Invariant versus context-specific representation of face shape and motion in the face network

A growing consensus suggests that separate brain pathways process facial shape and motion. However, a region may respond preferentially to one type of information (e.g., shape) while still being modulated by another (e.g., motion), similar to modulatory effects observed in non-classical receptive fields of visual neurons. To test this possibility in the face network, we applied a two-test strategy-combining cross-decoding and context-sensitivity analyses-to determine whether face-selective areas encode shape and motion in an invariant or context-specific manner. Twelve participants viewed videos generated from 3D synthetic face models in which facial shape and motion were manipulated independently. We report four key findings. First, shape and motion information could be decoded from all face-selective regions, suggesting overlapping encoding of shape and motion. Some, but not all, findings can be accommodated by models proposing separate pathways. Second, we observed distinct invariance patterns: OFA showed motion-specific shape representations, IFG encoded shape invariant to motion, and FFA encoded both shape and motion invariantly-challenging models that restrict motion processing to dorsal areas. Third, including motion involving different facial poses-in which motion and shape are confounded-inflated invariance estimates across the network, highlighting the importance of controlling for pose-motion confounds. Fourth, individual-level analyses revealed substantial variability. These results call for revised models incorporating representational overlap, context sensitivity, and individual variability.