The study investigates the spatiotemporal dynamics of semantic integration in the human brain. A key question in the neurobiology of sentence processing is whether multiple cortical regions are necessary for evaluating sentence meaning or if a single region with overlapping components can encode distinct aspects of meaning. Word retrieval from memory to refer to objects is crucial for language's contribution to cognitive abilities. Language's generative nature enables us to derive concepts from novel cues, which depends on rapid integration of multiple lexical objects. While broadly, language-related semantic processes engage the posterior temporal lobe, prefrontal and parietal cortices, there's no consensus on the specific regions and their interactions. Previous research using picture naming or scalp event-related potentials and fMRI lacked the spatiotemporal resolution to comprehensively map cortical responses. The study aims to isolate sites involved in distinct semantic processes by using an orthographic sentence comprehension and linguistic reference paradigm with a large cohort of subjects undergoing intracranial electrocorticography (iEEG).

Existing literature shows that language-related semantic processes engage the posterior temporal lobe and prefrontal and parietal cortices but lacks a precise understanding of the specific regions and their interactions. Much prior work focused on picture naming (overt reference), neglecting inferential meaning from definitions. The rapid and distributed nature of these processes made it difficult to disentangle retrieval from related computations like inferring semantic coherence. Studies using scalp event-related potentials or fMRI lacked the necessary spatiotemporal resolution. While some analyses suggest overlapping substrates for language and semantic processing, others indicate dissociability. The functional specialization of subregions within inferior frontal, inferior parietal, and posterior temporal cortices for diverse semantic processes remains unclear.

The study used a large cohort (n=58) of epilepsy patients undergoing intracranial electrocorticography (iEEG) with a total of 11,328 electrodes. Broadband high gamma activity (BGA; 70–150 Hz), strongly correlated with fMRI BOLD signal, indexed local cortical processing. A sentence comprehension paradigm with referential and non-referential sentences to common objects was used. Participants articulated common object names in response to written descriptions presented via rapid serial visual presentation (500 ms per word). The final word determined the referential nature of the description. Non-referential trials manipulated semantic coherence (coherent vs. incoherent). A norming study (n=80) quantified the point of semantic ‘narrowing’ for referential sentences, analyzing the timing and extent of lexical search effort. Four experimental conditions focused on semantic reference and coherence: Non-Referential (coherent vs. incoherent); Referential (strong narrowing vs. limited narrowing). Surface-based mixed-effects multilevel analysis (SB-MEMA) mapped cortical activity, examining BGA responses across the language-dominant cortex. Low-frequency dynamics (hippocampal theta) were also analyzed.

Behavioral performance showed significantly faster reaction times for referential trials. Spatiotemporal mapping revealed serially increasing activation across sentence duration in a distributed network. Three regions—ventral temporal cortex, inferior lateral temporo-occipital cortex, and IFS—were active throughout. Referential trials showed greater BGA in MFG, middle IFS, MPC, parahippocampal cortex, vmPFC, and OFC compared to non-referential trials. Non-referential trials showed increased BGA in posterior superior temporal cortex and alFG. For non-referential sentences, incoherent sentences showed increased medial frontal and superior medial parietal cortex activity, while coherent sentences showed greater BGA in IFS, alFG, angular gyrus, PMTG, and OFC. Referential trials with limited semantic narrowing showed greater BGA in pSTS, MPC, IFS, ATL, and OFC than strong narrowing trials. The IFS showed greater high gamma activity for all semantic contrasts, indicating a mosaic-like patchwork of activity across subregions. The pSTS and pMTG contributed to all semantic processes, showing a temporal progression of sensitivity, with early effects of reference followed by coherence effects. The MPC, hippocampus, PHG, OFC, intraparietal sulcus, and vmPFC were engaged during sentence processing. The study also found that coherent non-referential sentences resulted in greater BGA in various brain regions compared to incoherent sentences, potentially indexing the entry of the final word into the workspace and successful wrap-up for semantically legal structures.

The study identified distinct semantic roles for closely adjacent loci in the lateral inferior frontal and posterior temporal cortex. The IFS exhibited greater high gamma activity for all semantic contrasts, revealing a complex mosaic-like pattern of activity across its subregions. Anterior and posterior IFG showed separable semantic effects, but with some overlap. The pSTS and pMTG jointly contributed to all semantic processes, with a clear temporal progression of sensitivity. The study highlighted the joint role of pSTS-IFS in semantic composition demands. The findings are concordant with previous research implicating the MPC in context memory, situation model construction, and combining concepts from distinct categorical domains. The involvement of the medial temporal lobe, particularly the hippocampus and PHG, suggests a role in relating incoming words to sentence contexts and contributing memory-related representations for lexical search. The results support the involvement of a broader network, including medial cortical structures, in elementary components of sentence comprehension and lexical access.

This research provides extensive whole-brain intracranial mapping of semantically coherent orthographic sentence representations, revealing complementary cortical mosaics for semantic integration in posterior temporal and inferior frontal cortex, recruited for distinct semantic demands. The IFS and OFC uniquely responded to all semantic processes, potentially acting as higher-order lexico-semantic hubs or sites of semantic saliency computation. The findings challenge models assuming clear frontal, temporal, and parietal separation for semantic integration processes. Future research should explore the mosaic-like cortical organization for syntactic/grammatical processing and its implications for language disorders.

The study's findings are based on a patient population undergoing treatment for epilepsy, which might affect the generalizability of the results to the neurotypical population. The electrode placement in these participants was for clinical need, not specifically designed for the experimental purposes. Furthermore, the analysis focused primarily on the language-dominant left hemisphere; the conclusions regarding right hemisphere involvement are tentative due to limited electrode coverage in this region. The analysis of semantic narrowing did not strictly isolate lexical access.