ACT-R models of training data entry skills

This paper presents an important application of cognitive modeling, that of training perceptual, cognitive, and motor skills. We developed an ACT-R model for a simple, but important and general data entry task. This same ACT-R model was matched to three different published data sets. This exercise helped us understand the cognitive mechanisms involved in prolonged work and the suppression of subvocal rehearsal. The model provides an integrated explanation on the repetition priming and depth of processing effect in skill learning. Specifically, we explained how prolonged work results in distinctive effects on response time and accuracy using the ACT-R cognitive mechanisms of production compilation and activation noise. Production compilation is a mechanism that combines productions, reducing response time; increasing activation noise in retrieval produces increased retrieval problems, and thus typing error rate. Likewise we explained the cognitive processes in two other data sets. To us, however, the most important value of modeling is prediction. As such, we generated predictions about three important general questions: how performance deteriorates with different delays after training, how different amounts of rehearsal during training affect retention, and how re-training helps retention of skills. Although this effort is only a first step in an important research program, this research demonstrates that our ability to predict the effects of training manipulations will largely depend on the development of cognitive models.