In a variety of domains, children have been observed to over-regularize inconsistent input, while adults are more likely to "probability match" to any inconsistency. Many explanations for this have been offered, usually relating to cognitive differences between children and adults. Here we explore an additional possibility: that differences in the social assumptions participants bring to the experiment can drive differences in over-regularization behavior. We explore this in the domain of language, where assumptions about error and communicative purpose might have a large effect. Indeed, we find that participants who experience less pressure to be "correct" and who have more reason to believe that any inconsistencies do not correspond to an underlying regularity do over-regularize more. Implications for language acquisition in children and adults are discussed.