It's been in the labs since before I gave up my proper job because I worked on the original project. That was ooooh 2005.
And in English - serendipitous discovery means making connections through luck but actually doing it using semantic language.
So let's say for example you are a medical researcher working in the field of sudden heart failure. If you read all the medical literature you'll know what other folks are researching and you may have hunches of your own, but what if you could do reverse searches on any subject - not just medicine. Things like for example, cookery or burial rituals etc. Random things.
Let's imagine there is a tribe in Africa that never had sudden heart failure, and then by tracking back through all the books in the world (and all the papers etc) you could find out that it's because the tribesmen eat a certain berry and that the berry has a chemical in it that can eradicate the disease?
Now, that's where it gets cool (and profitable).
A more common example is the use by the US DoD where they used text mining to analyse ownership as well as content of websites and to find that a certain terrorist group were actually funding the creation of school websites for children as young as four in order to indoctrinate them.
Or the connections between kiddy porn and drugs.
The whole thing is fascinating.