The Computational Turn: Epistemology and Foundations of Digital Hermeneutics

The computational turn in the humanities is not merely an adoption of new tools but a profound reconfiguration of epistemological foundations, methodological frameworks, and the very definition of scholarly evidence. This shift, often encapsulated under the banners of Digital Humanities (DH) and computational literary studies, promises scalable analysis and novel insights unattainable through traditional close reading alone. Yet, it also risks epistemological naivety if the philosophical assumptions underlying algorithmic processing remain unexamined. This chapter establishes the critical bedrock for the technical work that follows by tracing the historical lineages of this turn, dissecting its core debates around representation, scale, and evidence, and formally introducing the concept of *algorithmic hermeneutics* as a synthesized paradigm. We will define the scope of "text" and "artifact" within computational contexts and map a detailed taxonomy of common DH research questions—attribution, influence, genre evolution, discourse analysis—to their appropriate computational paradigms, thereby setting a rigorous technical agenda that moves systematically from data to model to interpretation.

Historical Lineages: From Structuralism to Cliometrics to Digital Humanities

The impulse to apply systematic, often quantitative, methods to humanistic materials predates the digital computer by decades, rooted in early 20th-century intellectual movements that sought to uncover underlying structures governing cultural phenomena. Structuralism, with its emphasis on analyzing systems of signs and relations rather than individual content, provided a crucial conceptual precursor. Ferdinand de Saussure's *langue* versus *parole* distinction—the systemic rules of language versus individual speech acts—resonates deeply with the computational separation of a formal model (the grammar, the corpus statistics) from specific instances (a given text). Claude Lévi-Strauss's application of structural analysis to mythology, seeking binary oppositions and deep structures across disparate tales, prefigured the search for latent patterns across large document collections that topic modeling and network analysis now perform. This structuralist legacy imbued computational humanities with a goal: to move from the idiosyncrasy of the individual artifact to the discernment of governing systems.

A more directly quantitative lineage emerged in the 1960s and 1970s with cliometrics (or "new economic history") and the early humanities computing movement. Cliometrics, pioneered by economists like Robert Fogel and Douglass North, applied econometric techniques to historical data, famously using statistical analysis to challenge narratives about the economic efficiency of American slavery. This demonstrated that historical arguments could be grounded in formal, testable models rather than solely archival interpretation. Concurrently, the humanities computing pioneer Roberto Busa, S.J., embarked on a decades-long collaboration with IBM to create the *Index Thomisticus*, a machine-readable concordance of the works of Thomas Aquinas. Begun in the 1940s and culminating in a 56-volume print edition and later digital versions, this project proved that massive, complex Latin texts could be processed computationally to reveal lexical and conceptual patterns. Busa's work established two enduring pillars: the necessity of text encoding to represent hierarchical structure and semantic nuance, and the vision of the computer as a "slave" that could exhaustively perform tasks (like indexing every occurrence of a word) impossible for a human scholar, thereby *enabling* new forms of close reading on a grand scale.

The formalization of text encoding standards, particularly the Text Encoding Initiative (TEI) founded in 1987, was a watershed. TEI provided a shared, extensible XML-based grammar for marking up textual features—from paragraph structure and verse forms to named entities, editorial interventions, and textual criticism. This was not just a technical standard but an epistemological statement: it argued that a text's meaning is inextricably linked to its material and structural features (line breaks, marginalia, variant readings) and that these features must be formally represented to be computationally accessible. TEI made the "digital surrogate" a scholarly object in its own right, capable of supporting sophisticated analysis while preserving interpretive richness.

The term Digital Humanities coalesced in the 1990s and 2000s, broadening from text-centric humanities computing to include multimedia, GIS, network visualization, and maker culture. Early DH projects like the WordHoard (a tool for analyzing Early English poetry) or the Perseus Digital Library (integrating lexical tools, morphological analysis, and texts) exemplified a commitment to open tools, open data, and collaborative creation. The rise of web technologies and later, the big data era (circa 2010 onward), catalyzed a second wave. Mass digitization projects (Google Books, HathiTrust, Europeana) created corpora of unprecedented scale, while machine learning—particularly deep learning and natural language processing (NLP)—offered powerful new pattern recognition capabilities. This "computational turn" thus represents a convergence: structuralist ambitions for system-level understanding, cliometric faith in formal modeling, humanities computing's focus on encoded artifacts, and DH's culture of open, tool-based inquiry, all supercharged by algorithmic scalability.

Core Debates: Representation, Scale, and the Nature of Evidence

This historical trajectory brings us to three interlocking debates that define the current epistemological crisis and opportunity in computational literary studies.

1. Representation: The Lossy Transformation from Artifact to Data. Every computational process begins with a *representation* of the original artifact, and this transformation is inherently reductive. A physical book—with its paper grain, typeface variations, marginalia, binding, and smell—is rendered as a stream of characters, perhaps marked up with TEI tags for structural elements. A painting becomes a pixel matrix. The central question is: *What is lost, and what is preserved, in this translation?* The loss is not merely of sensory experience but of interpretive context. TEI can mark a phrase as "emphasized" but cannot encode *why* the author emphasized it—rhetorical force, irony, or printer's error? Optical Character Recognition (OCR) introduces errors that can systematically bias analysis, as seen in studies where 19th-century "wimmin" (a deliberate feminist spelling) is OCR'd as "women," erasing a key marker of ideological stance. Even perfect transcription loses paratext: the order of texts in a volume, the presence of advertisements, the physicality of the codex. The debate forces us to confront the ontological status of the digital artifact: is it a proxy, a model, or a new object of study in itself? A rigorous approach demands *reflexive encoding*: making explicit the decisions about what to represent (e.g., preserving original spelling vs. modernizing) and justifying them in light of the research question. For instance, a study on dialect would preserve orthographic variation meticulously, while a study on thematic evolution might lemmatize all words to their dictionary forms, sacrificing surface form for conceptual unity.

2. Scale: From Close Reading to Macro-Analysis and Back. The most celebrated promise of computational methods is *scale*: analyzing millions of documents, tracking trends over centuries, or comparing thousands of authors. Projects like the Google Books Ngram Viewer allow users to graph the frequency of words across a corpus of ~8 million books, revealing cultural trends with a sweep impossible for any individual reader. However, scale introduces a methodolog