Official URL: https://escholarship.org/uc/item/4zv1z80z

In this paper we outline a cognitive theory and model of surprise judgements which aims to explain how and why some events are considered to be surprising in a piece of text , while others are not. The model is based on a series of experiments carried out by Grimes-Maguire and Keane (2005a), which show that subtle changes in the predictability of a discourse can have a profound effect on a reader’s perceived surprise at certain events. Rather than defining surprise in terms of expectation, we conceive of it as a process involving Representation-Fit. We have implemented this theory in a computational model that has two stages: the Integration stage entails building a coherent representation of the scenario by means of an objective knowledge base rooted in WordNet. The Analysis stage then outputs a surprise rating for a specified event, based on the degree to which that event can be supported by the prior representation. Simulations reveal a strong correspondence between model and participant generated surprise ratings.