Abstract
The neural coding of faces is typically assumed to reflect a norm-based "face space" in which individual faces are represented by how far and in what direction they differ from the average. We asked to what extent observers have cognitive access to the metrics of this space. To explore a space with limited and well-defined dimensions, stimuli were restricted to basic expressions and their antifaces for a common average identity, with all faces created with FaceGen Modeller. Individual faces spanned different directions and magnitudes within planes defined by two different expressions and their anti-expressions. In one case we tested how well observers could identify the opposite direction or "antiface" of a given face within each plane. Observers were presented with a reference face angle in the plane and asked to select its opposite angle. Some pairs produced choices reflecting conceptual associations (e.g. sad is the opposite of happy), and configural shape opposites were sometimes also selected, yet in general the settings showed little clear bias toward the angular opposite. Most observers were consistent in their settings, while responses across observers were idiosyncratic. In a second experiment, observers adjusted the magnitude of a test expression until it matched a reference faces expression intensity. Both face/anti-face pairs and different-expression pairs were used. Settings could be made reliably for some expressions but did not always increase consistently or monotonically with reference intensity. Together these results suggest that although the neural representation of faces may reflect the geometry of a space, we are not in general able to consciously access this geometry to infer the metrical relationship between different facial configurations.
Meeting abstract presented at VSS 2014