Comparing action descriptions based on semantic preferences

We consider action domain descriptions whose meaning can be represented by transition diagrams. We introduce several semantic measures to compare such action descriptions, based on preferences over possible states of the world and preferences over some given conditions (observations, assertions, etc.) about the domain, as well as the probabilities of possible transitions. This preference information is used to assemble a weight which is assigned to an action description. As an application of this approach, we study the problem of updating action descriptions with respect to some given conditions. With a semantic approach based on preferences, not only, for some problems, we get more plausible solutions, but also, for some problems without any solutions due to too strong conditions, we can identify which conditions to relax to obtain a solution. We conclude with computational issues, and characterize the complexity of computing the semantic measures.