Psychology Publications
Document Type
Article
Publication Date
2017
Journal
Cognitive Science
Abstract
We present a connectionist model of event knowledge that is trained on examples of sequences of activities that are not explicitly labeled as events. The model learns co-occurrence patterns among the components of activities as they occur in the moment (entities, actions, and contexts), and also learns to predict sequential patterns of activities. In so doing, the model displays behaviors that in humans have been characterized as exemplifying inferencing of unmentioned event components, the prediction of upcoming components (which may or may not ever happen or be mentioned), reconstructive memory, and the ability to flexibly accommodate novel variations from previously encountered experiences. All of these behaviors emerge from what the model learns.
