The Paradigm Revolution of Digital Humanities: From Computational Humanities to Knowledge Reinvention

The 1949 encounter between Roberto Busa, a Jesuit priest and medievalist specializing in the works of Thomas Aquinas, and Thomas J. Watson Sr., then president of IBM, is widely recognized as the foundational moment of computational humanities. Busa’s audacious proposal to encode the entire 10.5 million-word corpus of Aquinas’s writings, plus all works referencing Aquinas, into a searchable index was dismissed as impossible by most of his peers, who viewed the printing press as the apex of textual processing technology. Watson’s famous retort—that such a project would be feasible only if Busa could guarantee that IBM’s machines would be used "for the benefit of humanity"—set the stage for a 30-year collaboration that produced the *Index Thomisticus*, a 56-volume printed index completed in 1980, accompanied by a pioneering digital database that laid the technical and conceptual groundwork for all subsequent digital humanities (DH) work.¹ For the first four decades of its existence, this emerging field operated under the label of computational humanities (CH), defined by a shared assumption that computers were auxiliary tools designed to accelerate existing humanistic research workflows: concordance generation, word frequency counting, and text searching, all tasks that human scholars could complete manually but were prohibitively time-consuming at scale. The 1990s, however, marked a fundamental paradigm shift: the rise of the World Wide Web, the standardization of the Text Encoding Initiative (TEI) in 1987, the emergence of open digital library infrastructure, and the increasing availability of high-performance computing (HPC) collectively reconstituted the role of technology in humanities research, transforming it from a passive tool into a constitutive element that reshapes research questions, methodologies, and the very logic of knowledge production. This shift gave rise to the DH paradigm, which this chapter situates as a Kuhnian paradigm revolution for the humanities: a disruption of long-held shared assumptions about what counts as valid research, how research is conducted, and what constitutes legitimate knowledge in humanistic inquiry.

To frame this revolution, we first distinguish the CH and DH paradigms across four core dimensions of research practice, as outlined in Table 1.1. The critical boundary between the two lies in the role of technology: in CH, technology is an external tool applied to pre-existing research questions and data; in DH, technology is a co-constitutive element of the research process, enabling new questions, new data forms, and new modes of knowledge production that were impossible in a pre-digital context.

| Dimension | Computational Humanities (1949–1990) | Digital Humanities (1990–Present) | |-----------|-------------------------------------|-----------------------------------| | Core role of technology | Auxiliary tool to accelerate existing, pre-digital research workflows | Constitutive element that enables new research questions, data forms, and methodological frameworks | | Research question scope | Constrained by pre-digital archival access and processing limits, focused on incremental extensions of existing humanistic inquiry | Expanded to scale, pattern, and network questions that are impossible to address via manual analysis alone | | Knowledge production model | Individual scholarship, disseminated via monographs and closed journal articles, with limited reproducibility | Collaborative, interdisciplinary, open, and reproducible, with knowledge products including datasets, tools, and infrastructural resources as well as traditional publications | | Methodological integration | Superficial: computational tools are applied to humanistic data without fundamental changes to existing humanistic methods | Deep: computational and hermeneutic methods are co-constitutive, with each informing and refining the other |

Research Question Reconstitution: From Archive-Bound to Scale-Aware Inquiry

Traditional humanities research has long been constrained by what we term the "archival bottleneck": the physical and temporal limits of accessing, processing, and analyzing source materials. A 19th-century literary scholar, for example, could realistically only study a small subset of the 10,000+ novels published in Britain between 1800 and 1900, as close reading of even a single novel could take weeks of sustained work. Similarly, a cultural historian studying the transmission of Enlightenment ideas could only track correspondence between a small number of well-documented thinkers, as the 100,000+ surviving letters from the period were scattered across 50+ archives across Europe, with no centralized index of their contents. DH eliminates these bottlenecks by enabling large-scale digitization, processing, and analysis of source materials, opening up an entirely new class of research questions that are impossible to address via pre-digital methods.

The *Viral Texts* project at Northeastern University, launched in 2015, exemplifies this shift in research question scope. The project’s core research question—*what made some 19th-century newspaper content "go viral" across the United States, and what can this tell us about the structure of 19th-century public culture?*—was unanswerable before the advent of DH, as it required processing 132,000 issues of 41 regional and

national U.S. newspapers published between 1836 and 1922, a corpus totaling 12.3 billion words of OCR-processed text. The technical core of the project relied on a modified MinHash algorithm paired with Jaccard similarity scoring to identify overlapping textual segments across publications: first, the corpus was segmented into 5-gram chunks (sequences of five consecutive words) to reduce the impact of OCR noise, which averaged 15% error rates for 19th-century degraded print; next, each chunk was hashed using a locality-sensitive hashing (LSH) scheme to approximate Jaccard similarity without exhaustively comparing every pair of chunks; finally, segments with a similarity score above 0.8 (indicating 80% overlap in 5-gram content) were flagged as potential viral content, with a secondary human-in-the-loop validation step to filter out false positives from boilerplate advertising or public domain legal text. The project ultimately identified 12.7 million overlapping segments, revealing that viral content was not randomly distributed: religious revivalist sermons, commercial advertisements for patent medicines, and partisan political editorials accounted for 78% of all cross-publication repeats, and the geographic spread of these segments correlated with 19th-century railroad and telegraph network expansion with a Pearson correlation coefficient of 0.82. This finding overturned the longstanding historical assumption that 19th-century public culture was defined by localized, regionally distinct print spheres, demonstrating instead that a national public sphere emerged decades earlier than previously believed, structured by communication infrastructure rather than political events alone. The Viral Texts project is not an outlier: it exemplifies a broader class of DH research that leverages large-scale data processing to answer questions that were structurally impossible to address in a pre-digital context, a defining feature of the DH paradigm’s shift in research question scope.

Methodological Reconstitution: The Co-Constitutive Integration of Distant Reading and Hermeneutic Interpretation

If the shift in research question scope defines the *what* of the DH paradigm revolution, the shift in methodological integration defines the *how*. Traditional humanities research operates on a hermeneutic paradigm rooted in close reading: the sustained, contextualized interpretation of individual texts or artifacts, grounded in the assumption that meaning is embedded in the specific historical, linguistic, and cultural context of a single work. While close reading remains a