Incoherent speech is a hallmark of formal thought disorder in schizophrenia, reflecting impaired semantic integration. However, objective measurement remains challenging. Production tasks confound semantics with motor/executive demands, whereas passive comprehension of manipulated narratives isolates integration processes. Using large language models, we generated graded incoherence and hypothesized dose-dependent reductions in language network integration. In a single-participant 7T fMRI study, GPT-4 generated parametric audio narratives (temperature T=0 coherent control, T=1.00–1.15 increasingly incoherent) from matched literary genres. Stimuli were presented passively across four 15-min runs. Nine participant-specific language-selective clusters were identified via functional localizer ( p FWE <10⁻⁴). Functional connectivity was assessed using pairwise Pearson correlations, Mantel t -tests compared dissimilarity between conditions. Graph metrics (global efficiency/modularity) quantified integration/segregation. Clusters were left-lateralized, spanning IFG, pMTG, and MFG. Coherent narratives elicited strong frontotemporal coupling ( r ∼0.7–0.8). Incoherence produced dose-dependent attenuation, with significant Mantel dissimilarity at T=1.00 ( p perm =.000) and T=1.10 ( p perm =.008), approaching significance at T=1.15 ( p perm =.060). Global efficiency decreased (trend: t (2)=-3.22, p =.084) and modularity increased ( t (2)=4.24, p =.050). Graded semantic incoherence alone induced progressive weakening of frontotemporal integration and increased modular segregation in the language network, suggesting a mechanistic model for patterns observed in thought disorder. High-resolution 7T imaging detected these subtle, predictable reorganizations within a single participant.
Tahedl et al. (Sun,) studied this question.
Synapse has enriched 5 closely related papers on similar clinical questions. Consider them for comparative context: