3 ms·
You'd think that the OCR process would somehow call attention to words that have a high probability of being wrong and especially of being wrong in a problemati
by robbrown451 8y ago
You'd think that the OCR process would somehow call attention to words that have a high probability of being wrong and especially of being wrong in a problematic way. You don't want to require humans to read and sign off on everything, but with something like that, it shouldn't be that hard to have something that is very quick for a human to see the scanned image and compare it to the transcription, simply on the basis of the word "anus" being in there.
- lathiat 8y agoI was reading “Creative Selection” by Ken Kocienda last week. Goes behind the scenes of him designing the iPhone keyboard early in its development (good read) In any case he mentioned there is a hate word dictionary specifically so that the autocorrect never suggests such words even if they seem to be a close match. You basically have to type those words perfectly. In another related bug Xerox document centres which weren’t even technically doing OCR were changing numbers from 1 thing to another in scanned IMAGES due to high level compression substituting numbers - much more dangerous! https://www.theregister.co.uk/2013/08/06/xerox_copier_flaw_means_dodgy_numbers_and_dangerous_designs/ https://www.theregister.co.uk/2013/08/06/xerox_copier_flaw_m...