Abstract
In the current study, we use the error-from-sample (EFS) paradigm (McEntire and Schwartz, VSS06) to determine which internal face dimensions are used in upright and inverted processing, respectively. EFS is a delayed match-to-sample XAFC paradigm in which the alternatives do not contain the target, but instead contain probes that each differ from the sample in a single dimension by a magnitude equal to the subject-specific JND for that dimension. Because the subject believes that a correct answer exists, the selection process requires subjects to passively discount any alternatives based on information about the sample that remains strongest in subject memory; the subject then selects the alternative that violates this information the least. As a result, the frequency of errors for a given dimension is inversely proportional to the relative strength of that dimension in subject memory, and is inversely proportional to the extent to which the subject is using each dimension to perform the matching decision.
JNDs for part dimensions (eyebrow, eye, nose, and mouth shapes) and configural dimensions (eyebrow-eyebrow, eyebrow-eye, eye-eye, eye-nose, and nose-mouth distances) were measured for each subject using morphed faces presented in an adaptive threshold estimation procedure in a same-different task. Subjects then performed a 3AFC-EFS task in which probes differed by 1 JND unit from the sample stimulus. Confirming earlier findings, subjects committed relatively fewer configural errors when viewing upright faces and relative fewer part errors when viewing inverted faces. However, this change accounted for a 5.1% shift in error rates with subjects still committing part errors on over 60% of inverted trials. These results suggest that while subjects may use part and configural information differentially as a function of face orientation, they still rely significantly more on configural information than part information regardless of the orientation of the face when making a matching decision.