We all know that there are many corners in a box that we can pick around and say "Hey, here's dust, why don't you clean it up?" or "It's too dark here, lighten it up for me."
I don't say I hate those petty improvements (I love them!).
But, what is our (and the developer's) grand design?
Do we really want a very smart, near-perfect AI translation engine, one that is affordable and hence could make us jobless?
And then again, it would be interesting to observe, how CafeTran's fix that MT result up with your own glossary feature performs its healing work here, to ensure consistency.
Just got controller, regelaar, kontroler suggested for one and the same source term.
One feature that statistical MT engines do miss, is a rigid terminological consistency mechanism. They suggest target term x in segment 234, singular, with article, and target term y in segment 236, plural without an article.
It's our task to polish that.
But, like I said: in the end these will be the only texts that will be offered to us human translators. The others (more simple, more boring?) will be translated automatically.
>Do we really want a very smart, near-perfect AI translation engine, one that is affordable and hence could make us jobless?
Well-written texts, preferably with controlled language, will be suited for AI (MT).
All other texts will require HI (human intelligence). They are the intellectual challenge en will keep us alive and kicking.