Polarization is typically conceptualized as increasing distance between opposing attitudes. We argue instead that it unfolds through asymmetries in the group-differentiating power of opinion language. In public discourse, actors use distinctive linguistic repertoires to signal alignment, mark boundaries, and render opinion landscapes intelligible. Conventional attitude scales estimate the distributions of private belief but cannot capture how these boundaries are enacted in expression or how normative asymmetries shape the visibility of positions. Across three studies of vaccination discourse in South Africa and the United States, combining human coding with machine learning models of natural language, we examine how majority and minority opinions differ in their capacity to differentiate social groups. Consistent with Moscovici's theory of minority influence, minority discourse became more linguistically distinctive— and more effective at separating communities—under politicized conditions, whereas majority discourse was more dispersed and aligned with institutional reporting. In both contexts, linguistic proximity to opinion poles was associated with heightened anger. These findings suggest that polarization is not only divergence in belief but also an asymmetrical process of boundary construction enacted through language.
Durrheim et al. (Thu,) studied this question.