Theory revision and probability (Q1183721)

The notion of ``epistemic entrenchment'' of propositions has been used by \textit{P. Gärdenfors} and the reviewer as a way of defining operations of contraction and revision over a theory [see their ``Revisions of knowledge systems using epistemic entrenchment'', in: Theoretical aspects of reasoning about knowledge, Proc. 2nd Conf., Pacific Grove CA 1988, 83- 95 (1988; Zbl 0711.03009)]. Of the five conditions on a relation \(\leq\) over propositions required for it to be an epistemic entrenchment relation, for a theory \(K\), only one depends on the choice of \(K\), but it does so in a very sensitive manner: if \(K\) is consistent then \(K\) should consist of just the propositions \(\alpha\) such that \(\alpha\nleq\beta\) for some \(\beta\). The purpose of the paper under review is to provide a computationally attractive way of constructing epistemic entrenchment relations that ``minimizes'' the role of \(K\). The author defines a ``preference relation'' over propositions to be any transitive and connected relation extending classical consequence under which only logical truths are maximal. Thus every epistemic entrenchment relation for every theory \(K\) is a ``preference relation'', but preference relations need not be epistemic entrenchments for any theory. The author shows however that: (1) There is a computationally simple procedure for constructing, out of any ``preference relation'' and any theory \(K\), an epistemic entrenchment relation for \(K\); (2) There is a procedure for constructing, out of any probability distribution over propositions of a countable language, a ``preference relation'' over those propositions.