Modelling Brain Activity Associated with Metaphor Processing with Distributional Semantic Models

In this study we investigate how lexical-semantic relations associated with the literal meaning (and abstract meaning) are being accessed across the brain during familiar metaphor comprehension. We utilize a data-driven whole-brain searchlight similarity-decoding analysis. We contrast decoding metaphoric phrases ("she's grasping the idea") using distributional semantic models of the verb in the phrase (Verb model) versus that of the more abstract verb-sense (Paraphrase Verb model) obtained from literal paraphrases of the metaphoric phrases ("she's understanding the idea"). We showed successful decoding with the Verb model across frontal, temporal and parietal lobes mainly within areas of the language and default-mode networks. In contrast, decoding with the Paraphrase Verb model was restricted to frontal-temporal lobes within areas of the language-network which overlapped to some extent with significant decoding with the Verb model. Overall, the results suggest that lexical-semantic relations closely associated with the abstract meaning in metaphor processing are largely localized to language and amodal (multimodal) semantic memory systems of the brain, while those more associated with the literal meaning are processed across a more distributed semantic network including areas implicated in mental imagery and social-cognition.