Social media and digital platforms: polarization and hate speech

1. Introduction

This special issue, Médias numériques et débats démocratiques, is the outcome of the workshop Which digital media spaces are open to democratic debate? held at the University of Stockholm in 2024. It is a collaboration with the International Seminar on Social Media Discourse Analysis (ISSMDA), supported by the LERMA laboratory (Aix-Marseille University), the European alliance CIVIS and Maison Française d’Oxford, as well as the ROMPOL research group of Stockholm University’s Department of Romance and Classical Studies. While the overarching aim is to examine how digital media shape the conditions for democratic deliberation, hate speech emerges as a privileged theme across the contributions, drawing attention to freedom of expression, platform affordances, regulation, and power relations. In this sense, hate speech functions as the special issue’s overreaching analytical lens, intersecting with domination–marginalization dynamics, conspiracism, and fake news.

A recurrent theme is also the real challenge of finding a balance in a system where the traditional printed or broadcast top-down professional media are the main channels, but destabilized by the flat structure of social media where every individual can communicate and express personal opinions in public forums, on different platforms. In between, of course, there is the growing heterogeneous variety of online media, including podcasts and blogs. Today’s media landscape is therefore a double-edged sword, since free speech on the one hand, and its control on the other, are not easy to handle, either from a political or from a technological perspective. In many parts of the free world, not least in Europe, we are currently engaged in political discussions about the regulation of social media, including questions such as age, distribution, and content, and their relation to citizens’ freedom of expression.

2. What Counts as Hate Speech? Definitions, Boundaries, and Power

The so-called “hate speech” seems to have emerged in public discourse as a major issue of our era; this evolution is coupled with the large expansion of social media. Social media is here understood as “Internet-based channels of mass personal communication” and as such mainly contains content that is generated and interacted with by the users and where self-presentation is an important added value (Carr & Hayes 2015: 8). Baptiste Nicaud, editor of the volume Discours de haine (2023a), assumes, among other linguists, discourse analysts, and other scholars, that the increasing number of hate speeches, mainly fuelled by social media, is the result of the liberalization and the expansion of internet content. The issue may, however, be more complex. Smyrnaios and Ratinaud (2023) argue, in their study of the genesis and circulation of the term “islamo-gauchisme” in the public sphere, that traditional media are often themselves a source of hate speech.¹ Although such discourse spreads online, its emergence often stems, in their view, from the selection, framing, or media treatment of a news item by a traditional media outlet. That said and regardless of this, public expression has undeniably undergone a profound transformation, becoming direct, massive, and accessible to anyone with online access, and has therefore become the most efficient tool ever for freedom of speech (Nicaud 2023b: 199). Moreover, anonymity, sharing, and media without editorial oversight reinforce the power of this informational revolution. This also comes with a price: expressions of language that before came with a sender, originated from established platforms, and whose reception was delayed relative to production can now, whoever you are, be spread in a second to the entire world. The sheer volume of information has also contributed to what Nicaud (2023b) calls “banalization” of hateful speech and resonates with the work of Rosier (2025), who shows how insults become normalized within contemporary public discourse. It is also worth mentioning Paveau’s (2017) concept of digital discourse, which highlights how online environments reshape the production, circulation, and interpretation of linguistic violence.

Konikoff (2021), Nagle (2009) and Woods & Ruscher (2021) emphasize in their respective studies the contamination effect and its potential for “pollution” in hate speech, reinforced by the ease of a simple login and the anonymous identity of haters. To quote Ermida, editor of the volume Hate Speech in Social Media: “[t]he disease it spreads is prejudice: the side effects are heightened bigotry and intolerance” (2023: 5). Another issue raised is the mentioned volume is the online permanence of written messages, allowing them to re-emerge and recapture support over time (idem: 5).

Perhaps the most important moral criterion for identifying hate speech is whether it is linked to real-world harm – for example, when the speaker simultaneously engages in, or calls for, violence offline. Scharlaj (2020), for instance, gathers actual studies on hateful speeches related to simultaneous wars and conflicts in countries like Ukraine and Belarus, among others. However, most hate speech today is produced behind the computer, and that is also the case for the material analyzed in most discursive studies on hateful messages, as is the case in this special issue. Behind computer-mediated discourses can, evidently, be rather devastating, on a mental level, for individuals or groups, some of these discourses therefore also being banned, although there is an ongoing legal, cultural, and political discussion regarding when, where, who, and for what reason certain discourses shall be banned and even subject to criminal prosecution. This raises the question of what a speech act actually is. Words become such acts punishable by law when they convey meanings of threat, defamation, and incitement to violence. Those types of discourses, as well as other harmful speech such as expressions of hate and violence, do sometimes lead to righteous physical violence, and this is why the question of broadening the criminalization of harmful speeches has been introduced in some countries and has become a burning political issue.

Faced with what is perceived as its expansion and dangerous impact, European governments have increasingly attempted to legislate on the subject, often in conflict with the equally important principle of freedom of speech.² Although the first laws and regulations in Europe appeared in 2003–2004, there are still no well-functioning systems to prevent hate discourse on the internet; the phenomenon continues to grow.³ The humanities at universities have naturally embraced this notion, which, however, has been the subject of criticism: while it is relatively easy to identify and describe overt hate speech, which takes the form of discriminatory verbal violence, it is much more complex to delineate hate speech when it assumes subtle, sophisticated forms or is presented as ideological argumentation – leading researchers to propose the concept of “covert hate speech” (Lorenzi Bailly et Moïse, 2021). Baider & Constantinou (2019) address this issue by deconstructing the linguistic mechanisms of exclusion and hate, while analyzing both their concealed forms and the counter-discourses that respond to them. Here it may also be worth noting that, regarding hate speech in general, the notion is hardly politically neutral since its use in research may implicitly advance a particular political agenda. We will examine these topics from different angles while linking them to the content of this special issue.

The definition of the term “hate speech”, per se, is an important question. It has been rephrased in many ways, not always with the same meaning, though it still vaguely conveys the nature of the linguistic devices we commonly refer to. Hate and harmful speeches may refer to harmful discourses that hurt the receiver, such that he or she doesn’t wish to reexperience it and may take action to stop it. Expressions such as prejudice, discrimination, insults, sexism, racism, ageism, homophobia, etc., are some more precise denominations circulating in studies on this matter. One of the biggest issues regarding hate and harmful speeches is where to draw the line between, on the one hand, strong but not unfair speech and, on the other, harmful and toxic speech. Some expressions are unquestionably hate speech, while others are regarded as fair criticism of a person or a phenomenon. This characterization is often due to the sender-receiver perspective, differences in worldview and culture, or, simply, the overall context. Experiencing harmful speech does not mean the person who delivers it considers it to be hate speech, that he has the intention to harm, but may rather consider it to be “the rightful truth”. Hate or harmful speech of a certain kind may therefore identify group identities and assemblies sharing the same values. Thus, certain discourses may construct an opposition between “we” and “them,” in which expressions perceived by some as hateful are associated with specific actors within one community and directed toward another community. Hate speech also changes over time, reflecting shifts in norms and attitudes. What are considered to be expressions of homophobia, misogyny, racism, or sizeism for the younger generations may not harm older people at all. In several countries, there have for instance been quite a few discussions on whether some literature should be banned from public libraries due to presumably racist and/or hateful content, debates shaped by differing norms and value systems concerning what constitutes hate discourse. Hate discourses also often develop through overgeneralizations of terms that may apply to specific individuals, such as “pedophile”, “rapist”, “terrorist”, but are used as derogatory terms to disqualify individuals or entire groups (ethnic, social, gender-based). Depending on how discourses are received, this may result in a large range of attitudes, from stereotyping and racism to defamation.

Hate speech may be a question of power and abuse of power, used to silence, oppress, and stigmatize persons or groups. This may have a double effect on the target; not only may a subjugated person or a group experience the negative impact of harmful speech, but they may also suffer from not being able to reply, or report the offender. Hate speech then serves as a means for socially dominated groups to regain power. One can therefore question the use of the concept of hate speech when wielded by dominant media or by political groups, particularly the performative use of it to silence social struggles staged by non-empowered groups. Could this invert the traditional understanding of “hate speech” by examining whether it can be framed as politically legitimate, as resistance and empowerment?

Finally, hate speech constitutes both a legal and a political issue, the regulation of which falls not only under national legislation but also under the responsibility of the platforms themselves. Moderation thus appears as a matter that is both closely tied to the legal framework and partly autonomous from it. Depending on the legal context, verbal attacks may therefore be regarded as lawful or unlawful, even though the boundary between hate speech, threats, and defamation remains a subject of intense debate in contemporary democracies. National laws on freedom of speech, and laws on hateful speech have, as a matter of fact, grown out of the culture, tradition, and ethnic principles within the community they emerge from, and discourses will be judged according to this. We know that in European countries’ constitutions, freedom of speech and other freedoms are viewed as rights granted to their citizens and guaranteed by the state authorities, whereas in the US, individuals are considered free by default, and the state shall only interfere when these inherently given freedoms, including freedom of expression, are threatened. This has been crystallized in the last decades, when many European countries have voted for laws criminalizing discourses of hate; the line for what is legal in the US is when it comes to direct threats and incitements to violence. However, beyond laws and regulations, within what is considered legal in each country, there is also a spectrum of cultural differences and views on religion, gender, age, parental authority, etc., which may shape how we perceive hateful speech. One community may view a certain discourse as hateful or discriminatory blasphemy, while another may regard it as a legitimate critique or even humor. Thus, conflicts arise, and cries for liberalizing as well as for hardening the laws are heard. In some countries (Sweden for instance), laws are written so that one may take personal experiences of hate or discrimination into account. In this case, speech without intention to harm may be recognized as such and, according to the circumstances, be subject to legal proceedings.

3. Platforms and Analysis: Algorithms and Linguistic Methods

Linguistic analyses are, of course, important for pinpointing the current state and the history of hate speech, as we are dealing with how language models on the Internet can reinforce discourses of hateful content and thereby trigger users in that direction. The number of studies within linguistics has grown, and the theoretical frameworks and methods used are numerous, although discourse analysis dominates the field of studies, with studies ranging from overt forms of hatred to covert, implicit forms of conveying prejudice, racism, and other harmful discourses. Exploring these phenomena often calls for an interdisciplinary approach, as evidenced by the edited volume on online hate speech directed by Longhi and Vernet (2023).

Zooming out before presenting the content of our special issue, several important perspectives emerge from earlier studies. Recurrent themes in studies of hate speech include gender prejudice, body shaming, religion-related hate, and, of course, racism, often viewed through the lenses of Critical Discourse analysis (Esposito & Breeze 2021; Raffone 2022; Sharifi et al. 2016; Van Dijk 2021). Cognitive linguistics and metaphor studies also contribute to research on (often covert) hate speech, misogyny, and anti-immigration themes (Musolff 2017; Prażmo 2020). Not surprisingly, many researchers in this domain find speech act theory rewarding because it allows them to examine (often implicit) speakers’ intentions and the fine line between words and acts (Austin 1962; Searle 1979). Sometimes, only by analyzing the illocutionary force – intention – behind a discourse can hate or harmful speech be identified, in studies such as Hornsby (1993) and Lederer & Delgado (1995) on pornographic and race-related issues in discourse. The performativity of speech acts, that is, the effect that speech has on the interlocutor, is also interesting, especially nowadays, when discussions focus on what, how, and when a discourse can incite illegal actions and may therefore be legally banned. In one of her books, Butler (1997) addresses the question of performativity directly, emphasizing that language can act upon subjects, constitute them, and inscribe them within relations of power. Macdonald & Lorenzo-Dus (2020) have also contributed with a study on performative speech acts and terrorism incitement.

It goes without saying that studies looking for explicit forms may conduct larger quantitative studies of frequency and collocation, for instance, whereas more implicit expressions of hate, such as the intentions behind them, may need to be analyzed through close reading. Regarding quantitative big data studies, Waseem & Hovy (2016) have studied predicative features of hate speech on X, and Fersini et al. (2018) have used computational methods for automatic detection of misogynistic expressions in Italian and English posts on X. Mostly, there is a need for both methods, which will be illustrated later on. This is certainly the case for harmful discourses since big data approaches cannot handle “figurative implicit or nuanced language, which skilfully phrases hate in a disguised way” (Ermida 2023 : 7).

The reasons we mentioned earlier – anonymity, accessibility, sharing, social media networks – foster the production of controversial discourses, as the algorithms of technological platforms amplify the virality of such emotionally charged posts. This tendency to circulate discourses without questioning their veracity promotes the proliferation of conspiracy theories and conspiracy narratives (exploiting deepfake technology). Media and political actors are apparently tempted to, and do, exploit technical devices to transmit unverified and misleading information (fake news, alternative truths), enabling them to construct and generate clickbait from discourses and counter-discourses arising from circulating dubious publications that can severely harm individuals and groups.

The renewal of digital discourse analysis studies has followed this new development. As Marie-Anne Paveau (2017) argues, online discourse cannot be analyzed independently from the technical architectures that enable its circulation, visibility, and interactional dynamics. It has embraced the web in its polysemiotic dimensions to capture the emergence and/or reconfiguration of new forms of discourse and discursive strategies. The expression of emotions on the web, including verbal and paraverbal violence, offers an unprecedented field due to its multimodal expressions, circulation, dissemination, and discursive reflexivity, where all discursive materialities intersect; these are aspects this special issue aims to address. Indeed, studies of mediatized hate speech dominate the articles in this special issue, exploring media corpora reflecting discriminatory and sometimes violent ideologies, such as xenophobia, antisemitism, classism, sexism, and LGBTQIA+-phobia, for example.

4. Summaries of individual articles

The section below contains brief summaries of the single articles. Together, they explore how digital media spaces – platform features, algorithms, and new online genres – shape the conditions for public debate and participation, and how the same dynamics can foster and amplify hate discourse. The abstracts underline what these spaces enable (visibility, access, circulation) and what they complicate (platform rules and their application: who gets attention, and the normalization of hateful and harmful speech).

Concerning the legal aspects of hate speech, Bornet examines the various procedures, methods, criteria, and approaches to the prevention and sanctioning of online hate speech, as addressed by the Cour européenne des droits de l’homme (CE). The performative aspect of hate speech is an important part of the CE’s work, and the article discusses the limits between hate speech in the form of using denigrating words to talk about someone (non-performative) and the attribution of despicable characteristics to someone that may incite violence (performative). The CE appears to be more severe than its American counterpart, although the author shows the relentless search for a balance between the protection of the freedom of speech, on the one hand, and the defence of individuals’ right to privacy and the fight against discrimination, on the other. The court’s handling of presumably hateful discourse consists, to a large extent, in contextualizing all and each of them by taking into account the instances of the communicative event, such as the identification of the nature of the speaker and the addressee, and also time, space, etc., with respect to the different national legal systems.

Some articles concentrate on specific types of hate speech. Mourier and Jackiewicz analyse the misogynistic cyberbullying of the female Formula racing influencer Manon Lanza. The analyses show the nature of a subgenre, “LOL-culture” in social media posts, a mix of harassment with sexual insinuations by male Internet users disguised as humour, often reinforced by platform dynamics and high-visibility targets. The video clips, in a joking manner, reveal the false causal link between the accident and her gender, and the comment threads further amplify and normalize this framing. The article also points to the multimodal circulation of such attacks across different online spaces and genres, and how these expressions of mockery work as a cover for misogynistic aggression.

Several articles examine hate discourses arising from clashes and confrontations between societal groups over questions of identity, class, age, gender, religion, culture, and traditions. In Atifi’s article on a Moroccan law that criminalizes sexual relations outside marriage, various parameters are at stake. The author shows how forms of diabolization and insults characterize the conservative leaders’ social media discourses towards the reformists, in order to delegitimize and stigmatize them, and create fear, when the legal, moral and religious arguments of the conservative leaders do not seem to stop these groups from fighting against repression and demands for freedom for women, equality and autonomy.

Vernet suggests in his article that hate speech must be considered in the light of relations of power and dominance. Thus, hate discourses seem to be used to (re)produce relationships of dominance in interactions between dominant and subordinate groups, such as ruling political parties or majority populations versus political opposition or national minorities, and economically dominating versus subdued groups, etc. In this perspective, the signalling and labelling of hate speech may be used by dominant groups for oppression, dominance and silencing of the discourses of other legitimate political groups, and even censoring them and closing the debate. This may lead to a seemingly (but not a real) shift in power relations where the stronger part finds themselves in the role of the oppressed, as a victim of hate speech, such as in the media discourse on the conflicts Kanaky-Nouvelle Calédonie, where expressions of anti-white racism or hate against whites may cover the struggle for decolonization and independence of the Kanaky people. Vernet suggests hate speech must not be understood as a merely violent or morally condemnable utterance, but argues for a materialistic definition of hate speech as a product and an instrument of social, historical, and material relations of domination.

Regarding the question of identities and hate speech, Ukhova’s article on polemic interactions on X between French and Russian-speaking Internet users, discussing the ongoing war in Ukraine in the light of official governmental Russian discourses published via YouTube and other channels, shows how the patriotic identity discourses produced from the Russian side also stage a diabolized image of the West. The self-image transmitted in the posts draws on stereotypical and collective identity markers of a strong nation with traditional values and a mythic history, whereas the West is pictured as liberal in the sense of weak, decadent, corrupt and without rules. These extreme positions of degradation and glorification of the russkiy mir (approximately translated as “the Russian World ideology”) reflect the official narrative of the leadership of Russia, cementing the polarized civilizational worldviews instead of a democratic exchange of visions.

When it comes to online discourses and democracies, some interesting studies depict how digital platforms are proved to be a double-edged sword: Malick Ba shows in his article on the political situation in authoritarian Senegal, how digital activism challenges the deficit of democracy in the struggle for freedom of expression including criticism of the government, while at the same time it also becomes a tool for the government to censure and manipulate its population. Here, digital spaces do empower citizens to challenge authoritarian leadership in the way that they facilitate access to information, but they also amplify misinformation, engendering polarization and hate speech by exposing them to state surveillance and censorship.

Rhetoric must always be taken into account in polemical and sensitive political discourse. The intention to legitimize one’s own discourse while simultaneously delegitimizing an opponent’s is rarely based solely on logos; pathos and ethos are often at play, so rhetorical devices are used to strengthen one’s position. In Terzi’s article, we can see how some discourse markers work together to produce offensive speech, moving away from fair argumentation and toward an ad hominem style marked by mockery, personal attacks, and delegitimization—developments that harm constructive political debate.

In the same vein, Cataldo and Rouby examine how the insult “soyboy” circulates on the pro-Trump platform patriots.win. The authors find that “soyboy” reinforces and polarizes in-group identity, that it uses “us vs them” labelling, and shifts “enemies”. What was first used for naming a “feminized man” lacking traditional “masculine” traits (meat-eating, toughness, dominance), now extends to a term that stigmatizes LGBTQIA+ identities but also weakness in general, corruption, and betrayal. The term “soyboy” is an example of “floating signifiers”, showing that this prejudiced name-calling undermines democratic debate by shifting political disagreement from political issues to identity, intensifying polarization and illiberal tendencies. These digital “agoras” often function as echo chambers for in-group communities, where algorithms and digital platforms can amplify hostile, delegitimizing talk.

Polarization is also the central theme in Lacaze’s article on the way Donald Trump structures his digital communication during the 2024 presidential campaign around the “border” theme in social media. This theme is linked to populist discourse: dramatization of the crisis through the mocking of his opponents regarded as ”enemies”, the use of derogatory terms such as “Crooked Joe Biden” and identifying scapegoats, enhancing polarization by using opposition features such as we/our vs. they/he, direct address through “you”, and typography in capital letters signalling shouting, and warlike metaphors (“final battle,” “destroy”). The study also shows Biden’s partly mimetic counter-discourse (the re-use of negative axiological terms and ad hominem attacks), contributing to an escalation of verbal violence by pushing the frontiers into a “new normal” discourse in the digital space.

Roitman’s corpus-based study examines how the French word “sioniste” circulates online, with unstable, sometimes competing meanings. It focuses on how repeated contexts make particular readings more available, especially readings that blur the line between “Zionists,” the state of Israel, and Jews as a social group. Following Galatanu’s theory of “dynamic meaning,” the analysis, with examples from a web news corpus, shows how a relatively stable semantic core in “sioniste” generates stereotypical associations in discourse and how frequent discursive realizations can, in turn, reshape the perceived “core” of the lexeme. The most frequent collocations – especially “entité sioniste” and “lobby sioniste” – tend to activate strongly negative frames (covert power, manipulation, global influence, etc.). The analysis shows how these pairings encourage conspiratorial thinking and make pejorative interpretations easier to justify.

Arrighi’s article does not primarily address explicit hate; it rather examines adjacent material to help understand how discursive frameworks capable of fostering hostile stances are formed and circulated. Drawing on a corpus of more than 2200 comments posted under a YouTube video devoted to Acadia, the thematic analysis – supplemented by tools from critical discourse and argumentation analysis – shows that the comments often reveal more about the commenters themselves (mostly French) than about the Acadians. The posts produce an identity-based discourse about France (nostalgia, calls to “resist,” paternalism, and conspiratorial motifs), within a dominantly non-dialogic space where positions are rarely challenged. The article’s contribution lies in documenting this process of banalization and these euphemized forms of “the other”, situated at the margins of explicit hate.

Competing Interests

Christophe Premat declares that he is co-editor-in-chief of the Nordic Journal of Francophone Studies. This article is part of a special issue published with the support of the “Språk och makt” network at Stockholm University and the Åke Wiberg Foundation (project H23-0123).