The Effectiveness of ChatGPT in Translating Chunky Construction Texts in Chinese Political Discourse

Abstract

From the theoretical perspective of Chinese-English temporal-spatial cognitive differences, this study employs a self-constructed bilingual parallel corpus of political discourse totaling one million words. It extracts chunky construction texts with latent agents in the source language (Chinese) and corresponding texts of target language translations (English) for a comparative experiment between ChatGPT translations and those of human translators. Driven by the “Human-AI Interaction Model”, three different prompts were set up to conduct three rounds of translation testing with ChatGPT. The study discovers: 1) ChatGPT has limitations in actively understanding the implicit elements of the source Chinese texts that are driven by a strong spatial preference, which are also reflected in the target language English translations; 2) The “Human-AI Interaction Model” can guide and train ChatGPT effectively through continuous optimization of prompts, enhancing AI abilities in understanding the source language and achieving effective target language translation; 3) Currently, ChatGPT cannot replace human translators in translating chunky discourses of Chinese, but the collaborative interaction between human translators and AI is highly effective. By optimizing the design of prompt instructions, this research offers cognitive reference and operational paradigms to improve ChatGPT’s recognition and cognitive reconstruction capabilities of Chinese-English temporal-spatial differences, providing insights for future research, translation practice, and teaching in related areas.