ABSTRACT
Listeners' brains align when they attend to the same speech, but what determines how strongly they align is not well understood. We asked whether that alignment tracks the communicative relation between a listener and a message, manipulating it from the speaker's side and from the listener's, and bounding the account with a manipulation that alters the social setting while leaving the relation intact. In two magnetoencephalography experiments, we measured inter‐subject correlation of band‐limited envelopes, which resolves shared processing by frequency rather than treating it as a single quantity. In Study 1, 40 listeners watched the same actor deliver the same speech charismatically and non‐charismatically, so that content and identity were held constant. Charismatic delivery raised correlation most strongly in the beta band, and after regression of the acoustic and visual features, the surviving effect was concentrated in right perisylvian cortex. In Study 2, listeners read a favorable or unfavorable description of a speaker they had never encountered and watched a second speaker either with or without a minimal cue that an audience was present. A newly formed attitude raised correlation in alpha and theta within the same right perisylvian region, whereas the audience cue, which left both the message and the listener's disposition unchanged, produced no detectable change in any band. Correlation also proved partly dispositional: how strongly a listener synchronized on one task predicted their synchrony on the other, specifically in alpha and independently of personality. Delivery and attitude thus converge on a common substrate for the speaker's expressive signal, reaching it through separate frequency channels.
Short abstract
Listeners' brains align when they attend to the same speech, but what determines how strongly is unclear. Resolving alignment by frequency, we find that a speaker's expressive delivery and a listener's newly formed attitude converge on the same right‐perisylvian cortex through separate frequency channels, whereas a cue altering only the social setting leaves it unchanged. Shared processing therefore tracks the listener's relation to the message, not social context generally.
1. Introduction
Human communication is a social process in which people continually adjust to one another. Listeners do not simply decode information; they develop shared attention and expectations, a process described for dialogue as interactive alignment (Pickering and Garrod 2004). At the neural level this coordination is manifested as similar cortical responses in people who engage with the same message, and during successful communication the listener's activity becomes coupled to the speaker's (Stephens et al. 2010; see Wheatley et al. 2012, for a review). Speakers help guide the process through prosody and co‐speech gesture, which organize listeners' attention and support shared representational states (Hasson et al. 2012; Holle et al. 2010; Marstaller and Burianová 2014; Redcay et al. 2016). Speaker and audience factors, such as personality, identity, leadership style and prior dispositions, also shape how similarly people interpret a message (Berson and Avolio 2004; Finn et al. 2020; Nummenmaa et al. 2018; van Baar et al. 2021). These influences are rarely separable outside the laboratory. A listener watching a recorded lecture brings an opinion of the speaker formed before a word is spoken, hears a delivery designed to hold attention, and knows that others are watching alongside them. Whether shared processing tracks all three, or only those that bear on the listener's relation to the message, is the question addressed here.
One approach for quantifying shared processing is inter‐subject correlation, which measures the similarity of neural time courses while people listen to or view the same material (for a methodological treatment, see Nastase et al. 2019). Early fMRI studies showed that it captures shared perceptual and narrative processing during naturalistic viewing and listening (Hasson et al. 2004, 2008; Honey et al. 2012; Wilson et al. 2008), and later work showed it to be sensitive to the temporal structure of a narrative (Lerner et al. 2011), to the emotional signals that accompany it (Dmochowski et al. 2012; Golland et al. 2017), and to the listener's attentional engagement (Ki et al. 2016). During spoken communication, listeners' activity becomes time‐locked not only to the auditory input but to the speaker's unfolding message, reflecting convergence in comprehension and internal modeling (Stephens et al. 2010). Much of this foundational work used fMRI, but MEG and EEG offer a complement that matters for speech: their millisecond resolution follows the dynamic structure of a message as it unfolds, and MEG adds relatively precise source localization. Both have revealed consistent correlation during films (Lankinen et al. 2014, 2018), televised and narrated content (Cohen and Parra 2016; Dmochowski et al. 2014), and continuous natural speech (Thiede et al. 2020). The measure is therefore well suited to naturalistic speech, and in MEG and EEG it need not be treated as a single quantity: it can be resolved by frequency.
MEG and EEG make it possible to examine ISC within specific oscillatory frequency bands, each of which has been linked to different roles in speech perception and the alignment process. Delta rhythms (1–4 Hz) are thought to track the prosodic structure of speech, marking phrase and sentence boundaries and providing a temporal scaffold for following intonational and rhetorical flow (Chalas et al. 2024; Kotz and Schwartze 2010; Oderbolz et al. 2024), although delta activity in this range is not exclusively prosodic and also reflects the segmentation of multi‐word units (Chalas et al. 2024). Theta activity (4–8 Hz) is entrained to the speech envelope at the syllabic rate and supports segmentation (Ding and Simon 2014; Luo and Poeppel 2007; Obleser and Kayser 2019), and has further been implicated in lexical retrieval and the encoding of incoming linguistic material into memory (Bastiaansen and Hagoort 2003). Alpha‐band activity (8–12 Hz) typically decreases in regions engaged in processing important semantic or syntactic information, a pattern associated with selective attention and the suppression of irrelevant input (Foxe and Snyder 2011; Klimesch 2012). Beta rhythms (12–25 Hz) have been linked to top‐down predictions about upcoming linguistic material, and more generally to the maintenance of the current sensorimotor and cognitive set (Arnal and Giraud 2012; Engel and Fries 2010; Lewis and Bastiaansen 2015). Gamma‐band activity (25–40 Hz) supports rapid perceptual and semantic integration, facilitating the binding of phonemic features and the matching of top‐down predictions against incoming lexical and conceptual information (Fries 2015; Giraud and Poeppel 2012; Lewis and Bastiaansen 2015). Examining ISC in these frequency ranges therefore indicates which processes become shared across listeners, and not only that sharing occurs.
Behavioral and social‐psychological research identifies three factors that plausibly govern how similarly listeners take up the same message. Expressive delivery, characterized by dynamic prosody, gesture and clear rhetorical structure, increases engagement and social influence (Antonakis et al. 2011; Niebuhr et al. 2017). Listener factors, such as an attitude toward the communicator held before the message begins, bias attention, evaluation and interpretation even when the content is identical (Hughes and Zaki 2015; Van Strien et al. 2016). Social context has its own effects: cues that suggest a shared audience or collective reception can alter motivation and perceived relevance, influencing how people adopt shared perspectives (Shteynberg 2015; Shteynberg et al. 2016). Whether these behavioral influences carry through to shared neural processing, and if so which of them, is not established. Two of them act on the listener's relation to the message itself, one from the speaker's side and one from the listener's. The third leaves that relation untouched and alters only the circumstances of reception, which is what makes it useful as a test.
The aim of the present research was to determine whether shared processing during naturalistic speech tracks the listener's relation to the message specifically, and if so, through which oscillatory channels. Two of these influences were manipulated directly and the third was used to bound the account, across two MEG experiments (Figure 1). The first varied the speaker's delivery style, charismatic against non‐charismatic. The second manipulated the listener's attitude toward the speaker and, in a crossed design, added a minimal cue that the speech was being delivered to an audience. Delivery and attitude altered what the message offered and what the listener brought to it; the audience cue altered neither and therefore tested whether shared processing tracks the communicative relation specifically or responds to social framing of any kind. Because both manipulations reached the listener through the speaker's prosodic signal, we expected any convergence between them to fall in right auditory‐perisylvian cortex, the region associated with the analysis of the voice and its paralinguistic properties (Belin et al. 2000; Belyk and Brown 2014; van Rijn et al. 2005). A region of interest was defined anatomically on that basis and applied identically to both experiments; it was motivated by theory and by the prior literature and was not preregistered. Participants viewed ecologically valid clips while inter‐subject correlation was computed across the delta, theta, alpha, beta, and gamma bands. Beyond asking whether each factor modulated shared processing, the combined design asked whether speaker‐ and listener‐side influences converge on a common cortical substrate or dissociate, and through which frequency channels, questions that no single manipulation can answer. We treated the canonical bands as candidate channels rather than committed mechanisms.
FIGURE 1.
Experimental designs. Study 1: a within‐subjects comparison of charismatic versus non‐charismatic delivery of the same speech by the same speaker, with a silent control. Study 2: a single cohort underwent two crossed, counterbalanced manipulations in one session—listener attitude (a positive or negative description of an unfamiliar speaker read before an identical lecture) and social context (a second, different lecture presented with or without a minimal visual audience cue).
2. General Methods
Both experiments were run on the same system, with the same preprocessing, the same measure of inter‐subject correlation and the same approach to inference, and those procedures are described once here. Study 1 compared the same actor delivering the same speech charismatically and non‐charismatically, within participants; Study 2 crossed a manipulation of listeners' attitude toward a speaker with a manipulation of the apparent presence of an audience, between participants (Figure 1). Participants, stimuli, designs and the study‐specific statistical contrasts are given with each experiment, in Sections 3.2 and 4.2.
2.1. Ethics Statement
All procedures in these studies were approved by the Ethics Committee of the Department of Psychology, Bar‐Ilan University (Approval No. 04/18). All participants provided informed consent prior to participation, in accordance with the Declaration of Helsinki.
2.2. MEG Acquisition
MEG data were recorded using a whole‐head 248‐channel magnetometer array (4‐D Neuroimaging, Magnes 3600 WH) at a sampling rate of 1017.25 Hz with an online bandpass filter of 1–400 Hz. Environmental noise reduction was achieved using reference coils. Participants were monitored by video and intercom throughout the recording.
2.3. Preprocessing
MEG preprocessing and source reconstruction were performed in MATLAB R2020b with the FieldTrip toolbox (Oostenveld et al. 2011). Line and building vibration noise and heartbeat artifacts were removed with a cancellation algorithm (Tal and Abeles 2013). Ocular artifacts were then removed by an objective, rule‐based independent component analysis (infomax, as implemented in EEGLAB and distributed with FieldTrip; Delorme and Makeig 2004), computed once per participant from all conditions concatenated and applied to all, with no manual component selection; components were classified as ocular by a fixed conjunction of topographic frontality, low‐frequency power and temporal kurtosis, removing a mean of 1.6 components per participant (full criteria in Supporting Information S5). All subsequent analyses (inter‐subject correlation, statistical testing and permutation nulls, low‐level feature extraction and regression, and the separation of periodic from aperiodic activity) were performed in Python 3.13 using NumPy, SciPy, librosa, FFmpeg and specparam/FOOOF (Donoghue et al. 2020). Word onsets were obtained by Hebrew automatic speech recognition followed by forced alignment with a wav2vec 2.0 CTC model (Baevski et al. 2020). Analysis code was partly written with the assistance of Claude Code (Anthropic; Claude Opus 4.8). The authors specified every analysis, verified all reported values against the data and, where a previous implementation existed, against its output, and take full responsibility for the analyses and their reporting.
2.4. Inter‐Subject Correlation Pipeline
Following Thiede et al. (2020), frequency envelopes were obtained by bandpass filtering the preprocessed data into five ranges: delta (1–3.9 Hz), theta (4–7.9 Hz), alpha (8–11.9 Hz), beta (12–24.9 Hz), and gamma (25–40 Hz). Voxel‐wise activity in each band was estimated using linearly constrained minimum variance (LCMV) beamforming. For each band, the spatial filter was computed from a data covariance matrix estimated on the band‐passed data for that band, calculated over the entire stimulus period and regularized at 10% of its trace. This yielded time series for 1169 voxels per participant across the five bands. Voxel activity was averaged within 72 cortical regions of the AAL(v1) atlas (Tzourio‐Mazoyer et al. 2002), Hilbert transformed, low‐pass filtered at 0.8 Hz, and down‐sampled to 100 Hz to produce the final ISC time courses (Thiede et al. 2020).
ISC was computed with a leave‐one‐out approach: for each participant, the regional time course was correlated with the averaged signal of the remaining participants in their condition, and coefficients were Fisher‐z transformed. Within‐condition significance was evaluated with a permutation null of circular time‐shifts of the averaged time series (1400 iterations), computed separately for each study and condition. The 95th percentile of the null was below r = 0.020 in Study 2 and below r = 0.033 in Study 1, whose clips are shorter and whose null is correspondingly wider; ISC values exceeding the relevant threshold were considered significant.
2.5. Inferential Approach
Differences in ISC and power between conditions were tested at the level of the experimental unit (participant) to avoid the statistical non‐independence of pairwise inter‐subject correlations (Chen et al. 2017; Nastase et al. 2019). Each participant contributed a single leave‐one‐out ISC value per region and band, and global effects were assessed on the 72‐region average. Regional effects were FDR‐corrected across the 72 regions within each band, and bands were treated as separate a priori hypotheses. Contrasts between groups were evaluated against a null formed by permuting group labels (10,000 permutations), which requires no distributional assumption; within‐subject contrasts were evaluated as described in Section 3.2.5. In addition, low‐level acoustic and visual features [amplitude envelope and its derivative, syllable rate (peakRate; Oganian and Chang 2019), fundamental frequency, four log‐mel spectrogram components, and visual motion and contrast] were reduced to their leading principal components and regressed out of the regional time courses before recomputing ISC, with a phase‐scrambled‐feature null; word onsets from forced alignment and a richer visual‐motion model were tested as additional regressors. Oscillatory activity was separated from aperiodic activity using spectral parameterization (specparam/FOOOF; Donoghue et al. 2020). Full extraction parameters are given in Supporting Information S2, and these controls are reported where relevant.
2.6. Right Auditory‐Perisylvian Region of Interest
As set out in the Introduction, a region of interest was defined anatomically before the region‐level analyses were examined, comprising four right‐hemisphere parcels of the AAL atlas: the rolandic operculum, the superior temporal gyrus, the middle temporal gyrus, and the insula. The temporal parcels carry the voice‐selective responses of the superior temporal sulcus (Belin et al. 2000) and the audio‐vocal network for affective prosody identified meta‐analytically (Belyk and Brown 2014); the operculum was included on causal grounds, repetitive TMS over the right fronto‐parietal operculum impairing discrimination of emotional prosody while leaving emotional semantics intact (Hoekert et al. 2008; van Rijn et al. 2005). The set corresponds to the temporal and opercular terminations of the right‐hemisphere dorsal and ventral prosody pathways identified by combined functional imaging and tractography (Sammler et al. 2015). The set was restricted to this auditory‐perisylvian core, and frontal nodes of the prosody network were excluded and are reported only in the whole‐brain analyses.
The same four parcels were applied to both experiments, so that neither study supplied the other's regions, and two properties of the set bear on its independence from the results. The insula was retained although the insular effect in Study 1 does not survive the low‐level and aperiodic controls, and the occipital and cingulate regions that do survive those controls in Study 1 were excluded. The composition of the region therefore runs against the observed maps rather than with them.
For each contrast, the per‐region effect map was averaged across the four parcels and located within the distribution of the same quantity over all 1,028,790 four‐parcel sets that can be formed from the 72 regions, enumerated exhaustively so that p is exact. The null is thus constructed over regions rather than over participants: the test asks whether an effect is more concentrated in this region than elsewhere in the brain, conditional on the observed map, and it does not test the region‐averaged effect against zero. Where a comparison against zero was required, the leave‐one‐out correlations were instead averaged across the four parcels to give one value per participant per band, and the group‐label permutation of Section 2.5 was applied. The two tests answer different questions and are reported separately throughout. Supporting Information S6 reports the full analysis, including a control in which the reference sets are restricted to spatially contiguous parcels.
3. Study 1: Delivery Style
3.1. Rationale and Predictions
Delivery style is a central aspect of spoken communication, shaping how audiences attend to a message, how they interpret it, and whether they revise their attitudes and behavior in response. Behavioral work shows that charismatic and expressive delivery can heighten attention, emotional involvement, and perceived credibility (Antonakis et al. 2011; Niebuhr et al. 2017), thereby increasing individual and collective performance (Banks et al. 2017; Bass et al. 2003). Neural evidence points in the same direction: ISC studies have reported that vivid, high‐engagement speeches produce stronger alignment across listeners, particularly in regions associated with attention, narrative processing, and social cognition (Imhof et al. 2017; Schmälzle et al. 2015), and naturalistic neuroimaging has shown that expressive cues in prosody and co‐speech gesture enhance multisensory and semantic processing (Holle et al. 2010; Marstaller and Burianová 2014). Delivery features can thus support more coordinated neural responses during communication.
However, many prior studies have compared speeches that differed not only in delivery but also in content, speaker identity, emotional tone, or the perceived importance of well‐known public figures. These factors make it difficult to isolate the specific influence of delivery style on the similarity of brain activity. To address this gap, we created two versions of the same speech, performed by the same unfamiliar actor, differing only in expressiveness: one charismatic and one intentionally flat and unexpressive. This design controls for content and identity, allowing delivery style to be examined as a causal factor in shaping shared neural processing during naturalistic communication.
The aim of Study 1 was to establish whether delivery alone modulates inter‐subject correlation, and in which frequency bands and regions. Given the functional roles sketched above, and because an expressive delivery differs from a plain one chiefly in what it offers to attention and prediction rather than in what it asserts, we expected the effect in the bands associated with attention, prediction and semantic integration, most notably alpha, beta and gamma, and not in the lower bands that track prosodic grouping and syllabic structure. We further anticipated that these effects would be strongest in cortical regions involved in speech comprehension (Hickok and Poeppel 2007), in the analysis of social and communicative signals in the superior temporal sulcus (Deen et al. 2015), and in the attribution of communicative intent (Redcay et al. 2016), including temporal, parietal and inferior frontal areas.
3.2. Method
3.2.1. Participants
Forty healthy undergraduate students from Bar‐Ilan University took part in the study (19 women and 21 men; age 21–32 years, M = 25.15, SD = 2.48). All participants were right‐handed according to the Edinburgh Handedness Inventory (Oldfield 1971), native Hebrew speakers, and reported normal or corrected‐to‐normal vision. Eligibility was assessed by questionnaire before the session. Respondents reporting metal implants or dental work likely to produce magnetic artifact or a neurological diagnosis were not invited to participate. Participants received either monetary compensation or course credit.
3.2.2. Stimuli
Participants viewed two ~3‐min video speeches performed by the same unfamiliar professional actor. In both videos, the actor presented his vision for the Bar‐Ilan University Student Council. The speech content was based on a preliminary survey in which 40 students identified the most prominent issues in their university experience, and these themes were integrated into both versions. The verbal content was identical, and only the delivery style differed. In the charismatic version, the actor used expressive prosody, gestures, and dynamic body language. In the non‐charismatic version, the delivery was deliberately flat, restrained, and minimally expressive. Both videos were filmed from a fixed camera angle with identical framing, lighting, and background, and were projected onto a 17‐in. screen located 60 cm above the participant's head (~30°), with a photosensitive diode marking stimulus onset and offset. Both clips were loudness‐normalized for playback. Characterized at matched presentation level, they differed principally in pitch dynamics and in visual motion energy. The charismatic version spanned an f0 interquartile range 3.8 times wider and a 5th–95th‐percentile range some eight semitones wider, with 3.2 times greater motion energy and a more modest difference in amplitude variability. The two versions were closely matched in spectral flux, spectral centroid, envelope modulation spectrum, mean luminance and spatial contrast (Supporting Information S1).
Prior to the MEG experiment, the two videos were evaluated in an online pilot with 30 participants using a Hebrew adaptation of the Conger–Kanungo Scale of Charismatic Leadership (Conger and Kanungo 1994). The scale demonstrated good reliability (Cronbach's α = 0.784).
3.2.3. Questionnaires
After each video, participants completed several questionnaires assessing their subjective responses. The modified Hebrew version of the Conger–Kanungo charismatic leadership questionnaire (Conger and Kanungo 1994) served as a manipulation check. Participants also completed the Positive and Negative Affect Schedule (PANAS; Watson et al. 1988), a collective efficacy questionnaire designed to assess students' confidence in their group abilities (Chen and Bliese 2002), and a cognition‐based trust questionnaire evaluating perceived reliability of the speaker (McAllister 1995). All questionnaires were presented electronically in a fixed order, and participants responded orally while the experimenter documented their answers.
3.2.4. Procedure
Upon arrival at the MEG laboratory, participants provided demographic information and signed an informed consent form. After removing all metal accessories, makeup, and jewelry, they entered the magnetically shielded room and lay in a supine position. Five head‐position indicator coils were attached to the scalp, and head shape was digitized using a Polhemus FASTRAK system to allow subsequent source localization. Earphones were positioned and secured, and participants were instructed to minimize movement and blinking throughout the session.
The session began with resting‐state recordings, including 2 min of eyes‐closed rest followed by 2 min of eyes‐open fixation. Participants then received recorded instructions informing them that they would watch a short video and later answer questions about it. They viewed the first video, either the charismatic or non‐charismatic version (order counterbalanced). After the video, questionnaires appeared on the screen in a fixed order, and participants responded verbally while the experimenter recorded responses via intercom. After a short break the same sequence was repeated for the second speech version. Next, participants watched a 2.5‐min video consisting of the actor standing silently in the same setting (used as a control). At the end of the experiment, participants exited the MEG room, were debriefed and received compensation. The entire session lasted approximately 1 h.
3.2.5. Statistical Analysis
Acquisition and preprocessing followed Section 2. The charismatic, non‐charismatic, and silent conditions were compared with repeated‐measures ANOVA and paired follow‐up tests across participants, corrected as described in Section 2.5. The delivery contrast was tested both whole‐brain and within the region of interest defined in Section 2.6.
3.3. Results
The manipulation check confirmed that the delivery‐style manipulation was successful. Questionnaire data were available for 39 of the 40 participants, and all comparisons are within‐subject. Participants rated the charismatic speech as more charismatic than the non‐charismatic version (M = 4.47, SD = 0.68 vs. M = 3.36, SD = 0.75), t(38) = 8.37, p < 0.001, Cohen's dz. (the mean difference divided by its standard deviation) = 1.34. Positive affect, as measured by the PANAS, was also higher following the charismatic speech (M = 25.8, SD = 7.1, summed across the 10 positive items) than the non‐charismatic speech (M = 20.1, SD = 5.5), t(38) = 6.44, p < 0.001, dz. = 1.03, while negative affect did not differ between conditions, t(38) = −1.14, p = 0.26, dz. = −0.18. Collective efficacy ratings showed a similar pattern, with higher scores after the charismatic speech (M = 4.15, SD = 1.06) than after the non‐charismatic speech (M = 3.70, SD = 0.96), though the difference was more modest, t(38) = 2.27, p = 0.029, dz. = 0.36. In contrast, ratings on the trust questionnaire did not differ between conditions, t(38) = −0.27, p = 0.79, dz. = −0.04. Overall, the behavioral measures indicate that charismatic delivery increased perceived charisma, positive emotional responses and, more modestly, collective efficacy, while leaving trust judgments unchanged.
3.3.1. Spectral Power
Analyses of spectral power averaged across regions revealed condition differences in the lower frequency bands (Figure 2A). Alpha‐band power was significantly lower in the charismatic than in the non‐charismatic and control conditions, F(2, 78) = 8.88, p < 0.001, η 2 p = 0.19, and a smaller reduction was present in delta, F(2, 78) = 4.68, p = 0.012, η 2 p = 0.11. Spectral parameterization showed the alpha reduction to be present in the periodic (oscillatory) component once the aperiodic fit was removed, F(2, 78) = 5.05, p = 0.009, η 2 p = 0.12, and more widespread there than in raw power (25 vs. 13 regions surviving FDR correction); a small difference in the aperiodic exponent was also present, F(2, 78) = 3.89, p = 0.024, so a broadband contribution cannot be excluded entirely. The alpha reduction localized to a posterior network: bilateral cuneus, calcarine, lingual and superior, middle and inferior occipital gyri, extending into precuneus, superior parietal, cingulate, angular and temporal regions (t = −5.59 to −2.50). The beta‐band reduction was weak and did not survive correction across the five bands, F(2, 78) = 3.38, p = 0.039. Charismatic delivery thus attenuated posterior alpha oscillatory power.
FIGURE 2.
Study 1: spectral power and global inter‐subject correlation (ISC). (A) Grand‐average power spectrum across the 72 atlas regions for each condition (left), and the spectral‐parameterization (specparam) decomposition of the grand‐average spectrum (right), showing an oscillatory alpha peak above the aperiodic (1/f) fit. Charismatic speech shows reduced alpha power; the accompanying beta reduction did not survive correction across bands. (B) Global ISC (leave‐one‐out, Fisher‐z) averaged across regions, by frequency band and condition. ISC is highest in the beta band within every condition, and higher for charismatic than non‐charismatic speech in alpha and beta (asterisks, p < 0.01). Error bars, SEM.
3.3.2. Inter‐Subject Correlation
Inter‐subject correlation analysis revealed robust differences in neural synchrony between the delivery conditions (Figure 2B). A repeated‐measures ANOVA showed significant effects of condition on global ISC in the alpha, beta, and gamma bands (F(2, 78) = 14.48, p < 0.001, η 2 p = 0.27; F(2, 78) = 93.66, p < 0.001, η 2 p = 0.71; F(2, 78) = 32.56, p < 0.001, η 2 p = 0.46, respectively), with no effect of condition in delta or theta (both F < 1). Post hoc comparisons indicated that global ISC was higher during the charismatic speech than during the non‐charismatic speech in the alpha (p = 0.007) and beta (p < 0.001) bands, but not in gamma (p = 0.11). Both speech conditions exceeded the silent control in alpha, beta, and gamma: charismatic in all three (all p < 0.001), non‐charismatic in alpha (p = 0.021), beta, and gamma (both p < 0.001). The non‐charismatic condition therefore showed intermediate synchrony. These findings demonstrate that expressive delivery enhances large‐scale shared neural activity across listeners, with the strongest effects emerging in the beta band.
Direct comparisons between the charismatic and non‐charismatic conditions, tested at the subject level, showed that charismatic delivery increased ISC most strongly in the beta band, where the effect was widespread (47 of 72 regions survived FDR correction) with maxima in left occipital and right‐perisylvian cortex (right superior and middle temporal gyri, supramarginal gyrus, rolandic operculum and insula; Figure 3). Because our ISC time series were low‐pass filtered at 0.8 Hz, we did not expect this signal to reflect rapid acoustic dynamics. Consistent with this, its correlation with the speech amplitude envelope did not exceed a circular‐shift null (p = 0.74). MEG activity did track the envelope at the faster timescales (1–8 Hz, p = 0.001) that the low‐pass filter removes, but this coupling did not differ between the two conditions (p = 0.43), nor did the slow measure (p = 0.86); full methods and results are given in Supporting Information S3. Envelope tracking is therefore not a candidate explanation for the delivery difference in ISC, although slower stimulus‐related contributions remain possible, and we tested these directly by regression.
FIGURE 3.
Study 1: charismatic & non‐charismatic beta‐band ISC across 72 cortical regions, before and after low‐level control (surface color = ISC difference, Fisher‐z; lateral and medial views of each hemisphere). Only regions surviving FDR correction at the corresponding stage are shown. Top: the raw effect, widespread with maxima in left occipital and right‐perisylvian cortex. Bottom: the core surviving both the low‐level acoustic and visual feature regression and the separation of periodic from aperiodic activity, comprising right rolandic operculum, right superior and middle temporal gyri, left mid‐cingulate and left superior occipital cortex. The three right‐perisylvian regions are those that recur in Study 2.
To test whether the ISC condition difference reflected the clips' low‐level properties rather than expressive delivery per se, we regressed the acoustic and visual features described in Section 2.5 out of each region's time course before recomputing ISC, with a phase‐scrambled‐feature null (Supporting Information S2). This reduced the beta effect from 47 to 9 regions surviving correction, and the reduction was strongly region‐specific: in temporal cortex 31% of the effect was retained and the frontal and insular effects were removed in full, indicating a large stimulus‐acoustic contribution there, whereas occipital cortex retained 79%, the least affected of any region. A phase‐scrambled null of matched complexity removed essentially nothing anywhere (96%–104% retained), confirming that the reductions were feature‐specific. We then applied a second control, removing the aperiodic component before regressing the same features. The two controls are not nested, since they act on different sources of variance. The aperiodic component is broadband and shared across listeners, so it contributes correlated variance to every band and dilutes a band‐specific effect; removing it can therefore leave a larger residual than feature regression alone. The temporal and mid‐cingulate effects, which do not survive feature regression on its own, are accordingly significant after the pair. Five regions survive both: the right rolandic operculum, the right superior and middle temporal gyri, the left mid‐cingulate and the left superior occipital gyrus (Table S5). Adding word onsets did not reduce these further, and an extended visual model removed no more than its own phase‐scrambled null in any region group except cingulate cortex. After low‐level control, a residual right‐perisylvian and occipital beta ISC increase, present on the aperiodic‐removed signal, remained specific to charismatic delivery (Figure 3 Bottom).
A linear regression with global ISC as the dependent variable and the questionnaire scores as predictors found that, for mean ISC, only PANAS positive ratings contributed significantly (β = 0.38, p = 0.04; model R 2 = 0.16, n = 39). The same predictor was significant for theta (β = 0.43, p = 0.02) and gamma (β = 0.48, p = 0.01) but not for alpha or beta; in delta, cognition‐based trust was the only significant predictor, in the opposite direction (β = −0.75, p = 0.02). Simple correlations between PANAS positive and mean ISC were positive under charismatic delivery and negative under non‐charismatic delivery, though neither reached significance (r = 0.24, p = 0.15; r = −0.22, p = 0.18); the only correlation that did was gamma ISC in the charismatic condition (r = 0.43, p = 0.006), with none significant in the non‐charismatic condition. A representational‐similarity analysis asked whether behaviorally similar pairs of participants also had more similar neural time courses. Under both nearest‐neighbor and Anna Karenina models, tested by Mantel permutation, the result was null across all bands, conditions, behavioral measures and region sets (Supporting Information S7). The individual‐difference signal in these data is therefore one of magnitude, how strongly a person's responses track the group, rather than of similarity structure.
3.4. Discussion
The behavioral consequences of delivery style were selective, and so were the neural ones. Charismatic delivery raised perceived charisma, positive affect and collective efficacy while leaving trust judgments untouched, which is what one would expect of a manipulation that alters how a message is delivered rather than what it asserts. In the MEG data, its clearest effect on oscillatory power was a reduction of posterior alpha, a pattern conventionally associated with externally oriented attention and perceptual engagement (Foxe and Snyder 2011; Klimesch 2012). That reduction was not an artifact of the broadband spectrum: it was present in the periodic component once the aperiodic fit was removed, and indeed more widespread there than in raw power. A small difference in the aperiodic exponent was nevertheless present, so a broadband contribution cannot be excluded, and we take the power result as indicating heightened perceptual engagement without resting the interpretation on it.
The correlation results provide clearer evidence of shared processing. Charismatic delivery raised global inter‐subject correlation in alpha and beta, most strongly in beta, and produced widespread regional effects in the temporal, parietal, and occipital cortex; gamma correlation exceeded the silent control but did not distinguish the two speech conditions. That engaging material elicits stronger correlation across listeners and viewers is itself well established, for film and video (Dmochowski et al. 2012; Hasson et al. 2004) and for spoken narrative and political speech (Regev et al. 2013; Schmälzle et al. 2015), but most of that work used broadband measures, which show that listeners are aligned without indicating what is aligned. Resolving correlation by frequency changes what can be asked of the measure. It also shows that power and correlation are separable quantities rather than two views of the same thing: alpha power fell while alpha correlation rose, so listeners were not simply producing more of the same signal but producing it more similarly to one another.
Whether the correlation increase amounts to anything beyond heightened attention to a more engaging stimulus is the natural objection, and the controls speak to it. Envelope tracking was present at the faster timescales removed by our low‐pass filter but did not differ between conditions, so the effect is not a difference in how closely listeners followed the acoustics. Nor is it a difference in the stimulus itself: the perisylvian effect withstood the feature regression and the aperiodic separation, and the phase‐scrambled control confirmed that the reductions elsewhere were feature‐specific (Section 3.3.2). Nor does affect account for it: subjective positive affect predicted correlation in theta and gamma but not in alpha or beta, the bands carrying the delivery effect. Beta has been linked to linguistic prediction, top‐down integration and sensorimotor coupling during speech (Arnal and Giraud 2012; Weiss and Mueller 2012; Lewis and Bastiaansen 2015), and we return to those accounts in Section 5.
One apparent effect should be set aside. A theta difference visible in the uncorrected data reflected blink artifact depressing correlation in the silent control condition rather than any difference between the speech conditions. Blinks are frontal, slow, and unsynchronized across participants, and they were more frequent while watching a motionless speaker; the difference disappeared once ocular components were removed.
Study 1 therefore shows that delivery alone, holding content and speaker identity constant, is sufficient to change how similarly listeners process a message. The effect surviving low‐level control is centered on right‐perisylvian cortex, together with a left mid‐cingulate effect and an occipital component considered in Section 3.5, and it is concentrated within the region specified in advance from the prosody literature (Section 2.6). What it cannot show is whether the same alignment arises when listeners come to the same speech with different predispositions, or when the social setting in which they hear it varies. Those are the questions the second experiment addresses, and it does so with the delivery style held constant.
3.5. Gaze Alignment in an Independent Sample
The delivery effect reported above raises a question that the MEG data cannot settle on their own. If a charismatic delivery draws listeners' eyes to the same places at the same times, then part of the correlation between their cortical time courses may reflect shared oculomotor and visual input rather than shared processing of the speech. We therefore asked whether gaze itself is more aligned under charismatic delivery, in a separate sample viewing the same two clips. Fifty‐six students at Bar‐Ilan University took part, all native Hebrew speakers screened for visual abnormalities and strabismus; two were excluded, one for repeatedly falling asleep during viewing and one for gaze values beyond three standard deviations from the sample mean, leaving 54 participants (42 women, aged 21 to 29, M = 24.3, SD = 2.3). Gaze position was recorded at 150 Hz with a Gazepoint GP3 eye‐tracker, and inter‐subject gaze synchrony was quantified with the multidimensional scanpath‐similarity algorithm MultiMatch (Dewhurst et al. 2012), each participant's score in a condition being their mean similarity to the remaining participants. This sample did not undergo MEG, so gaze cannot be regressed from the neural time courses; the analysis bears on the plausibility of an oculomotor account rather than removing it. The participants also viewed the clips seated at a monitor rather than supine under the MEG display, so the two datasets differ in viewing geometry as well as in participants.
Scanpaths were more synchronized while participants watched the charismatic speech than the non‐charismatic one, F(1, 53) = 8.90, p = 0.004, partial eta‐squared = 0.144, a difference driven by longer fixations on the speaker's face. Charismatic delivery therefore does align visual attention, and the oculomotor account is not idle speculation.
It does not, however, account for the effect reported in Section 3.3, and the pattern of regions indicates where it does and does not apply. Of the five regions surviving both controls, the left superior occipital gyrus is the one an oculomotor account predicts, and it is also the region that retained the largest share of its effect through the acoustic and visual feature regression. We accordingly regard the occipital component of the delivery effect as plausibly attributable to shared looking, and treat it as such throughout. The perisylvian regions are a different matter. Shared fixation on a face gives no reason to expect a beta effect in the rolandic operculum and the temporal gyri specifically, nor a right‐hemisphere asymmetry of the size observed; the ocular component was removed by a rule‐based independent‐component procedure applied to every recording (Section 2.3), and the effect survived regression of a visual‐motion model richer than gaze position alone. The left mid‐cingulate effect is accounted for by neither the oculomotor nor the prosodic reading, and we do not attempt to interpret it. Taken together, the gaze result narrows the delivery effect rather than undermining it: what survives the oculomotor account is the right‐perisylvian component, which is also the component that recurs in Study 2.
4. Study 2: Listener Attitude and Social Context
4.1. Rationale and Predictions
Study 1 showed that a property of the speaker, expressive delivery, shapes how similarly listeners process a naturalistic speech. The complementary question is what the listener brings to it. Study 2 manipulated listener attitude and, alongside it, the apparent presence of an audience, within a single cohort, so that an influence on the communicative relation and one that leaves it intact could be compared in the same participants and the same analytic frame.
Prior attitudes shape how messages are processed and evaluated. Pre‐existing impressions influence source credibility, selective attention, and perceived intentions before the content is weighed (Appelman and Sundar 2016; Chaiken 1980; Petty and Cacioppo 1986; Todorov et al. 2015). In the neural literature, ideological similarity predicts greater synchrony during naturalistic political content (de Bruin et al. 2023), partisan discourse appealing to policy beliefs and to partisan identity recruits dissociable systems (Jacoby et al. 2024), and listener traits modulate the degree of brain‐to‐brain synchrony under polarized viewing (van Baar et al. 2021). The listener side of this relation can outweigh the ideological one: shared extremity of view, irrespective of its direction, is itself sufficient to raise synchrony (de Bruin and FeldmanHall 2025). Because such studies rely on real figures or entrenched identities, the manipulated attitude is difficult to separate from the long‐standing beliefs, personality and media exposure with which political ideology is bound up (Amodio 2014; Jost et al. 2009). Source effects also depend on whether recipients already hold an attitude toward the issue (Kumkale et al. 2010). We therefore induced attitudes de novo toward an unfamiliar speaker, isolating the causal effect of a newly formed predisposition.
The audience cue serves a different purpose. Co‐presence alone can raise perceived relevance and sustained attention and encourage shared interpretations: physiological and emotional responses synchronize across individuals who are merely co‐present and not interacting (Golland et al. 2015), shared attention to the same object shapes how it is construed (Shteynberg 2015), and social contact between viewers alters neural synchrony during subsequent co‐watching (De Felice et al. 2024). In these studies the other people are physically present, whereas in everyday communication such context is often conveyed by minimal cues, a speaker shown before a visible audience being the commonest of them. We therefore kept the cue minimal, accepting that a manipulation of this kind trades salience for ecological realism, and asked whether so subtle a cue is by itself sufficient to change how similarly listeners process speech.
We reasoned that the two manipulations should not behave alike. The attitude induction changes the listener's stance toward the message (Hughes and Zaki 2015; Van Strien et al. 2016) and should modulate shared processing in speech‐relevant cortex; if it acts on the same substrate engaged by expressive delivery, it should do so in the right‐perisylvian regions identified above. The audience cue, by contrast, changes neither the message nor the listener's disposition toward the speaker, and therefore provides a test of specificity: if shared processing is driven by engagement with the message, the cue alone should leave ISC largely unchanged. Unlike the delivery manipulation, no band‐specific prediction followed from the attitude literature, and we did not commit a priori to a single frequency band. Finally, because the same participants completed both tasks, we could test whether the tendency to synchronize is stable across different speakers and materials rather than specific to a given speech.
4.2. Method
Participants and design. Sixty‐five right‐handed Hebrew‐speaking adults took part; after exclusions for earphone failure and MEG artifacts, 58 contributed to the attitude analysis and 61 to the social‐context analysis (overlap of 58; the three additional participants completed only the second task). The design was crossed: each participant was assigned to one attitude group (positive vs. negative description) and one audience‐cue group (cue vs. no cue), with the two factors counterbalanced, and completed both tasks in a single session while remaining in the scanner.
Attitude manipulation and stimulus. Before the first video, participants read a short description portraying the speaker positively (a generous philanthropist) or negatively (a racist, exploitative employer) and summarized it aloud. They then watched an 8‐min 36‐s video in which a personal trainer delivered a charismatic lecture on ‘the secret to a happy life.’ Afterward, they completed a speaker evaluation, the PANAS, the Big‐Five inventory, and a 20‐item content‐memory test.
Social‐context manipulation and stimulus. Immediately afterward, participants watched a second, unrelated video—a 5‐min 56‐s lecture on parenting delivered by a different (female) speaker—presented either with a minimal audience cue (dark silhouettes at the bottom of the screen) or without it. The two versions were identical in all auditory and linguistic features and in every visual feature except the cue.
Acquisition and preprocessing followed Section 2, and group differences were tested at the subject level with the group‐label permutation described in Section 2.5. The attitude and audience‐cue contrasts were tested both whole‐brain and within the region of interest defined in Section 2.6.
4.3. Results
Attitude manipulation check. Participants given the positive description evaluated the speaker more favorably, scoring lower on a scale on which higher values indicate a less favorable evaluation (M = 27.6, SD = 6.35, vs. M = 32.1, SD = 6.57), t(56) = 2.70, p = 0.009, d = 0.71. The groups did not differ in positive or negative affect (both p > 0.15) or in memory for the lecture (80.9% against 77.6%, p = 0.19). Differences in inter‐subject correlation are therefore unlikely to be attributable to engagement, comprehension, or mood.
Attitude modulated inter‐subject correlation in theta and alpha. Global correlation was higher in the positive‐ than in the negative‐attitude group in theta (p = 0.004) and alpha (p = 0.012); both survived Benjamini–Hochberg correction across the five bands, and theta also fell below the Bonferroni threshold. Delta did not differ (p = 0.06), nor did beta (p = 0.15) or gamma (p = 0.37). Spectral power did not differ in any band, so the group difference was not accompanied by a difference in the strength of the oscillations themselves. A regression of alpha correlation on positive affect and attitude accounted for 13% of the variance (R 2 = 0.13, p = 0.02), a relationship absorbed when group was entered (both predictors falling to p > 0.25; Figure 4A).
FIGURE 4.
Study 2. (A) Positive > negative attitude ISC (Fisher‐z difference) across 72 regions in the alpha and theta bands, showing regions surviving one‐sided FDR correction (six in alpha, nine in theta; Table S8). The alpha effect is concentrated in right perisylvian cortex (insula, rolandic operculum, inferior frontal operculum and superior temporal gyrus, the first three of which also survive two‐sided correction) while theta is broader, weaker and bilateral. (B) Global ISC by frequency band and group. Positive attitude raises alpha and theta ISC (asterisks); the audience cue produces no band‐specific increase; beta carries the highest baseline ISC in both. Error bars, SEM. *p < 0.05, group‐label permutation (10000) one‐sided in the predicted direction.
Regionally, the alpha effect was the strongest and the most extensive. Three right‐perisylvian regions survived two‐sided FDR correction across the 72 parcels: the right insula (Δ = +0.091, p = 0.0007), the right inferior frontal operculum (+0.069, p = 0.0010) and the right rolandic operculum (+0.089, p = 0.0012). The right superior temporal gyrus (+0.083, p = 0.001) survived under the directional test described in Section 2.5, as did the right orbital superior frontal gyrus and the left paracentral lobule (Table S8). Its two‐sided p, 0.003, falls essentially on the correction threshold, so its exclusion from the preceding set is not a substantive distinction. Theta yielded a broader but weaker set of nine regions under the same directional test, overlapping in part with the alpha network. Beta did not survive correction in any region. It was focal rather than distributed, and largest at the right rolandic operculum (+0.10, uncorrected p = 0.014), which together with the marginal global difference indicates that the attitude manipulation had at most a weak effect in this band. The right inferior frontal operculum lies outside the auditory‐perisylvian region defined in Section 2.6, which was restricted to temporal, opercular and insular parcels; it is reported here and in the whole‐brain analyses but does not enter the region‐level tests.
The attitude effect was not stimulus‐driven. Regressing out the low‐level acoustic and visual features left the alpha and theta effects intact: the global difference was undiminished in both (alpha p = 0.011 to 0.009, theta p = 0.004 to 0.003), while the phase‐scrambled null removed essentially nothing (94% to 99% retained). The global beta difference fell by more than half (+0.016 to +0.007), consistent with the beta effect being substantially stimulus‐locked; the focal effect at the right rolandic operculum was likewise more than halved (+0.103 to +0.045) but remained nominally significant (uncorrected p = 0.042), whereas averaged over the region of interest it did not (+0.031, p = 0.070).
Within the region of interest, the attitude effect was spatially concentrated as well as present. Two tests were applied to the four parcels defined in Section 2.6, and they answer different questions. Against a null constructed over regions, which asks whether an effect is concentrated in this cortex rather than elsewhere in the brain, the region‐averaged difference after feature regression exceeded all but a small fraction of the 1,028,790 possible four‐parcel sets in alpha (mean Δ = +0.047, p = 7.8 × 10–5), and was likewise concentrated in theta (+0.034, p = 0.016) and beta (+0.031, p = 0.0033); exceedance counts for every contrast and stage are given in Table S7. Against a null constructed over participants, which asks whether the region‐averaged difference departs from zero, the alpha and theta effects were reliable (+0.047, p = 0.001 and + 0.034, p = 0.003), whereas the beta difference was not (+0.031, p = 0.070). The beta effect is therefore not one we take forward, and what the region‐level analysis establishes is that the attitude effect in alpha and theta is both present in this cortex and concentrated there.
A stricter control reduces the analysis to alpha alone. The results above are reported after the feature regression, a stage fixed in advance on grounds of sensitivity and not chosen from among the outcomes: a between‐group contrast of this size has considerably less power to absorb the loss of correlation that the further control entails than does the within‐participant contrast of Study 1, which is reported at the fully controlled stage throughout. Separating periodic from aperiodic activity before computing correlation leaves between 13% and 51% of the baseline correlation itself in the periodic component, depending on band, and inflates gamma by some 30%. This is a property of the measure rather than of the group effect, and the two should not be conflated. At that stage the alpha effect remains concentrated in the region (+0.022, p = 0.0049), whereas theta (+0.004, p = 0.62) and beta (+0.005, p = 0.34) do not. We report this as a sensitivity analysis rather than as the primary result, because the reduction in effect size is of the order expected from the loss of correlation itself. Absence at this stage is therefore not evidence that the effects were aperiodic in origin. Table S7 gives every contrast at all three stages.
The audience cue did not modulate inter‐subject correlation. Under the same subject‐level test, the cue against no‐cue contrast was non‐significant in every frequency band, and no region survived FDR correction in any band (0 of 72 throughout); the global beta difference was small and, if anything, negative in sign. Spectral power likewise did not differ. Nor was the cue effect merely diffuse rather than absent: within the region of interest, where an effect was predicted, the cue contrast was indistinguishable from the reference distribution in both alpha (+0.005, p = 0.42) and beta (+0.000, p = 0.14), at no stage of the control sequence departed from it, and did not differ from zero. A minimal visual indication of an audience, holding the message and its delivery constant, thus did not detectably change how similarly listeners processed the speech (Figure 4B).
Awareness of the cue was not assessed. Because the manipulation was between subjects, participants saw only one version of the video and had no basis for comparison, and no debriefing question established whether the silhouettes had been noticed. The null is therefore consistent with two accounts that the present data cannot separate: a cue that was registered and did not alter shared processing, and a cue that was never registered at all.
The two manipulations were independent, and synchronization was a stable trait. Prior attitude did not carry over into the social‐context task: no region survived correction in any band, and no band survived correction across the five. Alpha showed a nominal difference in the same direction as the primary effect (d = +0.020, uncorrected p = 0.049), which is what the cross‐task stability reported below would predict; because the two factors were fully counterbalanced, it cannot bias the audience‐cue contrast. Because participants completed both tasks, it was possible to ask whether the propensity to synchronize is itself stable. A participant's correlation on one task predicted their correlation on the other, most strongly in alpha (r = 0.61), less so in beta (r = 0.49) and theta (r = 0.33), and not at all in delta or gamma (r = 0.08). The alpha relationship survived partialing out delta and gamma (r = 0.48) and overall correlation magnitude (r = 0.34, p = 0.010), and was not predicted by Big‐Five traits, suggesting a stable, stimulus‐general and alpha‐specific disposition to synchronize over and above any general signal‐quality factor. The quantity that is stable here is the magnitude of a participant's alignment with the group, not the idiosyncratic pattern of their response (cf. Finn et al. 2020). Figure 5 shows the cross‐task relationship.
FIGURE 5.
Neural synchronizability as a stable individual trait. Left: each participant's alpha ISC on the attitude video versus the (different) audience‐cue video; the positive relationship (Pearson r = 0.61, p < 0.001; Spearman r = 0.60; r = 0.56 with the most extreme point removed) indicates that a person's tendency to synchronize generalizes across different speech. Right: cross‐video ISC reliability by band, significant in theta, alpha and beta but not delta or gamma—ruling out a signal‐quality explanation. *p < 0.05.
4.4. Discussion
Within a single experiment, and in the same participants measured the same way, a manipulation of the listener's relation to the speaker and one that left that relation intact behaved differently. A description of the speaker read minutes beforehand raised inter‐subject correlation in alpha and theta, and the effect was concentrated in the right auditory‐perisylvian region defined in advance. It withstood regression of the low‐level acoustic and visual features, so it was not a consequence of the two groups hearing different sound or seeing different motion. Under the stricter separation of periodic from aperiodic activity, the effect reduced to alpha alone, theta and beta falling away and only the alpha concentration in the region surviving. A minimal cue that the speaker was addressing an audience did not alter correlation in any band, in any region, at any stage of the control sequence.
The contrast sharpens what the measure appears to track. Shared processing followed the listener's stance toward the speaker, and not a feature of the setting that left both the message and that stance unchanged. The null is not merely an absence of significance across the whole brain: within the region where an effect was predicted, the cue contrast was indistinguishable from the reference distribution in alpha and beta, did not differ from zero, and departed from neither at any stage of the control sequence. Two qualifications belong with that null rather than after it. Awareness of the cue was not assessed, so a cue that was noticed and proved inert cannot be distinguished from one that was never noticed; and because the two tasks were administered in sequence, the null arose in the later and less sensitive of them.
The cross‐task result identifies a trait dimension alongside these state effects. Some listeners synchronized more than others across two quite different speakers and topics, and did so specifically in alpha, in a manner not explained by signal quality or by personality as measured by the Big Five. Inter‐subject correlation therefore reflects both the communicative situation, meaning who is speaking and how the listener is disposed toward them, and a stable property of the listener that is carried across situations.
5. General Discussion
In general, the two experiments together support a conclusion that neither affords alone. Expressive delivery and a newly formed listener attitude, influences that enter the communicative chain from opposite ends, both raised inter‐subject correlation within the same right‐perisylvian cortex, whereas a cue that the speaker addressed an audience did not. In Study 1, analyzed whole‐brain and without any regional constraint, five regions survived both controls: the right rolandic operculum (+0.061), the right middle (+0.055) and superior (+0.052) temporal gyri, the left superior occipital gyrus (+0.055) and the left mid‐cingulate cortex (+0.041). Three of these are perisylvian, and it is these three that recur in Study 2, whereas the occipital and cingulate effects do not. Across the four parcels specified in advance (Figure 6, Section 2.6), the delivery effect was more concentrated than in sets of the same size drawn from elsewhere in the atlas (mean d = +0.043, p = 0.00044, after both controls). Study 2 applied that same region to a different cohort, listening to a different speaker. Its attitude effect was more widely distributed, but it was concentrated in that region in alpha (+0.047, p = 7.8 × 10−5) and theta (+0.034, p = 0.016) once the low‐level features were regressed out, the accompanying beta difference proving substantially stimulus‐locked and not distinguishable from zero thereafter, whereas the audience cue showed no such concentration in any band at any stage of the control sequence (p = 0.14 at its most extreme, with an effect size of zero) and no region survived correction in the whole‐brain analysis. The two message‐relevant factors thus reached the network and the context‐only cue did not, a dissociation obtained in the same participants and with the same measure, although the interpretation of the null is qualified by the fact that awareness of the cue was not assessed, a point we return to below.
FIGURE 6.
Two manipulations converge on an anatomically pre‐specified right auditory‐perisylvian region; a third does not. (A) Effect maps, right lateral view, on one color scale: Study 1 charismatic minus non‐charismatic, beta, after feature and aperiodic control (n = 40, within‐subject); Study 2 attitude, positive minus negative, alpha, after feature control (n = 58); Study 2 audience cue, cue minus no cue, alpha, after feature control (n = 61). Each is shown at the control stage reported in the text. Black outline, the four parcels of the region of interest—right rolandic operculum, superior and middle temporal gyri, insula—defined anatomically and independently of both results; the insula is buried in the sylvian fissure, so the outline effectively traces the other three. Scale as in m 0 to 0.09, negatives clipped, unthresholded, so color shows effect size rather than significance. Both message‐relevant manipulations are strongest inside the outline; the cue is not, though its map is not uniformly weak elsewhere. Study 1 is a beta effect on the aperiodic‐removed signal and Study 2 alpha effects on band amplitude, so the shared scale compares the size of the ISC difference, not the underlying signal. (B) The enrichment test (Table S7). Gray, the mean effect across all 1,028,790 four‐parcel sets in the atlas; vertical line, the observed mean in the region of interest. The reference set is enumerated in full, so p is exact: delivery +0.043 (456 sets exceed, p = 0.00044), attitude +0.047 (79, p = 0.000078), cue +0.005 (427,423, p = 0.42). Panels share an x‐axis. The test asks whether an effect is concentrated in this region not whether it differs from zero.
The convergence was anatomically specific. It fell in right perisylvian cortex, encompassing superior and middle temporal, opercular and insular regions, rather than in the left‐dominant frontotemporal language network. That this is cortex repeatedly implicated in the analysis of the voice and its paralinguistic properties is what makes the localization informative, and it is the basis on which the region was specified in advance (Section 2.6). The rolandic operculum bears most directly on the present findings. In the right hemisphere it is a somatosensory area representing the lips, jaw and tongue, engaged during listening as part of an audio‐motor loop (van Rijn et al. 2005; cf. Vigneau et al. 2006, for the left‐hemisphere case), and it is the region whose disruption selectively impairs the perception of emotional prosody while leaving emotional semantics intact (Hoekert et al. 2008; van Rijn et al. 2005). Both of our manipulations bear on that dimension. The two clips differed principally in their pitch dynamics (Section 3.2.2), and the attitude induction changed how listeners took up that same expressive signal.
The two manipulations did not, however, engage the region uniformly. The delivery effect had no left‐hemisphere counterpart, the same four parcels in the left hemisphere averaging −0.000, with the right exceeding the left in all four regions; the attitude effect was right‐dominant but more bilateral (laterality index +0.25 for alpha and + 0.13 for theta). Within the region, the two effects were also carried by different parcels. The delivery effect resided in the rolandic operculum and the temporal parcels and was negligible in the insula, whereas the insula yielded the largest single regional effect of the attitude manipulation. The insula was included in the region on prior anatomical grounds rather than because it carried the delivery effect, and its behavior across the two experiments is itself part of the dissociation. We therefore interpret the shared substrate as a right‐perisylvian network for processing the speaker's expressive signal, engaged from the auditory‐motor side by delivery and with a larger insular contribution when the listener's disposition is manipulated, and not as a language or a general predictive‐coding region.
The frequency profile admits more than one reading, and it is worth setting them out. The delivery manipulation acted chiefly through beta and the attitude manipulation through alpha and theta. On one account, beta reflects the tighter sensorimotor coupling afforded by dynamic delivery (Arnal and Giraud 2012; Engel and Fries 2010), while the alpha and theta effects reflect the attentional and segmentation adjustments described above, imposed by the listener's disposition on an identical input (Klimesch 2012; Luo and Poeppel 2007). The anatomy is congenial to this reading, since the parcel carrying the delivery effect is somatosensory cortex for the articulators in the right hemisphere and is thought to maintain a motor representation of the heard signal (van Rijn et al. 2005). On a second account, beta indexes the maintenance of a cognitive set (Engel and Fries 2010) or the feedback direction of cortical communication, which has been argued to run preferentially in beta while feedforward influences run in gamma and theta (Bastos et al. 2015; Fries 2015). The present data do not adjudicate between these accounts, and the frequency assignments should be read as candidate channels rather than as evidence for any one mechanism.
The findings can also be explained by a simple difference in attentional arousal, but three observations weigh against that account: the attitude groups did not differ in positive or negative affect or in memory for the lecture, yet differed in inter‐subject correlation; subjective positive affect in Study 1 predicted correlation in theta and gamma but not in alpha or beta, the bands carrying the two effects; and spectral power did not differ between groups in Study 2. The separation between the two channels is, moreover, sharper than it first appeared. The attitude manipulation did produce a weak beta difference, focal to the same operculum that carried the delivery effect, and an earlier reading of these data treated that as a component shared by the two manipulations. It does not survive the low‐level controls, halving under acoustic and visual regression and falling short of significance thereafter, whereas the delivery effect in beta survives both controls intact. The apparent overlap in beta was therefore largely stimulus‐locked, and the two influences are better described as reaching the shared network through channels that do not overlap at all.
A stricter analysis, in which periodic activity is separated from the aperiodic background before correlation is computed, narrows the attitude effect further to alpha alone. We do not read that as evidence that the theta effect was aperiodic in origin, because the separation itself retains only a minority of the baseline correlation and the reduction in effect size is of the order that loss predicts; the periodic estimate is also least stable at the lower edge of the fitting range, where theta sits. The comparison does indicate, however, that of the two attitude channels only alpha is demonstrably carried by oscillatory activity. In our view the present data cannot adjudicate among the oscillatory accounts, and we do not attempt to. What the design does establish, and what does not depend on any of them, is that a shared right‐perisylvian network is modulated by delivery and by attitude through distinct frequency channels.
Theoretically, the results extend interactive‐alignment accounts of communication (Pickering and Garrod 2004) in two respects. Convergence during naturalistic speech is shaped not only by the structure of the message but by what the listener brings to it, and the substrate on which the two influences converge can be localized to a network for the speaker's expressive signal rather than to language cortex. The attitude effect further shows that a predisposition need not be entrenched to matter; a description read minutes earlier, concerning a speaker the listener had never encountered, was sufficient to change how similarly an identical lecture was processed. These state effects are not the whole of the variance, however. That the same listeners synchronized more strongly across two unrelated speakers, specifically in alpha and independently of personality (Section 4.3), implies that alignment is jointly determined by the situation and by a disposition the listener carries into it, so a complete account of shared processing cannot be framed in terms of the stimulus alone (cf. Finn et al. 2020; van Baar et al. 2021). Notably, that trait is one of magnitude and not of similarity structure. Participants differed in how strongly their responses tracked the group, but behaviorally similar participants were no more similar to one another neurally than any others, so the measure appears to index the strength of entrainment rather than an idiosyncratic geometry of response.
Several limitations bear on the strength of these conclusions. The studies used a small number of speakers and topics, and although the same‐actor design of Study 1 removes the confounds of content and identity that have complicated earlier delivery comparisons, it cannot show that the effect generalizes across speakers; that would require a set of speakers each recorded in both styles. The spatial resolution of MEG source estimation, combined with a 72‐region parcellation, does not permit adjacent perisylvian regions to be separated with confidence, and we accordingly describe the substrate as a right‐perisylvian cluster rather than as a single parcel. The region‐of‐interest results carry a further qualification, since the enrichment test asks whether an effect is concentrated in this cortex relative to the rest of the brain and not whether the region‐averaged effect differs from zero (Section 2.6). The right‐lateralization of prosody processing is itself not settled. Meta‐analytic work on the neuroimaging literature finds emotional prosody to engage both hemispheres across temporal and frontal cortex, with consistent right‐sided convergence restricted to early auditory areas, and attributes that residual asymmetry to a right‐hemisphere advantage for acoustic dimensions of the speech signal that are not specific to prosody (Witteman et al. 2012). A direct test of functional lateralization yielded no evidence for it, with the authors concluding that lateralization effects in prosody perception are small (Witteman et al. 2014). The acoustic account bears directly on the present data, since the two deliveries differ in precisely those dimensions. Notably, the right‐perisylvian cluster reported here survived regression of fundamental frequency, spectral (log‐mel) components, the amplitude envelope and syllable rate from the regional time courses, and cannot therefore be attributed to a hemispheric preference for the acoustic form of the stimuli alone. Thus the present asymmetry should be read as consistent with one strand of this literature rather than as confirming it, although the account on which right‐lateralization is merely acoustic is the one these data fit least well.
The audience‐cue null carries a limitation of its own, since awareness of the cue was not assessed and the manipulation was administered between subjects, so a cue that was noticed and proved inconsequential cannot be distinguished from a cue that went unnoticed (Section 4.3). The gaze data suggest a mechanism for the latter possibility: viewers concentrated their fixations on the speaker's face, whereas the cue consisted of static silhouettes at the lower edge of the display, and a peripheral element of that kind may never have been foveated. Those data carry their own constraint, however, since they were collected in an independent sample without MEG and under a different viewing geometry, and cannot be regressed from the neural time courses (Section 3.5). Finally, because the attitude and social‐context manipulations were administered in sequence within a single session, they are not fully independent. We counterbalanced their order and verified that attitude did not carry over into the context task, but fatigue and time in the scanner may nevertheless have reduced sensitivity in the later task, which is the task that yielded the null.
That last consideration sets the most immediate priority for further work: the audience‐cue manipulation should be replicated in a fresh session, where it is not preceded by another task and where awareness of the cue is verified before its null is taken as a property of minimal social cues in general. A cue placed where listeners actually look, or one established by norming to be reliably noticed, would distinguish a genuinely inert manipulation from one that never reached the listener. Beyond replication, the present design manipulated who was speaking and how the listener was disposed while holding the linguistic content fixed; the complementary experiment would vary the predictability of the content itself while independently varying social and communicative factors, which would allow the predictive and the attentional accounts of the frequency profile to be disentangled. It would also be valuable to relate alignment to outcomes rather than to conditions, by asking whether the magnitude of shared processing predicts persuasion or convergence of memory for the message; group synchrony has already been shown to track engagement and interpersonal closeness in a real classroom (Dikker et al. 2017), and extending that logic to a controlled communicative manipulation is a natural next step. Such work will depend on the continued development of naturalistic paradigms and of the analytic tools they require (Sonkusare et al. 2019).
In sum, two experiments converge on an account of what inter‐subject correlation tracks during naturalistic communication. Shared processing was shaped by how the message was delivered and by how the listener was disposed toward the speaker, and these two influences converged on a common right‐perisylvian network through distinct frequency channels, whereas a minimal cue to social context did not engage it. Inter‐subject correlation therefore appears to reflect engagement with the message and with the speaker rather than social framing as such, and it carries, in addition, a stable and alpha‐specific trait component. With subject‐level statistics and control for low‐level features, frequency‐resolved inter‐subject correlation offers a tractable way of asking not merely whether listeners process a message together, but which of them do so, and through which channel.
Author Contributions
Barak Atia: data curation, project administration, investigation. Yair Berson: conceptualization, supervision, investigation. Abraham Goldstein: conceptualization, investigation, writing – original draft, supervision, resources, formal analysis, methodology.
Conflicts of Interest
The authors declare no conflicts of interest.
Supporting information
Table S1: Acoustic features of the two Study 1 clips, computed at matched presentation level.
Table S2: Visual features of the two Study 1 clips.
Table S3: Low‐level feature battery.
Table S4: Envelope coupling against a circular‐shift null (n = 40, pooled across conditions).
Table S5: Regions surviving both feature regression and aperiodic removal (beta band, one‐sided FDR). Values are the charismatic minus non‐charismatic ISC difference in Fisher‐z at each stage of control.
Table S6: Beta ISC condition effect retained after regressing each feature family, as a percentage of the uncontrolled effect, with each family's own phase‐scrambled null (mean of 20 draws).
Table S7: Convergence within the a priori right auditory‐perisylvian ROI, at every stage of control. Each cell gives the mean effect in the ROI and the exact enrichment p. The stage reported in the main text is shown in bold.
Table S7b: The ROI enrichment test against a contiguity‐matched null, each contrast at its reported stage.
Table S8: Regions surviving FDR correction for the attitude contrast (positive > negative). Effect is the group difference in leave‐one‐out ISC, Fisher‐z. Asterisk denotes regions also surviving two‐sided correction.
Table S9: Global inter‐subject correlation (72‐region mean, Fisher‐z) by condition and band, with the repeated‐measures ANOVA across conditions and the three post hoc contrasts (paired t, two‐sided p). n = 40.
Table S10: Spectral power across the three conditions (region‐averaged), as raw log power and as the periodic component isolated by spectral parameterization.
Acknowledgments
The authors would like to thank Yuval Harpaz, Assaf Paul Cohen, and Rotem Zecharia for their assistance in data collection.
Data Availability Statement
Analysis code is available on Zenodo at https://doi.org/10.5281/zenodo.22808469. The accompanying data deposit, available at https://doi.org/10.5281/zenodo.22808367, contains the source‐reconstructed regional time courses, stimulus features, questionnaire scores, and group statistics underlying every figure. Raw MEG data and the stimulus recordings are available on application, subject to ethics approval and actors' consent.
References
- Amodio, D. M. 2014. “The Neuroscience of Prejudice and Stereotyping.” Nature Reviews Neuroscience 15, no. 10: 670–682. [DOI] [PubMed] [Google Scholar]
- Antonakis, J. , Fenley M., and Liechti S.. 2011. “Can Charisma Be Taught? Tests of Two Interventions.” Academy of Management Learning & Education 10, no. 3: 374–396. [Google Scholar]
- Appelman, A. , and Sundar S. S.. 2016. “Measuring Message Credibility: Construction and Validation of an Exclusive Scale.” Journalism and Mass Communication Quarterly 93, no. 1: 59–79. [Google Scholar]
- Arnal, L. H. , and Giraud A. L.. 2012. “Cortical Oscillations and Sensory Predictions.” Trends in Cognitive Sciences 16, no. 7: 390–398. [DOI] [PubMed] [Google Scholar]
- Baevski, A. , Zhou Y., Mohamed A., and Auli M.. 2020. “Wav2vec 2.0: A Framework for Self‐Supervised Learning of Speech Representations.” Advances in Neural Information Processing Systems 33: 12449–12460. [Google Scholar]
- Banks, G. C. , Engemann K. N., Williams C. E., Gooty J., McCauley K. D., and Medaugh M. R.. 2017. “A Meta‐Analytic Review and Future Research Agenda of Charismatic Leadership.” Leadership Quarterly 28, no. 4: 508–529. [Google Scholar]
- Bass, B. M. , Avolio B. J., Jung D. I., and Berson Y.. 2003. “Predicting Unit Performance by Assessing Transformational and Transactional Leadership.” Journal of Applied Psychology 88, no. 2: 207–218. [DOI] [PubMed] [Google Scholar]
- Bastiaansen, M. , and Hagoort P.. 2003. “Event‐Induced Theta Responses as a Window on the Dynamics of Memory.” Cortex 39, no. 4–5: 967–992. [DOI] [PubMed] [Google Scholar]
- Bastos, A. M. , Vezoli J., and Fries P.. 2015. “Communication Through Coherence With Inter‐Areal Delays.” Current Opinion in Neurobiology 31: 173–180. [DOI] [PubMed] [Google Scholar]
- Belin, P. , Zatorre R. J., Lafaille P., Ahad P., and Pike B.. 2000. “Voice‐Selective Areas in Human Auditory Cortex.” Nature 403, no. 6767: 309–312. [DOI] [PubMed] [Google Scholar]
- Belyk, M. , and Brown S.. 2014. “Perception of Affective and Linguistic Prosody: An ALE Meta‐Analysis of Neuroimaging Studies.” Social Cognitive and Affective Neuroscience 9, no. 9: 1395–1403. [DOI] [PMC free article] [PubMed] [Google Scholar]
- Berson, Y. , and Avolio B. J.. 2004. “Transformational Leadership and the Dissemination of Organizational Goals: A Case Study of a Telecommunication Firm.” Leadership Quarterly 15, no. 5: 625–646. [Google Scholar]
- Chaiken, S. 1980. “Heuristic Versus Systematic Information Processing and the Use of Source Versus Message Cues in Persuasion.” Journal of Personality and Social Psychology 39, no. 5: 752–766. [Google Scholar]
- Chalas, N. , Meyer L., Lo C. W., et al. 2024. “Dissociating Prosodic From Syntactic Delta Activity During Natural Speech Comprehension.” Current Biology 34, no. 15: 3537–3549.e5. [DOI] [PubMed] [Google Scholar]
- Chen, G. , and Bliese P. D.. 2002. “The Role of Different Levels of Leadership in Predicting Self‐And Collective Efficacy: Evidence for Discontinuity.” Journal of Applied Psychology 87, no. 3: 549–556. [DOI] [PubMed] [Google Scholar]
- Chen, G. , Taylor P. A., Shin Y.‐W., Reynolds R. C., and Cox R. W.. 2017. “Untangling the Relatedness Among Correlations, Part II: Inter‐Subject Correlation Group Analysis Through Linear Mixed‐Effects Modeling.” NeuroImage 147: 825–840. [DOI] [PMC free article] [PubMed] [Google Scholar]
- Cohen, S. S. , and Parra L. C.. 2016. “Memorable Audiovisual Narratives Synchronize Sensory and Supramodal Neural Responses.” Eneuro 3, no. 6: ENEURO.0203‐16.2016. [DOI] [PMC free article] [PubMed] [Google Scholar]
- Conger, J. A. , and Kanungo R. N.. 1994. “Charismatic Leadership in Organizations: Perceived Behavioral Attributes and Their Measurement.” Journal of Organizational Behavior 15, no. 5: 439–452. [Google Scholar]
- de Bruin, D. , and FeldmanHall O.. 2025. “Politically Extreme Individuals Exhibit Similar Neural Processing Despite Ideological Differences.” Journal of Personality and Social Psychology 129, no. 5: 816–833. [DOI] [PubMed] [Google Scholar]
- de Bruin, D. , van Baar J. M., Rodríguez P. L., and FeldmanHall O.. 2023. “Shared Neural Representations and Temporal Segmentation of Political Content Predict Ideological Similarity.” Science Advances 9, no. 5: eabq5920. [DOI] [PMC free article] [PubMed] [Google Scholar]
- De Felice, S. , Hakim U., Gunasekara N., Pinti P., Tachtsidis I., and Hamilton A.. 2024. “Having a Chat and Then Watching a Movie: How Social Interaction Synchronises Our Brains During Co‐Watching.” Oxford Open Neuroscience 3: kvae006. [DOI] [PMC free article] [PubMed] [Google Scholar]
- Deen, B. , Koldewyn K., Kanwisher N., and Saxe R.. 2015. “Functional Organization of Social Perception and Cognition in the Superior Temporal Sulcus.” Cerebral Cortex 25, no. 11: 4596–4609. [DOI] [PMC free article] [PubMed] [Google Scholar]
- Delorme, A. , and Makeig S.. 2004. “EEGLAB: An Open Source Toolbox for Analysis of Single‐Trial EEG Dynamics Including Independent Component Analysis.” Journal of Neuroscience Methods 134, no. 1: 9–21. [DOI] [PubMed] [Google Scholar]
- Dewhurst, R. , Nyström M., Jarodzka H., Foulsham T., Johansson R., and Holmqvist K.. 2012. “It Depends on How You Look at It: Scanpath Comparison in Multiple Dimensions With Multimatch, a Vector‐Based Approach.” Behavior Research Methods 44, no. 4: 1079–1100. [DOI] [PubMed] [Google Scholar]
- Dikker, S. , Wan L., Davidesco I., et al. 2017. “Brain‐To‐Brain Synchrony Tracks Real‐World Dynamic Group Interactions in the Classroom.” Current Biology 27, no. 9: 1375–1380. [DOI] [PubMed] [Google Scholar]
- Ding, N. , and Simon J. Z.. 2014. “Cortical Entrainment to Continuous Speech: Functional Roles and Interpretations.” Frontiers in Human Neuroscience 8: 311. [DOI] [PMC free article] [PubMed] [Google Scholar]
- Dmochowski, J. P. , Bezdek M. A., Abelson B. P., Johnson J. S., Schumacher E. H., and Parra L. C.. 2014. “Audience Preferences Are Predicted by Temporal Reliability of Neural Processing.” Nature Communications 5, no. 1: 4567. [DOI] [PMC free article] [PubMed] [Google Scholar]
- Dmochowski, J. P. , Sajda P., Dias J., and Parra L. C.. 2012. “Correlated Components of Ongoing Eeg Point to Emotionally Laden Attention–A Possible Marker of Engagement?” Frontiers in Human Neuroscience 6: 112. [DOI] [PMC free article] [PubMed] [Google Scholar]
- Donoghue, T. , Haller M., Peterson E. J., et al. 2020. “Parameterizing Neural Power Spectra Into Periodic and Aperiodic Components.” Nature Neuroscience 23, no. 12: 1655–1665. [DOI] [PMC free article] [PubMed] [Google Scholar]
- Engel, A. K. , and Fries P.. 2010. “Beta‐Band Oscillations—Signalling the Status Quo?” Current Opinion in Neurobiology 20, no. 2: 156–165. [DOI] [PubMed] [Google Scholar]
- Finn, E. S. , Glerean E., Khojandi A. Y., et al. 2020. “Idiosynchrony: From Shared Responses to Individual Differences During Naturalistic Neuroimaging.” NeuroImage 215: 116828. [DOI] [PMC free article] [PubMed] [Google Scholar]
- Foxe, J. J. , and Snyder A. C.. 2011. “The Role of Alpha‐Band Brain Oscillations as a Sensory Suppression Mechanism During Selective Attention.” Frontiers in Psychology 2: 154. [DOI] [PMC free article] [PubMed] [Google Scholar]
- Fries, P. 2015. “Rhythms for Cognition: Communication Through Coherence.” Neuron 88, no. 1: 220–235. [DOI] [PMC free article] [PubMed] [Google Scholar]
- Giraud, A. L. , and Poeppel D.. 2012. “Cortical Oscillations and Speech Processing: Emerging Computational Principles and Operations.” Nature Neuroscience 15, no. 4: 511–517. [DOI] [PMC free article] [PubMed] [Google Scholar]
- Golland, Y. , Arzouan Y., and Levit‐Binnun N.. 2015. “The Mere Co‐Presence: Synchronization of Autonomic Signals and Emotional Responses Across Co‐Present Individuals Not Engaged in Direct Interaction.” PLoS One 10, no. 5: e0125804. [DOI] [PMC free article] [PubMed] [Google Scholar]
- Golland, Y. , Levit‐Binnun N., Hendler T., and Lerner Y.. 2017. “Neural Dynamics Underlying Emotional Transmissions Between Individuals.” Social Cognitive and Affective Neuroscience 12, no. 8: 1249–1260. [DOI] [PMC free article] [PubMed] [Google Scholar]
- Hasson, U. , Ghazanfar A. A., Galantucci B., Garrod S., and Keysers C.. 2012. “Brain‐To‐Brain Coupling: A Mechanism for Creating and Sharing a Social World.” Trends in Cognitive Sciences 16, no. 2: 114–121. [DOI] [PMC free article] [PubMed] [Google Scholar]
- Hasson, U. , Landesman O., Knappmeyer B., Vallines I., Rubin N., and Heeger D. J.. 2008. “Neurocinematics: The Neuroscience of Film.” PRO 2, no. 1: 1–26. [Google Scholar]
- Hasson, U. , Nir Y., Levy I., Fuhrmann G., and Malach R.. 2004. “Intersubject Synchronization of Cortical Activity During Natural Vision.” Science 303, no. 5664: 1634–1640. [DOI] [PubMed] [Google Scholar]
- Hickok, G. , and Poeppel D.. 2007. “The Cortical Organization of Speech Processing.” Nature Reviews Neuroscience 8, no. 5: 393–402. [DOI] [PubMed] [Google Scholar]
- Hoekert, M. , Bais L., Kahn R. S., and Aleman A.. 2008. “Time Course of the Involvement of the Right Anterior Superior Temporal Gyrus and the Right Fronto‐Parietal Operculum in Emotional Prosody Perception.” PLoS One 3, no. 5: e2244. [DOI] [PMC free article] [PubMed] [Google Scholar]
- Holle, H. , Obleser J., Rueschemeyer S. A., and Gunter T. C.. 2010. “Integration of Iconic Gestures and Speech in Left Superior Temporal Areas Boosts Speech Comprehension Under Adverse Listening Conditions.” NeuroImage 49, no. 1: 875–884. [DOI] [PubMed] [Google Scholar]
- Honey, C. J. , Thompson C. R., Lerner Y., and Hasson U.. 2012. “Not Lost in Translation: Neural Responses Shared Across Languages.” Journal of Neuroscience 32, no. 44: 15277–15283. [DOI] [PMC free article] [PubMed] [Google Scholar]
- Hughes, B. L. , and Zaki J.. 2015. “The Neuroscience of Motivated Cognition.” Trends in Cognitive Sciences 19, no. 2: 62–64. [DOI] [PubMed] [Google Scholar]
- Imhof, M. A. , Schmälzle R., Renner B., and Schupp H. T.. 2017. “How Real‐Life Health Messages Engage Our Brains: Shared Processing of Effective Anti‐Alcohol Videos.” Social Cognitive and Affective Neuroscience 12, no. 7: 1188–1196. [DOI] [PMC free article] [PubMed] [Google Scholar]
- Jacoby, N. , Landau‐Wells M., Pearl J., et al. 2024. “Partisans Process Policy‐Based and Identity‐Based Messages Using Dissociable Neural Systems.” Cerebral Cortex 34, no. 9: bhae368. [DOI] [PubMed] [Google Scholar]
- Jost, J. T. , Federico C. M., and Napier J. L.. 2009. “Political Ideology: Its Structure, Functions, and Elective Affinities.” Annual Review of Psychology 60: 307–337. [DOI] [PubMed] [Google Scholar]
- Ki, J. J. , Kelly S. P., and Parra L. C.. 2016. “Attention Strongly Modulates Reliability of Neural Responses to Naturalistic Narrative Stimuli.” Journal of Neuroscience 36, no. 10: 3092–3101. [DOI] [PMC free article] [PubMed] [Google Scholar]
- Klimesch, W. 2012. “Alpha‐Band Oscillations, Attention, and Controlled Access to Stored Information.” Trends in Cognitive Sciences 16, no. 12: 606–617. [DOI] [PMC free article] [PubMed] [Google Scholar]
- Kotz, S. A. , and Schwartze M.. 2010. “Cortical Speech Processing Unplugged: A Timely Subcortico‐Cortical Framework.” Trends in Cognitive Sciences 14, no. 9: 392–399. [DOI] [PubMed] [Google Scholar]
- Kumkale, G. T. , Albarracín D., and Seignourel P. J.. 2010. “The Effects of Source Credibility in the Presence or Absence of Prior Attitudes: Implications for the Design of Persuasive Communication Campaigns.” Journal of Applied Social Psychology 40, no. 6: 1325–1356. [DOI] [PMC free article] [PubMed] [Google Scholar]
- Lankinen, K. , Saari J., Hari R., and Koskinen M.. 2014. “Intersubject Consistency of Cortical Meg Signals During Movie Viewing.” NeuroImage 92: 217–224. [DOI] [PubMed] [Google Scholar]
- Lankinen, K. , Saari J., Hlushchuk Y., et al. 2018. “Consistency and Similarity of Meg‐And Fmri‐Signal Time Courses During Movie Viewing.” NeuroImage 173: 361–369. [DOI] [PubMed] [Google Scholar]
- Lerner, Y. , Honey C. J., Silbert L. J., and Hasson U.. 2011. “Topographic Mapping of a Hierarchy of Temporal Receptive Windows Using a Narrated Story.” Journal of Neuroscience 31, no. 8: 2906–2915. [DOI] [PMC free article] [PubMed] [Google Scholar]
- Lewis, A. G. , and Bastiaansen M.. 2015. “A Predictive Coding Framework for Rapid Neural Dynamics During Sentence‐Level Language Comprehension.” Cortex 68: 155–168. [DOI] [PubMed] [Google Scholar]
- Luo, H. , and Poeppel D.. 2007. “Phase Patterns of Neuronal Responses Reliably Discriminate Speech in Human Auditory Cortex.” Neuron 54, no. 6: 1001–1010. [DOI] [PMC free article] [PubMed] [Google Scholar]
- Marstaller, L. , and Burianová H.. 2014. “The Multisensory Perception of Co‐Speech Gestures–a Review and Meta‐Analysis of Neuroimaging Studies.” Journal of Neurolinguistics 30: 69–77. [Google Scholar]
- McAllister, D. J. 1995. “Affect‐And Cognition‐Based Trust as Foundations for Interpersonal Cooperation in Organizations.” Academy of Management Journal 38, no. 1: 24–59. [Google Scholar]
- Nastase, S. A. , Gazzola V., Hasson U., and Keysers C.. 2019. “Measuring Shared Responses Across Subjects Using Intersubject Correlation.” Social Cognitive and Affective Neuroscience 14, no. 6: 667–685. [DOI] [PMC free article] [PubMed] [Google Scholar]
- Niebuhr, O. , Tegtmeier S., and Brem A.. 2017. “Advancing Research and Practice in Entrepreneurship Through Speech Analysis: From Descriptive Rhetorical Terms to Phonetically Informed Acoustic Charisma Metrics.” Journal of Speech Sciences 6, no. 1: 3–26. [Google Scholar]
- Nummenmaa, L. , Lahnakoski J. M., and Glerean E.. 2018. “Sharing the Social World via Intersubject Neural Synchronisation.” Current Opinion in Psychology 24: 7–14. [DOI] [PubMed] [Google Scholar]
- Obleser, J. , and Kayser C.. 2019. “Neural Entrainment and Attentional Selection in the Listening Brain.” Trends in Cognitive Sciences 23, no. 11: 913–926. [DOI] [PubMed] [Google Scholar]
- Oderbolz, C. , Stark E., Sauppe S., and Meyer M.. 2024. “Concurrent Processing of the Prosodic Hierarchy Is Supported by Cortical Entrainment and Phase‐Amplitude Coupling.” Cerebral Cortex 34, no. 12: bhae479. [DOI] [PMC free article] [PubMed] [Google Scholar]
- Oganian, Y. , and Chang E. F.. 2019. “A Speech Envelope Landmark for Syllable Encoding in Human Superior Temporal Gyrus.” Science Advances 5, no. 11: eaay6279. [DOI] [PMC free article] [PubMed] [Google Scholar]
- Oldfield, R. C. 1971. “The Assessment and Analysis of Handedness: The Edinburgh Inventory.” Neuropsychologia 9, no. 1: 97–113. [DOI] [PubMed] [Google Scholar]
- Oostenveld, R. , Fries P., Maris E., and Schoffelen J. M.. 2011. “Fieldtrip: Open Source Software for Advanced Analysis of Meg, Eeg, and Invasive Electrophysiological Data.” Computational Intelligence and Neuroscience 2011, no. 1: 156869. [DOI] [PMC free article] [PubMed] [Google Scholar]
- Petty, R. E. , and Cacioppo J. T.. 1986. “The Elaboration Likelihood Model of Persuasion.” In Advances in Experimental Social Psychology, vol. 19, 123–205. Academic Press. [Google Scholar]
- Pickering, M. J. , and Garrod S.. 2004. “Toward a Mechanistic Psychology of Dialogue.” Behavioral and Brain Sciences 27, no. 2: 169–190. [DOI] [PubMed] [Google Scholar]
- Redcay, E. , Velnoskey K. R., and Rowe M. L.. 2016. “Perceived Communicative Intent in Gesture and Language Modulates the Superior Temporal Sulcus.” Human Brain Mapping 37, no. 10: 3444–3461. [DOI] [PMC free article] [PubMed] [Google Scholar]
- Regev, M. , Honey C. J., Simony E., and Hasson U.. 2013. “Selective and Invariant Neural Responses to Spoken and Written Narratives.” Journal of Neuroscience 33, no. 40: 15978–15988. [DOI] [PMC free article] [PubMed] [Google Scholar]
- Sammler, D. , Grosbras M.‐H., Anwander A., Bestelmeyer P. E. G., and Belin P.. 2015. “Dorsal and Ventral Pathways for Prosody.” Current Biology 25, no. 23: 3079–3085. [DOI] [PubMed] [Google Scholar]
- Schmälzle, R. , Häcker F. E. K., Honey C. J., and Hasson U.. 2015. “Engaged Listeners: Shared Neural Processing of Powerful Political Speeches.” Social Cognitive and Affective Neuroscience 10, no. 8: 1137–1143. [DOI] [PMC free article] [PubMed] [Google Scholar]
- Shteynberg, G. 2015. “Shared Attention.” Perspectives on Psychological Science 10, no. 5: 579–590. [DOI] [PubMed] [Google Scholar]
- Shteynberg, G. , Bramlett J. M., Fles E. H., and Cameron J.. 2016. “The Broadcast of Shared Attention and Its Impact on Political Persuasion.” Journal of Personality and Social Psychology 111, no. 5: 665–673. [DOI] [PubMed] [Google Scholar]
- Sonkusare, S. , Breakspear M., and Guo C.. 2019. “Naturalistic Stimuli in Neuroscience: Critically Acclaimed.” Trends in Cognitive Sciences 23, no. 8: 699–714. [DOI] [PubMed] [Google Scholar]
- Stephens, G. J. , Silbert L. J., and Hasson U.. 2010. “Speaker–Listener Neural Coupling Underlies Successful Communication.” Proceedings of the National Academy of Sciences 107, no. 32: 14425–14430. [DOI] [PMC free article] [PubMed] [Google Scholar]
- Tal, I. , and Abeles M.. 2013. “Cleaning Meg Artifacts Using External Cues.” Journal of Neuroscience Methods 217, no. 1–2: 31–38. [DOI] [PubMed] [Google Scholar]
- Thiede, A. , Glerean E., Kujala T., and Parkkonen L.. 2020. “Atypical Meg Inter‐Subject Correlation During Listening to Continuous Natural Speech in Dyslexia.” NeuroImage 216: 116799. [DOI] [PubMed] [Google Scholar]
- Todorov, A. , Olivola C. Y., Dotsch R., and Mende‐Siedlecki P.. 2015. “Social Attributions From Faces: Determinants, Consequences, Accuracy, and Functional Significance.” Annual Review of Psychology 66, no. 1: 519–545. [DOI] [PubMed] [Google Scholar]
- Tzourio‐Mazoyer, N. , Landeau B., Papathanassiou D., et al. 2002. “Automated Anatomical Labeling of Activations in SPM Using a Macroscopic Anatomical Parcellation of the MNI MRI Single‐Subject Brain.” NeuroImage 15, no. 1: 273–289. [DOI] [PubMed] [Google Scholar]
- van Baar, J. M. , Halpern D. J., and FeldmanHall O.. 2021. “Intolerance of Uncertainty Modulates Brain‐To‐Brain Synchrony During Politically Polarized Perception.” Proceedings of the National Academy of Sciences of the United States of America 118, no. 20: e2022491118. [DOI] [PMC free article] [PubMed] [Google Scholar]
- van Rijn, S. , Aleman A., van Diessen E., Berckmoes C., Vingerhoets G., and Kahn R. S.. 2005. “What Is Said or How It Is Said Makes a Difference: Role of the Right Fronto‐Parietal Operculum in Emotional Prosody as Revealed by Repetitive TMS.” European Journal of Neuroscience 21, no. 11: 3195–3200. [DOI] [PubMed] [Google Scholar]
- Van Strien, J. L. H. , Kammerer Y., Brand‐Gruwel S., and Boshuizen H. P.. 2016. “How Attitude Strength Biases Information Processing and Evaluation on the Web.” Computers in Human Behavior 60: 245–252. [Google Scholar]
- Vigneau, M. , Beaucousin V., Hervé P. Y., et al. 2006. “Meta‐Analyzing Left Hemisphere Language Areas: Phonology, Semantics, and Sentence Processing.” NeuroImage 30, no. 4: 1414–1432. [DOI] [PubMed] [Google Scholar]
- Watson, D. , Clark L. A., and Tellegen A.. 1988. “Development and Validation of Brief Measures of Positive and Negative Affect: The PANAS Scales.” Journal of Personality and Social Psychology 54, no. 6: 1063–1070. [DOI] [PubMed] [Google Scholar]
- Weiss, S. , and Mueller H. M.. 2012. ““Too Many Betas Do Not Spoil the Broth”: The Role of Beta Brain Oscillations in Language Processing.” Frontiers in Psychology 3: 201. [DOI] [PMC free article] [PubMed] [Google Scholar]
- Wheatley, T. , Kang O., Parkinson C., and Looser C. E.. 2012. “From Mind Perception to Mental Connection: Synchrony as a Mechanism for Social Understanding.” Social and Personality Psychology Compass 6, no. 8: 589–606. [Google Scholar]
- Wilson, S. M. , Molnar‐Szakacs I., and Iacoboni M.. 2008. “Beyond Superior Temporal Cortex: Intersubject Correlations in Narrative Speech Comprehension.” Cerebral Cortex 18, no. 1: 230–242. [DOI] [PubMed] [Google Scholar]
- Witteman, J. , Goerlich‐Dobre K. S., Martens S., Aleman A., Van Heuven V. J., and Schiller N. O.. 2014. “The Nature of Hemispheric Specialization for Prosody Perception.” Cognitive, Affective, & Behavioral Neuroscience 14, no. 3: 1104–1114. [DOI] [PubMed] [Google Scholar]
- Witteman, J. , van Heuven V. J. P., and Schiller N. O.. 2012. “Hearing Feelings: A Quantitative Meta‐Analysis on the Neuroimaging Literature of Emotional Prosody Perception.” Neuropsychologia 50, no. 12: 2752–2763. [DOI] [PubMed] [Google Scholar]
Associated Data
This section collects any data citations, data availability statements, or supplementary materials included in this article.
Supplementary Materials
Table S1: Acoustic features of the two Study 1 clips, computed at matched presentation level.
Table S2: Visual features of the two Study 1 clips.
Table S3: Low‐level feature battery.
Table S4: Envelope coupling against a circular‐shift null (n = 40, pooled across conditions).
Table S5: Regions surviving both feature regression and aperiodic removal (beta band, one‐sided FDR). Values are the charismatic minus non‐charismatic ISC difference in Fisher‐z at each stage of control.
Table S6: Beta ISC condition effect retained after regressing each feature family, as a percentage of the uncontrolled effect, with each family's own phase‐scrambled null (mean of 20 draws).
Table S7: Convergence within the a priori right auditory‐perisylvian ROI, at every stage of control. Each cell gives the mean effect in the ROI and the exact enrichment p. The stage reported in the main text is shown in bold.
Table S7b: The ROI enrichment test against a contiguity‐matched null, each contrast at its reported stage.
Table S8: Regions surviving FDR correction for the attitude contrast (positive > negative). Effect is the group difference in leave‐one‐out ISC, Fisher‐z. Asterisk denotes regions also surviving two‐sided correction.
Table S9: Global inter‐subject correlation (72‐region mean, Fisher‐z) by condition and band, with the repeated‐measures ANOVA across conditions and the three post hoc contrasts (paired t, two‐sided p). n = 40.
Table S10: Spectral power across the three conditions (region‐averaged), as raw log power and as the periodic component isolated by spectral parameterization.
Data Availability Statement
Analysis code is available on Zenodo at https://doi.org/10.5281/zenodo.22808469. The accompanying data deposit, available at https://doi.org/10.5281/zenodo.22808367, contains the source‐reconstructed regional time courses, stimulus features, questionnaire scores, and group statistics underlying every figure. Raw MEG data and the stimulus recordings are available on application, subject to ethics approval and actors' consent.