Interpreting Natural Language Database Updates

Although the problems of querying databases in natural language are well understood, the performance of database updates via natural language introduces additional difficulties. This thesis examines the problems encountered in interpreting natural language updates, and describes an implemented system that performs simple updates. The difficulties associated with natural language updates result from the fact that the user will naturally phrase requests with respect to his conception of the domain, which may be a considerable simplification of the actual underlying database structure. Updates that are meaningful and unambiguous from the user's standpoint may not translate into reasonable changes to the underlying database. The PIQUE system (Program for Interpretation of Queries and Updates in English) operates by maintaining a simple model of the user and interpreting update requests with respect to that model. For a given request, a limited set of "candidate updates"— alternative ways of fulfilling the request—are considered, and ranked according to a set of domain-independent heuristics that reflect general properties of "reasonable" updates. The leading candidate may be performed, or the highest ranking alternatives presented to the user for selection. The resultant action may also include a warning to the user about unanticipated side effects, or an explanation for the failure to fulfill a request This thesis describes the PIQUE system in detail, presents examples of its operation, and discusses the effectiveness of the system with respect to coverage, accuracy, efficiency, and portability. The range of behaviors required for natural language update systems in general is discussed, and implications of updates on the design of data models are briefly considered. This research, within the Knowledge-Based Management Systems (KBMS) project, has been supported by ARPA contract N39-82-C-250. The views and conclusions in this document are those of the author and should not be interpreted as representative of the official policies, either expressed or implied, of the U.S. Government This is a revised version of a thesis submitted to the Department of Computer Science and the Committee on Graduate Studies of Stanford University in partial fulfillment of the requirements for the degree of Doctor of Philosophy. Author's current address: Teknowledge, Inc. 1850 Embarcadero Rd Palo Alto, CA 94303 Phone:(415)424-0500