Comment by maxbond

The data processing inequality holds regardless of how many layers are in your neural net (processing data does not increase it's information content). You can impute missing data, and with something very regular text it could work pretty well, but that way lies hallucination.