Proceedings of the Annual Meeting of the Cognitive Science Society

Traditionally, linguists have treated ambiguity as a bug in the communication system, something to be avoided or ex-plained away. More recent research has taken notice of the efficiency ambiguity affords us. The current work identifies anadditional benefit of using ambiguous language: the extra information we gain from observing how our listeners resolveambiguity. We propose that language users learn about each others private knowledge by observing how they resolveambiguity. If language does not do the job of specifying the information necessary for full interpretation, then listenersare left to draw on their private knowledgeopinions, beliefs, and preferencesto fill in the gaps; by observing how listenersfill those gaps in, speakers learn about the private knowledge of their listeners. We implement this hypothesis as a com-putational model within the Rational Speech Act framework. We then test our hypothesis by using the model to predictbehavioral data from naive participants.