A Foundation of Solution Methods for Constraint Hierarchies

Constraint hierarchies provide a framework for soft constraints, and have been applied to areas such as artificial intelligence, logic programming, and user interfaces. In this framework, constraints are associated with hierarchical preferences or priorities called strengths, and may be relaxed if they conflict with stronger constraints. To utilize constraint hierarchies, researchers have designed and implemented various practical constraint satisfaction algorithms. Although existing algorithms can be categorized into several approaches, what kinds of algorithms are possible has been unclear from a more general viewpoint. In this paper, we propose a novel theory called generalized local propagation as a foundation of algorithms for solving constraint hierarchies. This theory formalizes a way to express algorithms as constraint scheduling, and presents theorems that support possible approaches. A benefit of this theory is that it covers algorithms using constraint hierarchy solution criteria known as global comparators, for which only a small number of algorithms have been implemented. With this theory, we provide a new classification of solution criteria based on their difficulties in constraint satisfaction. We also discuss how existing algorithms are related to our theory, which will be helpful in designing new algorithms.