The Chorus Effect: Narratives and Resonance on Dark Telegram

Team Members: Marc Tuters (UvA), Tom Willaert (VUB), Boris Noordenbos (UvA), Hanna Lauvli, Jurij Smrke, Alexandra Barancová, Lamia Putri Damayanti, Ray Dolitsay, Pepijn Stoop, Wilma Ewerhart, Furkan Dabaniyasti

View posters

Summary of Key Findings

Our analysis of a large, multilingual corpus of conspiratorial and geopolitically charged Telegram channels—drawn from over 4,000 channels and more than 30 million posts collected between mid-2022 and mid-2024—reveals a loosely synchronised “chorus” of accounts. These range from quasi-news outlets to influencer feeds to meme hubs, and they pick up, adapt, and circulate shared refrains. This distributed amplification produces coherence without direct coordination, making the discourse resilient to disruption across languages and national contexts.

Within this chorus, certain narrative frames recur with striking regularity: the West as morally corrupt and in irreversible decline, Ukraine as illegitimate and Nazi-controlled, and Russia or other counter-hegemonic actors as defenders of sovereignty and traditional values. These are interwoven with longer-standing conspiracist tropes—anti-NATO sentiment, historical revisionism, anti-globalist rhetoric, and claims of humanitarian crisis—allowing new events, whether political developments or battlefield incidents, to be absorbed seamlessly into an ongoing storyline.

Emotional appeal is central: outrage, pride, and resentment are sustained through memes and emoji as rhythmic affective anchors, and through capitalisation as a vernacular mode of urgency. The network’s architecture supports both scale and specialisation, with high-volume hubs generating core content and smaller channels remixing it for subcultural or linguistic niches, preserving thematic fidelity while expanding reach. Certain frames, such as “Ukraine-as-Nazi,” show remarkable persistence, resurfacing with each new incident to maintain continuity.

The “chorus effect” emerges not only in textual repetition but in its temporal rhythm, which we rendered audible by mapping keyword and emoji spikes to sound. The resulting resonant soundscape captures peaks of discourse swelling like refrains, making tangible the communicative logic at play: a decentralised yet synchronised system whose strength lies as much in its affective harmonies and infrastructural tempo as in its propositional content.

1. Introduction

This report investigates the dynamics of conspiratorial and geopolitically charged discourse on Telegram, using a longitudinal dataset collected between June 2022 and June 2024. The dataset—comprising 4,082 unique channels and more than 30 million posts—was assembled through a combination of keyword-based seeding and snowball sampling from Telegram’s internal channel-link graph. This method ensured coverage of multiple languages (EN, RU, DE, NL, and others) and channel types, including news-style outlets, influencer accounts, activist groups, meme hubs, and niche communities focused on conspiratorial, extremist, and geopolitical themes.

The initial seed consisted of two Dutch-language channels linked to the populist-right parties Forum for Democracy (Netherlands) and Vlaams Belang (Belgium). From there, we used Telegram’s “recommended channels” feature—a platform-native relational mapping tool—to recursively expand the network to ideologically adjacent spaces. The result is a corpus that spans dozens of languages, national contexts, and thematic preoccupations, allowing for cross-cultural and cross-linguistic comparison of discursive patterns.

From this corpus, multilingual keyword filtering was applied to generate thematic “bins” aligned with dominant discursive currents. Two bins form the focus of this report. The first captures the conspiratorial imaginary, especially contemporary versions of the “one world government” narrative—encompassing globalist elites, climate policy conspiracies, digital identity systems, the World Economic Forum, and the perceived erosion of sovereignty. The second focuses on the Russo-Ukrainian war, tracing how the conflict was narrated, instrumentalised, and mythologised. These domains often overlap, sharing symbolic repertoires and affective intensities that reflect a broader politics of suspicion.

Figure 1: thumbnails from conspiracy, including Dugin’s and Bannon’s channels

We frame these dynamics through the concept of the reactionary digital chorus. Drawing inspiration from the collective voice of Greek tragedy, the term describes a decentralised yet synchronised formation in which narratives emerge not through central control, but through repetition, symbolic mirroring, and infrastructural resonance across distributed actors. Like the tragic chorus, this formation both comments on events and shapes their meaning, its voice sustained through the rhythmic recurrence of shared refrains. In this mode, persuasion operates less through rational argument and more through affective saturation, emotional contagion, and symbolic overload—a process intensified by platform affordances that blur the line between organic uptake and inauthentic amplification.

In the contemporary digital environment, the chorus effect is amplified by Telegram’s affordances: pseudonymous identities, minimal moderation, a recommendation graph, and a broadcast architecture that together foster a populist media style privileging virality, thematic convergence, and moral dualism over ideological consistency. These features enable content to be both persistent—resurfacing over long durations—and adaptive, absorbing new events seamlessly into existing frames. Truth claims often intermingle with irony, esotericism, or religious invocation, collapsing disparate events into conspiratorial narratives. Within the conspiracy bin, climate politics frequently functions as a metonym for elite control, with tropes such as “climate scam,” “net zero,” and “Agenda 2030” bundled alongside fears of depopulation, CBDCs, and digital IDs.

The period from mid-2022 to mid-2024 encompasses heightened geopolitical tensions and institutional crises, notably around Ukraine and climate governance. Across this terrain, memes, slogans, and motifs circulate fluidly between languages and cultural registers, forming what we describe as a post-linguistic symbolic field. To account for this phenomenon, we introduce the concept of infrastructural populism. By this we mean a political modality in which the form, persistence, and reach of discourse are shaped less by centralised leadership or doctrinal coherence than by the affordances and rhythms of the media infrastructure itself. Telegram’s pseudonymous accounts, broadcast architecture, recommendation graph, and minimal moderation do not simply transmit populist messages; they actively structure a communicative environment that privileges thematic convergence, emotional contagion, and the rapid reframing of events into existing antagonisms.

The reactionary digital chorus is one expression of this infrastructural populism: a decentralised yet rhythmically aligned assemblage of channels that both comment on and help constitute unfolding events. Here, political meaning is carried as much by the recurrence, timing, and affective charge of shared refrains as by their propositional content. Antagonisms are sustained not through the slow work of persuasion but through continuous, infrastructurally-enabled resonance—a dynamic that blurs the distinction between organic uptake and inauthentic amplification. In this setting, narrative participation takes on a distinct form. Rather than advancing a coherent plot, participants sustain a shared antagonistic atmosphere by re-voicing familiar refrains, inserting recognisable motifs, and cycling through role positions embedded in the discourse—a distributed, affectively charged practice that erases the distinction between audience and author.

This report is therefore both an empirical mapping and a conceptual intervention. It examines how networked publics organise around suspicion, affect, and narrative fragments in an era of ambient political crisis, and proposes infrastructural populism as a lens for understanding the media-technical conditions that make the chorus possible.

2. Research Questions

*How does the “chorus” metaphor help to illuminate the relationship between narrative meaning-making and affective resonance, and to distinguish between inauthentic coordination and organic, emergent amplification?

* How do distinct choruses—commenting on the same events or thematic domains—form, interact, and influence each other within a distributed Telegram ecosystem?

3. Methodology

The dataset analysed in this report was assembled over a two-year period, between June 2022 and June 2024, and comprises 4,082 unique Telegram channels containing more than 30 million posts. Our aim was to capture a broad cross-section of conspiratorial and geopolitically charged discourse across multiple national contexts, languages, and subcultural milieus.

Data collection began with two Dutch-language channels linked to the populist-right parties Forum for Democracy in the Netherlands and Vlaams Belang in Belgium. From this seed, we employed Telegram’s built-in “recommended channels” feature—a platform-native relational mapping tool—to recursively expand the network. This snowball sampling process ensured coverage of ideologically adjacent channels, gradually producing a corpus that spans dozens of languages, including English, Russian, German, and Dutch, as well as a wide range of channel types. These include quasi-news outlets, influencer accounts, activist and campaign-focused groups, meme hubs, and smaller niche communities devoted to conspiratorial, extremist, or geopolitical themes.

Once collected, the dataset was segmented using multilingual keyword filtering to generate thematic “bins” corresponding to dominant discursive currents. Two of these bins form the focus of this report: one centred on the conspiratorial imaginary—particularly contemporary versions of the “one world government” narrative—and the other on the Russo-Ukrainian war. These domains often overlap, sharing symbolic repertoires, emotive registers, and rhetorical devices that reflect a broader politics of suspicion.

This project combines computational, linguistic, and artistic approaches to examine the chorus effect in such discourse. The methodology was designed to bridge large-scale quantitative mapping with experimental, affect-oriented techniques, tracing not only what is said but how it resonates over time and across networks. Structured as a series of complementary inquiries into the same dataset, it moves from identifying patterns in affective emoji use, to mapping recurring narrative anchors, to translating these patterns into sound. Each layer offers a distinct lens on the central question: how does a decentralised network of voices produce a shared discursive atmosphere?

Use of affective object emoji

To capture the non-verbal dimension of the chorus effect, we examined how emojis function as affective cues that punctuate and synchronise discourse across channels. Several computational quantitative analyses were launched using 4CAT and Python on the English and German language conspiracy datasets to observe the most commonly observed emoji in the Telegram channels. Emoji were treated here as proxies for affective signalling—small, repeatable units of feeling that punctuate text and serve as recognisable cues to the in-group. Interesting emoji for analysis were chosen based on which were most frequently used and which were affectively relevant. Notably, the prevalence of the use of the fire, explosion and exclamation emoji were observed over time in the German and English language datasets.

Emoji can be viewed as a digital-native affective vernacular. Their memetic nature allows networks of affective circulation to be mapped and tracked over time. On Telegram, particular emojis recur regularly, signifying a choral dimension to their movement. To analyze this distribution over time, a dataset of English-language Telegram posts related to conspiracy discourse was collected, covering the period from June 1, 2022, to June 30, 2024. The data was tokenized by month using 4Cat. From this, the top 500 words per month were extracted to identify the most frequently recurring emojis. The focus was deliberately narrowed to affective emojis, those signaling emotion or moo, excluding non-emotive symbols such as national flags. The dataset was then filtered to include only the selected emojis, allowing for a month-by-month visualization of usage frequency. This filtered data was subsequently visualized using RAWGraphs 2.0, resulting in the diagram shown below.

Choral refrains: “control,” “agenda,” and “globalist”.

To examine the verbal dimension of the chorus effect, we identified recurring keywords that serve as narrative anchors across disparate channels. We preprocessed all posts using standard NLP techniques—tokenization, stemming, and lemmatization—to group morphologically related terms, then aggregated token frequencies over time to track the emergence, persistence, and temporal clustering of these keywords across the corpus.

From this exploratory phase, the concept of “control” emerged as a particularly salient motif—appearing in a range of contexts, from claims of elite overreach to narratives about digital surveillance, pandemic governance, and technocratic power. We subsequently constructed a focused subcorpus of channels that frequently used the terms “control,” “agenda,” and “globalist.” These terms served as anchors for tracing what we interpret as refrains—recurring narrative fragments that, while not centrally coordinated, accumulate over time and across channels to form a shared discursive atmosphere.

Our hypothesis is that these refrains function as choral cues in the reactionary digital chorus: symbolic motifs that invite resonance, echo, and repetition. By mapping the proliferation of these terms across time and space, we aim to show how the effect of a collective voice emerges not through explicit alignment or ideological uniformity, but through distributed repetition and infrastructural amplification. This methodological approach thus links computational text analysis to a broader media-theoretical frame, foregrounding how stylistic cohesion and symbolic convergence operate in loosely networked, populist-aligned Telegram spaces. This step grounds the textual analysis in the broader media-theoretical frame of the “chorus effect,” connecting computational outputs with interpretive insight.

Data Sonification

To translate the chorus effect into an auditory form, we developed a data sonification process built directly on the outputs of our computational text and emoji analyses. Drawing from the Fantastic Little Splash’s work titled infocry, a chorus machine prototype, this project aims to map how affective resonance and narrative meaning-making can be understood in terms of sound.

Drawing from the Fantastic Little Splash’s work titled ‘infocry’, a chorus machine prototype, this project aims to map how affective resonance and narrative meaning-making can be understood in terms of sound. This approach integrates natural language data, CSV-to-MIDI transformation, and digital audio workstations to render the textual frequency as a sonic experience. Using the conspiracy dataset in English, the top three used words were delineated using 4CAT’s tokenisation function: “globalist”, “people” and “wef”. Over-time word counts were then produced to map the temporal distribution of those high-frequency words. The most used emoji, the fire emoji (🔥), was also included as a fourth input in this process (the process of attaining the over-time frequency for emojis is delineated previously).

The CSV files for each separate word were converted to MIDI files using the CSV to MIDI converter (https://csv-to-midi.evanking.io). In this transformation, rows correspond to time (mapped to musical bars), and word frequencies are mapped to MIDI note velocities or pitches. The resulting file allows for the experience of the choral progression of data in time. The data sonification was executed in Ableton Live 12 (trial version), with Max for Live installed to enable advanced MIDI modulation and parameter mapping.

Each keyword was rendered into a vocal audio sample using an AI-based text-to-speech generator, LuvVoice. The resulting audio clips were prepared for use in a sequencer. Two MIDI tracks were created: Track 1 was titled MIDI Globalist and Track 2 was titled Audio Globalist. The MIDI file was imported into both tracks. On Audio Globalist, a Drum Rack instrument was loaded, with each pad assigned a vocal sample of the keyword "globalist" corresponding to the mapped MIDI notes. Note assignment was limited to only those MIDI notes actually present in the track’s data, ensuring accurate triggering.

To sonify the intensity of discourse over time, the following modulation process was applied. First, an Expression Control MIDI effect was added to the MIDI Globalist track. The Keytrack parameter was set as the source. The target for modulation was defined as the Dry/Wet value of the Big Room Reverb audio effect on the Audio Globalist track. The fader was mapped using the Map function in Expression Control, enabling cross-track modulation. This mapping allowed note values in the MIDI data, proportional to term frequency, to dynamically adjust the reverb level, resulting in more spatialized or “echoed” playback during periods of heightened discourse activity. A successful mapping was confirmed by the appearance of a blue dot on the Dry/Wet control, indicating that it was being driven by MIDI input from the other track.

Upon playback, each MIDI note triggers the "globalist" sample via the Drum Rack, while the degree of reverb applied to each sample is modulated in real time based on the frequency data, embedding temporal and affective weight into the auditory rendering. The steps were repeated for each data sample, resulting in a soundscape portraying the choral dimension of the platform. This method creates an affective and dynamic soundscape directly informed by language patterns in the conspiracy Telegram channels. The soundscape can be found here.

4. Findings & Discussion

🔥Use of affective object emojis 💥

The use of emoticons—emoji mirroring facial expressions—to convey affect has been studied extensively, including in the context of alt-right and conspiratorial narratives (see Baider & Constantinou, 2024). In our dataset, however, such facial-expression emoji were comparatively rare. Instead, we observed a recurring reliance on object emoji—explosions, sirens, SOS signs—used to punctuate messages with affective force. As Riordan (2017) notes, object emoji can communicate emotion as effectively as emoticons, functioning as compact, repeatable affective cues.

Although not among the most frequently used symbols overall, flag emoji are an instructive case. In the German-language dataset, the German flag emoji appears consistently across time. Prior research shows that online political communication using the German flag tends to align with the political right and is associated with higher engagement (Kariryaa et al., 2022, p. 368). This is often read in light of Germany’s particular historical relationship to nationalism, which has fostered a degree of hesitancy toward overt display of national symbols both on- and offline (Lin, 2022, p. 26).

In the German-speaking channels, the single and double exclamation mark emoji are among the most prominent symbols. They are often used to divide lists or mark calls to action, giving posts a heightened sense of urgency. Affective emoji such as 🔥 and 💥 also appear with notable regularity, peaking around the New Year, when channel activity is at its highest.

In the English-language dataset, the fire emoji appears most frequently, outpacing other symbols. Overall posting volume spikes each January, coinciding with increased mentions of “Davos” and “wef” around the time of the World Economic Forum. As shown in Figure X, emoji usage peaks sharply in January 2023 and January 2024 before dropping off, reflecting a broader rhythm in channel activity.

One illustrative case is ‘Disease X,’ a placeholder term used by the WHO for hypothetical pathogens, which became a focal point in January 2024. On conspiracy-oriented Telegram channels, the term was reframed as the “New Pandemic” or described as “a concept used by vaccine developers for years entailing the use of gain-of-function research to create a serious disease in the lab and make a vaccine for it.” This rapid rise and fall in keyword frequency exemplifies the choral logics of Telegram (Cinelli et al., 2021), where narratives reverberate intensely for a short period before dissipating.

The fire emoji plays a central role in this dynamic. Often deployed to signal a “hot take,” it functions as an affective and aesthetic marker of new narrative formations. Some of these gain traction—becoming “sticky” with affect and spreading across the platform—while others fade quickly. Its cyclical surges in frequency mirror the rhythmic passage of Telegram discourse, highlighting the platform’s infrastructural tendency toward repetitive, affect-charged amplification.

TG channel analysis

Shifting from the micro-level of symbolic expression to the channel level, we examined the descriptive fields and content themes of the sampled Telegram channels. After extracting the Description field from the dataset CSV, we normalised the text (lowercasing, removing punctuation) and applied TF-IDF vectorisation (excluding common English stopwords) to identify high-value terms. We then used Truncated SVD to detect latent topics, each represented by its top 10 most representative terms.

The resulting topic clusters reveal a hybrid rhetorical style: conspiratorial ideation interwoven with populist mobilisation. Many channels foreground themes of awakening, truth, and freedom (Topics 2 and 4), positioning themselves as insurgent truth-tellers against a corrupt global order. Others fuse COVID-scepticism with world-systemic critique—martinezperspective and coronanuchterheid, for instance—while QAnon residue remains visible in terms such as WWG1WGA, NCSWIC, and trust the plan (Topic 5).

Religious and moral motifs (e.g., faith, God, kindness) frame the ideological struggle in spiritual terms, while recurring references to official, real, and current news (Topics 1 and 3) underscore a counter-informational identity rejecting mainstream narratives. Stylistically, these channels speak in an emotive, declarative register—urging followers to speak, hold the line, or stay awake—a mode that aligns with the “style” dimension of populism studies. But style here is inseparable from substance: these calls to action sustain a coherent imaginary of an awakened people resisting a globalist elite.

Within this larger ensemble, certain channels stand out as soloists whose “verses” crystallise the chorus effect. uncensoredtruths blends health misinformation with anti-globalist rhetoric; thepatrioticau inflects global conspiracy frames with distinctly Australian nationalist tones; and Dugin_Aleksandr introduces an explicitly geopolitical register grounded in multipolarity and resistance to Western hegemony. Each performs in a different register—prophetic, nationalist, geopolitical—yet all harmonise around shared antagonists and affects.

While loosely networked, these channels form a discursive ensemble grounded in distrust, sovereignty, and eschatological urgency. In choral terms, they are distinct voices singing in different registers—nationalist, prophetic, geopolitical—yet harmonised around shared antagonists and shared affects. Together they enact an infrastructure of political resonance, sustaining the chorus effect across ideological nuance and geographic distance.

SHOUTS!

In earlier internet culture, the excessive use of capital letters was often read as a form of “shouting” or yelling. More recent research reframes this as a form of prosody — a way to mark emphasis, rhythm, or identity in text (Heath, 2018; Chan & Fyshe, 2018). In the metaphor of the chorus, capitalisation marks the parts sung forte — the emphatic moments that cut through the noise.

Across the three language corpora, these emphatic peaks take different shapes.

* English-language channels capitalise governmental organisations and key individuals — e.g., JEFFREY EPSTEIN — transforming them into affective signposts in the conspiratorial score.

* Dutch-language channels more often emphasise popular media outlets, particularly TV broadcasters such as RTL 4, alongside several well-known Dutch alt-right websites.

* German-language channels lean toward capitalising country names, reinforcing geopolitical themes more than individual scandals.

These differences suggest that the “loud notes” of each chorus section are tuned to national registers: in English, scandal and corruption; in Dutch, media and information control; in German, the geopolitics of sovereignty. In all three, capitalisation functions less as random noise and more as a deliberate stylistic choice — a prosodic device for heightening emotional pitch, cueing collective response, and making certain terms resonate longer in the audience’s memory.

Frequency and Conspiratorial Network

The temporal rhythms of posting reveal how the chorus effect is sustained across the network. Activity surges in sync with key geopolitical or symbolic events — January peaks around the World Economic Forum meetings, spikes during major battlefield developments, or bursts of activity when new conspiratorial tropes (such as “Disease X”) emerge. These pulses are not confined to single channels; rather, they echo across the ecosystem, as multiple actors pick up the same cues, adapt them to their own idioms, and reintroduce them into circulation. The effect is not unlike the refrain in a song: familiar, recognisable, and primed for re-entry at just the right moment.

Within this temporal scaffolding, certain channels act as distinct “verses” in the chorus, each contributing a recognisable thematic line while keeping time with the same overarching beat:

* uncensoredtruths delivers its message in the key of apocalyptic revelation. Health misinformation, anti-globalist rhetoric, and prophetic warnings are woven into urgent, declarative posts that position the channel as a custodian of forbidden knowledge. The tone is insistent, almost sermonic, sustaining emotional intensity and priming audiences for ongoing vigilance.

* thepatrioticau localises the globalist frame to an Australian nationalist register. Posts refract anti-WEF, anti-surveillance, and anti-UN talking points through references to domestic politics, sovereignty, and the defence of the homeland. This localisation deepens audience identification, showing how global antagonists can be remapped into national struggles without losing the underlying narrative grammar.

* Dugin_Aleksandr offers a geopolitical bassline, layering philosophical commentary over narratives of multipolarity and civilisational struggle. Historical tropes such as the “Great Patriotic War” are reframed to connect past resistance to contemporary opposition against Western hegemony, reinforcing the sense of an unbroken, epochal conflict.

Although stylistically distinct — prophetic, nationalist, geopolitical — these channels move in concert. They amplify and adapt each other’s refrains, whether by reposting content verbatim, riffing on shared slogans, or remixing memes into new thematic contexts. The network thus operates as both an echo chamber and an improvisational ensemble: repetition sustains recognition, while variation keeps the material fresh and adaptable.

In this way, frequency is not merely a metric of output but an organising principle of conspiratorial discourse. Posting rhythms align across disparate actors, allowing the chorus to surge at moments of narrative opportunity and recede when the informational terrain is less favourable. The result is a decentralised but synchronised propaganda structure — an infrastructure of resonance — where the interplay of tempo, refrain, and thematic variation sustains a shared conspiratorial imaginary over time.

Semantic frame extraction and actantial networks: “involved” discourse vs. “narrative” structures

Starting from samples of 1,000 messages in English for the “conspiracy” and “Ukraine war” subcorpora of our Telegram datasets, we annotated each message for the semantic frames it contains. Semantic frame extraction represents the meaning of a sentence in terms of who does what to whom relations, allowing us to detect recurring roles, actions, and relationships. This was achieved using the semantic frame extraction module in the PyFCG library (Van Eecke & Beuls, 2025). Previous research has shown that this method can capture key argumentative relations in societal debates (Willaert et al., 2022) and inform the design of alternative reading devices and opinion facilitators for social media (Willaert et al., 2021). More recent work explores its use in mapping narrative structures in public discourse (Pournaki & Willaert, 2024).

We used the extracted semantic frames to map relations between key actors as graphs. In the Ukraine war subcorpus, the resulting graph resembles a Greimassian actantial network: clearly identifiable dramatis personae—Russian and Ukrainian troops, geopolitical figures like Putin, Prigozhin, and the Wagner Group—engaged in defined antagonisms. Here, the chorus effect manifests as distinct verses: each role enters at the right cue, interacting in a sequence of events that gives the discourse a recognisable narrative arc.

Figure 2: actantial narrative graph of Ukraine dataset (see Posters linked at the top of this blog post for a higher resolution version)

The conspiracy subcorpus, by contrast, produces a graph dominated by the pronouns “us” and “they.” These actants are less about specific individuals and more about collective positions: “they” as elites, globalists, or the WEF; “us” as the awakened people, patriots, or believers. Instead of sustained narrative arcs, this mode delivers a stream of rhythmic shouts—short, affectively charged refrains that repeat across messages and channels. In this register, resonance outweighs narrative: persuasion hinges less on the unfolding of events than on the repeated re-activation of shared affect.

Affective rhythms

The sonification revealed that conspiratorial discourse on Telegram is not only patterned semantically but also rhythmically. Peaks in the sonic rendering corresponded to narrative flashpoints—periods when globalist, people, wef, and 🔥 surged in tandem. These moments often aligned with geopolitical or domestic events, producing short-lived but intense bursts of affective resonance. The layered vocal samples and modulated reverb produced an audible sense of “chorus,” where multiple strands of antagonistic framing seemed to converge into a shared emotional register. The fire emoji in particular acted as a percussive accent, punctuating and intensifying the verbal refrains. This confirmed the hypothesis that repetition and affective cues work together to bind disparate messages into a unified sensory experience. In this way, the soundscape made tangible the chorus effect as more than a metaphor—it became an audible structure of distributed resonance, revealing how platform-native discourse can align audiences through rhythm and affect as much as through explicit argument.

Taken together, these findings reinforce the central insight of this project: some online discourse aligns with classical narrative structures, while other forms operate more like a chorus—cyclical, affect-driven, and role-saturated without requiring linear progression. In this latter case, semantic frames capture not the unfolding of a story, but the ongoing re-voicing of antagonistic positions, keeping the collective “song” going even in the absence of new events. Here, the primary logic is not narrative coherence but resonance—the capacity of repeated cues, slogans, and symbols to evoke and sustain a shared affective stance.

The data sonification underscores this distinction: when translated into sound, the conspiratorial “chorus” emerges as a rhythmic architecture of resonance, where peaks and accents in the audio mark moments of intensified alignment rather than narrative development. In such cases, the communicative force lies less in telling a story than in sustaining a mood, ensuring that the antagonism remains continuously felt, even when there is nothing new to narrate.

6. Conclusions

Our analysis of emoji use, capitalisation practices, and narrative structuring across conspiratorial and extremist Telegram channels reveals a complex interplay between form, affect, and infrastructure. While the data confirm prior findings about the role of affective cues in digital propaganda, they also extend these insights by showing how such cues operate not as incidental embellishments but as integral, rhythmic components in the “chorus effect” we describe. Like recurring motifs in a song, affective object emoji—explosions, sirens, SOS signs, flags, and especially the 🔥—function as anchors that punctuate and pace the discourse, marking peaks of urgency, anger, or revelation. Their patterned appearance—often peaking around cyclical events such as the World Economic Forum’s Davos meeting or the emergence of new “threat” narratives like Disease X—creates a tempo that is infrastructurally synchronised across channels. This cyclical rhythm reinforces the chorus effect: users encounter the same emotional beats across multiple voices, producing a sense of resonance and inevitability.

Capitalisation patterns add another instrument to this ensemble. Typographic style becomes a low-tech yet highly effective means of signalling urgency, emphasis, or outrage—a form of visual shouting that can strike the same emotional chord as a drum hit or cymbal crash. As Birchall (2024) notes, aesthetic sensibilities differ markedly across online publics; what might appear excessive or amateurish to one audience may, for another, function as a badge of authenticity or an index of sincerity. In our corpus, capitalised words often coincide with key antagonists (“ELITES”), existential threats (“FREEDOM”), or moral absolutes (“TRUTH”), contributing to a micro-rhetorical economy in which typographic emphasis harmonises with the visual affect of emoji.

The semantic frame and actantial network analysis revealed a further structural distinction between the Ukraine war discourse and conspiracy discourse. In the former, narrative still follows a recognisable arc, with named actors in defined roles, actions unfolding over time, and antagonisms mapped clearly—a series of “verses” that add plot points to an ongoing story. In the latter, the structure is closer to a sustained drone: collective actants (“us” vs. “they”) locked in an eternal antagonism, with no resolution and no finale. Here, the “story” is less about what happens next than about sustaining a shared key and tempo. The implication is that counter-messaging strategies effective in narrative-driven contexts—fact-checks, debunks, linear rebuttals—may fail in chorus-driven spaces, where affective resonance, not propositional content, keeps the discourse alive.

From a societal perspective, these findings complicate conventional models of misinformation control. Platforms and regulators often focus on removing harmful content, but in chorus-driven spaces, the form of communication—the rhythmic recurrence of affective cues—may be as important as the content itself. Interventions aimed purely at content moderation risk muting individual notes without altering the underlying rhythm section. More promising approaches might seek to disrupt synchronisation patterns or introduce counter-melodies that dilute the density of affective cues, though such measures raise significant normative and legal questions about speech and design.

For academic research, the results underscore the need to treat style, affect, and infrastructure as interlinked analytical categories. Our mixed-method approach—combining emoji frequency mapping, keyword refrain analysis, semantic frame extraction, and artistic re-encoding through sonification—offers one model for integrating computational, qualitative, and creative methods. This hybridity enables researchers to detect patterns that might otherwise remain inaudible, while also making them perceptible to broader publics. The approach also opens pathways for comparative cross-platform studies and for testing whether the chorus effect holds in other media ecologies, including video, audio, and livestream formats.

Technically, the analysis points toward new tools for early-warning systems in information integrity work. Monitoring not just what is said, but how it is said—the pacing of emoji bursts, the clustering of capitalised refrains, the repetition of antagonistic frames—could enable earlier detection of narrative mobilisation. Such tools could be embedded in dashboards for journalists, researchers, and civil society groups, allowing them to recognise when the band is tuning up for a new performance and to respond before the set begins in earnest.

Ultimately, the findings suggest that the chorus effect is not merely a metaphor but a structural property of conspiratorial communication in “dark” parts of a platform like Telegram. Its persistence across thematic and linguistic boundaries points to a shared communicative logic that is robust to content moderation and platform migration. If, as we have argued, the chorus is sustained by rhythm, repetition, and affective intensities rather than by linear narrative, then countering it will require interventions that operate on the same registers. This might involve composing alternative “counter-choruses” that mobilise humour, empathy, or collective joy rather than fear and grievance—an avenue that demands both creative experimentation and rigorous evaluation. In this sense, future work will need to think not only as analysts but as arrangers, capable of re-scoring the affective infrastructures that currently keep these movements in perfect sync.

8. References

Ableton 12 (2024). Ableton Live 12 [Digital audio workstation software]. https://www.ableton.com/

Baider, F., & Constantinou, M. (2024). Transnational Conspiracies Echoed in Emojis, Avatars, and Hyperlinks Used in Extreme-Right Discourse. International Journal of Communication, 18.

Birchall, C. (2024). Haute Baroque Bling: Style, Taste, and Distinction in the Study of Populist Conspiracism. In Populism and Conspiracy Theory (pp. 323-343). Routledge.

Chan, S. and Fyshe, A. (2018). Social and Emotional Correlates of Capitalization on Twitter. Proceedings of the Second Workshop on Computational Modeling of People’s Opinions, Personality, and Emotions in Social Media, 10–15. https://aclanthology.org/W18-1102.pdf.

Cinelli, Matteo, et al. "The echo chamber effect on social media." Proceedings of the national academy of sciences 118.9 (2021): e2023301118 Heath, M. (2018). Orthography in social media: Pragmatic and prosodic interpretations of caps lock. Proceedings of the Linguistic Society of America, 3, 55-1. Kariryaa, A., Rundé, S., Heuer, H., Jungherr, A., & Schöning, J. (2022). The role of flag emoji in online political communication. Social Science Computer Review, 40(2), 367-387. King, E. (n.d.). CSV to MIDI [Web-based data-to-MIDI converter]. https://csv-to-midi.evanking.io/

LuvVoice AI. (2024). LuvVoice [AI voice and speech synthesis platform]. https://luvvoice.com/

Lin, C. (2022). Was ist das Wahrzeichen Deutschlands?: Eine Studie zur Identifikation mit den nationalen Symbolen der Deutschen. Jahrbuch für Internationale Germanistik, 54(1), 23-40

Papacharissi, Z. (2015). Affective publics and structures of storytelling: sentiment, events and mediality. Information, Communication & Society, 19(3), 307–324. https://doi.org/10.1080/1369118X.2015.1109697.

Pournaki, A, Willaert, T., (2024) A graph-based approach to extracting narrative signals from public discourse. https://arxiv.org/abs/2411.00702

Riordan, M. A. (2017). Emojis as Tools for Emotion Work: Communicating Affect in Text Messages. Journal of Language and Social Psychology, 36(5), 549-567. https://doi.org/10.1177/0261927X17704238.

Van Eecke, P. and Beuls, K. (2025) PyFCG: Fluid Construction Grammar in Python. https://arxiv.org/abs/2505.12920 Willaert, T., Banisch, S., Van Eecke, P., & Beuls, K. (2022). Tracking causal relations in the news: Data, tools, and models for the analysis of argumentative statements in online media. Digital Scholarship in the Humanities, 37(4), 1358–1375. https://doi.org/10.1093/llc/fqab107

Willaert, T., Van Eecke, P., Van Soest, J., & Beuls, K. (2021). An opinion facilitator for online news media. Frontiers in Big Data, 4, 695667. https://doi.org/10.3389/fdata.2021.695667

9. Appendix A. — Coreference Resolution Example

One challenge in applying frame-semantic analysis to Telegram discourse is that key actors are often referred to only through pronouns (they, their, them), leaving the underlying semantic relationships ambiguous. This is especially common in conspiratorial rhetoric, where antagonists are cast as vague collectives—e.g., they as “globalists,” “elites,” or the “WEF.” Without resolving these pronouns, the resulting semantic graphs can misrepresent or obscure the actual actors involved in a narrative.

To address this, we implemented a process known in NLP as coreference resolution, which identifies when different words or phrases refer to the same entity and replaces ambiguous pronouns with their explicit referents. In our case, this step was critical for making actantial networks—maps of “who does what to whom”—more interpretable and analytically useful.

We began with frame-semantic outputs from the PyFCG pipeline, then applied the fastcoref library within a spaCy (en_core_web_sm) environment to detect and resolve coreference chains. For example, in the sentence:

Our children are merely pawns used by leftists as part of their larger pursuit of securing the profit and power necessary to advance their globalist/NWO agenda.

the pronoun their is replaced with leftists, resulting in:

Our children are merely pawns used by leftists as part of leftists’ larger pursuit of securing the profit and power necessary to advance leftists’ globalist/NWO agenda.

This transformation preserves the original meaning while making the actor–action–target relationship explicit. Across our dataset of 2,144 unique messages, this process greatly improved the clarity of semantic graphs—especially in the “conspiracy” subcorpus, where abstract antagonists dominate.

Step 1 — Original Message

Figure A.1 — Original message (excerpt)

“Our children are merely pawns used by leftists as part of their larger pursuit of securing the profit and power necessary to advance their globalist/NWO agenda. — The system runs like Tammany Hall: To be a part, you accept favors from the party, and in return you do them favors. They give you or your institution money; you get them elected. You get them elected, they give you or your institution money. No one in this backscratching circle considers the interests of kids, let alone of our country. They don’t have to. The kids arrive on their doorsteps automatically even if the schools neglect and abuse them. In this system, kids are merely pawns cynically used to obtain money and power.”

Step 2 — Frame Extraction Output

Figure A.2 — Frame extraction output

The initial frame extraction identifies the argumentative structure but leaves certain actor references ambiguous due to pronoun use (e.g., “their” in “their globalist/NWO agenda”).

Step 3 — Coreference Resolution Output

Figure A.3 — Coreference resolution output

Coreference resolution identifies entity clusters (e.g., ['leftists', 'their', 'their']) and replaces pronouns with their antecedents.

Step 4 — Resolved Text

Figure A.4 — Message after coreference resolution

NoReferenceText: Our children are merely pawns used by leftists as part of leftists’s larger pursuit of securing the profit and power necessary to advance leftists’s globalist/NWO agenda. — The system runs like Tammany Hall: To be a part, you accept favors from the party, and in return you do the party favors. The party give you or your institution money; you get the party elected. You get the party elected, the party give you or your institution money. No one in this backscratching circle considers the interests of kids, let alone of our country. They don’t have to. The kids arrive on The kids’s doorsteps automatically even if the schools neglect and abuse The kids. In The system, kids are merely pawns cynically used to obtain money and power.

-- TomWillaert - 05 Sep 2025