Generalisation

Background

It is frequently observed that individuals with autism show poor transfer of newly acquired skills or knowledge from a training context to a novel environment. That is, although new skills can be learned, and there can be dramatic improvements in the context in which they are taught, there are often severe limitations in how far these new skills are generalised to other contexts. Parents of children with autism often comment that the behaviour of their children in structured educational environments can be quite unlike their behaviour at home. Given the promising results in the ability of children with autism to learn new skills, for example in social interaction, a key focus of training studies now is to identify the reasons for the poor generalisation to new settings.

Enhanced discrimination

We have been conducting experimental research examining possible mechanisms for poor generalisation in autism. We have focussed on how far individuals with autism can generalise from one stimulus set to another, rather than between contexts. Generalising from one stimulus to another depends on being able to identify which aspects of the two stimuli are the same, or held in common with one another. We hypothesised that this ability to process common aspects or features between two stimuli may be compromised in autism. This makes the prediction that individuals with autism should show enhanced discrimination between similar stimuli. We tested this by giving adults with autism and typical adults very difficult discrimination task, in which the stimuli were highly similar to one another. Since these stimuli had many features in common, we predicted that individuals with autism would be better at learning the discriminations compared to the typical group because they would process the common features poorly. For stimuli that were entirely novel, we found enhanced discrimination in the group with autism compared to the typical group.

Reduced categorisation

Grouping different stimuli and objects into categories is a fundamental skill. It allows for the recognition of particular stimuli which have not been seen before as being like other stimuli which we have previously learnt about. It therefore allows us to make sense rapidly of stimuli in the environment. The ability to categorise is based on processing the similarity between stimuli -- the more similar they are, the more they will be treated as members of the same category. However, a deficit in processing common features between stimuli would result in poor categorisation. We have tested this in adults with autism. They showed worse categorisation during a training phase in a prototype task, and a reduced prototype effect. We are now conducting more direct tests of generalisation in autism.

Implications

One clear implication of these results are that if individuals with autism are trained to learn a skill with one set of stimuli, then they will find it difficult to transfer that skill if the stimuli in the new environment have fewer features in common from those in the training environment. Another is that individuals with autism are like to be able to perform well on any task which involves discriminating one stimulus from others. An example is visual search.

Related reading

Plaisted, K.C. (2000). Aspects of autism that theory of mind cannot easily explain. In Baron-Cohen S, Tager-Flusberg H & Cohen DJ (Eds). Understanding Other Minds: Perspectives from Autism and Cognitive Neuroscience (2nd Edition). Oxford University Press.

Plaisted, K.C. (2001). Reduced generalization: An alternative to weak central coherence. In: Burack, J.A., Charman, A., Yirmiya, N. & Zelazo, P.R. (Eds). Development and Autism: Perspectives from theory and research. Lawrence Erlbaum Associates, New Jersey.

Plaisted, K.C., O'Riordan, M., & Baron-Cohen, S. (1998). Enhanced discrimination of novel, highly similar stimuli by adults with autism during a perceptual learning task. Journal of Child Psychology and Psychiatry, 39, pp.765-775.

Plaisted, K.C., O'Riordan, M., Aitken, M., & Killcross, S. (submitted). Categorisation in autism: Evidence of a reduced prototype effect. Submitted to Quarterly Journal of Experimental Psychology.