Just as with redacted documents (consistently blocked terms) or bad OCR jobs (wrong or missing characters), even if only a certain percentage comes out unmangled it is more readable than having no data at all.
replyA stable base corpus and some dynamic programming will allow you to clean up the remainder[0].