Cumulative default logic: In defense of nonmonotonic inference rules (Q1187215)

The author's purpose is to modify Reiter default logic in such a way as to eliminate two apparent formal drawbacks. One is failure of cumulativity in the generated inference relation: the addition, to a set of premises, of conclusions inferred from them may change the set of conclusions. The other is failure of joint consistency of justifications: although each justification formula in each default rule applied in generating an extension must individually be consistent with that extension, the justifications taken together need not be so. The former drawback can arise even for normal default systems; the latter only for non-normal ones. The strategy followed by the author is to ``upgrade the ontology'': instead of dealing with plain propositions, he considers propositions \(p_ R\) indexed by sets \(R\) of propositions representing possible ``supports'' for \(p\). The author redefines the notion of an extension in such a way that in its generation the indices in \(R\) expand to keep track of the justification appealed to and also the intermediate conclusions passed through. It is then shown that the resulting system of ``cumulative default logic'' (CDL) is indeed cumulative and satisfies joint consistency of justifications. The author also outlines a variant of his construction, obtained by the reviewer, for ensuring cumulativity without requiring joint consistency of justifications. Reviewer's comments on subsequent developments: Systems such as CDL do however suffer from a formal drawback of their own --- the phenomenon of ``floating conclusions''. It can happen that a certain proposition occurs in every extension, but only with different indices, so that there is no indexed instance of it common to all the extensions, with the result that nothing about it can be inferred (sceptically) via the intersection of all extensions. This phenomenon was noted by the author, the reviewer and \textit{K. Schlechta} [``Cumulative inference relations for JTMS and logic programming'', in: Nonmonotonic and inductive logic (\textit{J. Dix} et al. (eds.)) (Berlin: Lect. Notes Artif. Intell. 543 (1991)]. To eliminate floating conclusions whilst retaining the other desirable properties, the author has recently proposed further refinement of the constructions insolved. See his technical report ``A framework for cumulative default logics'' [TR-92-042, Int. Comput. Sci. Inst., Berkeley/Calif. (1992)].