Causal language about social interactions

Causal language is central to our understanding of social interactions—whether someone “caused” or “allowed” another’s action shifts our impression of what happened. Yet models of causal language use have largely focused on physical events (e.g., billiard balls), ignoring the beliefs and preferences implicated in human action. We present a computational framework and three experiments investigating how people use causal expressions (“caused,” “enabled,” “allowed,” “made no difference”) across physical, epistemic, and preference-based interventions between agents. We find that people prefer different causal expressions across these intervention types: they describe removing a physical obstacle as a different form of facilitation than providing information. We capture people’s language use with a model that selects utterances based on counterfactual simulations of events, inferences about agents’ mental states, and utterance informativity. This model explains human judgments better than baseline models, suggesting that describing social influence involves reasoning about mental states, alternative actions, and alternative utterances.