Similarity learning and memory: A comparative analysis

All animals need to learn about relationships between stimuli that occur in their environment so that they can behave in an appropriate manner in the future. For example, they need to learn that ingesting some items is appropriate while ingesting others is not. Of course, the environments that most animals face are complex and there needs to be a basis upon which important relationships are learnt about whereas less important or spurious relationships are not. Across a number of academic disciplines (Artificial Intelligence, Neuroscience, Psychology) it has been popular to suggest that such relationships are learnt by animals forming mental or neural associations between stimuli. There is a great deal of laboratory research showing that associations become stronger as a consequence of repeated pairings of one stimulus with another, that associations form most readily between intense stimuli, and that the stimuli need to be presented close together in time. These three laws of association (frequency, intensity and contiguity) provide a clear basis upon which relationships that are likely to be important are most readily learnt. However, recent evidence has suggested that there is another constraint that influences associative learning. Thus, when the influences of frequency, intensity and contiguity are held constant, associations are more likely to form between stimuli that are similar than between those that are dissimilar. This is an intriguing finding because it is not anticipated by influential theories of learning and memory. This project will examine why animals (humans, rats and pigeons) are more likely to learn a relationship between two similar stimuli than between two dissimilar stimuli. The research will allow us to gain a better understand of the fundamental process of association formation in different species.

Animals learn about simple relationships between stimuli in their world by forming mental associations between the memories of those stimuli. Associative learning is held to be underpinned by three laws of association: the laws of frequency, intensity and temporal contiguity. Thus associations become stronger with repeated pairings and are most likely to form between stimuli that are intense and occur in close temporal proximity. These laws have been implemented in both behaviourally driven and neural inspired models of learning and memory; however, there is evidence that these laws are incomplete. Recent research in humans (Grand et al, 2006) and earlier research in pigeons (most notably by Rescorla & Gillan, 1980) indicate that, other things being equal, associations form more readily between similar stimuli than between dissimilar stimuli. However, our understanding of the influence of similarity on learning is impoverished and while it is possible to accommodate this influence within some models of learning (e.g., Wagner, 1981) the evidence that supports such an analysis is fragmentary and there remain other possible theoretical analyses. The proposed project will provide a comparative analysis of the influence of similarity on (associative) learning.