In a field as mature as Beowulf studies, the discovery of new data is a rare event. Possessed of a two-hundred-year critical history written by generations of sharp-eyed scholars, the text of Beowulf would seem to have yielded all extractable information, and indeed, the larger recent debates in the field have been over the interpretation of the evidence already accumulated. The possibility of discovering new evidence has not been much entertained, and reasonably so: the chance that a trove of Anglo-Saxon manuscripts relating to Beowulf could come to light is infinitesimal, and even the most remarkable archeological discoveries
1
are unlikely to be directly applicable to the text of the poem itself. Furthermore, in the past three decades philological and linguistic methods, previously so productive, have failed to uncover many new facts,
2
suggesting, perhaps, that their capabilities are exhausted or that Beowulf is simply mined out. But just as advances in technology allow valuable minerals to be recovered from abandoned tailings heaps, innovations in information processing may enable us to extract new data from old texts: in both cases the technology allows us concentrate that which was previously too diffuse to be useful.
In this study we present the results of our use of âlexomicâ
3
methods to analyze the text of Beowulf. These approaches, which use computer-assisted statistical analyses to augment traditional philological and literary techniques, have identified patterns of vocabulary distribution, meter and orthography that, when examined in light of our previous knowledge of the poem, shed new light on Beowulfâs structure, ontogeny and evolution. In this first half of a two-part investigation, we introduce methods of cluster analysis that allow us to calculate and represent visually the relative similarity of the vocabulary distributions of segments of a text. Previous research has shown that, in other Anglo-Saxon texts, these patterns of vocabulary distribution are often correlated with sources, structures and histories. Here, we show how the methods can be adapted to the particular problems posed by Beowulf and that, when combined with more traditional approaches to the poem, cluster analyses can identify relationships of similarity and difference between various sections of Beowulf that might otherwise be too subtle or diffuse to be detected by the unaided eye and mind.
4
Since the earliest days of Beowulf scholarship researchers have sought to use the poem in support of arguments about other questions (some only tangentially related to Beowulf itself).
5
This tendentious approach has regularly led to significant data being misconstrued or misunderstood and to the insights of previous scholars being neglected.
6
Even in more recent studies, in which Beowulf is not used for explicitly nationalistic or other overtly partisan political purposes, observation has often been subordinated to thesis, with the predictable result that some data are under- and others over-emphasized, making it difficult for a logical synthesis to evolve. We seek to avoid this problemâat least as long as possibleâby describing the tools and methods and then presenting the data acquired by them without trying to embed interpretive or synthetic conclusions in the discussion (these we will leave for the final sections of each part of the investigation).
7
The goal of this structure is to allow interested researchers to adopt those methodologies that they find congenial without by so doing being manipulated, rhetorically or logically, into pre-accepting a specific position on later interpretations.
8
To avoid having each piece of new evidence become, at the very moment of its production, entangled in the thicket of the major scholarly disputes, parts of this paper operate at a higher level of abstraction than has been traditional in Beowulf studies. We also, for the most part, reserve until the interpretative section of the paper references to the rich scholarship to be found on nearly every line of the poem. Readers should not infer from this arrangement any rejection of the painstaking, erudite analysis of the details of the poem, the accumulation of which is an intellectual monument. We are merely trying to make our own argument as clear, unencumbered and straightforward as possible, emphasizing the findings and the methods that have produced them.
Notes
- 1.
The National Endowment for the Humanities helped sponsor this research with three grants: NEH HD-50300-08, âPattern Recognition through Computational Stylistics: Old English and Beyondâ 2008-2009; NEH PR-50112011, âLexomic Tools and Methods for Textual Analysis: Providing Deep Access to Digitized Textsâ 2011-13; and NEH HD-228732-15. National Endowment for the Humanities, Digital Humanities Start-up grant: âEasing Entry and Improving Access to Computer-Assisted Text Analysis for the Humanities,â 2015-17. Any views, findings, conclusions, or recommendations expressed in this article do not necessarily reflect those of the National Endowment for the Humanities.
Â
- 2.
In great part due to their being perfected over multiple scholarly generations.
Â
- 3.
Derived by analogy from âgenomics,â the term âlexomicsâ was coined by Betsey Dexter Dyer in 2002 and first appeared in print in Genome Technology 1.27 (2002).
Â
- 4.
Clustering methods do have a weakness in that they, by necessity, require us to compare reasonably large segments of a text, limiting the sizes of features that we can resolve. Additionally, cluster methods detect relationships that are discrete rather than continuous, thus making the detection and selection of ideal segment boundaries a matter of trial and error. We overcome some of these limitations in the follow-up to this study, Beowulf Unlocked II: The Evidence of Rolling Window Analysis [forthcoming], in which we use rolling window analysis to allow us to resolve features that are both continuous and smaller than those identified by cluster analysis. Rolling window analysis enables us to depict the distribution of metrical, linguistic and orthographic features more accurately, identifying concentrations and allowing us to correlate them with the results of both cluster analysis and more traditional approaches. Combined, the two lexomic techniques compensate for each otherâs deficiencies and significantly enhance the knowledge already gained from traditional approaches.
Â
- 5.
See, among others, Allen J. Frantzen, Desire for Origins: New Language, Old English and Teaching the Tradition. New Brunswick: Rutgers University Press, 1990, 168â200; T. A. Shippey and Andreas Haarder, Beowulf: The Critical Heritage. London: Routledge 1998, 1â74 and passim; T. A. Shippey, The Shadow-Walkers: Jacob Grimmâs Mythology of the Monstrous. Tempe: Arizona Center for Medieval and Renaissance Texts and Studies, 2005; Joep Leerssen, National Thought in Europe: A Cultural History. Amsterdam: Amsterdam University Press, 2006; Andrew Wawn, Graham Johnson and John Walter, Constructing Nations, Reconstructing Myth: Essay in Honour of T. A. Shippey. Turnhout: Brepols, 2007.
Â
- 6.
Theodore Anderssonâs assessment that âthe institutional memory in Beowulf studies is about an even centuryâ may even be optimistic. Theodore Andersson, âSources and Analogues,â in Robert E. Bjork and John D. Niles, eds. A Beowulf Handbook. Lincoln, NE: University of Nebraska Press, 1998, 129.
Â
- 7.
In so doing we will try to live up to the example of Levin SchĂŒcking, who in 1905 showed that it was possible to present evidence both for and against a preferred argument and to be fair-minded when evaluating it and the arguments of others. Levin SchĂŒcking, Beowulfs RĂŒckkehr: eine kritische Studie. Studien zur englischen Philologie 21. Halle: Max Niemeyer, 1905.
Â
- 8.
We are trying to avoid presenting our argument as a âcase lawyerâs plea,â instead seeking to set out the evidence âin clear detail for a sober and dispassionate judgeâ; Johan Gerritsen, âHave with you to Lexington! The Beowulf Manuscript and Beowulf,â in In Other Words, Transcultural Studies in Philology, Translation and Lexicography Presented to Hans Heinrich Meier on the Occasion of his Sixty-Fifth Birthday, ed. J. Mackenzie and R. Todd, Dordrecht: Foris, 1989, 15â34 at 15.
Â