Introduction

The expression Distant Reading (DR) represents a wide array of practices in which large literary corpora are studied through algorithmic transformations, data visualizations, and statistical analysis. The concept has been mobilized heavily since it was coined by Franco Moretti in “Conjectures on World Literature” (2000), however, Distant Reading has a theory problem: it’s missing one.

This absence is strongly felt in the literature around DR: its relevance, scientificity, and legitimacy are questioned from within and outside Digital Humanities. Without a coherent or common understanding of DR among its practitioners, we cannot answer critics, adapt as afield, or advocate for the practice. The current state is that beyond Moretti’s initial texts (2000, 2013, 2005), theoretical papers on DR have mainly been critical of the current state of Distant Reading (see Trumpener 2009; Marche 2012; and Da 2019, among others), or offered limited strides towards a theory of DR (see Drucker 2017; Bode 2023). Distant Reading and other computational approaches to literary corpora (such as Matthew Jockers’ Macroanalysis, 2013) are the natural progression of formalism (Lvoff 2021; Gasparov 2016) and, later, literary and linguistic computing (Milic 1967; Deegan 1996). Massive digitalization projects and online repositories (Underwood 2019), as well as the democratization of programming and access to powerful personal computers lead (Irizarry 1988; Berry 2011) to increased possibilities for the quantitative study of large collections of literary texts: it is now possible for almost anyone to engage with DR.

A methodological model of Distant Reading

This presentation showcases a general methodological model of Distant Reading based on the last decades of practice in the field. The model explicits the general steps—corpus creation, literary modelling, algorithmic transformations, data analysis, etc.—and the design questions that are often invisibilized in articles and monographs on DR, such as corpus imbalance,random model initialization, and literary modelling. The model of Distant Reading pipelines combines the work and practices of multiple authors. It acts as an assemblage, drawing from, among others, Andrew Piper (2015, 2017) and Willard McCarty (2013) for literary modelling, from Johanna Drucker (2017, 2020) for data representation, and from Geoffrey Rockwell (2003; and 2022 with Stéfan Sinclair) for textual transformations. The result is a comprehensive and conceptual step-by-step guide to DR, summarized in Figure 1; a three-phase pipeline centered on the hypothesis or research question.

The goal of the suggested methodological model is not a standardisation of practices, but rather a tool to explicify the general DR pipeline. Each phase comes with design questions and algorithmic methods, some of which are invisibilized, some others that are normalized by the current state of the field. The plurality of computational methods, from corpus creation to visualisation, form a complex technical assemblage, and none of its elements should be left unexamined or taken for granted. Shining light on the actual practices of DR is a necessary step in its legitimization and theorization.

See main document for bibliography and Figure 1.

When Terry Eagleton declared the “end” of theory, he described not its failure but its integration into different academic routines (2003). At roughly the same time, DH began to cohere as a field. From this point on, questions once posed as problems of ideology, form, and critique reappear in DH as disputes over method, evidence, scale, and interpretation under computation. This could be understood, simplistically, as a polemic between close and distant readers, which in turn reflects a divide between that segment of the Humanities that embraced computational methods of analysis and that which did not.

By tracing the intellectual history of close reading from the New Critical tradition and distant reading from Franco Moretti’s formulation, this chapter proposes that the relationship between them is best understood as a productive, reciprocal misreading. Drawing on Harold Bloom’s concept of anxiety of influence, this paper argues that distant reading, in they way Moretti formulates it, operates as a parody of close reading: it exaggerates close reading’s interpretive commitments in order to open new space designed to provoke methodological innovation. Rather than representing mutually exclusive epistemologies, the two approaches secure disciplinary authority by productively misreading their predecessors, in doing so, reshape the terrain of literary study. This argument also challenges the assumption that traditional Humanities methods are holistic whereas Digital Humanities methods are reductivist. Recent debates in DH suggest instead that, at the epistemological level, both close reading and distant reading repeat the same operation at different scales of reading.

The paper concludes that the difference between the two approaches lies in the way each anchors its methodological authority by elevating distinct objects of investment—a process I term "fetishisation." Close reading locates this authority at the level of the minutia treating it as the privileged ground from which generalisations proceed; as opposed to distant reading, which relocates that authority to the archive as the primary locus of explanatory force.

Bibliography

Bloom, Harold. [1973] 1997. The Anxiety of Influence. 2nd ed. Oxford: Oxford University Press.

Eagleton, Terry. After Theory. London: Allen Lane, 2003.

Large language models increasingly present themselves as systems capable of translating across time, culture, and expertise, producing fluent outputs that appear to render complex human knowledge universally legible. In public, policy, and commercial discourse, this apparent translatability is framed as an opportunity: historical texts can be summarized, contextualized, and explained at scale with minimal human mediation. Digital humanities (DH) scholars, however, have long understood that digitization, modelling, and editorialization are not neutral processes, but interpretive acts that introduce bias, reduction, and conceptual loss.

This paper argues that the current moment does not represent a novel epistemic crisis so much as the amplification of a familiar one. What has changed is scale. The interpretive risks digital humanities has spent decades theorizing: probabilistic meaning, false equivalence, and the masking of uncertainty are now enacted through large language models (LLMs) that circulate far beyond scholarly contexts and claim authority through fluency alone. Drawing on established DH insights into modelling, remediation, and translation, I identify several recurring failure modes in LLM-generated knowledge production, including the flattening of culturally situated concepts, the suppression of uncertainty, and the substitution of interpretive judgment with confident approximation.

Rather than treating these failures as technical shortcomings to be resolved through improved accuracy or alignment, this paper reframes them as encounters with the untranslatable. In this sense, untranslatability is not an obstacle to be overcome, but a diagnostic signal, one that marks the limits of computational representation and the ethical stakes of meaning-making at scale. Digital Humanities, as a field that has long grappled with the consequences of translating cultural materials into computational form, is uniquely positioned to articulate why these limits matter and how they should inform responsible uses of AI.

The paper concludes by shifting from critique to leadership, asking how DH scholars might more effectively translate their hard-won methodological cautions into conversations beyond the academy. I suggest that treating untranslatability as a visible and meaningful feature, rather than an error to be smoothed away, offers a way for digital humanities to contribute not only to scholarly debates about AI, but also to broader discussions of governance, accountability, and interpretive responsibility in an era of automated knowledge production.