Exploring Bengali Creative Storytelling Capabilities of Large Language Models Across Cultural Variations

Abstract

Large Language Models (LLMs) excel in fluency but often struggle with originality, suspense, and emotional depth in storytelling. This study evaluates their creative storytelling capabilities in Bengali, a language with significant dialectal diversity. Using three narrative prompts across single-dialect and cross-dialect settings with initial results and story continuation, we analyze AI-generated content for coherence, creativity, and cultural relevance. Native Bengali speakers provide qualitative feedback, highlighting key challenges such as dialectal fidelity and narrative richness. Our findings emphasize the need for culturally adaptive NLP models to enhance storytelling in low-resource languages.