Proceedings of the Annual Meeting of the Cognitive Science Society

Previous research has found that comprehenders are willingto adopt non-literal interpretations of sentences whose literalreading is unlikely. Several studies found evidence that com-prehenders decide whether or not a given utterance should betaken at face value in accordance with principles of Bayesianrationality, by weighing the prior probability of potential inter-pretations against the degree to which they are (in)consistentwith the literal form of the utterance. While all of these re-sults are consistent with string-edit noise models, many errorprocesses are known to be sensitive to the underlying linguis-tic structure of the intended utterance. Here, we explore thecase of exchange errors and provide experimental evidencethat comprehenders’ noise model is structure-sensitive. Ourresults add further support to the noisy-channel theory of lan-guage comprehension, extend the set of known noise opera-tions to include positional exchanges, and show that compre-henders’ noise models are well-adapted to structure-sensitivesources of signal corruption during communication.