Claims 25
From the paper Distinct spatiotemporal patterns of syntactic and semantic processing in human inferior frontal gyrus · claims file · checked 2026-10-22: ready
- participants
Ten participants (four male, six female, 19 to 50 years) were recorded with two 128-electrode grids each; all ten did the CORR, SYN-P, SYN-C and SEM conditions, five of them the combined violation (COM) and the other five the non-linguistic task (NL).
This study included ten participants (age range, 19–50 years; four males, six females) ... All the ten participants performed under CORR, SYN-P, SYN-C and SEM conditions. Five participants performed under COM and the other five performed under NL conditions.
Methods, Participants, paragraph 1, and The Chinese language and stimuli design, paragraph 3; Supplementary Table 1How it will be judged
Counts of distinct subjects in the dataset, their sex, and the tasks each has. Tolerance zero, as these are counts the paper gives exactly. Reproduced when all match; partially reproduced when the dataset holds a subset of the participants with matching metadata. Ages are compared with Supplementary Table 1 (30, 39, 19, 30, 50, 34, 42, 36, 40, 31 for participants 1 to 10) where the dataset gives them.
- task-structure
Each of the CORR, SYN-P, SYN-C, SEM and COM conditions has 32 sentences, with six words per sentence in CORR, SYN-C and SEM and seven in SYN-P and COM, each word shown for 400 ms with a 100 ms interval after an 800 ms fixation and a 500 ms blank.
Each trial started with an 800 ms fixation followed by a 500 ms blank screen, and then each word of the sentence (or geometric shape) was displayed for 400 ms, with a 100 ms interstimulus interval. There were six words per sentence in the CORR, SYN-C and SEM conditions and seven words in the SYN-P and COM conditions.
Methods, The Chinese language and stimuli design, paragraph 4; Supplementary Information, stimulus sentence listsHow it will be judged
The 32 sentences per condition are read from the numbered lists in the supplement (no sentence of the text gives the count). Per participant, the number of trials of each condition (before bad-interval exclusion, where the dataset says which were excluded) and the word timing from the event times. The timing tolerance of 10 ms is one sample of the 100 Hz HGA. Reproduced when every count matches and the timings are within tolerance in every participant; partially reproduced when they do in some.
- behaviour-response-time
Most participants responded fast and accurately in the linguistic task, most response times lying around 0.5 to 1.0 s.
Most of the responding time is around 0.5-1.0 second.
Results, paragraph 1; Supplementary Figure 2b legendHow it will be judged
Response times of the linguistic trials of all ten participants pooled, measured from the onset of the question as the dataset times it. The paper gives no measure of "most"; the reading taken is that more than half of the response times lie in the range widened by 10% of each bound (0.45 to 1.1 s), and no other is named in the text. Accuracy is reported (proportion of correct judgements per participant) but has no number to match. Reproduced when more than half lie in the range.
- fig1b-high-gamma-best-band
High-gamma components above 70 Hz classify the linguistic conditions with the highest accuracy (above 80%), and the evoked responses are more prominent in the 70 to 150 Hz band than in other bands.
Results showed that high-gamma (HG) components (>70 Hz) discriminated conditions with the highest accuracy (>80%, Fig. 1b and Supplementary Fig. 3a) and ERPs were more prominent in the HG band (70–150 Hz) compared with other frequency bands
Results, paragraph 3; Figure 1b; Supplementary Figure 3a to 3eHow it will be judged
Needs the broadband signal, from which the analytic amplitude of bands from 0 to 150 Hz in 10 Hz steps is taken as the first principal component across band-passes, and classified with the SVM, logistic regression and k-nearest-neighbour settings of the Methods (Frequency component extraction) into correct, syntactic violation, semantic violation and resting state. If the dataset holds only high-gamma activity, as its description says, the claim cannot be tested. Otherwise, open, the cross-validation and the features (time samples of the band amplitude in the keyword window). The paper reports the band accuracies for the SVM (Fig. 1b legend) and for logistic regression (Supplementary Fig. 3a legend) only; reproduced when, under both, a band with lower bound at or above 70 Hz gives the highest accuracy and, for the SVM, it exceeds 0.72 (80% less the tolerance). The kNN result is shown as a variant. The ERP part is judged from the measure of Supplementary Fig. 3d (the peak power during violation against the peak power during CORR, per responding electrode), and holds when the mean difference is largest in the 70 to 150 Hz band.
- fig1c-responsive-in-ifg
Most electrodes showing enhanced activity to linguistic stimuli lie in the IFG.
Most of the electrodes showing enhanced activities to linguistic stimuli were located in the IFG
Results, paragraph 3; Figure 1c; Supplementary Figure 3fHow it will be judged
The rest against stimulus test of E (with readings E1a and E1b) applied to every electrode of every participant, regardless of location, with HGA higher during stimuli. Reproduced when more than half of the responsive electrodes carry an IFG label. Open, which labels form the IFG, (a) pars opercularis and pars triangularis, the labels of E, or (b) these and pars orbitalis.
- fig1d-only-after-keyword
The enhanced activity for violation conditions relative to CORR appears only after the keyword.
In the linguistic task, the enhanced activity of violation conditions relative to CORR was only observed after the keywords, indicating that the responses were highly linked to the violation manipulations (Fig. 1d).
Results, paragraph 3; Figure 1dHow it will be judged
A claim of absence before the keyword, judged on the paper's test, the test of V (the discriminating period of Fig. 1e and the ERP amplitude Methods), applied to the electrodes of E over the sentence before the keyword, aligned to the first word (words 1 to 3, 0 to 1,500 ms). Reproduced when no electrode and condition pair passes V before the keyword; not reproduced when any does. The count of passing pairs, before and after the keyword, is reported either way. The SEM verb (word 2) differs from the CORR verb; the test without SEM samples 500 to 1,000 ms is shown as a variant.
- fig2b-violation-electrodes
58 IFG electrodes respond to at least one violation type more strongly than to correct sentences.
we defined a significant effect in each electrode if its response to a certain type of violation was reliably enhanced compared with the CORR condition for more than 50 ms (P < 0.05, ANOVA, two tailed, Bonferroni correction for the number of electrodes, n = 256 for each participant); there were 58 electrodes satisfying this criterion
Results, paragraph 5; Figure 2bHow it will be judged
Electrodes of E (readings E1a, E1b) passing V (readings V1a, V1b, V2a, V2b) for at least one of SYN-P, SYN-C and SEM, over the ten participants. Reproduced when the count is within tolerance under at least one combination of readings (a count that a reading can move is a reported number read in more than one way, so any reading may match); not reproduced when none is. The count under every combination is reported.
- fig2b-single-type
Most responding electrodes (49 of 58) respond strongly to only one violation type; few (9 of 58) are co-activated by two types, mainly SYN-P and SYN-C.
We found that most of the electrodes (49 of 58) strongly responded to only one type of violation (Fig. 2b, one-way ANOVA F statistic > 4). Only a few electrodes were co-activated by two violation types (9 of 58) and were mainly co-activated by SYN-P and SYN-C
Results, paragraph 5; Figure 2bHow it will be judged
Among the electrodes of fig2b-violation-electrodes, an electrode strongly responds to a violation when its F (readings F1, F2, F3) exceeds 4. Counts of electrodes strongly responding to exactly one, and to two, of the three types. Reproduced when, under one combination of readings, both counts are within tolerance and the SYN-P and SYN-C pair is the most frequent pair among the co-activated electrodes; partially reproduced when the single-type electrodes are a majority and SYN-P with SYN-C is the most frequent pair, but the counts are outside tolerance under every reading. The 10% tolerance on 9 is below one electrode, so the second count must match exactly.
- suppfig6a-half-split
When half of the sentences select the responding electrodes and the other half are analysed, most electrodes (46 of 62) still respond to one violation type only and 16 of 62 to two, mainly SYN-P and SYN-C.
We used first half sentences to identify the linguistic responding electrodes and used the remaining sentences for further analysis, to resolve non-independence issue. We found that most of the electrodes (46/62) strongly responded to only one type of violation (below figure). Only a few electrodes were ... by two violation types (16/62) and were mainly ... by SYN-P and SYN-C
Results, paragraph 5; Supplementary Figure 6a legendHow it will be judged
As fig2b-violation-electrodes and fig2b-single-type, with E and V computed on the trials of the first half of the sentences and F on the remaining trials. Open, (a) "first half" as sentences 1 to 16 of each condition's list, or (b) the first half of each condition's trials in presentation order. Reproduced when, under one combination of readings, all three counts are within tolerance and SYN-P with SYN-C is the most frequent pair; partially reproduced when the single-type electrodes are a majority and the pair is the most frequent, with counts outside tolerance.
- suppfig6b-duration-metric
When the length of the period differing from CORR is used in place of amplitude, most electrodes (61 of 79) respond to one violation type only and 18 of 79 to two, mainly SYN-P and SYN-C.
We found that most of the electrodes (61/ 79) strongly responded to only one type of violation (below figure). Only a few electrodes were ... by two violation types (18/79) and were mainly ... by SYN-P and SYN-C
Results, paragraph 5; Supplementary Figure 6b legendHow it will be judged
The paper does not say how 79 electrodes are selected or what length counts as a response. Readings, (a) electrodes of E with a significant period (V, without the amplitude rule V2) of more than 50 ms for a violation count as responding to it; (b) as (a) with the amplitude rule kept. The length of an electrode's significant period is the number of significant samples times 10 ms. Reproduced when, under one reading (crossed with V1a and V1b), all three counts are within tolerance and SYN-P with SYN-C is the most frequent pair; partially reproduced when the single-type electrodes are a majority and the pair is the most frequent, with counts outside tolerance.
- fig2c-euclidean-similarity
The response time courses of SYN-P and SYN-C are more similar (Euclidean distance 48.43) than those of the syntactic and semantic conditions (62.17).
the similarity between SYN-P and SYN-C (Euclidean distance, 48.43) was greater than that between syntactic and semantic conditions (Fig. 2c, Euclidean distance, 62.17)
Results, paragraphs 5 and 6; Figure 2cHow it will be judged
Distances computed with scipy's hierarchical clustering on Euclidean distance (Methods, Electrode selection, paragraph 3), between condition vectors built from the responding electrodes of fig2b-violation-electrodes. Open, (a) each condition's vector is the trial-mean HGA from 0 to 1,000 ms of every responding electrode, concatenated over electrodes; (b) the trial-mean HGA time course from 0 to 1,000 ms averaged over the responding electrodes (the Methods call it a "time-course similarity", so a reading without time is not allowed); and the "syntactic to semantic" distance read as (i) SEM to SYN-P, (ii) SEM to SYN-C, (iii) the mean of the two, or (iv) the linkage height at which SEM joins the SYN-P and SYN-C cluster. Fig. 2c's y axis is a "relative distance", so absolute values depend on the vector. Reproduced when, under every reading, the syntactic to semantic distance is larger than SYN-P to SYN-C, and under at least one reading both values are within tolerance; partially reproduced when the order holds under every reading but no reading matches the values; not reproduced when the order fails under every reading; blocked otherwise. The values under each reading are reported.
- suppfig7a-com-overlap
Of the electrodes responding to the combined violation (COM), 78.3% also respond to SYN-P but only 26.1% to SEM.
for the electrodes responding to the combined violation condition (COM), 78.3% of them also showed enhanced activity to the syntactic violation (SYN-P), but only 26.1% of them responded to the semantic violation (SEM)
Results, paragraph 6; Supplementary Figure 7aHow it will be judged
In the five participants with COM, electrodes of E passing V for COM against CORR (COM keyword at word 5, as in SYN-P). Fractions of those that also pass V for SYN-P, and for SEM. Open, whether "responded" to SYN-P and SEM means passing V or having F > 4 (F1 to F3), as Supplementary Fig. 7a plots F values. The smallest count the two percentages fit is 18 and 6 of 23 electrodes (an inference; the paper gives no count), so the 10% tolerance on 26.1% is below one electrode there. Reproduced when both fractions are within tolerance under one reading; partially reproduced when, under every reading, the SYN-P fraction exceeds the SEM fraction but the values are outside tolerance.
- suppfig7b-not-nonlinguistic
The electrodes responding to linguistic violations do not respond to the non-linguistic violation in the high-gamma band.
at 70 -150 Hz, the linguistic responding electrodes were almost completely not responding to the non-linguistic task
Results, paragraph 6; Supplementary Figure 7b and 7c legendHow it will be judged
A claim of absence. In the five NL participants, the electrodes of fig2b-violation-electrodes tested for NL-VIO against NL-CORR with the test of V, NL trials aligned to the onset of the shape in the keyword's position. The paper's NL test is a one-way ANOVA at P < 0.01, corrected (Supplementary Fig. 4b); the threshold read is 0.01/256 per sample, with 0.05/256 (the linguistic test) as a variant. "Almost completely not" is read as at most 5% of those electrodes passing, the rate a test at the paper's alpha allows. Reproduced when no more than 5% pass; not reproduced otherwise. The count and the electrodes' NL-VIO minus NL-CORR differences are reported either way.
- fig2de-no-spatial-separation
Across participants, the activation centres and epicentres of the SYN-P, SYN-C and SEM electrodes are not spatially separated in the IFG.
Unlike the result at the electrode level, we failed to observe a tendency of spatial separation for the activation centre of syntactic and semantic conditions. Likewise, the epicentre of these responses failed to show separations using a permutation test of the clusters of electrodes (Fig. 2e).
Results, paragraph 7; Figure 2d and 2e; Supplementary Figure 8aHow it will be judged
A claim of absence, judged on the paper's test, a non-parametric permutation test of the spatial segregation among SYN-P, SYN-C and SEM electrodes on the MNI152 template (Methods, Spatial organization). The Methods do not give the statistic or the permutation scheme. Readings, (a) for each pair of conditions, the distance between the mean MNI coordinates of their electrodes; (b) the distance between the peaks of their kernel density estimates (bandwidth 20, as the Methods give, in mm); in both, condition labels permuted among the electrodes of the pair 1,000 times (the permutation count the Methods give for ALE), two-sided, p < 0.05. Co-activated electrodes count in each of their conditions. Reproduced when no pair is significant under every reading; not reproduced when a pair is under every reading; blocked otherwise. The distances are reported. The ALE analysis (GingerALE) is not repeated.
- suppfig8b-individual-no-separation
Even within individual participants the SYN-P, SYN-C and SEM electrodes show no clear spatial separation.
However, even at the individual level, we were unable to obtain a clear separation pattern (Supplementary Fig. 8b).
Results, paragraph 7; Supplementary Figure 8bHow it will be judged
The paper shows one representative participant and reports no test. The test of fig2de-no-spatial-separation, reading (a), run per participant on native or MNI coordinates, for pairs of conditions with at least three electrodes each (a standard minimum, not the paper's). A claim of absence, reproduced when no participant has a pair significant at p < 0.05 uncorrected; not reproduced otherwise. The number of participants with a significant pair is reported.
- fig2f-circuit-diameter
From the fraction of co-activated electrodes, the diameter of the functional circuits is estimated as no greater than about 3.25 mm.
The results showed that the diameter of functional circuits was no greater than ~3.25 mm (Fig. 2f)
Results, paragraph 8; Figure 2f; Methods, Functional circuit size estimationHow it will be judged
The Methods give p(co-activation) = 1 - (r - 0.5 mm)/r for an electrode of radius 0.5 mm and a circuit of diameter r, so r = 0.5 mm / p. p is the fraction of co-activated electrodes from fig2b-single-type (9 of 58 gives 3.22 mm, matching the paper's 15.6% and 3.25 mm on Fig. 2f). Reproduced when r from the dataset's fraction is within tolerance under a reading that also reproduces fig2b-single-type, or under any reading if none does.
- suppfig9-predict-from-corr
Electrode types (SYN-P, SYN-C, SEM, non-responding) can be predicted from responses to correct sentences above the 25% chance level, with accuracies of 45.6%, 73.0%, 69.3% and 38.4%.
our data showed the predicted electrode types with accuracies significantly higher than chance level (SYN-P 45.6%, SYP-C 73.0%, SEM 69.3%, and not-responding electrodes 38.4%; all P < 0.0001 relative to chance level 25%, Supplementary Fig. 9)
Results, paragraph 9; Supplementary Figure 9 legend; Methods, Predicting different types of electrodesHow it will be judged
k-nearest neighbours (Supplementary Fig. 9 legend) with the settings the Methods give for the frequency decoding (5 neighbours, Minkowski p = 2, uniform weights; the Methods state them for that analysis, and no other kNN settings are given), features from the CORR response of each electrode only, onset, offset, peak (L), area and amplitude at T0 (Methods, Predicting different types of electrodes). Open, (a) the electrodes and labels, single-type electrodes of fig2b-single-type labelled by their type, and non-responding electrodes either (a1) the electrodes of E that pass V for no condition or (a2) the IFG electrodes outside E; co-activated electrodes left out; (b) the cross-validation, (i) stratified 5-fold repeated 100 times with the t-test over repeats, or (ii) leave-one-out with the t-test over electrodes of each class (correct as 1, wrong as 0); (c) features standardized or raw; (d) L readings. Accuracy per class is the fraction of that class's electrodes predicted correctly. Reproduced when, under one combination of readings, all four accuracies are within tolerance and each exceeds 25% with two-sided t-test p < 0.0001; partially reproduced when, under every combination, all four exceed chance at p < 0.0001 but the values miss tolerance; not reproduced when any class fails to exceed chance under every combination.
- fig3-latency-values
SYN-P responses start and peak earliest (onset 113 ms, peak 480 ms), then SEM (122 ms, 524 ms) and SYN-C (283 ms, 579 ms).
The results suggested that the SYN-P (onset, 113 ms and peak, 480 ms) occurred earlier than SEM (122 ms and 524 ms) and SYN-C (283 ms and 579 ms) in terms of both onset and peak of the response, respectively.
Results, paragraph 10; Figure 3c, 3d and 3fHow it will be judged
Latencies L per electrode and condition, averaged (mean, as Fig. 3a's box centre is the mean) over electrodes. Open, the electrodes, (a) for each condition the electrodes passing V for it; (b) the electrodes of E+ (pars opercularis, pars triangularis and precentral, the set of the Fig. 3d legend, n = 106), each giving latencies for all three conditions; with L1, L2, L3 and the readings of E and V. Reproduced when, under one combination, all six values are within tolerance and SYN-P precedes SEM and SYN-C in both onset and peak; partially reproduced when the order holds under every combination but no combination puts all six within tolerance. The n of each reading is reported against 106.
- fig3d-latency-differences
Onset and peak latencies differ significantly between each pair of SYN-P, SYN-C and SEM.
we estimated the latency differences based on ANOVA (Fig. 3d, P < 0.001, one-way ANOVA, two tailed, Bonferroni, corrected for ten participants.
Results, paragraph 10; Figure 3d legendHow it will be judged
For each pair of conditions and for onset and peak, a one-way ANOVA (f_oneway) of the electrodes' latencies of the two conditions, p multiplied by 10 (Bonferroni "for ten participants", as printed). Electrodes and latencies under the readings of fig3-latency-values. Reproduced when, under every combination of readings, all six corrected p are below 0.001; partially reproduced when at least four are under every combination; not reproduced when fewer than four are under every combination; blocked otherwise. The paper's p for SEM against SYN-P onset (1.83e-5) with means 9 ms apart is noted as surprising; it is matched as printed.
- fig3-sync-latency-variability
Onset and peak latencies vary more across electrodes for SYN-C than for SYN-P.
This possibility may help explain the results that relatively later (and also greater variations of) onset and peak were observed for SYN-C than SYN-P.
Results, paragraph 10; Figure 3c and 3dHow it will be judged
The paper gives no measure or test of the variation. Electrodes and latencies under the readings of fig3-latency-values; the spread is the standard deviation of the latencies across electrodes (a standard choice; the interquartile range, which Fig. 3a's boxes show, as a variant). Reproduced when the SYN-C spread exceeds the SYN-P spread for both onset and peak under every reading; not reproduced when it does for neither under every reading; blocked otherwise.
- fig3e-latency-not-location
Response latencies do not depend on the location of the electrodes along the anterior-posterior or superior-inferior axis, or across pars opercularis, pars triangularis and precentral gyrus.
Moreover, response latencies did not depend on the spatial location of the electrodes (Fig. 3e and Supplementary Fig. 10c,d).
Results, paragraph 11; Figure 3e; Supplementary Figure 10c and 10dHow it will be judged
A claim of absence, judged on the paper's test, a linear regression of latency on electrode coordinate with its 95% confidence interval (Fig. 3e legend), for onset and peak, each condition, and each axis. The legend calls the axes the x and y coordinates, while the panel names the anterior-posterior and superior-inferior axes; on MNI152 these are y and z. Open, (a) MNI y and z; (b) MNI x and y, as the legend prints. Reproduced when every slope's 95% CI includes zero; not reproduced when any excludes it, under the readings of fig3-latency-values and of the axes (blocked when they differ). Slopes are reported in ms per mm. The comparison across the three areas (Supplementary Fig. 10c) has no test in the paper; a one-way ANOVA of latency on area at p < 0.05 is shown as a variant.
- suppfig10a-separable-400-700
The accuracy of classifying the conditions after the keyword peaks about 400 to 700 ms after keyword onset.
The accuracies for classifying different conditions peaked at about 400–700 ms after the keywords, which indicated they were most separable in this time frame
Results, paragraph 10; Supplementary Figure 10a legendHow it will be judged
The paper does not name the classifier, features, time resolution or cross-validation of this panel. Readings, each of the three classifiers with the Methods settings (SVM, logistic regression, kNN), classifying CORR, SYN-P, SYN-C and SEM trials together, or each violation against CORR (the legend says "classification accuracy of each violation type"), from the HGA of the responding electrodes of fig2b-violation-electrodes within each participant at each 10 ms sample (or in 100 ms windows), 5-fold stratified cross-validation (a standard choice), accuracy averaged over participants. Reproduced when, under every reading, the time of maximum accuracy lies in 360 to 770 ms (the range widened by 10% of each bound); not reproduced when it lies outside under every reading; blocked otherwise.
- suppfig10b-duration-differs
The total length of the significant difference from CORR differs between SYN-P, SYN-C and SEM electrodes.
The discrepancy in terms of the total length of the significant difference between respect violation and correct sentence of each electrode (ANOVA). This also indicated temporal differences in processing each linguistic component
Results, paragraph 11; Supplementary Figure 10b legendHow it will be judged
For each electrode and each violation it passes V for, the total length of significant samples (times 10 ms). One-way ANOVA of these lengths across SYN-P, SYN-C and SEM, and pairwise one-way ANOVAs Bonferroni-corrected for three pairs (the paper does not say what is corrected; the three pairs are read). Reproduced when the three-group ANOVA has p < 0.001 and at least one corrected pair has p < 0.001 under every reading of E and V; the group medians are reported.
- nl-low-frequency
In the non-linguistic task, the violation enhances activity mainly at lower frequencies (4 to 10 Hz), where many electrodes respond to both linguistic and non-linguistic tasks, unlike the high-gamma band.
In contrast, in the NL task, the violation condition showed an enhanced activity mainly at the lower frequency (Supplementary Fig. 4), indicating that neural oscillations during linguistic and NL tasks may arise in different frequency bands.
Results, paragraph 3; Supplementary Figures 4 and 7cHow it will be judged
Needs the 4 to 10 Hz amplitude, which a high-gamma dataset does not hold. If the broadband signal is present, the test of V for NL-VIO against NL-CORR at P < 0.01 corrected (Supplementary Fig. 4b) on the 4 to 10 Hz amplitude and on HGA; reproduced when more electrodes pass at 4 to 10 Hz than in high gamma. The measure of Supplementary Fig. 4a (peak power during NL-VIO against peak power during NL-CORR, per electrode and band) is shown with it.
- stimuli-word-frequency
Keyword frequencies are matched between CORR (shared by SYN-P and SEM; mean 31.88 per million) and SYN-C (30.25 per million).
the word frequencies were matched between CORR (same in SYN-P and SEM, mean frequency = 31.88 per million, s.d. = 35.14) and SYN-C (mean frequency = 30.25 per million, s.d. = 73.58) (t(62) = 0.11, P = 0.91)
Methods, The Chinese language and stimuli design, paragraph 3How it will be judged
Keywords from the supplement's sentence lists (word 4 of CORR and SYN-C), frequencies from the corpus the paper names (lingua.mtsu.edu/chinese-computing, the Modern Chinese word frequency list). Reproduced when both means are within tolerance and an independent two-sample t-test gives p > 0.05.
Pipeline
- Paper
- done (1 of 1)
- Claims
- done
- Claims check
- done (ready)
- Attempt
- todo
- Questions
- none
- Verified
- none