Skip to main content
NCBI home page
NIHPA Author Manuscripts logoLink to NIHPA Author Manuscripts
. Author manuscript; available in PMC: 2026 Sep 27.
Published in final edited form as: J Neurophysiol. 2026 Aug 18;136(3):1075–1085. doi: 10.1152/jn.00628.2025

Functional Identification of Language-Responsive Sensors in Individual Participants in MEG Investigations

Mathias Huybrechts 1,*, Rose Bruffaerts 1,2,3,*, Alvincé Pongos 3,4, Cory Shain 3,5, Benjamin Lipkin 3, Matthew Siegelman 3,6, Vincent Wens 7,8, Martin Sjøgård 9,10, Idan Blank 3,11, Serge Goldman 7,8, Xavier De Tiège 7,8, Evelina Fedorenko 3,12
PMCID: PMC13615564  NIHMSID: NIHMS2205370  PMID: 42611035

Abstract

Making meaningful inferences about the functional architecture of the language system requires the ability to refer to the same neural units across individuals and studies. Traditional brain imaging approaches align and average brains together in a common space. However, lateral frontal and temporal cortices, where the language system resides, are characterized by high structural and functional inter-individual variability, which reduces the sensitivity and functional resolution of group-averaging analyses. This issue is compounded by the fact that language areas lay in close proximity to regions of other large-scale networks with different functional profiles. A solution inspired by vision neuroscience is to identify language areas functionally in each individual brain using a ‘localizer’ task (e.g., a language comprehension task). This approach has proven productive in fMRI, yielding a number of robust and replicable findings about the language system. Here, we extend this approach to MEG. Across two experiments (one in Dutch speakers, n=19; one in English speakers, n=23), we examined neural responses to the processing of sentences and a control condition (nonword sequences). We demonstrate that the sensor and source topography of neural responses to language is spatially stable within individuals but varies across individuals. Consequently, analyses that take this inter-individual variability into account are characterized by greater sensitivity, compared to the group-level analyses. In summary, similar to fMRI, functional identification within individuals yields benefits in MEG, thus opening the door to future investigations of language processing including questions where whole-brain coverage and temporal resolution are both critical.

New & Noteworthy

Language areas vary across individuals, challenging traditional group-averaged brain imaging approaches. Using an fMRI-based strategy, this study validates that functionally localizing language responses within individuals improves sensitivity in magnetoencephalography (MEG). Across Dutch and English speakers, language-related sensor and source topographies were stable within individuals but varied between individuals. Individual-level functional identification thus enhances MEG analyses, enabling precise investigations of language processing with both high temporal resolution and whole-brain coverage.

Graphical Abstract

graphic file with name nihms-2205370-f0001.webp

Introduction

The functional architecture of the human language network is broadly consistent across individuals (1). However, the precise topography of this network varies substantially even within homogenous groups of neurotypical adults (2). Developing research methods that take these inter-individual differences into account by identifying functional areas in individual participants – an approach known as ‘functional localization’ – has proven vital in cognitive neuroscience across domains and has been shown to increase sensitivity, functional resolution, accurate effect size estimation, and interpretability (3–6).

Numerous studies over the last decade have provided evidence that the language network can be delineated in a robust and replicable way at the individual-subject level using a contrast between language comprehension and a perceptually matched control condition in fMRI (2, 7–11) (see (1) for data in > 800 individuals). Importantly, the language localizer contrast (language > perceptually matched control condition) is robust to input modality (written, spoken, or signed) (2, 12, 13, 10), stimulus content (hand-crafted sentences, sentences extracted from a corpus, or connected passages (12)), language (14), and the presence or absence of an active task (2, 15–17). Furthermore, this network has been shown to be strongly selective for linguistic input (18–20) (see e.g., (9) for a review). This broad generalization across paradigm variations and selectivity for language processing jointly suggest that the language brain areas support specifically linguistic computations. Indeed, evidence from dozens of studies has implicated these areas in lexical access, syntactic structure building, and semantic composition during both comprehension and production (21–24).

Although fMRI investigations of the language network have yielded important findings, fMRI’s poor temporal resolution limits its use for research questions where timing information is critical. Intracranial recordings provide high-spatial and high-temporal resolution data with high signal-to-noise ratio, but the coverage is sparse and idiosyncratic across individuals. In contrast, magnetoencephalography (MEG) enables whole-brain non-invasive measurements of neural activity at millisecond-level resolution. MEG studies have consistently identified activity within language-specific regions that broadly overlap with the network characterized in lesion studies of patients with aphasia and fMRI studies of healthy individuals (25–28). However, MEG’s spatial lateralization profiles show greater variability than fMRI, with some studies reporting more bilateral or right-hemispheric patterns (29–32), presumably reflecting MEG’s sensitivity to transient, synchronized neural activity and susceptibility to individual anatomical variation. Despite this variability, MEG has proven highly valuable for mapping lexical, syntactic, and compositional processes within the language network, revealing dynamic interplay between the temporal and frontal regions during sentence-level comprehension (33, 34).

These past MEG studies have relied on group-level source analyses that average responses across participants to characterize common loci of activation, providing a coherent yet population-averaged view of the network’s spatiotemporal dynamics. This group-averaging approach does not take into account the well-established inter-individual variability in functional topography (3–6, 2) and may therefore underestimate effect sizes or conflate nearby distinct networks.

In line with the shift toward individual-subject analyses in fMRI (5, 35), recent MEG research has begun to emphasize individual functional localization as a means to increase sensitivity, yield more accurate estimates of language-related activity, and improve correspondence between studies (36, 28, 37, 38).

However, to our knowledge, the functional localization approach has not been systematically evaluated for the domain of language in MEG investigations. Here, we attempt to fill this gap. Using data from two independent groups (across two languages: English and Dutch), we establish the feasibility of identifying language-responsive sensors and sources at the individual level in MEG investigations.

Methods

Participants

We recruited 42 healthy young volunteers: 19 self-reported native Dutch speakers (18 female, between 19 and 29 years old, mean 23.4 years old) and 23 self-reported native English speakers (between 19 and 53 years old, mean 26.7 years old). The study was approved by the local Ethics Committees, being the CUB-Hôpital Erasme Ethics Committee for the Dutch-speaking volunteers and MIT’s Committee on the Use of Humans as Experimental Subjects (COUHES) for the English-speaking volunteers. All participants provided written informed consent in accordance with the Declaration of Helsinki.

Functional Localizer Stimulus Materials

To identify language responses we use a design that has been extensively validated in fMRI (2), namely, a sentence reading task. Participants performed the experiment in their native language and original materials can be downloaded from https://osf.io/vc2bw/. For the sentence condition, 80 12-word-long sentences were constructed in English using a variety of syntactic structures and covering a wide range of topics. The sentences were translated into Dutch, with minimal changes to obtain 12-word-long sentences.

For the nonword condition, care was taken to minimize low-level differences in the phonological make-up of the stimuli compared to the sentence condition. In this study, the term ‘nonword’ follows the convention of the canonical sentences > nonwords fMRI localizer (2) and refers to pronounceable, pseudoword-like strings. More specifically, in English, the content words (nouns, verbs, adjectives, adverbs) of the sentence condition were syllabified to create a set of syllables that could be recombined in new ways to form pronounceable but meaningless strings. For syllables that formed real English words, a single phoneme was replaced (respecting the phonotactic constraints of English) to turn the syllable into a nonword. The syllables were then recombined to create nonwords matched for length (in syllables) with the sentence condition. In Dutch, nonwords were derived from the content words in the sentence condition by means of the pseudoword generator Wuggy (https://github.com/crr-ugent/wuggy). This program generates nonwords that match the original word in subsyllabic structure and respects the transition frequencies specific to Dutch (39). To verify that the relationship between words and nonwords was comparable across the two language groups, we performed a quantitative linguistic analysis of the stimulus materials. For both Dutch and English, words and nonwords were compared across a range of features including word length, syllable count, bigram and trigram frequencies, vowel ratio, and visual complexity. The stimulus sets showed highly similar results across these measures, indicating close comparability at the orthographic and sublexical level. Full results are reported in Supplemental Materials 1.

It is worth noting that these nonwords engage orthographic processing mechanisms that are also involved in processing words. This means that the sentence versus nonword contrast isolates higher-order combinatorial and lexical-semantic processes rather than outlining every cortical process involved in reading such as visual aspects. Accordingly, throughout this paper, ‘language-responsive’ refers specifically to sensors selective for higher-order language processing.

Functional Localizer Task

The words/nonwords were presented one at a time in a rapid serial visual presentation paradigm at a fixed rate per word/nonword (385 ms per (non)word in Dutch, 400 ms in English). Each word/nonword was presented in the center of the screen in capital letters without punctuation and appeared immediately after the previous word/nonword without a blank transient screen. At the end of each trial, participants in the Dutch group were presented with a memory probe (i.e. a single word or nonword), and asked to indicate via button press whether it had appeared in the preceding sentence. Half of all trials required a positive response. Participants in the English group instead saw a button-press icon and were asked simply to press a button, without a memory judgement. This difference between the Dutch and English version was introduced to reduce the duration of the English experiment by several minutes, in order to combine the paradigm with other, unrelated studies. As noted above, previous fMRI work demonstrated that the sentences>nonwords localizer contrast is robust to different tasks including passive reading/listening (2, 15, 17). Following the memory probe or button-press icon, a variable inter-trial interval, between 0.5s and 2.25s was inserted, during which participants could respond at any point after the probe or button-press icon appeared and until the next trial started. The experiment lasted ~20 minutes. Five participants in the English group were recorded using a similar paradigm but with a different set of materials (8-word-long sentences and 8-nonword-long sequences). Previous fMRI work has shown that the sentences>nonwords localizer contrast is robust to such variation (2).

MEG Data Acquisition

Continuous MEG data were recorded using a whole-head 306 channel (102 magnetometers, 204 planar gradiometers) TRIUX system (MEGIN, Espoo, Finland) either at the HUB - Hôpital Erasme (Brussels, Belgium) or at the Martinos Imaging Center (McGovern Institute for Brain Research at MIT, Cambridge, MA, USA). Participants were tested in an upright seated position (68° recline). Four head-position indicator (HPI) coils were used to record the head position within the MEG helmet every 200 ms. The participant’s head shape was digitally recorded by means of a 3D digitizer (Fastrak Polhemus, Inc., Colchester, VA, USA) along with the position of the HPI coils and fiducial points (nasion, left and right periauricular). To minimize inter-individual differences in head position within the dewar, a standardized seating and positioning protocol was followed across all participants. The MEG chair was adjusted so that the top of the head was in contact with the helmet, participants were seated upright with their back against the chair, and head position was verified to be consistent before recording commenced. MEG signals were recorded at a sampling rate of 1000 Hz with on-line band filter between 0.1 and 330 Hz.

MRI Data Acquisition

Individual T1-weighted structural MRI scans were only available for the Dutch-speaking cohort. Images were acquired on a 3T SIGNA™ PET/MR scanner (GE Healthcare, Chicago, IL, USA) at HUB - Hôpital Erasme using a 3D T1 BRAVO sequence (repetition time: 8.328ms, echo time: 3.104ms, inversion time: 450ms, flip angle: 12°, acquisition matrix: 240×240, reconstructed matrix: 256×256, 1mm isotropic resolution). Cortical surface reconstruction was performed using FreeSurfer software (40, 41), providing the individual cortical surface models used for forward modelling and source space reconstruction.

MEG Data Processing

Initial preprocessing of the raw MEG data used MaxFilter version 2.2 (MEGIN, Espoo, Finland): temporal signal space separation (42) was applied to remove noise from external sources and from HPI coils for continuous head-motion correction (correlation threshold: 0.98, 10 s sliding window), and to virtually transform data to a standardized head position. MaxFilter was used to automatically detect and virtually reconstruct noisy sensors. Further preprocessing was performed using MNE-Python version 1.8.0 (43). Preprocessing consisted of high-pass filtering at 0.1 Hz, low-pass filtering at 300 Hz and notch filtering at 50Hz or 60Hz (with respective higher harmonics) depending on the line frequency. Data were downsampled to 500 Hz and epoched at the level of the (non)word (200ms before stimulus onset until 400 ms after stimulus onset). Visual inspection of all epochs was performed, and epochs with clear artefacts were marked as bad and excluded from further analysis. Denoising was performed by visually identifying and removing ICA components consistent with eye blinks, eye movements, SQUID jumps and cardiac artefacts from a total of 60 components extracted using fastICA (44) with default parameters on the concatenated epochs, excluding bad epochs.

A quantitative marker of noise was derived per participant by first averaging the pre-trial baseline intervals (neutral interval of 500ms prior to trial onset when only the fixation cross is presented) across all sensors and then calculating the standard deviation of these values across all trials. Participants with markers of noise greater than two standard deviations from the average noise marker across participants recorded in the same MEG device were removed from further analysis. This procedure led to the removal of one participant from the English group. Two additional participants were removed from the English group: one due to excessive motion and one due to the presence of high-amplitude “mu rhythm” (45) apparent upon visual inspection.

For source reconstruction of the Dutch participants, we applied the same general pipeline for each participant individually. First, a single-shell boundary element model (BEM) was constructed from each participant’s T1-weighted MRI based on an inner skull surface derived using the watershed algorithm with a conductivity of 0.3S/m. A cortical surface-based source space was then defined on the white matter surface using an octahedral subdivision (oct6), resulting in approximately 4098 source locations per hemisphere. MEG and MRI coordinate systems were co-registered by aligning digitized head shape points and anatomical fiducials (nasion, left and right periauricular points) to each participant’s structural MRI using manual refinement of the initial coregistration. Forward solutions were computed by combining the source space, BEM model, and the coregistration transformation. The noise covariance matrix was estimated from the pre-stimulus baseline period (−500 to 0 ms relative to trial onset) and the rank was estimated using both the sensor information (rank from info) and empirical rank from the data (tolerance = 1e-3), with the minimum of these two values used in the inverse operator to account for rank deficiencies introduced by MaxFilter. The forward models were inverted using minimum norm estimation using a loose orientation constraint (0.2) and depth weighting (0.8). Source estimates were then derived for individual word-level epochs using the same epoching scheme as in the sensor space. Dynamic statistical parametric mapping (dSPM) was applied to obtain noise-normalized estimates of cortical activity over time (40).

Functional identification and validation of language-responsive sensors and sources

To test whether MEG allows for a robust identification of language-responsive sensors and sources at the individual level, we evaluated the stability of the language responses within participants across time (i.e., between the first half and the last half of trials). Sentence and nonword epochs were first equalized in number and then independently split into two halves based on trial order (first and last half). Language responses were then identified separately in each half through contrasting the sentence condition with the nonword control condition, similarly to fMRI. At the sensor level, differential evoked responses between conditions were quantified using spatiotemporal F-values derived from a one-way ANOVA (F-test) computed at each gradiometer sensor and time point across conditions. At the source level, F-values were computed analogously at each source location and time point.

The F-statistics derived from the one-way ANOVA test served as the basis for further analyses of within- versus between-participant stability. First, we defined a language-sensitive time window spanning 100–350 ms after word/nonword onset, motivated by reliable condition-specific differences in event related fields (ERF) and an additional spatiotemporal permutation cluster analysis reported in Supplemental Materials 2, which revealed significant clusters falling within this time window across participants. Then we computed Spearman correlations between time-averaged F-values from the first and last halves of trials within each participant, yielding a measure of within-participant splithalf reliability. To obtain a between-participant measure, we computed Spearman correlations between the first-half F-values of each participant and the last-half F-values of all other participants. In sensor space, this was done directly due to shared sensor layout. In source space, between-participant measures were obtained after morphing individual participant’s F-maps to fsaverage space to ensure vertex-wise correspondence. The resulting within- and between-participant correlation distributions were then compared using a Wilcoxon rank-sum test. This analysis tests whether the sensor and source distributions of the language response is more consistent within individuals than across individuals.

Second, in the same [100ms, 350ms] time window, we used the data from the first half of the trials to define sensors or sources of interest (SOIs) in each participant individually. SOIs were defined as the 10% of gradiometer sensors in sensor space and the 10% of sources in source space with the highest F-values, following fMRI-based approaches (1, 2). To test whether identification of language-responsive sensors and sources at the individual level is superior to identifying language-responsive sensors and sources at the group level, we then computed group-level SOIs as well. In sensor space, these were defined as the top 10% of gradiometer sensors with the highest F-values of the averaged spatiotemporal F-maps over all participants. Critically, unlike in the individual-level analyses, the SOIs in the group analysis are the same for all participants. In source space, group-level SOIs were defined by first morphing each participant’s F-map to fsaverage space, averaging the resulting maps across participants in fsaverage space, and selecting the top 10% of sources with the highest F-values. This group-level source mask was then projected back into each participant’s native source space to obtain subject-specific group-level SOIs. To compare the effect sizes in SOIs defined individually vs. based on a group-level map, we used a paired Student’s t-test or Wilcoxon signed-rank test, depending on whether normality assumptions were met. In both cases (individual-level or group-level), SOIs were defined using only the first half of the trials, and their summary F-statistics were then evaluated in the last half of the trials for each participant.

Results

Grand average event-related fields

For comparison to prior work, we report the grand average event-related fields (ERFs) for each condition (Figure 1A-B). In particular, for the sentence condition, the timecourses were averaged across all the word positions in the sentence, and for the nonword condition across all the nonword positions, to derive a single timecourse (spanning the duration of a single word/nonword) per participant per condition. For both conditions, these timecourses were subsequently averaged across participants yielding the grand average ERFs. At the group level, sentence and nonword conditions differed significantly in both the Dutch and English groups within the 100–350 ms time window, as assessed using paired, time-resolved cluster-based permutation tests on gradiometer Global Field Power (GFP; p<0.05).

Figure 1:

Figure 1:

Open in a new tab

Grand-average gradiometer responses (Global Field Power, GFP) for the sentence and nonword conditions in the A) Dutch and B) English groups. Traces show the mean GFP across participants with a 95% confidence interval, reflecting the overall magnitude of the evoked response. Grey shading indicates time windows in which sentence and nonword conditions differed significantly at the group level, as determined by a paired, time-resolved cluster-based permutation test on GFP (p<0.05).

The topography of neural responses to language is stable within but varies between individuals

After inspecting the average time-courses, we next asked our key question of stability and inter-individual variability of the neural responses. In the Dutch group, the average correlation of the F-values for the sentence versus nonwords contrast across all sensors within each participant across the first and last half of the trials was 0.47 (s.d. 0.16). The mean correlation between pairs of participants (mean rho: 0.20, s.d. 0.13) was significantly lower (p<0.0001) compared to the within-participant correlation as indicated by the Wilcoxon rank-sum test (Figure 2A). In source space, the same pattern was observed: the average within-participant correlation was 0.14 (s.d. 0.12), whereas the average between-participant correlation was 0.04 (s.d. 0.10), with the within-participant correlations again exceeding the between-participant correlations (p=0.004; Figure 2B). This finding was extended to the English group, where in sensor space the average within-participant correlation (0.26 s.d. 0.19) was also significantly higher (p<0.0001) than the average between-participant correlation (0.07 s.d. 0.13; Figure 2C).

Figure 2:

Figure 2:

Open in a new tab

Distribution of Spearman correlation scores of F-values for the sentences versus nonwords contrast across sensors and sources, calculated within individuals (first vs last half of the data; darker shades) and between different individuals (lighter shades) in the A-B) Dutch sensor and source space and C) English sensor space. Relative bin heights are shown after normalizing each histogram to its maximum value to facilitate visual comparison between within- and between-participant distributions, which differ in sample size. Vertical striped lines indicate the mean of each distribution.

Visual inspection confirmed the similarity between the topographies of the F-values in the first and last half of the trials at the individual level in both sensor and source space (Figure 3). When assessing the topography of the language-responsive sensors of interest that were selected in different participants, inter-individual differences were notable: although some sensors were consistently selected in a substantial fraction of participants, other sensors were only selected in a small fraction of individuals. In the Dutch group, certain sensors were consistently selected in up to 68%, or 13 of the 19 participants. Similarly, in the English group, some sensors were consistently selected in up to 55%, or 11 of the 20 participants.

Figure 3:

Figure 3:

Open in a new tab

Topographies of the F-values in the first and last half of the trials in sample participants and group averages of the Dutch and English groups with SOIs indicated by white dots in sensor space and black surface outline in source space.

Analyses based on individual-level sensors and sources of interest yield more sensitive responses compared to group-averaging analyses

Given that individual-level SOIs are defined to capture each participant’s strongest language-responsive sensors and sources, analyses based on these SOIs are expected to benefit from a greater sensitivity than group-averaging analyses. We examined whether this expected directionality is preserved when SOIs are evaluated on held-out data. Using the sentences versus nonwords contrast, we defined two sets of SOIs – individual-level SOIs and group-level SOIs – on one half of the data (i.e. first half) and evaluated their responses on the held-out data (i.e. last half).

In the Dutch group sensor space, the averaged or summary F-statistic across SOIs, evaluated in the held-out data, was significantly smaller for group-level SOIs than for individual-level SOIs (p<0.001, Figure 4A). At the level of individual participants, 17 of the 19 participants showed higher summary F-statistics when SOIs were defined individually, with an average increase of 0.35. Similarly, in source space, higher summary F-statistics were observed for individually defined SOIs compared to group-level SOIs (p=0.036, Figure 4B). A similar finding was observed in the English group’s sensor space analysis (p=0.032, Figure 4C), where on average the F-values increased by 0.30. At the individual-participant level, 13 out of 20 showed a numerically larger summary statistic when SOIs were selected at the individual level. In all cases presented here, SOIs were defined as the top 10% of sensors or sources with highest F-values. Confirmatory analyses using more conservative and lenient cutoffs (5% and 15%) yielded the same overall pattern of results (Supplemental Materials 3).

Figure 4:

Figure 4:

Open in a new tab

Summary F-statistics when the SOIs are defined at the individual level vs. at the group level in the A-B) Dutch sensor and source domain and C) English sensor domain. In all cases, SOIs are defined using only the first half of the trials and the response magnitudes are examined in the last half of the data.

Discussion

In two independent datasets, MEG recordings allowed the identification of language-responsive sensors at the individual-participant level. During sentence reading, robust within-vs-between participant effects were detected in native Dutch speakers and reproduced in native English speakers. We showed that in both groups, the response to this language comprehension task was spatially consistent over time within individuals: similar sensors and sources showed strong responses during language processing in both halves of the data. Yet importantly, the response to language was spatially variable across individuals, likely reflecting differences in functional neuroanatomy. This variability means that analyses that take into account these inter-individual topographic differences can achieve greater sensitivity than traditional group-averaging approaches. Overall, our findings indicate that the functional identification approach, where sensors and sources of interest are defined in individual participants, can provide advantages in MEG, including improved sensitivity, functional resolution, and interpretability.

Limitations of conventional group-averaging approaches

In group-based fMRI and MEG at the source-level, when using a typical preprocessing and analysis pipeline, voxel- and vertex-wise neural responses from each participant are warped into a common space based on a brain template and functional correspondence is assumed in each voxel or vertex. This assumption has long been shown to be flawed, especially when examining cognitive functions supported by the association cortex (4, 8, 46–48). In sensor-level analyses of brain recording approaches, like fNIRS and MEG at the sensor-level, where a fixed number of sensors are used, a similar assumption is typically made when data are pooled across participants: that the same sensor is functionally equivalent across individuals. However, because of inter-individual neuroanatomical differences in brain volume and folding patterns, and other physical characteristics like hair type and skull shape, size and thickness, the ‘same’ sensor may capture different underlying sources of neural signals. This variability in the sources captured by the same sensor across participants introduces noise when signals are averaged across participants and complicates interpretation, potentially leading to incorrect conclusions (49, 50).

Advantages of individual-level functional localization

Functional identification in individual participants offers a practical way to address these limitations. In the present study, we established the feasibility of individual-participant functional identification in MEG using an extensively validated language localizer paradigm and demonstrated the topographic stability of the language-responsive sensors and sources within individuals over time – the critical foundation of individual-level functional localization.

Applying this approach provides advantages in statistical sensitivity and functional resolution. For instance, grouping these sensors and sources of interest into a single set allows for one statistical test across the ensemble, reducing the need for multiple comparisons (4, 34). Treating this ensemble as a multivariate pattern in multivariate pattern analysis (MVPA) could then powerfully enable investigations of fine-grained meaning and structure representations within the language network. For example, with respect to semantic knowledge, explicit quantitative models reflecting single-word and contextualized semantic knowledge can be used to test hypotheses about how the brain processes meanings extracted from linguistic input (51). Focusing on language-responsive sensors and sources can increase both the sensitivity of such an analysis and help interpret the observed effects.

In line with prior work using the same language localizer task within fMRI (1), intracranial recordings (52), and fNIRS (50), our findings suggest that taking into account inter-individual topographic differences increases sensitivity: language responses are systematically stronger in the vast majority of participants when sensors or sources are selected individually rather than at the group level. This suggests that group-level analyses underestimate effect sizes, likely due to the inclusion of responses from non-task-responsive sensors or sources in some participants. Such gains in sensitivity are especially helpful for detecting subtle effects in new or critical conditions and may be particularly valuable when studying patients with neurological disease, where neural responses may be overall weaker.

Finally, the suggested approach improves interpretability. When a validated localizer is used to define the relevant functional subsets of the brain across individuals, studies can make stronger inferences about the origins of an effect (e.g. the ability to interpret some critical effect as arising within the language-responsive regions). Using the same paradigm across individuals, studies, imaging modalities, and species (in cases of shared cognitive capacities such as face processing), increases confidence that studies refer to the same system, supporting knowledge accumulation and comparability across labs (53).

In addition to these benefits, functional localization offers a built-in safeguard against false positives. By using one half of the data to identify sensors or sources of interest and the other half to quantify response magnitudes, the approach incorporates an internal replication component.

Using MEG to probe linguistic computations

In well-characterized systems such as the language network or the face recognition system, for which certain localizer paradigms have been shown (in fMRI research) to reliably identify the relevant underlying functional neuroanatomy, MEG offers a powerful complement by capturing neural activity at millisecond timescales. This allows to study the detailed time-course of information processing within these regions as dynamic language processing unfolds.

Our proposed method provides a means for the selection of language-responsive sensors and sources for which, in a second stage, different time-resolved cognitive processes relevant to sentence comprehension can be studied. Specifically, the fine temporal resolution of MEG offers the potential to study the incremental construction of sentence structure and meaning in real time (54–56). Previous MEG studies have suggested that the temporal dynamics of sentence processing entail both feed-forward as well as recurrent processing (57), that surprisal values or embeddings extracted from artificial neural network language models capture aspects of neural signals recorded with MEG (27, 54, 58–61), and that activity in different frequency bands may reflect distinct cognitive processes, such as lexical retrieval, semantic composition, and prediction of upcoming words (62–64). Functional identification of language-responsive sensors and sources could help ensure that the observed effects – including the time-varying sensitivity to surprisal and other factors affecting incremental comprehension difficulty – occur within the language-selective system.

Limitations

We opted to include the top 10% of most responsive sensors and sources (highest F-values) as SOIs. A consequence of this choice is that the identified language response has an equal extent in all participants. The advantage is that the sensors and sources of interest enable the selection of a multivariate response pattern for each participant, suitable for MVPA. An alternative would be to preset a fixed F-statistic threshold, but this may have an unpredictable effect on the extent of the language response pattern if the method is applied to a non-neurotypical population. The top 10% cutoff is consistent with common practice in language mapping, where such a threshold is considered appropriately sensitive for detecting language responses (29). Confirmatory analyses using cutoffs of 5% and 15% yielded the same overall pattern of results suggesting that the findings are robust to this parameter choice (Supplemental Materials 3).

Differences in the timecourses of the neural responses were observed between the Dutch and English datasets (Figure 1), which may arise from multiple sources of variance. Minor methodological differences between datasets may have introduced language-specific differences in lexical familiarity and working memory demands, although stimulus materials were closely matched across languages (see Supplemental Materials 1). Cross-linguistic work has shown that the timing, amplitude, and spatial distribution of language-related neural responses can vary across languages due to differences in linguistic structure and processing strategies (65, 66). Furthermore, differences in participant characteristics between the Dutch and English groups, such as multilingual experience, language background, and cognitive profiles may have contributed to the observed neural variability. For instance, Jouravlev et al. (67) demonstrated using fMRI that polyglots show reduced activation in both magnitude and spatial extent within the language network during native language processing compared to matched controls, suggesting that individual-level factors such as multilingual experience can substantially shape language-related neural responses. How such participant-level variables interact with neural measures of language processing across different language communities remains an open question that future work with more systemic characterization of participant profiles could help disentangle.

Conclusion

Using an extensively validated fMRI language localizer task, based on sentence reading, we showed that the neural responses recorded with MEG are reproducible at the individual-participant level and we replicated these findings across two groups in two different languages (English and Dutch). We validated that language-responsive sensors and sources are spatially variable across individuals, giving an individual-level approach an advantage over the traditional group-averaging analysis. This method has a wide range of applications: from the detailed characterization of the time-course of language processing to probing the language system in non-neurotypical populations (especially when only a small number of participants are available), and may generally encourage the use of MEG to study the functional neuroanatomy of human higher-order cognition.

Supplementary Material

Supplemental Material 1, Supplemental Material 2, and Supplemental Material 3 available at: https://doi.org/10.17605/OSF.IO/VC2BW

Acknowledgements

The authors wish to thank dr. Anna Ivanova, dr. Antonin Rovai, and the Athinoula A. Martinos Imaging Center at the McGovern Institute for Brain Research.

Funding information

This work was supported by an FWO research grant to RB (1509318N), an FWO junior and senior postdoctoral fellowship to RB (12I2117N, 12I2121N), an FWO travel grant for a short stay abroad to RB, and a MISTI Belgium seed fund to EF and RB. RB and MH are supported by the Francqui Foundation. MH is additionally supported by an FWO PhD fellowship (11AAS26N). AP was supported by a diversity supplement to EF’s NIH grant R01-DC016950. EF was additionally supported by NIH awards R01-DC016607, R01-DC016950, and U01-NS121471, and by research funds from the McGovern Institute for Brain Research, the Brain and Cognitive Sciences department, and the Simons Center for the Social Brain.

Footnotes

Disclosures

The authors wish to disclose no conflicts of interest, financial or otherwise.

Data and Code Availability Statements

Initial preprocessing of the data made use of MaxFilter version 2.2, while further processing was performed with MNE-Python version 1.8.0. All scripts used to generate the outputs for this work as well as the task stimuli are available on OSF (https://osf.io/vc2bw/). Raw MEG data can be made available upon reasonable request.

References

  • 1.Lipkin B, Tuckute G, Affourtit J, Small H, Mineroff Z, Kean H, Jouravlev O, Rakocevic L, Pritchett B, Siegelman M, Hoeflin C, Pongos A, Blank IA, Struhl MK, Ivanova A, Shannon S, Sathe A, Hoffmann M, Nieto-Castañón A, Fedorenko E. Probabilistic atlas for the language network based on precision fMRI data from >800 individuals. Sci Data 9: 529, 2022. doi: 10.1038/s41597-022-01645-3. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 2.Fedorenko E, Hsieh P-J, Nieto-Castañón A, Whitfield-Gabrieli S, Kanwisher N. New Method for fMRI Investigations of Language: Defining ROIs Functionally in Individual Subjects. J Neurophysiol 104: 1177–1194, 2010. doi: 10.1152/jn.00032.2010. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 3.Saxe R, Brett M, Kanwisher N. Divide and conquer: A defense of functional localizers. NeuroImage 30: 1088–1096, 2006. doi: 10.1016/j.neuroimage.2005.12.062. [DOI] [PubMed] [Google Scholar]
  • 4.Nieto-Castañón A, Fedorenko E. Subject-specific functional localizers increase sensitivity and functional resolution of multi-subject analyses. NeuroImage 63: 1646–1669, 2012. doi: 10.1016/j.neuroimage.2012.06.065. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 5.Fedorenko E The early origins and the growing popularity of the individual-subject analytic approach in human neuroscience. Curr Opin Behav Sci 40: 105–112, 2021. doi: 10.1016/j.cobeha.2021.02.023. [DOI] [Google Scholar]
  • 6.Brett M, Johnsrude IS, Owen AM. The problem of functional localization in the human brain. Nat Rev Neurosci 3: 243–249, 2002. doi: 10.1038/nrn756. [DOI] [PubMed] [Google Scholar]
  • 7.Mahowald K, Fedorenko E. Reliable individual-level neural markers of high-level language processing: A necessary precursor for relating neural variability to behavioral and genetic variability. NeuroImage 139: 74–93, 2016. doi: 10.1016/j.neuroimage.2016.05.073. [DOI] [PubMed] [Google Scholar]
  • 8.Braga RM, DiNicola LM, Becker HC, Buckner RL. Situating the left-lateralized language network in the broader organization of multiple specialized large-scale distributed networks. J Neurophysiol 124: 1415–1448, 2020. doi: 10.1152/jn.00753.2019. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 9.Fedorenko E, Ivanova AA, Regev TI. The language network as a natural kind within the broader landscape of the human brain. Nat Rev Neurosci 25: 289–312, 2024. doi: 10.1038/s41583-024-00802-4. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 10.Lee JJ, Scott TL, Perrachione TK. Efficient functional localization of language regions in the brain. NeuroImage 285: 120489, 2024. doi: 10.1016/j.neuroimage.2023.120489. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 11.Olson HA, Chen EM, Lydic KO, Saxe RR. Left-Hemisphere Cortical Language Regions Respond Equally to Observed Dialogue and Monologue. Neurobiol Lang 4: 575–610, 2023. doi: 10.1162/nol_a_00123. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 12.Scott TL, Gallée J, Fedorenko E. A new fun and robust version of an fMRI localizer for the frontotemporal language system. Cogn Neurosci 8: 167–176, 2017. doi: 10.1080/17588928.2016.1201466. [DOI] [PubMed] [Google Scholar]
  • 13.Richardson H, Koster-Hale J, Caselli N, Magid R, Benedict R, Olson H, Pyers J, Saxe R. Reduced neural selectivity for mental states in deaf children with delayed exposure to sign language. Nat Commun 11: 3246, 2020. doi: 10.1038/s41467-020-17004-y. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 14.Malik-Moraleda S, Ayyash D, Gallée J, Affourtit J, Hoffmann M, Mineroff Z, Jouravlev O, Fedorenko E. An investigation across 45 languages and 12 language families reveals a universal language network. Nat Neurosci 25: 1014–1019, 2022. doi: 10.1038/s41593-022-01114-5. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 15.Diachek E, Blank I, Siegelman M, Affourtit J, Fedorenko E. The Domain-General Multiple Demand (MD) Network Does Not Support Core Aspects of Language Comprehension: A Large-Scale fMRI Investigation. J Neurosci 40: 4536–4550, 2020. doi: 10.1523/JNEUROSCI.2036-19.2020. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 16.Ivanova A, Siegelman M, Chandler C, Pongos A, Kean H, Fedorenko E. Effect of Task on Sentence Processing. 2020. [Google Scholar]
  • 17.Gao R, Cheung C, Siegelman M, Pongos ALA, Tanner A, Fedorenko E, Ivanova AA. The language network responds robustly to sentences across diverse tasks.. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 18.Fedorenko E, Behr MK, Kanwisher N. Functional specificity for high-level linguistic processing in the human brain. Proc Natl Acad Sci 108: 16428–16433, 2011. doi: 10.1073/pnas.1112937108. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 19.Monti MM, Parsons LM, Osherson DN. Thought Beyond Language: Neural Dissociation of Algebra and Natural Language. Psychol Sci 23: 914–922, 2012. doi: 10.1177/0956797612437427. [DOI] [PubMed] [Google Scholar]
  • 20.Amalric M, Dehaene S. Cortical circuits for mathematical knowledge: evidence for a major subdivision within the brain’s semantic networks. Philos Trans R Soc B Biol Sci 373: 20160515, 2018. doi: 10.1098/rstb.2016.0515. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 21.Pallier C, Devauchelle A-D, Dehaene S. Cortical representation of the constituent structure of sentences. Proc Natl Acad Sci 108: 2522–2527, 2011. doi: 10.1073/pnas.1018711108. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 22.Hu J, Small H, Kean H, Takahashi A, Zekelman L, Kleinman D, Ryan E, Nieto-Castañón A, Ferreira V, Fedorenko E. Precision fMRI reveals that the language-selective network supports both phrase-structure building and lexical access during language production. Cereb Cortex 33: 4384–4404, 2023. doi: 10.1093/cercor/bhac350. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 23.Shain C, Kean H, Casto C, Lipkin B, Affourtit J, Siegelman M, Mollica F, Fedorenko E. Distributed Sensitivity to Syntax and Semantics throughout the Language Network. J Cogn Neurosci 36: 1427–1471, 2024. doi: 10.1162/jocn_a_02164. [DOI] [PubMed] [Google Scholar]
  • 24.Giglio L, Hagoort P, Ostarek M. Neural encoding of semantic structures during sentence production. Cereb Cortex 34: bhae482, 2024. doi: 10.1093/cercor/bhae482. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 25.Brennan J, Pylkkänen L. The time-course and spatial distribution of brain activity associated with sentence processing. NeuroImage 60: 1139–1148, 2012. doi: 10.1016/j.neuroimage.2012.01.030. [DOI] [PubMed] [Google Scholar]
  • 26.Coolen T, Wens V, Vander Ghinst M, Mary A, Bourguignon M, Naeije G, Peigneux P, Sadeghi N, Goldman S, De Tiège X. Frequency-Dependent Intrinsic Electrophysiological Functional Architecture of the Human Verbal Language Network. Front Integr Neurosci 14: 27, 2020. doi: 10.3389/fnint.2020.00027. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 27.Huizeling E, Arana S, Hagoort P, Schoffelen J-M. Lexical Frequency and Sentence Context Influence the Brain’s Response to Single Words. Neurobiol Lang 3: 149–179, 2022. doi: 10.1162/nol_a_00054. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 28.Matchin W, Brodbeck C, Hammerly C, Lau E. The temporal dynamics of structure and content in sentence comprehension: Evidence from fMRI‐constrained MEG. Hum Brain Mapp 40: 663–678, 2019. doi: 10.1002/hbm.24403. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 29.Coolen T, Dumitrescu AM, Wens V, Bourguignon M, Rovai A, Sadeghi N, Urbain C, De Tiège X. Comparison of lateralization and spatial patterns of hemodynamic and neuromagnetic language brain mapping. Clin Neurophysiol Pract 10: 511–523, 2025. doi: 10.1016/j.cnp.2025.10.007. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 30.Grummich P, Nimsky C, Pauli E, Buchfelder M, Ganslandt O. Combining fMRI and MEG increases the reliability of presurgical language localization: A clinical study on the difference between and congruence of both modalities. NeuroImage 32: 1793–1803, 2006. doi: 10.1016/j.neuroimage.2006.05.034. [DOI] [PubMed] [Google Scholar]
  • 31.Herfurth K, Harpaz Y, Roesch J, Mueller N, Walther K, Kaltenhaeuser M, Pauli E, Goldstein A, Hamer H, Buchfelder M, Doerfler A, Prell J, Rampp S. Localization of beta power decrease as measure for lateralization in pre-surgical language mapping with magnetoencephalography, compared with functional magnetic resonance imaging and validated by Wada test. Front Hum Neurosci 16: 996989, 2022. doi: 10.3389/fnhum.2022.996989. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 32.Youssofzadeh V, Babajani-Feremi A. Mapping critical hubs of receptive and expressive language using MEG: A comparison against fMRI. NeuroImage 201: 116029, 2019. doi: 10.1016/j.neuroimage.2019.116029. [DOI] [PubMed] [Google Scholar]
  • 33.Flick G, Abdullah O, Pylkkänen L. From letters to composed concepts: A magnetoencephalography study of reading. Hum Brain Mapp 42: 5130–5153, 2021. doi: 10.1002/hbm.25608. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 34.Neophytou K, Manouilidou C, Stockall L, Marantz A. Syntactic and semantic restrictions on morphological recomposition: MEG evidence from Greek. Brain Lang 183: 11–20, 2018. doi: 10.1016/j.bandl.2018.05.003. [DOI] [PubMed] [Google Scholar]
  • 35.Gratton C, Braga RM. Editorial overview: Deep imaging of the individual brain: past, practice, and promise. Curr Opin Behav Sci 40: iii–vi, 2021. doi: 10.1016/j.cobeha.2021.06.011. [DOI] [Google Scholar]
  • 36.Gwilliams L, Lewis GA, Marantz A. Functional characterisation of letter-specific responses in time, space and current polarity using magnetoencephalography. NeuroImage 132: 320–333, 2016. doi: 10.1016/j.neuroimage.2016.02.057. [DOI] [PubMed] [Google Scholar]
  • 37.Liu J, Harris A, Kanwisher N. Stages of processing in face perception: an MEG study. Nat Neurosci 5: 910–916, 2002. doi: 10.1038/nn909. [DOI] [PubMed] [Google Scholar]
  • 38.De Vries E, Baldauf D. Attentional Weighting in the Face Processing Network: A Magnetic Response Image-guided Magnetoencephalography Study Using Multiple Cyclic Entrainments. J Cogn Neurosci 31: 1573–1588, 2019. doi: 10.1162/jocn_a_01428. [DOI] [PubMed] [Google Scholar]
  • 39.Keuleers E, Brysbaert M. Wuggy: A multilingual pseudoword generator. Behav Res Methods 42: 627–633, 2010. doi: 10.3758/BRM.42.3.627. [DOI] [PubMed] [Google Scholar]
  • 40.Dale AM, Fischl B, Sereno MI. Cortical Surface-Based Analysis. NeuroImage 9: 179–194, 1999. doi: 10.1006/nimg.1998.0395. [DOI] [PubMed] [Google Scholar]
  • 41.Fischl B FreeSurfer. NeuroImage 62: 774–781, 2012. doi: 10.1016/j.neuroimage.2012.01.021. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 42.Taulu S, Kajola M, Simola J. Suppression of Interference and Artifacts by the Signal Space Separation Method. Brain Topogr 16: 269–275, 2003. doi: 10.1023/B:BRAT.0000032864.93890.f9. [DOI] [PubMed] [Google Scholar]
  • 43.Gramfort A MEG and EEG data analysis with MNE-Python. Front Neurosci 7, 2013. doi: 10.3389/fnins.2013.00267. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 44.Hyvarinen A. Fast and robust fixed-point algorithms for independent component analysis. IEEE Trans Neural Netw 10: 626–634, 1999. doi: 10.1109/72.761722. [DOI] [PubMed] [Google Scholar]
  • 45.Jones SR, Pritchett DL, Sikora MA, Stufflebeam SM, Hämäläinen M, Moore CI. Quantitative Analysis and Biophysically Realistic Neural Modeling of the MEG Mu Rhythm: Rhythmogenesis and Modulation of Sensory-Evoked Responses. J Neurophysiol 102: 3554–3572, 2009. doi: 10.1152/jn.00535.2009. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 46.Braga RM, Buckner RL. Parallel Interdigitated Distributed Networks within the Individual Estimated by Intrinsic Functional Connectivity. Neuron 95: 457–471.e5, 2017. doi: 10.1016/j.neuron.2017.06.038. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 47.Fedorenko E, Blank IA. Broca’s Area Is Not a Natural Kind. Trends Cogn Sci 24: 270–284, 2020. doi: 10.1016/j.tics.2020.01.001. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 48.Frost MA, Goebel R. Measuring structural–functional correspondence: Spatial variability of specialised brain regions after macro-anatomical alignment. NeuroImage 59: 1369–1381, 2012. doi: 10.1016/j.neuroimage.2011.08.035. [DOI] [PubMed] [Google Scholar]
  • 49.Powell LJ, Deen B, Saxe R. Using individual functional channels of interest to study cortical development with fNIRS. Dev Sci 21: e12595, 2018. doi: 10.1111/desc.12595. [DOI] [PubMed] [Google Scholar]
  • 50.Luo H, Yu T, Li Q, Sheng L. Functional Specialization for Language Processing in Inferior Frontal Regions During Early Childhood: Evidence from functional near-infrared spectroscopy individual functional channels of interest approach. Neuroscience: 2025. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 51.Bruffaerts R, De Deyne S, Meersmans K, Liuzzi AG, Storms G, Vandenberghe R. Redefining the resolution of semantic knowledge in the brain: Advances made by the introduction of models of semantics in neuroimaging. Neurosci Biobehav Rev 103: 3–13, 2019. doi: 10.1016/j.neubiorev.2019.05.015. [DOI] [PubMed] [Google Scholar]
  • 52.Fedorenko E, Scott TL, Brunner P, Coon WG, Pritchett B, Schalk G, Kanwisher N. Neural correlate of the construction of sentence meaning. Proc Natl Acad Sci 113, 2016. doi: 10.1073/pnas.1612132113. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 53.Poldrack RA, Baker CI, Durnez J, Gorgolewski KJ, Matthews PM, Munafò MR, Nichols TE, Poline J-B, Vul E, Yarkoni T. Scanning the horizon: towards transparent and reproducible neuroimaging research. Nat Rev Neurosci 18: 115–126, 2017. doi: 10.1038/nrn.2016.167. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 54.Heilbron M, Armeni K, Schoffelen J-M, Hagoort P, De Lange FP. A hierarchy of linguistic predictions during natural language comprehension. Proc Natl Acad Sci 119: e2201968119, 2022. doi: 10.1073/pnas.2201968119. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 55.Ten Oever S, Carta S, Kaufeld G, Martin AE. Neural tracking of phrases in spoken language comprehension is automatic and task-dependent. eLife 11: e77468, 2022. doi: 10.7554/eLife.77468. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 56.Desbordes T, King J-R, Dehaene S. Tracking the neural codes for words and phrases during semantic composition, working-memory storage, and retrieval. Cell Rep 43: 113847, 2024. doi: 10.1016/j.celrep.2024.113847. [DOI] [PubMed] [Google Scholar]
  • 57.Hultén A, Schoffelen J-M, Uddén J, Lam NHL, Hagoort P. How the brain makes sense beyond the processing of single words – An MEG study. NeuroImage 186: 586–594, 2019. doi: 10.1016/j.neuroimage.2018.11.035. [DOI] [PubMed] [Google Scholar]
  • 58.Caucheteux C, King J-R. Brains and algorithms partially converge in natural language processing. Commun Biol 5: 134, 2022. doi: 10.1038/s42003-022-03036-1. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 59.Choi HS, Marslen-Wilson WD, Lyu B, Randall B, Tyler LK. Decoding the Real-Time Neurobiological Properties of Incremental Semantic Interpretation. Cereb Cortex 31: 233–247, 2021. doi: 10.1093/cercor/bhaa222. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 60.Gwilliams L, Marantz A, Poeppel D, King J-R. Hierarchical dynamic coding coordinates speech comprehension in the human brain. Neuroscience: 2024. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 61.Slaats S, Meyer AS, Martin AE. Lexical Surprisal Shapes the Time Course of Syntactic Structure Building. Neurobiol Lang 5: 942–980, 2024. doi: 10.1162/nol_a_00155. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 62.Lam NHL, Schoffelen J-M, Uddén J, Hultén A, Hagoort P. Neural activity during sentence processing as reflected in theta, alpha, beta, and gamma oscillations. NeuroImage 142: 43–54, 2016. doi: 10.1016/j.neuroimage.2016.03.007. [DOI] [PubMed] [Google Scholar]
  • 63.Prystauka Y, Lewis AG. The power of neural oscillations to inform sentence comprehension: A linguistic perspective. Lang Linguist Compass 13: e12347, 2019. doi: 10.1111/lnc3.12347. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 64.Recasens M, Gross J, Uhlhaas PJ. Low-Frequency Oscillatory Correlates of Auditory Predictive Processing in Cortical-Subcortical Networks: A MEG-Study. Sci Rep 8: 14007, 2018. doi: 10.1038/s41598-018-32385-3. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 65.Bornkessel-Schlesewsky I, Kretzschmar F, Tune S, Wang L, Genç S, Philipp M, Roehm D, Schlesewsky M. Think globally: Cross-linguistic variation in electrophysiological activity during sentence comprehension. Brain Lang 117: 133–152, 2011. doi: 10.1016/j.bandl.2010.09.010. [DOI] [PubMed] [Google Scholar]
  • 66.Tune S, Schlesewsky M, Small SL, Sanford AJ, Bohan J, Sassenhagen J, Bornkessel-Schlesewsky I. Cross-linguistic variation in the neurophysiological response to semantic processing: Evidence from anomalies at the borderline of awareness. Neuropsychologia 56: 147–166, 2014. doi: 10.1016/j.neuropsychologia.2014.01.007. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 67.Jouravlev O, Mineroff Z, Blank IA, Fedorenko E. The Small and Efficient Language Network of Polyglots and Hyper-polyglots. Cereb Cortex 31: 62–76, 2021. doi: 10.1093/cercor/bhaa205. [DOI] [PMC free article] [PubMed] [Google Scholar]

Associated Data

This section collects any data citations, data availability statements, or supplementary materials included in this article.

Data Availability Statement

Initial preprocessing of the data made use of MaxFilter version 2.2, while further processing was performed with MNE-Python version 1.8.0. All scripts used to generate the outputs for this work as well as the task stimuli are available on OSF (https://osf.io/vc2bw/). Raw MEG data can be made available upon reasonable request.

RESOURCES