Live data from Hacker News

DeepL Translator – AI Assistance for Language

deepl.com

31–40 of 56 posts

Re: DeepL Translator – AI Assistance for Language

#34
> Specific details of our network architecture will not be published at this time. DeepL Translator is based on a single, non-ensemble model.

Kinda sad to hear, but completely understandable. I'm curious whether the difference in performance is due to their model specifics or just better training data.

Does anyone have more information?

Re: DeepL Translator – AI Assistance for Language

#35
post #16

I tried it with a German poem (Erlkönig) and it seems like Shakespeare is in the training data... at least it's not regular English: "And if thou wilt not, I shall need violence."

That said, in some ways Goethe was the Shakespeare of German, so it's not unreasonable for it to use a slightly archaic style.

Re: DeepL Translator – AI Assistance for Language

#38
Surprisingly good on Jabberwocky, though my French is too weak to really judge:

"Twas Brillig, et les fentes fendues tournoyaient et gimblaient dans l'épée. Tous les mimsy étaient des borogoves, et les mome raths dépassaient les ragots.

Méfie-toi du Jabberwock, mon fils! Les mâchoires qui mordent, les griffes qui attrapent. Et'ware l'oiseau Jubjub, et fuyez le bandersnatch frumieux.

It's curious that it didn't understand "'twas" for the French translation, but apparently did for the German one. (My German is almost nonexistent, though.)

Re: DeepL Translator – AI Assistance for Language

#39
post #32

"Fruit flies like an arrow" Google: La fruta vuela como una flecha. DeepL: La fruta vuela como una flecha. "Fruit flies like bananas". Google: La fruta vuela como plátanos. DeepL: Las moscas de la fruta son como los plátanos.

All of them are correct without context. I could imagine a Boris Vian novel where "las moscas de la fruta son como plátanos"

Re: DeepL Translator – AI Assistance for Language

#40

> Specific details of our network architecture will not be published at this time. DeepL Translator is based on a single, non-ensemble model. Kinda sad to hear, but completely understandable. I'm curious whether the difference in performance is due to their model specifics or just better training data. Does anyone have more information?

They have the perfect training data as this is a Linguee venture (https://www.linguee.com/). They have millions of translations of paragraphs from one language to another.

I have no information on the model, unfortunately.

Post reply on HN