Unclear Distinctions Lead to Unnecessary Shortcomings: Examining the Rule Vs Fact, Role Vs Ller, and Type Vs Predicate Distinctions from a Connectionist Representation and Reasoning Perspective

Unclear Distinctions lead to Unnecessary Shortcomings: Examining the rule vs fact, role vs ller, and type vs predicate distinctions from a connectionist representation and reasoning perspective Venkat Ajjanagadde Wilhelm-Schickard Institute, Universitaet Tuebingen Sand 13, D-72076 Tuebingen, Germany [email protected] Abstract This paper deals with three distinctions pertaining to knowledge representation, namely, the rules vs facts distinction, roles vs llers distinction, and predicates vs types distinction. Though these distinctions may indeed have some intuitive appeal, the exact natures of these distinctions are not entirely clear. This paper discusses some of the problems that arise when one accords these distinctions a prominent status in a connectionist system by choosing the representational structures so as to re ect these distinctions. The example we will look at in this paper is the connectionist reasoning system developed by Ajjanagadde & Shastri(Ajjanagadde & Shastri 1991; Shastri & Ajjanagadde 1993). Their1 system performs an interesting class of inferences using activation synchrony to represent dynamic bindings. The rule/fact, role/ ller, type/predicate distinctions gure predominantly in the way knowledge is encoded in their system. We will discuss some signi cant shortcomings this leads to. Then, we will propose a much more uniform scheme for representing knowledge. The resulting system enjoys some signi cant advantages over Ajjanagadde & Shastri's system, while retaining the idea of using synchrony to represent bindings. Introduction Given a particular piece of knowledge, can one unambiguously decide whether it is a rule or a fact? Are there entities which always act as roles and never as llers? Are there entities which always act as llers and never as roles? What is a type and what is a general predicate? In spite of the fact that the rule/fact, role/ ller, type/predicate distinctions get mentioned not too infrequently in general AI parlance, an attempt to clearly state the distinctions faces di culties (Some of the difculties will be listed in the following section). This paper illustrates that taking these rather unclear distinctions and according them prominent representa1This paper was written in third person for double-blind reviewing. tional status in a connectionist network may not be a desirable thing to do. Speci cally, the example we consider here is the connectionist reasoning system(Ajjanagadde & Shastri 1991; Shastri & Ajjanagadde 1993) developed by Ajjanagadde & Shastri (Henceforth A & S). Their system performs an interesting class of inferences extremely fast. A major idea underlying their approach is the use of activation synchrony to represent dynamic bindings. We consider the idea of using synchrony to represent bindings to be indeed e cient, elegant, and as discussed in (Ajjanagadde & Shastri 1991; Shastri & Ajjanagadde 1993), neurologically plausible. However, the system of A & S has some shortcomings. These shortcomings are due to the representational methodologies A & S have chosen and are not due to the use of synchrony itself. The major reason for the shortcomings of their representational schemes can be diagnosed to be the prominence A & S have accorded to the distinctions of rules & facts, roles & llers, types & predicates. The representational structures in their system directly re ect these distinctions. For example, Fig. 1 shows how A & S encode the following knowledge base: give(x,y,z) ) own(y,z) ; buy(x,y) ) own(x,y); own(x,y) ) can-sell(x.y) ; give(john,mary,book1); buy(mike,house3)