Item

Abstract

Biological movements can be easily recognised from point-light stimuli. It is still unclear how the visual system accomplishes this recognition. Some properties of biological-motion recognition, eg the inversion effect, suggest that recognition is based on learned templates. This hypothesis predicts that humans should be able to learn arbitrary new movement patterns, even if they do not correspond to movements of a real biological organism. We tested this hypothesis in a very direct way by creating artificial movements through motion morphing. Using a special technique, we computed linear combinations of very dissimilar prototypical movements that were obtained by motion capturing. Such linear combinations specify very similar low-level motion information as that specified by the prototypes, but define movements that cannot be realised by a human. The stimuli were presented as point-light walkers in a discrimination experiment with a pair comparison paradigm. By varying the weights of the linear combination we could gradually vary the spatiotemporal similarity of the stimuli. We found a robust learning effect for the discrimination task, but only if subjects received feedback during training. Also, we found partial transfer between upright and rotated figures. We interpret these results as evidence that learning might play a fundamental role in the recognition of biological motion.