Applications of CAPS Modeling to Strategy Competition and Flexibility in Discourse Comprehension

Abstract : Behavioral data obtained by Goldman and Saul (1990) suggest the need to incorporate competition among reading strategies into text comprehension models. REREADER is a computational model that does that. The model is implemented within the Collaborative Activation-Based Production System (CAPS) architecture (Just and Carpenter, 1990). In CAPS, productions must accumulate sufficient activation to fire and their activations fluctuate depending on other aspects of the system. Total activation is limited by working memory capacity; when this limit is reached active elements lose activation so additional processing, can occur. In REREADER, reading strategies are represented as productions and competition among them is governed by activation thresholds. REREADER operates on a linked set of input propositions that represent the text. Processing, occurs in 4 phases. The first three create a semantic network of propositions linked on the basis of argument overlap and propositional embedding and having differential levels of activation. In phase 4, if sufficient levels of coherence obtain (determined by an evaluation function that includes a motivation parameter), REREADER continues to process new input; otherwise, a rereading strategy operates. Initial tests of the system indicate the plausibility of the computational model. Additional development will incorporate prior knowledge and more sophisticated strategy competition mechanisms.