I'm a professional translator, have been for over 15 years now. I've seen a lot of technology companies try to perfect machine translation (i.e. "automatic translation," "computer translation" etc.) during the entire course of my career. Google wants to do it for "free", and has tied it into their visual processing AI so that you can point your phone camera at a menu and magic happens. A horde of specialized service providers want to do it for very large amounts of money, backed by enormous databases for their specific field of medical transcription, legal documents, patents etc. And the accuracy of machine translation has absolutely been getting better -- more slowly than I think a lot of folks outside the industry were expecting, but there's no denying that machine translation will one day lead to my early retirement.
But that day is not today, nor any time soon, really. And the first and most significant reason is that machine translation fails badly. I mean, when it translates accurately, it's fine, and these days you wouldn't be able to tell whether sentence A was translated by an algorithm or a professional typing away. But sentence B? Well, where I'll hit some dictionaries, search literature, and give you intelligible and accurate English, the algorithm will read like Yoda dropped acid and then rearranged words by whacking Ctrl-X and Ctrl-V with that cane of his. That isn't just Google Translate, that's very expensive services that are marketing themselves as just as good as hiring a human. Don't believe it: their trick is, there's still a human (sometimes me, these days) getting paid full rates to revise and edit the machine translation. Pay no attention to the man behind the curtain.
For those of you who are more on the software engineering side of things, why do all machine translation services have a failure state that reads like their backend hardware is an old Reverse-Polish Notation calculator that's been whacked with a hammer?