I tried a French language page, and the translation to English seemed excellent. It was fun to watch the text change on the page as it was translated.
Firefox translation is wrong: "Ursula von der Leyen elogia il compromesso sulle emissioni dei veicoli annacquati"
It is the compronise that is "watered-down" not the vehicles.
Google translation is better: "Ursula von der Leyen elogia il compromesso annacquato sulle emissioni dei veicoli"
Both Deepl and Papageo don't translate "watered-down" and it could make sense, since "watered-down" can be considered redundant.
I don't think that makes sense and isn't correct, a translation should be a translation without subjective copy editing. It may be redundant or not, but it's not the job of the translation tool/service/person to decide.
Just think about all the idioms that we use in English that would be totally baffling if translated literally.
These decisions naturally crop up all over the place during translation. It feels pointless to make blanket statements like you did about the removal of “watered down”. Without knowing that specific language intimately, I can’t say that it is wrong, and this is a very simple case.
Good quality translation will always be somewhat subjective.
It's a genuinely hard problem. Even the idea of "correct translation" is slippery.
We got in the car and drove across the island.
Nos subimos al coche y cruzamos la isla.
"We got into the car and crossed the island."
> Subimos al coche y condujimos a través de la isla
> We got in the car and drove across the island.
Making it explicit that you drove across the island :)
If I had to translate "We drove across the island", and make the car explicit without the first clause giving context, I'd say:
> Cruzamos la isla en el coche
> We crossed the island in the car
> "Ursula von der Leyen elogia il compromesso sulle emissioni dei veicoli attenuato"
> Note: "Watered-down" is translated as "attenuato" in this context, which means "weakened" or "diluted." The term can be subjective, so ensure the tone and implication match the intent in the original context.
ChatGPT 3.5:
> "Ursula von der Leyen elogia il compromesso sulle emissioni dei veicoli ridotto"
Additionally, the auto-discovery seems to appear only once per domain? I tested it on Wikipedia, and after the first article, it never shows up again. But I guess this is ok, after all that dialog is Fu*ing huge. Couldn't they choose something more subtle? It's their own browser, they should have all the ability to make proper integrations for something like this.
This is overall a really great feature, which could have great use for people, but as it's now, it's more likely to be removed again in a while, because "nobody is using it", like so often in the past.
You need to go to the settings, search for Translations, then click install all packs.
That it often doesn't show up at all is indeed a bit confusing. I can translate german to english or french to german, but english to anything doesn't seem to show up on english websites even thought it's an option when I'm on a non-english website.
Well, it's a beta, so I assume this will change.
And then? This would make it even worse, people will easily forget some random icon sitting somewhere.
https://addons.mozilla.org/en-US/firefox/addon/traduzir-pagi...
This works exactly like the respective Chrome feature.
>this is coming so late
Because it isn't the same thing. This is done locally. (Not the linked extension but the Firefox built-in.)
GP wants a translation service built in to the browser. Even if they appreciate it being local-only, coming back with “no, here’s some largely irrelevant technical difference that makes this better!” misses the point.
This seems to be about the first time I see Mozilla knowing its own users better than a random HN user know Mozilla's users.
Mozilla/Firefox is meant for users who care about privacy, it's literally the first taglines you see when visiting the landing pages for Mozilla or Firefox.
So, with this newfound knowledge about who is using Firefox and who is using other browsers like Google Chrome, maybe now it makes sense why Firefox wanted local translations instead of sending them off somewhere else?
Google has no qualms about sucking up all your data.
Google also has no qualms about you using Firefox. They pay Mozilla $450 mil a year to insure it stays this way.
- massive anti-trust lawsuit
- paying $450 million / year to Mozilla
- you using Chrome
- you using FireFox
It is very well possible that the preferred boxes to tick include you using Chrome and paying $450 million / year to Mozilla to recoup some of that money. Chrome is one of the more powerful planks in Google's control over the web, without Chrome they'd be fighting a continuous rearguard action, now they get to be the dictators. Think of the $450 million as a license to operate. Any users that end up using the product created using that $450 million are additional loss.
But if you need evidence that Google has qualms about you using other browsers, look at how poorly supported other browsers are for google services.
Different battleground, but could very well end up having a major impact on this.
Goggle payments have done nothing but go up as FireFox market share has done nothing but go down --- to the point of insignificance, now around 3% world wide --- less than Opera.
Paying a huge sum of money for a miniscule market share doesn't provide any reasonable hedge --- as evidenced by the fact that Google is in the middle of an anti-trust lawsuit and FireFox has played virtually no role.
Goggle payments have done nothing but go up as
FireFox market share has done nothing but go down
FireFox has played virtually no role [in Google's
antitrust case]
The DOJ has antitrust beef(s) with Google, but monopolizing the browser market is not one of those beefs. We can't read the minds of the DOJ, but it strongly suggests Google's support of Firefox is working within that context.
It does not, obviously, give them global blanket immunity against all conceivable antitrust shenanigans -- were you mistakenly believing this is what I was suggesting?
These obviously float up and down some over time but Firefox has been consistently under 3% globally for a while now.
See for yourself here: https://www.whatsmybrowser.org/
I'm excited to see a more open translation system that seems to perform pretty well and runs locally.
It definitely needs to add support for more languages, and based on other models I've seen recently... I have to wonder if building dedicated models for each pair of languages is still the best choice. I believe SeamlessM4T just uses a single model (available in different sizes), and I have definitely seen that Whisper only uses a single model for multiple language pairs as well (although it was only specifically trained to translate into English). Similarly, virtually all of the LLMs are multilingual. It seems like a single model is able to learn shared insights that apply across languages, reducing the amount of training data needed for each additional language (or increasing the accuracy with the same amount of data), but I admit that I could be wrong. This has just been my perception of how things are developing.
[0]: Some training info here, it looks like: https://github.com/browsermt/students/tree/master/train-stud...
My only complaint about it really is that it has no Asian languages.