Is there an objective measure of how well DL generates text (as in the article)? No. But there is one for how accurately it predicts test text, not in its trai

Is there an objective measure of how well DL generates text (as in the article)? No. But there is one for how accurately it predicts test text, not in its training corpus.