Summary: Two distinct frontotemporal networks in the brain activate and cooperate to integrate individual word meanings, enabling the construction of higher-order sentence meaning during reading.
Source: UT Houston
When someone reads a sentence, two separate but interacting brain networks—one anchored in the frontal lobe and the other in the temporal lobe—become active. Together, they integrate the meanings of individual words to build richer, more complex sentence-level meaning, according to research from UTHealth Houston.
The study was led by Oscar Woolnough, PhD, a postdoctoral research fellow in the Vivian L. Smith Department of Neurosurgery at McGovern Medical School, and by Nitin Tandon, MD, professor and interim chair of that department. Their findings were published in The Proceedings of the National Academy of Sciences (PNAS).
“This work clarifies how distributed hubs within the brain’s language network interact to support comprehension of complex sentences,” said Woolnough, first author on the paper and a member of the Texas Institute for Restorative Neurotechnologies (TIRN) at UTHealth Houston. “Language understanding depends on rapid, precisely timed processes that take place across multiple sites in the brain.”
To identify how specific brain regions contribute to reading, the research team recorded neural activity directly from patients who had intracranial electrodes implanted as part of clinical monitoring for epilepsy. These direct recordings allowed the team to track rapid neural dynamics that are difficult to observe with noninvasive methods.
Participants read three types of stimuli while neural signals were collected: normal, meaningful sentences; “Jabberwocky” sentences that preserve grammar and syntax but replace content words with nonsense terms; and unconnected lists of real words or pseudowords. Comparing responses across these conditions enabled the researchers to separate processes related to sentence-level meaning from those tied to individual word processing.
From these recordings, the investigators isolated two distinct frontotemporal networks with different timings, directions of information flow, and functional roles. The first network links the inferior frontal gyrus (IFG) in the frontal lobe with the middle temporal gyrus (MTG) in the temporal lobe. In this network, frontal activity precedes and drives temporal responses, and neural activity steadily increases as meaningful sentences unfold—consistent with a role in integrating words into sentence-level meaning. This pattern was reduced or absent when participants read Jabberwocky or lists of words, indicating sensitivity to semantic composition across a sentence.
The second network involves a different region of the temporal lobe, the superior temporal gyrus (STG), and connections back to frontal cortex including portions of the IFG. In this pathway, temporal responses lead frontal responses, and activity was stronger for individual words presented in lists than for the same words embedded in sentences. That pattern suggests that sentential context streamlines lexical and phonological processing of words, making individual word processing more efficient when context is available.
“Direct intracranial recordings give us unparalleled access to the fast, transient interactions that underlie reading,” said Tandon, senior author, who holds the Nancy, Clive and Pierce Runnels Distinguished Chair in Neuroscience at the Vivian L. Smith Center for Neurologic Research and is a BCMS Distinguished Professor in Neurological Disorders and Neurosurgery. “Our results show that comprehension and language generation arise from brief, coordinated states across multiple brain regions rather than from single isolated areas.”
Understanding these rapid frontotemporal dynamics offers a clearer picture of how normal reading functions and may inform research into reading difficulties such as dyslexia. The authors note that improved knowledge of these mechanisms could eventually contribute to better diagnostic tools and targeted interventions for people with reading disorders.
Funding: The research received support from a five-year, $4.4 million grant from the National Institutes of Health BRAIN Initiative, which funds work to develop and apply new technologies that reveal how the human brain works dynamically.
Co-authors from McGovern Medical School’s neurosurgery department included Cristian Donos, PhD; Elliot Murphy, PhD; Patrick Rollo; and Zachary Roccaforte. Murphy, Rollo, Roccaforte, and Tandon are members of TIRN. Additional contribution came from Stanislas Dehaene, PhD, of Université Paris-Saclay and the Collège de France.
About this neuroscience research news
Author: Caitie Barkley
Source: UT Houston
Contact: Caitie Barkley – UT Houston
Image: The image is in the public domain
Original Research: Closed access. “Spatiotemporally Distributed Frontotemporal Networks for Sentence Reading” by Oscar Woolnough et al., PNAS
Abstract
Spatiotemporally Distributed Frontotemporal Networks for Sentence Reading
Reading requires integrating the meanings of individual words to derive more complex, higher-order sentence meaning. Prior research has implicated the inferior frontal gyrus (IFG) and middle temporal gyrus (MTG) in language processing, but the distinct contributions and timing of these regions during sentence reading have remained unclear.
Using direct intracranial recordings, the study measured neural activity while participants read meaningful sentences, meaning-deficient Jabberwocky sentences, and lists of words or pseudowords. Analysis revealed two functionally and spatiotemporally distinct frontotemporal networks that are sensitive to different aspects of word and sentence composition.
The first network, engaging IFG and MTG with IFG activity preceding MTG, showed progressive increases in activity over the course of a sentence and diminished responses for Jabberwocky and word lists—supporting a role in deriving sentence-level meaning. The second network, involving the superior temporal gyrus and IFG with temporal activity leading frontal responses, exhibited stronger responses to isolated words in lists than to those same words in sentences, suggesting that sentential context enhances efficiency in lexical or phonological processing.
These adjacent but temporally dissociable mechanisms illuminate the layered semantic computations that enable fluent reading and point to distributed, dynamic processing across the frontotemporal language network rather than a strict frontal-versus-temporal division of labor.