The thing is, the people receiving the message "Swan 930" aren't confused. They know it means a pub called the Swan and not an invitation to see Swan Lake the ballet.
My point is that since siri can't accurately translate an 8 character text message that I send at least once a week, I'll just continue to send the text the "old fashioned way" with my fingers. And because siri can't handle one of the apparently simpler things that I do on the phone, I don't really have the trust to use it for anything more complicated.
I appreciate that siri is beta... but I'd be interested to know how it could or would be "trained" to an individual's voice. If siri doesn't translate, say, the body of the text message correctly, then I'm either going to cancel and send the message the old fashioned way, or manually correct what siri has come up with, and then send it. In either case, how is siri getting the feedback of what it thought I said vs. what I intended to say? Is the copy of the text message being sent up to the siri servers to allow comparison with the speech sample?