Towards the Automatic Detection and Correction of Errors in Automatically Constructed Ontologies

The Open Information Extraction Project is one of the most ambitious attempts in the area of automatically constructing ontologies by harvesting information from the web. What we will call their KnowItAll Ontology contains about 6 billion items, consisting of triples and rules. The downside of such automatically constructed ontologies is that they contain a vast number of errors: some arising from errors in the original web data and some from errors in extracting the data. In this project we explore whether techniques we have developed in the domain of ontology repair can be used to detect and correct some of these errors. In particular, we explore whether the errors in their ontology can be automatically detected by using a theorem prover. We also present a manual classification of the errors as a preliminary feasibility exploration, and discuss our future work towards automatically correcting the ontology based on the error classification.