ABSTRACT
The preceding chapters presented the main principles of the Construction-Integration Model (Kintsch, 1988, 1998), a hybrid architecture devoted to describing and modeling knowledge use in text comprehension. This model has two phases: a construction phase and an integration phase. During the construction phase, a large and complex associative propositional network is built. It results from the association of two propositional nets: the textbase network, directly built from the linguistic input, and the knowledge network, in which each textbase proposition serves as an independent retrieval cue for all close associates in the knowledge net. Although the construction of this last net leads to certain elaborations, another step requires generating bridging inferences and macropropositions. Finally, connection strengths are assigned to all of the element pairs that have been created, depending on theoretical and empirical considerations. One of the main differences between this model and previous models of text com-
prehension is that “weak” and “dumb” rules allow for the construction of this large associative network, assumed to be built during each processing cycle. Thus, at the end of the construction process, the propositional net is rich, but some of the propositions or concepts with their interrelationships, also part of this network, are irrelevant or even contradictory. The integration phase rapidly reinforces contextually appropriate elements while it inhibits those that are irrelevant or contradictory. This is done via an automatic spreading activation process using connectionist procedures. This second phase, which is automatic and not resource consuming, provides an integrated mental representation that corresponds to “a kind of situation model” (Kinstch, 1988, p. 180). Thus, consistent with a bottom-up view of the comprehension process, the construction of a propositional network is the raw material for understanding a text. However, it is not sufficient for attaining a deep level of textual interpretation, and relevant information has to be integrated into the resulting net while irrelevant information has to be deleted from the reader’s representation (i.e., the propositional net). In essence, these two phases describe two crucial mechanisms involved in the reading comprehension process: knowledge activation, which is independent of contextual relevance, and knowledge integration, which is dependent on the situational context. The integration process (i.e., integration phase) involves the updating of the representation, which brings into play several mechanisms such as information retrieval, inhibition of irrelevant or contradictory information, and reinforcement of relevant information. Updating reinforces relevant elements and inhibits those that are irrelevant and will be removed from long-term memory.
