Indeed, this approach is good idea! Comparative translation, probably also positively biased by the genre and author (to distinguish between, say, XIX and XX centuries etc.). Probably this is even possible using GPT-3.