The aim of this chapter is to demonstrate that an elemental model, using a relatively simple error correcting learning algorithm, can prove remarkably resourceful when it comes to simulating human and infra-human learning and memory. The basic premise behind all elemental models of category learning is that the representation of any stimulus comprises multiple components which can individually enter into associations with designated category labels or responses. Used to its full potential, this approach captures the strengths of both prototype- and exemplar-based approaches to categorization. The full range of resources that elemental associative theories have to offer are rarely taken into account in comparisons with models that use other forms of representation, such as the configural theories offered by Pearce (1987, 1994) in the animal domain and Nosofsky (1991) in the human domain. We are by no means the only theorists to adopt this position, and the reader will find considerable overlap between our approach and that of several others (Brandon, Vogel, & Wagner, 2000; Harris, 2006; Wagner & Brandon, 2001).

We first set out the formal details of a model that implements elemental representation within an associative network employing a modified delta rule (following McClelland & Rumelhart, 1985). The modifications transform the delta rule into the basic real-time learning algorithm used by McLaren, Kaye, and Mackintosh (1989). For simplicity, some of the complexities of the latter model (e.g., weight decay and salience modulation) will not be considered here.