machine translation: what is it good for?

I will get back to posting terms soon, but one more meta comment for now:

What I tell organizations, when they ask me about it, is:
Machine translation should be used on non-essential inbound texts ONLY - to get an idea (sort of) of what other folks are saying. If you don't want to run the risk of being wildly misunderstood and saying something offensive you should never use it for outbound texts. You should also never ever use it as the first step in a human translation project. For some reason a lot of activists seem to think that putting it through a machine first will make the human work easier, but honestly it makes it harder, because once you have seen it rendered wrong, it will be harder to think of how to say it right - and your human translation will never be as good if you start with the machine weirdness. Just start from scratch.