Representation of Requirements through Preference Orderings of Soft Constraints

Constraint satisfaction problems have generally been recognized as a useful framework for many AI problems. Several notions of soft constraints have been suggested to infer uncertain knowledge as well as formulate optimization tasks. Additionally, the ability to represent optional requirements of experts on schedules or plans has been claimed by certain forms of soft constraints. In opposition this paper shows, that the problem of representing optional requirements is not solved, yet. This problem must especially not mixed up with the representation of uncertain knowledge. This insight is founded by practical experience as well as theoretical re ections. On the one hand, we worked on a nurse scheduling application, that demanded the representation of more or less important requirements of nurses. On the other hand, we developed a unifying view on the well known notions of soft constraints, i.e. weighted and prioritized constraints and constraint hierarchies. Then, from this view we formulated some conditions for representations of optional requirements. Thus, a (preliminary) set of criteria is obtained to assess, whether a formalism is appropriate to represent requirements in scheduling or con guration systems. Finally, the several notions of soft constraints are discussed with respect to theses criteria. 1