One issue for context-aware applications is to identify without delay situations requiring reactions. The identification of these situations is computed from both dynamic context information and domain specific knowledge. This identification is the output of a process involving context interpretation, aggregation and deduction. In smart environments, these treatments have to be efficient since they may be partly performed on constrained mobile devices. Two main approaches exist in the literature: process-oriented and ontology-based context management. In this paper, we claim that they are complementary and we propose an architecture which integrates the two approaches. We show in a scenario how context-aware applications can benefit from this architecture both to scale to numerous mobile users and to identify complex situations.