Researchers have created a benchmark called MHER to help computers accurately identify individuals from the Mongol world who may be referred to by different names or scripts. The benchmark includes pairs of name attestations from various sources and evaluates how well AI models can reconcile these identities using historical evidence. Tests show that using source-grounded evidence, such as mention-by-source information, significantly improves accuracy in identifying distinct individuals. However, the study also highlights limitations, including cases where surface-level name similarities lead to incorrect identity merges. --- Why it matters: This matters because accurate entity reconciliation is crucial for tasks like historical record-keeping and cultural preservation. The MHER benchmark provides a controlled framework for studying how AI models use evidence to make decisions, which can inform the development of more effective NLP systems for historical applications. Source: https://arxiv.org/abs/2608.23507