生成式人工智能模型在處理低資源語言對的專業術語翻譯時,面臨一個根本性限制：高度專業化的詞彙在訓練語料庫中出現頻率過低。本文針對這一問題,提出了一個混合框架,將計算詞典學與神經機器翻譯相結合,以解決系統性的術語不一致現象。該方法包含四階段自動提取流程：基於詞典的匹配、語音相似性分析、統計顯著性檢測以及專有名詞解析。經專家驗證後,這些流程構建出富含元數據的術語數據庫。數據庫支持兩階段翻譯系統：首先通過標準神經微調處理風格和句法,隨後基於元數據（名稱、傳統、時期）進行自動後處理,實現術語控制。在古代語言至中文的聖經翻譯測試中（希伯來語、希臘語、拉丁語）,該系統的術語覆蓋率較詞典基線提升了1.13倍至1.68倍。本框架並非旨在取代商業大型語言模型,而是提供一種互補方法：通過元數據驅動的術語控制彌補現有模型在專業領域的系統性缺陷。此類方法論有望直接整合至未來模型的訓練或推理流程中。該模塊化架構或可擴展至其他具有相似特徵的專業領域,如有限並行語料庫和高術語變異性的法律、醫學及技術領域。
Abstract
Generative AI models face a fundamental limitation when translating specialized terminology in low-resource language pairs: highly specialized vocabulary appears too infrequently in training corpora. This paper addresses this problem by proposing a hybrid framework that combines computational lexicography with neural machine translation to resolve systematic terminological inconsistencies. The approach implements a four-stage automated extraction pipeline: dictionary-based matching, phonetic similarity analysis, statistical significance detection, and proper noun resolution. Following expert validation, these four stages construct a metadata-rich terminological database. The database supports a two-stage translation system: first, standard neural fine-tuning handles style and syntax, followed by automated post-processing based on metadata (denomination, tradition, period) to achieve terminological control. In tests on biblical translation from ancient languages to Chinese (Hebrew, Greek, Latin), the system achieved 1.13× to 1.68× improvements in terminology coverage over dictionary baselines. Rather than competing with commercial large language models, this framework proposes a complementary approach: metadata-driven terminological control addresses systematic gaps that current models cannot resolve. Such methodology could be integrated directly into future model training or inference pipelines. The modular architecture could extend to other specialized domains with similar characteristics: limited parallel corpora and high terminological variability, including legal, medical, and technical fields.
关键词
神經機器翻譯 /
計算詞典學 /
低資源語言 /
術語提取 /
專業翻譯
Key words
neural machine translation /
computational lexicography /
low-resource languages /
terminology extraction /
specialized translation
{{custom_sec.title}}
{{custom_sec.title}}
{{custom_sec.content}}
参考文献
中文部分(Chinese Section)
白雲曉(2015).《聖經人名辭典》.中央編譯出版社..
Bai, Yunxiao (2015). Shengjing renming cidian. Zhongyang bianyi publications.
輔仁大學神學著作編譯會(1987).《信理神學詞彙》.光啓出版社..
Furen daxue shenxue zhuzuo bianyi hui (1987). Xinli shenxue cihui. Guangqi chubanshe.
——.(2007).《信理神學詞彙》.光啓出版社..
——.(2007). Xinli shenxue cihui. Guangqi chubanshe.
英文部分(English Section)
Balmont, Alexis (2025). Syriac Christians in Tang China: Chinese Texts and Theologies. Lit-Verlag.
Bamman, David,Gregory Crane (2009). “Computational Linguistics and Classical Lexicography.”Digital Humanities Quarterly 3(1):1-49.
Basset, Jean, andJohann Su (ca. 1705-1707). Testamentum Novum Sinice [Chinese new testament]. Biblioteca Casanatense di Roma [Casanatense library of Rome], Mss. 2024.
Cheng, Yong, Yang Liu, Qian Yang, Maosong Sun,Wei Xu (2016). “Neural Machine Translation with Pivot Languages.” arXiv preprint arXiv:1611.04928.
Choi, Daniel Kam To (2023). “Issues Related to the Chinese Translation of ‘God’: A Study from a Historical Perspective.” In Translatology, Translation and Interpretation—Toward a New Scientific Endeavor, ed. by Noury Bakrim, 1-20. IntechOpen.
Garcia, Xavier, Yamini Bansal, Colin Cherry, George Foster, Maxim Krikun, Fangxiaoyu Feng, Melvin Johnson,Orhan Firat (2023). “The Unreasonable Effectiveness of Few-Shot Learning for Machine Translation.” arXiv:2302.01398.
Hock, Ronald (1995). The Infancy Gospels of James and Thomas. Polebridge Press. Hu, Junjie, Mengzhou Xia, Graham Neubig, and Jaime Carbonell (2019). “Domain Adaptation of Neural Machine Translation by Lexicon Induction.” Proceedings of the 57th Annual Meeting of the Association for Computational Linguistics, 2989-3001. Association for Computational Linguistics.
Johnson, Kyle, Patrick Burns, John Stewart, Todd Cook, Clément Besnier,William Mattingly (2021). “The Classical Language Toolkit: An NLP Framework for Premodern Languages.”Proceedings of the 59th Annual Meeting of the Association for Computational Linguistics and the 11th International Joint Conference on Natural Language Processing: System Demonstrations, 20-29.
Kay, Christian, Jane Roberts, Michael Samuels, and Irené Wotherspoon (2009). Historical Thesaurus of the Oxford English Dictionary. Oxford University Press.
Leeb, Leopold (2021). “Classical Languages in Chinese Biblical Studies.” In The Oxford Handbook of the Bible in China, ed. by Khiok-Khng Yeo, 147-162. Oxford University Press.
Li, Zhengrong (2021). “The Eastern Orthodox Bible in China.” In The Oxford Handbook of the Bible in China, ed. by Khiok-Khng Yeo, 63-78. Oxford University Press.
Liu, Yinhan, Jiatao Gu, Naman Goyal, Xian Li, Sergey Edunov, Marjan Ghazvininejad, Mike Lewis,Luke Zettlemoyer (2020). “Multilingual Denoising Pre-training for Neural Machine Translation.”Transactions of the Association for Computational Linguistics 8: 726-742.
Oxford FLRTP (2010). English-Chinese, Chinese-English Dictionary. Oxford University Press.
Saunders, Danielle (2021). “Domain Adaptation and Multi-domain Adaptation for Neural Machine Translation: A Survey.” arXiv:2104.06951.
Suee, Yan Yu (2021). “Bible Translations for Ethnic Minority Groups in China.” In The Oxford Handbook of the Bible in China, ed. by Khiok-Khng Yeo, 129-145. Oxford University Press.
Tan, Justin (2021). “Patristic Biblical Interpretation in China.” In The Oxford Handbook of the Bible in China, ed. by Khiok-Khng Yeo, 415-428. Oxford University Press.
Wu, Jingrong (2000). New Age Chinese-English Dictionary. Commercial Press.
Yang, Huilin (2021). “Study of the Bible in Chinese Academia.” In The Oxford Handbook of the Bible in China, ed. by Khiok-Khng Yeo, 399-414. Oxford University Press.
PDF(849 KB)
