http://www.zdnet.com/apple-stores-your-voice-data-for-two-years-7000014216/
Google does it too, & Microsoft probably does it, only they may not have a delete date attached. Translation -- speech to speech, speech to text, text to text etc. -- is extremely hard, & for the most part only improves by being corrected when it gets it wrong. Lots & lots of data [about when it's wrong] is helpful. The potential problem though, is not everyone bears in mind that what they want translated is going to be recorded, so they don't watch what they say as if it was going to be tomorrow's newspaper headline. Allegedly no one is ever going to mine all that data for something they'll find useful, but people do stuff that they shouldn't, that's against the rules, all the time [that's why we have law enforcement]. And while you may not be identifiable from the raw data, any specifics, names, dates etc. are as easily read as if you posted them to your public blog.
It's often been said, never post anything on-line you wouldn't want printed in a newspaper -- remember the same thing might well apply to talking to your cell phone, using Google Translate etc.
