Donald Bren School of Information and Computer Sciences

A computer system that results in a useful and quite natural vehicle in which fact retrieval systems can be constructed quite easily is presented and analyzed. The system, called TRAMP, consists of a simulated associative machine providing the storage structure, and a relational language for that associative storage structure providing a question-oriented data structure. The TRAMP system is a computer language with applicability to a class of problems that are best represented as associations between objects, or by a relational data structure. In particular, the language contains potent fact retrieval capabilities in the sense that it automatically deduces implications of facts resident in its store.

We use the term "fact retrieval" here to mean the extraction of facts from a data store regardless of whether or not the information explicitly resides in the store. Thus, fact retrieval includes "document" retrieval--simply locating and extracting a data item from memory (primary or secondary)--but is mainly concerned with strategies for making inferences from data in memory. As a very simple illustration: if we know that A is greater than B, then from that data item we should be able to conclude that B is less than A. To more clearly define the problem it is necessary to place constraints on the deduction. For our purposes these constraints will have to do with time, space, ease of application, and general utility in a realistic environment.

The TRAMP language and system is described and motivated. Comparison of this system to other fact retrieval approaches is discussed. The implementation of the system (on an IBM 360/67) is described. Of perhaps most importance is the algorithm, along with its proof of correctness, for processing the relational language, thereby yielding a very simple manner of handling what has proved to be an elusive problem. The language is critically evaluated, examples given of how it works and how it can be used, a general discussion and critique of the field of fact retrieval, and the user's manual are also included.