A Model for Assessing the Linguocultural Adequacy of AI-Generated Folklore Texts

Large language models (LLMs) such as ChatGPT, Claude and Gemini now retell, translate and invent folk tales in seconds, and their texts are already used in classrooms, children’s books and online media. Yet there is no established way to judge whether such a text is culturally adequate. Standard measures of machine translation quality, such as BLEU or general fluency ratings, can tell us whether a text is grammatical and close to a reference, but not whether the Dev is still a Dev, whether Zumrad’s sabr is still a virtue, or whether the hero still greets the monster as bobo. Research on AI-generated folklore has shown that cultural losses occur at many levels, from names and idioms to metaphors, values, emotions, speech etiquette, narrative structure and visual imagery, but these findings have not been integrated into a practical evaluation tool.