I used to work in ontologies, and what I learned is that people would rather get an 80% heuristic solution for dirty data rather than a 100% correct solution for data they have to clean.

The big job sites have resume parsers that work well enough from a PDF or Word doc, and then they don't have to worry about you forgetting a close tag or a mandatory field. Sure, stuff gets lost, but they get 80% of a billion resumes rather than 100% of a million of them. They can't exchange data or even trust what they have, but it's good enough for them to make money. Meanwhile, a competitor demanding good data from its clients never gets off the ground.

Anyway, every data format for human information ends up being either vague (to allow in everything) or impossible (see the myths that programmers believe about names, time, addresses, etc.) You end up giving a string for each field... Then give up, just accept any string, and hope for the best.

Everybody wants Google or some machine learning solution because the formats never work for the information people want to convey. Better solutions could exist but the hacky ones are first to market, in a natural monopoly where there really only needs one good enough product.

If you think a lack of a good resume format is bad, look at electronic health records. Those are far more important to be correct and exchangeable, and even there the cleanup effort is always enormous.