Providing students with appropriate additional texts from a course (i.e., readings) requires cognitively demanding human activity. Moreover, the content of such readings might not be adequate to the ongoing student understanding of the course. Systems that provide such readings are either traditional hypertexts using links designed by hand, or adaptive hypertexts using a sophisticated model of the student. The method described here uses LSA, a statistical procedure devoted to the semantic comparison of texts. The method mimics some characteristics of human discourse understanding through a simplified version of the Construction-Integration model. This method thus retrieves course readings from the web that take into account the ongoing comprehension of the course by the students