COMPILING SPECIFICITY INTO APPROACHES TO NONMONOTONIC REASONING

We present a general approach for introducing specificity information into nonmonotonic theories. Historically, many approaches to nonmonoto nic reasoning, including default logic, circumscription, and autoepist emic logic, do not provide an account of specificity, and so fail to e nforce specificity among default sentences. In our approach, a default theory is initially given as a set of strict and defeasible rules. By making use of a theory of default conditionals, here given by System Z, we isolate minimal sets of defaults with specificity conflicts. Fro m the specificity information intrinsic in these sets, a default theor y in a target language is specified. For default logic the end result is a semi-normal default theory; in circumscription the end result is a set of abnormality propositions that, when circumscribed, yield a th eory in which specificity information is appropriately handled. We mai nly deal with default logic and circumscription although we also consi der autoepistemic logic, Theorist, and variants of default logic and c ircumscription. This approach differs from previous work in that speci ficity information is obtained from information intrinsic in a set of conditionals, rather than assumed to exist a priori. Moreover, we deal with the ''standard'' version of, for example, default logic and circ umscription, and do not rely on prioritised versions, as do other appr oaches. The approach is both uniform and general, so the choice of the ultimate target language has little effect on the overall approach. ( C) 1997 Elsevier Science B.V.