HalluTruthQA: A Fine-Grained Benchmark for Hallucination Detection, Localization, and Explanation in Arabic Question Answering
HalluTruthQA is a fine-grained benchmark designed to evaluate how well Arabic large language models generate factual answers without making errors. It moves beyond simple yes/no detection by assessing where and why models hallucinate, including identifying specific erroneous text spans and providing human explanations for the mistakes.