Un système de maintenance de la vérité à propagation de contextes. (A context-propagation based truth maintenance system)

Hypothetical reasoning consists in completing available knowledge to achieve reasoning. In order to help hypothetical reasoning, specific algorithms are designed which efficiently deal with hypothesis and their consequences, and moreover allow an automatic hypothesis generation. This last consideration leads to implement a nonmonotonic reasoning. Truth maintenance systems (TMS) record the inferences, made by a reasoning system, as a dependency graph and insure the coherence of the formulas to be found in the knowledge base. Two kinds of truth maintenance systems were considered: • Propagation-based TMS deal with nonmonotonic inferences and propagate validity through the graph. Labels associated with formulas represent an interpretation of the graph. • Context-based TMS are restricted to monotonic inferences but use labels representing ! #"# $ % & ')( # *+ , . & / 0 #1 2 / 3 & contexts simultaneously. This work designs a new system that merges the advantages of both systems. It allows to reason simultaneously under several contexts with the help of nonmonotonic inferences. For this purpose, new environments able to take into account the absence of an hypothesis are defined. They are associated with an interpretation which is extended all over the dependency graph according to the propagation-based TMS’ interpretation. A system corresponding to this characterization, the CP-TMS was implemented as an extention of propagation-based TMS. That implementation is described here before to be discussed.