My Account

Poster A24, Tuesday, August 20, 2019, 10:15 am – 12:00 pm, Restaurant Hall

Domain-general neural mechanisms responsible for integrating incoming semantic content against multiple contextual aspects: evidence from semantic and nonsemantic tasks

Francesca Martina Branzi1, Matthew Lambon Ralph1;1MRC Cognition and Brain Sciences Unit, University of Cambridge

The integration of spoken or written narratives into a meaningful structure has been linked to a neural network that includes perisylvian temporal and left inferior frontal regions, but also parietal cortex and cerebellum. However, the question of whether activity in these regions reflects mechanisms for integrating information into a relevant context not uniquely implicated in language processing has received much less attention. In this functional magnetic resonance imaging (fMRI) study we addressed this question by testing healthy participants in semantic and nonsemantic tasks (between-subjects design) reflecting two important aspects of semantic integration during naturalistic language processing. The first refers to its time-extended dimension. Hence, integration processes were measured across two paragraphs of text (narrative reading task) and two paragraphs of number sequences (number pattern task). The second refers to the dynamic nature of integration, requiring sometimes to update information when context or situation changes. Therefore, we compared brain regions involved in integration when the second paragraph (target) was preceded by a low−congruent (LC) context paragraph (disruption of semantic or numerical coherence) versus when the same second target paragraph could be integrated in a highly congruent context paragraph (HC). To reveal brain regions involved in time-extended integration processes in the two tasks, we compared the above-mentioned two conditions against a no context control condition (NC) (i.e., when the same target paragraph could not be integrated with a contextual support). In both tasks, the resultant fMRI data were analysed only for the identical target paragraph, ensuring that any observed difference must reflect the influence of the preceding contexts. Key brain areas and networks involved in domain-general integration processes were established by using both univariate and multivariate (independent component analysis-ICA) analyses. In accord with previous evidence (Duncan, 2010), we hypothesised that domain-general mechanisms for semantic integration (LC&HC>NC) would have been supported by a fronto-parietal network, maximally engaged during update of contextual information (LC>HC). Furthermore, according to evidence on prediction-error (Moberget et al., 2016), we also hypothesised that the cerebellum would have been sensitive to integration, and particularly to shifts of context (LC condition). Univariate analysis’ results revealed that activity in inferior frontal gyrus-IFG, left dorsal angular gyrus-dAG and cerebellum was similarly modulated by integration (LC&HC>NC) in semantic and nonsemantic tasks. Accordingly, ICA’s results revealed a fronto-parietal network similarly engaged in the two tasks. Furthermore, we found that IFG, left precentral gyrus, right insula, and superior parietal lobe were sensitive to shifts of context (LC>HC) similarly in both tasks. Finally, despite some regions were similarly modulated by shifts of context in the two tasks (e.g. IFG), they were recruited by different networks in semantic and nonsemantic tasks (ICA). The present study established the core neurocomputations supporting integration processes not limited to language processing. These findings are relevant not only for basic research, but also for understanding co-occurring deficits in neurological patients.

Themes: Meaning: Combinatorial Semantics, Meaning: Lexical Semantics
Method: Functional Imaging

Back