Expression in human and illusory faces (pareidolia) shows cross-domain serial dependence: evidence for common processing

Recent data from primates suggests illusory faces (face pareidolia) and real faces are processed by common mechanisms. We use a serial dependence approach in human observers to test whether pareidolia and face images cross-prime each other for expression. Forty pareidolia and 40 face images spanning four levels of expression from negative (angry) to positive (happy) were selected from a pre-rated image-set. Presentation time was 250 ms and observers rated expression by clicking on a scalebar. Face condition: 40 faces presented 20 times each in random order. Pareidolia condition: 40 x 20 pareidolia images in random order. Cross-domain condition: 80 images (faces+pareidolia) presented 20 times in randomly interleaved order. Ratings for each face were combined into a mean expression for that face. Mean expression ratings for faces and pareidolia validated the pre-experiment ratings and clustered into 4 evenly spaced expression levels. To test for serial bias, the rating on the previous trial was subtracted from the current-trial’s mean rating. Serial independence would yield no systematic bias, yet both image categories produced a positive serial dependence: current trial ratings were biased towards the preceding trial’s expression rating. This was true for faces and for pareidolia, although the serial dependence was stronger for pareidolia. In the cross-domain condition with randomly interleaved image categories, the serial analysis was run separately for sequences where a face followed pareidolia and where pareidolia followed a face. Both orders of cross-domain sequences produced a significant positive dependence for expression. These results show that face pareidolia can be systematically and reliably rated for expression, comparably to ratings of real faces, and that face or pareidolia image sequences both exhibit a serial dependence for expression. Finally, interleaving both image types produces a clear cross-domain serial dependence and therefore implies common processing of human and illusory faces.