The Neural Processing of Face-like Stimuli.

Abstract

Facial expressions are a primary channel for conveying emotional and social information, yet the neural mechanisms underlying their perception across different types of face-like stimuli remain incompletely understood. This dissertation will report an EEG - multivariate pattern analysis (MVPA) study that investigated how emotional expressions are represented in the brain when conveyed through real human faces, symbolic emoji faces, and ambiguous pareidolic faces (face-like objects), across three experiments. Participants viewed faces displaying four emotions-happy, angry, sad, and neutral-and performed two-alternative forced-choice (2AFC) tasks tailored to each stimulus type. We examined whether emotional expressions could be decoded from EEG patterns within each stimulus category using leave-one-subject-out decoding, both for 4-class classification and pairwise emotion contrasts. Next, we assessed the generalizability of emotion-related neural patterns by performing cross-decoding across stimulus types. Our results demonstrate robust within-condition decoding of emotional expressions in all three experiments and reveal significant cross-condition generalization, indicating shared neural representations of emotion across biologically and symbolically rendered faces. We also explored whether ambiguous face-like stimuli elicit a neural bias toward male gender perception, as suggested by prior behavioural studies. Cross-decoding from real face gender signals to emoji and pareidolic stimuli revealed above-chance classification accuracy, suggesting that ambiguous faces are more likely to be processed as male at the neural level. Together, these findings demonstrate both the flexibility and the constraints of neural representations involved in face and emotion perception. They suggest that the human brain employs a generalised, modality-independent coding scheme for facial expressions, while also exhibiting categorical biases when interpreting ambiguous social cues.