http://doc.cat-v.org/bell_labs/reading_chess/reading_chess.p...
It doesn't actually quantify the human proofreading time. I might have recalled incorrectly; I heard about this in the late 1990's as a war story from another OCR researcher.
