Proceedings of the Annual Meeting of the Cognitive Science Society

A standard assumption in linguistic, psycholinguistic, and computational research on pronoun use is that production and interpretation are guided by the same set of contextual factors. Kehler et al. (2008) and Kehler & Rohde (2013) have argued instead for a Bayesian model, one in which pronoun production is insensitive to a class of semantically- and pragmaticallydriven contextual biases that have been shown to influence pronoun interpretation. Here we evaluate the model using a passage completion study that employs a subtle contextual manipulation to which traditional analyses are insensitive, specifically by varying whether or not a relative clause that modifies the direct object in the context sentence invites the inference of a cause of the event that the sentence denotes. The results support the claim that pronoun interpretation biases, but not production biases, are sensitive to this pragmatic enrichment, revealing precisely the asymmetry predicted by the Bayesian analysis. A correlation analysis further establishes that the model provides better estimates of measured pronoun interpretation biases than two competing models from the literature.