Otherwise I have no idea what he wants to talk about...
AI answers Factoid and Definition (Other) questions, and hence AI is the answer to the NIST TREC challenge.
In order to answer AI first structures a recieved search request, ie transform it to a certain format. Then AI filters the extended by synonyms the search request through a personal profile, purging all random phrases; where the profile consists of paragraphs, as of subdivisions of a written compositions (which consist of one or more sentences, deal with one idea or give the words of one speaker). Next AI annotates the request by the most relevant paragraphs, expanding the request to a full and meaningful text. And finally, all found prospective answers are filtered through the search query and profile paragraphs.
If the profile contains no relevant to the search query information Machine Learning can be applied, a search for new texts and paragraphs performed (which should be structured and added to the personal profile for further use).
The answer found can be displayed for a searcher or used to drive a driverless car.
Hopefully Sebastian Riedel would mention the above AI technology and my (Ilya Geller's) role into the creation of the technology. Otherwise I have no idea what he wants to talk about...
Biting the hand that feeds IT
