FastText Models 2022

Описание: Natalie Grills, supervised by Dr. Antonioli

The aim of this study is to evaluate the accuracy and effectiveness of word-embeddings in comparing thematic relationships between Baudelaire and other Romantic French poets. Two FastText skip-gram models are trained for comparison: one by the complete poetic works of Charles Baudelaire and the other by the combined poetic works of Marceline Desbordes-Valmore, Victor Hugo, and Alphose de Lamartine. Both the nearest thematic neighbors and the cosine similarities amongst thematic words provide information that suggest differences between Baudelaire and the other Romantic French poets. Overall, the information produced by the trained FastText models alone does not produce a comprehensive comparison of Baudelaire’s poems to those of Desbordes-Valmore, Hugo, and Lamartine, but the use of word-embeddings in comparative literary analysis has potential to indicate themes and literary tendencies within the texts of interest prior to an in-depth analysis.