In the context studied here, learning occurs in response to an interpretative task where the theoretical model is provided in a declarative format to the learners. Consequently, learning foremost implies setting up matches between the knowledge associated with the perception of the world and entities of the new knowledge domain, all being subjected to constraints from both knowledge domains. We have gathered evidence about failed attempts at matches by the subjects, and the "repair strategies" they use to overcome these obstacles, consisting mainly in changes of the knowledge representation primitives, activation of new classes, and creation of new ones.