Unsettling preferential semantics (Q1986085)

The well-known Hansson preferential semantics for conditional obligation, dating back to 1971, notoriously validates the formula \(O(A/A)\). It often felt that while such a validation may be no more than a harmless limiting case when the condition \(A\) is unchangeable, it is quite inappropriate when the condition can be undone. With this in mind, several alternative semantics have been proposed that avoid that theorem, whether by indexing the preference relation with worlds, replacing it with a selection or neighbourhood function, or by making a more radical abstraction as may be found in input/output logics. The purpose of the paper under review is to reach a similar result but within the preferential framework. With preference relations taken to be a strict transitive orders < with no infinite ascending chains, the simplest version of the Hansson valuation rule puts \(O(X,A)\) to be true at a world \(w\) iff \(X\) is true at every <-maximal \(A\)-world of the model; the author's proposed rule requires instead that \(X\) is true at all maximal worlds of the model that are either themselves \(A\)-worlds or <-better than some \(A\)-world. For both rules, the truth-value of \(O(X,A)\) at world \(w\) is independent of the choice of \(w\). The author notes that the logics generated by these two rules differ not only on \(O(A/A)\) but also in several other respects, for example, only the logic arising from the latter rule validates a monotony formula and is closed under `factual detachment', i.e. passage from \(A\) and \(O(X/A)\) to \(O(X)\) (under a suitable definition of the monadic operator \(O(X)\), different from the usual \(O(X/T)\)). A final section locates the author's logic with respect to the more abstract spectrum of input/output logics; roughly speaking, for an appropriate fragment of the logic's language, it corresponds in all but certain limiting cases to basic output.