This final kind of mapping is very much a free-for-all. Essentially, it allows you to map references to any of your other mapped entities interchangeably. This is done by providing two columns , one which contains the name of the table to which each reference is being made, and another which provides the ID within that table of the specific entity of interest.
You can't maintain any sort of foreign key constraints in such a loose relationship. It's rare to need this kind of mapping at all. One situation in which you might find it useful is if you want to maintain an audit log that can contain actual objects. The reference manual also mentions web application session data as another potential use, but that seems unlikely in a well-structured application.