Automated Cognitive Modeling

In this paper WC dcscribc an approach to automating the construction of cognitive process models. WC make two psychological assumptions: that cognition can bc modclcd as a production system, and that cognitive behavior involves starch through some problem space. Within this framework, WC employ a problem reduclion approach to constructing cognitive mod&, in which one begins with a set of indcpcndcnt, overly gcncral condition-action rules, adds appropriate conditions to each of thcsc rules, and then rccombincs the more specific rules into a final model. Conditions arc dctcrmincd using a discrimination learning method. which rcquircs a set of po’;itivc and ncgativc instances for each rule. Thcsc instances are based on infcrrcd solution paths that lead to the same lnswcrs as those obscrvcd in a human subject. We have implcmcntcd ACM, a cognitive modeling sjstcm that incorporates thcsc methods and applied the system to error data from the domain of multi-column subtraction problems.