A Study on Japanese-English Machine Translation Based on Large Language Models and Post-Editing Strategies

Sirui Wang, Xiang Li
Proceedings of 2025 2nd International Conference on Machine Learning and Intelligent Computing, PMLR 278:222-228, 2025.

Abstract

This study investigates the role of post-editing in enhancing neural machine translation (NMT) quality, focusing on Japanese-to-English translations in the Information and Communication Technology (ICT) sector. By analyzing outputs from three NMT platforms (DeepSeek, Youdao, DeepL) against an official benchmark, the research identifies persistent challenges, including inconsistent terminology (e.g., “critical infrastructure" vs. “core facilities"), tense inaccuracies (“had been restored" vs. “have returned"), and omissions of technical annotations (e.g., “Hikari (fiber-optic)"). While DeepL and DeepSeek demonstrate superior semantic and structural fidelity, their outputs require adjustments to align with domain-specific standards. The proposed post-editing framework prioritizes terminological alignment with authoritative references, temporal precision to emphasize ongoing actions, and structural coherence to restore source-text logic. Full post-editing is advocated for formal contexts to achieve human parity, whereas light post-editing suffices for rapid delivery with minimal quality compromises. Industry data highlights the dominance of the “machine translation + post-editing" model, adopted in 30.4% of projects in 2023, underscoring its efficiency and cost-effectiveness. However, human expertise remains irreplaceable in addressing nuanced challenges such as cultural adaptation and contextual dependencies. The study concludes by advocating for AI-augmented post-editing tools to streamline workflows while preserving the “humanistic core" essential for high-stakes translations. This synergy between technological advancement and human judgment is critical for advancing translation quality in the AI era, particularly in high-demand sectors like ICT.

Cite this Paper

BibTeX
@InProceedings{pmlr-v278-wang25c, title = {A Study on Japanese-English Machine Translation Based on Large Language Models and Post-Editing Strategies}, author = {Wang, Sirui and Li, Xiang}, booktitle = {Proceedings of 2025 2nd International Conference on Machine Learning and Intelligent Computing}, pages = {222--228}, year = {2025}, editor = {Zeng, Nianyin and Pachori, Ram Bilas and Wang, Dongshu}, volume = {278}, series = {Proceedings of Machine Learning Research}, month = {25--27 Apr}, publisher = {PMLR}, pdf = {https://raw.githubusercontent.com/mlresearch/v278/main/assets/wang25c/wang25c.pdf}, url = {https://proceedings.mlr.press/v278/wang25c.html}, abstract = { This study investigates the role of post-editing in enhancing neural machine translation (NMT) quality, focusing on Japanese-to-English translations in the Information and Communication Technology (ICT) sector. By analyzing outputs from three NMT platforms (DeepSeek, Youdao, DeepL) against an official benchmark, the research identifies persistent challenges, including inconsistent terminology (e.g., “critical infrastructure" vs. “core facilities"), tense inaccuracies (“had been restored" vs. “have returned"), and omissions of technical annotations (e.g., “Hikari (fiber-optic)"). While DeepL and DeepSeek demonstrate superior semantic and structural fidelity, their outputs require adjustments to align with domain-specific standards. The proposed post-editing framework prioritizes terminological alignment with authoritative references, temporal precision to emphasize ongoing actions, and structural coherence to restore source-text logic. Full post-editing is advocated for formal contexts to achieve human parity, whereas light post-editing suffices for rapid delivery with minimal quality compromises. Industry data highlights the dominance of the “machine translation + post-editing" model, adopted in 30.4% of projects in 2023, underscoring its efficiency and cost-effectiveness. However, human expertise remains irreplaceable in addressing nuanced challenges such as cultural adaptation and contextual dependencies. The study concludes by advocating for AI-augmented post-editing tools to streamline workflows while preserving the “humanistic core" essential for high-stakes translations. This synergy between technological advancement and human judgment is critical for advancing translation quality in the AI era, particularly in high-demand sectors like ICT.} }
Endnote
%0 Conference Paper %T A Study on Japanese-English Machine Translation Based on Large Language Models and Post-Editing Strategies %A Sirui Wang %A Xiang Li %B Proceedings of 2025 2nd International Conference on Machine Learning and Intelligent Computing %C Proceedings of Machine Learning Research %D 2025 %E Nianyin Zeng %E Ram Bilas Pachori %E Dongshu Wang %F pmlr-v278-wang25c %I PMLR %P 222--228 %U https://proceedings.mlr.press/v278/wang25c.html %V 278 %X This study investigates the role of post-editing in enhancing neural machine translation (NMT) quality, focusing on Japanese-to-English translations in the Information and Communication Technology (ICT) sector. By analyzing outputs from three NMT platforms (DeepSeek, Youdao, DeepL) against an official benchmark, the research identifies persistent challenges, including inconsistent terminology (e.g., “critical infrastructure" vs. “core facilities"), tense inaccuracies (“had been restored" vs. “have returned"), and omissions of technical annotations (e.g., “Hikari (fiber-optic)"). While DeepL and DeepSeek demonstrate superior semantic and structural fidelity, their outputs require adjustments to align with domain-specific standards. The proposed post-editing framework prioritizes terminological alignment with authoritative references, temporal precision to emphasize ongoing actions, and structural coherence to restore source-text logic. Full post-editing is advocated for formal contexts to achieve human parity, whereas light post-editing suffices for rapid delivery with minimal quality compromises. Industry data highlights the dominance of the “machine translation + post-editing" model, adopted in 30.4% of projects in 2023, underscoring its efficiency and cost-effectiveness. However, human expertise remains irreplaceable in addressing nuanced challenges such as cultural adaptation and contextual dependencies. The study concludes by advocating for AI-augmented post-editing tools to streamline workflows while preserving the “humanistic core" essential for high-stakes translations. This synergy between technological advancement and human judgment is critical for advancing translation quality in the AI era, particularly in high-demand sectors like ICT.
APA
Wang, S. & Li, X.. (2025). A Study on Japanese-English Machine Translation Based on Large Language Models and Post-Editing Strategies. Proceedings of 2025 2nd International Conference on Machine Learning and Intelligent Computing, in Proceedings of Machine Learning Research 278:222-228 Available from https://proceedings.mlr.press/v278/wang25c.html.

Related Material