HalluTruthQA: A Fine-Grained Benchmark for Hallucination Detection, Localization, and Explanation in Arabic Question Answering
The benchmark adds span-level error marking and explanations to Arabic hallucination testing.
HalluTruthQA contains 2,400 expert-curated Arabic QA examples across Islamic knowledge, history, science, and geography. Each item includes a verified answer, hallucination label, six factual-verification candidates, and, when needed, character-level erroneous spans with human explanations and hallucination types. The authors test Allam, Falcon-H1, Qwen32, and Silma zero-shot, finding different leaders for detection, localization, verification, and explanation. ArXiv · AI/CL/LG's note
HalluTruthQA contains 2,400 expert-curated Arabic QA examples across Islamic knowledge, history, science, and geography. Each item includes a verified answer, hallucination label, six factual-verification candidates, and, when needed, character-level erroneous spans with human explanations and hallucination types. The authors test Allam, Falcon-H1, Qwen32, and Silma zero-shot, finding different leaders for detection, localization, verification, and explanation. ArXiv · AI/CL/LG's note
score 5