Bias-Driven Revision of Logical Domain Theories