- A Pebble Smoothed by Tradition:Lines 607–61 of Beowulf as a Formulaic Set-piece
In lines 607–61 of Beowulf, just before the battle between the hero and the monster Grendel, the Danes and visiting Geats celebrate their comradeship in the great hall of Heorot.1 While venerable Hrothgar, king of the Danes, presides, Queen Wealhtheow, bedecked with gold, carries the ornamented cup of fellowship to each warrior in turn, old and young alike.2 The passage, which for convenience we will call "Wealhtheow's cup-bearing," is one of several depictions in Beowulf of the social happiness that Anglo-Saxon poetry often calls dream ("joy") and has been described as "the most detailed description we possess of the offering of the ceremonial drinking cup to an honored guest in early Germanic society" (Fulk, Bjork, and Niles 2008:155). But in contrast to Wealhtheow's later appearance in the poem (lines 1168b-231)—in which she thwarts Hrothgar's attempted adoption of Beowulf, promotes the king's nephew Hrothulf as a protector for her sons, and gives the legendary Brosing necklace to the hero—nothing much happens. Jeff Opland (1976:446–57) does not include the passage in his list of "joy in the hall" type-scenes.
Yet new computer-assisted "lexomic" methods of analysis3 show that these seemingly banal lines contain some of the highest concentrations of unusual lexical, metrical, grammatical, and formulaic features in Beowulf, and the overall distribution of vocabulary in the passage is so distinctive that it affects computer-assisted cluster analysis to a greater extent than any other similar-sized segment of the poem. In the discussion that follows, we introduce several techniques of lexomic analysis and explain how these approaches identify qualitative differences between lines 607–61 and the rest of the poem. We then show how all of these differences are best explained by positing that the passage has a source different from its surrounding textual matrix, a source that was most likely not a written text, but a traditional type-scene.4 A close reading of the lines in the light of recent approaches to the formula in Old English explains how the passage, so well polished by tradition that it preserved low-level linguistic features to almost the same degree as a written source would, could nevertheless have been easily adapted to other narrative contexts.
Lexomic methods combine computer-assisted statistical analyses with traditional literary approaches such as close reading, philological analysis, source study, and cultural interpretation. The specific techniques employed in this paper fall into two categories: hierarchical clustering, which uses the mathematical calculation of similarity and difference to create groups of texts or segments in which the members inside the group share more features than those outside, and rolling-window analysis, which produces a visual representation of the average frequency of particular words, letters or phrases within whole texts, allowing us to identify much smaller features within them.
In cluster analysis, we determine the relative frequencies of every word in a group of texts or text-segments, calculate the differences among these relative frequencies, square the resulting numbers, and uses the square-root of the sums of the differences to find what is called the "Euclidian distance" between each pair of segments. From this information, the Lexos software5 uses the free implementation of hierarchical, agglomerative clustering to group the segments, without pre-specifying the number of groups to be created, by clustering together those with the smallest overall differences in word frequencies (these have the most words in common) in a branching diagram, or dendrogram, that visually represents the relative similarities of the segments.6 Branches of the dendrogram are called clades, the similarity of which is represented by the vertical distance between the branch-points: the shorter the line, the more similar the clades. Because variations in the distribution of very common words (most often function words such as conjunctions, prepositions, and pronouns), more strongly influence dendrogram geometry than the presence or absence of rare words in particular segments, cluster analysis can often identify broad patterns of vocabulary distribution that are not always...