It's true that mistakes can be made by anyone, but screwing up someting like a Star Wars reference is a lot different from screwing up names and story lines. A human mistake is just that, a mistake, something unintentional. when a machine makes a mistake, it's not really a mistake but is infact what the machine "thinks" is right, and will provide the same mistranslation every time.
It won’t actually! Given a system that doesn’t learn, okay, but most MT solutions are under constant development. Give Google Translate a passage that it has trouble with. In a few weeks’ time, that passage will have a different wrong translation. The machine is never deluded into thinking it is right; that’s not how AI works. Actually, it is always coming up with something it considers “okay”. Such a suspiciously nondeterministic-sounding approach may sound imprecise, but it’s a good enough approach to fix problems with space probes that are too far away to make contact with Earth at the speed of light, and it’s good enough to make the U.S. Army logistics more efficient than any human can manage.
A TM system, on the other hand... garbage in, garbage out. The translations in a TM are typically human-supplied, so anything erroneous is also human-supplied. My own TM has only my errors in it, if it has any (and only contains text from the very same project, for that matter).
People that are incompetent tend to not just use but depend on translation software, so when both sides of the coin are useless it doesn't matter if it lands on heads or tails. Personally I don't like the whole conept of online translators and memory translation, but if you can find a way to use them safely and for something like video game fan translations, I guess there is no harm.
The thing is, a lot of really professional and competent text translators use these technologies all the time—you could say they
depend on them, too. Five years ago, a survey of translators revealed that over 80% used TMs. Four years ago, another survey revealed that of 430 translation job postings, 95% listed TM skills as a prerequisite. These trends have only increased since; studies have shown that the use of TMs increase (human) translator accuracy, throughput, and consistency. Since that time-saving factor makes more money, you have the good, the bad, and the ugly all using it these days.
Professional translators use MT, too. Not for anything coming close to a final translation, of course, but they might use it on, say, large bodies of text when they need a general idea without jumping in (for sanity checking, too, since computers don’t accidentally skip over lines). And many uses for MT have really only begun to be exploited.
So you can dislike these things all you want, but you really shouldn’t knock them. They do solve real problems, and yield helpful results in many places. For people who know what they’re doing, these technologies actually decrease the chance of errors.