HERIT: Enabling Global Access to Hanja-to-English Historical Translation by Mitigating Temporal Bias
Authors: Dayeon Jeong, Kyeongpil Kang
Organizations: Graduate School of Data Science, Kangwon National University · Department of Computer Science and Engineering, Kangwon National University
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.
Figures & tables
Hanja
Korean
English
疫疾
역질
Epidemic disease
世宗
세종
King Sejong
江原道
강원도
Kangwon Province
Table 1: Examples of glossary information: Hanja source, Korean pronunciation, and English translation.
Figure 1: Overview of the proposed framework for translating historical documents into English. The framework comprises three main stages: 1) pseudo-labeling for data augmentation, 2) fine-tuning the LLM by integrating human-labeled and pseudo-labeled data, and 3) translation of the untranslated corpus using the trained model.
Model
BLEU
chrF++
Cost $
Hunyuan-MT-7B
8.318
35.450
-
Qwen3 8B
10.631
38.738
-
Qwen3 32B
12.062
40.461
-
GPT-5.1
13.574
43.545
1.92
Kimi-K2 (R)
15.507
43.684
5.63
Gemini-2.5-Flash
15.720
43.971
0.42
Table 2: Translation performance on Dtest using basic prompts and the total API cost. (R) denotes the integration of reasoning capabilities.
Table 3: Ablation study of pseudo-labeling. The language in parentheses denotes the query language used for retrieval, while K and D represent the number of few-shot examples and translation drafts, respectively.
Model
BLEU
METEOR
chrF++
H2KE
24.62
-
-
Pivot-MT
35.539
0.596
59.804
With Glossary & RAG
Gemini-2.5-Flash
35.632
0.600
60.176
Gemini-2.5-Pro
34.847
0.598
60.139
Qwen3 32B ( Dtrain )
38.117
0.622
62.338
Table 4: Translation performance on Dtest . Our method is compared against baselines (H2KE Son et al. (2022) and Pivot-MT), zero-shot LLMs, and various ablation models. For H2KE, the BLEU score reported in the original paper is used.
Figure 2: Translation performance with respect to pseudo-labeling data size.
Model
Qwen3 8B
Qwen3 32B
With Glossary & RAG
Dtrain
6.539
7.003
Without Glossary & RAG
Dtrain
6.312
6.883
Daug→Dtrain
7.887
8.009
Table 5: LLM-based evaluation on the testset DNTtest .
Figure 3: Human expert evaluation scores and preference ratios for our model versus the baseline.
Figure 4: Differences in human expert scores between HERIT (ours) and Qwen3 32B trained solely on human-labeled data, ordered chronologically by each king’s reign. Positive values indicate that HERIT outperforms the baseline, demonstrating its robustness across different historical periods.
Input Hanja
兩司啓趙應奎事, 不允.
Hunyuan-MT
a petition was submitted by the two departments regarding Zhao Yingkui ’s case, but it was denied.
Qwen3 32B
The two investigative bodies reported on the matter of Jo Eunggyu , but the king did not approve their request.
HERIT
The two offices reported on the matter of Jo Eunggyu , but the king did not approve.
Input Hanja
宜給宣飯及房子, 炊飯, 汲水人等
Hunyuan-MT
It is appropriate to provide food, housing, cooking facilities, and water for the people.
Qwen3 32B
They should be provided with food, a house , a female palace servant , and a water carrier.
Table 6: Qualitative examples of translations by different models using samples in Dtest . Red denotes incorrect translations, while blue indicates correct ones.
Appendix figures & tables11 assets
Supplementary material from the paper’s appendix.
Appendix
Case
Language
Sample in Dtest
Searched Sample in Dtrain
Example 1 (Sim: 0.900)
Hanja
宗貞盛遣人來獻 土宜 .
宗貞盛遣人來獻 土物 .
English
Sō Sadamori sent a messenger to present local products.
Sō Sadamori sent a man to present his local products to the king.
Example 2 (Sim: 0.857)
Hanja
上王詣健元陵, 行 端午 別祭.
上王詣健元陵, 行 中秋 別祭.
English
The abdicated king went to the royal tomb of King Taejo and held a special ancestral rite of the fifth day of the fifth month .
The abdicated king visited the royal tomb of King Taejo to conduct the special sacrificial rite for the Mid-Autumn .
Appendix
Table 7: Qualitative examples of test–train sample pairs with high Levenshtein similarity.
Figure 5: Distribution of test set samples across similarity intervals against the training data.
Figure 6: Performance and cost of the Proposer and Fusor across different numbers of few-shot examples.
BLEU
METEOR
chrF++
No aggregation
37.905
0.616
61.534
BoN
38.337
0.622
62.117
FusioN
38.462
0.625
62.501
Appendix
Table 8: Comparison of pseudo-labeling performance among FusioN, BoN, and Baseline.
Method
BLEU
METEOR
chrF++
Cost ($)
Multiple
36.250
0.613
61.507
234.304
Single
38.462
0.625
62.501
58.517
Appendix
Table 9: Performance comparison between diverse-model and single-model FusioN approaches.
Input Hanja
御勤政殿受朝.
Ground Truth
The King proceeded to Geunjeong Hall and received officials in the morning court assembly.
Translation by Candidates
(1) The king proceeded to Geunjeong Hall and received officials in the morning court assembly .
(2) The king proceeded to Geunjeong Hall to receive the morning court assembly .
…
Translation by Fusor
The king proceeded to Geunjeong Hall and received officials in the morning court assembly.
Input Hanja
傳旨兵曹: 自今司僕寺常養馬, 冬節則三百匹, 夏節則二百匹.
Appendix
Table 10: Qualitative case studies showing how FusioN synthesizes candidate translations.
Model
Accuracy
Qwen3 32B ( Dtrain )
0.818
Qwen3 32B ( Daug∪Dtrain )
0.875
Qwen3 32B ( Daug→Dtrain )
0.878
Appendix
Table 11: LLM-based evaluation of proper noun accuracy across the baseline, pseudo-labeling, and training strategies.
Figure 7: Web-based evaluation tool for human experts.
Figure 8: An example of the base prompt.
Figure 9: An example of the prompt format used in the Proposer.
Figure 10: An example of the prompt format used in the Fusor.
Machine translation develops computational methods for converting text, speech, and visual content between languages while preserving meaning and purpose.
Translating historical cultural heritage poses profound challenges for machine translation systems due to the acute semantic asymmetry across eras and cultures, where terse classical syntax frequently obscures domain-specific cultural knowledge. While Large Language Models have demonstrated remarkable prowess in generic translation benchmarks, they struggle to negotiate Culturally-Specific Items and diachronic shifts in domain-rich historical texts, such as Yellow River cultural records. To bridge this critical gap, we introduce YRC-Bench, the first human-curated, multi-parallel benchmark dedicated to historical heritage translation across Classical Chinese, Modern Chinese, and nine target foreign languages. YRC-Bench establishes a tri-level parallel corpus incorporating modern explanatory glosses as explicit semantic bridges and operationalizes translation strategies through paired conserved versus substituted CSI variants categorized by cultural translatability. Systematic evaluations of nine leading LLMs across bidirectional translation directions reveal that leveraging modern Chinese glosses - especially human-curated ones - dramatically elevates both lexical fidelity and semantic adequacy. Furthermore, multidimensional human preference audits by native experts unveil distinct cross-cultural preferences between Sinographic-tradition and non-Sinographic language groups.
We introduce TypewriterLM, a 7.24B History language model (LM) trained exclusively on English text predating 1913. Developing History LMs requires addressing challenges in data quality and availability, preventing temporal leakage, designing temporally consistent post-training pipelines, and constructing reliable evaluations. To address these issues, we construct TypewriterCorpus, a 54B-token historical corpus collected from diverse archival and linguistically annotated sources with extensive data cleaning and leakage mitigation procedures. Furthermore, we introduce lexically grounded instructing tuning, a post-training framework that constraints responses to remain directly grounded in historical source documents. Using this framework we construct two historical instruction tuning datasets: History-LIMA and History-SelfInstruct. To evaluate capability and temporal consistency, we introduce History-Event, a benchmark suite for evaluating competence, temporal grounding and data leakage. We release TypewriterLM and all associated resources to support future research on historical language models.
Despite remarkable progress in machine translation, Vision Language Models (VLMs) struggle on historical manuscripts, a domain that stresses core Natural Language Processing (NLP) capabilities: low-resource transliteration, archaic vocabulary, and noisy input signals. We present a systematic framework for evaluating the full image-to-translation pipeline on medieval Latin manuscripts, a setting in which scribal shorthand, ligatures, and parchment degradation expose failure modes that are invisible in clean-text benchmarks. Benchmarking on the CATMuS Latin dataset reveals a specialization gap: domain-specific Optical Character Recognition (OCR) models reduce character error rate by up to 4.3× compared to general-purpose VLMs, despite operating at orders of magnitude fewer parameters. We introduce the Interpres-Parallel-Corpus (IPC), a novel dataset comprising 1,383 aligned manuscript image lines, transcriptions, and expert translations, to our knowledge the first of its kind for medieval Latin. Our experiments uncover a complexity paradox: the simplest pipeline, a specialized OCR model feeding directly into a VLM, outperforms all multi-component variants. Adding retrieval-augmented generation (RAG) or post-OCR correction introduces prompt saturation and error propagation that degrade aggregate translation quality. These findings offer both a new benchmark and practical guidance for deploying translation systems in low-resource historical settings.
Nguyen Kim Hai Bui, Md. Easin Arafat, Tamás Gábor Orosz +1
Eötvös Loránd University · King Fahd University of Petroleum and Minerals