James Gawley, Caitlin Diddams, Elizabeth Hunter, Tessa Little
The discovery of allusions is one of the enduring vocations of the philologist. It is also the subject of continuous critical debate. Julia Kristeva coined the term ‘intertextuality’ to describe connections that are drawn in the mind of the reader, regardless of the author’s intent (1969). Gian Biagio Conte asserted that allusion is a message from the author to the reader that a trained philologist can correctly identify most of the time (1986). Stephen Hinds showed that a reader is always imagining the intention of the author, and an allusion is convincing when we can imagine the author’s intent to reference a particular passage (1996). Neil Coffee set out to catalogue which formal features of similarity can identify convincing allusions (2012).
In the present study, we set out to discover whether the formal similarity features discussed in the literature can predict whether a potential allusion will be convincing. Taking Vergil’s allusions to Homer as a case study, we performed a survey of the literature and identified the similarities that philologists use in order to argue for the presence of an allusion. We grouped these similarities into four categories: shared language, subject matter, sound patterns, and syntax. Then we looked for those features in parallels drawn from the appendix of G.N. Knauer’s Die Aeneis und Homer. Our goal was to determine how often those features are present in known similarities between Vergil and Homer, and whether they can be used to predict a reader’s judgment about a possible allusion. We limited the scope of our test to the first book of the Aeneid and included all connections with either the Iliad or the Odyssey. This data set amounts to over 1,000 entries in Knauer’s appendix.
We also categorized each textual parallel on an impact scale from one to five. In order to determine the impact, each annotator performed a thought experiment. We asked ourselves two questions: first, how strongly did we believe that Vergil intended us, as readers, to recall the specific passage of Homer cited in Knauer’s index? Second, how likely did we think it was that Vergil’s Roman audience would make the connection? A rank of five represented complete confidence in both the author’s intent and the ancient audience’s perception. Ideally, using this thought experiment each annotator would assign the same rank to the same parallel without consulting the others. In order to test this, 10% of parallels were assessed by all four annotators in a blind study.
We used a linear regression model to determine whether topic, sound, syntax, and word similarity could be used to predict the impact ranking of a parallel. We found that the correlation between ranking and feature was extremely significant for words and topic, and somewhat significant for sound. There was a correlation between syntax and ranking, but it failed to meet the cutoff for statistical significance. However, these features alone were not able to fully account for the impact rankings. We measured inter-annotator agreement with Robinson’s method to determine whether we accurately predicted each other’s rankings. We found that shared impact rankings had a Robinson’s score significantly higher than would be expected from chance.
Strong correlation between formal features and impact ratings means that allusion is partly signaled by formal similarities in the parallel text. The fact that the formal features are not capable of precisely predicting an annotator’s impact assessment means that an allusion is also partly signaled by something beyond formal similarity. The fact that annotators are able to predict each other’s impact rankings suggests that whatever is not conveyed by the formal features of similarity is nevertheless received consistently in the mind of the expert reader. Our model provides quantitative support for the use of these features in arguments surrounding potential allusions. Nevertheless, normal similarity must combine with impact to cross some threshold of plausibility before an intertext should be accepted as a deliberate allusion.