It covers the limitations of search engines today. Documents and queries aren't really understood at all, because they're still viewed as tokens.
They tested a method where they consult Wikipedia for knowledge. It hasn't worked so far but there has been a lot of research on it. Wikipedia is useful for semantic relatedness (see Wikirelate).
The idea is treated like an ontology here. Wikipedia how ever is not a formal structure so it's not easy. It's believe that it can be used in this way, using HCI rather than AI or NLP.
Koru is introduced for exploratory search. It works well, although improvements are necessary. "Wikiminer" is also demoed.
For an awful lot more detail and interesting information, take an hour, sit back and enjoy the talk.