It was a first for an inaugural address: the president naming “white supremacy” as a force driving the politics of the country, and that its citizens “must confront and will defeat.” Perhaps because of Joe Biden’s insistence, throughout his campaign, that “words matter,” the mention seemed momentous: a national figurehead speaking aloud, and therefore recognizing, the racist ideology that is largely responsible for the centuries of prejudice, discrimination, and outright violence faced by nonwhite people—and Black people in particular—in the land now called the United States.
Words do matter, precisely because they carry with them specific rhetorical significance, and because they index larger concepts and debates. But the work of elevating these words to the level of national discourse, such that they can be uttered—and, crucially, understood—by a sitting president, is performed not by individuals but by collectives and coalitions and takes place not in any particular public statement but, instead, over the course of many years.
Can this process be tracked with any degree of precision? And if so, what can be learned about the path by which certain ideas move from the margins to the mainstream?
Our team is currently compiling a dataset that can help to trace the emergence of certain words and phrases, like white supremacy, that have become central to contemporary conversations about racial justice.1 And as we consider the range of sources that might document these efforts, and the range of voices that our dataset must contain, we have turned to a racial-justice movement from a previous era—the abolitionist movement of the 19th-century United States—to begin to formulate a computational model of how new words enter into conversation, how existing words are transformed, and which groups are responsible for those new or changed meanings.
In many ways a direct antecedent of today’s racial-justice movement, the 19th-century US abolitionist movement was born of a fundamental injustice: the persistence of slavery in a country defined by its democratic ideals. And while the movement coalesced around the common goal of ending slavery, its constituents—who were diverse in terms of race and gender, as well as political sensibility—often disagreed about the means by which that end should be accomplished. They also often disagreed about what the meanings of words like freedom and justice truly entailed.
These debates played out in person, at organizing meetings known as the Colored Conventions.2 They also played out in the press, through the hundreds of (mostly) weekly newspapers that were established with the aim—in some cases quite explicitly—of promoting the abolitionist cause. These newspapers were compiled by both Black and white editors, and aimed at both Black and white readers. The majority of the editors were men, but some were women. And while most of these newspapers have been lost to time, several prominent titles (and a scattered sampling of others) have been preserved. In recent years, these newspapers have begun to be digitized, opening them up to a new reading public—and to computational research.
To begin our research into the changing language of the abolitionist movement, we were required to assemble a dataset, or corpus, of these digitized newspapers.3 Figure 1 shows the 90,000 unique articles (from the original 224,160 articles) that we selected for analysis, as distributed by newspaper title and over time.4
Our corpus includes some titles, such as The Liberator and the North Star, that were explicit in their aim of advancing the abolitionist cause. It also contains additional titles associated with the Black press, such as the Colored American and the Provincial Freeman, that placed the issue of abolition in the context of broader discussions about Black life. We decided to include a women’s suffrage newspaper, The Lily, so as to probe possible word changes in relation to the suffrage movement. (We would have incorporated more women’s suffrage titles had they been available to us.) We also included several less overtly political titles, such as the National Era, a more general newspaper; Godey’s Lady’s Book, a literary magazine aimed at white women; and the Christian Recorder, the newspaper of the African Methodist Episcopal (AME) Church. Each title functions as a reference point, allowing us to place our core set of abolitionist newspapers in the context of a wider, overlapping set of cultural and political conversations.
How, then, to identify the conceptual keywords like freedom or justice that animated our corpus? And how to determine if their meanings had changed?
To probe these questions, we drew inspiration from the linguistic theory of distributional semantics, which holds that the meaning of any particular linguistic element—in our case, a word—can be inferred by the context in which it appears.5 Building on computational implementations of this theory, known as word embedding models, we developed our own model that could track changes in the meanings of the words in our corpus.6 We also developed a metric that could determine which newspaper was responsible for introducing each changed meaning, and which newspaper was the next to pick it up. Together, the model and the metric allowed us to explore how certain political concepts evolve as they move from the margin to the mainstream.7
We found that the word freedom enters the corpus with associations to words like humanity, people, and country—words related to the generalized nature of the nation’s foundational ideals. By 1861, however, after the abolitionist movement had largely coalesced, the associations of the term become more specific, including direct references to liberty and to the rights that ensure it, as well as to the institutions that enforce their application (figure 2).8
Our model determined that it was The Liberator, one of the most prominent abolitionist newspapers, that introduced this change in the context—and thus the meaning—of the word freedom. Our model also determined that the National Anti-Slavery Standard, the official newspaper of the American Anti-Slavery Society, was quick to follow the newer, more targeted meaning of the term.
The term justice follows a reverse trajectory, from specific to abstract, as it becomes increasingly connected with the abolitionist cause. It enters the corpus with narrow associations to the legal system, as evinced by words that initially surround it, including judges, trial, offence, and crime. But by the midpoint of the corpus, the term has significantly expanded, commanding a more ideological frame. Its associations in 1847 include the terms liberty, equality, oppression, and rights, which track qualitative arguments about how justice transformed into a much more capacious sociopolitical concept over the course of the 19th century (see figure 3).9
In our corpus, this transformation is reinforced by the appearance of the terms universal, humanity, citizenship, and nation, which may point to the success of early advocates for criminal-justice reform in expanding not only legal protections, but also ideas about what justice properly entailed. It also may speak to the growing realization, on behalf of abolitionists, of how issues of incarceration intersected with those of enslavement.10
It was The Lily, the women’s suffrage newspaper, that introduced this changed meaning. And it was The Liberator that was quick to adopt it, showing how the ideas advanced by these two movements often, if uneasily, intertwined.
But one of our main findings was unexpected. Despite significant shifts in words like freedom and justice, the majority of words that changed in meaning (with a high degree of statistical significance) were not those that carried any particular political or conceptual charge. Instead, our model detected a high degree of semantic change in words like growing and writing, hoped and courage.
These words and others, which denote actions and emotions, were associated with the strongest statistical signals in the entire corpus. This suggested to us that the form of language change best captured by our model might be better understood at the level of discourse—a larger system of meaning-making—rather than at the level of individual words. What we were witnessing, it seemed, was a shift—at a time of great social and political change—in how people talked about the emotions and experiences that brought them to coalesce around the abolitionist cause.
what we can affirm is how the discourse of any social movement is shaped by a range of sources, not all of them fully acknowledged or allied in their efforts, or even centered on a common cause.
To get a better sense of this changing discourse and the role of each newspaper in shaping it, we decided to take an aggregate view. For each word, we noted which newspaper was responsible for introducing its changed meaning, and which paper then followed that change. We then counted up the number of leader-follower pairs for each newspaper and visualized them as a flow diagram, pictured in figure 4. The newspapers involved in the highest number of leadership events, as we termed them, confirmed some well-known findings.
For example, the relationship between The Liberator and the National Anti-Slavery Standard rises to the top of the list, with 32 words (of 435 total) on which The Liberator leads and the Standard follows. Historical circumstance suggests these two titles should indeed be in close conversation with each other, since the latter was founded with the explicit aim of appealing to a more moderate audience than the more radical Liberator. It makes sense that the Standard would follow the lead of others rather than actively work to shape the movement’s discourse.11
Using this form of analysis, we can also isolate certain subsets of newspapers: for example, the titles associated with the Black press, as pictured in figure 5. Here we can see how publications like the North Star, which Frederick Douglass edited, influenced all the others.
More specifically, we see Douglass’s influences in both the Standard, which was aimed at a national audience of both Black and white readers, and the Christian Recorder, one of the major publications of the AME Church. Its religious focus would seem to hold it far from the political fray. And yet here, too, is evidence of the paper adopting certain linguistic elements from the discussion surrounding abolition, even if it rarely treated the subject head-on.12
Another interesting finding relates to the prominence of the Provincial Freeman, a Canadian newspaper edited by Mary Ann Shadd. According to our semantic analysis, its influence is nearly as far-reaching as Douglass’s three titles. And yet, in its own time, the Freeman struggled for both recognition and readership.
In her farewell editorial, Shadd was explicit about how, when the Freeman had initially failed to “prosper,” she had “traveled to arouse a sentiment in favor of it,” while at the same time encountering “obstacles” that “few, if any [other] females had had to contend against.”13
The influence of both Black and white women editors on the abolitionist discourse is further confirmed in a network analysis of the leader-follower pairs. To perform this analysis, we employed an algorithm known as hubs and authorities (HITS), which was developed as a means of ranking content on the web.14 The underlying premise is that certain web pages, known as hubs, serve as aggregators of other content, while others, known as authorities, produce the content that hubs then aggregate. We mapped this concept onto our leader-follower pairs and then plotted the hub and authority scores against each other. The result is an overall picture (figure 6) of the position of each newspaper within the discourse network.15
Once again, The Liberator stands out with the largest authority score, while the Standard possesses the largest hub score. This further confirms what scholars know to be true about the role that each played in this newspaper network.
But below those two titles, several interesting clusters emerge. Among them are two of the titles edited by women, the Provincial Freeman and The Lily. Their position at the top left of the chart indicates that they innovated new word meanings in far greater proportion to the meanings they adopted from others.
What is the significance of these two papers, one edited by a Black woman and the other by a white woman, emerging as the semantic leaders of the group? And what does it mean for our understanding of the often vexed relationship between the social movement—abolition and women’s suffrage, respectively—that each represents?
The historian Manisha Sinha has suggested that “women were abolition’s most effective footsoldiers,” and that “black and white women” had, at times, “marched arm in arm” in response to anti-Black violence.16 And yet then, as now, this alliance was not without significant flaws. In The Trouble with White Women, Kyla Schuller explains how, when it came time for white women abolitionists “to expand their horizons beyond the existence of slavery and recognize the pervasive violence of antiblackness, many instead remained invested in racism.” For certain key figures, including known contributors to The Lily, such as Elizabeth Cady Stanton, this racism was unabashed and overt: “White supremacy,” Schuller explains, “became her choice strategy for advancing women’s suffrage.”17 Anti-Black racism, then, subtended much of the women’s suffrage movement—and areas of the abolitionist movement as well. As a nation, we have only begun to account for this racism and its life-altering consequences.
Clearly, words do not always reflect the actual beliefs of those who speak them. And yet what we can affirm is how the discourse of any social movement is shaped by a range of sources, not all of them fully acknowledged or allied in their efforts, or even centered on a common cause.
The nation, as Biden explained, continues to grapple with the impact of white supremacy. And as we undertake this necessary work, our hope is that this computational view of the discourse of the abolitionist movement of the 19th-century United States can lend additional credence to the complexity of the coalitions, past and present, that have helped to work toward justice. More important, we hope this view will prompt those of us who seek to contribute to the current racial-justice movement—especially as allies—to hold ourselves accountable, both individually and collectively. Together, we can push more capacious ideas about freedom and justice to the front of the political stage.
The project on which this article is based was conducted in collaboration with Jacob Eisenstein. See Sandeep Soni, Lauren F. Klein, and Jacob Eisenstein, “Abolitionist Networks: Modeling Language Change in Nineteenth-Century Activist Newspapers,” Journal of Cultural Analytics (2021).
This article was commissioned by Richard Jean So.
- On the evolution of the phrase white supremacy in particular, see, for example, Michael Powell, “‘White Supremacy’ Once Meant David Duke and the Klan. Now It Refers to Much More,” New York Times, October 17, 2020. ↩
- To learn more about these in-person organizing events led by and for Black Americans, see the Colored Conventions Project, as well as The Colored Conventions Movement: Black Organizing in the Nineteenth Century, edited by. P. Gabrielle Foreman, Jim Casey, and Sarah Lynn Patterson (University of North Carolina Press, 2021). ↩
- The source for our dataset was Accessible Archives, which has transcribed the contents of a curated set of titles and made them available on the web. Through a paid licensing agreement, we were able to obtain the text of these newspapers—224,160 articles, drawn from 14 titles, totaling just over a million unique words. For the purposes of this project, we further refined our corpus so as to remove all articles that might contain reprinted content, as we wanted to avoid possible false signals about word adoption. We also eliminated titles that published fewer than 500 articles over the course of their run. Finally, we removed all articles published after 1865, so as to focus more closely on antislavery activism before the end of the Civil War. This date, of course, is one of many possible break points. The 13th Amendment, which abolished the legal condition of slavery, was passed on January 31, 1865, and was ratified after the final state, Georgia, approved the amendment on December 6 of that year; we could have chosen to truncate our corpus on that exact date. Another alternative would have been to end it after the ratification of the 14th Amendment, which granted citizenship and equal civil rights to Black Americans; it was not passed by the Senate until 1866 and took two more years to ratify. But because our corpus experienced a significant drop-off in volume after 1865, we opted for an earlier break point. We thought that the end of the calendar year would allow for some discussion following the ratification of the 13th Amendment and also make a year-by-year accounting of word changes easier to track. ↩
- In two instances, we created sets of related titles in order to maximize the overall number of articles that we could include in our analysis. “The Colored American” includes articles from both the Weekly Advocate and the Colored American, since the former changed its name to the latter two months into its publication run. “Douglass Papers” includes three titles edited by Frederick Douglass: the North Star, Frederick Douglass’ Paper, and Douglass’ Monthly. While there were other editors involved in each of these papers, both named and unnamed, we grouped them together so as to be able to compare Douglass’s publications to others across a longer span of time. ↩
- The theory of distributional semantics has been around since the 1950s, and linguists used to perform this type of statistical analysis by hand. See John R. Firth, “A Synopsis of Linguistic Theory, 1930–1955,” Studies in Linguistic Analysis (1957). ↩
- For an introduction to word embeddings written for a humanities audience, see Laura Johnson, with Julia Flanders and Sarah Connell, “Introduction to Word Embedding Models,” in Women Writers Vector Toolkit (2019). For an overview of computational applications of word embeddings, particularly as they have been used to track change over time, see Andrey Kutuzov et al., “Diachronic Word Embeddings and Semantic Shifts: A Survey,” in Proceedings of the 27th International Conference on Computational Linguistics (2018). ↩
- For more on our methods, see Sandeep Soni, Lauren F. Klein, and Jacob Eisenstein, “Abolitionist Networks: Modeling Language Change in Nineteenth-Century Activist Newspapers,” Journal of Cultural Analytics (2021). Prior work with Kristina Lerman also contributed to the development of the metric. See Sandeep Soni, Kristina Lerman, and Jacob Eisenstein, “Follow the Leader: Documents on the Leading Edge of Semantic Change Get More Citations,” Journal of the Association of Information Science and Technology, vol. 74, no. 4 (2021). ↩
- To create figure 2, we took a vector representation—or embedding—of the word freedom, along with the embeddings of the words that surround it, and reduced them to two dimensions using singular value decomposition (SVD). The changing position of the embedding and its changing proximity to those of related words help to show how the meaning of freedom shifted from more abstract to more concrete. ↩
- As in figure 2, we created figure 3 by reducing the embeddings of the word justice and those that surrounded it into two dimensions so that they could be plotted on an x-y plane. Here, the changing position of justice and its changing proximity to other words show how its meaning shifted from specific to abstract. ↩
- For a discussion of how abolitionists, including Frederick Douglass, began to identify connections between criminal justice (or the lack thereof) and slavery, see Angela Y. Davis, “Slavery, Civil Rights, and Abolitionist Perspectives toward Prison,” in Are Prisons Obsolete? (Seven Stories, 2003). ↩
- For more on the origins of the National Anti-Slavery Standard and its early years, as well as its relationship to other newspapers in this corpus, see Lauren F. Klein, “Dimensions of Scale: Invisible Labor, Editorial Work, and the Future of Quantitative Literary Studies,” PMLA, vol. 135, no. 1 (2020). ↩
- For more on the significance of the publications of the AME Church, see Francis Smith Foster, “A Narrative of the Interesting Origins and (Somewhat) Surprising Developments of African-American Print Culture,” American Literary History, vol. 17, no. 4 (2005). ↩
- Mary Ann Shadd, “Adieu,” Provincial Freeman, August 22, 1855. Jane Rhodes, Mary Ann Shadd Cary: The Black Press and Protest in the Nineteenth Century (Indiana University Press, 1998), offers an in-depth biography. Jim Casey, “Parsing the Special Characters of African American Print Culture: Mary Ann Shadd Cary and the * Limits of Search,” in Against a Sharp White Background: Infrastructures of African American Print, edited by Brigitte Fielder and Jonathan Senchyne (University of Wisconsin Press, 2019), explores Shadd’s authorship and editorial practices in detail. ↩
- The HITS algorithm is documented in Jon M. Kleinberg, “Authoritative Sources in a Hyperlinked Environment,” Journal of the ACM (JACM), vol. 46, no. 5 (1999). ↩
- For information on Pagerank, mentioned in figure 6, see Larry Page et al., “The PageRank Citation Ranking: Bringing Order to the Web,” Stanford InfoLab (1999). ↩
- Manisha Sinha, The Slave’s Cause: A History of Abolition (Yale University Press, 2016), p. 266. ↩
- Kyla Schuller, The Trouble with White Women: A Counterhistory of Feminism (Bold Type, 2021), p. 18. ↩