Ancient Greek Word Vectors

Word vectors are a particularly useful tool for the investigation of ancient discourses. They allow us to explore associations of concepts that pervade either an entire culture or a group of people. Recently, Kozlowski, Taddy, and Evans (2019) used word vectors and embedding to study the semantics of class structure in 20th-century American cultures, a publication that shows exciting results. Using diachronic word embeddings trained on large corpora of books, the authors traced how the semantic associations of class-related terms (like “rich,” “poor,” “working class,” or “elite”) evolved over time, revealing shifting ideological alignments in American discourse. They show, for instance, that the term “rich” became increasingly associated with meritocratic ideals, while the term “poor” shifted from connotations of moral failing to structural disadvantage. Their method quantifies how class-related concepts are embedded within cultural narratives and offers a replicable model for studying cultural change through language.

Similar approaches can be applied to the ancient world, where semantic change may not always be traceable diachronically in the same way, but where embeddings trained on large, representative corpora can still illuminate the conceptual proximity of key terms. For example, we can use word vectors to investigate how doulos (“slave”) and gynē (“woman”) are positioned relative to other concepts across genres or authors—whether they cluster with terms associated with domesticity, morality, subjugation, or agency. Such analysis can reveal implicit ideological structures in texts, highlight differences in how gender and slavery were framed in tragedy versus historiography or philosophy, and help us reconstruct ancient worldviews that are often subtly embedded in language use. This computational lens complements traditional philological interpretation by offering new, scalable ways to analyze cultural semantics in antiquity.

For this porpuse, I have trained Ancient Greek word embeddings using the Fasttext library and I have made these embeddings available in Zenodo as Ancient Greek Fasttext Word Embeddings. An article on the topic is in preparation.