It stores the data as English text, parseable into Prolog predicates (ie, key/value data chunks), and you have a special editor which shows you the predicates generated by the English you just wrote.
Does this sound useful to anyone? I'd love to pick up on it.
The parser for English I wrote is in Lisp, and works very well (it's recursive, etc).
And of course, you don't have to make a new database, you just do the usual Wiki editing and just make sure the text you enter is parseable by the latest parser.