Slide Slam M9
Overlap in neural processes in the comprehension and production of discourse.
Matias Morales1, Tanvi Patel1, Andres Tamm1, Martin Pickering1, Paul Hoffman1; 1University of Edinburgh
When comprehending discourse, people construct a situation model of its content (Kintsch & van Dijk, 1978). In the brain, this process engages the default mode network (DMN), which shows greater activation during the comprehension of coherent relative to incoherent text (Ferstl & von Cramon, 2002), and maximal activity when it is difficult, but not impossible, to construct a situation model from discourse (Kuperberg et al., 2006). Importantly, situation model processes are thought to guide production as well as comprehension of language (Garrod & Pickering, 2004). However, very little is known about how these processes overlap in the brain during the comprehension and production of speech within participants. To address this issue, we compared the neural correlates of coherence during the production and comprehension of discourse. In a fMRI study, 25 participants produced and listened to speech about a range of different topics. We computed the coherence of this discourse as a time-varying measure that indexed the natural fluctuations of coherence that occurred in the speech. As discourse processing relies on distributed networks, we took a network-level approach to analyse our data. At the whole brain level, we investigated the effects of coherence along the principal connectivity gradient, which describes the functional organisation of cognitive processes in the brain (Margulies et al., 2016). This gradient situates the DMN at one extreme of this spectrum, associated with multimodal high-level abstract thought, with sensorimotor networks engaged in more basic processes at the other end. To confirm and extend the results from this analysis, we also investigated the effects of coherence on three specific networks: the DMN, the semantic control network (SCN), and the multiple demand network (MDN). Results showed that overall, production and comprehension of discourse activated the same left-lateralised neural networks. In addition, people activated the DMN more when they listened to less coherent speech, suggesting they frequently updated and reconfigured their situation models during comprehension of low-coherence speech. Critically, our study reveals that these effects were also present when people produced their own speech, suggesting a role for situation models in planning and generating speech content just as it occurs in comprehension. In contrast, coherence effects diverged between production and comprehension processes in the SCN, with greater activation when participants listened to less coherent discourse, but not when participants produced less coherent speech. In addition, the sensorimotor part of the gradient showed greater activation when discourse was less coherent during speech comprehension, which was not present during speech production, thus indicating that low-level sensorimotor networks were more engaged when participants listened to speech that deviated from the expected topic. Finally, results showed no coherence effects in the MDN for either task, suggesting that higher levels of language processing draws on specific neural resources distinct from those engaged in other cognitive domains. Overall, our results show that neural mechanisms implicated in the construction of situation models are largely shared between the production and comprehension of discourse. However, they also indicate that the SCN plays a different role in comprehension and production.