Start a new topic

DeepL connection for CafeTran


see also: http://www.proz.com/forum/business_issues/318095-deepl.html (which is where I first heard about it just now)

Probably the system trained by DeepL does not contain a completely new technique. However, the combination of Convolutional Network, Attention and Beam Search corresponds to the latest research results.
Yes, Michael, but I was the first to post the news here. You cannot take that away from me :P

Yes, but have you mentioned it to Jost on Twitter yet? Ha!

Okay, you win. But only because Twitter is obsolete ;) Hey dude have you yet converted the units conversion macro to AHK?

1 person likes this
I hope that the announced API will be integrated in CafeTran. Until then, we can use a web resource definition for DeepL. Would be fun to visit them in Cologne. I'll send them a message.

I don't like it, it's too good.


I tried three sentences, EN>NE, current job (cannot be in their resources), marketing BS. Two were excellent, in one a preposition was left out that would have been left out by many native speakers of Dutch as well. I'm going to try out more, it's frightening.


H.

Tried a few more sentences, DE>NE. A bit disappointing.


H.

Some more comfortable segment/phrase searching in DeepL/Google Translate/Microsoft Translator resources via their web browser interfaces is coming in the next update.


3 people like this

It would also be great to have a feature to harvest these great alternatives they often give. (Other than by macro.)


image



Often they are so nice that I take the time to add them manually ;).

That's not possible via the web interface as they are generated dynamically. Anyway, what's the point of harvesting any web resources (especially the ones included a pretty good MT service) if they are freely available?


1 person likes this
>Anyway, what's the point of harvesting any web resources (especially the ones included a pretty good MT service) if they are freely available? // The fact That when they are presented in sentence A, doesn't implicate that they are presented in segment B too. Just like with real humans.
Login to post a comment