The Effects of Order: A Constraint-Based Explanation
Stellan Ohlsson · 2007
Abstract This chapter presents a computational model that shows how information migrates from declarative to procedural knowledge and provides a powerful new learning mechanism for machine-learning algorithms. The model is used to examine the effects of learning three different counting tasks. The model predicts order effects that vary in several dimensions, including the number of times the model has to revise its knowledge and how long it will take to learn. Although some of these effects are quite large within a subtask, the overall effect is muted by other aspects of the task, including interaction. This model suggests that the complexity of a task's constraints is important for computing transfer between similar tasks. The model's behavior has been compared to human performance, and a general summary is provided.