In this paper, the authors present a representation model used for Large Knowledge-Based Systems (LKBS). Since the design of such systems falls into a new research area, the choice of a “best” representation model is as yet an open question. The authors begin by stating a few criteria for such a model, and from there they describe their proposed approach.

The Qualified Binary Relationship Model of Information (QBRM) is based on, but is an improvement over, semantic networks. Overcoming the disadvantages of the latter is achieved by attaching to each arc a label and a truth value. The labels allow each Well-Formed Formula (WFF) in the model to appear as a term in another WFF. Since each arc is explicitly qualified, the truth value of an assertion cannot necessarily be defined implicitly by its presence. To be noted in particular is their usage of the truth value “UNDEFINED.” Thus, as opposed to the semantic networks, where the basic unit of information is a triple (two nodes and a connecting arc), in QBRM this unit is a quintuple of the type: <label> <entity> <relation-name> <entity> <logical value>.

Finally, the authors propose the low-level hardware support: an Associative Predicate Store (APS). Also presented are some data about the access times of operations for a 4M prototype. The data are attractive, but it should be born in mind that it is with respect to triples, and that the label referencing mechanism is not implemented. The real benefits of the system can be assessed when the missing components are added and new tests are performed.

The present paper is a highly technical one proposing a new formalism for models which are on the (as yet) delicate bridge between conventional DBMS and classical AI. Though not too long, it makes very heavy reading and should not be taken up as a fill-in article. The reader is supposed to have a good background in mathematical logic and reasoning and should be acquainted with formal languages such as PROLOG. It should also be noted that the paper is well presented, having no typing errors and a good visual set-up. The references are ample and are well marked. Overall, the paper promises a new trend in this relatively unexplored area, but further experimentation is needed before an unbiased judgment can be formed.