DIFFERENCES IN THE NEURAL BASIS OF SEMANTIC ORGANIZATION FOR CONCRETE AND ABSTRACT CONCEPTS IN HUMANS

According to neurocognitive models of semantic processing, concrete as opposed to abstract concepts recruit modality-specific cortical regions in addition to shared contributions from fronto-parieto-temporal semantic hubs. It remains unclear whether beyond processing differences, abstract and concrete concepts also differ in how semantic similarity maps onto multivoxel activation patterns across the semantic network. We hypothesized that modality-specific regions represent fine-grained semantic similarity of concrete, but not abstract concepts, whereas shared semantic hubs represent the semantic similarity of both. During functional magnetic resonance imaging, twenty-two young adults (12 females, M_age = 21 y, SD = 1.69 y) made semantic decisions about baseline (e.g., 'police uniform'), semantically close ('commander uniform') and distant ('railway uniform') variations of abstract and concrete expressions, where similarity was quantified using a word2vec word-embedding. Expressions were median-split by concreteness based on ratings from an independent sample (n = 240, 193 females, M_age = 23 y, SD = 4.03 y). As expected, preliminary representational similarity analyses revealed significant correlations between multivoxel pattern similarity and word2vec semantic similarity within the modality-specific lateral occipital cortex for concrete concepts only. Surprisingly, in semantic hub areas – temporal pole, angular gyrus, middle and superior temporal and inferior frontal cortices – semantic similarity correlated significantly with the neural similarity for concrete – but not abstract – expressions. We propose that the shared perceptual basis of concrete concepts yields more consistent meaning representations that align with word2vec across the semantic network. Conversely, abstract concepts are more context-dependent and variable in meaning, which may produce greater neural variability even within semantic hubs.