14.1 Introduction

This paper deals with a general issue in the study of animal behaviour that we call path dependence. The expression refers to the fact that different histories of experiences (paths) may at first seem to produce the same behavioural effects yet reveal important differences when further examined. For instance, two training procedures may establish the same discrimination between two stimuli yet produce different responding to other stimuli, because the two paths have produced different internal states within the animal. There are several reasons why path dependence is an important issue. First, it comprises many phenomena that can provide stringent tests for theories of behaviour. Second, path dependence is at the root of several controversies, for instance whether animals encode absolute or relative characteristics of stimuli (Spence, 1936; Helson, 1964; Thomas, 1993) or whether learning phenomena such as backward blocking and un-overshadowing imply, in addition to basic associative learning, stimulus–stimulus associations or changes in stimulus associability (Wasserman & Berglan, 1998; Le Pelley & McLaren, 2003; Ghirlanda, 2005).

In this paper we use a simple neural network model of basic associative learning (Blough, 1975; Enquist & Ghirlanda, 2005) to show how path dependence can arise from fundamental properties of associative memory. The model has two core components: (1) distributed representations of stimuli based on knowledge of sensory processes and (2) a simple learning mechanism that can associate stimulus representations with responses. We consider examples of path dependence in experiments on generalisation (or ‘stimulus control’).