Toward Learning High-Level Semantic Frames from Definitions

We describe work toward learning semantic frames from definitions in WordNet. Our ultimate purpose is to build a resource of common sense knowledge with at least the depth of FrameNet but with wider coverage. The approach uses the TRIPS parser to identify relationships between concepts which we evaluate with human judges. From those relations we generate a very small domain of frames with encouraging results. We discuss how the relationships can be further exploited in the future and identify obstacles and possible solutions.