Feature inference in perceptual categorization

The ability to make inferences about the properties of a category instance based on knowledge of its category membership is a crucial cognitive ability. The purpose of this thesis was to evaluate feature inference learning of categories and feature inference decision-making for category instances in an attempt to clarify the nature of the category representation underlying feature inference. Specifically, three experiments evaluated feature inference learning of the classic Shepard, Hovland and Jenkins (1961) category structures compared to standard classification learning. The results supported a label bias hypothesis in terms of an advantage in feature inference learning tasks that allowed using label-based unidimensional rules over classification learning tasks that did not. However, both classification and feature inference learning resulted in predominantly rule representation, consistent with participants’ self-reported learning strategies. A further five experiments evaluated feature inference decision-making in terms of analogues for standard categorical induction effects--notably premise typicality--in the perceptual categorization paradigm. However, there was no evidence of a premise typicality effect when similarity was controlled for. Possible conceptual and methodological reasons for failing to find this effect are discussed. While results do not support prototype representation as the underlying basis for feature inference, methodological explanations and/or a lack of power to detect a potentially small effect cannot be ruled out as explanations for the absence of premise typicality. The results of these eight experiments tentatively support a bias for label-based rules as the representation underlying feature inference learning and decision-making, but future research will need to more definitively differentiate this from prototype and exemplar representation