Abstract
Individuals differ in the stimulus boundaries they select for categorizing faces on dimensions such as gender and ethnicity (Webster et al. Nature 2004). We asked whether these differences reflect differences only in criteria or have links to sensory differences in how faces are perceived. To test for these links, we used adaptation to identify the face image that was neutral with regard to the adapting category. Adapting to a male face causes an ambiguous face to appear more female, while adapting to a female face biases appearance in the opposite direction. Thus an intermediate adapting stimulus can be found that does not shift the judgments, and defines the neutral point for the visual processes affected by the adaptation. We tested whether these perceptual neutral points are correlated with subjects' category boundaries before the adaptation. Stimuli were morphs between a male and female face forming a graded series of 100 images. A 2AFC staircase was used to determine the gender boundary in the sequence, before adaptation or after adapting to male/female blends in proportions of 0/1, .2/.8, .4/.6, .6/.4, .8/.2, or 1/0. Adapt stimuli were shown initially for 30 s and then for 3 s before each 0.5 s test trial. The neutral adapt level was estimated from linear fits of the after-effect vs. adapt level. Results for 23 subjects tested with two face pairs showed comparable ranges of variation in the zero-crossing for adaptation and the subjective gender boundary (though the two means differed), and a significant correlation between the two alternative ways of defining a gender-neutral image (r = 0.63, p < 0.01). At intermediate adapt levels near the mean for the group, there was also a high correlation (r∼−0.5, p < 0.05) between the direction of shift and the individual's pre-adapt neutral point. These results suggest that at least part of the variation in how observers categorize faces may depend on actual differences in how faces are perceptually encoded.