Predicting Learning and Retention of a Complex Task

We use an ACT-R model of a complex task to explore the implications of ACT-R's learning and forgetting mechanisms to better understand learning and retention. The model performs a task that has 14 non-iterated subtasks that takes approximately 25 min. to perform the first time. The results show that a typical learning curve is generated by the model that is well fit to human data. When decay is examined we find that the retention curves basically match the shapes predicted by the KRK theory, and that training and testing have been confounded in many studies. From these results we see that the previously hypothesized mixed declarative procedural stage of learning actually starts on the first trial and is never completely exited, so we will need to propose other thresholds to mark transitions between declarative, mixed, and proceduralized knowledge. We predict based on this model that learning and retention will vary greatly by task components, practice schedule, and learner's strategy.