Google pulls back the curtain on its new voice search, sums it up in this graph
 

Google has revealed just how it harnesses your idle curiosity on every subject imaginable to supercharge its voice search. A database of 230 billion googled words was fed into a language model that can then work out the probability of what you're going to say next. Mountain View researcher Ciprian Chelba explained that one example of this is if you say "New York," you're statistically more likely to say "Pizza" than "Granola," regardless of any new year's resolutions. If you'd like to learn more, you can find the algebra-packed original paper down at the source link.

http://www.engadget.com/2012/11/01/g...h-predictions/