Ancient Words, New Demands: Lessons from Recent Work on Greek, Latin, and Biblical Reception. Retrieving Biblical References in Greek and Latin Patristic Texts

Identifying biblical quotations and allusions in early Christian literature is hard for computational methods: the Bible was never a single fixed text, ancient authors reshaped what they reused, and Ancient Greek and Latin are low-resource languages. We report on recent work that frames the problem as ranked retrieval, with Clement of Alexandria, Nonnus of Panopolis, and Augustine as test cases, and we argue that the decisive factor is knowledge enhancement: making philological knowledge operational within the pipeline. This happens at three points. A graded annotation scheme replaces the quotation/allusion dichotomy with nested levels of similarity. A rule-based parser turns the critical apparatus of the Septuagint into a knowledge graph and into virtual versions of each verse, which are indexed as historically attested data augmentation. Parallel Latin versions of the Bible supervise contrastive fine-tuning with self-mined hard negatives. Signals grounded in the textual tradition improve retrieval, while generic ones, such as paraphrases generated by large language models or unconstrained synonym expansion, degrade it. An error analysis maps the remaining failures to their causes and shows that the hardest cases require explicit exegetical knowledge. Throughout, the system ranks candidates and the scholar judges their relevance.