Patrick J. Burns
In his study of Latin poetic diction Unpoetische Wörter, Bertil Axelson presents a compelling argument that Horace uses a particularly prosaic vocabulary in the Odes (Axelson 1945: 108–11; cf. Wilkinson 1959; McDermott 1982). So convinced is Axelson of the prosaicness of the closing lines of Odes 4.9 that he prints these verses “without line breaks to mark its stylistic character, which may be safely called ordinary prose (ordinären Prosa).” Since publication, Unpoetische Wörter has produced a productive debate, often critical of Axelson's work, on the use of Latin poetic diction (e.g. Ernout 1947; Williams 1968; Watson 1985; Lyne 1989). Ernout (1947, 68) expresses the critique well, namely that Axelson’s “simple listing [of word frequencies], purely statistical and comparative in nature, without an examination of usage in context" is insufficient for judging poetic style.
In this paper, following recent work in computational literary study (Underwood 2019; Piper 2017; Piper 2018; So 2017; also, McCarty 2004), I revisit Axelson’s argument about prosaism in Odes 4.9 precisely with an “examination of usage in context” by recasting the study in new methodological terms, namely by leveraging digital text analysis to construct a literary model of “poeticness” in Latin literature. The method for this paper is as follows: 1. a tagged corpus of Latin verse and prose is used to model the probability of words occurring in either category; 2. each word in a given work is assigned a poeticness weight; and 3. these weights are plotted in narrative space, that is they are plotted such that the weight is given along the y-axis and the position of the word on the x-axis (i.e., first word at x=1, second at x=2, etc.). An example of the resulting visualization for Odes 4.9 can be seen at this link: http://bit.ly/scs-2020-unpoetic-abstract-img-1. Using a literary model to define poeticness allows us to “describe continua instead of sorting everything into discrete categories” (Underwood 2019: 20). What if rather than label a word like uxor as unpoetic, we instead label it as 10% poetic and 90% unpoetic? Modeling literary vocabulary as falling along a spectrum of poeticness leads to both a more nuanced measurement of individual words—one greatly assisted by computational power and storage—and a measurement that can be extended more productively to surrounding context.
I begin with a reading of Odes 4.9 that concludes that Axelson’s readerly instinct is correct and that the diction in the second half of this poem is, according to the model, comparatively unpoetic. In the remainder of the paper, I move towards increasingly distant readings of Horace, from the other poems in Odes 4 to the Odes overall to his entire multigeneric corpus, as an exploratory analysis of quantifiable moments of unpoeticness in his works.
In the wake of recent debate on computational literary studies (e.g. Da 2019 and responses in the corresponding Critical Inquiry forum), this paper demonstrates how a computational approach to a Latin literary critical problem can help, as one critic writes (Algee-Hewitt 2019), “[present] new kinds of evidence, often invisible to even the closest reader, alongside carefully considered theoretical arguments, both working in tandem to produce new critical work.” Accordingly, I revisit by way of conclusion Gian Biagio Conte’s distinction between “exemplary” models and code “models” (Conte 1986). Much large-scale digital literary critical work on Latin poetry (e.g. Coffee et al. 2012) has focused on diction matching and the search for explicit allusive activity, what Conte sees as the exemplary model. This paper uses a systematic valuation of poeticness to gesture toward the “poetic memory” aspect of Conte’s work, that is the code model, or a "series of phenomena that could be otherwise registered only piecemeal, in uncoordinated, discrete details," expanding the range of Latin literary critical problems that can be addressed through computational analysis.