Search Abstracts | Symposia | Slide Sessions | Poster Sessions
Neural correlates of spontaneous speech production: insights from whole-brain fMRI
Poster Session D, Thursday, October 1, 4:30 - 6:30 pm, Wangari Maathai
Giulia Li Calzi1, Narly Golestani2,3,4, Remo Nitschke1, Agata Marcante1, Balthasar Bickel1, Martin Meyer1, Laura Giglio1; 1Institute for the Interdisciplinary Study of Language Evolution, University of Zurich, 2Department of Behavioural and Cognitive Biology, Faculty of Life Sciences, University of Vienna, 3Department of Developmental and Educational Psychology, Faculty of Psychology, University of Vienna, 4Department of Psychology, Faculty of Psychology and Educational Sciences, University of Geneva
To communicate with others, we must convert our ideas into linear sequences that can be externalised through speech or sign. The vast majority of studies investigating the neural correlates of this process have relied on highly controlled paradigms, which differ considerably from everyday, naturalistic language production. Here, we analysed an existing fMRI dataset in which participants (n=19) spontaneously recalled the contents of short movies (Lee et al., 2023) to investigate which brain regions support different aspects of speech production planning. Using the Stanza constituency parser and the spaCy dependency parser, we derived four word-by-word predictors capturing a) the number of verbs in the sentence (propositional complexity), b) the number of phrasal nodes being opened at each word (syntactic structure building), c) the number of verbal dependencies being maintained simultaneously (linearisation), and d) the number of syllables within lexical noun phrases anchored to the phrase onset (phonological encoding). Each word-by-word predictor was convolved with the haemodynamic response function and entered as a parametric modulator of the response to word onset in a whole-brain model, alongside word frequency and lexical surprisal as control regressors. Group-level inference was performed using one-sample t-tests with cluster-level correction for multiple comparisons (p < 0.05 FWE-corrected, threshold p < 0.001 uncorrected). Results revealed that propositional complexity modulated brain responses in the bilateral precunei and temporo-parietal junctions. Because these regions have been associated with the formation of conceptual representations (Menenti et al., 2012), our findings suggest these regions support the segmentation of continuous experiences into the discrete propositional units, which are then mapped onto a syntactic structure. Syntactic structure building was associated with higher responses in the bilateral superior temporal sulci and inferior parietal gyri. The effects in the temporal lobe converge with prior findings from controlled production paradigms, corroborating the role of this region in syntactic structure building across speech contexts (Yeaton, 2025). The linearisation of verbal dependents was associated with increased responses in the posterior left middle temporal gyrus and left supramarginal gyrus. The former region has previously been associated with lexical retrieval, and its engagement may reflect the fact that the maintenance of verbal dependencies is always highest at the verb itself. The left supramarginal gyrus is known to underlie domain-general serial order retention. This result could therefore suggest that linearisation in a language with limited morphological demands, such as English, relies on domain-general resources for the ordering of sequential information (cf. Matchin & Hickok, 2020). Finally, phonological encoding was associated with increased responses in the left precentral gyrus extending into the pars opercularis of the left inferior frontal gyrus. This finding is consistent with prior studies of syllabification during single-word production. Taken together, these results demonstrate that model-based approaches can track multiple distinct aspects of spontaneous speech production. They also provide the first comprehensive characterisation of the neural infrastructure that supports this process, thus laying the groundwork for future work on the temporal dynamics of these processes using neuroimaging methods with a higher temporal resolution.
Topic Areas: Language Production, Computational Approaches