Slide Slam O16
Neural mechanisms of non-canonical sentence comprehension: A study of effective connectivity in healthy adults
Brianne Chiappetta1, Elena Barbieri1, Christine Kim1, Cynthia K. Thompson1; 1Northwestern University
Introduction. Neurocognitive models of sentence comprehension1-4 involve similar left hemisphere (LH) regions but differ in the role assigned to such regions in processing syntactically complex (non-canonical, e.g., Theme-Verb-Agent) sentences, and in the way regions are inter-connected. During sentence comprehension, healthy adults employ an Agent-first strategy5 that ensures successful comprehension of canonical (Agent-Verb-Theme) sentences, but fails for non-canonical sentences, triggering reanalysis/revision processes. The present study evaluated four neurocognitive models of sentence comprehension1-4 with respect to their ability to predict neural connectivity within the LH language network during comprehension of non-canonical sentences, using Dynamic Causal Modeling (DCM), a hypothesis-driven approach that estimates task-induced neuronal interactions. Methods. Twenty-one right-handed healthy adults (aged 24-61 years) performed a sentence comprehension fMRI task where they indicated whether spoken sentences matched/mismatched concurrently presented pictures of semantically reversible actions. Task conditions included: (1) canonical sentences; (2) non-canonical sentences; and (3) a control condition (reversed speech and scrambled images). DCM models included six LH regions of interest (ROIs) common to all four neurocognitive models: Inferior Frontal Gyrus pars triangularis and pars opercularis (IFGtri and IFGoper), posterior Superior Temporal Gyrus (pSTG), posterior Middle Temporal Gyrus (pMTG), Angular Gyrus (AG), and Anterior Temporal Lobe (ATL). Within each ROI, significantly active clusters (p<.05, uncorrected) for the sentences > control contrast were identified, and participants’ timeseries were extracted from 4mm-radius spheres centered on the peak coordinates of the cluster (k>=10) with the highest t-value, adjusted for effects-of-interest. Six participants were excluded because at least one ROI had suprathreshold voxel-wise activation <10 voxels. All DCM models assumed intrinsic connectivity between all regions. Driving inputs reflected sentence comprehension (canonical + non-canonical), and modulatory effects on between-region connectivity reflected syntactic complexity (non-canonical > (canonical + control)). Models were specified as bilinear and deterministic, with two-state neuronal equations employed to model excitatory and inhibitory effects within regions. Fifteen DCM models, reflecting the 4 main neurocognitive models of sentence comprehension and their variations, were compared using random-effect Bayesian Model Selection. Results. The group data was best fit by Bornkessel-Schlesewsky & Schlesewsky’s model4, which assumes driving inputs to pSTG and ATL, i.e., the starting point of the dorsal and ventral streams, respectively. In the dorsal stream, following segmentation of the auditory input (pSTG), the sentence syntactic structure is computed (pMTG) and thematic roles are mapped (AG). In the ventral stream, semantic representations for each word are activated and concatenated in the ATL. Both streams converge on the IFG for integration and control processes. Results showed that syntactic complexity modulated the inhibitory connection from the pMTG to AG, suggesting that for non-canonical sentences, sentence structure building (pMTG) results in inhibition of canonical thematic mapping (AG) and subsequent release of the inhibitory constraint from AG to IFGoper, ultimately increasing cognitive control and initiating revision processes. References. 1. Friederici, 2012, Trends in Cognitive Sciences, 16(5), 262-268. 2. Matchin & Hickok, 2020, Cerebral Cortex, 30(3), 1481-1498. 3. Thompson & Meltzer-Asscher, 2014, Structuring the Argument, 141-168. 4. Bornkessel-Schlesewsky & Schlesewsky, 2013, Brain & Language, 125(1), 60-76. 5. Mack & Thompson, 2017, JSLHR, 60(5), 1299-1315.