During the Reconstruction era through the twentieth century, many jurisdictions in the United States drove out their Black communities. When this happened, these jurisdictions came to fit Loewen’s (2018) definition of sundown towns; that is, any “jurisdiction that for decades kept African Americans or other groups from living in it and was thus ‘all-white’ on purpose” (p. 4). Yet the means through which these places forced out marginalized groups varied. Sometimes, sundown towns were established through legal systems, and in this sense, their borders were coterminous with their given jurisdictions. More often, however, sundown towns were (and continue to be) demarcated through threats of racial violence. In this broader sense, sundown towns are harder to map. They are places defined not solely by physical space, but rather by the racist systems of oppression that overlay their geospatial dimensions. They are, in other words, phenomenological and epistemological places, formed through lived experience and passed down by culture. It is subsequently the means of enacting or threatening racial violence that demarcate sundown towns. Their borders are the points at which non-white people feel threatened based on their race and proximity to dangerous white communities.

In this paper, I argue that the phenomenological and epistemological dimensions of sundown towns require innovative methodological approaches in order to delineate and study them. To address this challenge, I present a pipeline for identifying sundown towns via historical records of racial violence and digitized newspaper archives. This pipeline relies on several data sources: 1) the Tolnay-Beck-Bailey Inventory and the Seguin-Rigby Dataset, two authoritative lynching records, 2) the Dataset of U.S. Lynching Reports (DUSLR), and 3) the Chronicling America archive of digitized newspapers. These sources are cross-referenced to isolate places where known lynchings occurred and where we have contemporaneous newspapers that have been digitized. These places are taken as the ground truth for the rest of the pipeline. They are considered sundown towns because they are places where racially motivated lynchings occurred. They are also critical for the project because of their available contemporaneous newspaper coverage, offering the possibility of identifying local reporting of racially motivated events.

With these places isolated, I then text-mined their contemporaneous newspapers for known lynching victim names. After hand-keying the extracted articles, I identified 300 reports describing local events of racial violence. These hand-keyed reports were then paired with a random sample (also hand-keyed) of newspaper text from Chronicling America, creating a training set used to fine-tune BERT to classify newspaper reports as containing local threats of racial violence or not. Once fine-tuned, I then deployed BERT on a series of local newspapers, finding that it was effective at identifying local coverage of racial violence. Finally, I used these steps to map newspapers with strong signals of racial violence, thereby identifying other potential sundown towns geospatially and temporally. Taken together, these methods show how integrating historical lynching records with computational newspaper analysis can systematically delineate the otherwise elusive boundaries of sundown towns. Doing so can also help to identify examples of sundown towns even when threats of racial violence–as expressed through newspapers–did not necessarily culminate in atrocities or expulsions of communities, but were nonetheless threatening and pervasive.

This project builds upon other data-driven work on sundown towns. Scholars at Tougaloo College, for example, have maintained Loewen’s wide-ranging project of identifying sundown towns using census records. Their system involves checking census data for precipitous drops in Black populations then reviewing local documents for more information. The locational data compiled through those initiatives has also been reapplied computationally by Rigby et al. (2025), making it more easily accessible via digital maps (p. 1-6). I do not argue against these methods of demarcating sundown towns. They are perfectly viable ways to identify the geospatial dimensions of these places. In my paper, however, I argue that sundown towns should be understood as more highly dimensional–defined by complex social relations expressed through various means, such as newspaper publications or racial atrocities, as much as by any jurisdictional boundaries. And while my paper maps sundown towns based on its analyses of newspaper content, it does not argue that these maps showcase the complete boundaries of sundown towns. Rather, they loosely represent the existence of sundown towns in the sense of place expressed by its local newspapers.

This theoretical framework is informed by scholarship in human geography. Massey (1994), for example, notes how places are the result of complex social relations, and understood this way, they need not be conceptualized with geospatial boundaries (p. 155). This is particularly important in the case of sundown towns because so much of their impact extends beyond physical space. For Black communities, sundown towns are embodied and evoked through an awareness of the dangers of racial violence within a given area, shaping part of what Lipsitz (2011) describes as the Black spatial imaginary: a “socially shared understanding of the importance of public space” among Black people in America (p. 52). In the case of sundown towns, these segregated places emphasize how place can eliminate opportunities or destroy the most basic sense of safety. They result in changed behavior, diminished freedoms, and in the worst cases, horrific events of racial violence.

That is why this work seeks to innovate on the study of sundown towns. Its theoretically informed expansion of the definition, coupled with its innovative computational methods, simultaneously opens the topic for closer analysis and demonstrates how to increase those analyses to a national scale. These approaches, moreover, also showcase how to overcome one challenge of the digital humanities: to “take up [humanistic] theoretical principles and engage them in the production of methods” (Drucker, 2012, para. 27). The project addresses this challenge by showing how theories of place from human geography and Black studies can meaningfully guide computational analysis. By shaping my pipeline on the basis of identifying and analyzing sundown towns via digitized newspaper records, this paper illustrates how computational methods can be designed to surface the lived, relational, and racialized dimensions of place that traditional geospatial approaches often obscure.