Functional segregation of conversational production and comprehension when using word predictability

The extent to which the language production and comprehension systems overlap remains debated. We address this debate using a dataset where participants engaged in unscripted conversations, while scanned with fMRI. Word predictability was hypothesized to rely on different processes, depending on whether the word was uttered or heard. We employed the information-theoretic measure of surprisal (the negative log probability of a word occurring, given the preceding context) as a parametric modulator, controlling for the words overall frequency. The results for production surprisal revealed activation in the left superior and inferior frontal gyri and motor areas. A large bilateral cluster in the posterior part of the medial prefrontal cortex extended from the supplementary motor area to the anterior cingulate cortex. The results for comprehension surprisal replicated findings from non-conversational contexts, showing involvement of the bilateral superior temporal gyrus/sulcus, presumably supporting bottom-up processes for prediction error detection. Importantly, no overlap in the neural infrastructure of production and comprehension was observed, suggesting that word predictability processes in production and comprehension differ. We suggest that while the comprehension system handles prediction errors, the production system minimizes these errors through adaptation, all to achieve successful communication.