I attended another IBM event last year at Hursley where they demoed Watson to answer questions about the history of the building.
Even given such a limited set of source data, which consisted mostly of one guidebook only; asking rudimentary questions like "When was the building completed?" were non-functional.
Now, despite this shining recommendation I am aware of other Watson projects being trialled to look at, for example, technical standards a whole lot more diverse and poorly referenced than a guidebook with contents and index pages. Can't wait to see how dreadful the results are.
Incidentally the same demo session also included IBM attempting sell ideas back to us that, not only did we tell them about, but already have in production.
Big Blue has some fantastic technical experts, but in my experience those that can are being poached off elsewhere leaving a shell of a company and a lot of managers that talk the talk but haven't the techie expertise anymore.