Hello all! I’m looking for a good alternative to Google Translate. I would prefer one that isn’t using a modern LLM, but I understand if that’s not possible. I would also like it to not be collecting personal data.
DeepL is probably the best option.
Definitely DeepL! Translations have an excellent quality and you can alter words and easily rephrase sentences within the web interface. The smartphone app is good very good.
Deepl was the first using an LLM though richt? They have always advertised that it was AI.
OP specifically doesn’t want that if possible.
Or is it a different class of “AI”?
Look at this answer OP
Just for awareness, Google Translate is based on LLMs. Translation is actually why LLMs were initially made and you won’t find a better method for translation (besides hiring a translator). The thing you’re looking for is an LLM only for translation.
OP specifically requested non-LLM.
Reverso is a french translation service, even better than DeepL imo. Their context translation is especially useful.
Yeah, with Reverso you can see how the word has been translated in different context, super useful if you actually want to learn.
And their general translator is also good for longer texts. I’ve been using DeepL for that, though, and haven’t made a proper comparison.
Not really european (the main dev is from florida), but free and open source without collecting your data:
I use that, too, and you can run it locally.
Firefox’s built in translator from Project Bergamot is made by a co-operation of several European universities and Mozilla. It’s a local LLM that you download so the translation never leaves your computer. You can translate whole pages with it or select text and right click and “Translate selection”. The only problem is that if you had some raw text you copied from somewhere. You’d need to go to some pastebin site, paste the text, select it and translate selection. I guess it’s not that inconvenient.
https://en.wikipedia.org/wiki/Project_Bergamot https://firefox-source-docs.mozilla.org/toolkit/components/translations/resources/03_bergamot.html https://bergamot.mt/
There used to be a desktop version somewhere but I couldn’t get it to work. You can run the wasm version from their git repo maybe
As other said DeepL is European and a good option. I have no idea how much data they collect.
If you want to guarantee zero data collection, running a model like Ministral locally on your device using a trustworthy front end is probably the only option.
I mean LLMs kind of dropped out of the work to build translation engines. There are models which are better trained towards doing translation than being an agent of bullshit, but the architecture is usually the same.
This might not be a perfect fit or off topic but I had to chime in and recomend Transcribro Transcribro is a private and on-device speech recognition keyboard and service for Android. It uses whisper.cpp to run the OpenAI Whisper family of models and Silero VAD for voice activity detection. It features a voice input keyboard, enabling you to type with speech. It can also be used by other apps either explicitly or when set as the user-selected speech to text app which some apps may use for speech to text. Transcribro is available on the Accrescent app store and GitHub releases. Accrescent is the recommended way to get Transcribro as it is more secure than GitHub releases.
Lingvanex, DeepL
If you want something open-source/self-hostable there is LibreTranslate
Why not LLM? When ChatGPT came out I was living a country where I didn’t speak much of the local language. GPT was a game changer, it didn’t just translate the words, it understood the context and it was much more useful.
AI does not “undestand”
Especially not the context. LLMs are Mathematical Methods.
But I agree: They are actually good for translating especially when focused on translation - like deepl
