This article proposes an extension to Kintsch's Construction-Integration (CI) model of text comprehension, which changes its mathematical implementation and emphasizes the connectionist features of the model. Specifically, the extension proposed here (a) simulates the learning process in a connectionist manner by making explicit changes in the connecting values among processing units; (b) takes into account individual differences in prior background knowledge activation and individual ability to make logical inferences, using these parameters to adjust the results of the simulation; and (c) implements an algorithm that constructs the connectivity matrix W from text processing. The proposed extension is tested on existing recall and contradiction-detection data from readers of science texts, and its predictions fit the empirical data better than the CI model's prediction.