OCR and AI

An interesting project which I hope takes off. Since the probate calendars are printed, OCR seems to be quite viable with checking as proposed. However, this got me thinking about the use of OCR on HANDWRITTEN documents, for example the wills themselves. I have always believed this to be on the fringes of viability, but clearly times move on. I Googled "ocr of handwriting" and from a plethora of offerings, I picked one of the first listings handwritingocr.com which claims to use artificial intelligence. This is a commercial site which will transcribe100 pages for a $15 plus $0.15 per page for additional pages. For heavier or regular users, there are more economical plans. (I add that I have no commercial interest in this or any other site). The site offers to OCR five free pages to allow you to test it. I chose to try a handwritten 1902 will which I had recently ordered and a page from one of the children's refuge admission registers on which we are working at present. The will was written in a flowing hand and was quite readable. It was transcribed very nearly word-perfect. The only error was the mysterious appearance of the word "cox" in the middle of a sentence. In the original the "word" was unclear until with magnification I discovered it was the letters "con", the start of the word "contained" written prematurely in error by the author and crossed through. Even the testatrix and witness signatures were correctly read as was the rather unclear probate registry stamp, which was slanted at about 45 degrees. The admission form was a mixture of printed text and manuscript. Again, the handwriting was quite readable and again, the OCR rendering quite faithful. One interesting error was the misreading of "E T Home" as "S T Home" while a little further on, "Egerton St" which used the same "curly" "E" was correctly rendered. A couple of rather badly written capitals were mis-read - "Jany" became "Davy" and "Oates" became "Dalt". The difference between traditional OCR and AI enhanced OCR is clear. Where traditionally unclear letters would be individually translated into "best guess" often leading to gibberish, AI clearly tries to map what it finds into something which is (or might be) a real word (this might prove a pitfall for proofreaders). The tests were on clear handwriting and at some point I will challenge it with Secretary Hand, but my experience is encouraging that these tools will be of practical value for future projects. I already have a lengthy family manuscript which I have in mind. Have other members used online OCR for handwriting? I'll be interested to hear.