pith. sign in

arxiv: 2605.01537 · v1 · submitted 2026-05-02 · 💻 cs.CL

The grip of grammar on meaning uncertainty: cross-linguistic evidence, neural correlates, and clinical relevance

Rui He , Claudio Palominos , Samuele Vallisa , Ni Yang , Han Zhang , Miguel \'Angel Santos Santos , Neguine Rezaii , Sergi Valero
show 11 more authors
Yonghua Huang Huan Li Hong Jiang Yongjun Peng Maria Francisca Alonso-S\'anchez Frederike Stein Tilo Kircher Philipp Homan Iris Sommer Lena Palaniyappan Wolfram Hinzen
This is my paper

Pith reviewed 2026-05-09 14:06 UTC · model grok-4.3

classification 💻 cs.CL
keywords grammarmeaning uncertaintysurprisalcross-linguistic analysisfMRIlanguage disordersaphasiaschizophrenia
0
0 comments X

The pith

Grammar reduces uncertainty in word meanings across 20 languages, with corresponding brain activity and selective clinical disruptions.

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

Isolated words carry uncertain meanings that become clearer when placed in grammatical context. The paper demonstrates this uncertainty reduction through comparisons of word frequency predictions versus grammar-aware contextual models in stories from 20 languages. This compression effect aligns with brain signals during language tasks and is diminished in conditions affecting language processing like aphasia and schizophrenia. It also connects to sentence structure complexity and word variety. These observations frame grammar as a tool for managing informational uncertainty in communication.

Core claim

Grammar compresses meaning uncertainty cross-linguistically, reflected in brain activity and selectively disrupted in disorders. This was measured as the relative difference between non-contextual surprisal from lexical frequency and contextual surprisal from grammar-sensitive models. In narratives from 20 languages, the contextual surprisal reduced the frequency-based one, tracking word order reversal costs and scaling with richer lexis and optimal dependency structures. During fMRI, this reduction explained BOLD activity in comprehension and production regions, and it was attenuated in aphasia, dementia, and schizophrenia.

What carries the argument

The relative difference between non-contextual surprisal from lexical frequency and contextual surprisal estimated by grammar-sensitive models, which quantifies grammar's role in reducing meaning uncertainty.

If this is right

  • Contextual surprisal reduction closely tracks the cost of reversing word order in sentences.
  • This reduction scales with richer, non-redundant vocabulary organized by complex but optimal dependency structures.
  • Surprisal and its reduction explain BOLD activity in overlapping but distinct brain regions for comprehension and production.
  • Uncertainty reduction via grammar is significantly attenuated in aphasia, dementia, and schizophrenia, but intact in disorders without primary language deficits.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • If grammar's compression of uncertainty is foundational, it may explain cross-linguistic preferences for certain syntactic patterns.
  • This could lead to new ways of modeling language acquisition or evolution based on uncertainty minimization.
  • Clinically, it suggests potential for assessing language disorders by measuring uncertainty reduction rather than traditional tests.
  • Connections might exist to how other cognitive systems handle uncertainty, like in decision making or perception.

Load-bearing premise

The difference between surprisal based on word frequency alone and surprisal from models using grammar actually captures grammar's unique contribution without being influenced much by meaning or other factors.

What would settle it

Observing no reduction in uncertainty from grammar in a new set of languages or finding that brain activity correlations disappear when controlling for other variables.

Figures

Figures reproduced from arXiv: 2605.01537 by Claudio Palominos, Frederike Stein, Han Zhang, Hong Jiang, Huan Li, Iris Sommer, Lena Palaniyappan, Maria Francisca Alonso-S\'anchez, Miguel \'Angel Santos Santos, Neguine Rezaii, Ni Yang, Philipp Homan, Rui He, Samuele Vallisa, Sergi Valero, Tilo Kircher, Wolfram Hinzen, Yonghua Huang, Yongjun Peng.

Figure 1
Figure 1. Figure 1: Cross-linguistic uncertainty reduction from lexical to contextual surprisal. (A) Example of uncertainty estimation. Each word is assigned a non￾contextual surprisal based on its lexical frequency (WordFreq) and a contextual surprisal from a grammar-sensitive language model (BERT). A third row shows surprisal for the same words presented in reversed order, which preserves the multiset of words but disrupts … view at source ↗
Figure 2
Figure 2. Figure 2: Formal syntactic organization and lexical s view at source ↗
Figure 3
Figure 3. Figure 3: Neural correlates of linguistic uncertainty and its reduction. view at source ↗
Figure 4
Figure 4. Figure 4: Selective disruption of uncertainty optimization in clinical populations. Panel A indicated the direction of changes in surprisal reduction in different disorders. In panels B–F, each cell shows a Spearman’s correlation coefficient between an uncertainty measure and a behavioral or clinical variable; numbers indicate the correlation value, and color encodes its sign and magnitude. Colored cells indicate si… view at source ↗
read the original abstract

Isolated word meanings are inherently uncertain. This uncertainty reduces when they are combined and anchored in context. We propose that grammar compresses meaning uncertainty cross-linguistically, which is reflected in brain and selectively disrupted in disorders. Compression was operationalized as the relative difference between non-contextual surprisal estimated from lexical frequency, and contextual surprisal from grammar-sensitive models. In narratives from 20 languages, contextual surprisal reduced frequency-based surprisal. This reduction closely tracked the surprisal cost of reversing word order, and scaled with richer, non-redundant lexis as organized by more complex but optimal dependency structure. During fMRI, surprisal and its reduction explained BOLD activity for comprehension and production in overlapping but distinct regions. Uncertainty reduction was significantly attenuated in aphasia, dementia, and schizophrenia, but remained intact where primary deficit is not language. These findings position uncertainty reduction via grammar as a foundational concept that illuminates principles, brain basis, and disruptions of language.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Referee Report

1 major / 2 minor

Summary. The paper claims that grammar compresses meaning uncertainty cross-linguistically. This is operationalized as the relative reduction from non-contextual surprisal (lexical frequency) to contextual surprisal (grammar-sensitive models). Evidence includes this reduction in narratives from 20 languages, its tracking of word-order reversal costs and scaling with dependency complexity, overlapping fMRI BOLD correlates during comprehension and production, and selective attenuation in aphasia, dementia, and schizophrenia (but not non-language disorders).

Significance. If the attribution to grammar holds after appropriate controls, the work would provide a unifying information-theoretic account linking grammatical structure to uncertainty reduction, with direct neural and clinical implications. The cross-linguistic scope and multi-method approach (behavioral, neuroimaging, patient data) would strengthen its potential impact in computational linguistics and cognitive neuroscience.

major comments (1)
  1. [Abstract] Abstract and operationalization section: the central metric (lexical-frequency surprisal minus contextual surprisal from grammar-sensitive models) does not isolate grammar's contribution. Models trained on full natural text encode semantic, pragmatic, and lexical co-occurrences in addition to syntax; without explicit baselines (e.g., semantics-only embeddings or purely syntactic parsers) or controls holding semantics constant while varying structure, the reported reductions, correlations with reversal cost and dependency complexity, fMRI overlaps, and clinical selectivity could be driven by non-grammatical factors. This is load-bearing for every downstream claim.
minor comments (2)
  1. [Results] Results section: directional effects are stated but quantitative statistics, sample sizes per language, model architectures, multiple-comparison corrections, and effect sizes are not reported in the provided abstract or summary; these details are required for reproducibility and evaluation of the strength of the cross-linguistic and clinical findings.
  2. [Methods] The manuscript should clarify the exact definition and training data of the 'grammar-sensitive models' (e.g., whether they are standard LMs or syntax-augmented) to allow readers to assess the degree of circularity in the surprisal subtraction.

Simulated Author's Rebuttal

1 responses · 0 unresolved

We thank the referee for their insightful comments, which identify a key methodological issue in attributing our findings specifically to grammar. We address this concern directly below and outline revisions to strengthen the isolation of grammatical contributions.

read point-by-point responses

  1. Referee: [Abstract] Abstract and operationalization section: the central metric (lexical-frequency surprisal minus contextual surprisal from grammar-sensitive models) does not isolate grammar's contribution. Models trained on full natural text encode semantic, pragmatic, and lexical co-occurrences in addition to syntax; without explicit baselines (e.g., semantics-only embeddings or purely syntactic parsers) or controls holding semantics constant while varying structure, the reported reductions, correlations with reversal cost and dependency complexity, fMRI overlaps, and clinical selectivity could be driven by non-grammatical factors. This is load-bearing for every downstream claim.

    Authors: We agree that models trained on natural text capture semantic, pragmatic, and lexical co-occurrence information in addition to syntax, so the contextual surprisal estimates reflect multiple levels of structure. Our core operationalization quantifies the reduction from non-contextual (frequency-based) surprisal to contextual surprisal, thereby measuring the benefit provided by context as it occurs in natural language, where grammar organizes that context. The manuscript already includes targeted evidence linking the reduction to grammar: the reduction closely tracks the surprisal cost of word-order reversals (a direct syntactic manipulation) and scales with dependency complexity (a syntactic measure of structure). These correlations would be unexpected if non-grammatical factors were the sole drivers. Nevertheless, to address the referee's concern more explicitly, we will add control analyses in the revised manuscript, including (i) semantics-only baselines using static embeddings without sequential context and (ii) purely syntactic models such as dependency parsers that supply structure while minimizing semantic content. These controls will be applied to the same 20-language narratives and used to re-evaluate the fMRI and clinical results. We view this as a necessary strengthening of the claims rather than a fundamental flaw in the current design. revision: yes

Circularity Check

0 steps flagged

No significant circularity; empirical measurements remain independent of inputs

full rationale

The paper explicitly operationalizes compression as the difference between unigram frequency surprisal and contextual model surprisal, then reports empirical findings that this difference tracks word-order reversal costs, scales with dependency structure, overlaps with fMRI activity, and attenuates selectively in certain disorders. These downstream correlations and clinical contrasts are not forced by the operationalization itself; they constitute separate measurements on the same narratives, brain data, and patient groups. No self-citations, uniqueness theorems, or ansatzes are invoked to justify the core attribution, and the models are treated as standard tools rather than fitted parameters renamed as predictions. The derivation chain is therefore self-contained against external benchmarks.

Axiom & Free-Parameter Ledger

0 free parameters · 2 axioms · 0 invented entities

The central claim rests on the premise that surprisal differences cleanly isolate grammar's contribution and that clinical attenuation reflects language-specific impairment rather than general cognitive decline.

axioms (2)
  • domain assumption Lexical-frequency surprisal provides a valid baseline for non-contextual meaning uncertainty.
    Used as the subtractive baseline for measuring contextual compression.
  • domain assumption Grammar-sensitive language models capture syntactic structure without being dominated by semantic or lexical factors.
    Required for interpreting the surprisal reduction as grammar-driven.

pith-pipeline@v0.9.0 · 5541 in / 1623 out tokens · 61961 ms · 2026-05-09T14:06:15.440797+00:00 · methodology

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Reference graph

Works this paper leans on

22 extracted references · 22 canonical work pages

  1. [1]

    Olness, G. S. & Ulatowska, H. K. Personal Narratives in Aphasia: Coherence in the Context of Use. Aphasiology 25, 1393–1413 (2011)

    work page 2011

  2. [2]

    L., Fromm, D

    Minga, J., Johnson, M., Blake, M. L., Fromm, D. & MacWhinney, B. Making Sense of Right Hemisphere Discourse Using RHDBank. Top Lang Disord 41, 99–122 (2021)

    work page 2021

  3. [3]

    & Gibson, E

    Rezaii, N., Mahowald, K., Ryskin, R., Dickerson, B. & Gibson, E. A syntax –lexicon trade-off in language production. Proceedings of the National Academy of Sciences 119, e2120203119 (2022)

    work page 2022

  4. [4]

    2020 , booktitle =

    Luz, S., Haider, F., Fuente, S. de la, Fromm, D. & MacWhinney, B. Alzheimer’s Dementia Recognition Through Spontaneous Speech: The ADReSS Challenge. in Interspeech 2020 2172–2176 (ISCA, 2020). doi:10.21437/Interspeech.2020-2571

    work page doi:10.21437/interspeech.2020-2571 2020

  5. [5]

    F., Limongi, R., Gati, J

    Alonso-Sánchez, M. F., Limongi, R., Gati, J. & Palaniyappan, L. Language network self -inhibition and semantic similarity in first -episode schizophrenia: A computational -linguistic and effective connectivity approach. Schizophrenia Research https://doi.org/10.1016/j.schres.2022.04.007 (2022) doi:10.1016/j.schres.2022.04.007

    work page doi:10.1016/j.schres.2022.04.007 2022

  6. [6]

    He, R. et al. Navigating the semantic space: Unraveling the structure of meaning in psychosis using different computational language models. Psychiatry Research 333, 115752 (2024)

    work page 2024

  7. [7]

    F., Sholomskas, D., Thompson, D., Belanger, A

    Leckman, J. F., Sholomskas, D., Thompson, D., Belanger, A. & Weissman, M. M. Best Estimate of Lifetime Psychiatric Diagnosis: A Methodological Study. Archives of General Psychiatry 39, 879–883 (1982)

    work page 1982

  8. [8]

    Liddle, P. F. et al. Thought and Language Index: an instrument for assessing thought and language in schizophrenia. The British Journal of Psychiatry 181, 326–330 (2002)

    work page 2002

  9. [9]

    G., Yang, L

    Opler, M. G., Yang, L. H., Caleo, S. & Alberti, P. Statistical validation of the criteria for symptom remission in schizophrenia: Preliminary findings. BMC Psychiatry 7, 35 (2007)

    work page 2007

  10. [10]

    Schneider, K. et al. Syntactic complexity and diversity of spontaneous speech production in schizophrenia 40 spectrum and major depressive disorders. Schizophr 9, 1–10 (2023)

    work page 2023

  11. [11]

    Tustison, N. J. et al. N4ITK: Improved N3 Bias Correction. IEEE Transactions on Medical Imaging 29, 1310–1320 (2010)

    work page 2010

  12. [12]

    B., Epstein, C

    Avants, B. B., Epstein, C. L., Grossman, M. & Gee, J. C. Symmetric diffeomorphic image registration with cross-correlation: Evaluating automated labeling of elderly and neurodegenerative brain. Medical Image Analysis 12, 26–41 (2008)

    work page 2008

  13. [13]

    & Smith, S

    Zhang, Y., Brady, M. & Smith, S. Segmentation of brain MR images through a hidden Markov random field model and the expectation-maximization algorithm. IEEE Transactions on Medical Imaging 20, 45–57 (2001)

    work page 2001

  14. [14]

    M., Fischl, B

    Dale, A. M., Fischl, B. & Sereno, M. I. Cortical Surface -Based Analysis: I. Segmentation and Surface Reconstruction. NeuroImage 9, 179–194 (1999)

    work page 1999

  15. [15]

    Klein, A. et al. Mindboggling morphometry of human brains. PLOS Computational Biology 13, e1005350 (2017)

    work page 2017

  16. [16]

    & Collins, D

    Fonov, V., Evans, A., McKinstry, R., Almli, C. & Collins, D. Unbiased nonlinear average age -appropriate brain templates from birth to adulthood. NeuroImage 47, Supplement 1, S102 (2009)

    work page 2009

  17. [17]

    & Smith, S

    Jenkinson, M., Bannister, P., Brady, M. & Smith, S. Improved Optimization for the Robust and Accurate Linear Registration and Motion Correction of Brain Images. NeuroImage 17, 825–841 (2002)

    work page 2002

  18. [18]

    Greve, D. N. & Fischl, B. Accurate and robust brain image alignment using boundary -based registration. NeuroImage 48, 63–72 (2009)

    work page 2009

  19. [19]

    Power, J. D. et al. Methods to detect, characterize, and remove motion artifact in resting state fMRI. NeuroImage 84, 320–341 (2014)

    work page 2014

  20. [20]

    & Liu, T

    Behzadi, Y., Restom, K., Liau, J. & Liu, T. T. A component based noise correction method (CompCor) for BOLD and perfusion based fMRI. NeuroImage 37, 90–101 (2007). 41

    work page 2007

  21. [21]

    Satterthwaite, T. D. et al. An improved framework for confound regression and filtering for control of motion artifact in the preprocessing of resting-state functional connectivity data. NeuroImage 64, 240–256 (2013)

    work page 2013

  22. [22]

    Evaluation of Noisy Data

    Lanczos, C. Evaluation of Noisy Data. Journal of the Society for Industrial and Applied Mathematics Series B Numerical Analysis 1, 76–85 (1964)

    work page 1964