No label defined (Q4236375)

Default logic was invented by R. Reiter in 1980 to describe statements like ``If \(A\) is true, and it can be consistently assumed that \(B\) is true, then we can infer \(C\)'' (e.g., ``if a person has a motive and no alibi, then he is a suspect''). Reiter's original semantics for a theory consisting of a set of formulas \(W\) and a set \(D\) of defaults \({A:B\over C}\) is as follows: For every set of formulas \(S\), we can define an extension \(\Gamma(S)\) as the smallest deduction-closed set of formulas \(S'\supseteq W\) for which, for any default \({A:B\over C}\in D\), if \(A\in S'\) and \(\neg B\not\in S\), then \(C\in S'\); then, a ``model'' is defined as a set \(E\) for which \(E=\Gamma(E)\). This semantics covers many useful examples, but in some cases, it leads to counter-intuitive conclusions and has, therefore, to be modified. These examples and the corresponding modifications are described by the author; e.g., one of the problems with the standard default theory is that if we add the conclusions of default reasoning to the set of facts \(W\) and start anew, we may (unexpectedly) get yet new conclusions. The author shows, rather convincingly, that to make formalizations more adequate, it is not sufficient to simply modify Reiter's semantics, it is also necessary to modify the definition of a default theory by explicitly adding to its two previous components \(W\) and \(D\) a third component -- the context. He shows it by giving two examples in which \(D\) and \(W\) are the same, but the intuitive conclusions are different. In the first example, we have two robotic arms. We know that one of them is not functional (broken): \(W=\{\neg F_1\vee \neg F_2\}\), and that an arm is usable (\(U_i\)) unless it's broken: \(D=\{{:F_1\over U_1},{:F_2\over U_2}\}\). In this example, we expect the following two (minimal) ``models'': \(\{F_1,\neg F_2,U_1,\neg U_2\}\) and \(\{\neg F_1,F_2,\neg U_1,U_2\}\). In the second example, we are packing for a trip. We know that the weather cannot be both hot (\(H\)) and cold (\(C\)), i.e., \(W=\{\neg H\vee \neg C\}\), and that if it possible that it will be hot we should take a T-shirt (\(T\)), and if it is possible that it will be cold, we should take a sweater (\(S\)): \(D=\{{:H\over T},{:C\over S}\}\). In this case, the reasonable strategy is to take both \(T\) and \(S\), i.e., the reasonable model is \(\{T,S\}\). Formally, the two situations are equivalent: \(F_1\) corresponds to \(H\), \(F_2\) to \(H\), \(U_1\) to \(T\), and \(U_2\) to \(S\) -- but the desired conclusions are different. The author describes the formalization of contextual default logic, and shows that practically all previously proposed modifications and extensions of classical default logic can be described as particular cases of this new formalism. For the entire collection see [Zbl 0908.90002].