I was struck by the example pasted below, in which Leo transcribed an identically written final “a” three different ways in three successive words. I have some inkling of how machine learning works, so I’m not sure how helpful it is to provided such an example. (Hopefully, somewhat!)
But this example relates to a basic question I’ve been wondering while experimenting with Leo. To what extent does Leo already, or might Leo in the future, be able to learn from a specific manuscript or scribe and apply that to reading other works in the same manuscript or by the same scribe. For instance, in the example below, if Leo was confident that the final letter of pericula was a, but less certain about the final letters of damna and exilia (transcribed as piliae), but recognized that the three words had the same final letter, would it be able to improve its accuracy? And does Leo learn from the corrections that users make? And if knows that other images are written in the same hand, can it apply such knowledge selectively, rather than it become part of its universal method of reading all manuscripts?
