Characterizing Cultural Localization in AI-Generated Stories

The paper introduces a method to evaluate cultural localization in AI-generated stories, distinguishing between templated and holistic localization. By analyzing narratives from five models across 125 topics for 193 nationalities, the study finds that only 9-17% of the vocabulary accounts for national variation, indicating reliance on a shared narrative template. Additionally, it assesses cultural markers for stereotypicality and offensiveness, revealing that markers from many Global South countries are often deemed offensive, which is crucial for practitioners aiming to create culturally sensitive AI content.