructions for a motor task
Takeuchi Ryoto, Tomoya Tamei · 2020
Expert instructors frequently use verbal instructions in motor learning coaching. The instructors find points to be improved in learners' forms and give advice. However, the verbal expressions strongly depend on instructors' characteristics, and the process of choosing instruction is not systematic in many fields. This study aims to develop an agent that can automatically generate proper verbal instructions from learners' motion data, and to systematize the instructions given by various expressions. We collected shooting motion data of free-throws in basketball from novice players, and verbal instructions to the novices' shooting motions from experienced players. Using multi-label classification, a type of supervised learning, we developed an agent that generates essential instructions to the novices' motions. We also suggested a possibility to extract semantic relationships among instructions with various expressions in the context of motor learning by using network structure inference in graph theory.