Challenging the predictive coarse-to-fine hypothesis in emotional face perception

Abstract

Introduction Predictive processing (PP) accounts conceptualize perception as an interaction between top-down predictions and bottom-up sensory inputs. In the visual domain, low spatial frequencies (LSF) convey coarse information, supporting early predictions, refined by high spatial frequencies (HSF). Emotional expressions may thus be processed coarse-to-fine (CtF), although evidence remains inconsistent. Methods Across two online experiments (N=56), participants judged the valence (positive/ negative) of unambiguous (happy, sad) and ambiguous (morphed) faces filtered in LSF or HSF, and presented in CtF or fine-to-coarse (FtC) sequences. As sad faces were poorly recognized in Exp. 1 (LSF cutoff = 4.6 cpi), Exp. 2 used a higher cutoff (9.2 cpi). Results Contrary to our CtF hypothesis, HSF and FtC sequences led to more accurate and faster responses to happy and sad faces (Exp 1.) than LSF and CtF sequences, but only for happy faces in Exp. 2. In Exp. 1, positive mood facilitated positive morph interpretations, while in Exp. 2, LSF morphs showed faster and more frequent negative categorizations. Conclusion Findings varied with small methodological changes, highlighting that emotional face perception is highly complex. Unlike the robust CtF pattern in scene perception, our results challenge a CtF account for facial expressions and suggest that SF manipulations alone may not probe predictive mechanisms in emotional perception.