Neural populations in the language network differ in the size of their temporal receptive windows

Nature Human Behaviour, Journal Year: 2024, Volume and Issue: 8(10), P. 1924 - 1942

Published: Aug. 26, 2024

Language: Английский

---

Linguistic inputs must be syntactically parsable to fully engage the language network

bioRxiv (Cold Spring Harbor Laboratory), Journal Year: 2024, Volume and Issue: unknown

Published: June 22, 2024

Abstract Human language comprehension is remarkably robust to ill-formed inputs (e.g., word transpositions). This robustness has led some argue that syntactic parsing largely an illusion, and incremental more heuristic, shallow, semantics-based than often assumed. However, the available data are also consistent with possibility humans always perform rule-like symbolic simply deploy error correction mechanisms reconstruct when needed. We put these hypotheses a new stringent test by examining brain responses a) stimuli should pose challenge for reconstruction but allow complex meanings be built within local contexts through associative/shallow processing (sentences presented in backward order), b) grammatically well-formed semantically implausible sentences impede heuristic processing. Using novel behavioral paradigm, we demonstrate backward- indeed recovery of grammatical structure during comprehension. Critically, backward-presented elicit relatively low response areas, as measured fMRI. In contrast, areas similar magnitude naturalistic (plausible) sentences. other words, ability build structures both necessary sufficient fully engage network. Taken together, results provide strongest date support generalized reliance human on parsing. Significance statement Whether relies predominantly structural (syntactic) cues or meaning- related (semantic) remains debated. shed light this question areas’ where semantic pitted against each other, using find respond weakly composition cannot parsed syntactically—as confirmed paradigm—and they strongly sentences, like famous ‘Colorless green ideas sleep furiously’ sentence. These findings accounts suggest can foregone favor shallow

Language: Английский

---

The neurobiology of sentence production: A narrative review and meta-analysis

Brain and Language, Journal Year: 2025, Volume and Issue: 264, P. 105549 - 105549

Published: Feb. 20, 2025

Although there is a sizeable body of literature on sentence comprehension and processing both in healthy disordered language users, the production remains much more sparse. Linguistic computational descriptions expressive syntactic deficits aphasia are especially rare. In addition, neuroimaging (psycho) linguistic literatures operate largely separately. this paper, I will first lay out theoretical land with regard to psycholinguistic models production. then provide brief narrative overview large-scale meta-analysis as it pertains computation, followed by an attempt integrate findings from functional clinical neuroimaging. Finally, surrounding propose path forward close some existing gaps.

Language: Английский

---

The cerebellar components of the human language network

bioRxiv (Cold Spring Harbor Laboratory), Journal Year: 2025, Volume and Issue: unknown

Published: April 19, 2025

Abstract The cerebellum’s capacity for neural computation is arguably unmatched. Yet despite evidence of cerebellar contributions to cognition, including language, its precise role remains debated. Here, we systematically evaluate language-responsive regions using precision fMRI. We identify four that respond language across modalities (Experiments 1a-b, n=754). One region—spanning Crus I/II/lobule VIIb—is selective relative diverse non-linguistic tasks 2a-f, n=732), and the rest exhibit mixed-selective profiles. Like neocortical system, language-selective region engaged by sentence-level meanings during comprehension production 3a-c, n=105), but it less sensitive than neocortex individual word grammatical structure. Finally, all regions, especially I/II/VIIb, are functionally connected system (Experiment 4, n=85). propose these constitute components extended network, with one supporting semantic processing, other three plausibly integrating information from regions.

Language: Английский

---

Constructed languages are processed by the same brain mechanisms as natural languages

Proceedings of the National Academy of Sciences, Journal Year: 2025, Volume and Issue: 122(12)

Published: March 17, 2025

What constitutes a language? Natural languages share features with other domains: from math, to music, gesture. However, the brain mechanisms that process linguistic input are highly specialized, showing little response diverse nonlinguistic tasks. Here, we examine constructed (conlangs) ask whether they draw on same neural as natural or instead pattern domains like math and programming languages. Using individual-subject fMRI analyses, show understanding conlangs recruits areas language comprehension. This result holds for Esperanto (n = 19 speakers) four fictional [Klingon 10), Na’vi 9), High Valyrian 3), Dothraki 3)]. These findings suggest critical allow them representations computations, implemented in left-lateralized network of areas. The differentiate languages—including recent creation by single individual, often an esoteric purpose, small number speakers, fact these typically learned adulthood—appear not be consequential reliance cognitive mechanisms. We argue shared feature is symbolic systems capable expressing open-ended range meanings about our outer inner worlds.

Language: Английский

---

Intuitive physical reasoning is not mediated by linguistic nor exclusively domain-general abstract representations

Neuropsychologia, Journal Year: 2025, Volume and Issue: unknown, P. 109125 - 109125

Published: March 1, 2025

Language: Английский

---

Beyond Markov: Transformers, memory, and attention

Cognitive Neuroscience, Journal Year: 2025, Volume and Issue: unknown, P. 1 - 19

Published: April 15, 2025

This paper asks what predictive processing models of brain function can learn from the success transformer architectures. We suggest that reason architectures have been successful is they implicitly commit to a non-Markovian generative model - in which we need memory contextualize our current observations and make predictions about future. Interestingly, both notions working cognitive science rely heavily upon concept attention. will argue move beyond Markov crucial construction capable dealing with much sequential data certainly language brains contend with. characterize two broad approaches this problem deep temporal hierarchies autoregressive transformers being an example latter. Our key conclusions are benefit their use embedding spaces place strong metric priors on implicit latent variable utilize direct form attention highlights most relevant, not only recent, previous elements sequence help predict next.

Language: Английский

---

Neural populations in the language network differ in the size of their temporal receptive windows

bioRxiv (Cold Spring Harbor Laboratory), Journal Year: 2022, Volume and Issue: unknown

Published: Dec. 30, 2022

Abstract Despite long knowing what brain areas support language comprehension, our knowledge of the neural computations that these frontal and temporal regions implement remains limited. One important unresolved question concerns functional differences among populations comprise network. Leveraging high spatiotemporal resolution intracranial recordings, we examined responses to sentences linguistically degraded conditions discovered three response profiles differ in their dynamics. These appear reflect different receptive windows (TRWs), with average TRWs about 1, 4, 6 words, as estimated a simple one-parameter model. Neural exhibiting are interleaved across network, which suggests all have direct access distinct, multi-scale representations linguistic input—a property may be critical for efficiency robustness processing.

Language: Английский

---

A 3.5-minute-long reading-based fMRI localizer for the language network

bioRxiv (Cold Spring Harbor Laboratory), Journal Year: 2024, Volume and Issue: unknown

Published: July 3, 2024

Abstract The field of human cognitive neuroscience is increasingly acknowledging inter-individual differences in the precise locations functional areas and corresponding need for individual-level analyses fMRI studies. One approach to identifying networks within individual brains based on robust extensively validated ‘localizer’ paradigms—contrasts conditions that aim isolate some mental process interest. Here, we present a new version localizer fronto-temporal language-selective network. This similar commonly-used reading sentences nonword sequences (Fedorenko et al., 2010) but uses speeded presentation (200ms per word/nonword). Based direct comparison between standard (450ms word/nonword) versions language 24 participants, show single run (3.5 min) highly effective at areas: indeed, it more than given leads an increased response critical (sentence) condition decreased control (nonwords) condition. may therefore become choice network neurotypical adults or special populations (as long as they are proficient readers), especially when time essence.

Language: Английский

---

A low-activity cortical network selectively encodes syntax

bioRxiv (Cold Spring Harbor Laboratory), Journal Year: 2024, Volume and Issue: unknown

Published: June 20, 2024

Abstract Syntax, the abstract structure of language, is a hallmark human cognition. Despite its importance, neural underpinnings remain obscured by inherent limitations non-invasive brain measures and near total focus on comprehension paradigms. Here, we address these with high-resolution neurosurgical recordings (electrocorticography) controlled sentence production experiment. We uncover three syntactic networks that are broadly distributed across traditional language regions, but focal concentrations in middle inferior frontal gyri. In contrast to previous findings from studies, process syntax mostly exclusion words meaning, supporting cognitive architecture distinct system. Most strikingly, our data reveal an unexpected property syntax: it encoded independent activity levels. propose this “low-activity coding” scheme represents novel mechanism for encoding information, reserved higher-order cognition more broadly.

Language: Английский

---