This could be done by a combination of statistics/histograms to identify empty columns and heuristics, but can't be perfect.

Small number of data would result in ambiguity.

But a semi automatic approach, where you choose a result of your liking sounds very feasible.

Think of the way Excel asks about the best way to import CSV after showing a best guess. °

Would that be good enough?

The first step would require collecting sufficient training data.

updates

°) e.g. Text-Import-Wizard