For images, the user can say complicated combinations – at Google I/O they demoed “Obama with the French president” and the browser returned relevant results.
Voice powered translation works much the same – choose a language, speak your sentence, and Froyo will speak back to you (and display the sentence on the screen) in the translated language.
“Human intentions”, however, might be the most interesting, as basically it allows the user to just say what they want the phone to do (ok, it can’t do anything, but within the limits of the phone) and Android interprets it and carries out the action. It will be very interesting to test the limits of this feature.
All very cool stuff.