A Word Embedding Based Approach for Emending Books 3 and 4 of Propertius' Elegiae

Abstract

The Elegiae of the Roman poet Sextus Propertius have fascinated classicists for centuries.However, the manuscripts of the Elegiae face a staggering number of errors, and scholars’ ability to study the text is in many ways hindered in many places due to an inability to determine the optimal way to correct the text. The quantitative tools used to study similar problems (authorship verification) have been primarily applied to English corpora, and there has been little work so far applied to Latin texts. This work harnesses word embeddings (in particular, theWord2Vec algorithm developed by Mikolov et al.) to develop a scoring model for ranking the various emendations proposed by scholars to different lines of the text. We measure cosine distance between a proposed emendation word and the sum of the context words for the line in which it resides. We apply our model to 71 lines across Books 3 and 4 of the Elegiae and, given a list of emendations proposed by scholars to each line, determine the most probable emendation. We conclude by comparing samples of the scores our model gives to each emendation with modern theories about how Propertius’ work can best be emended.