How do societies create shared categories? This longstanding question in social science has been approached through two dominant and competing paradigms. The first paradigm is the cognitivist approach which holds that people come to share categories because they possess innate predispositions to represent the world’s structure in similar ways. The second paradigm is the social constructivist approach which holds that people construct meaning dynamically in social networks, where interaction can lead to radically different category systems across societies. The social construction account is consistent with recent experiments which contend that communication in dyads and small social groups can amplify individual variation, leading separate communities to converge on dissimilar representations of the same stimuli. However, recent formal models illustrate that agents with minimal biological constraints can converge on similar categories when interacting in social networks. In this study, we use formal models and online experiments to challenge both of these paradigms in accounting for how categories emerge in social networks. Specifically, we develop novel theories and analyses to explain how categories emerge in social networks of various sizes and topologies.