Pupillometry reveals resting state alpha power correlates with individual differences in adult auditory language comprehension

Although individual differences in adult language processing are well-documented, the neural basis of this variability remains largely unexplored. The current study addressed this gap in the literature by examining the relationship between resting state alpha activity and individual differences in auditory language comprehension. Alpha oscillations modulate cortical excitability, facilitating efficient information processing in the brain. While resting state alpha oscillations have been tied to individual differences in cognitive performance, their association with auditory language comprehension is less clear. Participants in the study were 80 healthy adults with a mean age of 25.8 years (SD = 7.2 years). Resting state alpha activity was acquired using electroencephalography while participants looked at a benign stimulus for 3 min. Participants then completed a language comprehension task that involved listening to 'syntactically simple' subject-relative clause sentences and 'syntactically complex' object-relative clause sentences. Pupillometry measured real-time processing demand changes, with larger pupil dilation indicating increased processing loads. Replicating past research, comprehending object relative clauses, compared to subject relative clauses, was associated with lower accuracy, slower reaction times, and larger pupil dilation. Resting state alpha power was found to be positively correlated with the pupillometry data. That is, participants with higher resting state alpha activity evidenced larger dilation during sentence comprehension. This effect was more pronounced for the 'complex' object sentences compared to the 'simple' subject sentences. These findings suggest the brain's capacity to generate a robust resting alpha rhythm contributes to variability in processing demands associated with auditory language comprehension, especially when faced with challenging syntactic structures. More generally, the study demonstrates that the intrinsic functional architecture of the brain likely influences individual differences in language comprehension.


Introduction
While our ability to use language reaches functional maturity by middle childhood (Hoff, 2009), individual variability in linguistic ability persists into adulthood (Kidd et al., 2018).This is especially the case with respect to understanding spoken language (e.g., Chipere, 2001;Da ˛browska & Street, 2006;Kuperman & Van Dyke, 2011).Research to date has primarily examined the extent to which domain-general cognitive abilities and demographic factors contribute to this variability (e.g., Da ˛browska, 2018;Just & Carpenter, 1992;Street, 2017;Wells et al., 2009).Much less is known, however, about the neural basis of individual differences in this domain.The current study addressed this gap in the literature by examining whether variability in adult sentence comprehension is related to the brain's intrinsic activity as revealed through resting state electroencephalography (EEG).
Intrinsic, or non-evoked, brain activity refers to spontaneous neural fluctuations that occur in the absence of task or sensory stimulation (Llin as, 2001).Rather than representing unimportant background noise, non-tasked related brain activity is increasingly recognised as reflecting the intrinsic functional architecture of the brain, which in turn, defines processing capabilities and limitations (Deco & Corbetta, 2011;Deco et al., 2011;Raichle, 2009;Raichle & Mintun, 2006;Sadaghiani & Kleinschmidt, 2013).In line with this perspective, intrinsic brain activity has been found to correlate with individual differences on a range of learning, memory and cognitive tasks (e.g., Cross et al., 2022;Immink et al., 2021;Lum et al., 2023;Mahjoory et al., 2019;Oswald et al., 2017;Prat et al., 2016;Sugata et al., 2020).
Intrinsic brain activity is examined using 'resting state' or 'task free' neuroimaging paradigms.In this research, brain activity is recorded while participants look at a benign stimulus (e.g., a dot) on a computer display or while they close their eyes for a few minutes (Anderson & Perone, 2018).During this time, neural oscillatory activity can be observed in the resting state EEG and magnetoencephalogram (MEG).This type of oscillatory activity arises from cycles of synchronised-desynchronised firing by neuronal populations.More specifically, the dendritic electrical activity generated by synchronised firing can be observed as rhythmic fluctuations, or oscillations, of varying frequencies (Cohen, 2017).In the human brain, oscillatory activity manifests in distinct frequency bands termed delta (1e4 Hz), theta (4e7 Hz), alpha (8e12 Hz), beta (12e30 Hz), and gamma (>30 Hz) (Buzs aki et al., 2013).This type of brain activity cannot be detected using fMRI or resting state fMRI paradigms.fMRI captures changes in cerebral blood flow and volume rather than direct electrical activity, changes that occur on the order of seconds and are too slow to capture neuronal firing dynamics (Logothetis & Pfeuffer, 2004).Thus, resting state M/EEG is required to study these frequency bands.In M/EEG resting state research, oscillatory activity within a frequency band is typically quantified in terms of either amplitude or power (which is amplitude squared).In general terms, as more neurones synchronously fire at the same frequency, amplitude/power increases (Pfurtscheller & Aranibar, 1977).
Alpha oscillations are the dominant frequency band, in terms of power, present in the human M/EEG frequency spectrum (Buzsaki, 2006).Oscillatory activity in this frequency band has been studied extensively in resting state research, especially in relation to cognitive functioning (Clark et al., 2004;Doppelmayr et al., 2002;Oswald et al., 2017;Reichert et al., 2016).This focus stems from evidence that suggests alpha oscillations play a pivotal role in regulating cortical activity and information processing in the brain (Foxe & Snyder, 2011;Klimesch, 1999Klimesch, , 2012;;Sadaghiani & Kleinschmidt, 2016).Specifically, during sensorimotor or higher-order operations, increased alpha synchronisation (or power) serves to inhibit irrelevant neural networks, while decreased synchronisation releases task-relevant networks from inhibition (Klimesch, 1996(Klimesch, , 2012;;Klimesch et al., 2007).
Alpha oscillatory activity appears to support language comprehension (Prystauka & Lewis, 2019).First, increases in alpha power have been observed when comprehending sentences that are highly demanding on working memory.For example, Meyer et al. (2013) found alpha power, localised to the left parietal lobe, was higher when comprehending sentences with long distance subject-verb dependencies (e.g., 'The coach [subject] has after a season on the German soccer league the striker honored[verb]'), compared to short (e.g., 'After a season on the German soccer league has the coach [subject] the striker honored[verb]').The evidence suggests increases in parietal related alpha activity serve to shield taskfocused content from irrelevant information by suppressing non-task related neural processes (Bonnefond & Jensen, 2012).In the context of the above comprehension task, increases in alpha power may have served to ensure the sentence's subject was retained in working memory until the verb was encountered.
Second, decreases in alpha power have been observed when comprehending syntactically difficult sentences.For example, Meltzer and Braun (2011) found alpha power was lower after participants comprehended object relative clauses (e.g., 'The man who the woman is teaching is discussing a hard problem.')compared to subject relative clauses (e.g., 'The man who is teaching the woman is discussing a hard problem.').This reduction in alpha activity was localised bilaterally to frontal and temporal regions.Since decreases in alpha power (or synchronisation) increase cortical excitability (Klimesch et al., 2007), the results suggest a greater allocation of neural resources to comprehend the object relative clauses.This is in line with a well replicated finding that object relative clauses are not only more difficult to comprehend than their subject counterparts (Gordon et al., 2001;King & Just, 1991;Staub, 2010;Traxler et al., 2002;Ueno & Garnsey, 2008), but they are also associated with greater bilateral cortical activation (Just et al., 1996).In general terms, this effect arises because of the non-canonical syntactic structure.Specifically, in object relative clauses, the object moves to the position typically occupied by the subject.This makes the sentence less predictable (Roland et al., 2007), which may explain greater cortical activation and therefore decreased alpha activity.In sum, there is evidence to suggest the brain's ability to synchronise (i.e., increase in alpha power needed to protect information held in working memory) and desynchronise (i.e., decrease in alpha power needed to increase neural resources) in the alpha band supports auditory language comprehension (Prystauka & Lewis, 2019).If this is the case, we might expect individual differences in sentence comprehension to covary with differences in the intrinsic or resting state alpha rhythm given its role not only in language processing, but regulation of cortical activity.
Surprisingly, very few studies have observed an association between individual differences in sentence comprehension and resting state alpha power.In children aged between 4 and 6 years, Kwok et al. (2019) found resting state alpha power correlated with a composite measure of language functioning encompassing both comprehension and productive language skills.In adults, however, the association is less clear.Beese et al. (2017) examined the association between resting state power and comprehension of working memory demanding sentences in young and old adults aged between 25 and 65 years.A supplementary analysis did not find a correlation between alpha power and comprehension accuracy, although significant correlations were observed in the theta band.The association between resting state theta power and sentence comprehension is in line with research demonstrating theta band oscillations support working memory (Roux & Uhlhaas, 2014) and semantic retrieval (Bastiaansen et al., 2008), both of which are needed during sentence comprehension (Prystauka & Lewis, 2019).Wang et al. (2022) investigated the relationship between resting state alpha power and the ability to comprehend sentences that were demanding in terms of length and syntactic complexity.Participants in this study were aged around 27 years.Again, no significant correlations between resting state alpha power and comprehension accuracy were found.
To explore the relationship between resting state brain activity and individual differences in comprehension, we may require 'online' measures, that capture moment-to-moment changes in sentence processing demands.Previous research has only correlated resting state alpha power with 'offline' measures like response accuracy, which are obtained postsentence presentation (Beese et al., 2017;Kwok et al., 2019;Wang et al., 2022).Such measures reflect the cumulative cognitive and linguistic processes involved in listening to a sentence and then, formulating an explicit judgement about semantics or grammar (Pliatsikas & Marinis, 2022).Offline measures may not be sensitive enough to the dynamics influenced by alpha activity, as these oscillations primarily affect brain activity during the real-time presentation of sentences, rather than in post-presentation assessments.
Pupillometry has emerged as a useful online measure for assessing sentence comprehension, with increases in pupil size indicating completion of cognitively demanding tasks (Sirois & Brisson, 2014).For example, larger pupil dilation is observed when trying to remember seven digits, compared to three (Beatty, 1982).Modulation of the pupillary response by cognitively demanding tasks is related to neural activity of the locus coeruleus norepinephrine system (Samuels & Szabadi, 2008).The locus coeruleus, which is a structure found in the brainstem, releases the neurotransmitter norepinephrine, which affects oscillatory activity throughout the cortex.Norepinephrine increases the responsivity of cortical neurons, permitting neuronal firing or spiking to occur at lower thresholds (Aston-Jones & Cohen, 2005;Sara, 2009).This can have the effect of facilitating functional connectivity between networks, thereby increasing neural resources available to complete a task.The release of norepinephrine also stimulates postganglionic neurons of the sympathetic nervous system, which in turn releases norepinephrine which bind to receptors present in the iris, causing pupils to dilate (Ferencova et al., 2021).As a consequence of this mechanism, pupil dilation/constriction exhibits a close relationship with the firing rate of the locus coeruleus (Rajkowski et al., 2004).In pupillometry research, increased pupil dilation during a cognitive task is often interpreted as reflecting increased cognitive load or processing demands, suggesting that more neural resources might be engaged in the task (e.g., Borghini & Hazan, 2018;Cabestrero et al., 2009;Fietz et al., 2022;Sirois & Brisson, 2014).
Sentence comprehension also modulates pupil size, in at least two ways (Aydın & Uzun, 2023;Beatty, 1982;Ben-Nun, 1986;Engelhardt et al., 2010;Just & Carpenter, 1993).First, pupil size increases as words in a sentence are processed (Aydın & Uzun, 2023;Piquado et al., 2010;Wendt et al., 2016).That is, the typical finding in the literature is that pupil size is smaller at the start of the sentence compared to the end.The interpretation of this effect is that processing load increases as more words in the sentence are parsed.Second, comprehending syntactically complex sentences is associated with larger pupil size.For example, pupils dilate to a larger extent when comprehending the more 'difficult' object relative clause, compared to the 'easier' subject relative clause (Just & Carpenter, 1993;Piquado et al., 2010).The pupillometric response associated with sentence comprehension appears to be specific to the demands associated with syntactic processing.Indeed, there is evidence to suggest changes in pupil size during comprehension are modulated by syntax, rather than prosody or background noise (Aydın & Uzun, 2023;Wendt et al., 2016).For example, Wendt et al. (2016) examined pupil size as participants listened to easier subject-first sentences (e.g., 'The angry penguin will film the sweet koala') and more difficult object-first sentences (e.g., 'The sweet koala, the angry penguin will film').Significant differences in pupil size only emerged after the initial noun (i.e., 'The angry'/'The sweet') was auditorily presented, at which point subject and object elements of the sentence become defined.This literature demonstrates the sensitivity of pupillometry to real time sentence processing demands.Moreover, pupillometry data from different points in the sentence can be correlated with resting state alpha power.This approach can be used to evaluate whether the relationship between intrinsic brain activity and sentence comprehension might have a time sensitive element.For example, the ability of the brain to generate alpha oscillations might only be observable at points during comprehension that place the greatest demands on processing load.Notably, such an analysis cannot be undertaken using accuracy and reaction time data since they can only be collected after a sentence has been presented.

The current study
This study examined the relationship between resting state alpha power and individual differences in processing demands associated with sentence comprehension.In this research, resting state EEG data were first acquired from 80 adults.A sentence comprehension task was then administered in which participants listened to subject-and objectrelative clauses.Pupillometry data was acquired during this task.The decision to study these sentence types was informed by research showing that not only do they reliably modulate the pupillometric response (Just & Carpenter, 1993), but they are also linked to individual differences in auditory language comprehension among adults (Da ˛browska & Street, 2006).The analyses investigated correlations between resting state alpha power and the pupillometry data.

Method
We report how we determined our sample size, all data exclusions, all inclusion/exclusion criteria, whether inclusion/ exclusion criteria were established prior to data analysis, all manipulations, and all measures in the study.

Participants
The participants in this study were 80 adults (51 females; 28 males), ranging in age from 18.2 to 56.3 years (M ¼ 25.8 years; SD ¼ 7.2 years).Details of sample size determination are outlined in the 'Overview of Data Analysis' section.All participants spoke English as their first language.The sample largely comprised university-educated students.The highest education level of 63 of the participants was either an undergraduate or post-graduate university degree.Almost all participants were right-handed, as assessed by the Edinburgh Handedness Inventory (Oldfield, 1971).This instrument measures handedness on a scale of À100 to 100, where positive scores indicate a right-hand preference and negative scores, a left-hand preference.A positive score was observed for 75 (out of 80) participants.The mean handedness score for the sample was 73.5, with a standard deviation of 47.2 (Range: À100 to 100).All participants provided written consent and received a $30 AUD shopping gift card as reimbursement for their time participating in the study.The research was approved by the Deakin University Human Research Ethics Committee and adhered to the guidelines stipulated by the Declaration of Helsinki.

Sentence comprehension task
The sentence comprehension task aimed to capture the pupillary response associated with listening to grammatically correct, semantically plausible sentences.Participants listened to 110 sentences that were auditorily presented via computer speakers (details on sentence stimuli presented below).After each sentence, participants were asked to indicate whether the sentence's semantics matched a picture shown on a computer display.The semantic judgement served to encourage participants to listen to each sentence and maintain eye contact with the computer display/eyetracker.As participants completed the task, a Tobii 120 eyetracker (Tobii Technology, Sweden) continuously recorded the size of the left and right pupil at a sampling rate of 120Hz.
This apparatus consists of an eye-tracker fixed underneath a 17-inch monitor.This eye-tracker permits pupil and eyemovement data to be acquired without a chin rest.Participants sat comfortably in a chair at a distance of approximately 60 cm from the monitor as they completed the comprehension task.The resolution of the display was set at 1280 Â 1024 pixels.
Each sentence presentation or trial commenced with a central fixation point ('þ' sign) appearing on the computer display for 150 ms.Next, the picture was displayed on the screen for 800 ms, followed by the sentence presented via computer speakers.Each picture's size on the computer display was approximately 12 cm Â 12 cm.Following sentence presentation, participants were asked to indicate whether the sentence's semantics matched the picture using a button box.The button box comprised two horizontally arranged buttons operated with the index and middle fingers from the same hand.A picture of a 'green tick' appeared on the left button and 'red cross' on the right button.The 'green tick' was pressed when the picture was congruent with the sentence's semantics, and the 'red cross' when the semantics were incongruent with the picture.After participants made a response, the screen went blank for 350 ms.Both accuracy and reaction times were recorded as participants made their response.Only reaction times, measured in msec (ms), associated with a correct response were analysed.The presentation order of sentences was randomised.The task was presented to participants using E-Prime 2 (Psychology Software Tools, Pittsburgh).Fig. 1 presents an overview of a single trial.
The test sentences analysed were 30 subject relative clauses and 30 object relative clauses that were semantically congruent with the stimulus picture.The filler sentences were 15 subject-relative clauses and 15 object-relative clauses that were incongruent with the stimulus picture.An additional set of filler items was also presented comprising active-conjoined sentences (e.g., 'The woman is following the man and is holding the flower').In this set, there were equal numbers of items that were congruent (n ¼ 10) and incongruent (n ¼ 10) with the stimulus picture.These filler sentences were included to reduce expectation effects concerning forthcoming syntactic structures.A list of all sentences presented to participants is presented in 'Supplementary Materials A'.
All test and filler sentences were semantically reversible.The test and filler subject and object relative clause sentences comprised, on average, 11.6 words.All object relative clauses adhered to the syntactic structure: for example, 'The boy who the man is pushing is wearing the hat'.The structure for subject relative clauses was [Det NP1 [RelPron V Det NP2] V Det NP3], for example, 'The boy who is pushing the man is wearing the hat''.For both subject and object test sentences, the nouns used in the NP1 or NP2 position were either 'boy', 'girl', 'man', 'woman', 'cat', 'dog', or 'elephant'.The main verb of the relative clause (e.g., 'pushing') used in both sentence types was either 'chasing', 'following' 'hugging', 'pulling', 'pushing', 'touching' or 'watching'.The main verb of the matrix clause (e.g., wearing) for both sentence types was either 'eating', 'holding', 'looking', 'next', 'pointing', 'sitting', 'standing', or 'wearing'.Finally, the object of the preposition was either 'banana', 'bird', 'box', 'cake', 'cat', 'dog', 'flower', 'hat', 'stick', or 'tree'.Each noun and verb appeared with equal frequency in each sentence type.For example, the noun 'boy' appeared in the NP1 position for both subject and relative clauses seven times.Filler sentences were incongruent with the stimulus picture in one of two ways.First, the picture reversed the role of the NP1 and NP2.For example, if the filler sentence was The boy who is pushing the man is wearing the hat, the picture showed the man pushing the boy.Second, the picture incorrectly assigned the attribute of the main clause to the object of the embedded clause; using the same example above, the picture incorrectly showed the man wearing the hat.Finally, we also examined the average word frequency for each sentence.Using the database provided by Gimenes and New (2016), the word frequency (per one million words) was estimated for each word in the sentence.The average individual word frequency was then computed.Word frequency information for all sentences is presented in Supplementary Materials A. The average word frequency for the subject and object test sentences were similar.For subject sentences the average word frequency per sentence was 6393 per million words (SD ¼ 263), for object sentences 6380 (SD ¼ 264).

2.3.
Resting state protocol EEG resting state data were collected using an 'eyes open' protocol with 64 electrodes positioned according to the 10-10 system.Participants were seated comfortably in front of a computer display.After fitting the EEG cap (Twente Medical Systems International, The Netherlands) onto the participant's head, resting state data were acquired for 3 min.During this time, participants were instructed to look at a black 'þ' symbol displayed against a grey background.EEG data were sampled at 2048 Hz with no online filters using a TMSi RefB Amplifier (Twente Medical Systems International, The Netherlands) and a PC computer running Polybench software (Version 1.30; Twente Medical Systems International, The Netherlands).Prior to recording, impedances were reduced to under 10 kU.The EEG data were recorded using a common average reference, with AFz as the ground.

Procedure
Each participant was individually tested in a laboratory setting.After completing a demographic and handedness survey, the resting state protocol was administered followed by the sentence comprehension task.

Pre-processing of pupillometry data
Pre-processing of the pupillometry data first involved averaging left and right pupil sizes at each time point.Time points with missing data from either the left or right eye were not averaged.Next, pupil data were averaged into separate bins or segments.This process is now described.As participants completed the comprehension task triggers were inserted into the continuous recording of pupil data that marked the onset and offset of the: 1. Baseline Segment (i.e., 800 ms period in which only the picture was shown on the computer display).2. Pre-Relative Clause Segment (e.g., 'The boy').3. Relative Clause Segment (e.g., 'who the man is pushing').4. Post-Relative Clause Segment (e.g., 'is wearing a hat').5. Response Period Segment (the time between sentence offset and when a manual response was made).
To control for the influence of luminance on the pupillary response the pre-relative clause, relative clause, post-relative clause, and response period segments were all baseline corrected using data from the baseline segment.This involved subtracting the average pupil size of the baseline segment from all subsequent sentence segments (e.g., pre-relative clause segment e baseline segment).For each participant, averaged baseline corrected pupil size from the pre-relative clause, relative clause, post-relative clause, and response c o r t e x 1 7 7 ( 2 0 2 4 ) 1 e1 4 period segments were then averaged separately for objectand subject-relative clause test sentences.Data from each of these segments were correlated with resting state alpha power (details of EEG pre-processing are presented below).Only sentences associated with a correct response were analysed.Pre-processing of pupillometry data was performed using Visual Basic scripts implemented in Microsoft Excel.

2.6.
Pre-processing of EEG data EEG data were pre-processed using custom scripts and EEGLAB (version #2023.0;Delorme & Makeig, 2004) run in MATLAB (version #2023A).Data were first-down sampled to 250Hz to reduce processing time and then bandpass filtered at .25e80Hz.Oculomotor, line noise and muscle and other artifacts were removed using the 'Reduction of Electroencephalographic Artifacts (RELAX)' pre-processing pipeline using default settings (Bailey, Biabani, et al., 2023;Bailey, Hill, et al., 2023).Next, data from the left and right mastoids were removed.Time-series data for all remaining 62 electrodes were converted to the frequency domain using Welch's Method (2-s window with a Hann Taper).The amplitude at each frequency point was computed in terms of mV 2 or power.
In the following step, the aperiodic signal was subtracted from the power spectrum using the FOOOF toolbox (Donoghue et al., 2020) run in Python 3.10 via a MATLAB wrapper.The aperiodic component of the power spectrum can be described in terms of the offset and aperiodic slope.The 'offset' describes the baseline level or amplitude of the aperiodic activity, providing an indication of the general power level without considering the frequency content.Conversely, the 'slope' (exponent) quantifies how the power of aperiodic activity decreases as frequency increases, essentially detailing its spectral slope.The role of aperiodic activity in sensorimotor and cognitive functioning is an active area of research (e.g., Cross et al., 2022;Hill et al., 2022;Merkin et al., 2023;Thuwal et al., 2021).However, it has also been demonstrated that failure to remove the aperiodic component from the power spectrum exaggerates the relationship between resting state alpha power and cognitive functioning (Ouyang et al., 2020).
The effects of removing aperiodic activity from the power spectrum are depicted in Fig. 2. Panel A and B in this figure show the power spectrum averaged across all channels and participants before and after removing aperiodic activity respectively.Finally, for each participant and channel, spectral power was averaged between 8 and 13 Hz, which is commonly used to study alpha oscillations (see Panel B).Panel C shows the distribution of alpha power, averaged over participants, at each electrode after removing the aperiodic signal.

Overview of Data Analysis
An exploratory analytic approach was employed to examine the correlation between pupillometry and resting state alpha power.This approach was deemed necessary in the absence of past research to generate predictions about which electrodes, sentence type (subject vs object) or segment within each sentence a correlation, along with its direction, would be observed.Spearman's r was first used to compute the correlation between pupil data and resting state alpha power.Correlations were calculated separately for each electrode (n ¼ 62), sentence segment (n ¼ 4; pre-relative clause, relative clause, post-relative clause, response period) and sentence type (n ¼ 2: subject, object) resulting in a total of 496 (62 Â 4 x 2) Spearman's r and pvalues.To correct for multiple comparisons, an FDR correction (Benjamini & Hochberg, 1995) was applied to the uncorrected pvalues to ensure alpha was maintained at .05 (two tails).This correction was implemented using a MATLAB function developed by Groppe (2023).A sample size of 80 was selected for this study since it provides 80% power to detect correlations as low as .3,given the number of tests being corrected (n ¼ 496) and assuming 15% of the tests are truly significant (Izmirlian, 2020).
The concern with a larger sample is that potentially trivial correlations (arbitrability set at r < .3),may be found to be significantly different from zero.All other statistical analysis were conducted using JASP (Version .17.3; JASP Team., 2023).
The MATLAB script used to pre-process, analyse the data, generate figures along with EEG and pupillometry datasets can be accessed via the OSF site (https://osf.io/5q48r/)that accompanies this work.The raw data on the OSF site has been collated into single MATLAB and csv files.The original files could not be uploaded owing to the conditions of our ethics approval.No part of the study analyses or procedures were preregistered prior to the research being conducted.The E-Prime code used to run the comprehension task is also available from the OSF site, however, the audio and visual stimuli could not be uploaded since the individuals who generated this content did not provide consent for us to distribute their work.
The conditions of our ethics approval do not permit public archiving of anonymised study data.Readers seeking access to the data should contact the lead author.Access will be granted to named individuals in accordance with ethical procedures governing the reuse of sensitive data.Specifically, requestors must complete formal data sharing agreement to obtain the data.In addition, legal copyright restrictions do not permit us to publicly archive the full set of stimuli used in this experiment.Readers seeking access to the stimuli are advised to contact the lead author.Stimuli will be released on completion of a formal collaboration agreement.

Preliminary analyses of behavioural and pupillometry data
Initial analyses tested for significant differences between subject-and object-relative clause sentences in terms of accuracy, reaction time, and baseline-corrected pupil size.These tests acted as a verification to ensure the comprehension task replicated the typical finding that object relative clauses tend to be more difficult to understand compared to subject relative clauses.Overall, the proportion of correct responses (i.e., indicating the sentence matched the stimulus picture) for both subject and object sentences approached ceiling.However, the mean proportion of correct responses for object sentences (M ¼ .95;SD ¼ .09;Range: .70,1.0) was found to be significantly lower compared to subject sentences (M ¼ .97;SD ¼ .04;Range: .83,1.0) when tested using a Wilcoxon signed-rank test (z ¼ 2.447, p ¼ .013).Mean reaction times (in ms) were significantly slower for object sentences (M ¼ 625; SD ¼ 321) compared to subject sentences (M ¼ 551, SD ¼ 280), again when tested using a Wilcoxon signed-rank test (z ¼ 2.715, p ¼ .007).For completeness, we also present summary statistics for reaction times associated with incorrect responses.The mean RT associated with incorrect responses for subject sentences was 1193 ms (SD ¼ 1100) and object sentences 1413 (SD ¼ 1057).It should be noted that not all participants are represented in the incorrect response reaction time data since not every participant made errors on the task.Fig. 3 shows baseline corrected pupil sizes reported by sentence type (i.e., subject vs object) and sentence segment (i.e., pre-relative clause, relative clause, post-relative clause, and response period).Panel A in the figure shows group level data and Panel B subject level data.
Post-hoc tests employed Bonferroni corrections for multiple comparisons that compared all pairwise combinations of means computed across and within the two sentence types (i.e., a total 28 comparisons).Results from all pairwise comparisons of means are presented in Table 1.Object relative clause sentences were associated with larger pupil size, compared to subject sentences during the relative clause and post-relative clause segment (p's < .001).The difference in pupil size between subject and object sentences at the prerelative clause and relative clause segments was not significant (p's ¼ .999).In terms of the change in pupil size across sentence segments, for both sentence types, pupil sizes significantly increased from pre-relative clause to relative clause (p's < .001).There was also a significant increase in pupil size from the relative clause to post relative clause segment for both sentence types (p's < .001).A further significant increase in pupil size was observed for both sentences from post-relative clause to response period (p's < .001).In sum, the preliminary analyses replicated past research.For object sentences, relative to subject, accuracy was lower, reaction times were slower, and pupil size larger.

Correlation between resting state alpha power and the pupillary response
Correlations between resting state alpha power and pupillometry data are now presented.The results from these analyses are summarised in Fig. 4.This figure shows topographic head plots of the correlation (Spearman's r) between resting state alpha power and baseline corrected pupil size at each sentence segment.Panel A of Fig. 4 shows correlations reported by sentence type (subject sentences, object sentences) and by sentence segment (pre-relative clause, relative clause, post relative clause, response period).Electrodes highlighted in green were found to be statistically significant after applying the FDR correction.Results from the subject relative clause sentences revealed a significant positive association between alpha power and pupil size, but only during the postrelative clause segment.The correlations were predominantly significant at occipital electrodes.For illustrative purposes, Panel B in Fig. 4 shows a scatterplot of pupil sizes from the post-relative clause segment of the sentence and alpha power, averaged over significant electrodes.In this figure Spearman's r ¼ .326(p ¼ .003).
Results pertaining to the object relative clause sentences reveal positive significant correlations between pupil size and c o r t e x 1 7 7 ( 2 0 2 4 ) 1 e1 4 resting state alpha at the relative clause and post-relative clause segments.During the relative clause segment, correlations were primarily significant at central parietal electrodes.During the post-relative clause segment, significant electrodes were found across the scalp.Panels C and D of Fig. 4 show scatterplots of resting state power averaged over significant electrodes and pupil size for the relative clause and post-relative clause segments, respectively.In Panel C, the correlation was r ¼ .353(p ¼ .001)and in Panel D the correlation was r ¼ .354(p ¼ .001).

Exploratory analyses
Five sets of exploratory analyses were undertaken that examined the relationship between: (1) pupillometry data and neighbouring frequency bands (i.e., theta and beta), (2) resting state alpha power and the behavioural measures (i.e., accuracy and reaction times), (3) pupillometry data and behavioural measures (i.e., accuracy and reaction times), (4) baseline pupil size and resting state alpha power, and (5) handedness with both resting state alpha power and pupil size.
The results of all exploratory analyses are presented in 'Supplementary Materials B'.A summary of these analyses is presented in this section.First, using the same statistical approach adopted for the alpha band, no significant correlations were observed for either sentence type at any electrode or sentence segment in the theta (4e7 Hz) or beta (14e30 Hz) bands (see Figs. S1 and S2 in Supplementary Materials B).Second, resting state alpha power was not significantly correlated with reaction times or accuracy for either sentence type or electrode (see Fig. S3 in Supplementary Materials B).Third, pupil size was not significantly correlated with accuracy or reaction times (see Table S1 in Supplementary Materials B).Fourth, we also tested whether the average pupil size during the baseline period of each sentence correlated with resting state alpha power.Baseline level pupil size captures general arousal level (Lloyd et al., 2023).It is possible the significant correlations presented in Fig. 4 reflect an association between resting state alpha power and general arousal levels, rather than language comprehension, given all pupil data were baseline corrected.To test this possibility, correlations between resting state alpha power and baseline pupil size were examined.Correlations between these variables were tested at each electrode.None of the correlations were found to be significant, even before applying an FDR correction (see Fig. S4 in Supplementary Materials B).Finally, correlations  between handedness and the data were examined.This analysis tested whether the handedness of the participants may have influenced the results, since both left-and righthanded adults participated in the study.To test for potential effect of handedness on the results, we examined the correlation between handedness, as quantified by the Edinburgh Handedness Inventory, with both resting state alpha power and pupil size measured during the sentence comprehension task.None of the correlations were found to be significant, even before applying the FDR correction (see Fig. S5 & Table S2 in Supplementary Materials B).

Discussion
This study investigated the relationship between individual differences in auditory language comprehension and resting state alpha power.Using pupillometry, we captured real-time processing load dynamics as participants listened to subject and object relative clauses.The analyses revealed resting state alpha power positively correlated with pupil size.For subject relative clause sentences, significant correlations between alpha power and pupil size were only observed during the post-relative clause segment.For object sentences, significant correlations between alpha power and pupil size were observed in both the relative clause and post-relative clause segments.Exploratory analyses did not indicate these associations were related to baseline arousal level or handedness.Also, the effects were specific to alpha band activity; resting state theta and beta band power were not found to correlate with pupil size.These findings provide new evidence linking the brain's functional architecture to individual differences in auditory language comprehension.Object sentences demonstrated the most pronounced association between intrinsic alpha activity and online sentence processing demands.For these sentences, the association became observable when participants encountered the noncanonical syntactic frame.As noted earlier, in a number of languages, object relative clause sentences tend to be more difficult to comprehend, when compared to subject relative clauses that follow the canonical subject-verb-object structure (Gordon et al., 2001;King & Just, 1991;Staub, 2010;Traxler et al., 2002;Ueno & Garnsey, 2008).Replicating these past findings, we found a similar trend in our data.The preliminary analyses showed the object sentences were associated with lower accuracy, slower reaction times and larger pupil size (e.g., Just & Carpenter, 1993;Traxler et al., 2002).During the relative clause segment for object sentences, we note that significant correlations were largely observed at parietal electrodes.Alpha activity recorded at parietal electrodes reflects neural processes that protect information held in working memory (Jensen et al., 2002;Meyer et al., 2013).This mechanism may be especially important in comprehending object relative clauses, which disrupt the standard subjectverb-object sequence and necessitate prolonged retention of sentence elements until agent-patient relationships can be defined (Gibson, 1998).The significant correlations observed during the relative clause segment may suggest a link c o r t e x 1 7 7 ( 2 0 2 4 ) 1 e1 4 between individual differences in this memory protection mechanism and the brain's capacity to generate a robust alpha rhythm, potentially influencing processing load.
Significant correlations observed at the post-relative clause segment from object sentences link resting state alpha oscillations to a more extensive role regulating cortical activity.At the post-relative clause segment, the information provided in the relative clause is integrated, in real-time, with the main clause to understand the full meaning of the sentence.Additionally, in the current study, participants also needed to evaluate whether the auditory information matched a visual stimulus.This likely requires coordination of communication between functional networks across the cortex that support both language and visual processing.Also, networks not relevant to the task need to be inhibited to avoid disrupting task relevant processes.This may explain the widespread positive correlations observed across the scalp during the post-relative clause segment for object sentences, potentially indicative of the heightened neural resources required at this stage.This proposal aligns with findings from fMRI studies on language comprehension.A meta-analysis of this literature by Walenski et al. (2019) revealed comprehension activates the left prefrontal cortex along with left and right temporal lobes.Also, the absence of global cortical activation revealed in this review highlights the selective engagement of language areas.Based on our data, resting state alpha power appears to correlate with the regulation of cortical activity, which could underpin the observed variability in processing load during comprehension of object sentences.
Interpreting the pattern of significant and non-significant correlations from subject sentences requires caution.We found no significant correlations at the relative clause segment for these sentences.However, at the post-relative clause segment, significant correlations mainly appeared at occipital electrodes.This might suggest a difference in the emergence and distribution of correlations between the subject and object sentences.Yet, a closer look at our data contradicts this interpretation.Panel A of Fig. 4 shows that both sentence types exhibit a similar distribution of correlations across segments and electrodes, albeit generally weaker for subject sentences.This could be because subject sentences are more predictable, making them less cognitively demanding compared to object sentences and thus, less regulated by alpha oscillations.Statistically, this would result in smaller correlation coefficients, unlikely to be significant post-FDR correction.An ad-hoc analysis, presented in Fig. 5 below, investigated this by examining the distribution of significant correlations for the subject sentences, at each sentence segment, without applying the FDR correction.Here, electrodes marked in green indicate significant correlations, based on uncorrected p-values.In this figure the distribution of significant electrodes is strikingly similar to the results observed for object sentences (see Fig. 4).Based on this result, we suggest the relationship between intrinsic alpha oscillations and processing load is fundamentally similar for subject and object sentences.The effect, however, appears to be more pronounced in object sentences than subject sentences, likely due to their higher cognitive demands.
We acknowledge resting state activity did not account for all the variability on the comprehension task.Consistent with previous research (Beese et al., 2017;Wang et al., 2022), no significant correlations were found between resting state alpha power and measures of accuracy or reaction times.Also, correlations involving the pupillometric response during the pre-relative clause and response segments were not significant.These findings suggest a reduced impact of intrinsic alpha activity during offline aspects of sentence analysis and in the initial stages of sentence comprehension, such as when processing a single noun phrase.These nonsignificant correlations could highlight instances where resting state alpha power plays a less dominant role in sentence comprehension.
Finally, further research is needed to discern whether higher or lower resting state alpha power indicates proficient language comprehension.The positive correlations we found might suggest that higher resting state alpha power indicates a neural system that demands more cognitive effort for language comprehension.From this perspective, one could hypothesise that individuals who have difficulties understanding spoken language would exhibit higher resting state alpha power.This interpretation, however, contrasts with prior research indicating that higher levels of resting state alpha power correlate with superior cognitive performance (Doppelmayr et al., 2002;Grandy et al., 2013;Prat et al., 2016;Vogt et al., 1998).The use of online and offline measures of individual differences could help reconcile these findings.Strictly speaking, the online pupillometry data used in the current study does not directly address performance metrics like accuracy or processing speed.Additionally, the absence of correlations between our behavioural and pupillometry data makes it challenging to delineate their relationship.It is also worth noting that participant accuracy in our study approached ceiling.Hence, our findings highlight a relationship between resting state alpha power and pupillometry predominantly during successful comprehension.The positive correlations we observed might be indicating individuals with higher resting state alpha power were able to allocate more neural resources to accurately comprehend the Fig. 5 e Results from ad-hoc analyses examining correlations between resting state alpha power and pupil size for subject sentences.Electrodes highlighted in green are significant, without applying an FDR correction.
sentences.Given this context, if we aim to assess whether resting state alpha power can be a reliable marker for language processing impairments, examining resting state brain activity and language comprehension in individuals with and without language disorders might overcome this limitation.

Conclusion
This study investigated the neural basis of individual differences in language comprehension.The results revealed individual differences in sentence processing demands correlate with resting state alpha power.This association, however, appears to be strengthened when comprehending language that is syntactically complex.Collectively, our results demonstrate that the brain's intrinsic functional architecture correlates with individual differences in adult auditory language comprehension.Additionally, the present study paves the way for additional targeted investigations involving populations with language disorders to determine the extent to which the brain's ability to generate a robust alpha rhythm may be indicative of suboptimal language processing.

Fig. 1 e
Fig. 1 e Summary of a single trial from the comprehension task.

Fig. 2 e
Fig. 2 e Resting state power spectrum.Panels A & B shows power spectra before and after removing aperiodic activity, respectively.The highlighted region in Panel B denotes alpha band range (8e13 Hz).Panel C shows grand average 8e13 Hz power plotted across the scalp.

Fig. 3 e
Fig. 3 e Baseline corrected pupil size reported by subject-and object-relative clauses and sentence segment.Panel A shows summary group data and Panel B subject-level data.

Fig. 4 e
Fig. 4 e Correlation (Spearman's r) between resting state alpha power and baseline adjusted pupil sizes.Panel A shows head plots of correlations reported by electrode, sentence type and sentence segment.Electrodes where the correlation is significant are highlighted in green.Panels B e D, show scatter plots of resting state alpha power (averaged over significant electrodes) and baseline corrected pupil sizes at the post-relative clause segment (subject sentences; Panel B), relative clause segment (object sentences; Panel C) and post-relative clause segment (object sentences; Panel D).

Table 1 e
Bonferroni Corrected p-values (correcting for 28 comparisons).Comparing Mean Difference in Pupil Size Between Sentence Segments for Subject and Object Relative Clauses.