1. Introduction

We discussed the challenges of machine translation in terms of lexical diversity and syntactic complexity, pointed out the shortcomings of traditional evaluation methods, and emphasized the potential of ChatGPT in translation tasks. This article introduces the significant progress of ChatGPT-4 in understanding logical structures, handling language complexity, enhancing vocabulary richness, and improving text coherence. This study adopts quantitative analysis methods to explore the advantages and limitations of ChatGPT translation, manual translation, and DeepL translation in different application scenarios, in order to improve translation quality and efficiency.

2. Literature review

This article summarizes the application and research progress of ChatGPT in the field of translation. ChatGPT performs well in dialogue scenarios and content creation, and its translation ability is particularly noteworthy in post editing teaching and translation instruction design. Despite its limitations in professional translation tasks, ChatGPT outperforms Google Translate in reducing translation errors. Vocabulary diversity is an important indicator for evaluating translation quality, and ChatGPT has made progress in simulating the use of rich vocabulary, but machine translation is usually not as good as manual translation in terms of vocabulary diversity. Syntax complexity is crucial for evaluating translation quality, affecting the comprehensibility and expressive richness of the text. Translated texts are usually simpler in syntax than non translated texts, and the performance of ChatGPT translation in terms of syntactic complexity still needs to be evaluated. This study aims to explore whether ChatGPT translation surpasses manual translation and DeepL translation in terms of lexical diversity and syntactic complexity.

3. Research Design

Introduced the research design aimed at evaluating the performance of ChatGPT translation, manual translation, and DeepL translation in terms of lexical diversity and syntactic complexity. The study constructed three corpora, including ChatGPT English translation corpus, artificial English translation corpus, and DeepL English translation corpus, to ensure the comprehensiveness of the evaluation. Use TAALED tool to analyze lexical diversity and explore syntactic complexity through Python programming and natural language processing tools. The study used JASP software to perform Bayesian Wilcoxon sign rank test and compare the performance of different translations. The research tools include TAALED, Python programming language, and JASP software. The research hypothesis suggests that CGT may be superior to HT or DT in terms of lexical diversity and syntactic complexity, and the validity of the hypothesis is evaluated through Bayesian factors and Wilcoxon rank sum statistics.

4. Research Results and Discussion

Through comparative analysis, the differences in lexical diversity and syntactic complexity between ChatGPT translation (CGT), human translation (HT), and DeepL translation (DT) were demonstrated. In terms of vocabulary diversity, CGT outperforms HT in terms of content word diversity and vocabulary density, but does not significantly surpass HT in other indicators. Compared with DT, CGT has significant advantages in lexical diversity and richness of vocabulary usage, especially in processing content words and maintaining lexical diversity in long texts. In terms of syntactic complexity, CGT exhibits higher complexity in certain specific syntactic structures, such as the proportion of parallel phrases and verb phrase complexity, but does not surpass HT in indicators such as average sentence length. Compared with DT, CGT shows higher complexity in the indicator of average T unit length, which means it tends to generate longer and structurally more complex sentences. The study also explored how to comprehensively apply CGT, HT, and DT to improve translation quality. By using specific translation instructions, the translation results of CGT can be optimized, while combining the deep understanding of human translation to achieve higher quality translation output. In addition, the dynamic and continuously evolving translation process of CGT promotes collaboration between humans and machines, forming a new mode of human-machine cooperative translation.

5. Conclusion

Summarized the advantages of ChatGPT translation over manual translation and DeepL translation in terms of vocabulary diversity and syntactic complexity, including vocabulary density, content word usage, complexity of parallel and verb phrases, as well as the ability to construct long and complex sentence structures. The research results provide important references for effectively utilizing ChatGPT to improve translation quality, and suggest combining ChatGPT translation with manual translation to improve translation efficiency and quality. At the same time, it is pointed out that ChatGPT may have limitations in specific contexts and text genres, and future research should explore its ability to adjust translation output and its adaptability to different text genres.