Abstract
Many recent studies have revealed that face recognition heavily relies on the processing of horizontal spatial orientations. However, most of those studies used tasks where it is difficult to dissociate the impact of physical face information from that of identity-specific information. To investigate this issue, we used a method designed to precisely control the physical difference between stimuli, and verified the horizontal tuning for faces of identical distances with regard to low-level properties but of different perceptual distance with regard to identity. Ten participants each completed 2,880 trials in a 2-ABX match-to-sample task. On each trial, the participants saw a target and two response alternatives, both sampled with the same orientation bubbles (Duncan et al., 2017). One response choice was visually identical to the sample (i.e. the correct response) whereas the other was either on the same side (within-identity [WI]) or on the other side (between-identity [BI]) of the categorical barrier. Thus, the physical distance between the target and the different (WI or BI) alternative was always the same, but the perceptual distance was not. As expected, WI trials were more difficult than BI trials for all participants, as indicated by the higher number of bubbles needed for the former (WI: M=101.66, SD=83.50) than the latter (BI: M=15.85, SD=14.94). Orientation tuning in the BI and WI conditions was revealed by computing a weighted sum of the orientation filters across trials, using participant accuracies as weights. In the BI condition, horizontal orientations between 62 and 101degrees were significantly associated with accuracy (Zcrit=2.101; Zmax=4.25, p< 0.05, peak at 84 degrees); whereas no orientation reached the threshold in the WI condition (Zmax=1.41, p>0.05). Comparing horizontal tuning between the two conditions using a paired sample t-test reveals an identity-specific horizontal tuning for faces, t(6) = 2.8, p < 0.05.