Comparing DeepL and ChatGPT in Translating Chinese Tourist Public Signs: Accuracy and Pragmatic Appropriateness

Abstract

AI-assisted translation is becoming increasingly visible in tourism communication, yet the translation of Chinese tourist public signs into English remains underexplored. Unlike longer tourism texts, public signs are condensed, function-sensitive, and pragmatically constrained. Many Chinese tourist signs also contain figurative, descriptive, or rapport-oriented wording, which complicates translation. Against this background, this study compares DeepL and ChatGPT in the translation of 25 Chinese tourist public signs into English, focusing on accuracy and pragmatic appropriateness. ChatGPT is examined under two conditions—a minimal-prompt condition and a domain-sensitive prompt condition—whereas DeepL is tested through direct source-text input. Using an exploratory qualitative comparative design, the study identifies recurrent tendencies across formulaic, warning, environmental, and rhetorically marked signs. The analysis suggests that DeepL is relatively stable on conventional signs, whereas domain-sensitive ChatGPT more often produces concise and context-appropriate English signage, albeit sometimes at the cost of reduced rhetorical texture or weakened rapport-related nuance.