Text reuse has a rich and varied history for literary culture, encompassing a range of practices from direct citation, commentary and criticism to indirect references, paraphrases and allusions. ‘Nous ne faisons que nous entregloser’ goes the line by Montaigne, who warns us that all literature is in essence just a rehashing of what came before (2004, p.1069). This underlying tension between commentary and creation, or imitation and innovation, is in many ways the dialectic that drives literary practice and theory forward. In France, this conflict comes to a head at the turn of the 18th century, with the Querelle des Anciens et des Modernes ushering in a long and sometimes painful process of literary modernisation. The 18th century is thus an important transitory period in the long history of text reuse, one in which related concepts such as copyright and plagiarism begin to be discussed and codified.

This article aims to explore text reuse from both a theoretical and technical perspective, while demonstrating the potential of computational methods for the automatic detection of text reuses as a means for exploring the growing digital archive of the 18th century. As digital resources and collections reach a critical mass for Enlightenment studies – a field that has benefited greatly from the mass digitisation efforts of the past two decades (Burrows and Roe 2020) – new data-intensive methods are needed in order to exploit these resources to their fullest. Text-reuse detection is one such method. Technically, it is deceptively simple in its approach to text analysis: find two similar strings in at least two different documents. Theoretically, however, this simple implementation of text reuse is in need of both contextualisation – providing identified passages with an appropriate amount of historicity – and interpretation, which requires new forms of computational hermeneutics (Dobson 2019).

More generally, the need for deep domain expertise – philologist, historicist, hermeneutical – is essential for understanding the output of even the most robust computational models and methodologies; and it is here that humanists have much to offer. This is especially true in our current age of so-called artificial intelligence (AI), when large language models (LLMs) capable of reading whole universes of text and generating human-like responses based on this almost infinite archive, are in desperate need of theoretical grounding from the humanities (Dobson 2023). These models are the fever-dream, I would argue, of post-structuralism and we need scholars conversant in theory to understand these new artificial textualities. This article is thus a modest attempt at bridging this computational-hermeneutical divide; providing examples of text-reuse practices ‘in the wild’ as it were, along with both technical and theoretical reflexions on text reuse as an important, and perhaps overlooked, component of 18th-century literary culture.

From a computational perspective, text-reuse detection as a technique is best understood as a text-based instance of the ‘longest common subsequence’ (LCS) problem in computer science (Bergroth et al. 2000). Determining the LCS in texts is a long-standing challenge for natural language processing and forms the foundation for data comparison tools like the ‘diff’ utility (Hunt and McIlroy 1976). It is also a significant challenge in fields such as computational linguistics and bioinformatics. LCS algorithms are integral to version control systems like Git, which use them to merge various modifications made to a set of revision-controlled files (Blischak, Davenport and Wilson 2016). Similar approaches applied specifically to text reuse applications date back to the early 2000s, when Paul Clough et al. (2002) at the University of Sheffield began experimenting with measuring text reuse in large journalistic corpora. Its initial definition is quite pragmatic: ‘A topic of considerable theoretical and practical interest is that of text reuse: the reuse of existing written sources in the creation of a new text’ (p.152). Over the past 15 years or so, text-reuse projects and applications developed in the digital humanities have grown both in terms of number and methodological variety. Freely available tools for automatic text-reuse detection include BLAST,¹ passim,² Tesserae,³ Text-PAIR,⁴ textreuse⁵ and TRACER,⁶ among others. These tools were developed to address a wide range of scholarly use-cases ranging from: bioinformatic gene sequencing; reprint detection in antebellum American and Finnish newspapers; identifying allusions in classical poetry; early-modern French dictionaries and encyclopaedias; 18th-century print culture in England; and biblical reception, to name but a few.

In the field of 18th-century studies, text reuse has also found fruitful applications for understanding the growing digital cultural record. Mass datasets such as Gale’s Eighteenth Century Collections Online (ECCO) have in particular inspired teams working in Chicago (Gladstone and Cooney 2020) and Helsinki (Ryan, Mahadevan and Tolonen 2023) to explore text-reuse practices from philosophical, editorial and book-historical perspectives at a scale frankly unimaginable just two decades ago. These projects have not only moved research forward in their respective disciplines, but also represent important technical interventions for the wider digital humanities community, highlighting concerns over sustainable tool development, the accessibility of proprietary data and the use and design of data-rich interfaces for complex humanities corpora (Rosson et al. 2023).

From a theoretical perspective, text reuse can be seen as a concrete form of intertextuality – a concept that emerged in the 1960s and 1970s in France. Julia Kristeva was the first to use the term, translating Bakhtin’s notion of dialogism, and it subsequently gained traction within the Tel Quel circle (Kristeva 1969). As an expansive literary-theoretical concept, there are numerous and varied approaches to both the theory and practice of intertextuality.⁷ A conservative perspective might align with Harold Bloom’s theory of ‘influence’, where the art of criticism involves understanding the underlying connections that link one poem to another. In this view, writers grapple with the great giants of the past, necessarily borrowing from their predecessors to assert their own poetic authority. For Bloom: ‘Criticism is the art of knowing the hidden roads that go from poem to poem’ (1973, p.96).

A more radical take on intertextuality, appearing at the same time, comes from Roland Barthes, who posits that: ‘Tout texte est un intertexte; d’autres textes sont présents en lui, à des niveaux variables, sous des formes plus ou moins reconnaissables: les textes de la culture antérieure et ceux de la culture environnante; tout texte est un tissu nouveau de citations révolues’ (1973). Somewhere between these two extremes lies a more modest middle ground, one that is perhaps more tractable in computational terms, and which aligns itself with Gérard Genette’s later interpretation of intertextuality as (1982, p.7):

Today, reuse is widely considered as a form of plagiarism. This stems from the modern belief that words (and by extension, ideas) belong to specific authors and cannot be reused without their permission (Mallon 1989). This was not always the case, however, especially in the 18th century, as we shall see. While accusations of plagiarism were somewhat commonplace in the early-modern period, the widespread practice of creative reuse was an established aesthetic paradigm (Gevrey 1995, p.5):

Indeed, the culture of citation during this period oscillated between the opposing poles of originality and imitation. In the early 18th century, for example, the classical notion of imitation still held sway, and was predicated on the belief that the value of a work derived from its conformity to the great works of the past. Through these works, one could access truth, beauty, nature and other higher ideals that the ancients exemplified (Norman 2015). For instance, Alexander Pope, speaking of Virgil, conflates the ancient author with nature itself, both of which merit imitation (1713, v.133–36): Perhaps he [Virgil] seem’d above the Critick’s Law, And but from Nature’s Fountains scorn’d to draw: But when t’ examine ev’ry Part he came, Nature and Homer were, he found, the same.

And yet, from the mid-18th century onwards literary works gradually began to be understood as the emanation of what was unique in each individual – their genius (Schöch 2011). Concurrently, poets developed a new and immediate relationship with nature, one that did not need to be mediated by representations of the ancients.

To observe this shift in action, we need only compare Racine’s second preface to Britannicus, published in 1674, in which the classical playwright praises the conformity of his work with that of Tacitus: ‘J’avais copié mes personnages d’après le plus grand peintre de l’Antiquité, je veux dire d’après Tacite, et j’étais alors si rempli de la lecture de cet excellent historien, qu’il n’y a pas un trait éclatant dans ma tragédie, dont il ne m’ait donné l’idée’ (2015, p.9). Contrast this with Rousseau, who, just a century later, proudly declares in the opening of his Confessions that he had neither predecessor nor successor for his undertaking: ‘Je forme une entreprise qui n’eut jamais d’exemple, et dont l’exécution n’aura point d’imitateurs’ (Rousseau 2022, p.41). Never mind the fact that Rousseau’s work is generically borrowed from Saint Augustine and, thematically, continues the semi-autobiographical work of Montaigne (pace Jean-Jacques). Indeed, Montaigne is perhaps best understood as a key figure in the history of text reuse, exemplifying both the inherent ambiguity of a writer deeply imbued with the tradition of imitation – inserting his ‘fleurs étrangères’⁸ into his text as he goes along – while simultaneously aware of his identity as a unique and original author, intimately presenting himself to the reader (Compagnon 1979; Compagnon 1980).

Montaigne’s project, filtered through Pascal and the 17th-century moralists, reaches a certain apogee with Rousseau, whose work underscores the gradual transition away from a culture of imitation towards one of individual originality; an evolution that will ultimately transform our understanding of authorship and textuality as the literature of the Enlightenment gives way to the Romantic era. In this manner, the pre-Romantic notion of originality was in fact tied to the various practices of text reuse in the 18th century. Roland Mortier has demonstrated that the concept of originality is relatively new and, more significantly, should be understood as an offshoot of French Enlightenment aesthetics (Mortier 1981).

Following Mortier’s argument, ‘originality’ as an aesthetic category did not emerge in French until the very end of the 17th century. Roger de Piles first uses the term in his Abrégé de la vie des peintres (1699) to describe the originality of paintings – unique works that could not be copied: ‘S’il y a des choses qui semblent favoriser l’originalité d’un Ouvrage, il y en a aussi qui paroissent la détruire; comme la répétition du même Tableau, l’oubli où il a été durant beaucoup de tems, & le prix modique qu’il a coûté’ (p.100). Diderot later employs the term to describe a person of a very original character, specifically the Danish minister to France, Baron de Holberg, in a 1759 letter to Sophie Volland: ‘Nous dinâmes tous d’appétit. Notre baron, le nôtre, fut d’une folie sans égale. Il a de l’originalité dans le ton et dans les idées. Imaginez un satyre gai, piquant, indécent et nerveux, au milieu d’un groupe de figures chastes, molles et délicates. Tel il étoit entre nous’ (1950, p.28).

As a lexicographical item, ‘originality’ made its first dictionary appearance in the 1743 edition of the Dictionnaire de Trévoux, with the Jesuit editors proudly proclaiming that ‘Ce mot ne se trouve point encore dans les Dictionnaires’ (Dictionnaire universel françois et latin 1743, vol.4, p.1491).⁹ This was followed by a somewhat laconic entry in the fourth edition of the Dictionnaire de l’Académie française, which defined originality simply as the ‘Caractère de ce qui est original. Il se dit Des personnes & des choses’ (1762, vol.2, p.266). The semantic shift between ‘original’ as a descriptor and ‘originality’ as an aesthetic or epistemological category – moving from character to quality – reflects the evolution of a concept whose modern acceptation will only crystallise at the beginning of the 19th century.¹⁰

The great mid-century Encyclopédie, edited by Diderot and D’Alembert, in many ways exemplifies the above transition. While certainly still informed by the aesthetics of imitation inherited from their 17th-century forebears, the encyclopédistes nonetheless hint at its gradual exhaustion over the course of the work’s 20-year publication. The unsigned article ‘Original’, for example, still holds firm to the doctrine of imitation, defining an original as ‘le premier dessein, ou instrument authentique de quelque chose, & qui doit servir comme de modele ou d’exemple à être copié ou imité’ (Diderot and D’Alembert 2022, vol.11, p.648). And yet, it also acknowledges the rarity of true originality, noting that ‘aujourd’hui l’on trouve à peine aucun titre ancien de possession, inféodation, &c. qui soit original; ce ne sont que des vidimus, ou copies collationnées sur les originaux’.

The second article for the headword ‘Original’, in its grammatical sense, includes nothing more than a cross-reference to the article ‘Originalité’. As we know, cross-references in the Encyclopédie are far from neutral, and Diderot’s use (and abuse) of cross-references is particularly striking (Starobinski 1995). Thus, at the other end of the cross-reference – following an early-modern hyperlink as it were – we are brought to the article ‘Originalité’, which, while technically unsigned, is likely written by Diderot.¹¹ Here we find that originality is considered the ‘maniere d’exécuter une chose commune, d’une maniere singuliere & distinguée: l’originalité est très-rare. La plupart des hommes ne sont en tous genres, que des copies les uns des autres. Le titre d’original se donne en bonne & en mauvaise part (Diderot and D’Alembert 2022, vol.11, p.648). Diderot thus underscores the fundamental tension between imitation and originality that plagues any work of compilation such as a dictionary or his very own Encyclopédie.

From the outset, Diderot and D’Alembert faced numerous accusations of impiety, improper use of authorities and outright plagiarism (Albertan 1992). D’Alembert responded to these detractors in the editors’ foreword to the third volume (Diderot and D’Alembert 2022, vol.3, pp.xvii–xviii), explaining that:

If we turn to the Encyclopédie article ‘Compilateur’, which was authored, rather ironically, by one of the great text reusers of the Encyclopédie, the Abbé Mallet, we see emerge an interesting counter perspective. Mallet defines a compiler as an ‘écrivain qui ne compose rien de génie, mais qui se contente de recueillir & de répeter ce que les autres ont écrit (Diderot and D’Alembert 2022, vol.3, p.762). While Mallet’s compiler is very much not an advocate of originality, he goes on to note that for the most part lexicographers are also merely compilers called by a different name. This viewpoint likely irked Diderot, and, to rebut any ambiguity concerning lexicographical compilation, a cross-reference is included at the end of Mallet’s article, subtly directing readers to the entry ‘Plagiaire’.

‘Plagiaire’ begins by defining the term as an ‘écrivain qui pille les autres auteurs, & donne leurs productions comme étant son propre ouvrage’. However, in a clear rebuke of Mallet’s previous assertion, the article author, most likely Diderot, elaborates further (Diderot and D’Alembert 2022, vol.12, p.680) that:

Thus, in the context of the Encyclopédie, plagiarism is portrayed as both negative, in its most restrictive and legalistic sense, and, given the right context, potentially positive, especially when it comes to dictionary-making, where the judicious reuse of past authorities becomes an almost moral imperative. This positive spin on plagiarism will evolve further in the work of later writers. Voltaire, for example, in his Questions sur l’Encyclopédie – a significant work of text reuse in its own right – redefines the term in a more commercial sense (2012, pp.438–39):

In Panckoucke’s Supplément à l’Encyclopédie (1777), Marmontel argues that being labelled a plagiarist should by no means give one pause, as it is ‘une sorte de crime littéraire, pour lequel les pédans, les envieux, et les sots ne manquent pas de faire le procès aux écrivains célèbres’ (p.388). While it might entail the theft of thoughts, the outcry against plagiarism is often exaggerated, according to Marmontel. On the contrary, he continues, ‘Quiconque met dans son vrai jour, soit par l’expression, soit par l’à-propos, une pensée qui n’est pas à lui, mais qui sans lui serait perdue, se la rend propre en lui donnant un nouvel être, car l’oubli ressemble au néant’ (p.389). In this context, the plagiarist serves a productive purpose, saving the best bits of our shared culture from obscurity. Much like public law allows the cultivation of fallow land for the common good, Marmontel thinks literature should follow the same principle (p.389):

Viewed as such, plagiarism – or, more generously, text reuse – becomes not only permissible for Enlightenment writers, but can also be seen as a useful tool for shedding light on otherwise forgotten authors, both ancient and modern. Moreover, the productive poetics inherent in this expanded notion of text reuse is also in line with Diderot’s preference for philosophical ‘eclecticism’ over and above the more systematic philosophies of the 17th century (Mulsow 1997). The ambition of the eclectic, Diderot tells us (Diderot and D’Alembert 2022, vol.5, p.270):

Caught somewhere between the pull of imitation and tradition and the lure of originality and genius, the eclectic will become a transitory literary device of the late Enlightenment, applying Diderot’s philosophical ideas to other, more varied domains such as aesthetics, literary creation and, as we shall see, ‘illuminated’ mysticism (Antoine-Mahut 2024).

This historical divagation leads us back to text-reuse detection as a viable method for uncovering the various intertextual practices of 18th-century authors. With colleagues at the University of Chicago, we began testing these methods on the Encyclopédie itself, the text of which was first digitised in the late 1990s by the ARTFL Project and has been freely available on the web since 2007.¹² Our first experiments were aimed at automatically identifying potential sources of the Encyclopédie, testing various tried-and-true information retrieval techniques such as lexical similarity, distributional semantics and vector-space analysis to identify similar articles in other dictionaries. While these ‘bag-of-words’ techniques yielded fairly interesting results, confirmation that ‘similar’ articles were in fact borrowed or reused from previous dictionaries continued to elude us. Thus, we began developing an interest in more literal text-reuse detection, experimenting with the BLAST and Text-PAIR tools mentioned earlier, in order to apply sequence-based alignment (similar to those used in bioinformatic DNA analysis) to our datasets (see Allen et al. 2010; Olsen, Horton and Roe 2011; Roe 2012).

Building off these initial experiments, we have continued using the Text-PAIR system to identify regions of text reuse in the Encyclopédie and, progressively, other literary-historical corpora. A robust and flexible system for text-reuse detection, Text-PAIR effects scalable and fast comparisons of text collections of varying size and quality. Conceived specifically for digital humanities applications, Text-PAIR is highly fault-tolerant in its matching parameters, which allows it to find similar passages with significant changes and/or errors in transcription such as those generated by optical character recognition (OCR).¹³ Although the identified pairwise passages represent a deceptively straightforward notion of intertextuality – where the same text (or roughly the same text) appears in two different places – our experience over the years has indicated that this simple approach can reveal a large and varied range of text-reuse practices, from direct quotations to literary allusions, and everything in between.

For example, it is unsurprising that the Encyclopédie cites or uses Pierre Bayle’s Dictionnaire historique et critique (1697), a known predecessor and inspiration for the project. And indeed, we find a long sequence at the end of the article ‘Spinosa, philosophie de’ (Diderot and D’Alembert 2022, vol.15, pp.473–74) that is clearly derived from Bayle. What is surprising, however, is that the encyclopédiste (thought to be the Abbé Yvon) concludes the article in Bayle’s voice, reusing the first-person pronoun that begins the sentence, ‘Je finis par dire…’, without any quotation marks to indicate a direct quotation. While Bayle is mentioned at the beginning of the article, this blending of voices signifies something more nuanced: Bayle’s voice becomes part of the authorial fingerprint of the Encyclopédie in this and other articles. Whether readers at the time recognised this remains uncertain; but it is now clear that this is an example of a non-explicit reuse, which was only unearthed using computational methods.

This phenomenon is observed with numerous other works. The use of John Locke, for instance, in the Encyclopédie is particularly intriguing. Locke’s Second Treatise was banned and burned in France and was put on the Index of prohibited books. Explicitly citing Locke would have attracted the censor’s attention, risking the loss of publishing privilege for the Encyclopédie. However, the chevalier de Jaucourt – author of some 17 000 Encyclopédie articles – seemed determined to incorporate Locke’s ideas into his contributions and did so surreptitiously. The article ‘Gouvernement’ (Diderot and D’Alembert 2022, vol.7, pp.788–91), for example, is littered with passages taken from the French translation of Locke and woven into Jaucourt’s arguments seamlessly with no attribution whatsoever. This covert incorporation of Locke further exemplifies how text reuse in the 18th century functioned as a means of generating meaning through the often-unacknowledged redeployment of authoritative or subversive models.

These examples were identified as part of a collaborative research project which uncovered a sophisticated system of citation strategies within the Encyclopédie (Edelstein, Morrissey and Roe 2013). The use of loose, or non-existent, citations and naming conventions, coupled with other subtle cues, meant that the encyclopédistes were able to smuggle in passages from prohibited works while avoiding (sometimes) the censor’s oversight. This system of implicit references may have been clear to some 18th-century readers but is less obvious to contemporary audiences. Specialists might discern these nuances, but for the most part we must rely on computational analysis to reveal these intricate layers of text reuse. Viewed in this light, the Encyclopédie functions not only as a repository of knowledge but also as a complex network of interwoven texts, connected through both direct and indirect citations, each reflecting the deep intertextual currents of its time.

Uncovering these hidden voices in the Encyclopédie helped us better understand its status as a polyphonic philosophical work, wherein authors such as Bayle and Locke become de facto encyclopédistes in their own right. While approximately 140 named contributors to the Encyclopédie are known, there are likely many more whose contributions remain unidentified. One such figure is Émilie Du Châtelet. As Judith Zinsser has stated, many Enyclopédie authors: ‘simply took whole passages from [Du Châtelet’s] Institutions de physique for subjects as general as “Time” and “Hypothesis” and as specific as the descriptions of the metaphysical concepts of Leibniz and Wolff. Few identified the original author, and so Du Châtelet acquired the title of so many other women before and after her: “anonymous”’ (2006, p.285). Using a digitised version of the Institutions, we tested this hypothesis using Text-PAIR, discovering some 13 articles that contain substantial content ‘borrowed’ from Du Châtelet (Roe 2018). The article ‘Contradiction’, to take but one example, is actually 95 per cent Du Châtelet. Its stated author, Samuel Formey, was the Secretary of the Academy of Berlin, and originally intended to create his own philosophical dictionary but, after abandoning the idea, he later handed over his papers to D’Alembert. Unbeknownst to D’Alembert, Formey’s articles heavily cribbed from Du Châtelet’s Institutions (and undoubtedly other sources), and these unattested borrowings were then incorporated into the Encyclopédie with Formey credited as sole author. This complicated lineage begs the question of Du Châtelet’s status as an author in the Encyclopédie, restoring a lost voice to the polyphonic text that had been muted for the last few centuries.

Moving beyond the Encyclopédie, and working with colleagues in Oxford, we wanted to explore text-reuse practices in other text collections and contexts, tracing their participation in the broader communication networks of 18th-century print culture, and drawing inspiration from Robert Darnton’s now-famous communication circuit model (see Figure 1), which demonstrates the flow of information through various media channels in 18th-century France (Darnton 1996, p.189). With this model in mind, our aim was to use text reuse as sort of relay through which literary value was formed and debated in both the public and private spheres. To do this, we turned our focus to private correspondences, in particular those assembled by the Electronic Enlightenment project at Oxford.¹⁴ As we know, private letters acted as a platform for sociability and a means of diffusing information and building literary authority. They were also, as we will see, a great vector for text reuse.

Perhaps no author in the 18th century was more attuned to the literary and social uses of correspondence than Voltaire. Recently, we have begun exploring the extraordinary richness of his more than 21 000-letter correspondence using digital methods (Cronk and Roe 2020). Unsurprisingly, turning Text-PAIR loose on this corpus also provides excellent examples of how text reuse functions in the (semi-) private sphere. Throughout his letters, Voltaire exhibits extraordinary reading and sharing practices, acting much like an early-modern Reader’s Digest, excerpting works and sharing these passages with his friends. Take, for example, Helvétius’ book De l’homme, published posthumously between 1771 and 1773, from which Voltaire takes snippets and sends them to D’Alembert, commenting on their quality (D18450, see Voltaire 1975, p.42–43).

Voltaire remarks on the brilliance of certain passages despite his overall low opinion of the work, calling it a bad book but noting the several ‘très bonnes choses’ he nonetheless found within it. He sends similar excerpts of the same work to Madame Du Deffand, who was blind (D18607, Voltaire 1975, p.164).

With this last paragraph in particular, Voltaire sums up the valuable service he has provided to his correspondent by sending her only the best parts of a book, thereby sparing her the effort of reading (or, in the case of Mme Du Deffand, having her secretary read to her) what he considered subpar content.

This anecdote captures another striking form of the poetics of text reuse in the 18th century: the process of accruing knowledge through the dissemination of excerpts. Voltaire’s method involved not just extracting valuable passages from texts but also sharing them widely, thereby facilitating their circulation and, as we saw Marmontel arguing above, safeguarding them from future oblivion. There is thus a deep sociability inherent to 18th-century text-reuse practices, contingent upon someone actively reusing and sharing content through early-modern social networks.

Finally, moving on to a more recent project exploring practices of text reuse, a team of researchers at the Sorbonne are aiming to identify and analyse these 18th-century intertextual networks on a vast scale. The project, generously funded by the European Research Commission and titled ModERN (Modelling Enlightenment: reassembling networks of modernity using data-driven research), participates in the so-called ‘network turn’ of digitally-inflected humanities research that has been on the rise over the past ten years.¹⁵ Social network analysis (SNA) – borrowed from the computational social sciences – has proven particularly useful for understanding cultural exchanges across various domains and linguistic traditions. Here, Ruth and Sebastian Ahnert have been instrumental in promoting this approach for historical and literary studies using correspondence networks (Ahnert and Ahnert 2015; Ryan and Ahnert 2021; Ahnert and Ahnert 2024). Further applications of SNA in the realm of Enlightenment studies have also been highlighted in a recent volume edited by Chloe Edmondson and Dan Edelstein (2019). Additionally, the 2021 Cambridge Element, The Network Turn, co-authored by the Ahnerts with Nicole Coleman and Scott Weingart, serves as an excellent primer on this methodology and its potential applications across the humanistic disciplines (Ahnert et al. 2021).

Beyond SNA as a guiding methodology, ModERN also aims to incorporate Bruno Latour’s Actor-Network Theory (ANT) as an underlying theoretical framework (Latour 2007). ANT has two main advantages: first, it avoids presuppositions and a priori assumptions when considering networks and their behaviour, aligning well with a ‘data-driven’ research philosophy. Secondly, ANT considers any entity (human or non-human) that exerts an influence on network behaviour as an actor; so, authors and readers, certainly, but also books themselves, letters, pamphlets, extracts, cafés, salons, ad infinitum. Another key insight from ANT is that mediators – those actors that facilitate communication and exchange – can often be the most critical nodes in a network, moving focus away from the originators or receptors of the information flow. This sets up the ‘mediator’ as a key actor in our intertextual networks, one that navigates the waters between originality and imitation outlined above.

To build these networks we obviously need extensive data. Thus, the first years of the project have been focused on acquiring as many 18th-century sources as possible, with the sole caveat that the texts must already be digitised in some form or another. We rely on a mixture of corrected or transcribed works, which we call the Canon sub-corpus, and uncorrected texts that have been digitised using OCR technologies, which we call the Archive. Our main combined database (Canon + Archive) includes just over 13 000 documents (manly printed books), which we will compare iteratively with other secondary corpora of 18th-century pamphlets, letters, dictionaries and newspapers.¹⁶ The variety of our collections speaks to our goal of considering Darnton’s communication network in its totality, recognising that books are just one part of the 18th-century information flow. As we have seen above, dictionaries and private letters both play crucial, albeit distinct, roles in the formation and exchange of intertextual networks, and we expect pamphlets and the press to do the same.

Our overall goal for the project is twofold: first, our database of 18th-century text reuses will act as a supplement to traditional ‘close-reading’ methods, and, ideally, offer new ways of performing historical and literary analyses with significantly greater speed and breadth. By leveraging the power of computational tools such as Text-PAIR to analyse a massive textual corpus, we can begin to tackle essential questions about the literary history of the 18th century at an unprecedented scale. These tools will thus not only enrich our understanding of the past but also demonstrate the dynamic interplay between texts and their historical contexts, offering a more nuanced and comprehensive picture of French literary history as it was being debated and codified. Secondly, and in a more ‘revolutionary’ spirit, the ModERN project aims to deploy new ‘distant-reading’ methodologies in order to challenge received ideas about the Enlightenment and its main actors.¹⁷ A greater focus on intertextual networks will enable us to remap the literary history of the 18th century using SNA and other computational approaches to trace and categorise how texts and ideas move through different contexts. This approach embraces the more radical version of intertextuality espoused by Roland Barthes, one in which every text is a potential (or real) intertext (Barthes 1973):

As Barthes suggests, intertextuality is not merely about the reproduction of some other text, but rather a latent productivity that informs all texts. An intertext or cluster of intertexts thus becomes a productive force, often independent of any authorial intent.¹⁸ These reuses traverse texts and contexts, sometimes attached to author-names, but often not, a fact that invites a re-evaluation of authorship in our intertextual model of literary dissemination. For us, an author (named or unnamed) is but one attribute of a text-object; the model’s primary focus is on the text itself and its productive circulation.

When we do generate author (or ego) networks, we understand an author’s importance not from the perspective of traditional literary history but from their position within the network and their author-function as text reusers. Our model thus defines an author-object as the result of the combination of all the characteristics attributed to it by the relations between text-objects; its ‘influence’ is measured by way of its position within the intertextual network. Fénelon, for example, while certainly an important ‘authority’ in terms of text reuse, is more influential in our model as a ‘mediator’, i.e. a node that bridges several different communities that are normally separate, such as biblical exegetes and the philosophes, thereby underscoring the fluidity and interconnectedness of 18th-century literary culture.

This methodology also prompts us to reconsider what constitutes a ‘text-object’ in our model. Prosaically, we define a text-object as a co-occurrence of at least four similar trigrams within a defined interval that has been identified in at least two different documents. This definition raises several issues, especially since not all reuses are equal. In Figure 2 we see two pairs of identified ‘reuses’ from our database. In the first, we have the curious example of Marat, in the midst of the revolutionary furore, reusing – whether consciously or not we cannot say – a passage from Fénelon’s Télémaque in his Chaînes de l’esclavage. The passages are nearly identical, though Marat replaces ‘lion de Numidie’ with ‘tigre’ and omits three words, ‘de faibles brebis’, but the meaning of the passage is completely transformed in its revolutionary context. This example is exactly the kind of productive reuse we seek to identify. The second example, however, would seem superficially important, given its length, near-exact replication, and the similar book titles in which it occurs. But, on examination, this match might fall into the category of ‘noise’: a formula which is so much of a commonplace that it has little or no significant intertextual value. These sorts of decisions – what constitutes a ‘reuse’ in our database – along with establishing a typology of all reuses, will be one of the main research axes for the project going forward.

These examples bring us back to Barthes’ concept of intertextuality. For Barthes, ‘tout texte est un intertexte’, and, as such, ‘intertextualité, condition de tout texte, quel qu’il soit, ne se réduit évidemment pas à un problème de sources ou d’influences; l’intertexte est un champ général de formules anonymes, dont l’origine est rarement repérable, de citations inconscientes ou automatiques, données sans guillemets’ (Barthes 1973). The challenge for us thus lies in identifying and classifying these different sorts of intertexts, from direct quotations, indirect borrowings, loose paraphrases and allusions, down to the aforementioned ‘anonymous formulas’, all the while bearing in mind the various author-functions (anonymity, pseudonymity, collaborative authorship, etc.) attached to them.

To identify and classify the most highly intertextual texts or ‘hubs’ in our data, we rely on ‘graph profiles’ that combine several well-known metrics for network centrality: in- and out-degree centrality (Freeman 1978), betweenness centrality (Barthelemy 2004), and PageRank (Page et al. 1999), among others. While these SNA measures are in need of proper assessment concerning their applicability for networks constructed out of text-reuse data, we are nonetheless encouraged by our preliminary results. As text-reuse networks are not ‘natural’ networks in the sense that (some) social networks are, using standard SNA measures on them requires rigorous reflection on what is actually being measured and analysed. While these assessments are outside the scope of this paper, our initial deployment of these graph profiles helps us identify and classify a large variety of intertexts, i.e. those that are reused the most (authorities), those that reuse the most (observers), those with high betweenness that bridge communities (mediators) and those whose reuses are the most influential based on their PageRank score (influencers). A work’s graph profile can thus give us an indication of how it behaves in the network and, ideally, bring lesser-known works to light.

To give a salient example, a text previously unknown to us that repeatedly emerges in our graph analyses is a certain anonymous volume entitled Pensées républicaines (1794), which was part of the Bibliothèque bleue de Troyes collection.¹⁹ Initially, we noticed its relatively high ranking in several network measures without fully understanding its significance.²⁰ Upon closer inspection, it became clear why it scored so highly: it is essentially a text-reuse machine. Written at the beginning of the Revolution, Pensées républicaines is a collection of maxims and aphorisms intended for children, with one for each day of the Republican calendar. None of the maxims are attributed in the original text, and our general alignment only found around 100 of the 360 odd entries in our main corpus. Nonetheless, the sources we did identify attest to its syncretic nature, drawing on a wide range of sources, including Rousseau, La Bruyère, Benjamin Franklin and a history of China, among others (see Table 1).

If we examine just one ‘decade’, the seventh of Frimaire, we see a fascinating intertextual mosaic emerge (see Figure 3).

Benjamin Franklin is reused twice (nos 1 and 7), along with Voltaire (no. 3), La Bruyère (no. 5), Seneca (no. 8), Duclos’ Histoire de Louis XI (no. 10), and Servan’s Discours sur l’administration de la justice criminelle (no. 6), with each maxim loosely related thematically (work/idleness/pleasure). What is especially notable in the text is the total absence of all authorial authority; the maxims stand on their own, disconnected from any origin or source.

Moving from quantitative to qualitative analysis, or from distant- to close-reading modes, there are numerous deformations to identified maxims that are altogether remarkable. For instance, the passage drawn from Duclos’ History of Louis XI is actually altered, changing its meaning entirely (differences in bold):

Duclos, Histoire de Louis XI, Préface:

Ce que j’avance au sujet d’une nation peut s’appliquer aux particuliers. Les hommes privés de lumières sont toujours dans l’occasion du crime; au lieu qu’un homme d’esprit, n’eût-il que des vues d’intérêt, sent qu’il n’a point de meilleur parti à prendre que d’être honnête homme. On est bien près de suivre la vertu, quand on est obligé de rougir du vice.²¹

Pensées républicaines, Frimaire, septième décade, pensée 10:

Un homme d’esprit, n’eût-il que des vues d’intérêt, sent qu’il n’a point de meilleur parti à prendre que d’être uni à l’intérêt commun.²²

One need not be a historian of the Revolution to understand the importance of placing the common good above the individual interest of the ‘honnête homme’.

The text becomes even more intriguing when considering examples from elsewhere in the calendar. In the sixth decade of Brumaire, we find the following pensée (Pensées républicaines 1794, pp.13–14):

This maxim is actually drawn from two different sources, and then slightly altered (the relevant text is shown in bold in each case):

Corneille, Le Cid, IV.v: Qu’il meure pour mon père, et non pour la patrie, Que son nom soit taché, sa mémoire flétrie; Mourir pour le pays n’est pas un triste sort; C’est s’immortaliser par une belle mort. J’aime donc sa victoire, et je le puis sans crime, Elle assure l’État, et me rend ma victime,²⁴

Voltaire, La Henriade, Chant 2: Quelques-uns, il est vrai, dans la foule des morts, Du fer des assassins trompèrent les efforts. De Caumont, un enfant, l’étonnante aventure Ira de bouche en bouche à la race future. Son vieux père, accablé sous le fardeau des ans, Se livrait au sommeil entre ses deux enfants; Un lit seul enfermait et les fils et le père.²⁵

What are we to make of this amalgamation? Does the pensée belong to Voltaire or Corneille? Or neither? This unexpected juxtaposition – not to mention the insertion of Barras to replace Caumont – again raises questions about the validity of our understanding of intentionality when viewed through the prism of intertextuality. These subtle, yet significant changes, also highlight the limitations of computational analysis; while the computer can identify these connections, understanding their significance requires deeper interpretive work. The hybridity of the Pensées républicaines is a testament to the complexity of intertextuality in the 18th century, where texts constantly intersect and transform, challenging our modern notions of authorship and originality along the way.

Finally, another text that has surfaced repeatedly in our analyses is a fascinating work by a certain Jean-Marie Chassaignon (1779), whose title is ‘tout un programme’: Cataractes de l’imagination, déluge de la scribomanie, vomissement littéraire, hémorragie encyclopédique, monstre des monstres, par Epiménide l’Inspiré… dans l’antre de Trophonius, au pays des visions. Chassaignon was a poet from Lyon who was highly active in the illuminist and antiphilosophe circles assembled around Louis-Claude de Saint-Martin (Poirson 2017). Chassaignon’s ‘frenetic’ and ‘excessive’ writing style was indicative of a certain stylistic ‘énergie’ that pervades the late Enlightenment according to Michel Delon (1988). His four-volume magnum opus, the Cataractes, published in 1779 and never re-edited, is a strange and fascinating work; taking Diderot’s notion of eclecticism to a rare extreme, it mixes genres from the poetic to the philosophical, mystical and theosophical, and enacts the productive poetics of text reuse we have been discussing in this essay.

Chassaignon begins the book with a preface ‘qui n’en est pas une’, in which he immediately denigrates the very idea of any prefatory introduction: ‘On ne lit plus les préfaces: elles ne sont pour l’ordinaire qu’un surcroît d’ennui pour le lecteur à qui elles n’apprennent rien’ (1779, vol.1, p.5). And yet, he then proceeds over the next hundred pages or so to introduce his work in painstaking detail: ‘Mais, moi qui crie contre les préfaces, je m’en permets cependant une, & me laisse surprendre en flagrant délit; oui, parce que mon opuscule ne pourroit s’en passer. J’écris dans un genre inconnu à ce siécle & il faut que je donne la clé de l’énigme, si je veux être compris’ (pp.6–7).

Though it was presented as being in a wholly new genre, Chassaignon’s work was not without predecessors: ‘Plutarque, [La Moth] Lévayer, Bayle & Montagne qui ont entassé pêle & mêle sur le papier tout ce qui leur venoit dans la pensée & ont submergé leurs écrits d’annotation, m’ont communiqué le goût de leur genre. Le dernier sur-tout m’a enhardi à suivre ses traces; on ne peut résister à sa maniere naïve & entraînante’ (p.18). As we have seen, Montaigne is key figure in the history of text reuse, though Chassaignon admires him most, it seems, for writing chapters that diverge entirely from their titles, following his disparate thoughts wherever they may lead, and including as many citations as needed to demonstrate his way of thinking. In Pierre Bayle – avowed predecessor to the Encyclopédie and the ‘arsenal of Enlightenment’ according to Cassirer (1951, p.167) – Chassaignon finds structural inspiration for his work, and particularly Bayle’s extensive system of notation, weaving text, remarks, footnotes, marginal notes, references, etc. into the fabric of his Dictionnaire historique et critique (Van der Lugt 2016). Like Bayle, Chassaignon’s notes can range from a single phrase to several pages of self-commentary, often running longer than the main text, which is punctuated by note calls, both foot and marginal, and discursive asides marked with an asterisk – perhaps recalling Diderot’s authorial mark in the Encyclopédie (see Figure 4).

Further on in his preface, Chassaignon outlines his own theory, philosophy and practice of text reuse, all in one go (1779, vol.1, pp.34–36):

The two Latin quotations, included without attribution, deftly demonstrate Chassaignon’s poetics of reuse. The first comes from Virgil’s Aeneid (Book 1, lines 133–35), and serves as an illustration of Chassaignon’s eclectic project of mixing genres and elements: ‘Do you now dare, winds, without command of mine, to mingle earth and sky, and raise confusion thus?’²⁶ This passage would certainly have been recognisable by most 18th-century readers and perhaps needed no attribution. The second citation, however, is a bit murkier. It would seem to be a deformation of a verse written by Charles-Jean-François ‘le Président’ Hénault and published as an epigraph to the first edition of his Nouvel abrégé chronologique (1744): ‘Indocti discant et ament meminisse periti.’ It was initially thought to be from Horace, a misattribution that persisted well into the 19th century (Longpérier 1875);²⁷ although Hénault would eventually disclose the origin of the verse in the third edition of his work: namely his own translation of lines 741–42 of Pope’s Essay on Criticism (1713): Content if hence the unlearned their wants may view, The learned reflect on what before they knew

Whatever its paternity, Chassaignon’s replacement of ‘unlearned’ with ‘children’ makes the verse his own, all the while keeping its intertextual link with Hénault (not to mention Pope and Horace) for those in the know.

Chassaignon’s eclectic virtuosity in reusing texts thus forms an integral part of his poetic enterprise (Moore 2010). It is through the interweaving of old and new, ancients and moderns, known and unknown, that new forms and meanings can emerge (Chassaignon 1779, vol.1, pp.33–34):

It is an almost alchemical approach to literature, hoping that the right mixture of elements might transform literary lead into gold. The productive intertextuality mentioned early is thus very much operative in Chassaignon’s work, which is another example of a text-reuse machine, weaving together a vast array of sources into a cohesive, albeit chaotic, whole. Like the Encyclopédie before him, which generated new knowledge through the critical reuse of sources, Chassaignon’s project is to generate enlightenment itself, the deep illumination that can only come from literary creation. It is not an easy road to travel, and one can easily get lost, but given the right tools and enough time, anything is possible: ‘un bisarre débordement de pensées errantes, & de passages dérobés n’exige pas un effort bien étrange de l’imaginative; avec des dictionnaires, des recueils, de la patience, des copistes & du temps, le scribe le plus borné est en état de peupler l’univers littéraire de monstres encyclopédiques’ (p.36). This conception of creation through reuse should resonate with our times, where the right (digital) tools, such as ChatGPT and other LLMs, can also create poetic ‘monsters’ of text reuse.

Diderot, of course, already foresaw these sorts of monstrosities in the making when he took stock of the encyclopédistes’ ‘universalising’ enterprise in 1755. Far from the perfect system of human understanding proposed by the mathematician D’Alembert, what they had created in reality was uneven, unwieldy and ultimately uncontrollable (Diderot and D’Alembert 2022, vol.5, pp.641–42):

Despite these reservations, the Encyclopédie became an intertextual dynamo, driven forward by the potentiality of critical text reuse and the creative productivity of its inherent intertextuality. In this way Diderot’s dictionary can rightly be seen as predecessor not only to modern encyclopaedic projects like Wikipedia and the internet (Bianco 2002; Fernandez 2019), but also, and perhaps more significantly, it shares a direct lineage with the imaginary information machines of Charles Babbage, Ada Lovelace, Alan Turing and Vannevar Bush. The poetics of reuse outlined in the Encyclopédie and put into practice throughout the 18th century is thus a resolutely modern, if not contemporary, phenomenon. In our age of AIs and digital information machines, recognising both the historicity and contemporaneity of these intertextual practices is crucial for our understanding of the legacy of the Enlightenment as well as our place in it. In some ways, one could go so far as to argue that works like the Encyclopédie, Voltaire’s correspondence, the Pensées républicaines and Chassaignon’s Cataractes – literary monsters one and all – are only fully actualised in and by the intertextual affordances of the digital.

Returning to our illuminist poet, finding more information about Chassaignon has proven challenging. A 19th-century biographical entry in Quérard’s La France littéraire simply states his birth and death dates and lists his published works, including the Cataractes. Curiously, the entry ends with an editorial note that claims that Chassaignon ‘avait laissé beaucoup de manuscrits que son frère, épicier à Lyon, a employés pour envelopper les drogues de sa boutique’ (Quérard 1828, p.145). Less conventional, and certainly far from computationally tractable, this tangible example of text reuse serves as a final reminder of the enduring relevance and diversity of text reuse as a cultural practice.

Our goal with this article has been to present a thorough, but by no means exhaustive, investigation into the practice of text reuse in the 18th century, demonstrating its significance not only as a literary phenomenon but also as a cultural mechanism that shaped intellectual exchange. By examining both historical and computational perspectives, we highlight how the boundaries between plagiarism, citation and innovation were fluid during this period. Our analysis reveals that text reuse was not merely a form of borrowing, or worse, plagiarism, but rather a means of contributing to the dialogic process of early-modern knowledge production, reflecting the broader socio-political and intellectual currents of the time.

As we have seen, automatic text-reuse detection tools such as Text-PAIR, combined with large-scale digital collections, can offer new insights into how intertextuality functioned historically, revealing hidden connections between texts and authors that traditional close reading might overlook. These methods enable scholars to trace the intricate ways in which ideas circulated and were reshaped within ‘intertextual hubs’ such as the Encyclopédie and Voltaire’s prodigious correspondence. In this context, the blending of voices, particularly through the identification of unacknowledged contributions of figures such as Émilie Du Châtelet, underscores the complexity of authorship practices and the aesthetics of eclecticism prevalent in the 18th century.

Ultimately, this article not only provides a deeper understanding of historical text reuse but also demonstrates the potential of the digital humanities to reconfigure our traditional approaches to literary and cultural history. Through ongoing projects that aim to map these intertextual networks, the study offers a nuanced view of the Enlightenment as a period of rich intellectual exchange, driven by both overt and covert forms of collaboration. This approach reaffirms the importance of considering both humanistic and technological perspectives when examining the literary past, thus opening up new avenues for future research.

We expect text-reuse approaches to continue to evolve rapidly as machine learning and AI models become more prevalent in the investigation and exploitation of our cultural collections. LLMs will certainly have their place in this evolution and might in fact guide both technological and theoretical reflection about intertextuality moving forward. As suggested above, the ‘poetic monsters’ of generative AI systems such as ChatGPT function primarily as massive intertextual machines; those that combine and remix past texts probabilistically in order to generate ‘new’ textual artefacts delivered through dialogical interfaces. From a theoretical perspective we are thus closer today than ever to Roland Barthes’ adage that ‘tout texte est un intertexte’, and these same LLMs will no doubt prove helpful not only from a generative perspective, but also analytically, allowing us to detect broader, more nuanced forms of text reuse that go beyond the current sequence-based models (Kanerva et al. 2024).