Modelling developmental disorders

Introduction: Why build models of developmental disorders? A six-year-old child is given a vocabulary test where he has to name pictures of objects. He scores poorly, and makes mistakes such as naming a picture of a guitar as a 'piano'. One explanation of the child's difficulty is that he has impoverished semantic (meaning) representations. Another child has problems reading words out loud, especially new words that she has not seen before. For example, she reads the novel word 'slear' as 'sear'. One explanation of her difficulty is that she has poorly specified phonological (speech sound) representations. What do these terms 'impoverished' and 'poorly specified' mean? How did the mental representations of meaning and speech sounds get this way? If a speech and language therapist or a specialist teacher wanted to intervene to remediate these problems, what intervention would be appropriate, and at what age? Should it be a different intervention if representations are impoverished versus poorly specified? Computational models of developmental disorders represent one technique to improve our understanding of the nature of deficits, their origin, long-term outcomes and possible pathways for remediation. The majority of the models we consider in this chapter are developmental, in that they learn abilities by exposing a developmental system to a structured learning environment. The act of building a model-specifying the nature of the developmental system, its input and output systems, and the information present in the learning environment-forces a theory to be specified in 3 much greater detail than would normally be the case in a verbal formulation. In the context of disorders, models allow detailed consideration of what could be affecting development-what is it that is different about the developmental system that is preventing strong learning? Is the information different in the learning environment? What exactly is different about meaning representations that are 'impoverished' or phonological representations that are 'poorly specified' and how does this lead to the kinds of error one sees in children with developmental impairments in productive vocabulary and reading, respectively? The process of constructing a model by its nature involves simplification. The aim is to build a working system that embodies key constraints of the phenomenon under consideration, and simplifies aspects that are taken to be unimportant. For the study of behaviour in children and adults with developmental disorders, empirical data from psychology form the most frequent constraints. For example, we might find that a child with a …