Presumably OCR on the Google Books scans then compare to some known text from a different source (Gutenberg or something)