Proceedings of the Annual Meeting of the Cognitive Science Society

Cognitive modelling uses computer models to investigate psy-chological theories. To conclude from executions of a cogni-tive model to the theory, the model needs to be a correct im-plementation of the theory since a defective cognitive modelmay yield wrong statistical figures. We consider three commonreasons for a model to be incorrect wrt. a theory: situationswhich unintentionally do not enable any production rule, ruleswhich erroneously construct undesired declarative knowledge,and wrongly chosen architecture parameters. Defects of thesekinds are hard to detect since repeated execution and observa-tion of the model does not guarantee to uncover these defects.In this work, we give formal definitions of the three kinds ofdefects in terms of an existing abstract formal semantics of thehybrid architecture ACT-R. We demonstrate the application offormal analysis techniques to ACT-R models to reliably detectthe considered defects and to thereby increase the confidencethat the model behaves according to the psychological theory.