Giants were a serious problem for early modern British historians. For example, in a chapter titled “Whether it be likely that there were ever any Gyaunts inhabiting in this Isle or not” from his “Historical Description of the Island of Britain,” William Harrison offers a lengthy meditation on the historical plausibility of giants, arguing against the idea that the presence of fables in a nation’s historical record should irredeemably discredit that nation’s history. Harrison writes that to “some mens eares,” accounts that Britain was first settled by giants seem so implausible that they unfairly “suspect the credit of our whole historie, & reject it as a fable, unworthie to be read” (8). Giants were just one of a great many issues faced by early modern British historians—they also had to account for figures like Brutus of Troy, the purported grandson or great-grandson of Aeneas; Albina, a princess who, with fifty female companions, was set adrift as punishment for murder and then populated Britain by sleeping with giants; Albion, the son of Neptune and a giant himself; and even King Arthur, whose court contained the wizard Merlin and who was credited with victories (including several over giants—there is no shortage of giants in Britain’s early histories) that early modern writers feared were well beyond mortal abilities.

The presence of such materials in most early accounts of British³⁴ history made matters quite difficult for historians who wanted to trace their national records back to their origins, or who wished to include figures such as Brutus and Arthur, both of whom remained useful in constructing national identities, validating the antiquity of the British crown, and authorizing imperial claims. In discussing the challenges that they faced, many historians turned, just as Harrision did, to credit as a framework for evaluation. Harrison himself also complains that is a “pitie” that “incredible and fond fables” have brought Arthur and other early British rulers “out of credit” (1587, vol. 1, 120). Defending Arthur’s historicity, Richard Baker writes in his 1643 Chronicle of the Kings of England that many of the deeds attributed to Arthur might seem so “wonderful” that the king would be “reckoned amongst the Fabulous” were there not “enow true to give them credit” (5). Samuel Daniel is far from optimistic about the possibility of recovering Britain’s origins, writing in 1612 Historie of England he “cannot see” what credit can be cleared to “the account of above a thousand yeares from Brute to Casseuellaunus”;³⁵ he therefore leaves this period “on the booke to such as will bee creditors, according to the substance of their understanding” (2).³⁶ In discussing the twelfth-century historian Geoffrey of Monmouth—who was widely recognized as the source for most of the Arthur and Brutus traditions, and who was equally widely criticized for writing the sort of “incredible” fables that so troubled Harrison—William Tyrrell concludes in his 1696 General History of England that “the whole relation of the Actions of Brutus, and the Succession of all the Princes that followed him, do all depend upon the Credit of Geoffrey and the truth of his transaction” (7).

References to credit were far from the only way that early modern historians discussed their efforts—they also describe their work as following a thread through a labyrinth, attempting to bring light to darkness, or separating gold from dross³⁷—but this framing is nevertheless quite prominent, and clearly something that these scholars found useful in discussing their endeavors. As part of a larger project examining the ways that early modern historians made deliberate and very critically-engaged uses of materials like the “incredible” accounts of Brutus and Arthur, I have been investigating the discourse around credit in seventeenth-century British histories. In particular, I am trying to understand what made this credit framing such an appealing one—what are the affordances of “credit” in describing historical research? I am also trying to determine what this language can reveal about how early modern historians set boundaries around their discipline during a period when the pressures of empire and the disruptions of the English Civil War made national origin stories and figures like Arthur particularly necessary, while at the same time the historicity of Britain’s early traditions was being vigorously questioned.

Methods: Corpora, Models, Applications

While I have been using a fairly wide range of methods to pursue my inquiries into early modern historiographies, here I will focus on word embedding models, which are a machine-learning technique for studying relationships between words in large corpora of texts. These models are useful for my work because I am trying to understand the breadth of seventeenth-century historiographic discourse and because I am studying complex concepts that cannot be expressed in single terms or straightforward binaries. For instance, I am examining the distinctions that early modern writers made between proper “history” and forms of historical writing deemed less satisfactory, which might be termed: annals, chronicles,³⁸  legends, fables, romances, fictions, lies, or one of many other terms, each of which assumes different criteria for historiographic gatekeeping.

Word embedding models are able to identify connections between words (and, by proxy, the concepts represented by those words) across large corpora of texts. For this project, I am using several corpora I have built from the Early English Books Online–Text Creation Partnership collection. My primary corpus is a collection of 52 seventeenth-century histories, comprising 10.9 million words. I collected these texts by hand, based on my own research, as well as extensive searching in the EEBO-TCP database; the texts are primarily national histories but also include world histories, church histories, chorographies, king lists, and chronicles. My secondary corpus is a collection, following the same selection principles, of fifteenth- and sixteenth-century histories—this comprises 21 texts and 10.2 million words. I have also gathered three comparison corpora by randomly selecting seventeenth-century texts from EEBO-TCP with a modified version of a Python script developed by my colleague Bill Quinn; these have approximately 9 million words each. I trained models on each of the above corpora, using the “WordVectors” R package developed by Ben Schmidt and Jian Li, which is based on the original word2vec code developed by Mikolov et al.

There are many excellent and detailed explanations of word embedding and vector space models available—see, for example, work by Ben Schmidt, Ryan Heuser, Gabriel Recchia, and Lynn Cherny—the essentials are that word embedding models allow for a spatial understanding of the words in a corpus of text, making it possible to query both the proximity of words in vector space, and the directionality of relationships between words. Words that are close in vector space are those that the model predicts are likely to appear in similar contexts. For example, in many corpora, it would be expected that the closest words to “Tuesday” might be “Friday,” “Thursday,” “Wednesday” and so on—while words like “anglerfish,” “defenestration,” and “yogurt” would be less likely to appear in the same contexts and therefore less proximate in vector space. Word embedding models are also able to represent relationships between words in large numbers of dimensions,³⁹  and thus can allow for inquiry into the complex ways language intersects—for instance, “bride” and “woman” share a relationship as female-gendered words while “bride” and “groom” share a relationship as words related to marriage.

Assessing the Credit of Histories

Word embedding models are powerful because they allow you to understand a corpus as a whole—as Schmidt explains, they “try to ignore information about individual documents so that you can better understand the relationships between words.”⁴⁰ For example, a simple query looking at the words nearest to “credit” in vector space shows a close relationship between credit and language on evaluating historical reliability in my corpus of seventeenth-century histories. The closest words, with their cosine similarities,⁴¹  to “credit” are: reputation (0.729), historian (0.679), testimony (0.677), certainty (0.671), antiquity (0.668), relation (0.659), truth (0.653), pains⁴² (0.642), authour⁴³ (0.640), opinion (0.632), impartiality (0.632), authors (0.631), account (0.628), herein (0.625), critiques (0.622), fictions (0.614), shew (0.613), hector’s⁴⁴ (0.610), authours (0.610), romance (0.609), historians (0.609), deserves (0.603), histories (0.602), disparage (0.602), story (0.602).

The words with the highest cosine similarities to a particular term aren’t necessarily synonyms (in fact, antonyms are common) but they are terms that tend to be used in similar contexts. This list shows that the words that tend to be used with “credit” include ones that connect with discussions of history as a discipline and historical research as a process (historian, historians, authors, authour, authour, pains, antiquity); that describe different kinds of historical sources (testimony, relation, account, story, romance, fictions); and that evaluate historical sources or make arguments about the past (reputation, opinion, certainty, truth, shew, disparage, deserves, impartiality, critiques). Importantly, these are not just words connected with the credibility of histories and historians—rather, the discourse around “credit” is closely linked with contemporary debates about historical methodologies and the boundaries of history as a discipline.

These results are from a model I trained with 100 dimensions and a window size of 12, which means that the 12 words to the left and right of each word in the corpus are treated as relevant context when the model is predicting which words are likely to appear together.⁴⁵ I also trained models with several different parameters, partly to determine which provided the best results, but also because I was interested in seeing how my results might differ from one model to another. Word embedding models are non-deterministic, which means that I would still get different results even if I trained multiple models on the same corpus with the same parameters,⁴⁶ so I wanted to make sure I wasn’t basing all my conclusions on a single model. Trying the same query in a model that had a smaller window size (6 words), showed a similar investment in credibility, but some variation in the individual top terms: reputation (0.731), credite (0.657), testimony (0.654), opinion (0.652), honour (0.643), conduceth (0.631), antiquity (0.631), relation (0.628), account (0.621), certainty (0.620), disparage (0.608), pertinently (0.603), confidence (0.598), quote (0.595), warinesse (0.594), merit (0.593). There is a particularly strong thread of evaluation in this list, evident in terms like merit, warinesse, confidence, and pertinently.

The other models I tried all had some variety in the individual terms, but quite a lot of consistency in the overall type of associations, all of which connect with historians’ attempts to wrestle with the credibility of their sources, bolster the credit of their own work, or discredit the histories of their competitors. What my results with word2vec show that I couldn’t yet see from reading individual texts is how thoroughly the discourse of credit in this corpus is bound up in language around credibility. Even extending the terms in my results set to the top 250 (beyond which the cosine similarities drop below 0.5), still gives only language related to credit and credibility, without any terms connected to more commercial meanings of the word.

To determine whether it might just be the case that this particular corpus did not include any financial language, I tried adding the vectors for terms like “sell” and “buy” to the vector for “credit.”⁴⁷ These results showed language much more related to commerce than credibility; for example, here are some of the words closest to V(credit) + V(sell): sell (0.843), buy (0.719), credit (0.691), carry (0.679), purchase (0.648), get (0.645), overplus (0.635), gain (0.628), profit (0.625), chapmen (0.617), make (0.617), mony (0.611), money (0.609), creditours (0.606), goods (0.603). As already suggested by the presence of commercial terms in the discussions of historical credit quoted above (“on the books,” “transactions,” “creditors”), the more financial connotations of “credit” are very much present in this collection of seventeenth-century histories, and even seem to be part of the appeal of credit as an evaluative framework. It’s just that the purely financial meanings aren’t as central to the ways the word tends to be used in this corpus as the historiographic meanings are.⁴⁸

Is this a seventeenth-century preoccupation? Comparing with my corpus of earlier texts suggests that credit was equally connected with reputation and reliability, but the intensive focus on disciplinary debate and evaluation is somewhat less prominent: truth (0.735), credite (0.678), mind (0.629), estimation (0.629), praise (0.629), ghesse⁴⁹ (0.627), affoorded (0.626), report (0.624), vaine (0.623), dooings (0.617), opinion (0.610), disalow (0.609), reproach (0.608), shew (0.608). Broadening generically rather than chronologically also indicates that the discourse around credit is distinctive in my primary corpus. The sets of terms in my seventeenth-century EEBO–TCP comparison corpora vary considerably (as is to be expected, given that these are randomly-selected sets of texts from a very diverse collection), and thus can illustrate the broader range of associations for “credit” in this period. For example, the closest terms in one model primarily relate to finance and commercial exchanges: reputation (0.775), advantage (0.668), mony (0.668), money (0.667), acceptor (0.660), fund (0.657), profit (0.642), bills (0.632), adventure (0.615), borrower (0.606).⁵⁰ What these results tell me is that there is a strong association between credit and historiographic evaluation in my corpus of seventeenth-century histories and that this association is distinctive when compared with other texts from the period and with earlier histories.⁵¹

Great Credit, Some Credit, No Credit

Given that this collection of seventeenth-century histories has a measurable discursive link between “credit” and the language of historiographic assessment, what do the specifics of that language reveal? In particular, I am interested in identifying why this framework held such appeal for early modern historians and what that might show about the shapes that the discipline was taking during the tumultuous seventeenth century. I’m hoping to gain more insight into the questions asked by my larger project, which argues that early modern historians’ uses of medieval materials, even—especially—those that had been challenged as implausible or “fabulous,” were far from naïve or uncritical, but instead show a deep engagement with contemporary debates about how the boundaries of history should be drawn.

Aside from the obvious resonance between “credit” and “credibility,”⁵² my results suggest that the credit framing appealed to historians because it could be leveraged in many different kinds of claims about what, exactly, belonged in history. Assigning varying degrees of credit to historical texts allowed writers to establish rubrics for determining what should be considered reliable history, as well as what might not be reliable but could still be useful to the discerning historian. Importantly, “credit” as a framing allows for a sliding scale, in which sources could be marked as problematic but still allowed some credit. The flexibility and usefulness of credit can be seen in the characteristics of its closest neighbors in the model, which include positive, negative, and neutral language for evaluating sources—certainty, truth, impartiality, fictions, romance, story—as well as words for referencing (and evaluating) other scholars: historian, historians, author.

The breadth of evaluative moves that references to credit could authorize are also indicated by other methods for investigating my corpus, such as counting its common bigrams. Many of these are terms for assigning different amounts of credit: “great credit” (46), “no credit” (37), “more credit” (35), “little credit” (23), “good credit” (20), “much credit” (18), “any credit” (14), “some credit” (13), and “small credit” (11). These are just the most commonly-appearing evaluative terms appearing with “credit”—moving down the list also gives: small, best, greater, lesse, warrantable, greatest, most, suspicious, all, deserved, and sufficient. There were many ways that credit could be measured, and early modern historians seem to have used most of them.

Given the level of abstraction at which word2vec operates, I wanted to confirm my conclusions by looking at some examples from my corpus in which “credit” appears in proximity to the closely related terms from my initial query. While I found many useful and interesting passages when I turned back to direct searching within my corpus, I will focus here on a few results with “credit” and “author.”⁵³ Looking at the cases where “author” and “credit” appear within 20 words of each other demonstrates two things: first, seventeenth-century historians are near-unanimous in asserting that it is important to determine what should and should not be permitted within historical texts; second, these writers are equally consistent in their disagreement about precisely how historical boundaries should be drawn. A few examples can provide a glimpse of this divergence.

Despite the many historians who validated their materials simply by saying that they were found in authors of good credit, there was little consensus on who, exactly, could be accounted creditable. For example, Aylett Sammes writes that he has “not made use of any of the British Histories, because their credit in the World is but small” but instead grounds his assertions “upon the Authority of Greek and Roman Authors” (2). By contrast, James Tyrrell defends Britain’s own historians: “If our Saxon Annals were not a good Foundation for succeeding Historians to build upon, I desire to know what Credit the Antient Greek and Roman Authors can claim” (xxi). Similar debates are evident in writers discussing Scottish, Welsh, and Irish sources, as well as prominent figures from British History, such as Geoffrey of Monmouth, Nennius, and Bede.⁵⁴

Historians sometimes attempted to reconcile disagreements about historical credibility by validating suspect materials against multiple sources. For example, Peter Heylyn writes of the highly contentious Geoffrey of Monmouth⁵⁵ that, although he is “a Writer of no great credit with me, when he stands single by himself, yet when I finde him seconded and confirmed by others, I shall not brand a truth by the name of falshood, because he reports it” (16). However, even finding multiple sources in agreement cannot solve this conundrum if all of those sources derive from a single text. Edmund Campion writes that: “as for the multitude of writers that agree thereon, they are in effect but one writer, seeing the latest ever borrowed of the former, and they all of Cambrensis,⁵⁶ who affirmeth it not, but onely alleadgeth the received opinion of Irish Histories, yea rather in the foote of that Chapter, he seemeth to mistrust it, and posteth it over to the credit of his authors” (21–22). Cambrensis was one of a great many historians who tackled the problems of indeterminacy they faced by assigning questionable materials to the credit of their sources: for example, John Spottiswood says of Bede that “this and the other miracles he reporteth, I leave upon the credit of the writer, who is too lavish oftentimes in such fables and fictions” (15). Clearly wishing to be less lavish, Spottiswood acknowledges and then passes by Bede’s accounts of miracles, without including them in his history (which is not quite the same thing as ignoring them altogether).

In fact, this deep concern with what can be considered reliable history should not be taken as evidence that all early modern historians defined their discipline as having space only for historically verifiable materials. References to credit were also deployed as an authorizing move to justify including materials that were recognized as suspect, such as when Tyrrell writes of Britain’s early history that “we are forced in this Period, not only to make use of Authors who lived long after the Things they treat of were done, but also are otherwise of no great Credit; such as Nennius, and Geoffery of Monmouth, whom we sometimes make use of for want of those of better Authority” (114). In fact, Tyrrell goes on to write that, because there remains “nothing more ancient than Bede, and the Saxon Chronicle” he has chosen to include this material “almost entire” in his own history (114). For some authors, at least, history with bad credit was better than no history at all.

The Rules of Horsemanship and History

As I develop this project, I will continue building out my corpora, so that I can be confident my findings are not impacted by any idiosyncrasies in the data collection. I will also be systematizing the results of my basic queries and investigating more sophisticated ways of working with embedding models. Ben Schmidt has done some excellent work with examining binaries in vector space and I think that his methods have a lot of potential for the divisions of credibility—and boundaries between histories and fictions—that my authors are both defending and subverting. Even the preliminary forays I’ve been discussing here have already provided me with a clearer sense of the strong association between credit and historiographic evaluation in seventeenth-century British histories, as well as the flexibility and utility of the credit framework for historians’ competing claims about their discipline.

In all of this work, I’ve found that large-scale analysis methods like word2vec are useful not only because they allow me to ask comprehensive questions about my texts taken as a whole, or because they offer new ways to understand the histories I’ve been working on for years (although, I value both of these capabilities quite a lot). These methods are also useful because they bring a degree of unpredictability to the work that I do—which sometimes is just part of the pleasure of research and sometimes can provide an effective method for preventing me from only ever finding the things I have gone looking for.

I’d like to close with one such unpredictable result, which I found by querying V(credit) + V(impartiality). Adding the vector for “impartiality” causes a shift toward language for assigning credibility: impartiality (0.935), credit (0.865), reputation (0.720), critiques (0.687), candour (0.677), candor (0.674), truth (0.669), unmask (0.659), credibility (0.658), exactness (0.656). Since I had already identified a close connection between credit and evaluation, and since impartiality was one of the mechanisms by which historical sources were evaluated, these results were not particularly surprising. However, I was not expecting anything like the twenty-second closest term in the results from this query: “horsemanship,” with a cosine similarity of 0.626. Searching through my corpus to explain this unexpected proximity (in addition to a few less interesting results), I located a section in Thomas Fuller’s 1655 Church-history of Britain that considered whether a particularly troublesome passage related to Saint Augustine was original to Bede. Fuller notes that this passage might have been added “by some in after-Ages,” an effort that only succeeds in making matters worse because “this Passage checketh the Pen of Bede in the full Speed thereof (no lesse against the Rules of History, then of Horsemanship)” (64).

Horsemanship aside, that Fuller was able to speak so confidently about the rules of history in 1655 is notable, because all of the research I have done indicates that there was the farthest thing from consensus on what, precisely, the “rules of history” might be. However, as my work has also shown, historians were very deeply invested in attempting to establish those rules. In their efforts, they wrote histories that weighed the credit of giants and of kings, and though they did not often agree with each other, their words still demonstrate how flexible and effective credit could be as a scale for histories.

Sources

Baker, Richard. A Chronicle of the Kings of England. London, 1643.

Campion, Edmund. A Historie of Ireland. Collected in Two Histories of Ireland. Ed. James Ware. Dublin, 1633.

Cox, Richard. Hibernia Anglicana: Or the History of Ireland from the Conquest Thereof by the English to this Present Time. London, 1689.

“credit” OED Online. Oxford University Press, June 2019. Web. 25 June 2019.

Daniel, Samuel. The Collection of the Historie of England. London, 1618.

Enderbie, Percy. Cambria Triumphans. London, 1661.

Fuller, Thomas. The Church-history of Britain. London, 1655

Harrison, William. “An Historical Description of the Island of Britain.” Chronicles of England, Scotland, and Ireland. Ed. Raphael Holinshed. Vol. 1. London, 1577.

Heylyn, Peter. Examen Historicum, or, A Discovery and Examination of the Mistakes, Falsities and Defects in Some Modern Histories. London, 1659

Heuser, Ryan. “Word Vectors in the Eighteenth Century, Episode 2: Methods.” Virtue and the Virtual, 1 June 2016. http://ryanheuser.org/word-vectors-2/

Sammes, Aylett. Britannia Antiqua Illustrata. London, 1676.

Schmidt, Ben. “Vector Space Models for the Digital Humanities.” Ben’s Bookworm Blog, 25 Oct. 2015. http://bookworm.benschmidt.org/posts/2015-10-25-Word-Embeddings.html

Speed, John. The History of Great Britaine. London, 1611.

Spottiswood, John. The History of the Church of Scotland. London, 1655.

Stanihurst, Richard. “The Historie of Ireland.” Chronicles of England, Scotland, and Ireland. Ed. Raphael Holinshed. Vol. 3. London, 1577.

Temple, William. An Introduction to the History of England. London: Richard Simpson, 1695

Tyrrell, James. The General History of England. London, 1696.

Word Vectors for the Thoughtful Humanist has been made possible in part by a major grant from the National Endowment for the Humanities: Exploring the human endeavor. Any views, findings, conclusions, or recommendations expressed in this project, do not necessarily represent those of the National Endowment for the Humanities.

Notes