Murder in the Archive

A confession: as someone who works with digitized texts, I have a romanticized view of physical archives. There is the sensory experience of an archive—that remarkable combination of decay and order. The worrying smell of decomposing paper and the reassuring sight of tidy gray boxes. There is the intellectual experience of confronting texts within an intentionally assembled collection.

Consider the Thomason Tracts. This collection of about 22,000 texts was assembled by the London bookseller, George Thomason, between 1640 and 1661. The texts include pamphlets, newspapers, books, plays, and other materials. It is the largest collection of texts from one of the most turbulent periods of English history (Mendle 2009).

The nineteenth-century historian, Thomas Carlyle, called the collection “the most valuable set of documents connected with English history.” He believed that the Thomason Tracts held “the whole secret of the seventeenth century” (Great Britain 1850, 274). Uncovering this secret is essential for those scholars who are trying to understand political thinkers such as Thomas Hobbes in their English contexts.

The physical tracts are housed in the British Library. However, I use this collection in the way most contemporary scholars do: through the Early English Books Online (EEBO) portal. There is something miraculous about this digital access. Yet it abandons the collection as a collection. What Thomason so patiently assembled and catalogued is now “just another incomplete pile of books” (Mendle 2009).

Encountering the texts in this way foregoes all the pleasures of the physical archive: no skirmishes with prickly but efficient librarians; no nods of recognition to other scholars working with the same collection; no walks through the very London streets on which the authors of these texts lived. Working with the EEBO is convenient and efficient, but it is not romantic.

When I start doing things with the texts, the exercise shifts from the mundane to the murderous. To get a sense of how English political discourse was changing during the time in which Hobbes was thinking about and writing Leviathan (1651), Jacqueline Basu and I used a computational approach called topic modeling (Basu and McQueen forthcoming). At the most general level, this form of “macroanalysis” helps to surface themes in a corpus of texts by identifying recurring patterns of word clusters (Jockers 2013, 2014).

These approaches begin by mutilating the corpus. We set aside all of the images contained in the texts. This includes some of the most evocative visual products of the time, such as the topsy-turvy woodcut that graces the 1646 ballad, “The World Turned Upside Down” (T. J. 1647).

We then assaulted the texts themselves. We standardized their gloriously irregular early modern (or “earlie moderne” or “erly moderne”) spelling. We eliminated punctuation; removed commonly used words (e.g., “the” and “and”); and stemmed the corpus, converting words with the same stem into a single word (e.g., “political” and “politics” became “politic”). The topic model then treated each document as a “bag of words,” without regard for word order (Jockers 2014, 137). The result was something monstrous—and certainly illegible—to a human reader.

At this point, we were a long way from that glorious and fragile collection in the British Library. There were losses, to be sure—the loss of a physical encounter with the archive, the loss of the visual features of the texts, the loss of their peculiarities of spelling and punctuation, and eventually the loss of all linguistic coherence.

So, why do it? Why assault the archive? The simple answer is that we can uncover patterns that otherwise might elude us. We can capture the thematic content of thousands of texts and see which themes are especially salient over the 21-year life of the corpus. Not surprising, themes about the relationship between the King and Parliament and the course of the civil wars are particularly prominent. More surprising, comedies and comic themes also are pronounced. Perhaps amid so much bloodshed, laughter was precious.

Why assault the archive? The simple answer is that we can uncover patterns that otherwise might elude us.

We also can see how the prevalence of these themes changes over time. This offers suggestive contextual evidence that may answer thorny textual puzzles. For instance, Hobbes tells us that he wrote Leviathan in response to political and religious discourse in England (Hobbes 1839, xcii). He also wrote it in English, with the hope that it would be read by his countrymen and taught in the universities (Hobbes 2012 [1651], 1140).

It seems reasonable to assume that some features of Leviathan speak to the English public discourse of the late 1640s—the period in which Hobbes was thinking about and writing the work. For example, in Leviathan, he adds an entirely new account of the Christian doctrine of the Trinity (Hobbes 2012 [1651], 776–78). He had not addressed this question in his earlier political works, Elements of Law (1640) and On the Citizen (1642/1647).

His account of the Trinity is heterodox, to say the least. Hobbes manages to imply, for instance, that Moses is a member of the Trinity (Hobbes 2012 [1651], 776). The account exposed him to criticism. The Presbyterian critic, George Lawson (1657, 161), claimed that Hobbes’s doctrine of the Trinity was “blasphemous” and “deserve[d] no answer but detestation.” Furthermore, nothing in Hobbes’s political theory required him to weigh in on the Trinity. So, why did he bother? The fact that the Trinity was “trending” in popular English discourse in the late 1640s offers suggestive—although hardly conclusive—evidence. Perhaps Hobbes was speaking to debates on the ground.

More broadly, computational approaches have the potential to expand the ambitions of contextualist work in the history of political thought. The opening gambit of contextualists in the 1960s and 1970s was to shift the history of political thought away from a conception of the Western canon as the context (e.g., Skinner 1969). For instance, reading John Locke’s Two Treatises of Government (1690) as if it were a response to Hobbes, rather than to the politics of the Exclusion Crisis, is deeply misleading (Dunn 1969; Laslett 1960). To understand what political thinkers meant by any given argument or utterance requires a deep knowledge of the public discourse of their time.

So, context matters. But what counts as context? Although the question remains unsettled, there is little doubt that the contextualist turn expanded the range of documents that might count—not only canonical works of political thought but also pamphlets and sermons, diary entries and plays, and illustrations and frontispieces. For those scholars working on twentieth- and twenty-first-century political thought, contextual evidence also may include tapes, films, memes, and social media posts.

However, any honest contextualist must admit that the sheer volume of material is overwhelming. Faced with such a large and varied archive, how can our expertise possibly expand to meet it? We run up against predictable cognitive limits (Blaydes, Grimmer, and McQueen 2018). Without realizing it, we may focus on the items that seem familiar or that confirm our intuitions. We will pattern the archive according to our priors and, in so doing, we will reduce the chance of finding the unexpected.

I view the role of macroanalysis with digitized archives as one way to resist these tendencies. Where rich digitized archives are available, computational approaches allow us to retain an expansive answer to the question of what might count as context, to allow new patterns to present themselves, and to expand the context beyond our current expertise (London 2016). Perhaps most of all, these approaches preserve the possibility that the archive will surprise us.