NativQA Framework: Enabling LLMs and VLMs with Native, Local, and Everyday Knowledge

arXiv:2504.05995v3 Announce Type: replace-cross Abstract: The rapid progress of large language models (LLMs) raises concerns about cultural bias, fairness, and performance in diverse languages and underrepresented regions. Addressing these gaps requires large-scale resources grounded in multilingual, local, and cultural contexts. We systematize and extend the earlier NativQA framework to multimodality by adding image, audio, and video support, enabling scalable construction of culturally and reg