I don't know how to determine this, quantitatively or otherwise. It's an interesting question once machine translation gets better, but for now I consider it obvious that something like Google translation does not know what it's doing. Having access to and having translated a large existing courpus of text is obviously not enough, as Google certainly has analyzed more text than a human translator does, and still is wrong whenever there is the slightest possibility of ambiguity (i.e., all the time, in practice).
Anyway, TFA was not about machine translation, but AI. A human translator who translates a text knows that he is translating a text. I am not worried that a computer will, by 2045.