So this will be the first in what is planned to be a series of articles looking at automatic and machine translation, both in the context of Serbian-English translation and of translation in general. In this article we will look briefly at the quality of Google’s automatic Serbian-English translation and explain why we do not think translators and translation companies working in the Serbian-English pair should be too concerned for their livelihoods right now.
An example of Google’s Serbian English translation
Let’s carry out a little experiment first. We took a paragraph of Serbian text (taken from a Serbian Wikipedia article) and pasted it into the Google Serbian-English translation tool.
A human translation from the Serbian to English would read something like this:
Language Translation Services is comprised of segments of text in the source language and of their translation into one or more target languages. These segments can be passages, paragraphs, sentences or phrases. Individual words are not handled by translation memories, these are dealt with by terminology bases. Research has shown that many companies using multilingual documents use translation memory-based systems.
Within a few seconds, Google Translate outputs the following translation into English:
Translation memory consists of segments of the text in the original language and their translation into one or more target languages. These segments can be passages, paragraphs, sentences or phrases. Individual words are not in the field of translation memory, but they deal with terminology database. Research shows that many companies have multilingual documentation systems used to translating memory.
Can you understand it? Apart from a few problems the translator had in identifying passive/active constructions and an unknown word, of course you can! It’s certainly a lot better than any Serbian-English machine translation tool we’ve tried before. If you look at what an old-style machine translation (which shall remain nameless) did to this paragraph, maybe you can begin to appreciate how good Google Translate is:
Prepackage store Astoria oneself off segmentation textual on omnivore Jezebel too njihovog predator on unity whether over civilian Zika. Those segment might lie finders,passim,Berenice whether graze. Pojedinacne res did not of do menu prepacked memorize,vex oneself to them nave terminology baize. Sitar%u017Eivanja pokazuju ought gnome companion quo there are visejezicku documentation terrorist sister with prevaricator memorize.
I beg your pardon? That was supposed to be English, in case you were wondering! And NO, we did not doctor this in any way! Also, if anyone can tell us what”finders” are, they then understand more Middle English than we do!
Google Translate is possibly somewhat less effective with all texts since it had been with this one, however it’s definitely a significant improvement over the above mentioned case in practically all instances! So maybe translators need to think twice before blowing machine translation from Serbian to English (and other languages, even if that is anything to go by).