Abstract
Introduction
There is escalating concern, across national governments, inter-governmental organisations and think tanks, about the rise of far-right extremism globally (Chazan, 2019; Guterres, 2018; Jones, 2018). Public authorities, commentators and civil society groups have reported that this rise is being cultivated in online spaces, such as 4chan, 8chan and Gab.ai (Arthur, 2019; Europol, 2019). Yet, to date, there is limited academic research examining these spaces. Even basic information remains estimated or unknown (Hine et al., 2017).
Greater research of these spaces can assess the extent and nature of abusive, manipulative, and transgressive activities increasingly associated with them (Rainie et al., 2017). It can expose such behaviours and activities so that their targets can be better protected from harm. It can help explain why these cultures develop in certain online spaces and not others (Beyer, 2014). It can provide evidence regarding radicalisation processes that observers link to these communities (Palmer, 2019). Research can also illuminate the extent to which members of these communities represent a threat to outside groups, specific individuals, political events or society more widely.
The authors of this article organised a workshop for researchers and practitioners, hosted at King’s College London in 2019, to discuss methodological challenges studying 4chan and far-right politics online. Informed by their experiences and prior research, participants at the workshop agreed on three central challenges which help to explain the sparsity of academic research on these online spaces. These challenges concern the difficulties analysing (1) site content, (2) site users and (3) challenges researchers face regarding ethics and validity.
Content-related challenges concern difficulties analysing discourse on sites such as 4chan. Participants highlighted fast evolving language and image use on the site; the active attempts of 4chan users to manipulate research findings, and the difficulty of differentiating genuine and ironic content (Phillips and Milner, 2016). User-related challenges concern how the design of 4chan, and related platforms, makes it difficult to identify and assess site users. Difficulties include user anonymity, decentralised posting and dialogue, and limited or non-existent archiving (Bernstein et al., 2011). These elements make it hard to identify individuals, or even retain data for later analysis. Researcher challenges include the danger of further promoting extremist content and harming groups it targets. They also include researcher exposure to extreme content and possible retaliation, and how this may shape how they conduct their research.
During the workshop – unbeknownst to the authors at the time – a participant created a 4chan/pol/ thread to discuss the event, which they subsequently archived. Examining this later, we realised that it provided a unique dataset revealing how the 4chan/pol/ community talks about itself and responds to being observed. It also illuminated many of the challenges that the workshop was set up to discuss. We examined the thread using qualitative discourse analysis – an underutilised method in 4chan research. This enabled us to interrogate in depth the nuances of language use on the site, and explore the limits of what could be learnt about individual users from the content they post. Through this, we determined that our three central challenges could be broken down into seven more specific research questions. This article illustrates how far discourse analysis can help answer these questions, and establishes avenues for further research.
Content-related challenges analysing /pol/ are as follows:
1. How can we analyse language and images given their dynamic evolution and the difficulty differentiating serious and ironic content?
2. How consistent and coherent are identities and ideologies expressed in site content?
User-related challenges analysing /pol/ are as follows:
3. How far does 4chan’s structure shape the ability to differentiate /pol/ users?
4. How can we best interrogate the relationship between 4chan and the Alt-Right or far right?
5. To what extent is discourse on /pol/ shaped by influencers and opinion formers?
Researcher-related challenges analysing /pol/ are as follows:
6. What are the ethical challenges to researchers studying 4chan and similar free-extremist sites?
7. How does /pol/’s response to observation affect research validity? How can this be resolved?
The article is structured as follows. First, it offers context and justification for the research, conscious that publishing about these spaces can amplify and ‘oxygenate’ them (Phillips, 2013, 2018). It then outlines existing scholarship, before detailing our research design and methodology. This is followed by the discourse analysis itself, addressing the research questions outlined above. Since discourse analysis primarily focuses on language in use, we focus first on site
This article makes a distinct contribution in several ways. Its discourse analytic approach provides a more nuanced understanding of 4chan/pol/ and other free-extremist communities modelled on it, such as 8chan and Gab.ai. It illustrates the challenges of applying traditional discourse analytic approaches to platforms such as 4chan, since anonymity and the distinct culture on the site make firm inferences about content, users and context harder to make (Ludemann, 2018). It provides novel insights into the role of influencers on the site, and how this might shape the ability to trace radicalisation processes. It assesses how the /pol/ community responds to being observed, considering reflexively how this shapes research into the community. Finally, it suggests how future research can overcome challenges faced when analysing such communities.
Advisory warning
Before proceeding, readers should note that some of the content we reproduce from 4chan is extremely offensive and can be classified as hate speech. We are conscious of the risks involved in spreading such content, since doing so potentially contributes to attempts to subvert the existing political order and provides the amplification the community seeks (Phillips, 2018). To counter this, we only reproduce the minimum content we deem necessary to answer the study’s questions. We are reflexive in explaining these decisions as we go, conscious that such decisions are inherently political (McKinnon et al., 2016).
When we do refer to specific content, we consider it necessary to reproduce some of the actual language as used. We do this for three reasons. First, discourse analysis concerns the analysis of language
Yet, recognising the deeply offensive, anti-Semitic and racist language used, we have only used certain terms once, after which we obscure these.
Context and justification
4chan has become notorious as being ‘one of the darkest corners’ and ‘seediest zones’ of the Internet’ (Coleman, 2014: 47; Hine et al., 2017) or, more charitably, the Internet’s ‘id’ (Woolf, 2016). Simultaneously, it is credited with being one of the most ‘impactful generators of online culture’ (Hine et al., 2017), with distorting democratic politics (Moore, 2018), with incubating the Alt-Right (Nagle, 2017), and with developing almost every viral meme on the Internet up to 2011 (Phillips, 2015). The site has earned its reputation in part due to its self-consciously offensive culture, characterised by the endemic use of racist, misogynist, homophobic, anti-Semitic, bullying and aggressive language, images and videos. This culture is most closely associated with its /pol/ (politically incorrect) board, one of over 70 boards on the site.
Founded in 2003, interest in 4chan developed from its aggressive attempts to spread its culture across the web, its association with trolling, and its incubation of the hacker collective, Anonymous (Beyer, 2014; Coleman, 2014; Olson, 2012; Phillips, 2015). Early on, the 4chan community was more likely to be described as libertarian, left-wing or anti-establishment, advocating free speech fundamentalism, taking on corporations and supporting democratisation movements in the Arab Springs (Phillips, 2015). Emphasising extreme free speech led to the community being described as ‘free-extremists’ (Moore, 2018). Only later did the community’s fundamentalist approach to free speech lead it to evolve into a more reactionary community, fighting online ‘culture wars’ against left-wing ‘Social Justice Warriors’ (SJWs) and becoming associated with the Alt-Right (Nagle, 2017). Politically, 4chan became more prominent in the 2016 US election, with media reports and 4chan users claiming the platform was so influential that it ‘actually elected a meme as a president’ in the form of Donald Trump (Ohlheiser, 2016).
The site, along with similar and overlapping ‘free-extremist’ communities at 8chan, Gab.ai and Reddit, has also been accused of radicalising users and promoting right wing extremism and terrorism (Beyer, 2014; Coleman, 2014; Hine et al., 2017; Moore, 2018; Olson, 2012; Phillips, 2013; Zannettou et al., 2018a). Several alleged killers have reportedly discussed plans, posted manifestos, photographs or footage on 4chan, 8chan or Gab.ai, including David Kalak, Alek Minassian, Chris Harper-Mercer, John T. Earnest, Lance Scarsella, Brenton Tarrant and Patrick Crusius (BBC, 2014; Feldman, 2015; Hume, 2019; Lavender and Burke, 2017).
Research to date
Before 2016, most research on 4chan and similar sites was anthropological and ethnographic. Quantitative research was minimal. Gabriella Coleman (2014) combined observation with participation in her analysis of Anonymous, combining detailed descriptions with extensive chatlogs. Whitney Phillips (2015) mapped the relationship between online trolling and mainstream culture using a ‘qualitative, mixed-methods’ approach employing extensive interviews and ‘thousands of hours’ of observation. Jessica Beyer (2014) examined ethnographically the evolution of political consciousness in four different online communities, including Anonymous/4chan. There is less qualitative or ethnographic research charting 4chan’s evolution after Anonymous grew apart from it in 2011, with the exception of Angela Nagle’s influential Kill All Normies (2017).
The only quantitative research examining 4chan before 2016 was by Michael Bernstein et al., who looked at anonymity and ephemerality on the site. More has emerged since but remains limited to a relatively small research community. We now have statistics on the ephemerality of posts and the extent to which users maintain anonymity (Bernstein et al., 2011). We know that 4chan/pol/ users habitually discuss ethnicity, often using hate speech (Zannettou et al., 2018a). Hine et al. (2017) found that 12% of posts contained a term defined in the
While individually valuable, the limited number of these quantitative analyses, the difficulty of comparing them or extending their findings over time, and the necessarily constrained ambitions of each, means that there remain numerous unanswered questions about these online communities. We know little about 4chan beyond /pol/. We know little about users and even less about the relationships between them. We lack evidence tracing the development of users’ political views on the site. We do not understand fully the connections between images and text, although Nissenbaum and Shifman (2017) provide useful insights into how memes on 4chan’s /b/ board function simultaneously to trigger conflict, and also help generate shared culture and identity. Philippe-Joseph Salazar (2018) has written about how the Alt-Right have consciously sought to ‘formalise a gamut of memes’ with embedded meanings that broach the prohibitions of contemporary culture. Yet we are also no closer to knowing how to distinguish between satire and seriousness in these forums and we are only beginning to recognise the origins of co-ordinated activity (Mariconti et al., 2018).
An even more substantial gap exists between qualitative and quantitative research. There is a particular absence of qualitative discourse analytic approaches. One exception is Ludemann (2018), although the study’s detailed analysis focuses only on three posts among the 100 threads the author collected over 4 months. /pol/ is widely recognised for its unique and evolving language, with acronyms and slang used both seriously, ironically and as identity markers. Qualitative discourse analysis would enable researchers to examine language in context, while conversation analysis would provide a deeper understanding of how the site’s users interact. In-depth qualitative analysis could reveal whether there are particular influencers within these communities shaping the formation of culture and views. Discourse analysis may help illustrate the ideologies used to radicalise or ‘red pill’ community members.
Research design and methodology
Our research design is based on analysing a single, large thread on /pol/, created without our knowledge during an academic workshop we ran in 2019 concerning how best to study 4chan and assess their association with the Alt-Right. At the end, a delegate approached, explaining that they were a regular 4chan user, and that they had created a thread on the site about the event. The thread began with a screenshot of the event invitation. Throughout, the ‘OP’ – ‘original poster’ in chan parlance – had photographed various PowerPoint slides, posting these for comment. The conversation developed from there.
The thread generated extensive interest. A study of 11 million 4chan posts found that the median number of posts on each /pol/ thread was only 7.0, and the mean 38.4 (Hine et al., 2017: 4). In contrast, ours generated 333 posts, from 174 different posters. 98 files were uploaded – a combination of pictures, ‘gifs’ and videos. It was the largest thread on /pol/ throughout the period it was active. The opportunity to comment on how members were being studied was clearly of interest.
Since the dataset contained rich insights about how the /pol/ community talks about itself, we employed qualitative discourse analysis – a method used little in 4chan research. Using it positions our research in a space between ethnographic studies elucidating ‘chan culture’ and informatic approaches using Big Data to quantify and track the spread of memes and hate speech. As mentioned, the near-anonymity afforded to 4chan users, and the short duration in which content remains on the site before deletion/archiving, complicate the research process and limit inferences one can draw about site users. Nevertheless, as Ludemann (2018) explains in their discourse analysis of 4chan, ‘it is important to see how people create and perform digital identities, and mediate their offline ones in a space where identity has been all but stripped away’ (p. 97). Our examination of the practical and ethical challenges of researching such communities advances the literature further, facilitated by the unusual circumstances in which our dataset emerged.
Our first issue was to ensure the preservation of the thread given its impermanence (Bernstein et al., 2011). Fortunately, the participant had archived the thread, but we then had to quickly save, predominantly using screenshots, its contents before its expiry a week later. We then gathered elementary quantitative data on thread size, the number and location of participants, and separated out the different conversation strands within it. Sometimes, a participant may post and a direct reply to their post may only come many responses later. Following these sub-threads allowed us to identify distinct conversational themes. We then coded these thematically for further analysis.
We combined this with a deliberately reflexive approach, for, as will become clear, our perceptions as researchers being analysed by /pol/ are just as pertinent as /pol/’s responses to being studied by us. This reflexive approach includes being explicit about our own reactions to the thread, since we believe these not only to be relevant to our interpretation of its content, but also necessary if one is to assess thoroughly the challenges of researching these spaces. We deliberated carefully on decisions of what to reproduce or omit, conscious of the impact of our research on ‘the people we study and broader communities implicated or impacted by those we study’ (McKinnon et al., 2016: 568). We hope that considering these ethical and practical challenges will serve ‘a pedagogical function’ (McKinnon et al., 2016: 560), as part of an ongoing conversation with other scholars about how best to collect, analyse and represent free-extremist discourse.
The following discussion illuminates how discourse analysis can help address the seven research questions outlined earlier, and what additional research is needed to understand 4chan content, users and their association with the Alt-Right.
Data analysis
Challenges analysing /pol/ content
1. How can we analyse language and images given their dynamic evolution and the difficulty of differentiating serious and ironic content?
Analysing content on /pol/ presents numerous challenges. Language use evolves quickly on the site – which makes it difficult to assess the meaning of terminology. Extensive use of irony by users makes it hard to differentiate serious and ironic content. In turn, as the analyst cannot easily tell if content reflects serious or ironic intent, it is hard for researchers to avoid imposing their prior assumptions and political perspectives onto the text. We address these issues in turn.
A major challenge analysing 4chan/pol/ discourse concerns the community’s dynamic and evolutionary language and imagery. This signals in-group identity under conditions of anonymity. Employing chan parlance, ‘anons’ differentiate themselves from ‘newbies’, while new members are quickly outed and abused as ‘newfags’ when they fail to demonstrate understanding of recognised terminology (Phillips, 2015). Researchers have shown how readily communities employing hate speech can avoid detection using typos, word boundaries or synonyms, such as the use of multiple ‘(((brackets)))’ to imply that whoever is in the brackets is Jewish (Gröndahl et al., 2018). Updated dictionaries to analyse site content are therefore essential (Zannettou et al., 2018a).
Ironically, the challenge of analysing evolving discourse on the site is highlighted by one of the users themselves in a discussion about a workshop presentation slide revealing the most common hate speech terms on /pol/:
– AHAHAah [on] this graph, ‘trash’ is a ‘hate word’, what in the fuck right up there with niggers. I suppose what about ‘poo’? someone didnt do their research. – But ‘trash’ is used to indicate racial animus here, anon. ‘Retard’ on the other hand is used to say ‘that idea is the kind of idea a low-intelligence person would have’. Normies use stupid, dumb, foolish, idiotic, etc. in the same way. Is there any sign English is evolving alternatives to calling ideas low-intelligence? Why would it? We follow smart ideas, the kind of ideas a person who is smart would come up with. Training AI on data with massive categorisation errors like thinking ‘retard’ has something to do with mentally disabled people is retarded. AI’s going to change everything but not without some hilarious blind spots.
Looking beyond the obviously offensive terminology, the user raises valid criticisms about the difficulty large-n quantitative studies have detecting nuances in textual meaning. Indeed, the exchange shows contention between site users about meaning, as well as the challenges newer users face learning site language. The first user seeks to denigrate the research by suggesting that ‘trash’ should not be considered a hate term, and is then corrected by the second user, who points out that on /pol/ it connotes ‘racial animus’. The challenge in analysing sentiment will be expanded on later when discussing the role of influencers on the site (see question 5), when we illustrate that only by examining language use in combination with the images and video footage users post can we develop a more accurate and nuanced reading of the platform’s discourse.
Beyond terminology, another challenge is differentiating serious and ironic communication. This has been referred to as Poe’s law – which states that it is impossible to differentiate serious and parodic online content if author intent is unknown (Phillips and Milner, 2016). To illustrate, consider the following comment responding to the workshop title ‘4chan and the Alt-Right: how best to study them’:
– ‘Study them’. Like fucking animals. These snobbish pseudo-intellectuals deserve to be shot in the street.
Should we as academics fear that a user genuinely believes that we should be shot in the street? Should authorities infer intent to harm from this? How should we feel about it? Should we report it? Should we dismiss it? Or is the user merely ‘shitposting’, 1 by saying ironically something an extremist might say? We cannot know with certainty, and would not be able to without contextualising these comments with an individual’s broader activities, online and offline.
Being unable to determine intent creates a further research challenge: assessments of the site’s dangers risk merely reflecting researchers’ prior assumptions. In the radicalisation literature, for instance, Archetti (2013) argues that much research is ‘stuck in the early twentieth century’, assuming that audiences are largely uncritical of information, and can therefore be ‘injected’ with radicalising content, as per ‘hypodermic’ theories of communication (Esser, 2008). From this perspective, /pol/’s content will seem especially dangerous. Conversely, observers adopting a ‘minimum effects’ perspective (Bennett and Iyengar, 2008) will more likely assume that most /pol/ content will be dismissed as ironic and intended purely for effect.
Even if content on /pol/ is highly offensive, judging how ‘extreme’ the site’s content is depends partly on the observer’s political perspective. In what Salazar (2018) describes as the ‘assumed “normal” public sphere’ within liberal democracies, such content is routinely considered extreme (p. 141). Commentators self-associating with the ‘Alt-Right’ acknowledge this with their stated aim of shifting the ‘Overton Window’: by which they mean normalising discourse that would otherwise be interpreted as extreme within society (Daniels, 2018: 62).
If one instead adopts a critical race theory perspective that sees racism as systemic – routinely embedded throughout the social structure of societies (Bonilla-Silva, 2015; Martinot, 2010), the racist discourses commonly observed on /pol/ (see questions 4 and 5) could be read not as ‘extreme’ necessarily, but as an inherent feature of the system (Flores-Yeffal et al., 2019). As Daniels (2018) summarises, the difference is between writing ‘as if racism is a “bug” rather than a “feature” of the system’ (p. 64). Interpreting it as an inherent feature, what is novel is not that such sentiment pervades and shapes society, it is that the platform provides a space for a community to express openly views they might not in other public contexts. A danger, from this perspective, is that such spaces could reverse the acknowledged trend away from public expressions of racism (Delgado and Stefancic, 2001: 9–10).
These observations are important because they show how researchers’ prior assumptions may shape how they interpret the meaning and intent behind site content. This is potentially even more of an issue when one cannot tell if content is serious or ironic. A researcher trying to identify the next violent extremist may threats of violence on the site at face value; others may read them as non-serious ‘shitposting’. Reflexivity about researchers’ prior assumptions is therefore crucial to explain how they interpret site content.
2. How consistent and coherent are identities and ideologies expressed in site content?
Our analysis finds four recurring political identities and ideologies: commitment to extreme freedom of speech; belief that the community possesses a superior understanding of the world to outsiders; critique of the hypocrisies of liberalism, and belief in the community’s ability to ‘redpill’ site visitors. These were consistent throughout our thread – although further research is needed to substantiate whether similar consistency is found across /pol/ messaging boards, or 4chan more generally.
The thread’s creator or ‘OP’ began by uploading an image of the workshop invitation. They then posted pictures of three slides from a presentation within the workshop on the spread of memes on social media, including from /pol/. One image presented the audience with examples of prominent /pol/ memes, which many posters found highly amusing. There was otherwise little direct information about the workshop’s content. This was revealing, because in its absence, posters instead shared their imagined assumptions of what the content would be:
– ‘4chan is a hateful meme factory, promotes free speech, and needs to be shutdown’. – Academic fags. These people are idiot savants that can’t handle the bantz. – Goddammit, these people won’t quit speaking their minds and forming their own conclusions. Motherfucker!!! This is literally what the whole conference will be. How can we stop independent thoughts from forming and being represented. – Today class, we try to understand and define pure, unadulterated chaos in its natural form. Why do they continue to try?
Here, we see two different ideas through which users identify with the chan community. The first and second posters reject the assumption that the community promotes hate speech, instead seeing it as promoting free speech and independent thought. This reflects a cyber-utopian notion that the Internet should be a place for the free exchange of information, free of state interference, an idea long-associated with 4chan (Nagle, 2017; Turner, 2006). Second, the comments reflect the idea that the group lacks a coherent ideology but its activities instead represent ‘pure, unadulterated chaos’. The terms ‘pure’ and ‘unadulterated’ reinforce the idea of extreme freedom of information, but when combined with ‘chaos’, they suggest that the community represents an elemental force. Traces of this can be found in broader commentary about the group, with their apparent ability to achieve political effect through ‘meme magic’ (Asprem, 2020; Ohlheiser, 2016). These conceptualisations of the /pol/ ‘self’ are contrasted with us as ‘cuckademics’, portrayed as representing a liberal-left consensus failing to grasp the platform’s nature and appeal:
– This is fucking hilarious. The Academics are trying to wrap their heads around the meme culture and failing so miserably. – This [photograph] is one of many ‘researchers’ presenting . . . the general consensus is as you’d expect from the mainstream academia studying and not understanding the fundamental issues behind this phenomenon. – Now offering women’s lib with a BA in white supremacy awareness. Only 75k if you act now!
The community’s self-identification as possessing a superior understanding of the world to outsiders exists alongside participants’ openness about, indeed celebration of, being disenfranchised and marginalised by mainstream society (Marwick and Lewis, 2017). The key link is the idea of being ‘redpilled’. Redpilling is largely a new manifestation of an old phenomenon – the process through which movements convert, or for some, radicalise, people to their cause. As Eric Hoffer (1951) explains, mass movements generate support by denigrating the existing order, drawing on people’s frustration and alienation while offering them a new sense of belonging. Tellingly, users describe red-pilling as empowering; a moment of awakening where, thanks to /pol/, they realised the truth about ‘how the world really is’ (Mountford, 2018). These assertions contrasted us as ‘cuckacademics’ as being ‘naïve’ ‘idiots’, ‘brainwashed’ into false consciousness. Indeed, many posters imagined /pol/’s influence as so great that they could ‘redpill’ workshop participants:
– At least some of these pseudo-academics will get hooked on /pol/ memes and get red-pilled. – They mean ‘how can we learn about 4chan without going there because we’ll be unbrainwashed’? – We’ve got them now. They’ve gotten a taste, and they know the source. Get ready to redpill curious but naive academics and argue with useful idiot marxbots. – They will come here to ‘study’ pol and end up staying here forever. There is no escape from /pol/. Once you taste freedom of thought exchange, there is no going back. – How best to study /pol/. Come on in the water’s fine.
Such posts reveal the community’s belief that /pol/ has unique power to ‘convert’ people. This raises questions concerning the platform’s role in radicalisation – and whether its claims of influence reflect reality. This invites attention beyond platform content to the characteristics and behaviours of its users.
Challenges analysing/pol/users
3. How far can we distinguish between different types of /pol/ user?
Previous studies of 4chan and free-extremist platforms have emphasised the difficulty identifying coherent discourse on the platforms and in differentiating between users (Marwick and Lewis, 2017). Scanning our thread, it is difficult to identify coherence amid a mixture of the analytical, inane, offensive and irrelevant. However, each post is given a code (e.g. 213263611). When someone replies, the reply includes the code of the original post – showing to which comment a user is responding. We can therefore follow separate conversation strands within the overall thread. Doing this we discovered several distinct conversational themes. Some were very brief – maybe one poster and one respondent. Others sustained far longer conversations. Coding these thematically and then combining our original codes into larger categories, we refined the discourse into seven strands, indicated in Table 1. Where a comment contained content relevant to two categories, we coded them based on the majority of the post’s content. We make no claims about the statistical representativeness of this small sample, only that the range of discussion reflects what others have identified as typical on /pol/ (Beyer, 2014; Marwick and Lewis, 2017).
Breakdown of different conversation strands on the thread (
The quality of commentary within each strand varies. For instance, category 2 – specific critiques – contained a mix of astute sociological and technical commentary and more basic criticism of memes others posted as the conversation continued. Shitposts were either generally abusive or not noticeably relevant. The final category, the ‘Influencer strand with white supremacist theme’, specifically references a conversation strand initiated by a single poster, who attempted to divert the thread onto the topic of white racial superiority. We coded this separately not because its content was significantly different from that found on /pol/ more generally. Such content has been commonly expressed on the platform in recent years (Marwick and Lewis, 2017; Nagle, 2017). Rather, we coded it separately because it appeared to illustrate different user behaviours: a focused, persistent attempt by an individual to influence the discussion topic, which produced distinctly different dynamics.
These diverse conversations suggest that 4chan users are far from homogeneous, deriving gratification from different uses of the site. Consider the critique of the workshop, which dominated the first third of the thread. Many early points were general critiques of us ‘cuckademics’, employing age-old stereotypes about academia:
– There’s already books written about 4chan and pol. Who gives a shit. – Have they figured out anything new? Looks like they’re reviewing old data. – Old news, published ~1 year ago and based on data from who knows when. Things move too fast for them. – This is the absolute most pretentious and retarded diagram I have ever seen. Please fuck off and never make a PowerPoint ever again.
Readers may note that some of these are not too dissimilar to comments one might get in journal article reviews, albeit with more expletives. Some, we admit, we found amusing, just as we found other content highly disturbing. As Kathleen Blee (2007) and Tina Askanius (2019) have noted in their fieldwork with different far right groups, it is important for academics to talk openly about the feelings engendered by their research. This seems especially important in our case as emotional provocation is a central aim of much discourse on /pol/.
In addition to the above points, posters demonstrated clear knowledge of some of the sociological and technical nuances of meme propagation:
– The academics don’t get how memes are spread in the first place? It’s because the message is simple enough to be interpreted and can be appended with new information over time. There’s a reason why Minecraft and lego are popular. – Anyone considering ‘memes’ as finite discrete objects to be observed needs to retake sociology 101.
These points were interspersed with crudely transgressive content, characteristic of ‘shitposting’.
– I bet the special ed classes don’t have this retarded shit, raze this place to the fucking ground. – POO POO PEE PEE 4chan IS COMIN FOR YOU
Those engaging in technical discussion tended to comment multiple times, creating longer conversations. Most users posting ‘shitposts’ only tended to post once, confirming that /pol/ users engage in different ways for different purposes. Some engage for genuine debate: the board’s homepage does call (however ironically) for ‘a high level of discourse’ and ‘quality, well thought-out, well-written posts’. 2 Others engage in behaviours more closely associated with trolling, deploying transgressive content to assert their in-group identity, or to provoke others. Or, as we shall see when discussing the ‘Influencer’ conversation strand (see question 5), to promote certain perspectives by leading conversation onto particular topics. Overall, these findings illustrate how discourse analysis makes it possible to identify different categories of user behaviour, even as anonymity limits what can be learnt about individuals.
4. How can we best interrogate the relationship between 4chan and the Alt-Right?
Researchers contest how closely /pol/ and the Alt-Right are linked. For Malmgren (2017), the online culture associated with /pol/ is intrinsic to the Alt-Right, which comprises a ‘gaggle of nihilistic troublemakers indistinguishably aligning themselves with committed racists’ (p. 3). The problem is telling which is which. Our findings more closely echo other researchers who suggest that associating the two is defensible since at least 2014 (Moore, 2018; Nagle, 2017), but doing so would be oversimplistic before that given 4chan’s varied history of political activism. Much of this concerned causes more readily associated with the political left and, through the actions of Anonymous, forms of political activism far harder to categorise (Coleman, 2014; Olson, 2012). Conflating 4chan with the Alt-Right also risks downplaying the existence of historical and contemporary contention about the identity(ies) of the /pol/ community. More fundamentally, it ignores that 4chan contains dozens of different messaging boards, of which the ‘politically incorrect’ /pol/ is only one. Still, that the /pol/ community is a site in which discourse associated with the Alt-Right circulates is widely recognised, with researchers citing significant increases in hate speech on 4chan since 2015 especially (Arthur, 2019). The community’s own memes allude to this. Figure 1, for instance – taken from a separate /pol/ discussion – charts Pepe the Frog’s evolution from an obscure cartoon character in 2004 3 to a ‘symbol of hate’, as labelled by the Anti-Defamation League in 2015. 4
See http://archive.4plebs.org/pol/thread/90738931, accessed 11 October 2019.
Our thread contained copious hate speech readily associated with the Alt-Right and/or far-right. Nevertheless, the link between 4chan and the Alt-Right implied by our workshop focus was questioned by several posters:
– Leftypol has a large presence. Why they only focus on altright? – What is this ‘alt-right’ they keep talking about? – 4chan is not one person or ideology, and you should ask why you are being taught this. Literally, who is funding this study and why.
These points may be being made seriously or ironically. They may be being made by users attempting to discredit the association between the community and the Alt-Right or far-right. Again, we cannot tell. Some even suggest the Alt-Right link reflects a conspiracy against them:
– Remember when CNN made a custom KKK pepe to smear us as dumb evil racists on tv? These people think it’s a noble pursuit to manufacture evidence to suit an agenda. – It’s because the left is literally creating these to smear us.
When seeking a deeper explanation, posters look beyond mainstream media and ‘the left’ and instead suggested that the Alt-Right link, and the workshop itself, reflected an anti-Semitic conspiracy. This can be seen in users’ self-perceptions of the workshop’s purpose, using variants of the anti-Semitic Happy Merchant meme, common on /pol/ (Figure 2).
‘Happy Merchant’ meme implying an anti-Semitic conspiracy theory.
This is ironic – perhaps intentionally so. In rejecting the claim that they are racist and associated with the Alt-Right, users credit a conspiracy theory that is racist and associated with the Alt-Right. Returning to Malmgren (2017), what we do not know is whether these are posted by ‘nihilistic troublemakers’ or ‘committed racists’.
Rather than fixating on a given user’s intent, or whether an individual user is racist, an alternative approach would be to focus on such discourse’s individual and societal effects. Studying discourse on 4chan through that lens, individual intent matters less. As Ludemann summarises (2018), much content on /pol/ ‘can be understood as racist, racialized, or otherwise unnecessarily vitriolic and violent, yet is viewed by many in this context as everyday discussion, removed from social policing by politically “correct” others’ (p. 92). A key element of ‘free-extremism’ is for users to express whatever they like, however offensive, regardless of whether they believe it (Moore, 2018). Near-anonymity reduces the potential personal cost of doing so, and irony can be used as an excuse to avoid responsibility for one’s words or the harm they might cause (Palmer, 2019).
However, even if racist content is used ironically, it reproduces discourses that could normalise systemic racism within societies (Lawrence and Keleher, 2004). The extent to which racism is embedded in (and by) the digital world outside 4chan is beyond the scope of this study – for more on this see Daniels (2018) and Flores-Yeffal et al. (2019). Still, it seems clear that /pol/ represents a space in which such content is reproduced and normalised, even if the sincerity with which individual users hold such views varies. When interpreted as a form of discursive violence, hate speech has the potential to cause ‘real harm to real people’ (Matsuda et al., 1993), whether communicated ironically or not.
Overall then, discourse analysis can show that many 4chan/pol/ users post content generally characteristic of the Alt-Right and far-right. Moreover, it can be shown that the community actively functions to normalise the expression of racist discourses, among a wide variety of other content. Ignoring the extensive use of irony and parody would provide an incomplete picture of discursive practices on the platform, which are far more varied than the reproduction of long-established racist tropes. Nevertheless, in terms of the effects of such discourse, the harm to individuals and societies may be significant, regardless of the intent of the individual user.
5. To what extent is discourse on /pol/ shaped by influencers and opinion formers?
A related, but as yet unresearched question, concerns how far influencers and opinion formers shape discourse on 4chan. The role of influencers is a prominent issue in social media research and has reinvigorated interest in older communication theories such as the two-step flow model (Katz and Lazarsfeld, 1955; Roch, 2005). A recent article identifies a community termed the ‘Bakers Guild’ on 4chan, responsible for keeping conversations going by posting new threads on the same general topic after a previous one expires (Bach et al., 2018). Beyond this, however, the role of influencers shaping 4chan discourse has received little attention. On the surface, anonymity makes identifying influencers difficult, while Big Data approaches have tended to eliminate the individual poster in favour of mass analysis of text frequency.
In contrast, our analysis reveals that identifying attempts to influence discourse on a /pol/ thread is possible, and a potentially useful research area. We illustrate this by analysing an ‘Influencer conversation strand’ that emerged approximately a third of the way through our thread. Up to then the thread had mainly been focused on critiquing the workshop, with a mixture of technical discussion, individual expressions of abuse and crude ‘shitposting’ not obviously related to other users’ comments. Then, a single user, named ‘pwex5ur4’ on the site, posted 14 videos in quick succession, accompanied by extensive commentary. This conversation thread was distinct from what had come before – a series of highly focused videos on a single topic by one user. We coded the theme of the videos as white supremacist. This was because each of pwex5ur4’s videos presented situations apparently designed to racialise whites as superior, moral actors, and blacks as comparatively inferior and immoral (Martinot, 2010).
We cite a limited selection below (and not the video content itself), conscious of the need to minimise harm, but also because of the importance of illustrating how pwex5ur4 sought to shape debate. Examples included the following:
A group of black adult males looting a canteen.
Multiple videos of black females in fights and brawls, seemingly on college campuses.
A video entitled ‘fetal hammer punch’ in which a pregnant black female hits her own abdomen with a hammer.
A white male repeatedly placing a piece of rubbish into a bin, and two black females repeatedly removing it.
A black adult male drowning, and a white adult male jumping in to save him, and both drowning.
The content is consistent with what Flores-Yeffal et al. (2019) describe as a ‘white racial frame’, selected to portray black people as inferior, criminal and immoral. The video of the white male placing garbage into a bin and black females removing it lacks context – it may be that it is a recycling bin in which the content should not be placed, but that is unclear to the viewer. Lacking context, the intended impression seems to be of the white male acting morally by discarding rubbish appropriately, and the black females doing the opposite. Commentary on the drowning video illustrates the general idea of white supremacy even more strongly. As one poster described it, the black male in the video ‘can’t swim and a white guy tries to save him, but gets dragged down in the process. Western civilisation summed up’.
Basic quantitative analysis suggests that pwex5ur4 had some success in influencing the thread’s focus. Resultant conversation about the video content comprised 68 posts – 23.1% of the thread’s total. Interestingly, no comments challenged them. This may provide circumstantial evidence of tacit acceptance – it may not. The lack of pushback from other users speaks to the routine nature of such discourse on the platform. The community is recognised to police the boundaries of acceptable discourse by verbally abusing members who appear unfamiliar with norms of expression (Phillips, 2015). Consequently, any who disagreed with it might feel unwilling to respond for fear of abuse. Alternatively, other users may simply have ignored them and continued with their own conversations about the workshop and other issues.
These examples illustrate how 4chan/pol/ is not just a community where individuals vie to produce provocative content to generate attention before their posts are rapidly superseded by others. It is a community in which individual influencers promote discourses of white supremacy, potentially contributing to normalising their expression.
While discourse analysis can identify this behaviour, we cannot tell for certain that this was pwex5ur4’s intent. They may indeed be posting to radicalise others, but they might instead be posting ironically in response to the workshop’s association of the Alt-Right and 4chan. They might have just been trying to provoke, for ‘the lulz’. Further analysis reveals additional telling information. Their video uploads occurred very quickly – six were posted in 4 minutes. This suggests that they were uploading pre-existing items from their device(s) rather than having to find content, view it and decide it was worth uploading. It is therefore probable that they were sufficiently comfortable with such content that they were willing to store it on a personal device. This raises the possibility of whether tracing the spread of specific video content, combined with detailed analysis of patterns of language use, might make tracking users possible across different threads and platforms.
pwex5ur4’s sub-thread illuminates important questions about influencers’ roles on largely anonymous platforms, such as 4chan/pol/. Studies on 4chan have mainly focused on quantifying memes or specific terms, which removes individuals from analysis. Alternatively, they have used ethnographic methods to assess the culture of such communities. Our discourse analysis advances understanding by illustrating attempts by individuals to dominate and shape conversations. Measuring influencer impact on beliefs or behaviours will always be immensely challenging on a pseudo-anonymous site. However, tracing the actions of individuals may help inform how the community’s culture and norms evolve, and the role individuals play in the normalisation of extremism.
The white supremacist-themed sub-thread also illustrates a crucial point concerning how we analyse extreme language. pwex5ur4’s commentary was explicitly racist, including mimicking the language of nature shows to dehumanise black people by depicting them animalistically. Extremely offensive as this is, their nature show-imitating commentary only contained one hate speech term that quantitative content analyses tend to track. The content would, therefore, almost certainly have been missed by quantitative analyses of hate speech, even though arguably it is far more insidious than a list of expletives used in jest. This demonstrates the importance of combining quantitative content analysis and qualitative discourse analysis to understand how language is used in context on sites such as 4chan/pol/ to reinforce extremist discourses. Otherwise, misinterpretations or omissions from findings are inevitable.
Challenges to research ethics and validity
6. What are the ethical challenges to researchers studying 4chan and similar free-extremist sites?
Addressing the previous research questions involves overcoming multiple ethical challenges. Extremist groups are increasingly moving to encrypted platforms, making ‘lurking’ harder. This raises the dilemma about whether to participate clandestinely in such groups, or even adopt a new identity or pseudonym, given that transparency would likely lead to retaliation from participants.
Reproducing extremist material potentially disseminates it further, a dilemma several researchers express (Martineau, 2019; Philips, 2018). One presentation slide in our workshop contained examples of offensive memes used on 4chan. Once the OP posted these, several users indicated that they had not seen them before, even suggesting that they should be posted elsewhere:
– other than the first one, which is rare, I have never seen these others on 4chan. – So the memes are true top fucking kek, I feel like this could be on /g/ and get serious laughs. – Don’t they understand that showing these will just increase the influence of /pol/?
In effect, we unwittingly brought back to prominence offensive memes that were otherwise marginalised. We inadvertently spread extremist content – hence our decision not to reproduce these memes here.
A further ethical issue concerns potential harm to researchers examining extreme content. Whether researching such platforms might lead to academics being ‘redpilled’ is questionable, though should not be dismissed. Academics have acknowledged these concerns themselves. Julia Ebner (2020), for example, describes fearing that she was being radicalised while researching the ‘Red Pill Women’ online group. A more prominent ethical issue concerns the mental health implications of prolonged exposure to extreme content. Media coverage has highlighted this among workers removing extreme social media content (Solon, 2017). Ethnographers have highlighted the despondency, numbness, intimidation and isolation they have experienced studying far-right platforms immersively (Askanius, 2019: 884).
Potential harm extends beyond researchers to their audiences too. /pol/ members were calling not just for our identities to be revealed but our workshop participants too. We had neither anticipated this nor pre-warned participants. Screening participants based on self-reported online activity would be unethical and impractical. Restrictions on permitted classroom behaviour may be more acceptable. Either way, our example illustrates the urgent need for more scholarship examining the ethical implications of researching and teaching about free-extremist platforms known for responding aggressively to observation and critique.
Extremist material may offend or harm any individual. The threats of violence on our thread were targeted at us and our workshop participants generally, without posters knowing anyone’s gender or ethnicity. But since /pol/ content is frequently racist, anti-Semitic and misogynistic, it is important to note that ethnic minorities, women and certain other identity groups are at particular risk. The thread on our workshop was notable for its lack of misogynistic content, which researchers have previously identified as a common aspect of discourse and culture on platforms such as /pol/ (Baele et al., 2019; Marwick and Lewis, 2017). That this received little attention here is simply because this was largely absent from our thread, though we recognise that it remains a vital consideration for researchers too. The two authors of this study are white males, something we recognise puts us in a less vulnerable position as researchers. Still, we found certain content highly disturbing, including a specific video that still haunts us today, which we choose not to discuss in order to minimise harm. In this way our affective responses during the research process shaped our decisions of what to reproduce here. However, they may also have affected unconsciously how we responded to the discourse in other ways. Such examples illustrate the futility of pretending to be emotionally detached or ‘objective’ observers when studying platforms such as 4chan/pol.
7. How does /pol/’s response to observation affect research validity? How can this be resolved?
A further ethical and epistemic challenge in analysing 4chan/pol/ is that it is not an object that can be passively observed. The community has been repeatedly shown to be sensitive to its public portrayal, with a history of collective action against critics and researchers (Hine et al., 2017; Martineau, 2019; Phillips, 2015). This includes altering their language to confound research. When combined with researchers knowing that retaliation is possible, this creates a mirroring dynamic. Academics may alter /pol/ by observing it, but in turn academics may be influenced by knowing that /pol/ is watching.
These dual observer effects present a challenge for the validity of findings and an ethical dilemma. To illustrate, one workshop presentation contained a content analysis of the board’s most frequent hate speech terms. Once images of this were uploaded, /pol/ users immediately started distorting the figures – in many cases, adding dozens of their preferred expletives after their posts:
– Those are rookie numbers, pump them up a bit /pol/ [ – Surely we can do better. Let’s get those numbers higher [ – lets get those [ – We should use steghide each time we post to hide a random string.in the image giving it a different hash without affecting its appearance.
This data manipulation illustrates how the community could game quantitative research to reproduce their preferred hate speech terms, for instrumental political purposes or simply for ‘the lulz’. The /pol/ community has a record of this, be it gaming opinion polls and Reddit threads during the 2016 US presidential election or Anonymous’s past attacks on the Church of Scientology and HB Gary (Olson, 2012). These examples illustrate the need to design research that recognises the interaction between /pol/ and the researcher rather than seeing it as a community that can be externally observed without consequence.
The flip side of academics altering an object of study by observing it is that academics may be influenced by knowing that 4chan is watching. /pol/, and associated communities, have long been associated with attacks on those trying to study or criticise them (Martineau, 2019), be it organised ‘raids’ on YouTube channels or ‘doxxing’ people by revealing personal information about them (Hine et al., 2017; Marwick and Lewis, 2017). Traces of these practices can be found in our thread:
– See if you can grab the presentation and upload it OP. Super interested in this. Also can you list the presenters/cuckademics? Thanks. Have a qt. – Somebody should report this lecture as hate speech and have them arrested for not having the proper licence. – Bongbros, call the police and report them for distributing offensive materials. – Given that we know Nazis have infiltrated this talk, can we use the ‘members’ list in that Facebook group to deduce the identities of these people?
This raises legitimate questions about the effects on research if such responses, or anticipation of them, shape how academics study or write about the community. Should researchers make themselves and their objectives visible? Anthropological ethical codes emphasise that researchers should be visible and transparent, though this is based on communities that are themselves visible (or not anonymous) and where visibility will not expose researchers to manipulation or danger. Researchers are routinely treated as hostile by the chan community, and as a potential threat to its continued freedoms. This complicates the application of traditional ethnographic or anthropological approaches.
There is no straightforward solution. The student who began the thread on our workshop may have been doing so opportunistically, just for ‘the lulz’, or with a pre-determined agenda to influence future academic research. We cannot know for sure. We do know that at least one user expressed the assumption that ‘I bet those academics are in this thread’ – suggesting at least that the thought had crossed some users’ minds. This may have shaped the behaviour of users who may actively desire the publicity that would be gained from offensive content being reproduced verbatim by researchers.
Factoring these interactions between researcher and subject into research design involves making difficult choices about what to include and what to omit. That the site’s content is routinely offensive, and its users may be posting it to gain publicity, adds layers of complication to decisions of what discourse researchers should reproduce. Here we have made a series of choices that attempted to strike a difficult balance, recognising that these choices will be inherently selective and hence political (McKinnon et al., 2016).
We concur with Whitney Phillips’ warning about the dangers of amplification and have sought to reproduce only the minimum text necessary to provide a detailed grasp of these discursive practices, while limiting harm to readers (excluding, for example, all videos and most images). We have also sought to balance /pol/ users’ apparent desire to get egregiously offensive content reproduced outside the site with the desire not to reinforce the community’s commonly expressed view that academia is stifling free speech through excessive political correctness. By being open about our decisions, we hope to spark further conversation about how to navigate these ethical dilemmas. Whichever decisions are made, findings will be more valid if researchers – quantitative and qualitative – consider these interaction effects, and are openly reflexive about them.
Conclusion
This article has sought to demonstrate the value of discourse analysis in deepening understanding of free-extremist platforms, such as 4chan/pol/. It has done so through a unique dataset: a 4chan/pol/ thread created without our knowledge during a workshop we ran on studying 4chan and the Alt-Right. Addressing seven challenges analysing /pol/’s content, users and the broader research challenges scholars face, we have illustrated how discourse analysis can reveal nuances unavailable to Big Data approaches. We have highlighted different categories of /pol/ user, ranging from crudely offensive shitposters to those engaged in nuanced technical discussion. We have revealed the sense of superiority that unites the community, and its belief in its power to ‘redpill’ or radicalise others. We have examined for the first time the significance of influencers on the platform, demonstrating that extremist discourses on /pol/ may reflect individual efforts to shape conversation rather than emergent, collective behaviour. In the process, we have established the possibilities and limitations of what can be learnt about individual users. Finally, we have reflected on an under-researched ethical dilemma: how to research free-extremist platforms when their participants may be responding to – and seeking to influence – that research.
Our findings suggest several practical ways policymakers and researchers might proceed. Regarding radicalisation, our analysis suggests limited potential in identifying behavioural radicalisation (Neumann, 2013) and offline trigger events using the site. Any analysis is likely to be plagued by false positives and by the difficulty tracking individual users. Our thread contained eight threats or calls to action, none of which – to our knowledge – was taken up at the time. We see greater potential in analysing cognitive radicalisation on 4chan and similar sites. Partly this is because our analysis reveals the possibility of identifying the contribution of ‘influencers’ to specific threads and their attempts to lead conversations towards extreme topics, or to promote extremist ideas, language or memes.
Future research would benefit significantly from investment in online storage of platform content. This would help address the problem of ephemerality and could enable consistent classification and categorisation. Researchers still use manual screenshots to archive 4chan content and store datasets locally. Practical methods for capturing, grouping and analysing threads – and conversations within them – would make research faster and more accessible.
Ideally, future research into free-extremist platforms would combine quantitative and qualitative approaches, encompassing ethnographic, informatic and discourse analytic methods. Here we have reinforced existing research that 4chan/pol/ is a community in which expressions of racism and white supremacy appear normalised, whether used seriously or ironically. But we also show that to understand how users seek to achieve this discursively requires going beyond quantifying patterns of hate speech, either because they are too easy to manipulate, or that users employ other language to avoid detection. Multi-disciplinary approaches could better assess the relationship between 4chan users and the users of other sites and services, such as Gab.ai, Discord, Telegram, Reddit and 8chan. Quantitative research could benefit from the contextualisation of images and text that qualitative approaches provide. The combination would help assess claims about such platforms’ societal influence and their ability to radicalise individuals and make broader political discourse more extreme.
Research is also necessary into the appropriation into ‘mainstream politics’ of rhetorical methods associated with 4chan and /pol/. Just as 4chan/pol/ evolved from association with Anonymous to a closer association with the Alt-Right, the methods used on /pol/ now appear to have been appropriated and re-interpreted by ‘mainstream’ political parties and governments (Bradshaw and Howard, 2017). References to the QAnon conspiracy theory – which began on 4chan – are increasingly prominent in US party politics. Both the Australian and British general elections in 2019 saw traditional parties employing tactics based on a reinterpretation of ‘shitposting’, whereby they deliberately posted low-quality content on social media to provoke attention (Stokel-Walker, 2019; Williams, 2019). Pepe the Frog was found daubed on government buildings during Hong Kong protests as a symbol of non-conformity and dissent, though in a very different context from its previous appropriation by the Alt-Right. The appropriation and re-interpretation of discursive tactics and symbols from such platforms by others seeking political change reinforces the importance of further study, mindful of the potential harms and amplification involved in doing so.
The users on our /pol/ thread are right about several things. We do have much to learn to understand what one described as ‘le mystical 4chan’ and similar free-extremist communities. Their communication dynamics may not represent ‘pure chaos in its natural form’ as they suggest, but it is highly complex. They are right that we know too little about ‘the fundamental issues behind the phenomenon’. This is the task for researchers as we seek to understand better a key source of political extremism in the digital age.
Footnotes
Funding
ORCID iD
