OCR is the hard
part and I've yet to hear of anything that is
even close to remotely acceptable. At say 1000 words/page
a success rate of 99.9% still leaves you with one fix up
per page. That's a good chunk of work for even a small manual
(say 200 pages). It's a lot of work for an RT-11 manual set
or similar!
Sounds like an ideal thing for a distributed project.
Give everyone who registers a few pages to proof read, combine into
finished work. If you wanted cross-checking you'd just make sure that
different people got different batches at different times, and diff the
results.
Gordon.
However, for technical documents, I'd like to be able to see the
original in case an OCR scanno occurs in just the wrong place.