Well (or walmer 🙂 ), Jott does seem to be able to correctly transcribe some messages, but I’ll be very conscious of diction and slang (both of which are addressed by the hints Jott.com gives on maximizing accuracy) in the future. I’ve not been able to find any definitive answer on how the service works, but I’ve read a few things that lead me to believe that the decisioning process might be two-tiered with software taking point and human cognition handling backup when that fails (or at least when confidence is low). The details of when a living person actually interferes with the process may not be as easy to pin down as all that, but the logic remains: something happens after something else happens. Prove me wrong if you can.
If this thinking is correct, the next logical assumption would be that the human contingent would cost as little as possible. And, of course, that leads to other assumptions… and explains why phrases like “Freaked out cats” might be eliminated and others might get turned into words like “walmer.” Of course, some of the results have been absolutely brilliant, with certain expletives spelled out as if by native English speakers. The resulting text of one test message I left myself ended with “who-ha, who-hoo,” which was actually a pretty good representation of the jovial nonsense sounds I recorded. The whole text of that particular message was absolutely perfect, and that lends reinforcement to my thinking. Some people are just more intuitive than others, and some definitely speak and understand spoken English better than others.
Like my friends Perry and Wayne. One of them is way better at most things than the other one.