#BLM Insurgent Discourse, White Structures of Feeling and the Fate of the 2020 "Racial Awakening"

Abstract

Working with Twitter data, this paper offers new findings on the #BlackLivesMatter movement and "racial awakening" of summer 2020. Framing methods to address this important moment, this paper contends that cultural studies and critical race studies can be enriched through an engagement with new computational approaches. We analyze how white and racial minority voices talked about race and track their fraught contestation for leadership of racial discourse over the summer of 2020. We uncover a surprising story of white colorblindness even in the midst of a "racial awakening," a story that questions claims that the Trump presidency and the summer of 2020 ushered in a new era of US racial consciousness. And we show how a Black and minority discourse with transformative potential surged and receded. For cultural studies, our data and analysis revise Raymond Williams's influential model of cultural evolution by introducing a new concept: the insurgent, a long-building minority cultural strain that surges to contest the dominant culture in a moment of crisis. For critical race studies, our findings revise prominent theorizations of colorblindness, racial ideology, and hegemony. By revealing the messy and unconscious feelings characterizing colorblindness, our data contest theorizations of colorblindness as an ideology and counter the focus on articulate beliefs in theories of racial hegemony. Ultimately, this paper shows that bringing data methods focused on moments of cultural contestation and mass communication into dialogue with field-specific theory and qualitative analyses can expand our models of how race, discourse, and culture operate.

In the wake of George Floyd's murder at the hands of police on May 25, 2020, the largest social justice protests in US history erupted. Millions took their bodies into the streets and their words online to show their support for Black Lives Matter (BLM). The summer of 2020 appeared to many to be a "racial awakening" that promised transformative change in US racial discourse and the racialized structures of the nation. As everyday Americans, celebrities, and major corporations avowed their support on social media, the transformation in online racial discourse seemed particularly promising and appropriate for a movement that had emerged seven years earlier on Twitter. Figure 1 shows the explosion of public interest in Black Lives Matter on the platform that summer. In May 2020, the number of #BlackLivesMatter (#BLM) tweets reached an unprecedented peak—more than three times greater than its previous apex in 2016.

Looking back at this moment three years later—with many reforms to racialized police violence stalled and the centrality of BLM discourse faded—we can ask, was there really a "racial awakening"? Taking advantage of access to Twitter data made possible by the introduction of the Twitter Academic Research access API in 2021, this paper offers new findings on the unfolding of the 2020 "racial awakening." With a wealth of publicly available data, scholars of culture, race, and discourse have a valuable chance to analyze the high point of the most important racial justice movement of the twenty-first century, a moment that exemplified how social media platforms transform the pace and nature of cultural change. In contrast to much commentary on the summer of 2020, we can investigate without relying on anecdote or intuition: as a discourse, how successful was #BLM in 2020? How pervasive and enduring was this discourse in online publics? How did white and racial-minority voices interact, influence others, and form communities?

In what follows, we offer a layered account of the shifts in US racial discourse in the summer of 2020 that complicates received narratives

Fig 1.

of a "racial awakening" and fleshes out this story with previously unobserved textures. Our findings add substance to the claim that Black Lives Matter discourse achieved a historic breakthrough while offering a dramatic account of its dissipation. We find striking distinctions between how white and racial-minority voices talked about race and track their fraught contestation for leadership of racial discourse as it evolved over the summer (Table 1 offers a taste of the distinct ways that white versus nonwhite users tweeted about the moment). In the content of this discourse, we uncover a surprising story of white colorblindness even in the midst of a "racial awakening," a story that revises established theories of colorblindness and questions widespread claims that the Trump presidency and the summer of 2020 ushered in a new era of US racial consciousness. And we show how a Black and minority radical discourse with transformative potential both surged and receded.

These findings are important in and of themselves, but #BLM in 2020 also presents an important case study that challenges the theories and methods that cultural studies brings to the analysis of cultural change, racial ideology, and hegemony. The intertwining of social movements with new communications technologies, such as social media, affords a rich opportunity to study the relationships between and among race, discourse, and social change—a topic of longstanding interest. Perhaps the greatest affordance, leveraging computational methods, is that we can now study this relationship at multiple scales: at the scale of large

Table 1¹.

online populations and the scale of a single utterance. For scholars of culture interested in the relationship between discourse and race, the opportunity is significant. We can now ask with greater precision, how do new cultural discourses of race and identity emerge and circulate? How do such discourses vary across different racial groups? How do they achieve hegemony and fade away?

This essay brings together three different scholarly frameworks: (1) cultural studies, particularly theories of cultural change and emergence from Raymond Williams; (2) critical race studies, particularly theories of whiteness and colorblindness from Eduardo Bonilla-Silva; and (3) information science, particularly recent work in natural language processing and text-as-data. Our major contention is that the first two scholarly frameworks—historically rooted in qualitative methods and, thus far, largely untouched by recent innovations in data science—can be powerfully enriched through an engagement with new computational approaches.

For cultural studies, our data and analysis of #BLM 2020 significantly revise Williams's influential model of cultural evolution, particularly his three-part schema of "dominant, emergent, and residual" forms of culture.⁸ Our study introduces and theorizes a new concept: the insurgent, a long-building minority cultural strain that surges to contest the dominant culture in a moment of crisis. We show how a powerful insurgent culture reframed the terms of US racial discourse before it was contested and absorbed by the dominant racial culture of post-Civil-Rights-era America: colorblindness.

Our findings also revise prominent theorizations of colorblindness, racial ideology, and hegemony from critical race studies scholars such as Bonilla-Silva, Ian Haney López, Michael Omi, and Howard Winant. By revealing the messy and unconscious feelings shaping colorblindness, our data contest theorizations of colorblindness as an ideology and counter the focus on articulate or explicit beliefs in theories of racial hegemony.

Our essay calls on scholars of race and culture to take seriously social media, and the data-driven methods used to study it. Williams himself predicted more than sixty years ago that new media and electronic communications would transform how cultural and social transformations unfold.⁹ We have little doubt that, had he lived to see the emergence of social media, he would have found it to be a crucial object of study; platforms such as Twitter exemplify the changes in communications he anticipated. But these platforms also make clear the need to update his methods and models to meet the nature of cultural and social change now. In 1961, Williams had already identified the present-day issues of scale that "distant reading" practitioners emphasize—the problem of generalizing about culture based on a small sample of texts: "Nobody has read all the English novels of the 1950s; the fastest reader, giving twenty hours a day to this activity alone, could not do it" (LR 81, 71). This problem has only intensified since Williams with the massive volume of cultural artifacts produced on the web today. While it's crucial to keep in mind his caveat that even comprehensive access to the records of a culture is not the same as access to the culture as lived, data methods that can analyze large corpora of texts serve to amend Williams's model so that they approach the scale at which he understood and theorized culture to transform, develop, and expand.

Data and Methods

Our study focuses on Twitter, and our data draws entirely from this platform. Researchers have established the importance of Twitter to the formation and development of Black Lives Matter, as well as the power of the #BlackLivesMatter hashtag in its impact.¹⁰ It goes without saying, however, that we limit our claims to Twitter users—our claims do not encompass "the public" more generally. In our conclusion, we discuss how our findings relate to new, survey-based research on the impact of Black Lives Matter on popular opinion.

Our dataset consists of 4,383,780 tweets from 5,244 unique Twitter users posted between January 1, 2020 and December 31, 2020. The tweets represent all of the original tweets (nonretweets, nonreplies, and non-quote tweets) posted by this group of users during this period. These users were selected by two criteria: (1) they are Twitter "verified" users, and (2) between May 1, 2020 and August 31, 2020—often referred to as the "BLM" Summer of 2020—the users posted a "high impact" tweet using the #BLM hashtag. We define "high impact" as a tweet that fell in the top decile of most liked and/or retweeted tweets that used the #BLM hashtag during this period. We focused on verified users because we aimed to collect information on their social identities based on public records and online sources, and these users are more likely (though not guaranteed) to be public figures. We focused on users who posted a high impact #BLM tweet because we were interested in studying users who were highly engaged with this online discourse and played a nontrivial role in shaping its development.

Using public records and online sources, we collected demographic information on this group of users, focusing on gender and race. 49% of the users in our dataset use he/him pronouns while 51% use she/her pronouns. 57% identify as white, 24% as Black, and 10% as non-Black racial minority (Asian American, Latinx, and Indigenous). Our gender breakdown aligns closely (49% male and 51% female) with a recent study that examines the demographics of #BLM tweeting on Twitter during this period.¹¹ However, our racial breakdown significantly over-represents Black and other nonwhite users; Sarah Shugars and their collaborators identify #BLM tweeters to be 83% white, 11% Black, and 6% non-Black racial minority (numbers that align with a Pew study of the overall racial demographics of Twitter).¹² Most likely, our dataset under-represents white users relative to a more general sample of Twitter users because we focus on users who were particularly influential with regard to Black Lives Matter discourse on the website (and, indeed, Shugars and her collaborators find that white users have less impact than nonwhite users when looking at retweet numbers: nonwhite users are more likely to be retweeted than white users relative to their share of original tweets).¹³

Research on #BlackLivesMatter online activism often looks at gender and/or political affiliation in studying the effects of racial identity on discourse.¹⁴ For this essay, however, we focus exclusively on racial identity in order to test several race-specific hypotheses from the critical race theory literature. In our conclusion, we outline suggestions on how future research might incorporate information on gender and political affiliation. Further, while #BLM research often also analyzes different racial groups as distinct entities, we aggregate these groups as a single "nonwhite" category. We are primarily interested in testing claims regarding whiteness and related social discourses of "colorblindness," as well as their relation to discourses that emerge from peoples marked as outside of this space—individuals racially marked as nonwhite. We do not aim to test claims specific to any racial group, although in our conclusion, we also sketch out ideas for follow-up work in this area.

To analyze our corpus of tweets, we rely on three computational tools that allow us to model large-scale trends and pick out representative and outlier tweets. The first is topic modeling. An unsupervised machine learning method, a topic model takes a large corpus of documents (in our case, tweets) and identifies a set of topics or themes (the number of which is predetermined by the researcher). Each topic is represented by a probability distribution over the dataset's vocabulary, and each document is represented by a probability distribution over the topics. Topic modeling is well-established in the digital humanities and computational social sciences.¹⁵ We trained our topic model using MALLET, a popular software implementation of the Latent Dirichlet Allocation (LDA) algorithm.¹⁶ We tested different numbers of topics and found that 150 topics produced the most legible results (meaning it produced fewer topics that combine multiple themes and more topics representing racial themes).¹⁷ The themes in our 150 discovered topics range widely. Last, we manually identified twenty-three topics as explicitly race-related and limited our analysis to these topics.

The second method we use is Kullback-Leibler divergence (KLD). KLD is an information-theoretic measurement that quantifies the difference between two probability distributions. This method has been previously used in the digital humanities to compare topic modeling distributions.¹⁸ Given two probability distributions, KLD returns a nonnegative, nonsymmetric score, where higher scores indicate greater differences.¹⁹ These scores allow us to compare sets of tweets using their averaged topic distributions. Following Alexander T. J. Barron et al., we compute KLD in two different ways: first, we compute the KLD score for each tweet compared to the aggregate mean of topic distributions for the seven preceding days, measuring how "surprising" the tweet is compared to its seven-day past. Barron et al. refer to this measurement as novelty. Second, we reverse the time comparison: we compare each tweet to the next seven days, measuring how "surprising" the tweet is compared to its seven-day future. Barron et al. refer to this measurement as transience. Novelty tells us how innovative a tweet's themes are; transience tells us how ephemeral a tweet's themes are. Last, we compute resonance, which is novelty minus transience. Resonance quantifies how innovative and enduring a tweet's themes are. To be clear, high resonance does not necessarily indicate that other users were directly influenced by that tweet. Rather, resonance simply says that the tweets posted after the tweet of focus were similar in their thematic content, though these themes could have arisen due to external factors not included in our dataset.

Our final method is pointwise mutual information (PMI), a statistical measure of association. PMI quantifies how likely two events are given their joint distributions and their individual distributions. By comparing the aggregate topic distributions of each group, this method allows us to calculate which topics are most distinctive or probable to one group in contrast to other groups. It tells us which topics are most associated with different groups.²⁰

Dynamics of the "Racial Awakening"

In the days following May 25, the video of George Floyd's murder captured the world's attention and set off an eruption of racial-justice protests and discourses on race. The intense energy of this surge can be felt in the skyrocketing of #BLM-related tweets from the users in our dataset. We looked first at all users in our dataset (Figure 2) and then users in our dataset broken down by racial identity (Figure 3).

But for such energy to coalesce into a discourse and a movement required voices to direct it. Many people were struggling with how to talk about the explosive subjects of racial violence and racial justice. To see who led this directing and to observe how the social form of this discourse organized itself, we turn to a computational analysis of resonance.

Fig 2.

Fig 3.

Fig 4.

Figure 4 divides the results of this analysis by identified white versus nonwhite users. This plot shows the evolution over the summer of these two groups' resonances compared to all users in our dataset. The higher the resonance score, the more this group of users is forecasting the future directions of this discourse as a whole. Figure 4 also marks two subperiods (Period 1 and Period 2) of particular interest during the #BLM summer. We identify these two subperiods in order to flag specific moments in which patterns of resonance based on our two racial subgroups dramatically alter.

Figure 4 tells a story of influence and contestation across white voices as well as Black and other racial-minority voices as they struggle for influence within the racial justice discourse of social media. During the three weeks following George Floyd's murder, Black and other racial-minority voices led the way. For most of Period 1 (May 25 to June 16), nonwhite resonance scores were positive and higher than white resonance scores, which remained largely in negative territory. The skyrocketing in overall #BLM tweets in the days after May 25 overlaps with this period of increasing Black and racial-minority resonance. As the surging energy of the discourse sought direction, nonwhite voices on Twitter were most resonant.

What did this period of minority-led discourse sound like? What was its actual content? Using our PMI method, we can identify the race topics that were distinctive of nonwhite users during this time—the content that defined this short era of racial resonance. Table 2 lists the top five distinctive topics of nonwhite users in this subperiod:

Table 2.

Top 5 Distinctive Topics of Nonwhite Users in Period 1

The topics call for radical change and protest to address issues of white supremacy, racialized police violence, and the criminal justice system. The consciousness of race in their discourse is crucial, as one characteristic tweet makes clear:

The Color Of My Skin IS NOT A WEAPONThe Color Of My Skin DOES NOT MAKE ME DANGEROUSThe Color Of My Skin SHOULD NOT MAKE YOU HATE METhe Color Of My Skin DOES NOT WANT YOUR BULLETS IN ITThe Color Of My Skin, The Color Of My Skin, The Color Of My Skin#BlackLivesMatter²¹

In this initial period, each of these topics specifically reference Blackness, Black organizing, Black victims, and/or police killings of Black people. Black and minority voices direct the discursive energy in clear ways: toward support of Black communities, Black victims, and Black-led movements and against white supremacy and a police system that defends it. They placed at the center of social media discourse the thoroughly racialized nature of the US. "Join the #StrikeForBlackLives," another tweet enjoined. "Workers are withholding their most valuable asset—their labor—in support of dismantling racism and white supremacy, to bring about fundamental changes in our workplaces, economy, and society."²² Their discourse is distinguished by radicalism and specificity, calling for direct action and transformation targeted at the central issue of police violence against Black people. They called for justice for specific victims—George Floyd, Breonna Taylor—and for arrests of specific perpetrators. "We will continue," one tweet declared, "to stand in solidarity with the people of Louisville, Kentucky to demand the arrest and conviction of the murderers of Breonna Taylor. #SayHerName #NoJusticeNoPeace."²³

The Textures of Colorblindness

Yet, as Figure 4 indicates, this Period 1 of racial-minority-led discourse was short lived. Within weeks, this discourse was overtaken and displaced by white-user-led discourse around #BLM and race more generally. During Period 2 (June 17 to June 29), tweets by white users become more resonant, relative to nonwhite users, amongst all users in our dataset—a complete and dramatic reversal of the trend we see in Period 1.

And this period of white-led discourse resonance has a distinctive set of topics or themes of interest, as distinctive as the patterns that define Period 1's minority-led discourse. This discourse is characterized by its "colorblindness." Race theorists such as Haney López and Bonilla-Silva have identified colorblindness as a post-Civil-Rights-era ideology that mediates the disjuncture between a discourse of racial egalitarianism and a reality of continuing racial inequality. Haney López notes that civil rights movements expelled overt racism from public discourse while their material progress toward racial equality was far more incomplete. The result was a "rejection of White supremacy as rhetoric" alongside the continued de facto dominance of white people.²⁴ The challenge colorblindness meets, Bonilla-Silva argues, is how to explain patent racial inequalities when one holds that color no longer matters.²⁵ Both Haney López and Bonilla-Silva illuminate a colorblind repertoire that has developed to meet this challenge. One strategy is to define racism as discursive. Colorblind ideology defines racism "as any direct invocation or use of race" (WW 159). This cynical conception brands race-conscious policies needed to achieve racial equality as themselves racist. Race and racism act like "magic words" that "spring into being" when spoken "but not otherwise" (WW 160–61). "This magic-word formalism" abstracts racism from material social practices, leaving a purely discursive definition of racism (WW 161). A second move also operates through abstraction: supporting racially progressive goals in general while remaining vague about or undermining specific policies designed to accomplish those goals (RR 142).

Our findings corroborate the idea of a colorblind repertoire. In Table 3, we list the top seven most distinctive topics of white users during Period 2—the period in which the discourse of white users became most resonant in our dataset relative to all users. In contrast to the topics distinctive of minority voices, these topics avoid specific mentions of race, racism, Blackness, and white supremacy. Only one of the most distinctive topics for white voices featured "race" or any racialized group in its top-ten words. This topic (Topic 6) is quite revealing. It mentions race, but this is not race-conscious radicalism replacing colorblindness. Topic 6 focuses on the most tepid form of action imaginable in the face of systemic racial violence: reading books.

Topic six reinforces the discursive definition of racism and reveals a corollary: if racism is conceived as discursive, then the battle against racism can unfold purely on the discursive level. Another distinctive topic (Topic 7) focused on general calls for change but without reference to specific racial groups, thus aligning with the strategy of abstract support while avoiding specifics. Bonilla-Silva argues that abstract liberalism and its universalism is foundational to colorblindness. Colorblindness helps maintain white advantages "without naming those who it subjects and those who it rewards" (RR 54, 4). In this light, one can observe the complete evasion of naming racialized groups, victims, and perpetrators across all topics. These evasions extended to the primary issue at stake and the protests that addressed it. In contrast to minority voices, who made clear the issue of racist police violence and called for arrests and transformations, white voices equivocated about the primary issue. Nowhere in the top words characterizing their most distinctive topics are police killings or racial violence mentioned. While minority distinctive topics showed support for protests, white distinctive topics focused on the disruptions resulting from the protests and on law enforcement responses to them.

Our data thus offer a wide opportunity to examine colorblindness at a moment that more than any other moment in recent US history tested its powers to span the disjuncture between colorblind rhetoric and racialized reality. This moment reveals the scope of colorblindness's powers to reconcile and absorb. For decades, colorblind discourse set a public norm of professing colorblindness and nonracism, a norm that papered over the continuing realities of systemic racism and diffused efforts to address them. White discourse on Twitter during this summer shows a different variation to assimilate and co-opt the power of a minorityity

Table 3.

Top 7 Most Distinctive Topics for White Users in Period 2

discourse: professing antiracism and race-conscious support for BLM to elide the realities of systemic racism against Black people and diffuse transformations to address them. That this variation is oxymoronic—an antiracist, race-conscious, radical colorblind moderation—attests to the remarkable pliability of colorblindness. In Haney López's account, colorblindness is opposed to race-conscious transformations: "Colorblindness … curtail[s] race-conscious efforts to promote racial justice" (WW 162). But our findings suggest a maddening kind of race-conscious colorblindness, a race-conscious stance that in its substance does all the work of colorblindness. Colorblindness flexes its powers of disjuncture, able to mediate not just between acceptance of racist realities and an "I'm not a racist" stance, but also between acceptance of racist realities and an "I'm an antiracist" stance.

What's fascinating are the ways that otherwise earnest white supporters of BLM slip into colorblindness and undermine the movement. To grasp this pattern, we argue that colorblindness cannot be understood simply as a racial ideology as theorists like Bonilla-Silva and Haney López hold. The ideological position of these white users is to articulate support for BLM. To dismiss all white support for Black Lives Matter in 2020 as cynical performance or "virtue signaling" would be too easy and, of the complexity of colorblindness, reductive. The knottier question is how colorblindness delimits the thinking, feeling, and writing of white people who genuinely feel themselves to be committed supporters of BLM. To see how colorblindness did this, we cannot treat it as an ideology alone. It is more fully a culture encompassing tones and styles, underlying feelings and textures, contradictions and messiness.

For example, white distinctive topics reveal textures that focus on civil order. This focus is apparent in one distinctive topic that emphasizes the disruptive effects of the protests on the functioning of cities and streets (Topic 2). A large number of the tweets with the highest probability for this topic were from users identified as white journalists. It's crucial to scrutinize their seemingly objective descriptions of the protests. In one characteristic tweet, a reporter wrote, "appears to be a gaggle of protesters heading west over the middle bridge from the island of Palm Beach. Town authorities sent an alert about middle bridge warning residents about traffic on Okeechobee Boulevard, per @PalmBeachPolice spokesperson."²⁶ On one hand, this user is simply reporting traffic conditions and the statements of police and municipal authorities; on the other hand, tweets like this show an alignment of police, city, and media discourses around the disruptiveness of racial-justice protests to everyday civic life. Many tweets in this topic emphasized looting, vandalism, and property destruction. Doug Sovern, a reporter for KCBS Radio in San Francisco, tweeted this description on May 30: "Lots of vandalism, smashed windows, small fires, up and down Broadway in downtown #Oakland. Worst I've seen from a protest in a while."²⁷ Note the syntactic slippage from descriptions of property damage to protest, a slippage that solidifies a media narrative of the racial-justice protests as destructive looting. What is not considered newsworthy according to such tweets are the goals of the protests. Instead, all that appears is an image of civic disruption and property damage. These tweets reveal an instinctive valuation of the sanctity of property and the normal workings of the civic order as primary goods to which any disruption, regardless of purpose (which need not be mentioned), constitutes "news." This underlying instinct for civic order over the causes of justice that justify civic disruption led many white users to align with and amplify police efforts to restore order, as seen in another distinctive topic focused on law-enforcement responses to protests (Topic 1).

These example tweets come from users who otherwise express support for Black Lives Matter. At another point in the summer, Doug Sovern, the reporter in San Francisco, tweeted a reflection by Doc Rivers, the Black head coach of the Philadelphia 76ers, on the police killing of Jacob Blake, calling Rivers's reflection "powerful."²⁸ Across the white distinctive-topics on civic disruptions and law-enforcement response were white reporters who covered BLM sympathetically in some tweets while emphasizing disorder and destruction in others. We see a duality: support for BLM on the level of articulated position and an underlying instinct for civic order that undermines the movement. For white US journalists, who tend to be in the middle class and, as relatively affluent whites, benefit from the civic order working as usual, it would feel natural to see property destruction and disruption as what is newsworthy about racial-justice protests. Because these journalists are instinctively attached to the white comforts of the current order, their attention gravitates toward actions that disrupt that order—even if the actions belong to a movement that they support. The lived experience of affluent whiteness does not readily equip one to perceive how the everyday workings of the US civic order is violent and destructive to Black lives. It's from this way of being in the world that the protests would appear violent, and disorderly, protests disrupting peace, law, and everyday order, instead of as protests against an everyday order of racial violence that continually disrupts Black lives in the name of peace, law, and order. To understand one crucial way that colorblindness unconsciously undermines racial-justice movements, we need to recognize white comfort in and attachment to the current order, as well as the invisibility of the order's racialized violence from this perspective.

Under white leadership of racial discourse on Twitter and its colorblind textures, the resonance story of the latter part of #BLM summer 2020 (Period 2) is a sobering one of fading energy. The rising intensity of #BLM discourse in early summer falls off dramatically in Period 2. It is the period when white voices are most resonant and Black and minority voices fall away from discursive resonance that the energy of BLM discourse (as indexed by the number of overall #BLM tweets) falls off most steeply (compare Figures 3 and 4). White voices remain dominant as the discourse's overall energy continues its decline over the summer. While the resonance of minority voices is associated in our data with the intensification of the BLM discourse, the resonance of white voices is associated with its dissipation. Given the colorblindness and relative moderation distinctive of white voices—hardly inspiring stuff—it's not difficult to imagine why the energy of the discourse died under their discursive leadership. As the discourse settled into a long fade, the resonance of white voices also began to fade, reaching near zero in late July.

The Insurgent, or How #BLM 2020 Expands Theories of Cultural Change

The story our data and models tell about the #BLM 2020 summer is important in and of itself, but it also offers a useful case study for expanding prevailing cultural studies theories and frameworks for studying cultural change, racial ideology, and hegemony. For example, recent attempts by both scholars and journalists to make sense of the 2020 "racial awakening" often do so with reference to the Trump presidency. President Trump's openly racist statements and policies led commentators to declare that the Trump presidency had opened a new stage in the US racial order in which open racism returned and the post-Civil-Rights-era colorblind consensus crumbled. Ibram X. Kendi wrote, "No president has caused more Americans to stop denying the existence of racism than Donald Trump."²⁹ Scholars pronounced the shattering of Obama-era postracial illusions.³⁰

Scholars of whiteness, such as Bonilla-Silva, however, were not taken in by the Trump era and refute this notion of a "new stage." Bonilla-Silva argues that even when Trump was making his most outrageous statements on race, he still had to go through the motions of denying racism and avowing his commitment to racial equality. Trump's rhetoric shows that he had to navigate a persistent social norm that makes overt racism impermissible. Considering this and the ways that seemingly nonracial US policies continue to reinforce racial inequalities (what he calls "new racism"), Bonilla-Silva concludes that "the racial regime of post-civil rights America is still the 'new racism' and the dominant racial ideology that glues this order is color-blind racism" (RR xiii). Bonilla-Silva was well equipped to perceive the continuation of colorblindness because his analysis recognizes multilayered temporal dynamism. He acknowledges that the dominant racial regime of any era is not totalizing: "The new racism is dominant, but it is not the only way of maintaining racial order. Jim Crow never died one hundred percent, and its ideology has remained important" (RR xiv). Racial regimes change, but "remnants" of the dominant racial regimes of earlier periods remain. He uses the term "resurgent" to describe how forms of overt racism associated with the past can crest again under conditions like the Trump administration without displacing the reality that the dominant racial regime remains colorblindness (RR 19). Colorblind racism can operate as the dominant scheme even while forms of racism never fully eradicated continue to exert force.

To readers in literary and cultural studies, Bonilla-Silva's terms should have a ring of familiarity. They align with one of the discipline's most influential models of cultural change: Williams's scheme of dominant, residual, and emergent. Just as Bonilla-Silva recognizes that colorblind racism is dominant but not comprehensive, Williams insists that the dominant formation in a culture is never totalizing. We also need to recognize the "residual" cultural strains that "formed in the past" but are "still active" (ML 122). Residual cultures, Williams explains, are not ensconced in the past; they are continuing processes and actors in the present. The future is also unfolding in the present. Williams's "emergent" refers to new cultural developments, "new meanings and values, new practices, new relationships" that are continually created, either as new extensions of the dominant or as cultures alternative or oppositional to the dominant (ML 123). Bonilla-Silva, for his part, notes how the dominant racial regime coexists and struggles with "oppositional" racial logics from minority groups (RR 9).

Both Bonilla-Silva's and Williams's models are useful, but they cannot fully explain the pattern of discourse explosion and fade we find in our data. Williams's model is attentive to newly developing cultural forces, especially those associated with the emergence of a new class. But the #BLM radical discourse that led the conversation at the height of summer 2020 was not new.³¹ Its trajectory reveals a gap in Williams's model: the lack of a robust category to describe a long-building minority cultural strain that erupts in a particular moment to become a new dominant by setting up a pitched battle with the established dominant. (Williams does refer briefly to ideas of alternative and oppositional cultures, but these are linked in his model to the new and emergent rather than the long-building [ML 123–24].) Even if temporary, the resonance of Black and minority voices articulating radical race consciousness was pervasive, so it was clearly a momentary dominant. And if we insist on historical contingency, a possibility existed in that moment for this new dominant to endure and shift the terms of the US racial order. The focus of Williams's emergent on newness—"new meanings and values, new practices"—cannot do justice to the long historical struggle of Black and minority communities to move race consciousness and radical transformation into the center of US discourse (ML 123). To do justice to our findings, we propose expanding Williams's model and, by extension, theories of the racial order such as Bonilla-Silva's model that align with Williams's.

We argue for the insurgent, a category of cultural change distinct from the emergent. The connotations of "insurgent" express the revolutionary potential in movements such as BLM to transform the social order. The word captures the sense of embattled struggle that has fueled the history of Black-led uprisings. From the viewpoint of the established dominant, insurgencies are dangerous; they demand contestation as well as efforts to diffuse their energy. The etymology of "insurgent" differs in important ways from that of "emergent." "Emerge" denotes a process of rising from or out of a medium, usually water.³² Williams pictured a chemical process: a precipitate emerging out of solution, a new cultural formation crystallizing out of a period's culture (ML 134). "Insurgent" suggests a different metaphor. The verb "insurge" is the process of the sea surging, rising up, or rushing in. The force and tumult of the image makes it easy to understand its figurative use to describe a revolt.³³ The two images of change and coming into being are very different. Emerging is more harmonious; something new precipitates out of the existing social solution. Insurging is the entire existing social system, the sea itself, thrown into upheaval. Unlike the emergent, which may or may not contest the dominant, the insurgent necessarily threatens to engulf and overthrow the dominant. It is clear that the transformative force of the BLM movement in the summer of 2020 cannot be described as emergent. It was an insurgent.

With this new distinction, we can reserve the emergent to describe new cultural formations that precipitate out of the existing dominant and use the insurgent to describe processes in which the entire social order is thrown into upheaval. Insurgent addresses the gap in Williams's model: what to call a long-building minority strain that erupts to contest the dominant. Insurgencies can feel new to ruling groups, but they are often long brewing, spurred by crisis to rise up. To call an insurgency simply emergent is to suggest that the uprising is a new, isolatable element in a society: a precipitate is separable from its solution. The metaphor of the insurging sea, however, insists that its conditions were already and always a part of the existing social order.

The insurgent threatens to overthrow the existing dominant, but, as a contingent process of historical change, its outcome is never assured. We see three possible outcomes:

1. 1. A powerful insurgent under the right conditions achieves an enduring overthrow of the existing dominant to become a new dominant.

2. 2. A less powerful insurgent, or an insurgent under difficult conditions, recedes, leaving the existing dominant intact and returning the social order to its unstable equilibrium.

3. 3. The energy of an intense insurgent breaks through and overwhelms the dominant for a period. This energy tests and stretches the integrity of the existing dominant and is ultimately transferred to and absorbed by the dominant. The process of stress and absorption leaves the dominant in its central position but exposes its parts, reveals its cracks, and deforms its elements.³⁴

While more evidence and analysis are required to demonstrate this point, we believe the third outcome is what happened in summer 2020. We speculate that the energy of an insurgent BLM discourse, which temporarily overwhelmed the colorblind dominant, was absorbed by white voices invested in that existing dominant. Such absorption comes with stresses, which raises a key question: how does a dominant absorb an insurgent with so much energy?

Colorblindness Is a Structure of Feeling

Our findings suggest that a dominant discourse capable of absorbing such a powerful insurgent must be understood as far more expansive than simply a dominant racial ideology. In pushing us to this conclusion, our findings revise theories of colorblindness and broader concepts of racial ideology and hegemony central to US race theory (as we describe earlier in our overview and discussion of Haney López and Bonilla-Silva).

The textures revealed in our topics data show that the articulated positions of white users—their racial ideology—are far from the full story. If colorblindness mediates between a white rhetoric supportive of racial equality and the reality of racial inequalities that don't budge because race-conscious efforts are diffused by whites, race theory needs a thick account of the mechanisms that allow colorblindness to span this disjuncture, especially when it yawns as large as it did in the summer of 2020. It's not a simple thing for a group to believe earnestly that they are supporters of antiracist movements while simultaneously undermining the movement's efforts. A lot of cultural and affective work is necessary to encompass this duality.

Colorblindness, we argue, is more capacious than a racial ideology. It is a racial structure of feeling. The concept of "structures of feeling" comes from Williams, who believed that social rule operates through consciousness as lived and practiced (ML 128–35). Ideology is a limited idea because it reduces consciousness to "explicit and finished forms" of thought (ML 128). He proposed the term structure of feeling to name the workings of consciousness, culture, and rule that exceed "formally held and systematic beliefs" (ML 132). It opens a way to analyze "the relatively mixed, confused, incomplete, or inarticulate consciousness" of actual people in a specific period and society (ML 109). On one hand, a structure of feeling "operates in the most delicate and least tangible parts of our activity"; on the other hand, it is "as firm and definite as 'structure' suggests," a pervasive culture forming in specific patterns (LR 69). This approach means widening the evidence analyzed beyond explicit beliefs to include forms and "elements of impulse, restraint, and tone" that unconsciously express aspects of a period's lived experience (ML 132).

The concept is powerfully suited for describing colorblindness because a structure of feeling encompasses the mixed feelings generated in the disjuncture between explicit ideals and lived experience—between, say, colorblind beliefs and a deeply racialized society. It includes the ways that people fumble and grasp to reconcile strongly held values with social realities. We can sense complex ambivalences beneath the ideals, feelings that reveal a grappling, "if only unconsciously, with a practical world in which things [are] not so simple" (LR 85, 87, 88). In this attempted reconciling, magical thinking—forms of fantasy driven by a desire to resolve ideals with reality—emerges. The strength of this desire indexes just how much a social group unconsciously feels the yawning gap between the two. Magical thinking is an inarticulate, felt awareness of how society doesn't live up to one's ideals. Such mixed feelings are why the concept of structure of feeling speaks so powerfully to colorblindness. Looking at our data through this lens, we can interpret the duality of support for BLM and repulsion to the civic disruptions it generates as a form of white magical thinking: an unarticulated desire for a movement that would somehow achieve transformative progress on racial justice but with minimal disruptions to the everyday order that white people are comfortable with. The attachments to order, the comforts of colorblindness—these clue us into a structure of feeling centered around the desire for a world exempted from racial conflicts and disruptions. It's because colorblindness is not just an ideology but a structure of feeling, one with multiple conduits for channeling the strong feelings and desires generated by grappling with disjuncture, that it was able to absorb and diffuse a minority insurgent culture.

The textures in our topics data occasion a rethinking not only of colorblindness but also of ideological approaches more broadly in race theory. These approaches would benefit from recognizing racial structures of feeling that include the unconscious, inarticulate layers beneath racial ideologies. This vocabulary would be more effective, for instance, at making sense of the rich interview data that powers Bonilla-Silva's theory of colorblindness. His interviews with white subjects show contradictions, ambivalences, tones, and underlying feelings as central to colorblind discourse, but his vocabulary of racial ideology does not capture these textures. Bonilla-Silva and other scholars in US race studies have been influenced by concepts of racial ideology and hegemony laid out by Omi and Winant's field-shaping theory of racial formation. This influence helps explain why Bonilla-Silva foregrounds "ideological positions" and why Haney López thinks of colorblindness as a set of coherent arguments "adopted" as a "rhetorical weapon" by groups seeking to stymie race-conscious efforts at greater equality (RR 9; WW 158). While such intentional uses of colorblindness are important to combat, our findings point to the unintended, unconscious, and harder-to-combat ways that colorblindness functions as a groundwork shaping the feelings, desires, and ways of being of many white people sympathetic to racial-justice movements.

Omi and Winant's model of racial ideology is limited because it reduces the textures of a dominant racial culture to articulated views. Interestingly, they draw on the same intellectual genealogy that feeds Williams's structure of feeling: Antonio Gramsci's theory of hegemony. But in a revealing substitution, they call this an ideology: "Gramsci's treatment of hegemony … argued that … ruling groups must elaborate and maintain a popular system of ideas and practices—through education, the media, religion, folk wisdom, etc.—which he called 'common sense.' It is through its production and its adherence to this 'common sense,' this ideology (in the broadest sense of the term), that a society gives its consent to the way in which it is ruled."³⁵ They acknowledge Gramsci's notion of hegemony as something that pervades everyday life. But by making the term interchangeable with ideology, they allow a slippage that occurs throughout their theory: that hegemony is reducible to articulated worldviews and beliefs. For instance, in their account of the evolution of "modern racial awareness," they focus on prominent thinkers such as David Hume, Immanuel Kant, and Thomas Jefferson and their explicit views of race.³⁶ This focus on ideology orients their theory around the contestation between "dominant racial ideology" and the "oppositional ideology" of minority social movements; here, racial projects attempt the "rearticulation of pre-existing racial ideology," which suggests that the battle for racial hegemony takes place on the field of articulated ideas.³⁷

We are drawn instead to Williams's understanding of hegemony, which insists on attention to a more complete sense of lived experience. Williams lauds Gramsci's concept for refusing to reduce actual consciousness to abstract ideology and formal systems of belief. The consciousness that builds consent to social rule saturates the whole lived experience. "Shaping perceptions of ourselves and our world," hegemony is "a lived system of meanings and values" that "constitutes a sense of reality for most people in the society" (ML 110). It is "in the strongest sense a 'culture'" (ML 110). From what we've seen in the textures of colorblindness, we're convinced that race studies would benefit from expanding Omi and Winant's theory of racial ideology and hegemony into a Williams-informed model of racial structures of feeling. In particular, when revised with a notion of the insurgent, Williams's approach can offer a broader vision of the dynamics of change and modes of struggle central to the racial order. The struggle is broader and deeper than racial ideology; it involves transforming a dominant racial culture, a racial structure of feeling that grounds the sense of lived reality for many in the ruling racial group.

Conclusion: Caveats and Future Directions

With these findings and expanded theoretical vocabulary, we can offer a new summation of the trajectory of #BLM in 2020. As the emotion and energy unleashed by George Floyd's killing flared into a social movement, Black and minority voices became most resonant and led the movement in radical and race-conscious directions. This insurgent culture helped move protest against white supremacy into the centers of US online discourse, challenging a dominant culture of colorblindness. If by "racial awakening" we mean the awakening of many Americans to the salience of race and racism within American life, then the "racial awakening" was real, at least in terms of online discourse on Twitter. But it lasted only three weeks. After three weeks, white voices regained control of Twitter #BLM discourse. The textures of their discourse revealed the absorptive powers of colorblindness as a racial structure of feeling that could take in and rechannel the emotions, desires, and conflicts unveiled by the minority-led insurgent challenge. With this challenge diffused, the momentary possibility of a transformed racial culture in online discourse faded, and the discourse resumed the dominant strain of white racial thinking in the post-Civil-Rights era: colorblindness.

Our study joins a growing body of research that studies the evolution of public opinion and discourse around #BLM during and after the summer of 2020. It aligns with an important new political science study, based on survey data, that demonstrates that perceived discrimination against Black Americans among white respondents spiked after May 25 but that this increase was relatively short lived, quickly regressing to pre-May 25 levels within weeks.³⁸ Our study enriches these findings by demonstrating how these effects play out in terms of online public discourse on an important social media platform—the ephemerality of radical-race consciousness in the weeks following George Floyd's death. How people talk about racial justice and their reported opinions about it follow similar trends during this period.

This study has several limitations. The most obvious is our single variable focus in classifying users: race. As already mentioned, the majority of recent studies of #BLM tend to include other identity variables, such as gender and political affiliation. Studying the interaction between race and gender with dynamics of discourse resonance is an obvious next step. Further, including political affiliation would help distinguish between tweets that are for or against #BLM, a distinction that our analysis necessarily glosses over. However, mobilizing that data would not entirely address this problem: conservatives sometimes tweet in favor of #BLM and vice versa. Further, the ideological disposition of a tweet is often ambiguous; a tweet's support for BLM could be partial or ambivalent, or a tweet can express both support and criticism. In any case, finding a way to quantify this distinction would be a useful next task for this analysis. And overall, future work in this vein will require a more multidimensional account of user identity.

Further, a premise of this paper, supported by existing scholarship, is that #BLM during the summer of 2020 was substantively different from earlier moments of BLM on social media platforms, particularly in terms of racial demographics.³⁹ The majority of BLM participants on Twitter in 2013 and 2016 were Black and minority; in the 2020 moment, the majority of Twitter users and participants were white.⁴⁰ It would be productive to understand how dynamics of discourse resonance played out in those earlier periods compared to the period we study in this article. A longer diachronic analysis of the ebbs and flows of white versus nonwhite resonance would likely prove highly informative.

Finally, in terms of methodology, we hope to have shown the affordances of using large datasets and computational methods to study race, discourse, and social change. Historically, critical race scholars have been skeptical of quantitative methods, particularly the way they often reify racial categories. We share these concerns and are aware of their potential pitfalls. But what our research has demonstrated, we believe, is that contrary to common assumptions that "big data" methods erase the particularity or nuance of individual language and experience, data methods can be powerful allies in revealing the granular textures, intricacies and messiness of racial consciousness, even at the scale of large groups of users. As this study's theoretical conclusions show, bringing data methods focused on moments of cultural contestation and mass communication into close dialogue with field-specific theory and qualitative analyses can expand our models of how race, discourse, and culture operate.