Friday, March 08, 2013
3:30 PM
CSB 703
Mary Swift
University of Rochester
Automatically Deriving Event Ontologies for a CommonSense Knowledge Base
We describe work aimed at building commonsense knowledge by reading word definitions using deep
understanding techniques. The end result is a knowledge base allowing complex concepts to be reasoned
about using OWL-DL reasoners. We show that we can use this system to automatically create a
mid-level ontology for WordNet verbs that has good agreement with human intuition with respect to
both the hypernym and causality relations. We present a detailed error analysis that reveals areas of
future work needed to enable high-performance learning of conceptual knowledge by reading.