Prior knowledge and statistical models of learning
The research reported here describes the effects of prior knowledge on how people form categories and learn continuous mappings. Chapter 2 is a review of the past research on knowledge effects in the statistical and psychological literature. Chapter 3 presents simulations of a set of experiments carried out by Heit and Bott (2000) into how knowledge is selected in a category learning task. The model was shown to account for the results of Heit and Bott and generate several new predictions concerning blocking effects with the use of prior knowledge. However, empirical testing of these predictions failed to demonstrate these effects. Chapter 4 describes work testing Delosh, McDaniel and Busemeyer's (1997) model of function learning, the Extrapolation Associative Learning Model (EXAM). Experiments were carried out demonstrating that a model that assumes only linear extrapolation, such as EXAM, is inadequate as a generic model of function learning. An alternative model to EXAM is presented which is constructed of several components, each module applying different quantities of prior knowledge to the task. Chapter 5 presents experiments investigating the extent to which participants abstract and apply functions in transfer-tasks. The results demonstrate that models of function learning must be able to restrict their range of allowable solutions in psychologically plausible ways.