LITERA: An LLM Based Approach to Latin-to-English Translation
Paul Rosu
TL;DR
LITERA introduces a multi-layer, literal-focused Latin→English translation pipeline built on fine-tuned GPT-4o-mini and revision-grade GPT-4o modules, guided by Mixture-of-Agents principles to iteratively refine outputs. It achieves state-of-the-art results on Classical Latin with $BLEU$ and $BLEURT$ improvements and demonstrates cross-LLM compatibility, including open-source LLaMA 8B, while maintaining a transparent, traceable translation process. The work is anchored by a small, high-quality Duke-curated Latin–English dataset and emphasizes reproducibility, licensing clarity, and a freely accessible translation platform. Limitations include the modest dataset size, potential translation errors without human checks, and variable performance on Early Modern Latin, signaling avenues for broader data collection and evaluation across time periods.
Abstract
This paper introduces an LLM-based Latin-to-English translation platform designed to address the challenges of translating Latin texts. We named the model LITERA, which stands for Latin Interpretation and Translations into English for Research Assistance. Through a multi-layered translation process utilizing a fine-tuned version of GPT-4o-mini and GPT-4o, LITERA offers an unprecedented level of accuracy, showcased by greatly improved BLEU scores, particularly in classical Latin, along with improved BLEURT scores. The development of LITERA involved close collaboration with Duke University's Classical Studies Department, which was instrumental in creating a small, high-quality parallel Latin-English dataset. This paper details the architecture, fine-tuning methodology, and prompting strategies used in LITERA, emphasizing its ability to produce literal translations.