In cross-language evaluation, literal translation may produce queries that are awkward or do not match local expression habits, reducing evaluation realism; a more reliable approach is for native speakers or localization teams to rewrite the queries according to the same intent.