Re-evaluating Computational Models of Semantic Cognition

Several computational models of the semantic cognitive system have been developed. This thesis considers four such models: the hub-and-spoke, the conceptual structure, the modality-specific, and the conceptual topography model. These models account for both generalised and category-specific semantic impairments. The models encapsulate different, and partially mutually exclusive, theoretical positions, but still account for similar, if not overlapping, semantic impairments. However, no single theory explains the full spectrum of both healthy and impaired semantic cognition. In order to better understand the space of theories and their inter-relations, this thesis reports results from (re)implementing the four theories and attempting to simulate both types of semantic impairment within each implementation. These four implementations shed light on the various computational and modelling assumptions and implications of each theoretical position. Compatibilities (and incompatibilities) between each theory (and model) are also discussed. It is additionally argued that some assumptions within each account, even though superficially different, are shared, and, conversely, some seemingly minor or background assumptions are centrally important. Examples of the latter include: limiting attention within any model to just two input modalities; evaluating model and patient behaviour with non-naturalistic semantic tasks; ignoring input from executive and affective systems; or (within the hub-and-spoke model) appealing to emergent properties of the implementation.