Machine translation progress
Machine translation seems to have made impressive progress in the last few years. Running e.g. a Wikipedia Article article through Google Language Tools or Systranbox.com (which produces the almost identical translation) actually creates a more than just rough translation (as I think they did maybe 5-6 years ago); it's not really "correct" of course (as a human written equivalent article), but... "reasonably understandable" definitely, I'd say.
Now, I always thought of MT as an end-user tool; one would consciously use it, knowing fully well that it's a machine translation. Apparently not everybody agrees though... For example, I came across an article on Microsoft MSDN in German, which at first glance got me thinking, wow, these guys translate MSDN content? Very kind. Then I read more carefully... and somehow it didn't sound right (we don't say "seit wir uns Ihnen in der ...-Kolumne MITGETEILT haben" -- do we??). Hey, wouldn't some statement like "This is machine translated content. We hope the content is of interest and use to you in this form. (And maybe also:) Click here to order a high-quality human translation of this page" - using some Service Façade to a translation services company.
posted by Michael Vorburger.ch at 13:38
2 Comments:
Interesting, do you think this a spamming machine which can decode CAPTCHAs, or actually a "real" poor soul, somewhere?
PS: Thinking about this in the context of my actual post above... machines and humans!
Yes, MSDN machine translation services does work, sometimes up to 90 percent compared to a human translation of the same text. I've used http://www.gotdotnet.ru/DotNet/MSDNTranslate/default.aspx for translating MSDN and Microsoft SQL server doctumentation into Russian.
It's really a good application for MT, for many reasons: the texts are large and of the same tenor, it's quite feasible to customize the system (I mean the dictionary and some grammatical rules) for translating these texts. On the other hand, the MSDN Labrary is constanly being brought up to date, and it's not a rational to translate it in tranditional way, whereas a translation system, especially combined with a Translation Memory system, can get around this problem fast and accurately.
Post a Comment
<< Home