Preferred Explanations for Quantified Constraint Satisfaction Problems

Abstract

The Quantified Constraint Satisfaction Problem(QCSP) is a generalization of the classical constraint satisfaction problem in which some variables can be universally quantified. This additional expressiveness can help model problems in which a subset of the variables take value assignments that are outside the control of the decision maker. Typical examples of such domains are game-playing, conformant planning and reasoning under uncertainty. In these domains decision makers need explanations when a QCSP does not admit a winning strategy. We present an approach to defining preferences amongst the requirements of a QCSP, and an approach to finding most preferred explanations of inconsistency based on preferences over relaxations of quantifiers and constraints. This paper unifies work from the fields of constraint satisfaction, explanation generation, and reasoning under preferences and uncertainty.