Classical approaches such as frames, scripts,... have been unable to deal with the kind of inferences necessary in natural language processing situations such as text comprehension. Shastri & Ajjanagadde (1993), proposed a local connectionist model for the sort of reasoning requiring such a fast inference. The problem with this system is that it controls only the adequacy of argument- fillers, leaving untouched the activation control issue, namely, why we perform certain inferences, and not others, in a given situation. The aim (...) of this paper is to examine how a focus, in the sense of Grosz (1981). could operate as a theoretic constraint with the above reasoner system to handle two aspects: inference control and anaphora resolution (i.e., antecedent activation), during text understanding. (shrink)