While the keynote is scripted, I doubt that its done to that level that they would fake it - the timing would have to be too perfect where its clear that there is some flexibility (you saw that with Scott talking about things). Its possible that several things could have been at fault - processing error (due to data swamping) or the system not hearing it due to a mumble from Craig.
Not to mention that dictation is inherently prone to such errors even with common words. Thats why its hard to pull off - there are so many ways for errors to creep in randomly. There are tons of examples where things go awry during a keynote despite Apples best efforts to control things.
But voice dictation isn't Siri. There's no parsing of the content, and it's probably not doing the recognition in the cloud.
I'm sure it was scripted, as was the Game Center face-off with the "Stig" ... I mean, Racer OSX.
I can see why they scripted it, it helps keep things predictable. Unfortunately, when the guy commits a gaffe like this, it would have actually looked better for Apple/Siri for Dictation to pick up the extra word, or a cough, a sneeze, a burp or any sort of mistake... rather than ignore it and clearly show the dictation was staged.
Anyone else notice that the voice dictation tweet demo was faked? Craig said "show off mountain lion" but the text that appeared was "show mountain lion" (59:29 in video).
Seems like it would be easy to just force Siri to return the scripted response regardless of what the input sounds were.
There were several uses of Siri during the demo, and there were no errors of any kind.
Again, trivial for apple to fake, and in their interest to try to overcome the general perception that Siri isn't perfect.
Why was this downvoted? Some simple programming on Apple's side could have easily faked this. Just like how Scott explicitly told the audience that he ran a simulator to demo the Turn-By-Turn directions.