Tag
This paper measures the self-referential evaluation loop when a knowledge base is used as the gold standard for entity-level machine translation in low-resource historical domains, showing that gains from KB injection are confined to overlapping segments and do not reflect true translation quality.