Writing a Natural Language Data Base System

We present a model for processing English requests for information from a relational data base. The model has as its main steps (a) locating semantie constituents of a request; (b) matching these constituents against larger templates called concept case frames; (c) filling in the concept case frame using information from the user's request, from the dialogue context and from the user's responses to questions posed by the system; and (d) generating a formal data base query using the collected information. Methods are suggested for constructing the components of such a natural language processing system for an arbitrary relational data base. The model has been applied to a large data base of aircraft flight and maintenance data to generate a system called PLANES; examples are drawn from this system. 1. Introduct ion The language processing model for the PLANES system for natural language access to a large data base [9,18-21] has evolved over the past two years to a point where we feel that it is now practical to begin constructing such systems for other data bases. Key ideas and assumptions in the model are described below. The model is designed to handle requests by real, casual users, whose only programming language is English, but who have some knowledge of the material in the data base.t We have assumed that users will ask questions which are often ungrammatical, which include many abbreviations, both standard and non-standard, and which use ellipsis (omission of information to be understood in context) and pronouns extensively. (See Malhotra [13] for ideas about the types of things users are likely to type in.) The model is designed to work with a rela-tional model |3,8], Data is viewed as being divided into relations which correspond to files or sets of files in conventional data base terminology. Each relation contains a collection of tuples which correspond to records; each tuple contains one or more domains or fields. A relation can conveniently be thought of as a table, tin the PLANES systems, we have provided easily accessible HELP files to bring a user without data base knowledge to a point where he can use the rest of the system. with each row being a tuple and each column a domain. There are two important reasons for using the relational approach: (1) The relational approach stresses data independence. This means that the user and front end programs are effectively isolated from …