Abstract
The ventral visual pathway contains rich representations of objects, with information about both their visual properties and category membership at multiple hierarchical levels, including animate versus inanimate. These neural representations show general agreement with behavioral similarity judgments and with representational similarities in "deep" convolutional neural networks. In this event-related functional neuroimaging study (n = 16), we challenge this state-of-the-art by dissociating object appearance (how does the object look like?) from object category (which object category is it?). The stimulus set includes animate objects (e.g., a cow), typical inanimate objects (e.g., a mug), and, crucially, inanimate objects that look like the animate objects (e.g., a cow-shaped mug). Behavioral judgments and deep neural networks showed a strong effect of the animacy dimension, setting the lookalike (and inanimate) objects apart from the animate ones. In contrast, neural activity patterns in ventral occipitotemporal regions were strongly biased towards object appearance: animate entities and lookalikes were similarly represented, and separated from the regular inanimate objects. Animacy, despite its importance for human behavior and neural networks, is not well represented in ventral visual cortex.
Meeting abstract presented at VSS 2018