Originally Posted by bobcdy
In the past I've tried (and tried and tried) to train FineReader 10 to handle old English (with ligatures) texts, and I concluded from my efforts that FR10 was a very slow (no?) learner. Eventually I gave up and just used MS word auto-correct options, eventually reaching a fully corrected version of the text - but this was very time consuming because FR10/11 is confused by the ligatures; for example, it would interpret the c-t ligature (probably the most difficult for it) in many different ways.
Finally I purchased the upgraded FR XIX Fraktur edition (Recognition Server v3) when ABBYY greatly reduced its price. Still pretty expensive but certainly worth the price if one works with 18th/early 19th century books. What a difference! RS3 reduced correction/proofing time by many days per book!
Well, some of our "blackletter libraries" in Finereader (I have about twenty for different font varieties) started life in Finereader 6.0 ten years ago, and even if we can't get 99.9 correctness (those "long" s's and h, and k,l, and t will probably never be anything but guesses), they are by now pretty good at what the do. The Finereader Fraktur edition's price is still pretty stiff for a small-time publisher as we are. Our latest project is a danish edition of Walter Scott's "Old Mortality" from 1870 - at almost 600 pages the cheapest FR XIX licence would only last 4 volumes ... Sometimes we also have to work with bleak photocopies of magazine pages from the early 19th century, which I doubt even the Fraktur Edition would make much of.
It's funny how the human eye and brain is able to fill out the voids in such document, so we actually can get something legible out of it, isn't it?