NLP for Literary Latin and Ancient Greek Texts

This presentation introduces the Daidalos Project, an interdisciplinary initiative at Humboldt-Universität zu Berlin that bridges classical philology, corpus linguistics, and digital humanities through the application of natural language processing (NLP) to Latin and Ancient Greek literary texts. The project supports digital literary research by evaluating and adapting existing NLP models for ancient languages, with a focus on methodological triangulation, participatory research, and user-driven tool development. Key research questions addressed include the semantic and stylistic differences in epitaphs across Thucydides, Lysias, and Plato; the distinction between authentic invectives and declamationes in Ps.-Sallustius; and the identification of information gaps in ancient historiography. The project employs a range of NLP techniques—including word embeddings, named entity recognition (NER), morphological analysis, part-of-speech tagging, and sentiment analysis—using both rule-based and neural network approaches, including state-of-the-art models like XLM-Roberta and LatinCy. A central innovation lies in the use of research tandems, which bring together domain experts and non-specialists in a co-creation process to develop and refine NLP pipelines. The workflow emphasizes transparency, reproducibility, and accessibility, with results visualized through network graphs, tables, and annotated texts. Findings demonstrate that combining multiple methods enhances interpretive power, while LLM-based approaches enable rapid prototyping and exploration, particularly in low-resource settings. The project contributes to the development of digital literacies in classical scholarship and advances a new research paradigm—augmented reading—by integrating computational methods into traditional philological practice. The talk was given at the colloqium "Variation Linguistique et Linguistique Computationnelle" (Sorbonne Université).