The Politeness Paradox: Why LLMs Fail at Persian Taarof and How Researchers Are Closing the Cultural Gap
New research exposes critical flaws in large language models' ability to navigate Persian taarof—a complex system of ritual politeness—with accuracy rates 40-48% below native speakers. The study introduces TaarofBench, the first cultural benchmark for Iranian social norms, and demonstrates how fine-tuning can boost alignment by up to 42%, revealing the limitations of Western-centric AI training.