CoChat

The DOI matches its Crossref record; 80 of the 81 references it cites were found in a registry or an index, and 1 are not checked yet; none of the 34 sources its claims cite is checked for support yet.

High-level prediction of continuous speech during mind-wandering

Open access

Abstract

Abundant evidence shows that when listening to speech or reading text, we continuously make high-level predictions about upcoming words – their content, meaning, and relationship to the overall narrative. Does this process stop when our mind wanders away and conscious experience becomes decoupled from the content of the speech? Conversely, is word prediction sufficient for conscious experience, or can these processes dissociate? We conducted an EEG study where participants (N = 25) listened to audiobooks (>12,000 words) while occasionally providing self-reports of mind-wandering. In line with current accounts, mind-wandering was associated with spectral changes in the signal, compared to subjectively attentive listening, as well as decreased early response to word onset. However, neural markers of word-level contextual surprise (indexed by the contextual surprise of large language models) remained intact during mind-wandering, alongside significant encoding of semantic content. Last, we measured the neural encoding of predictive context, reflected by the correlation between brain activity and the vector embeddings that language models use to predict each word given prior context. This encoding also persisted during mind-wandering with a weak decline compared to on-task periods, pointing towards a decreased ability to integrate information from multiple words. Our findings show that the brain predicts and monitors relevant inputs even when the subjective experience of the external environment is dimmed during mind-wandering. It carries implications for theoretical accounts of both mind-wandering and conscious awareness. This study examines how spontaneous drifts in listening relate to EEG responses to continuous speech, using encoding models of semantic and contextual representations. It finds that neural markers of prediction, surprisal, and word meaning persist when the mind wanders away from speech contents, even as perceptual responses weaken.

Checked 2026-11-19 · 1 reference not checked yet

Overview

CoChat Summary

TL;DR

In an EEG study with 25 Hebrew-speaking participants listening to >12,000 words of audiobooks, mind-wandering was linked to spectral changes and reduced early word-onset responses, but neural markers of word-level contextual surprise and semantic encoding persisted. Contextual embedding encoding also persisted during mind-wandering, with only a weak, temporally narrow decline around word onset.

Key findings

  • Mind-wandering reports were associated with spectral changes: on-task periods showed stronger theta and low beta power, and mind-wandering could be classified in all bands except delta and high beta (all Bonferroni-corrected p < 0.01).
  • Context-independent semantic encoding persisted during mind-wandering: on-task encoding was significant between 292–694 ms (permutation p < .001; mean 300–500 ms r = 0.0083, SD = 0.009), and mind-wandering encoding was significant between 51–171 ms, 312–473 ms, and 514–695 ms (permutation p < 0.01; mean 300–500 ms r = 0.0136, SD = 0.0118), with no reduction versus on-task (BF10 = 0.089 against on-task being stronger).
  • Word-onset response was weaker during mind-wandering: on-task versus mind-wandering cluster appeared during 55–371 ms (permutation p = .012); in the 100–300 ms window, mean difference = 0.119 μV, t(24) = 3.10, p = .005, Cohen’s d = 0.62, BF10 = 8.85.
  • Contextual surprise response persisted during mind-wandering: significant clusters for mind-wandering 152–628 ms (permutation p = .002) and on-task 191–941 ms (permutation p < .001), with no significant amplitude difference between states (max cluster p > .375); contextual embedding encoding showed a narrow on-task advantage around 30–151 ms (permutation p = .023), with mean encoding −100 to 100 ms r = 0.0244 (SD = 0.0135) on-task versus r = 0.0162 (SD = 0.0181) during mind-wandering (t(24) = −3.07, p = .005, Cohen’s d = −0.61, BF10 = 8.4).

Methods

  • EEG study: 25 native Hebrew speakers (17 females, 8 males; mean age 23.95, SD = 2.26) listened to six audiobook chapters totaling 132 minutes and >12,000 words; probes every 40–120 seconds asked about experience in the preceding 12 seconds, with 28.1% (SD = 10.0%) of responses labeled mind-wandering and post-listening comprehension accuracy 82.4% (SD = 7.4%).
  • Analyses used cluster-based spatial-temporal permutation tests, power spectral density classification, regression-based ERP modeling of word onset and LLM-derived contextual surprisal, and encoding models predicting EEG from word2vec semantic embeddings and Mistral 7B contextual embeddings.
  • Sample size, procedure, and analyses were not pre-registered; a Monte Carlo power analysis (10,000 iterations) estimated >80% power at 0.11 SD and 96% at 0.15 SD.

Limitations

  • Stated: the study treats “prediction” as a single entity rather than separating specific semantic and linguistic dimensions; mind-wandering was coded as a binary choice although theoretical and empirical accounts often describe it on multi-dimensional and continuous scales.
  • Stated: the 12-second pre-probe window may be contaminated by momentary on-task periods, and mind-wandering reports cannot be equated with reports of unawareness; the authors addressed the window issue by re-analyzing with 8–20 second windows, where the onset effect was more substantial in shorter windows (interaction slope t(148.3) = 2.48, p = .014) but the surprise effect had a near-zero slope (t(148.3) = 0.24, p = .81).
  • Stated/evident: the contextual encoding advantage for on-task periods was less robust across parameter choices (significant clusters in 4/7 comparisons), and the class imbalance was 72% on-task versus 28% mind-wandering, addressed by under-sampling on-task data; cluster-based permutation tests do not guarantee exact onset or offset significance.

Model-written summary. Check it against the paper before relying on it.

Integrity signals

Registry

Crossref · 10.1038/s44271-026-00518-4

Cites

80 of 81 references verified0 citation errors · 0 not checkable

Statistics

Figures

Image 1
CC-BY-NC-ND · From the PDF
Image 2
CC-BY-NC-ND · From the PDF
Image 3
CC-BY-NC-ND · From the PDF
Image 4
CC-BY-NC-ND · From the PDF

Verification

Claims the paper makes

0 Valid · 2 Not valid · 15 Not checked · 75 Own results and interpretations

48 more claims

Create a free account to see all 92 claims and every source verdict.

See all claims — free account
  • The quote only partly states this claim. 1
  • It may state two separate claims as one. 28
  • Probably not a substantial claim. 15

Create a free account to see these claims.

Statistics recomputed

Every reported test result, recomputed from its test statistic and degrees of freedom. A result shows as not adding up only once an independent check has confirmed it.

  • Not confirmedt(24) = 0.39reported p = .69 · recomputed p = .700

    “t(24) = 0.39, p = .69”

  • Not confirmedt(24) = 0.35reported p = .72 · recomputed p = .729

    “t(24) = 0.35, p = .72”

  • Consistentt(24) = -1.63reported p = .11 · recomputed p = .116

    “t(24) = -1.63,p = .11”

    The paper states an exact test

  • Consistentt(24) = 3.1reported p = .005 · recomputed p = .005

    “t(24) = 3.10, p = .005”

  • Consistentt(24) = 2.209reported p = .037 · recomputed p = .037

    “t(24) = 2.209, p = .037”

5 of 8 · the rest aren’t loaded

3 more recomputed tests

Create a free account to see every test the Stats check recomputed.

See all tests — free account

Self-check

Percentages, sums and reported differences re-computed from the paper’s own numbers.

Each contradiction between two passages listed here was confirmed by an independent check.

1 of 1

Citations

Cites

Show all 80 references

Research works printed but not linked: 3

  1. Ihde, D. Listening and Voice: Phenomenologies of Sound. (State University of New York Press, 2012).
  2. Puffay, C. et al. Large Language Models Reveal the Neural Tracking of Linguistic Context in Attended and Unattended Multi-Talker Speech. 2025.04.24.648897 Preprint at https://doi.org/10.1101/2025.04.24.648897 (2025). · doi:10.1101/2025.04.24.648897
  3. Makeig, S., Bell, A., Jung, T.-P. & Sejnowski, T. J. Independent component analysis of electroencephalographic data. Adv. Neural Inf. Process. Syst. 8, (1995).

Cited by

Still being checked.

Metadata

Authors
Chen, G. R., Finkelstein, R., Goldstein, A., Hassin, R. R., & Deouell, L. Y.
Affiliations
Hebrew University of Jerusalem (IL)
Corresponding author
Chen, Gal R.
Venue
Communications Psychology
Published
2026-08-15
ISSN
2731-9121 (ISSN-L)
OpenAlex
W7203546136
Cited by
0
References
80
FWCI
0.00
Type
journal article
Language
en
Open access
Licence
Locations
  • Communications Psychology (publishedVersion, cc-by-nc-nd)
  • bioRxiv (Cold Spring Harbor Laboratory) (submittedVersion, cc-by-nc-nd)
Topics
  • Life Sciences › Neuroscience › Cognitive Neuroscience › Mind wandering and attention
  • Life Sciences › Neuroscience › Cognitive Neuroscience › Neurobiology of Language and Bilingualism
  • Life Sciences › Neuroscience › Cognitive Neuroscience › Neuroscience and Music Perception
Keywords
mind wandering, word prediction, semantic encoding, conscious awareness, EEG, language model embeddings, audiobook listening, perceptual responses

About these checks

Registry, retraction, Cites and statistics results are computed, not judged, with one exception: a printed reference that no index lists is read once by a model before it counts as a citation error. Claims are extracted by a model and kept only when their quote appears verbatim in the text; each cited source is then checked, with a model’s help, for whether it says what the claim relies on. Those verdicts can be wrong — read the claim’s quote and the cited source before relying on one.

A language model drew the claims from the paper on 2026-11-19, under version 3 of the claim definition (model deepseek-v4.1-flash, prompt extract-v3).

A second model labelled each claim on 2026-11-19 (model jev-1.13-20260917, definition revision gold-guideline-v0.2).

Statistics and arithmetic are recomputed, not judged.

Confidence bands are shown only for fields calibrated on our annotated test set. None is calibrated yet, so no band is shown.

What counts as a claim ›How we measure confidence ›How we check this ›

Are you an author, or do you see a problem with this page? Request a review

How to cite

Cite CoChat for the checks on this page, and the paper for its own findings.

This page
CoChat. (2026). Citation check: High-level prediction of continuous speech during mind-wandering. https://app.cochat.ai/papers/10.1038/s44271-026-00518-4
The paper
Chen, G. R., Finkelstein, R., Goldstein, A., Hassin, R. R., & Deouell, L. Y. (2026). High-level prediction of continuous speech during mind-wandering. Communications Psychology. https://doi.org/10.1038/s44271-026-00518-4

Data sources

Bibliographic data: Semantic Scholar, OpenAlex, Crossref.

Contains information from the Semantic Scholar Academic Graph Datasets, made available under the ODC Attribution License (ODC-BY 1.0).

Semantic Scholar’s data is described in: Kinney, R., et al. (2023). The Semantic Scholar Open Data Platform. arXiv:2301.10140.