How hard would it be to do some kind of OCR (maybe combined with machine learning) to “scan” anything that looks like a table into a more accessible format? It seems like there’s a lot of room for ML to improve various aspects of screen readers and other assistive technology.
I claim such a solution would be morally wrong (I am not going to comment on whether such a solution would be good in practice or not) because latex largely throws away the data it has on the table structure and this solution tries to piece it together again. I claim a morally better solution is to have latex not throw away the information it had, but I concede it’s often harder to modify something big and old and complex than to produce something new which is supposed to always work.
Reconstructing content and formatting from a PDF through OCR, machine learning, etc. may not be the best possible solution. But it would still make more information accessible to more people. So I must vehemently disagree that it would be morally wrong.
Morally wrong? What the actual f. The proposed is a perfectly reasonable and decoupled solution that would solve the problem for troves of existing, unstructured paper documents, too. In fact, it likely already exists.
