The imperfect nature of context in Ambient Intelligence (AmI) environments, and the special characteristics of the entities that possess and share the available context information render contextual reasoning a very challenging task. The accomplishment of this task requires formal models that handle the involved entities as autonomous logic-based agents, and provide methods for handling the imperfect and distributed nature of context.

In this talk, we descrinea solution based on the Multi-Context Systems formalism, in which local context knowledge of AmI agents is encoded in rule theories (contexts), and information flow between agents is achieved through mapping rules associating concepts used by different contexts. To handle the imperfect nature of context, we extend Multi-Context Systems with non-monotonic features: local defeasible theories, defeasible mappings, and a preference relation on the system contexts. We present this novel representation model, called Contextual Defeasible Logic, describe its argumentation semantics, propose a sound and complete algorithm for distributed query evaluation, and a number of variants for this algorithm. We conclude with a review of some ongoing work.