Moving Beyond Dyadic Similarity and Towards Latent Structure Learning in Social Group Inference

Abstract

Successful groups require coordination specifically amongst its members. One crucial process underlying this is social categorization—the ability to accurately categorize others as in-group or out-group members. Much research examining social categorization has done so through the lens of specific, static groups, wherein targets are explicitly labeled. Indeed, previous cognitive neuroscience research examining social categorization has suggested that simple dyadic similarity is enough to underlie social categorization. These paradigms, however, do not account for the fact that in the real world, 1. we have multiple group memberships, and the group membership brought to bear in a given situation is context-dependent, and 2. novel others do not often come explicitly labeled. We will address these problems through three papers examining social categorization. Paper 1 demonstrates that a domain-general network, rather than areas associated with self-referential processing, is associated with explicit, generalized in-group categorization. Paper 2 describes an account as to how we might accumulate information about novel others in order to gauge group memberships—inferring latent group structures—and tests this against dyadic similarity and category label accounts of social categorization. Paper 3 examines the neural correlates of tracking both latent group structures and dyadic similarity and demonstrates that separable areas track each. Given the sharp increase in polarization across societies today, understanding the mechanisms underlying how we form these concepts of group memberships and their impacts on our judgements and decisions remains as important as ever.