To say almost the same thing? A study on cross-linguistic variation in ancient texts and their translations.

This article examines translation alignment patterns in parallel corpora of ancient and historical languages, derived from Ugarit, a crowdsourcing platform designed to manually align texts in low-resourced languages. The study analyzes bilingual alignments involving languages such as Ancient Greek, Latin, Classical Persian, Coptic, and Georgian, focusing on the ratios of word correspondences classified as one-to-one, one-to-many, many-to-one, and many-to-many. Findings indicate that linguistic structures—such as inflection and agglutination—and cultural factors, including translation practices and textual genres, significantly influence alignment patterns; for example, English translations of Classical Persian often show expansions and paraphrases reflecting cultural tendencies, while Georgian’s polysynthetic nature leads to different alignment dynamics with Ancient Greek. The research highlights the complex interplay between linguistic typology, cultural context, and translation strategies in the creation of aligned corpora, offering insights valuable for developing training data and guidelines for automatic translation models in ancient language studies.