INFORMATION ABOUT THE MASTER'S THESIS
1. Student's full name: WEI ZE MIN
2. Gender: Male
3. Date of birth: March 22, 1992
4. Place of birth: Guangxi, China
5. Decision No. 6998/QD-XHNV dated December 18, 2024, of the Rector of the University of Social Sciences and Humanities, Vietnam National University, Hanoi, recognizing the student.
6. Changes in the training process: None
7. Thesis Title: Application of AI in Chinese-Vietnamese Translation of Legal Documents on Food Packaging Labels and its Significance for Vietnam-China Trade Integration
8. Major: Vietnamese Studies;
9. Code:8310630
10. Scientific supervisor:
(Please specify your scientific title, academic degree, full name, and workplace.)
Dr. Vu Minh Hai – University of Social Sciences and Humanities
11. Summary of the thesis results:
This thesis evaluates the applicability of ChatGPT and DeepSeek in the Chinese-Vietnamese translation of legal documents on food labeling, and examines the significance of translation for the adoption of standards by Vietnamese businesses in trade with China. The study uses three data sources: a survey of 212 bilingual users, an analysis of translations by the two tools for GB 7718-2025 and GB 28050-2025, and interviews with two factory representatives and an expert. The survey results show that AI is primarily used for generating draft translations, looking up terminology, suggesting phrasing, and supporting self-learning. Participants appreciated the time-saving capabilities but also recognized the risks of misinterpretation, dependence, and data leakage. Analysis across six established dimensions shows that ChatGPT creates natural-sounding, easy-to-read Vietnamese sentences, but it can omit details, simplify definitions, or reduce the level of mandatory requirements. DeepSeek tends to retain more source content, data, and structure, but its translations often bear traces of Chinese characters and still contain terminology errors. Therefore, no tool is reliable enough for direct use in compliance purposes without being cross-referenced with the original source and post-edited by experts. Comparisons of standards between the two countries also reveal differences in required information, allergens, nutritional components, and label presentation, necessitating data updates, packaging adjustments, and coordination within businesses. From this, the thesis proposes a six-step process based on the "human design - AI support - human verification and accountability" model. However, due to the corpus consisting of only two standards, the convenient survey sample, the small number of interviewees, and the lack of independent evaluation by multiple experts, the results are only exploratory and cannot be generalized to all AI models or food businesses.
12. Practical applications:
The research has potential applications in translation work.management processThis study examines the work of businesses and training in Chinese-Vietnamese translation. The six criteria and six-step post-editing process can assist translators in checking the completeness, terminology, normative level, data, and presentation of AI translations. Businesses can refer to and apply this process when receiving standards, preparing labels, and controlling export documentation. In training, the research results help develop exercises for evaluating AI output and post-editing specialized texts. However, the proposed tools need further testing before widespread application.
13. Future research directions:
Further research could expand the scope to include standards for additives, quarantine, traceability, and import/export documentation in Vietnam-China trade, while also surveying more industries and businesses. A parallel Chinese-Vietnamese corpus needs to be developed, AI translations evaluated by multiple independent experts, and stability tested across models, commands, and output generation cycles. Real-world testing of the post-editing process would also help measure the impact of AI on translation quality, processing time, compliance costs, and customs clearance efficiency.
14. Published works related to the thesis:Do not have
INFORMATION ON MASTER'S THESIS
1. Full name: WEIZE MIN 2. Sex: Male
3. Date of birth: March 22, 1992 4. Place of birth: China
5. Admission decision number:No.6889/QD-XHNV dated December 18, 2024, issued by the Rector of the University of Social Sciences and Humanities, Vietnam National University, Hanoi.
6. Changes in academic process:None
7. Official thesis title:The application of AI in the Chinese-Vietnamese translation of legal documents on food packaging labels and its significance for Vietnam-China trade integration.
8. Major: Vietnamese Studies
9. Code:8310630
10. Supervisors: VuMinh HaiPhD
11. Summary of the findings of the thesis:
The thesis evaluates the applicability of ChatGPT and DeepSeek to Chinese–Vietnamese translation of food-labeling regulatory texts and examines the implications of such translation for Vietnamese enterprises' adoption and implementation of standards in trade with China. The study draws on three sources of data: a survey of 212 bilingual users, an analysis of the translations generated by the two AI tools for GB 7718-2025 and GB 28050-2025, and interviews with two factory representatives and one expert. The survey results indicate that AI is mainly used to produce draft translations, look up terminology, suggest alternative expressions, and support self-directed learning. Participants highly value AI's ability to save time, while also recognizing risks associated with mistranslation, over-reliance, and data leakage. The analysis based on the six established evaluation dimensions shows that ChatGPT tends to produce natural and readable Vietnamese, but may omit details, simplify definitions, or weaken the normative force of mandatory requirements. DeepSeek tends to preserve more of the source content, numerical information, and structural features, but its output often retains Chinese-language traces and continues to exhibit terminological errors. Therefore, neither tool is sufficiently reliable for direct use in compliance-related purposes without reference to the original text and post-editing by a qualified professional. A comparison of the standards of the two countries also differences in mandatory information requirements, allergens reveal labeling, nutrition information, and labeling presentation, thereby creating requirements for data updating, packaging adjustments, and interdepartmental coordination within enterprises. On this basis, the thesis proposes a six-step workflow based on the principle of “human-designed – AI-assisted – human-verified and accountable.” However, because the corpus is limited to two standards, the survey employed a convenience sampling method, the number of interviewees was small, and independent evaluation by multiple experts was not conducted, the findings are exploratory in nature and cannot yet be generalized to all AI models or food enterprises.
12. Practical applicability:
The findings of the study have potential applications in translation practice, enterprise management processes, and Chinese–Vietnamese translator training. The six evaluation criteria and six-step post-editing workflow can assist translators in checking the completeness, terminology, standardized force, numerical information, and formatting of AI-generated translations. Enterprises may consider applying this workflow when adopting standards, preparing product labels, and reviewing export documentation. In translator training, the findings can support the development of practical exercises focusing on the evaluation of AI-generated output and the post-editing of specialized texts. Nevertheless, the proposed tools and procedures require further validation before they can be applied on a wider scale.
13. Further research directions:
Future research could expand the scope of investigation to standards regarding food additives, quarantine and inspection, traceability, and import–export documentation in Vietnam–China trade, while also examining a wider range of product sectors and enterprises. There is also a need to develop Chinese–Vietnamese parallel corpora, evaluate AI-generated translations through multiple independent expert assessors, and examine the consistency of translation outputs across different AI models, prompts, and generation attempts. Empirical testing of the proposed post-editing workflow in real-world settings will further enable researchers to measure the effects of AI on translation quality, processing time, compliance costs, and customs-clearance efficiency.
14. Thesis-related publications:None
Newer news
Older news
