Turning null responses into quality responses

Natural language interfaces to database systems free the user from the undue formalism of learning a query language. However, the more important issue from the system's point of view. besides correctly interpreting the natural language query. is to respond correctly and cooperatively. In particular, the generation of quality responses has proven problematical in situations when null values arise. If a user's query cannot be answered by the system because of an incorrect assumption that the user has made. i t would be appropriate for the system to inform the user what has gone wrong. The query. "Whut grade did John Doe get in MATHIOO?", might result in a null answer because John is. not enrolled in MATH100. It is important for a system to realize these kinds of complementary issues in order to be truly "natural". Most recent work on cooperative responses has relatively overlooked null value events. In this thesis, we present an initial classification of null event problems in 8 natural language database systems. and methods for responding with appropriate answers to some classes of null events. Assuming that the database is relational and that the database query language is SQL, we develop and incorporate a knowledge base into the system based on the RM/T model, an extended relational model proposed by E. F. Codd, to furnish information for diagnosis of failed queries. The knowledge base, which is also in the relational model, consists of meta-level data information. This knowledge base provides information such as: entity concept hierarchies: generalization and aggregation hierarchies: entity relationships and entity relationship constraints;' event precedence in the database domain; and knowledge about null values existing in the database. Algorithms which further explicate the function of the knowledge base model are given to demonstrate the kind of quality responses we can obtain instead of a simple null answer.