Chenye (Peter) Shi
In this project, I employ an artificial neural network (ANN) to analyze the relationship between dictional choices and emotional expressions in Ovid’s epistolary poems (Tristia, Epistulae Ex Ponto, and Heroides). Using sentiments as the contextualizing criteria, I perform a comprehensive analysis of the semantic and stylistic changes in Ovid’s exilic poetry as compared to his early works.
Ovid’s mastery of language has been a central topic in the study of his poetic genius. Previous scholars have discussed in-depth the poet’s visual language (Hardie 2002), originality of word usages (Nagle 1980), revolutionary influence on the elegiac genre (Conte 1994), and his shift in vocabulary preference before and after exile (Claasen 1999). These studies usually base their discussions on preselected words that are considered thematically representative. Word frequency, when included in the analyses, served only as a secondary proof of the chosen vocabulary as part of the poet’s repertoire.
The growth of computational technologies offers new potential. Computational linguistics and machine learning can contextualize our readings of Ovidian poetry: they provide an analytical framework that is able to include the whole corpus and examine the function and significance of each word instantaneously. The use of information technology is not new in classics (McGillivray 2014). Online databases have both changed the experience of learning Latin and offered convenience to students and researchers alike. However, the ubiquitous use of digital tools has not significantly influenced literary analyses. Other than word frequencies, computational approaches have not significantly advanced our understanding of Latin literature despite the vast amount of materials easily accessible to the public.
This project aims at a further incorporation of computational technologies and traditional literary analysis. Using the Tristia as the training dataset, I first tag each sentence according to Robert Plutchik’s fine-grained “Wheel of Emotions” (thirty-two emotions in total, with eight basic emotions and twenty-four secondary or combinational ones). These data are processed with the artificial neural network (ANN) technique to develop an analytical model that detects the emotions associated with the words based on the words’ meanings (roots), inflections, and positions in the sentences.
The model will then be tested for its accuracy using the Ex Ponto as the test set. This evaluation process also yields a quantitative description of Ovid’s emotional changes throughout his exile. Furthermore, I apply the same model to the Heroides. These fictional letters' format and gloomy content make them comparable to the exile poetry. The computational model, however, reveals more thoroughly the subtle differences of word choice and sentiments between Ovid’s imagined writing and that originating in his personal experience.
Finally, I discuss this project’s applicational value. Compared with traditional research, the machine learning approach has the advantage of being exhaustive, quantifiable, and visually representable. It offers a comprehensive picture of the stylistic, dictional, and expressional features of the author. Moreover, the fine-grained tagging system, originally employed in cognitive science, lowers the barriers between psychological and literary discussions of emotions. Ultimately, the quantitative model also offers a more objective means to evaluate the faithfulness of translations by comparing results from similar sentiment analyses performed on translations in other languages.