pith. sign in

arxiv: 2512.20481 · v4 · submitted 2025-12-23 · 🧬 q-bio.NC · cs.CL

Coherence in the brain unfolds across separable temporal regimes

Davide Staub , Finn Rabe , Akhil Misra , Yves Pauli , Roya H\"uppi , Ni Yang , Nils Lang , Lars Michels
show 5 more authors
Victoria Edkins Sascha Fr\"uhholz Iris Sommer Wolfram Hinzen Philipp Homan
This is my paper

Pith reviewed 2026-05-16 20:09 UTC · model grok-4.3

classification 🧬 q-bio.NC cs.CL
keywords language comprehensionfMRI encodinglarge language modelstemporal regimescoherencedrift and shiftdefault mode networkevent boundaries
0
0 comments X

The pith

The brain implements language coherence through distinct slow drift and rapid shift neural regimes.

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

The paper shows that to maintain coherence in language, the brain must balance gradual accumulation of meaning across context with rapid updates at event boundaries. Using signals extracted from a large language model processing narratives, the authors test these in high-resolution fMRI data from one subject listening to stories for over seven hours. Drift predictions align with default-mode network hubs supporting contextual integration, while shift predictions align with auditory and language areas for quick reconfigurations. This reveals that coherence emerges from separable yet co-occurring temporal regimes in the brain.

Core claim

Coherence during language comprehension is implemented through distinct but co-expressed neural regimes of slow contextual integration and rapid event-driven reconfiguration. Drift signals derived from an LLM were prevalent in default-mode network hubs, whereas shift signals were evident bilaterally in the primary auditory cortex and language association cortex, as shown in voxelwise encoding models fitted to densely sampled 7T fMRI data.

What carries the argument

Annotation-free drift and shift signals derived from a large language model processing the narrative input, which capture contextual accumulation and boundary-driven changes respectively, fed into regularized encoding models to predict hemodynamic responses.

If this is right

  • Drift and shift can be dissociated in their regional expression across the brain.
  • Language coherence relies on both slow integration in association areas and fast updates in sensory-language areas.
  • The approach offers a way to study disturbances in language coherence without manual annotations.
  • These regimes provide a mechanistic basis for understanding how the brain handles competing temporal demands in naturalistic settings.

Where Pith is reading between the lines

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

  • Similar drift-shift separation might apply to other domains involving narrative or sequential processing, such as memory consolidation.
  • Disruptions in one regime over the other could explain specific symptoms in language-related psychiatric conditions.
  • Future experiments could test if these signals generalize across different languages or story types.

Load-bearing premise

That the LLM-derived drift and shift signals reflect the brain's actual temporal processing requirements instead of incidental correlations with text statistics that drive the hemodynamic response.

What would settle it

A finding that drift and shift models do not show distinct regional prediction patterns, such as both performing equally across all brain areas or failing to predict the specified hubs and cortices separately.

Figures

Figures reproduced from arXiv: 2512.20481 by Akhil Misra, Davide Staub, Finn Rabe, Iris Sommer, Lars Michels, Nils Lang, Ni Yang, Philipp Homan, Roya H\"uppi, Sascha Fr\"uhholz, Victoria Edkins, Wolfram Hinzen, Yves Pauli.

Figure 1
Figure 1. Figure 1: Annotation-free mapping of narrative coherence to brain dynamics. [PITH_FULL_IMAGE:figures/full_fig_p003_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: Across stories activation and consistency maps for drift and shift. a) [PITH_FULL_IMAGE:figures/full_fig_p004_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: Unique predictive contributions of drift and shift. a) [PITH_FULL_IMAGE:figures/full_fig_p005_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: Replicable drift effects across integration timescales. [PITH_FULL_IMAGE:figures/full_fig_p006_4.png] view at source ↗
read the original abstract

To maintain coherence in language, the brain must satisfy key competing temporal demands: the gradual accumulation of meaning across extended context (drift) and the rapid reconfiguration of representations at event boundaries (shift). How these processes are implemented in the human brain during naturalistic listening remains unclear. Here, we tested whether both can be captured by annotation-free drift and shift signals and whether their neural expression shows distinct regional preferences across the brain. These signals were derived from a large language model (LLM) processing the narrative input. To enable high-precision voxelwise encoding models with stable parameter estimates, we densely sampled one healthy adult across more than 7 hours of listening to crime stories while collecting 7 Tesla fMRI data. We then modeled the feature-informed hemodynamic response using a regularized encoding framework validated on independent stories. Drift predictions were prevalent in default-mode network hubs, whereas shift predictions were evident bilaterally in the primary auditory cortex and language association cortex. Together, these findings show that coherence during language comprehension is implemented through distinct but co-expressed neural regimes of slow contextual integration and rapid event-driven reconfiguration, offering a mechanistic entry point for understanding disturbances of language coherence in psychiatric disorders.

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

2 major / 2 minor

Summary. The paper claims that annotation-free drift and shift signals extracted from an LLM processing narrative text can be used in voxelwise encoding models to reveal separable neural regimes supporting language coherence: slow contextual integration (drift) preferentially expressed in default-mode network hubs and rapid event-driven reconfiguration (shift) in bilateral auditory and language association cortex. This is demonstrated via regularized linear encoding models fit to >7 hours of 7T fMRI data from a single densely sampled subject listening to crime stories, with validation on held-out stories.

Significance. If the reported regional dissociation survives controls for low-level text statistics, the work would supply a concrete, mechanistic entry point for studying how the brain balances gradual context accumulation against boundary-driven updates during naturalistic language comprehension, with potential relevance to coherence disturbances in psychiatric conditions.

major comments (2)
  1. [Methods] Methods and Results: The central claim of distinct but co-expressed neural regimes rests on encoding-model predictions from a single subject. While dense sampling (>7 h) reduces within-subject variance, the absence of an independent replication cohort or cross-subject generalization test leaves open whether the DMN-drift versus auditory/language-shift dissociation generalizes beyond this individual.
  2. [Results] Results: No variance-partitioning analysis or comparison against baseline regressors (word rate, sentence boundaries, or lexical surprisal) is reported. Without such controls it remains possible that the LLM-derived drift and shift features largely proxy low-level input statistics known to drive BOLD responses, undermining the interpretation that they specifically index competing temporal demands of coherence.
minor comments (2)
  1. [Abstract] Abstract: The phrase 'annotation-free' is used without clarifying that the LLM itself was pretrained on large text corpora that overlap the narrative domain; a brief qualification would improve precision.
  2. [Figures] Figure legends: The color scales and significance thresholds for the encoding-model maps are not stated explicitly; adding these details would aid reproducibility.

Simulated Author's Rebuttal

2 responses · 1 unresolved

We thank the referee for their constructive and detailed comments, which have helped us identify areas for improvement. We address each major comment point by point below, providing the strongest honest defense of the current work while acknowledging genuine limitations.

read point-by-point responses

  1. Referee: [Methods] Methods and Results: The central claim of distinct but co-expressed neural regimes rests on encoding-model predictions from a single subject. While dense sampling (>7 h) reduces within-subject variance, the absence of an independent replication cohort or cross-subject generalization test leaves open whether the DMN-drift versus auditory/language-shift dissociation generalizes beyond this individual.

    Authors: We acknowledge that the study relies on a single densely sampled subject. This design choice enables stable voxelwise encoding model fits and high-precision within-subject inference, which is a recognized strength in naturalistic fMRI paradigms requiring extensive data per participant. However, we agree that the absence of cross-subject generalization tests is a limitation for claims of broader applicability. In revision we will expand the Discussion to explicitly note this constraint and propose future multi-subject extensions, but we cannot add new cohort data to the current manuscript. revision: partial

  2. Referee: [Results] Results: No variance-partitioning analysis or comparison against baseline regressors (word rate, sentence boundaries, or lexical surprisal) is reported. Without such controls it remains possible that the LLM-derived drift and shift features largely proxy low-level input statistics known to drive BOLD responses, undermining the interpretation that they specifically index competing temporal demands of coherence.

    Authors: We agree this control is necessary to strengthen the interpretation. We will add variance-partitioning analyses that quantify the unique variance explained by the drift and shift features after accounting for baseline regressors including word rate, sentence boundaries, and lexical surprisal. These results, with statistical comparisons, will be incorporated into the revised Results section to demonstrate that the LLM-derived signals capture coherence-related temporal structure beyond low-level text statistics. revision: yes

standing simulated objections not resolved
  • The current single-subject dataset does not permit direct testing of cross-subject generalization without new data collection.

Circularity Check

0 steps flagged

No significant circularity in derivation chain

full rationale

The paper extracts drift and shift signals directly from an LLM applied to the raw narrative text, without any fitting to the fMRI measurements. These independent features are then fed into regularized linear encoding models whose parameters are estimated on training stories and evaluated on held-out stories. The reported regional dissociation (DMN hubs for drift predictions, auditory/language cortex for shift predictions) is an empirical outcome of this mapping rather than a quantity that reduces by construction to the input features or to any self-citation. No self-definitional equations, fitted parameters renamed as predictions, or load-bearing self-citations appear in the provided derivation steps. The central claim therefore remains externally falsifiable against the brain data.

Axiom & Free-Parameter Ledger

1 free parameters · 2 axioms · 0 invented entities

The central claim rests on the assumption that LLM next-token statistics can be decomposed into drift and shift components that correspond to brain mechanisms, plus standard fMRI encoding assumptions. No new entities are postulated.

free parameters (1)
  • regularization strength in encoding models
    Chosen to stabilize voxel-wise fits on the dense single-subject data; value not reported in abstract.
axioms (2)
  • domain assumption Hemodynamic response can be modeled as a linear convolution of feature time series with a canonical HRF
    Invoked in the regularized encoding framework section.
  • domain assumption LLM hidden-state trajectories contain separable slow and fast components that align with human temporal integration demands
    Core premise for deriving drift and shift signals from the model.

pith-pipeline@v0.9.0 · 5544 in / 1503 out tokens · 43384 ms · 2026-05-16T20:09:19.936786+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

49 extracted references · 49 canonical work pages · 2 internal anchors

  1. [1]

    N. C. Andreasen, Thought, language, and communication disorders. i. clinical assessment, definition of terms, and evaluation of their reliability, Archives of GeneralPsychiatry36(12)(1979)1315–1321.doi:10.1001/archpsyc.1979. 01780120045006

    work page doi:10.1001/archpsyc.1979 1979

  2. [2]

    Kircher, H

    T. Kircher, H. Bröhl, F. Meier, J. Engelen, Formal thought disorders: From phenomenology to neurobiology, The Lancet Psychiatry 5 (6) (2018) 515–526, doi: https://dx.doi.org/10.1016/s2215-0366(18)30059-2. URL https://doi.org/10.1016/s2215-0366(18)30059-2

    work page doi:10.1016/s2215-0366(18)30059-2 2018

  3. [3]

    Cavelti, T

    M. Cavelti, T. Kircher, A. Nagels, W. Strik, P. Homan, Is formal thought disorder in schizophrenia related to structural and functional aberrations in the language network? A systematic review of neuroimaging findings, Schizophrenia Re- search (199) (2018) 2–16, doi: https://dx.doi.org/10.1016/j.schres.2018.02.051. URL https://doi.org/10.1016/j.schres.2018.02.051

    work page doi:10.1016/j.schres.2018.02.051 2018

  4. [4]

    URL http://dx.doi.org/10.1093/schbul/sbac159

    L.Palaniyappan,P.Homan,M.F.Alonso-Sanchez,Languagenetworkdysfunction andformalthoughtdisorderinschizophrenia,SchizophreniaBulletin49(2)(2023) 486–497.doi:10.1093/schbul/sbac159. URL http://dx.doi.org/10.1093/schbul/sbac159

    work page doi:10.1093/schbul/sbac159 2023

  5. [5]

    Ojemann, J

    G. Ojemann, J. Ojemann, E. Lettich, M. Berger, Cortical language localization in left, dominant hemisphere: An electrical stimulation mapping investigation in 117 patients, Journal of Neurosurgery 71 (3) (1989) 316–326.doi:10.3171/ jns.1989.71.3.0316. 12

    work page 1989

  6. [6]

    Fedorenko, N

    E. Fedorenko, N. Kanwisher, Neuroimaging of language: Why hasn’t a clearer picture emerged?, Language and Linguistics Compass 3 (4) (2009) 839–865. doi:10.1111/j.1749-818X.2009.00143.x

    work page doi:10.1111/j.1749-818x.2009.00143.x 2009

  7. [7]

    doi:10.1016/j.neuroimage.2012.04.062

    C.J.Price,Areviewandsynthesisofthefirst20yearsofPETandfMRIstudiesof heard speech, spoken language and reading, NeuroImage 62 (2) (2012) 816–847. doi:10.1016/j.neuroimage.2012.04.062

    work page doi:10.1016/j.neuroimage.2012.04.062 2012

  8. [8]

    A. G. Huth, W. A. de Heer, T. L. Griffiths, F. E. Theunissen, J. L. Gallant, Natural speech reveals the semantic maps that tile human cerebral cortex, Nature 532 (7600) (2016) 453–458.doi:10.1038/nature17637

    work page doi:10.1038/nature17637 2016

  9. [9]

    E.Simony,C.J.Honey,J.Chen,O.Lositsky,Y.Yeshurun,A.Wiesel,U.Hasson, Dynamic reconfiguration of the default mode network during narrative compre- hension,NatureCommunications7(2016)12141.doi:10.1038/ncomms12141

    work page doi:10.1038/ncomms12141 2016

  10. [10]

    Yeshurun, S

    Y. Yeshurun, S. Swanson, E. Simony, J. Chen, C. Lazaridi, C. J. Honey, U. Hasson, Same story, different story: The neural representation of inter- pretive frameworks, Psychological Science 28 (3) (2017) 307–319.doi: 10.1177/0956797616682029

    work page doi:10.1177/0956797616682029 2017

  11. [11]

    C.H.C.Chang,C.Lazaridi,Y.Yeshurun,K.A.Norman,U.Hasson,Information flowacrossthecorticaltimescalehierarchyduringnarrativecomprehension,Pro- ceedings of the National Academy of Sciences of the United States of America 119 (49) (2022) e2209307119.doi:10.1073/pnas.2209307119

    work page doi:10.1073/pnas.2209307119 2022

  12. [12]

    M.Schrimpf,I.A.Blank,G.Tuckute,C.Kauf,E.A.Hosseini,N.Kanwisher,J.B. Tenenbaum, E.Fedorenko,Theneuralarchitectureoflanguage: Integrativemod- eling converges on predictive processing, Proceedings of the National Academy of Sciences of the United States of America 118 (45) (2021) e2105646118. doi:10.1073/pnas.2105646118

    work page doi:10.1073/pnas.2105646118 2021

  13. [13]

    Caucheteux, J.-R

    C. Caucheteux, J.-R. King, Brains and algorithms partially converge in natural language processing, Communications Biology 5 (2022) 134.doi:10.1038/ s42003-022-03036-1

    work page 2022

  14. [14]

    A.Goldstein,Z.Zada,E.Buchnik,M.Schain,A.Price,B.Aubrey,S.A.Nastase, A.Feder,D.Emanuel,A.Cohen,U.Hasson,Sharedcomputationalprinciplesfor language processing in humans and deep language models, Nature Neuroscience 25 (3) (2022) 369–380.doi:10.1038/s41593-022-01026-4

    work page doi:10.1038/s41593-022-01026-4 2022

  15. [15]

    Antonello, A

    R. Antonello, A. Huth, Predictive coding or just feature discovery? an alternative account of why language models fit brain data, Neurobiology of Language 5 (1) (2024) 64–79.doi:10.1162/nol_a_00087

    work page doi:10.1162/nol_a_00087 2024

  16. [16]

    Y.Lerner,C.J.Honey,L.J.Silbert,U.Hasson,Topographicmappingofahierar- chyoftemporalreceptivewindowsusinganarratedstory,JournalofNeuroscience 31 (8) (2011) 2906–2915.doi:10.1523/JNEUROSCI.3684-10.2011. 13

    work page doi:10.1523/jneurosci.3684-10.2011 2011

  17. [17]

    C. J. Honey, T. Thesen, T. H. Donner, L. J. Silbert, C. E. Carlson, O. Devinsky, W. K. Doyle, N. Rubin, D. J. Heeger, U. Hasson, Slow cortical dynamics and the accumulationofinformationoverlongtimescales,Neuron76(2)(2012)423–434. doi:10.1016/j.neuron.2012.08.011

    work page doi:10.1016/j.neuron.2012.08.011 2012

  18. [18]

    J. M. Zacks, N. K. Speer, K. M. Swallow, T. S. Braver, J. R. Reynolds, Event perception: A mind-brain perspective, Psychological Bulletin 133 (2) (2007) 273–293.doi:10.1037/0033-2909.133.2.273

    work page doi:10.1037/0033-2909.133.2.273 2007

  19. [19]

    C. A. Kurby, J. M. Zacks, Segmentation in the perception and memory of events, Trends in Cognitive Sciences 12 (2) (2008) 72–79.doi:10.1016/j.tics. 2007.11.004

    work page doi:10.1016/j.tics 2008

  20. [20]

    C.Baldassano,J.Chen,A.Zadbood,J.W.Pillow,U.Hasson,K.A.Norman,Dis- covering event structure in continuous narrative perception and memory, Neuron 95 (3) (2017) 709–721.e5.doi:10.1016/j.neuron.2017.06.041

    work page doi:10.1016/j.neuron.2017.06.041 2017

  21. [21]

    J. Chen, Y. C. Leong, C. J. Honey, C. S. Yong, K. A. Norman, U. Hasson, Shared memories reveal shared structure in neural activity across individuals, Nature Neuroscience 20 (1) (2017) 115–125.doi:10.1038/nn.4450

    work page doi:10.1038/nn.4450 2017

  22. [22]

    Nguyen, T

    M. Nguyen, T. Vanderwal, U. Hasson, Shared understanding of narratives is correlated with shared neural responses, NeuroImage 184 (2019) 161–170.doi: 10.1016/j.neuroimage.2018.09.010

    work page doi:10.1016/j.neuroimage.2018.09.010 2019

  23. [23]

    H. Song, E. S. Finn, M. D. Rosenberg, Cognitive and neural state dynamics of narrative comprehension, Journal of Neuroscience 41 (20) (2021) 4420–4431. doi:10.1523/JNEUROSCI.0037-21.2021

    work page doi:10.1523/jneurosci.0037-21.2021 2021

  24. [24]

    C.Whitney,W.Huber,J.Klann,S.Weis,S.Krach,T.Kircher,Neuralcorrelatesof narrative shifts during auditory story comprehension, NeuroImage 47 (1) (2009) 360–366.doi:10.1016/j.neuroimage.2009.04.037

    work page doi:10.1016/j.neuroimage.2009.04.037 2009

  25. [25]

    Geerligs, M

    L. Geerligs, M. A. J. van Gerven, K. L. Campbell, A partially nested cortical hierarchy of neural states underlies event segmentation in the human brain, eLife 11 (2022) e77430.doi:10.7554/eLife.77430

    work page doi:10.7554/elife.77430 2022

  26. [26]

    Anurova, S

    I. Anurova, S. Vetchinnikova, A. Dobrego, N. Williams, N. Mikusova, A. Suni, A.Mauranen,S.Palva,Event-relatedresponsesreflectchunkboundariesinnatural speech,NeuroImage258(2022)119346.doi:10.1016/j.neuroimage.2022. 119346

    work page doi:10.1016/j.neuroimage.2022 2022

  27. [28]

    How, When and Why Proteins Col- lapse: The Relation to Folding

    U. Hasson, G. Egidi, M. Marelli, R. M. Willems, Grounding the neurobiology of language in first principles: The necessity of non-language-centric explanations for language comprehension, Cognition 180 (2018) 135–157.doi:10.1016/j. cognition.2018.06.018. 14

    work page doi:10.1016/j 2018

  28. [29]

    S. A. Nastase, A. Goldstein, U. Hasson, Keep it real: Rethinking the primacy of experimental control in cognitive neuroscience, NeuroImage 222 (2020) 117254. doi:10.1016/j.neuroimage.2020.117254

    work page doi:10.1016/j.neuroimage.2020.117254 2020

  29. [30]

    Narratives

    S.A.Nastase,Y.-F.Liu,H.Hillman,A.Zadbood,L.Hasenfratz,N.Keshavarzian, J. Chen, C. J. Honey, Y. Yeshurun, M. Regev, M. Nguyen, C. H. C. Chang, C. Baldassano, O. Lositsky, E. Simony, M.-A. Chow, Y. C. Leong, P. P. Brooks, E. Micciche, G. Choe, A. Goldstein, T. Vanderwal, Y. O. Halchenko, K. A. Norman, U. Hasson, The “Narratives” fMRI dataset for evaluating ...

    work page doi:10.1038/s41597-021-01033-3 2021

  30. [31]

    S. Jain, A. Huth, Incorporating context into language encoding models for fMRI, in: Advances in Neural Information Processing Systems, Vol. 31, 2018

    work page 2018

  31. [32]

    Michelmann, M

    S. Michelmann, M. Kumar, K. A. Norman, M. Toneva, Large language models can segment narrative events similarly to humans, Behavior Research Methods 57 (1) (2025) 39.doi:10.3758/s13428-024-02569-z

    work page doi:10.3758/s13428-024-02569-z 2025

  32. [33]

    R. S. Desikan, F. Ségonne, B. Fischl, B. T. Quinn, B. C. Dickerson, D. Blacker, R. L. Buckner, A. M. Dale, R. P. Maguire, B. T. Hyman, et al., An automated labelingsystemforsubdividingthehumancerebralcortexonmriscansintogyral based regions of interest, Neuroimage 31 (3) (2006) 968–980

    work page 2006

  33. [34]

    Hasson, E

    U. Hasson, E. Yang, I. Vallines, D. J. Heeger, N. Rubin, A hierarchy of temporal receptive windows in human cortex, Journal of Neuroscience 28 (10) (2008) 2539–2550.doi:10.1523/JNEUROSCI.5487-07.2008

    work page doi:10.1523/jneurosci.5487-07.2008 2008

  34. [35]

    Ben-Yakov, R

    A. Ben-Yakov, R. N. Henson, The hippocampal film editor: sensitivity and speci- ficity to event boundaries in continuous experience, Journal of Neuroscience 38 (47) (2018) 10057–10068

    work page 2018

  35. [36]

    M.Silva,C.Baldassano,etal.,Rapidmemoryreactivationatmovieeventbound- aries promotes episodic encoding, Journal of Neuroscience 39 (43) (2019) 8538– 8548

    work page 2019

  36. [37]

    Steinhauer, K

    K. Steinhauer, K. Alter, A. D. Friederici, Brain potentials indicate immediate use of prosodic cues in natural speech processing, Nature Neuroscience 4 (2) (2001) 191–196.doi:10.1038/84014

    work page doi:10.1038/84014 2001

  37. [38]

    Meyer, K

    M. Meyer, K. Steinhauer, K. Alter, A. D. Friederici, D. Y. von Cramon, Brain activity varies with modulation of prosodic boundaries in natural speech, Journal of Cognitive Neuroscience 14 (4) (2002) 520–536.doi:10.1162/ 08989290260045817

    work page 2002

  38. [39]

    A. K. Ischebeck, A. D. Friederici, K. Alter, Processing prosodic boundaries in naturalandhummedspeech,NeuroImage39(2)(2008)714–724.doi:10.1016/ j.neuroimage.2007.09.019. 15

    work page 2008

  39. [40]

    , author Poeppel, D

    A.-L. Giraud, D. Poeppel, Cortical oscillations and speech processing: emerging computationalprinciplesandoperations,NatureNeuroscience15(4)(2012)511– 517.doi:10.1038/nn.3063

    work page doi:10.1038/nn.3063 2012

  40. [41]

    N. Ding, J. Z. Simon, Neural coding of continuous speech in auditory cortex during monaural and dichotic listening, Journal of Neuroscience 32 (46) (2012) 16293–16304.doi:10.1523/JNEUROSCI.2596-12.2012

    work page doi:10.1523/jneurosci.2596-12.2012 2012

  41. [42]

    K.J.Forseth,G.Hickok,P.S.Rollo,N.Tandon,Languagepredictionmechanisms in human auditory cortex, Nature Communications 11 (1) (2020) 5240.doi: 10.1038/s41467-020-19010-6

    work page doi:10.1038/s41467-020-19010-6 2020

  42. [43]

    Scaling speech technology to 1,000+ languages,

    V. Pratap, A. Tjandra, B. Shi, P. Tomasello, A. Babu, S. Kundu, A. Elkahky, Z. Ni, A. Vyas, M. Fazel-Zarandi, A. Baevski, Y. Adi, X. Zhang, W.-N. Hsu, A. Conneau, M. Auli, Scaling speech technology to 1,000+ languages, arXiv (2023).arXiv:2305.13516,doi:10.48550/arXiv.2305.13516. URL https://arxiv.org/abs/2305.13516

    work page doi:10.48550/arxiv.2305.13516 2023

  43. [44]

    PyTorch Audio Team, Forced alignment for multilingual data (mms fa), https://docs.pytorch.org/audio/main/tutorials/forced_alignment_for_ multilingual_data_tutorial.html, accessed 2025-09-03 (2024)

    work page 2025

  44. [45]

    The Llama 3 Herd of Models

    A. Grattafiori, et al., The llama 3 herd of models, arXiv (2024).arXiv:2407. 21783,doi:10.48550/arXiv.2407.21783. URL https://arxiv.org/abs/2407.21783

    work page internal anchor Pith review Pith/arXiv arXiv doi:10.48550/arxiv.2407.21783 2024

  45. [46]

    O.Esteban,C.J.Markiewicz,R.W.Blair,C.A.Moodie,A.I.Isik,A.Erramuzpe, J. D. Kent, M. Goncalves, E. DuPre, M. Snyder, H. Oya, S. S. Ghosh, J. Wright, J. Durnez, R. A. Poldrack, K. J. Gorgolewski, fmriprep: a robust preprocessing pipeline for functional mri, Nature Methods 16 (1) (2019) 111–116.doi:10. 1038/s41592-018-0235-4

    work page 2019

  46. [47]

    Jenkinson, P

    M. Jenkinson, P. Bannister, M. Brady, S. Smith, Improved optimization for the robust and accurate linear registration and motion correction of brain images, NeuroImage17(2)(2002)825–841.doi:10.1016/S1053-8119(02)91132-8

    work page doi:10.1016/s1053-8119(02)91132-8 2002

  47. [48]

    Zhang, A

    J.Hwang,M.Hira,C.Chen,X.Zhang,Z.Ni,G.Sun,P.Ma,R.Huang,V.Pratap, Y. Zhang, A. Kumar, C.-Y. Yu, C. Zhu, C. Liu, J. Kahn, M. Ravanelli, P. Sun, S.Watanabe,Y.Shi,Y.Tao,etal.,Torchaudio2.1: Advancingspeechrecognition, self-supervised learning, and audio processing components for pytorch, arXiv (2023).arXiv:2310.17864. URL https://arxiv.org/abs/2310.17864

    work page arXiv 2023

  48. [49]

    Lanczos, Applied Analysis, Prentice-Hall, 1956

    C. Lanczos, Applied Analysis, Prentice-Hall, 1956

    work page 1956

  49. [50]

    LLM.int8(): 8-bit Matrix Multiplication for Transformers at Scale

    T. Dettmers, M. Lewis, Y. Belkada, L. Zettlemoyer, LLM.int8(): 8-bit matrix multiplicationfortransformersatscale,arXiv(2022).arXiv:2208.07339,doi: 10.48550/arXiv.2208.07339. URL https://arxiv.org/abs/2208.07339 16 Supplementary Information Supplementary Methods MRI acquisition and preprocessing (full) Scanner: 7T Siemens MAGNETOM Terra; 32-channel head co...

    work page internal anchor Pith review Pith/arXiv arXiv doi:10.48550/arxiv.2208.07339 2022