The relationships between texts are multifaceted and complex, ranging from directly attributed quotations to indirect influences and faint allusions. Tracing these links at all levels is a core humanistic endeavor that can illuminate the meaning and reception of texts through the identification and examination of references, commonplaces, borrowings, re-workings, and even plagiarism. To aid in this sort of intertextual discovery, computational methods for examining literary text reuse can draw upon sequence alignment techniques developed primarily for gene sequencing programs in bio-informatics and plagiarism detection systems, with specific adaptations to suit the needs of humanities scholars. This paper examines just such an approach, first describing an open-source software package developed by the ARTFL Project at the University of Chicago for text reuse discovery in digitized text collections, and then offering a concrete application of this technique for literary critical research.

The software, named ‘PAIR: Pairwise Alignment for Intertextual Relations’ (Horton et al. 2010), applies sequence alignment techniques to large collections of literary and historical texts over several languages. The PAIR system first indexes documents by breaking them into overlapping sequences of words, called ‘n-grams’ or ‘shingles,’ and then creates a database of the occurrences of each shingle in a given corpus. This index allows for the discovery of text reuse by looking for occurrences of the same word sequences shared between documents or within different parts of the same document. Through various tuning parameters, this technique can be made flexible enough to find sequence matches with minor differences in word order, missing words, orthographic variations, misrecognized characters, and other textual variants. The first part of this paper will thus examine the philosophy behind the PAIR approach, the algorithmic design, its implementation as an open-source Perl module, and its eventual application to a variety of tasks relevant to humanities and social science research.

While technology is a relatively new entrant to the domain of textual studies, it nonetheless offers scholars important tools for tracing the currents of literary text reuse. Indeed, this form of ‘intertextuality’ can be considered a specific case of the more general problem of sequence alignment; that is, the task of identifying regions of similarity shared by two strings or sequences, often thought of as the longest common substring problem. This technique is widely applied in the field bio-informatics, where it is used to identify repeated genetic sequences, and for plagiarism detection in texts or computer programs (Clough et al. 2002; Lyon et al. 2001; Bourdaillet & Ganascia 2007).

In creating PAIR, we attempted to adapt existing techniques to suit the particular needs of humanities scholarship. Many algorithms such as BLAST exist for identifying duplicated DNA, for example, but these tend to emphasize speed over completeness. In text analysis, our corpora are small enough and our interest deep enough that we can emphasize retrieving as many hits as possible, since computation time is cheap and literary scholars are not traditionally in a hurry. The scholar’s time, however, is quite valuable, so we want to return results that are of maximal interest, avoiding a preponderance of banal commonplaces such as ‘Your devoted, humble servant.’

PAIR is thus based on k-tuple heuristics that provide a suitable balance of efficiency against completeness. Applied to text data, this involves the generation of ‘shingles’ (or n-grams), which are overlapping sequences of words. Preprocessing, such as removal of function and short words and reduction of orthographic variants (accents, spelling changes, case, etc.), is performed during shingle generation. This has the effect of folding numerous shingles into one underlying form for matching purposes, thus eliminating minor textual variations, which makes matching more flexible or ‘fuzzy.’ It also somewhat reduces the overall number of unique shingles, which aids speed of search (Seo & Croft 2008).

Once identified, the shingles within a defined window surrounding the shared shingle in each document are compared and evaluated. PAIR allows the user to set criteria for the acceptability of matches, such as the minimum overlap in shingles between the two sets, the minimum length of a shared shingle sequence, or the maximum number of consecutive gaps allowed between matching sequences in either set. If the criteria are met, the match is expanded, examining wider contexts in each document, until the criteria are violated, at which point the match is terminated and recorded. Furthermore, user configurable parameters for match retention and expansion allow for the fine-tuning of results, which is particularly important given the often ‘noisy’ information space of many humanities text collections.

As a concrete use case for the PAIR approach outlined above, we examined the intertextual relationships of two of the 18th-century’s most important text collections: Denis Diderot and Jean d’Alembert’s philosophic war machine, the Encyclopédie (28 in-folio volumes, published from 1751 to 1772 – digital edition provided by the ARTFL Project, University of Chicago), and the Complete Works of Voltaire (over 100 volumes, data provided by the Voltaire Foundation, University of Oxford). Both resources represent monuments of Enlightenment thought as well as model digital humanities databases: highly curated collections of historically significant texts built using the open-source PhiloLogic search and analysis software developed at the University of Chicago. By comparing these two data sets using the PAIR sequence alignment approach, we can come to a better understanding of the multifaceted and at times problematic relationship – one of influence, anxiety, and intertextuality – between the French Enlightenment’s most emblematic writer and its most widely-read text.

Though initially enthusiastic about the promise and ambitions of Diderot and d’Alembert’s project, Voltaire nonetheless contributed only 45 articles to the enterprise (Voltaire 1987). This curious lack of engagement on the part of one of the leading philosophes in the most important publication of the mid-18th century has led many to conclude that this diminished role was due to philosophical differences between Voltaire and the Encyclopédie’s editors (Jacob 2006). Or, as Jonathan Israel has recently contended, as a leading figure of the ‘mainstream’ brand of enlightenment (essentially Lockean-Newtonian in nature), Voltaire was in no way eager to follow the Encyclopédie as it steadily moved in the direction of a more ‘Radical Enlightenment,’ fundamentally Spinozist-materialist in inspiration (Israel 2006). The validity of this interpretation, however – which relies on a superficial reading of Voltaire as ‘author’ in the Encyclopédie rather than as ‘authority’ – begins to break down once one is presented with the results of the PAIR comparison. Indeed, even a cursory examination of the more than 10,000 matching sequences between Voltaire’s Complete Works and the Encyclopédie, demonstrates the preponderance of Voltaire’s textual presence as an authority over and against his relatively restrained role as an encyclopedic author; a fact that without sequence alignment would have likely gone largely unnoticed or at least greatly underestimated.

Nowhere is this interaction between Voltaire and the Encyclopédie more pronounced than in his last, longest, and perhaps least known work, the Questions sur l’Encyclopédie (1770-74). Here, Voltaire revisits the ‘encyclopedic moment’ of the 1750s and recasts many of the concerns still relevant to he and his fellow philosophes some 20 years later. Adopting the same textual strategies as the encyclopédistes before him – indirect citation, playful borrowings, veiled references, etc. – Voltaire’s reassessment of the encyclopedic texts is a treasure-trove of hidden intertextual associations. Our exploration of the Questions using the PAIR system will thus, for the first time, give us a more general idea of the scope and scale of Voltaire’s prolonged engagement with the Encyclopédie and its contributors. Finally, this extended vision (or version) of Voltaire’s ‘encyclopedism’ – made possible through the application of the sequence alignment techniques outlined above – implies a certain continuity of Enlightenment thought (at least by its main protagonists) from 1750 to 1775. By way of this ‘intertextual’ continuity, we thus arrive at a more comprehensive understanding of the French Enlightenment than is currently reflected by recent attempts (Jacob 2006; Israel 2006) at dividing its participants into binomial ‘mainstream’ and ‘radical’ camps.