HERIT: Enabling Global Access to Hanja-to-English Historical Translation by Mitigating Temporal Bias
Dayeon Jeong, Kyeongpil Kang
Kangwon National University
Published November 3, 2026 · on the feed November 5, 2026 · ▲ 1 · arXiv 2611.04949
The abstract
Historical archives are invaluable for multidisciplinary research but remain inaccessible globally due to severe language barriers. While manual translation is prohibitively expensive and time-consuming, simply applying existing machine translation models is inadequate due to unique linguistic nuances and the temporal bias inherent in scarce human-labeled corpora. To address these challenges, we propose HERIT, a systematic domain-adaptation pipeline that leverages retrieval-augmented generation (RAG) to generate high-quality pseudo-labeled data from abundant Hanja-Korean corpora. This approach substantially expands the training dataset, mitigating both data scarcity and temporal overfitting across different historical periods. Extensive automated and human evaluations demonstrate that HERIT consistently outperforms baselines. Finally, we translate previously untranslated portions of the archives using our model, thereby enabling global access to these resources for researchers.
The Brief
1:45A short podcast summary of the paper. Written and voiced by AI.
Transcript
HERIT. Enabling Global Access to Hanja-to-English Historical Translation by Mitigating Temporal Bias. By Dayeon Jeong and one other, from Kangwon National University. Published on the third of November.
Historical archives serve research across many disciplines, yet for most of the world they stay out of reach because of severe language barriers. Translating them by hand is prohibitively expensive and slow. Running an existing machine translation model over them is not enough either. The texts carry unique linguistic nuances, and the human-labeled corpora that exist are scarce and carry a temporal bias: a model trained on them overfits to the historical periods they happen to cover.
HERIT is a systematic domain-adaptation pipeline built to get around that scarcity. Its central move is to lean on what is abundant. Hanja-Korean corpora exist in quantity, and HERIT uses them as raw material for new training data.
It does so with retrieval-augmented generation, using it to produce high-quality pseudo-labeled data from those Hanja-Korean corpora. The result substantially expands the training set, and the abstract credits that expansion with easing two problems together: data scarcity, and temporal overfitting across different historical periods.
Both automated and human evaluations are reported, described as extensive, and HERIT consistently outperforms the baselines in them. The advantage is consistent, and it is checked by people as well as by automatic metrics.
The work does not stop at the evaluation. The authors use their model to translate portions of the archives that had never been translated before, and so put the pipeline to the use it was built for. The bet underneath HERIT is simple to state: where human-labeled data is scarce and skewed toward certain periods, a far larger body of related text can be turned into training data that covers more of history. The outcome is the global access the title promises, opened to researchers wherever they are.
More with an account
Listen to every day’s papers
Daily playlists, deep dives, quests for the papers, and quizzes on their key terms.
Join free