Abstract
Recognizing faces is thought to rely on holistic processing, where the entire face is recognized as a unitary visual object (e.g., Farah et al., 1998). Recent work using analyses from General Recognition Theory (Ashby & Townsend, 1986; GRT) that distinguishes between possible loci of holistic effects has found stronger evidence for a decisional rather than perceptual component (Wenger & Ingvalson, 2002, 2003; Thomas, 2001; Richler et al., 2006); these results introduce important constraints for models of face processing. Using this same GRT framework, the present research identified the loci of holistic processing in simulations produced by one well-known model of face recognition by Cottrell and colleagues (Cottrell et al., 2002) that accounts for the typical kinds of holistic effects observed in people. While this model accounts for holistic effects, is the locus of this holistic processing perceptual or decisional? We first trained the model to identify a collection of individual faces, and then tested the model using a same-different composite face task, using other faces. In this task, a study face is shown, then a test face is shown where the top and bottom can be the same or different as the test face; same-different judgments are made to both parts in turn. As with data from people, the model's same-different responses to the tops and bottoms of the test faces were analyzed using Multidimensional Signal Detection Analysis (Kadlec, 1999). While the results showed some evidence for a decisional component, as observed with people, evidence for a perceptual component was far more consistently found, a divergence from holism localized in people. These results suggest the necessity of refining existing models of face recognition to incorporate potential decisional sources of holism in face processing.
This research was funded by a grant from the James S. McDonnell Foundation and NSF Grant HSD-DHBS05