CoChat

Learning New Vocabulary Implicitly During Sleep Transfers With Cross-Modal Generalization Into Wakefulness

Open access
Koroma, Matthieu, Elbaz, Maxime, Léger, Damien, Kouider, Sid · 2022 · Frontiers in Neuroscience

The DOI matches its Crossref record; 43 of the 43 references it cites were found in a registry or an index; none of the 20 sources its claims cite is checked for support yet.

Abstract

New information can be learned during sleep but the extent to which we can access this knowledge after awakening is far less understood. Using a novel Associative Transfer Learning paradigm, we show that, after hearing unknown Japanese words with sounds referring to their meaning during sleep, participants could identify the images depicting the meaning of newly acquired Japanese words after awakening (N = 22). Moreover, we demonstrate that this cross-modal generalization is implicit, meaning that participants remain unaware of this knowledge. Using electroencephalography, we further show that frontal slow-wave responses to auditory stimuli during sleep predicted memory performance after awakening. This neural signature of memory formation gradually emerged over the course of the sleep phase, highlighting the dynamics of associative learning during sleep. This study provides novel evidence that the formation of new associative memories can be traced back to the dynamics of slow-wave responses to stimuli during sleep and that their implicit transfer into wakefulness can be generalized across sensory modalities.

Overview

CoChat Summary

TL;DR

Using an Associative Transfer Learning paradigm, participants heard Japanese words paired with meaning-related sounds during sleep and later identified corresponding images above chance after awakening, with confidence ratings indicating implicit memory. Frontal slow-wave EEG responses during NREM sleep predicted later memory and shifted from early to late slow-wave activity over the sleep learning phase.

Key findings

  • Memory for NREM-presented words was above chance (59%, CI 52–66; d = 0.58; t(21) = 2.7, p = 0.013), whereas REM (53%, p = 0.201) and control (45%, p = 0.204) lists did not differ from chance; repeated-measures ANOVA showed an arousal-level effect [F(2,21) = 4.66, p = 0.021], with NREM vs. control p = 0.011.
  • Learning was implicit: confidence scores did not differ between NREM and control lists (−0.03, CI −0.11–0.05, p = 0.444) or between correct and incorrect NREM items (0.00, CI −0.12–0.12, p = 0.959); in a wake control (n = 12), confidence was higher for learned vs. control and correct vs. incorrect items (Supplementary Figure 1).
  • EEG: correct vs. error trials differed after stimulation (3.36–3.93 s, d = −0.44, Pcluster = 0.029); for the 3rd slow wave, correct trials had lower amplitude than error trials (−3.22 μV, CI −6.3–−0.2, d = −0.46, t(21) = −2.2, p = 0.039).
  • Dynamics: in the first half, a correct-vs.-error cluster covered the 1st slow wave (0.59–1.48 s, Pcluster = 0.025); in the second half, late clusters emerged over the 3rd slow wave (3.16–3.56 s, Pcluster = 0.047; 3.67–4.05 s, Pcluster = 0.046), with interactions for the 1st SW [F(1,21) = 4.52, p = 0.046] and 2nd SW [F(1,21) = 4.45, p = 0.047].

Methods

  • ATL paradigm: 25 French-native speakers were recruited and 22 analyzed (age 23.7, range 20–29, 15 females); 48 Japanese words were assigned to NREM, REM, and control lists of 16. Awake familiarization used 48 sound-image pairs; during sleep, sounds alone were played when awake/NREM1 and Japanese translations were added during NREM/REM; after awakening, participants performed a 2AFC image-selection task plus confidence ratings (null/low/high). A separate wake control had n = 12.
  • EEG: 64-channel EEG and chin EMG were recorded; sleep was scored online and offline; analysis focused on a frontal electrode cluster and slow-wave frequency at 0.85 Hz, using cluster-based permutation statistics (1,000 permutations), t-tests, ANOVA, and generalized linear mixed models.

Limitations

  • Stated: REM vs. NREM comparison is limited because REM received about 3 times fewer presentations (13 vs. 40), preventing direct comparison of sleep learning across states; the study did not explain why some items or participants learned better, with individual preferences, language ability, memory capacity, and motivation suggested as possible factors.
  • Stated: the origin of cross-modal generalization remains unclear—whether it reflects cross-modal retrieval at awakening or cross-modal encoding during sleep—and whether ATL extends to REM sleep requires further experiments.
  • Evident: the analyzed sample is modest (22; wake control n = 12), and the paper notes explicit wake learning was much more efficient (4 vs. 40 repetitions; 88% vs. 59% accuracy), so the sleep-learning effect is small and implicit.

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

Integrity signals

Registry

Crossref · 10.3389/fnins.2022.801666

Cites

43 of 43 references verified0 citation errors · 0 not checkable

Statistics

Figures

FIGURE 1 Associative Transfer Learning paradigm. (A) During the first phase performed awake, participants were familiarized with the 48 sound-image pairs. Participants were then allowed to fall asleep. Whenever participants entered NREM or REM sleep, sounds from the corresponding list were played simultaneously with their corresponding Japanese translations. Upon awakening, participants performed a memory test. Japanese words were played with two images belonging to the same list (i.e., NREM, REM, or control list). Participants first chose which image correctly depicts the Japanese word (left or right), and then provided their confidence on a three-level scale (null, low, high). (B) For each participant, the 48 Japanese words (e.g., the Japanese equivalent for “dog”) were randomly assigned into 3 lists, resulting in a NREM, a REM and a control list of 16 items each. (C) Sleep scoring (hypnogram) of the experiment for one participant.
FIGURE 1 Associative Transfer Learning paradigm. (A) During the first phase performed awake, participants were familiarized with the 48 sound-image pairs. Participants were then allowed to fall asleep. Whenever participants entered NREM or REM sleep, sounds from the corresponding list were played simultaneously with their corresponding Japanese translations. Upon awakening, participants performed a memory test. Japanese words were played with two images belonging to the same list (i.e., NREM, REM, or control list). Participants first chose which image correctly depicts the Japanese word (left or right), and then provided their confidence on a three-level scale (null, low, high). (B) For each participant, the 48 Japanese words (e.g., the Japanese equivalent for “dog”) were randomly assigned into 3 lists, resulting in a NREM, a REM and a control list of 16 items each. (C) Sleep scoring (hypnogram) of the experiment for one participant.CC BY · PubMed Central
FIGURE 2 Implicit associative transfer learning for words presented during NREM sleep. (A) Memory performance across NREM, REM and control list. Student’s T-test against chance level (50%, dotted line) were computed for each list and corrected for multiple comparisons (*P < 0.05). (B) Confidence score on a 3-level scale (null, low, high) for memory response of the NREM list. Statistical tests reveal no difference between trials that were successfully identified (correct) and those associated with a mistake (error).
FIGURE 2 Implicit associative transfer learning for words presented during NREM sleep. (A) Memory performance across NREM, REM and control list. Student’s T-test against chance level (50%, dotted line) were computed for each list and corrected for multiple comparisons (*P < 0.05). (B) Confidence score on a 3-level scale (null, low, high) for memory response of the NREM list. Statistical tests reveal no difference between trials that were successfully identified (correct) and those associated with a mistake (error).CC BY · PubMed Central
FIGURE 3 Differential entrainment of frontal neural responses at a slow-wave frequency to stimulation for correctly and incorrectly identified items during NREM sleep. (A) Neural responses were computed over a frontal cluster of electrodes depicted in the lower left panel. Time-course of the stimulus presentation was indicated in a top panel. Time-courses of the brain responses (lower panel) were averaged across participants and smoothed for visualization purposes only, using a 500-ms wide Gaussian kernel. Mean and standard error of the mean (SEM) are represented, respectively, with solid lines and shaded areas for the correct (green) and the error (red) trials. Statistical tests were performed on brain responses before smoothing. Green, red and black horizontal lines denote significant clusters of neural responses differing from baseline (0, dotted line) across participants for, respectively, correct trials, error trials, or the difference between both conditions (P < 0.05 after cluster correction). Gray bars indicate time clusters corresponding to negative half-periods of a slow wave rhythm at 0.85 Hz. (B) Neural responses were averaged over each negative half-period of slow waves and compared between correct (green) and error (red) trials. Mean and standard error of the mean (SEM) are represented, respectively, with bar plots and solid lines. Student’s T-test against baseline (0, dotted line) was computed for correct and error trials and corrected for multiple comparisons. Paired Student’s T-test were computed between correct and error trials (***P < 0.001, **P < 0.01, *P < 0.05). Topography of amplitudes were indicated in lower panels for correct, error and the difference between correct and error trials.
FIGURE 3 Differential entrainment of frontal neural responses at a slow-wave frequency to stimulation for correctly and incorrectly identified items during NREM sleep. (A) Neural responses were computed over a frontal cluster of electrodes depicted in the lower left panel. Time-course of the stimulus presentation was indicated in a top panel. Time-courses of the brain responses (lower panel) were averaged across participants and smoothed for visualization purposes only, using a 500-ms wide Gaussian kernel. Mean and standard error of the mean (SEM) are represented, respectively, with solid lines and shaded areas for the correct (green) and the error (red) trials. Statistical tests were performed on brain responses before smoothing. Green, red and black horizontal lines denote significant clusters of neural responses differing from baseline (0, dotted line) across participants for, respectively, correct trials, error trials, or the difference between both conditions (P < 0.05 after cluster correction). Gray bars indicate time clusters corresponding to negative half-periods of a slow wave rhythm at 0.85 Hz. (B) Neural responses were averaged over each negative half-period of slow waves and compared between correct (green) and error (red) trials. Mean and standard error of the mean (SEM) are represented, respectively, with bar plots and solid lines. Student’s T-test against baseline (0, dotted line) was computed for correct and error trials and corrected for multiple comparisons. Paired Student’s T-test were computed between correct and error trials (***P < 0.001, **P < 0.01, *P < 0.05). Topography of amplitudes were indicated in lower panels for correct, error and the difference between correct and error trials.CC BY · PubMed Central
FIGURE 4 Differential entrainment of frontal neural responses to stimulation for correctly and incorrectly identified items for the first half and the second half of NREM trials. (A,B) Mean and standard error of the mean (SEM) are represented, respectively, with solid lines and shaded areas for the correct (green) and the error (red) trials for the first (A) and second (B) half trials in NREM sleep. Green, red and black horizontal lines denote significant clusters of neural responses differing from baseline (0, dotted line) for, respectively, correct trials, error trials, or the difference between both conditions (P < 0.05 after cluster correction). Gray bars indicate time clusters spanning over the three-fourth of a slow wave rhythm at 0.85 Hz centered over its trough. (C–E) Neural responses for correct (green) and error (red) trials for the first and second half of trials were averaged and compared for the first (C), second (D) and third (E) slow wave. Mean and standard error of the mean (SEM) are represented, respectively, with bar plots and solid lines. Student’s T-test against baseline (0, dotted line) was computed for correct and error trials and corrected for multiple comparisons. Paired Student’s T-test were computed between correct and error trials for and corrected for multiple comparisons. Interaction between trial type (correct vs. error) and period of the night (first vs. second half) for each slow wave was computed with repeated measures ANOVA (***P < 0.001, **P < 0.01, *P < 0.05).
FIGURE 4 Differential entrainment of frontal neural responses to stimulation for correctly and incorrectly identified items for the first half and the second half of NREM trials. (A,B) Mean and standard error of the mean (SEM) are represented, respectively, with solid lines and shaded areas for the correct (green) and the error (red) trials for the first (A) and second (B) half trials in NREM sleep. Green, red and black horizontal lines denote significant clusters of neural responses differing from baseline (0, dotted line) for, respectively, correct trials, error trials, or the difference between both conditions (P < 0.05 after cluster correction). Gray bars indicate time clusters spanning over the three-fourth of a slow wave rhythm at 0.85 Hz centered over its trough. (C–E) Neural responses for correct (green) and error (red) trials for the first and second half of trials were averaged and compared for the first (C), second (D) and third (E) slow wave. Mean and standard error of the mean (SEM) are represented, respectively, with bar plots and solid lines. Student’s T-test against baseline (0, dotted line) was computed for correct and error trials and corrected for multiple comparisons. Paired Student’s T-test were computed between correct and error trials for and corrected for multiple comparisons. Interaction between trial type (correct vs. error) and period of the night (first vs. second half) for each slow wave was computed with repeated measures ANOVA (***P < 0.001, **P < 0.01, *P < 0.05).CC BY · PubMed Central

Verification

Claims the paper makes

0 Valid · 0 Not valid · 13 Not checked · 39 Own results and interpretations

35 more claims

Create a free account to see all 52 claims and to verify the 14 references still open.

See all claims — free account
  • It may state two separate claims as one. 8
  • Probably not a substantial claim. 4

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.

  • ConsistentF(2, 21) = 4.66reported p = .021 · recomputed p = .021

    “F(2,21) = 4.66, p = .021”

  • Consistentt(21) = 2.7reported p = .013 · recomputed p = .013

    “t(21) = 2.7, p = .013”

  • Consistentt(21) = 1.32reported p = .201 · recomputed p = .201

    “t(21) = 1.32, p = .201”

  • Consistentt(21) = -1.31reported p = .204 · recomputed p = .204

    “t(21) = −1.31, p = .204”

  • ConsistentF(2, 1053) = 6.31reported p = .002 · recomputed p = .002

    “F(2,1053) = 6.31, p = .002”

5 of 19 · the rest aren’t loaded

14 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.

2 of 2

Citations

Cites

Show all 43 references

Cited by

See all 25

Metadata

Authors
Koroma, M., Elbaz, M., Léger, D., & Kouider, S.
Affiliations
  • Centre National de la Recherche Scientifique (FR)
  • Université Paris Cité (FR)
Corresponding author
Koroma, Matthieu
Venue
Frontiers in Neuroscience
Published
2022-03-09
Volume
16
Pages
801666-801666
ISSN
1662-453X (ISSN-L)
OpenAlex
W4220832644
Cited by
24
References
43
Citations by year
2026: 4 · 2025: 4 · 2024: 3 · 2023: 9 · 2022: 4
FWCI
2.05
Type
journal article
Language
en
Open access
Locations
  • Frontiers in Neuroscience (publishedVersion, cc-by)
  • PubMed (publishedVersion)
  • PubMed Central (publishedVersion, cc-by)
  • HAL (submittedVersion, other-oa)
  • ORBi (University of Liège) (submittedVersion, other-oa)
  • DOAJ (DOAJ: Directory of Open Access Journals) (submittedVersion, cc-by-sa)
Funders
  • European Research Council
  • Agence Nationale de la Recherche (ANR-10-LABX-0087)
  • Direction Générale de l’Armement
  • Canadian Institute for Advanced Research
Topics
  • Life Sciences › Neuroscience › Cognitive Neuroscience › Sleep and Wakefulness Research
  • Life Sciences › Neuroscience › Cognitive Neuroscience › Memory and Neural Mechanisms
  • Life Sciences › Neuroscience › Cognitive Neuroscience › Memory Processes and Influences
Keywords
sleep-dependent memory consolidation, implicit learning, cross-modal generalization, associative memory

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: Learning New Vocabulary Implicitly During Sleep Transfers With Cross-Modal Generalization Into Wakefulness. https://app.cochat.ai/papers/10.3389/fnins.2022.801666
The paper
Koroma, M., Elbaz, M., Léger, D., & Kouider, S. (2022). Learning New Vocabulary Implicitly During Sleep Transfers With Cross-Modal Generalization Into Wakefulness. Frontiers in Neuroscience. https://doi.org/10.3389/fnins.2022.801666

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.