any intuition of why that that would be the case for NLP? Certainly when you have something as large as the datasets being used to train large transformer mode

any intuition of why that that would be the case for NLP? Certainly when you have something as large as the datasets being used to train large transformer models, SOME of the data is already repeated. Why would one more epoch make it worse?