I only skimmed over their ANN when they published it in our local (Croatian) board so I can't be 100% sure, but from what I remember, it looked to me like it was translated by a native Croatian speaker.
Whoever reported it in our local board probably was too eager.
For me they were deleted rather on a hunch, because others had already been deleted.
That's the answer to your question - because if you want the translation to be good, then you need to find native speakers from each local board to translate your ANN. That "agency" most likely uses some online translation tools, and such translations are generally very poor and far from the standards they should have.
Not long ago I had to pay a sworn translator to translate documents from English into Spanish, who has to have a degree in English philology or equivalent and pass a very complex accreditation exam . Well, the translation I had to pay for had some mistakes that I could tell he had used an online translator and had missed a couple of words when revising. I would have translated it better but I am not a sworn translator.
While I find the work of the AOBT commendable, I don't see why the only valid translations for bitcointalk have to be theirs and not those of an agency.
What worries me more is that there is no method by which it can be established that a translation has been done with an automatic translator or not, and that the line between human and non-human translation is getting thinner every second, but in spite of that we have an obsolete rule that is applied depending on the hunch of the moderator on duty and not on objective criteria or a verifiable method.
