The neural basis of semantic cognition: converging evidence from neuropsychology, neuroimaging and TMS.

Recent studies suggest that a complex, distributed neural network underpins semantic cognition. This article reviews our contribution to this emerging picture and traces the putative roles of each region within this network. Neuropsychological studies indicate that semantic cognition draws on at least two interacting components: semantic representations [degraded in semantic dementia (SD)] and control processes [deficient in patients with multimodal semantic impairment following stroke aphasia (SA)]. To explore the first component, we employed distortion-corrected functional magnetic resonance imaging (fMRI) and transcranial magnetic stimulation (TMS) in healthy volunteers: these studies convergently indicated that the anterior temporal lobes (ATLs; atrophied in SD) combine information from different modalities within an amodal semantic "hub". Regions of cortex that code specific semantic features ("spokes") also make a critical contribution to knowledge within particular categories. This network of brain regions interacts with semantic control processes reliant on left inferior frontal gyrus (LIFG), posterior middle temporal gyrus (pMTG) and inferior parietal cortices. SA patients with damage to these regions have difficulty focussing on aspects of knowledge that are relevant to the current goal or context, in both verbal and non-verbal tasks. SA patients with LIFG and temporoparietal lesions show similar deficits of semantic control, suggesting that a large-scale distributed cortical network underpins semantic control. Convergent evidence is again provided by fMRI and TMS. We separately manipulated the representational and control demands of a semantic task in fMRI, and found a dissociation within the temporal lobe: ATL was sensitive to the number of meanings retrieved, while pMTG and LIFG showed effects of semantic selection. Moreover, TMS to LIFG and pMTG produced equal disruption of tasks tapping semantic control. The next challenges are to delineate the specific roles of each region within the semantic control network and to specify the way in which control processes interact with semantic representations to focus processing on relevant features of concepts.