This model focuses on procedural representations to model analogical learning in the simple physics task. For each problem, the model learns a procedural schema for the "quantity equation" that represents a mapping from a physics problem to its solution. The model studies the sample problem and learns the quantity, constant, or operator that goes in each position of the desired answer expression. This mapping is initially represented as a dependency which links the solve-equation subgoal to the appropriate response subgoals. When popped from the goal stack, the dependencies generate new productions that execute the correct steps for any new problem.
The terminal model represents subject behavior on the final (eighth) set of five problems. This model provides a good fit to the empirical data in terms of total latency per problem, number of references for each area item, and average reference times for each area.
The learning model represents subject behavior on all eight sets of problems. This model differs from the terminal model in only two respects: the enabling of strength learning, and the modification of one goal slot value. The model provides a good fit to the empirical data in terms of average latency and correctness per set. Because of the time needed to run a simulation, only the ACT-R model is provided; this model can be run locally as described in the model file.