AfrIFact: Cultural Information Retrieval, Evidence Extraction and Fact Checking for African Languages
AfrIFact is a benchmark dataset with over 18,000 claims in ten African languages and English across health and news domains. It tests embedding models for information retrieval, evidence extraction precision, and LLMs for fact-checking. Findings show significant gaps in cross-lingual retrieval and multilingual fact verification but demonstrate performance gains through few-shot prompting and parameter-efficient fine-tuning.