CHANGING THE THEORY OF THEORY CHANGE - TOWARDS A COMPUTATIONAL APPROACH

The theory of theory change has contraction and revision as its centra l notions. Of these, contraction is the more fundamental. The best-kno wn theory, due to Alchourron, Gardenfors, and Makinson, is based on a few central postulates. The most fundamental of these is the principle of recovery: if one contracts a theory with respect to a sentence, an d then adds that sentence back again, one recovers the whole theory. R ecovery is demonstrably false. This paper shows why, and investigates how one can nevertheless characterize contraction in a theoretically f ruitful way. The theory proposed lends itself to implementation, which in turn could yield new theoretical insights. The main proposal is a 'staining algorithm' which identifies which sentences to reject when c ontracting a theory. The algorithm requires one to be clear about the structure of reasons one has for including sentences within one's theo ry.