Abstract
Imitation is a common and effective way for humans to learn new behaviors. Until now, the study of imitation has been hampered by the challenge of measuring how well an attempted imitation corresponds to its stimulus model. We describe a new method for quantifying the fidelity with which observers imitate complex series of gestures. Wearing a data glove that transduced movements of their digits, subjects viewed and then reproduced a sequence of gestures from memory. The velocity profile of each digit’s flexion or extension was used to segment movements made during an imitation into gestures that can be compared against corresponding gestures in the stimulus model. The outcome is a multivariate description of each imitation, including its temporal characteristics, as well as spatial errors (in individual gestures and in the ordering of those gestures). As a demonstration, we applied this method to data from an imitation learning experiment with gesture sequences. With repetition, overall fidelity of imitation improved, with various aspects of the imitation improving at different rates. Confirming the approach’s usefulness, when we varied the complexity associated with imitation, that variation was robustly reflected in our measures of imitation quality. Finally, we describe a simple way to extend our methods to make them useful not only in assessing imitation and imitation learning, but also in various settings in which the detection and characterization of subtle abnormalities in movement production is paramount.