The Reciprocity of Semantic Alignment: When Misunderstanding Begets Misunderstanding

ABSTRACT Establishing shared meaning is widely considered the normative, almost ubiquitous trajectory of linguistic interaction. Dominant models in dialog research typically treat alignment as an automatic process or an intentional effort to build common ground, where misalignment is merely a temporary disruption prompting repair. However, it is not uncommon to experience communication breakdowns that actively escalate rather than resolve. This raises a fundamental question: does interaction naturally lead people toward shared meaning, or is semantic alignment a reciprocal, socially contingent process? I will present findings from four experiments utilizing a novel multiplayer word association task designed to probe semantic alignment and track how individuals adjust their behavior in response to a partner's representations. We found that while social interaction generally motivates individuals to produce more conventional or "typical" semantic associations to support alignment, this tendency is highly context sensitive. Particularly, when individuals interact with an atypical or difficult-to-align-with partner, they actively reduce their own typicality, effectively sacrificing semantic alignment. This "reciprocal misalignment" persists even when participants are explicitly instructed and incentivized to maximize agreement. Furthermore, interacting with a misaligned partner inherently diminishes an individual's motivation to coordinate and results in more negative social evaluations. By reframing divergence not as a failure of coordination, but as a reciprocal, socially motivated modulation of similarity, this work sheds light on the interactional mechanisms behind interpersonal distancing, communication breakdowns in neurodivergence, and potentially broader patterns of social polarization.