Revising Relational Domain Theories

This paper presents the AUDREY system which can revise incomplete and incorrect relational domain theories given a set of examples and non-examples of concepts in the theory. Using an abductive process, the system identifies clauses that must be added, removed or generalized to modify the theory to be consistent with the training set. The generalization process generates first-order Horn clauses. We present experimental results demonstrating the system's ability to repair incorrect domain theories as well its ability to learn relational concepts.