The information is probably lost, training on very long inputs is expensive, we usually train on short inputs and only send longer ones at the end. So the author name would not appear together with all the chunks unless it appears on every page in the header, assuming the headers are not stripped out first.