Ordinary database systems answer a query by listing the items that satisfy the query. They can be more helpful, however, if they can point out regularities in the data that was examined while answering the query. In this paper, the authors show how this can be done by gathering all the facts in the database that mention the entities appearing in the query answer, and then finding patterns in this set of facts. The technique used is a relatively simple form of induction logic programming that computes least-general generalizations from the collected facts. The authors also report on their extensive testing of their technique and discuss the potential value and utility of such generalizations. In addition to augmenting the usual query answer, these generalizations can also help with the debugging and refinement of databases (knowledge bases).

Overall, this well-written paper is more readable than many inductive logic programming papers. It will interest anyone working with database systems, knowledge bases, natural language question-answering systems, or inductive logic programming.