Oooh, DeepL has added 13 European languages! Including Danish, Swedish, Hungarian, …

DeepL quality consistently outperforms Google Translate. It's also great because DeepL doesn't use English as common ground between non-English languages.


Google Translate can include hilarious errors between non-English languages that clearly show that everything has been translated to English first.

· · Web · 3 · 2 · 14

@rixx Wait, is this how Google Translate works?! I never knew!

@rixx Honestly, when AltaVista’s BabelFish was still a thing, I thought it worked better than Google Translate (though Translate obviously improved a lot since).

@Sylvhem @rixx honestly, I could probably count the number of knowledgeable linguists actively working on Google Translate with zero hands

they probably do have direct translation between a few, mostly European languages, but I'd be surprised if that were the case between less common paths like Japanese to Farsi

@rixx The other day, I ran something through Google Translate that translated "Transfeindlichkeit", which I would understand to mean "transphobia", to "transhumanism" 😕

@rixx this isn't by design, but because the majority of Google Translate's training data involves English (because of the Internet).

I think tech-wise, Google Translate and DeepL are fairly similar, but DeepL uses a lot more human-annotated training data.

@russss Yeah, you're right!
I found out that DeepL also fails on corner cases – But, and this is kind of important, it fails by substituting an English term in the A → B translation, whereas Google Translate goes via English into B and loses information. I've seen "to watch" become "a watch" when translating from Hungarian to German (in both languages these words don't share anything).

@russss That's probably just better safeguards and a development team that is more focused on quality, rather than Google-ism, I'd guess.

Sign in to participate in the conversation – a Fediverse instance for & by the Chaos community