Learning Structures to Represent Verb Meaning

Abstract

A program that learns structures that represent verb meaning is described. Inputs are combinations of surface sentences and environment snapshots. The program learns to associate surface verbs with conceptual dependency networks that describe the changes that occur in the environment when an action is performed. Learning is performed by one of five processes, determined by now well current knowledge accounts for an input. Each process is described as well as the circumstances under which is it performed and the effects it has on the conceptual dependencies.