Semantic feature dissociation: A new hypothesis concerning autism

Abstract

This thesis introduces and defends a new hypothesis concerning autism:Semantic Feature Dissociation (SFD). The claim is that some autistic people only store information about strong correlations in semantic memory. I begin by arguing the most promising theories of autism currently on offer are Bayesian theories. However, these omit important details, especially about the underlying format of world knowledge, and its role in social cognition. The SFD hypothesis bridges this gap, linking autism traits explicitly to research on concept structure.After critically reviewing key literature, I defend the hypothesis in two ways.First, I report a methodologically novel qualitative study of autism autobiographies, from which the hypothesis was abducted. This reveals that it can potentially account for many real-world autism traits. Crucially, most social and language differences can be attributed to general changes in the structure of world knowledge, without implicating a specialised mechanism for identifying mental states. Second, I show SFD is better than other accounts at predicting important lines of experimental evidence concerning social cognition, language and perception in autism. I conclude by tentatively suggesting SFD might reconcile the two leading Bayesian accounts of autism: HIPPEA and weak priors.3