| The research in this dissertation describes how to optimize the learning of factual information. To optimize memory for facts in a principled manner, this work uses and develops an ACT-R (Adaptive Character of Thought - Rational) (Anderson & Lebiere, 1998) model that describes the underlying paired-associate memory task. This model is applied to determine the schedule of practice during learning with the goal of maximizing performance at test. The overall result of these investigations strongly favored the efficacy of the optimization algorithm described. For the final experiment in this dissertation, the algorithm was compared with other learning options including a replication of Atkinson (1972a). This comparison suggested that the algorithm results in significant benefits with relatively large effect sizes for both improved recall and improved recall latency. The algorithm was able to achieve these benefits through careful attention to the economics of the unit task to enable a principled maximization of the learning rate. To do this, the problem of optimizing learning was decomposed into two concerns regarding practice scheduling. First, one wants to allow as much spacing as possible between practices to maximize the spacing effect (the advantage for distributed practice). Second, one wants to allow as little spacing as possible to prevent longer retrieval latencies and to reduce failure costs. The optimization algorithm achieved its effects by balancing these opposing concerns. While many researchers have advocated attending to spacing effects, the fine-grained attention to the cost of spacing in this dissertation is novel. |
