Are interlocutors as sensitive to over-informativeness as they are to under-informativeness?

Research by Engelhardt, Bailey and Ferreira (2006) suggests that speakers avoid under-informativeness and listeners penalise it, but that neither of these behaviours apply to overinformativeness. We argue that what may appear as lack of sensitivity to over-informativeness is in fact preference for information that, even though it is not contributing to the unique identification of a referent, it is nevertheless addressee-oriented and pragmatically motivated because it minimizes the risk of ambiguity. In experiments 1A and 1B, where we remove the addressee-oriented motivation for over-informing that was implicit in Engelhardt et al’s experiments, we document that speakers avoid and listeners penalise overinformativeness (though less than under-informativeness). This supports the Gricean prediction of some leniency towards over-informativeness relative to under-informativeness but nevertheless sensitivity to both. In a further experiment (2), increased visual salience of an attribute, typically understood as a non-Gricean, non-addressee oriented factor, was also found to contribute to increased rates of over-informativeness in production. We conclude that a careful study of the factors that affect the production and comprehension of referring expressions is needed to identify cases where over-informing is in fact the optimal choice for speakers, or for listeners, or for both. Computational modelling of the relevant constraints could lead to further testable predictions.