Proceedings of the Annual Meeting of the Cognitive Science Society

The current study aims to investigate cross-cultural and gender-related differences in infants' facial expression using a novel, infant emotion detection model – the Hierarchical Multimodal Attention Neural (HMAN) Network – for automated Action Unit (AU) prediction. Analyses investigate how facial AUs (1) are distributed across positive and negative emotion classes, (2) co-occur within positive and negative emotion classes, and how (3) their temporal dynamics unfold over time. Compared to Singaporean infants (n=48), Brazilian infants (n=59) were more likely to lower brows (AU04) during negative emotion, and took longer to fully develop a facial expression. Male infants across countries were more likely than female infants to pull lip corners (AU12), and less likely to lower brows (AU04) during positive emotion. These findings advance theoretical understanding of early emotional development and highlight the need for culturally and contextually sensitive tools for interpreting infant emotion, particularly for the design of diagnostic and intervention tools.