That’s a tough question. You see, the thing is that some machines are built to translated specific types of contents only. For example, if I build machine to translate user manual of dell laptops (very specific) and I have loads of good quality data I can feed the machine, it is very likely that the translation will be very good, fluent and enough for the user.
But so far there isn’t a single machine that can translate all the contents fluently because they need always more good quality data and training. LIke Google translate is good for a broad types of contents. Sometimes it is really good, sometimes it is not. So no, not fluent for all the content types we have!
Another thing that I would add to Sheila’s answer is that typically in machine translation you have two components – one that is responsible for translating (words or phrases or whole sentences) and another that optimises this sentence to the linguistic specifics of the target language. If the second model is very good then the translation can be very fluent and sound very good. But actually be a very bad translation of the original text. That both Sheila and I have seen and researched extensively at the end of last year. And yes, we are working on fixing all types of errors and make machine translation better. So, stay tuned đŸ˜‰
Comments