On Rules vs. Paths in SNePS 8 November 2008 William J. Rapaport http://www.cse.buffalo.edu/~rapaport ======================================================================== There's no clear principle on when to use rules vs. when to use paths. In some cases, you can use either one: Suppose: Fido is a dog. Dogs are mammals. So, Fido is a mammal. "Dogs are mammals" can be represented either as: (assert forall ($x $y $z) &ant (build member *x class *y) &ant (build subclass *y superclass *z) cq (build member *x class *z)) Or using paths, as: (define-path class (compose class (kstar (compose subclass ! superclass)))) Which to use? Rule-based inference is "conscious", in the sense that Cassie explicitly believes the rule. When she uses it, she **infers** the conclusion. Path-based inference is "subconscious", in the sense that she has no explicit beliefs about the paths. When she uses it, she "already" **believes** the conclusion. For a demo, see rules-vs-paths.demo and rules-vs-paths.script. For discussion, see: Stuart C. Shapiro. Cables, paths and ``subconscious'' reasoning in propositional semantic networks. In John F. Sowa, editor, Principles of Semantic Networks, chapter 4, pages 137-156. Morgan Kaufmann, San Mateo, CA, 1991. http://www.cse.buffalo.edu/~shapiro/Papers/cablesnpaths.ps Arguably, you might use rules when you firmly believe that a relationship is best expressed by a universally quantified rule and paths when you think that the relationship might be defeasible (e.g., "birds fly"). You could even use path-based inference with rules. E.g., given the rule about membership in superclasses, above, you could define a *path* through the rule and use path-based inference instead of rule-based inference. I'll leave the details to your imagination. For a paper on when this might be appropriate, see: S. L. Peters, S. C. Shapiro, and W. J. Rapaport. Flexible natural language processing and Roschian category theory. In Proceedings of the Tenth Annual Conference of the Cognitive Science Society, pages 125-131. Lawrence Erlbaum Publishers, 1988. http://www.cse.buffalo.edu/~rapaport/Papers/Papers.by.Others/peters.10annual.pdf