GENERATIVE-AI-ORIENTED MULTIMODAL LINGUISTIC LANDSCAPE: EMPIRICAL EVIDENCE FROM YANCHENG CULTURAL-TOURISM SITES

Traditional linguistic landscape research largely concentrates on the semiotic analysis of physical signage, while dynamic digital linguistic symbols generated by large language models remain understudied. Taking three types of cultural-tourism spaces in Yancheng as field sites, this study collects three corpora: physical signs, digital navigation interfaces and AI-generated texts, and constructs a local multilingual cultural-tourism corpus. Controlled experiments are conducted to compare the outputs of a general large language model and a local-knowledge RAG model on sea-salt cultural proper terms and intangible heritage narratives. Field data show that the physical linguistic landscape of Yancheng tourism follows a pattern of “dominant Chinese, limited English coverage and fragmented Korean distribution”, and translations of historical cultural concepts commonly suffer from semantic simplification and erosion of cultural information. Model results reveal that general large models exhibit high hallucination rates when processing low-resource local cultural concepts such as Zaohu and Tuanjian; after connecting to the local corpus knowledge base, term translation accuracy and cultural fidelity improve. This study identifies a prominent semiotic disjuncture between digital and physical linguistic landscapes in tourism settings: generative AI brings instrumental empowerment alongside risks of cultural drift. By incorporating algorithmically generated texts into the multimodal linguistic landscape framework, this paper provides an empirical case for research on local cultural-tourism linguistic signs.