This is a really brilliant idea. I've been learning a language recently and almost everything is either too hard or way to easy to be useful.

This is great! I wonder if you could make a text-to-speech button as well

Really cool! I wonder if you could use AI to do other useful things in the text of webpages.

This is awesome!

Very cool idea!

awesome! i am also exploring “low effort language learning” space and making an app. would be interested in bouncing some ideas, got a social handle to share?

Yep, my Discord account is `alexc.j`

Toucan has a similar concept, but it works on individual words rather than sentences. Beyond a smaller scope, it's also quite difficult to accurately translate single words without context - Nuenki has it disabled by default.

The tradeoff is that Toucan is free, and translation is quite expensive.

looks cool,it would be useful to have transliterations for non latin alphabet languages,or at least pinyin for Chinese, since learning characters is somewhat orthogonal to the words.

I'll put that on the todo list!

Would you typically have the pinyin above the Chinese characters (like Furigana) or show it separately?

I'm also learning Chinese, and I honestly wouldn't care for Pinyin all that much. (Once you get to an intermediate level of Chinese, you'll want to learn characters anyway, because that's how Chinese is written in the real world.) That said, a hover option to transliterate and translate individual words (instead of the whole sentence at once) would be great.

For a more practical complaint, in the Nuenki browser extension, the settings window has a primary-colored (green) button in the bottom right that says "Log out", but has no "Save" button. I almost clicked "Log out" when I just meant to confirm the settings I had made!

Edit: Also one question, if I may - how large is the context window for the translations? I noticed that the Chinese it generates sometimes feels a little unidiomatic, but I'm not sure whether that's due to too little context, the fault of the translation engine(s) used, or just my feeling as a non-advanced Chinese learner being off.

Edit 2: Some of the translations are definitely just wrong, too. Again, might be the fault of the engines - after all, machine translation, even in the age of LLM, is not perfect.

I can't really translate individual words (translation is done on a sentence-by-sentence basis, so I don't have that data), but I could do transliteration. Perhaps hovering could transliterate, clicking could translate the whole sentence, and clicking again could reset it. It's difficult balancing the amount of things people want to do with the text (listen to audio, translate the word, translate the sentence, and transliterate the word) with how few input methods there are + text shifting.

The context window is only that sentence. It would be technically possible to increase the context window, but it would make it impossible to cache and increase translation costs even without that. Unfortunately the development of Nuenki has been defined by mitigating the cost of translation - before I had my many mitigations my prototype burned through six euros of credit in a day - so it's not practical to expand that context window.

I'll definitely recolour that button. Thanks for mentioning it!

The incorrect translations are the engine's fault. The version you're using uses https://www.deepl.com/en/whydeepl . Looking at my quality data for the new update (which adds Claude and Gemma as translation sources in some circumstances), that'll improve a little soon but ultimately not by much.

Would be really useful to have some way to hear the translated sentences so that you can also learn how it's supposed to be said - would be very useful for learning Japanese.

How cool would it be if it worked together with an adblocker. Instead of blocked ads, it would show language learning. Two-for-one.

Very clever.

Toucan [1] does something similar (but I think only at a word level, not sentences), so might be worth looking at what they do in terms of marketing.

Also, have you considered throwing in a spaced repetition component to the process? Really helpful when building an active voc.

[1] https://jointoucan.com/

Are you adding safari support? It's fairly trivial technically as there is a convertor template for xcode to chrome extension -> safari extension.

I'd like to, but I don't have a Mac. I'll probably do it in the future, but right now I'm focused on other parts of it.

Some feedback: I was about to try now (paying) but currently I can only install it in my desktop (no iOS support), and I am very worried about the idea of giving full access to a close-source unpopular extension with not-very-clear ownership.

I think it would go a long way to create some trust to show who are you in the webpage and maybe open source the extension too (the backend is not as importa nt).

That's a fair point, I should mention who I am on the website. I didn't even think to include that!

Here's my github: https://github.com/Alex-Programs

I'm considering open sourcing the extension. I don't want people to be able to copy it wholesale, but realistically that's unlikely and I'm going to write some blog posts on the technical details anyway.

Please do! i would love to contribute, ive been looking for something like this for some time but didnt want to install toucan since its closed-source and i dont trust it

There was/is an extension for FF to do a simple dictionary replace.

Not good for anything serious, but fun.

Hebrew please :)

DeepL doesn't support Hebrew, but I just did some testing and my dev build can do it via Claude, albeit with higher latency.

I'm not sure what the quality is like though.

If you'd like I can contact you when I've added support for it. What's your email/discord?

Mine is [email protected] / alexc.j on Discord.

Love it! Like Toucan but with LLMs.

Looking forward to being able to save specific words/phrases for future review + audio functionality + pinyin for Chinese please :)