The conceptual structure of cabbages and things

Abstract

review of semantic memory impairments and their implications for the representational format and organisation of conceptual knowledge. In that paper, they argued that conceptual knowledge is distributed across a network of different sub-systems or ‘‘attribute domains,’’ including at least a perceptual and functional or propositional system. Saffran’s theoretical and empirical development of this approach has contributed to its becoming an influential framework for the investigation of semantic memory, throughout both the neuropsychological and neuroimaging literature. The work that we will discuss addresses the question of how conceptual knowledge is organised and represented in the brain. It asks whether domains (such as living things or man-made artefacts) and categories (such as tools or fruit) are represented explicitly or whether domain and category structure emerges out of a unitary distributed system. Evidence from patients with ‘category-specific deficits’ and neuroimaging studies seem to suggest that conceptual knowledge is explicitly structured in independent content-based stores. Many studies point to these as being organised by property type into the distinct ‘‘attribute domains’’ posited by Saffran and many others, rather than by category/domain of concept per se (e.g., living vs. non-living things). However, we have recently developed a different theoretical account, based on analyses of the fine-grained details of semantic impairments, connectionist modelling and neuroimaging studies, in which concepts are represented as patterns of activation over multiple semantic properties within a unitary distributed system, which is not necessarily differentiated, functionally or neurally, into separate sub-stores for different kinds of semantic property. Within this context, categoryspecific deficits emerge as a result of differences in the structure and content of concepts rather than from damage to one or more propertyspecific sub-systems. The critical variables in explaining patterns of deficit within this system are correlation and distinctiveness rather than property type. We claim that living things have many shared, correlated properties with only weakly correlated distinctive properties, while artefacts have fewer, more weakly correlated properties which tend to be more distinctive. Since correlated properties are more robust in the face of damage, this predicts that the shared properties of living things will tend to be preserved while distinctive properties are lost. For artefacts the shared-distinctive dissociation should be less marked, since distinctive properties are protected by form–function correlations and shared properties are fewer and less inter-correlated.