My guess - and it is nothing more - is that there may be a possibility of learning procedural tasks, making them more and more automatic, provided that one begins the procedural task prior to falling asleep, as in this experiment.

As noted in the article, brain centers associated with attention are deactivated during sleep, so the possibility of acquiring semantic or episodic knowledge seems remote, as does the possibility of acquiring procedural knowledge without the "start it just before falling asleep" condition.

I thought the same thing, especially right after they explained that the subjects had no memories of the words they "responded" to while they were asleep. I know learning != memorization, but it seems like being able to remember a lesson is at least a part of learning.