Contextual Reasoning

It is widely agreed on that most cognitive processes are contextual in the sense that they depend on the environment, or context, inside which they are carried on. Even concentrating on the issue of contextuality in reasoning, many different notions of context can be found in the Artificial Intelligence literature, see for instance [Giunchiglia 1991a, Giunchiglia & Weyhrauch 1988, Guha 1990, Guha & Lenat 1990, Shoham 1991, McCarthy 1990b]. Our intuition is that reasoning is usually performed on a subset of the global knowledge base; we never consider all we know but only a very small subset of it. The notion of context is used as a means of formalizing this idea of localization. Roughly speaking, we take a context to be the set of facts used locally to prove a given goal plus the inference routines used to reason about them (which in general are different for different sets of facts). Our perspective is similar to that proposed in [McCarthy 1990b, McCarthy 1991]. The goal of this paper is to propose an epistemologically adequate theory of reasoning with contexts. The emphasis is on motivations and intuitions, rather than on technicalities. The two basic definitions are reported in appendix A. Ideas are described incrementally with increasing level of detail. Thus, section 2 describes why contexts are an important notion to consider as part of our ontology. This is achieved also by comparing contexts with situations, another ontologically very important concept. Section 3 then goes more into the technical details and proposes that contexts should be formalized as particular mathematical objects, namely as logical theories. Reasoning with contexts is then formalized as a set of deductions, each deduction carried out inside a context, connected by appropriate "bridge rules". Finally, section 4 describes how an important example of common sense reasoning, reasoning about reasoning, can be formalized as multicontextual reasoning.