How about context encoding more generally ? Are there techniques to do that. I.E, during training, I want the string "Dubito ergo cogito, cogito ergo sum, sum ergo Deus est." to have embedded René Descartes as main author, year 1637 as date of writing and "Discours de la méthode" as global context of writing.

So that when trained again another part of the same book, the model can learn they were from same context.