>   raw, unproofed text by OCR.

i love the way this sounds -- "raw unproofed text by ocr".

the implication is that the results are rather disastrous...

and sometimes, granted, they can be.

the chain is only as strong as its weakest link.

however, if the paper-book is in fairly good shape, and
its text is rather straightforward, and a best-of-breed
scanner is used, and the scans are done carefully, and
then properly treated (e.g., deskewed and regularized),
and o.c.r. is done with a best-of-breed program, then
auto-clean-up tools combined with normal spell-check
will produce quite accurate text, thank you very much...

and given advances in o.c.r. technology, and other tricks,
a need for "human" proofreading of every word on a page
could be eliminated for all but the most difficult of books.

and my bet is that that day will come long before the one
when we have machine-generated language-translations.

-bowerbird