Abstract
The own-race recognition advantage has been attributed to other-race faces being densely clustered in multi-dimensional face space (Valentine, 1991) rendering discrimination especially difficult. However, training protocols that emphasize discrimination (between-person variability) had only modest effects (McGugin et al., 2011). Relying on tightly controlled images ignores another important challenge: recognizing identity despite within-person variability in appearance (make-up, expression, viewpoint, lighting). When faces are unfamiliar, two images of the same identity often are perceived as belonging to different people, especially when viewing other-race faces (Laurence et al., 2015), perhaps because they have smaller attractor fields (Tanaka et al., 1998). We hypothesized that incorporating both between- and within-person variability among young Black female faces would increase training effectiveness. During training, Caucasian participants (n=20 per group) learned either 1 image of 12 identities (between-person variability only) or 5 images of 6 identities (both between- and within-person variability) to criterion (100% correct). A control group (n=13) received no training. Before and after training participants made same/different judgments for pairs of novel identities (not seen during training) comprising two different pictures of the same identity or pictures of two different people. Identities were paired randomly and presented simultaneously. Accuracy (d') showed a modest improvement, p = .02, an effect driven by improved performance on same trials, p=.03. However, the magnitude of improvement did not vary across groups, p=.83, suggesting training was ineffective. In an ongoing study, we are investigating the impact of more intensive training by extending training over 5 days with new identities introduced daily. During test phases, identities are matched for physical description, stimuli are presented sequentially (to avoid image matching), and include both novel identities and new images of trained identities. These results have implications for models of perceptual expertise and our representation of faces in multi-dimensional face space.
Meeting abstract presented at VSS 2016