ABSTRACT
A theory of discourse comprehension is described that is based on the idea that the construction of mental representations is largely a locally controlled, bottom-up process, governed by weak, inaccurate rules, and thus not very sensitive to contextual constraints. Context exerts its influence after an approximate but incoherent and contradictory network representation has been generated through a constraint satisfaction process. Those elements of the network that hang together begin to dominate the network and deactivate the irrelevant or contradictory parts. How this model works is illustrated by simulating the comprehension of a brief text. Predictions are derived for the free recall of the text, and for the existence of both contextual, discourse priming and non-contextual, associative priming effects. Experimental data support these predictions.
