Using CSPs to encode TMSs

A truth maintenance system (TMS) is a general problem solving facility designed to work in tandem with an inference engine. Justification-based TMSs (the most commonly used TMSs) have not a sufficient expressive power for many applications. Logic-based TMSs overcome this limitation but are untractable. So, McAllester, or Forbus and de Kleer have proposed to encode formulae in clauses and to run the efficient Boolean Constraint Propagation algorithm (BCP), what lose the completeness of the deductions. In this paper, we introduce the model of Constraint Satisfaction Problems (CSPs) underlining that the notion of local completeness (called local-consistency in CSPs) has already been widely treated in CSPs. We then propose an encoding of McAllester's TMS in Dynamic CSP to be able to use CSPs techniques. We show that not only we can compute as many deductions as BCP for the same timeconsuming, but also we can outperform the local completeness computed for a slight more running-time. Finally, we say that we can run other well known CSPs techniques dealing with the disposition of the constraints between them to solve with a good improvement the central problem in TMSs: finding all contradictions.