1 Introduction

ChatGPT is an NLP system developed by OpenAI, which possesses deep learning and semantic analysis capabilities, promoting the development of interactive post editing practices. The release of ChatGPT-4o marks the advancement of NLP in human-computer interaction. In translation practice, ChatGPT feedback helps to identify and avoid translation errors. This article uses technology translation texts as materials and conducts a comparative experiment on translation quality using ChatGPT-4o. The aim is to compare the changes in translation quality after translators modify their translations based on ChatGPT feedback under manual evaluation and machine evaluation modes, and explore the role of ChatGPT feedback in improving the quality of manual translation.

2 ChatGPT feedback and translation quality evaluation

The role of ChatGPT in improving the quality of technology translation was discussed, emphasizing its advantages as a large language model in providing feedback. ChatGPT has fast feedback speed and high efficiency, providing a large number of personalized modification suggestions, which helps learners learn in a timely manner and self correct errors. At the same time, the chapter also explores methods for evaluating translation quality, including qualitative and quantitative evaluations, and points out the shortcomings of manual and machine evaluations in existing research. This study used the error scoring method for manual evaluation and combined BLEU, TER, and METEOR indicators for machine evaluation, aiming to quantitatively analyze the impact of ChatGPT feedback on improving the quality of manual technology translation.

3 Research Design

A quantitative study was introduced to verify the effect of ChatGPT feedback on improving the quality of artificial technology translation. The research subjects are 28 master's students from the School of Foreign Languages at a comprehensive university in Central China, who possess strong translation skills and rich translation experience. Four scientific texts were selected for the experiment, and participants were required to manually translate them before modifying the translations based on ChatGPT feedback. The evaluation methods include machine evaluation and manual evaluation: machine evaluation calculates BLEU, METEOR, and TER values through NLTK and sacreBLEU libraries; Manual evaluation is conducted based on technology translation standards to develop classification and scoring criteria for error types. Four MTI master's students, under the guidance of professors, mark and count errors in the translated text. Finally, a questionnaire survey was conducted to understand the participants' evaluation of ChatGPT feedback.

Evaluation and analysis of translation quality before and after ChatGPT feedback

The experiment conducted machine and manual evaluations on 224 translations to analyze the impact of ChatGPT feedback on the quality of technology translation. The machine evaluation results show that the BLEU, TER, and METEOR values of the translation have all improved after ChatGPT feedback, indicating an improvement in the quality of the translation. Specifically, in the tasks of translating from Chinese to English and from English to Chinese, the probability of ChatGPT feedback helping improve the quality of the translation is approximately 77% and 74%, respectively. Manual evaluation found that after students modified their translations based on ChatGPT feedback, the total number of errors and error scores decreased, especially in terms of completeness, accuracy, and language. However, ChatGPT has a slightly weaker ability to address style related errors.

5 Conclusion

This chapter summarizes the quantitative evaluation results of technology translation quality before and after ChatGPT feedback using automatic evaluation indicators and error scoring method, and finds that ChatGPT feedback significantly improves the translation quality of Chinese to English and English to Chinese translations. ChatGPT-4o, due to its large-scale corpus foundation and interactive performance, can effectively identify mistranslations and omissions, optimize expression, standardize terminology translation, and correct grammar errors. The study emphasizes that translators and translation learners should leverage the advantages of large language models to enhance their translation skills through personalized feedback. Meanwhile, translation instructors can use tools such as ChatGPT to assist in teaching and improve feedback efficiency. The study also pointed out limitations, including small sample size, limited subject range, and the possibility of optimizing the experimental process. It is recommended to expand the sample size and subject range in the future, and further explore the role of ChatGPT feedback in improving translation quality.