Situated Language Understanding: An Irrational Analysis
Accurate interpretation of spoken language requires combining two different kinds of information: generic knowledge about language, and situational inferences about the beliefs and intentions of a given speaker. Rational models grant these two kinds of information equal status in spoken language comprehension, assuming that the impact of a given information source depends solely on its accessibility and validity. I will review recent experimental results from visual-world eyetracking that fundamentally challenge these ideas, showing a dissociation in listeners' ability to integrate generic versus situational information. In line with the predictions of rational models, listeners were observed to easily and efficiently integrate generic information. However, inconsistent with these models, they were unable to integrate situational information, in spite of clear evidence they had accessed it in time to impact comprehension. This dissociation in processing efficiency implies that comprehension systems have evolved to optimize the processing of generic rather than situational information.