ABSTRACT

Generally, our model for length is about 80% correct in its predictions. In this chapter and the next we will try to account for the 20% of wrong predictions, especially the poor results associated with predictions 46–52 of Table 4 (i.e., predicting success on one theorem-task from success on two other theorem-tasks). To achieve this, a “competence-performance,” or “competence-automaton,” distinction will be invoked. We shall not simply explain away embarrassing failure by the model, however, but will extend the model so as to yield testable predictions.