A Comparative Study on the Translation Quality of Eco-cultural Terms in Tourism Texts of Qinghai Province: An Empirical Analysis Based on Large Language Models

Abstract

This study conducts an empirical comparative analysis of translations generated by four mainstream large language models (ChatGPT, Google Gemini, DeepSeek, and Qwen) for 300 Qinghai ecotourism-related eco-cultural terms across four categories (Natural Landscapes, Ecological Resources & Conservation, Tourism Activities & Facilities, and Culture & Folk Customs). A multi-dimensional evaluation system covering accuracy, cultural connotation retention, terminological consistency, and readability is employed to assess each model’s performance. The results show that Google Gemini achieves the highest overall performance, excelling particularly in accuracy for natural landscape and ecological resource terms, and demonstrating strong capabilities in cultural connotation retention and readability. Qwen ranks second, with notable strength in cultural connotation retention and balanced performance across other dimensions. DeepSeek follows, maintaining stable and consistent output with good terminological uniformity. ChatGPT offers acceptable readability and natural expression. Based on these findings, the study proposes targeted optimization strategies and constructs a reference translation corpus, aiming to support the international dissemination of Qinghai’s ecotourism culture and to enrich empirical research on LLM-based translation in regional ecotourism contexts.