WhyNot: Debugging Failed Queries in Large Knowledge Bases

Abstract

When a query to a knowledge-based system fails and returns unknown, users are confronted with a problem: Is relevant knowledge missing or incorrect? Is there a problem with the inference engine? Was the query ill-conceived? Finding the culprit in a large and complex knowledge base can be a hard and laborious task for knowledge engineers and might be impossible for non-expert users. To support such situations we developed a new tool called as part of the PowerLoom knowledge representation and reasoning system. To debug a failed query, WhyNot tries to generate a small set of plausible partial proofs that can guide the user to what knowledge might have been missing, or where the system might have failed to make a relevant inference. A first version of the system has been deployed to help debug queries to a version of the Cyc knowledge base containing over 1,000,000 facts and over 35,000 rules.