terminological bleg: "structure learning"

When people say "structure learning", does this refer strictly to learning the structure of graphical models? Or more broadly learning distributions over structures of any kind?