This is a talk I gave with Stephen Sansom at the Intersections graduate student conference on November 6, 2015. We presented the results of some investigation into automatic detection of sound features in ancient poetry. We examined frequency distributions of letters (word-initial, and within-word), using letters as a proxy for sounds (mapping φ/π to the same bucket, similarly κ/χ). We also tried using entropy as a measure of alliteration (low word-initial letter entropy ⇒ higher alliteration) with interesting discoveries. Our text corpus consisted of the Iliad, the Odyssey, the Theogony, the Works and Days, and the so-called Homeric Hymns.

Abstract

The sound of poetry has recently gained much scholarly attention, whether in the burgeoning fields of Sound Studies and Auditory Culture (e.g. Sterne 2012, Bull and Black 2003) or phenomenological and formal approaches to the interpretation of poetry (Jarvis 2011, Bernstein 1998). Few, however, have attempted to devise and use digital tools to ‘read’ sound in texts. In this paper, we present computer techniques to identify textual sound features, including alliteration and accentuation. We demonstrate these techniques by application to poetry from Archaic Greece (8th–6th century BCE), particularly that of Hesiod. Not only do these approaches assist in identifying sound features in texts but they raise questions of formalism, historicity, selectivity, definitions of sound and figures of sound, and the use of technology in reading literature.

Here's a sample of the entropy-based alliteration output on Works and Days. Shorter bars mean less entropy and more alliteration.