Thanks for pointing this out. You're absolutely right that 'Zero Privacy Trace' was a bit too absolute.
I’ve just updated the website to be more precise: we use enterprise-grade APIs where they said data is not used for training by default, and we don't log any original content on our own servers.
We also only send the specific text snippets needed for processing to minimize exposure. I really appreciate the feedback, it helps make the project better!
This looks really cool but I can't justify using it because of privacy concerns. Running this with a local ai and a strongly worded guarantee of no tracking/reselling of my metadata is something I would pay for.
You could always run it in a sandboxed browser (e.g. a Firefox profile) and limit the use to just language learning (e.g. reading neutral articles online). The data you would leak wouldn't be much different than what you risk any time you browse any website.
At the moment, Lingoku browser extension does send text to LLms for processing, we don't sell user data, but i agree that simply saying 'trust us' isn't sufficient.
Supporting local LLM is something we are actively considering, though there are still real constraints on the user experience.
Have you done any work on trying to make the opposite? Injecting English words into Japanese text to make it easier to read?
I find that students of Japanese often have enough grammar to read widely after finishing a couple of beginner textbooks, but they are completely held back by vocabulary.
I have a deep understanding of this point, a lack of vocabulary makes reading Japanese materials very difficult.
For this scenario, we will translate the Japanese text completely into English first, then inject japanese words in to the english text, the translated text with the injected Japanese words is displayed next to the original material.
This is the main feature I've been using myself, you can try it out and see if it's the feature you want.
I can second this, after finishing my intro Japanese classes I was able to parse the grammar of most sentences. Memorizing vocab was the hard part, so I used OCR on manga pages and then Yomitan to hover over and see word definitions (in English).
Nice job! There have been quite a few of these language substitution extensions over the years. (Language Immersion, Polyglot, MindTheWord, etc.)
I have a personal extension that I wrote (close to 12 years ago at this point) which does the same thing - translates random words on websites as you browse according to your linguistic level. It vastly predates LLMs though so it's all built on sentence segmentation, POS analysis for stemming, and other NLP techniques.
I've written a bunch of integrations for it so it works with websites, documents, even Kindle books.
The site is visually a bit of a mess. The nav bar anchors but not to the top of the viewport (scroll and watch). Some of the cards are also different sizes. Some of the text isn't properly spaced (look for the colons).
LLM's makes this kind of words substitution more easier and accurate. we have also tried some methods like NLP, but effect is mediocre, but if we want use it in specific scenarios, NLP maybe more efficient.
The website's visual design definitely needs improvement. we are currently work on it.
The concept of "injecting" Japanese into text written in a different language is interesting. But I feel the presentation of word definitions are not great. Something similar to https://yomitan.wiki/ or https://jisho.org/search/kotoba would be preferred. E.g. 言葉ーことばーLanguage, word or phraseーKanji definitionsーSample sentence
The current version of word definition is somewhat rudimentary. right now we supported four languages(English, Chinese, Japanese, Korean), but we dont take much time to optimize for each language yet.
Our thought is that we will have a shared template across all the learning language, but also have abilities to customize the word definition template for language-specific needs.
It's a very good investment to spent a week or a month (typically depending on age) to really learn hiragana. Learn hiragana and katakana in parallel. Use an app like 'Kana mind', for example, to enforce memorization.
Using romaji will, due to how the brain works, forever keep you out of learning kana properly, even when you see both at the same time. Learning kana is a small effort for a lot of gain.
Same, I love the idea - but I cannot read the hiragana/katakana (heck even: はい hai yes <- would work well).
Edit: Decided to make my own firefox addon to do it, no worry about daily limits and I can simply update a json file with more words when I feel like I'm remembering things.
I've just whipped one together and am currently testing it out, I won't have time tonight but check back tomorrow and I should have it up by then. (I'll post a reply to my own comment here)
Interesting. I'd love to get something like this into things I use frequently. Primary issue for me is, though I often browse HN for example, I only do so using an app. Generally I rarely use the browser for anything beyond research and linked articles.
After a certain age many people, and I'm definitely one of them, stop using phones for anything serious. Too small screen, unreable fonts, or alternatively zoomed-out fonts make pages unreadable anyway.
So I'm basically back to using phones for, well, calls.. and a tablet or, if more than just casual, a real PC with a huge monitor if I want to actually spend some time on the net.
In short - I'm definitely now in the "browser for everything" group, I don't use apps if I can avoid it, and definitely not on phones.
I'm going in the other direction. I find myself increasingly wanting to do more things with my phone, since it's always with me everywhere, and am currently somewhat actively looking into what a truly mobile-friendly UX would look like, like for example development.
You're absolutely right, we typically spend more time our our phones for most people. we're also preparing to port this feature to mobile, possibly based on the mobile version of Firefox browser.
we've considered using local llm, but the problem is that for a better user experience, we will add user's new vocabulary list, then inject words based on the list, it's hard to do this on local.
We will seriourly consider the point of support local llm, this will also allow more users to utilize our basic functions.
I hear this often but haven’t seen too many translation-free alternatives for the non-immersion tasks (eg: memorizing vocabulary for a standardized exam, daily study in a non-immersive environment). Have you seen any good monolingual techniques beyond “just get tons of exposure”?
Thanks for feedback, in some cases, we use a NLP lib to detect to language of the word since we support multiple languages, this may be due to language detection failed on some words.
Why is it using romaji to show the pronunciation, instead of furigana? Any serious Japanese learner will learn hiragana and katakana very early on, and these are better for reading pronunciation than romaji.
Thanks for the feedback, actually we use furigana to show the pronunciation,
we use LLM to produce the word explanation, this may be due to LLM instability, could you help tell me the word of this case on your side.
I tried the pronunciation feature, which works less than awesome on my system. I am happy to share that "語彙" is pronounced "chinese letter chinese letter", while for "効果的" it is "chinese letter chinese letter chinese letter".
Is that just my Debian/Firefox system? Or is "AI slop" the reason here?
I tried the above words in Chrome, and got the same problem. sorry about that, our tool is far from perfect. this is a bug in the extension, we will fix it asap.
I'm unsure however what you mean by
> No Ads · No Social · Zero Privacy Trace
and
> Web data is used only for real-time AI parsing, transmitted via SSL encryption, and the server never stores any original content.
Because the APIs for commercial LLMs, if you're not hosting the models yourself, definitely grab and store everything.
I’ve just updated the website to be more precise: we use enterprise-grade APIs where they said data is not used for training by default, and we don't log any original content on our own servers.
We also only send the specific text snippets needed for processing to minimize exposure. I really appreciate the feedback, it helps make the project better!
https://www.youtube.com/watch?v=dcolM6W5Odc
TL;DW you have to use the words
At the moment, Lingoku browser extension does send text to LLms for processing, we don't sell user data, but i agree that simply saying 'trust us' isn't sufficient.
Supporting local LLM is something we are actively considering, though there are still real constraints on the user experience.
I find that students of Japanese often have enough grammar to read widely after finishing a couple of beginner textbooks, but they are completely held back by vocabulary.
For this scenario, we will translate the Japanese text completely into English first, then inject japanese words in to the english text, the translated text with the injected Japanese words is displayed next to the original material.
This is the main feature I've been using myself, you can try it out and see if it's the feature you want.
I have a personal extension that I wrote (close to 12 years ago at this point) which does the same thing - translates random words on websites as you browse according to your linguistic level. It vastly predates LLMs though so it's all built on sentence segmentation, POS analysis for stemming, and other NLP techniques.
I've written a bunch of integrations for it so it works with websites, documents, even Kindle books.
https://mordenstar.com/projects/linguaswap
Now onto some feedback:
The site is visually a bit of a mess. The nav bar anchors but not to the top of the viewport (scroll and watch). Some of the cards are also different sizes. Some of the text isn't properly spaced (look for the colons).
LLM's makes this kind of words substitution more easier and accurate. we have also tried some methods like NLP, but effect is mediocre, but if we want use it in specific scenarios, NLP maybe more efficient.
The website's visual design definitely needs improvement. we are currently work on it.
BTW, i like the blog
The current version of word definition is somewhat rudimentary. right now we supported four languages(English, Chinese, Japanese, Korean), but we dont take much time to optimize for each language yet.
Our thought is that we will have a shared template across all the learning language, but also have abilities to customize the word definition template for language-specific needs.
Using romaji will, due to how the brain works, forever keep you out of learning kana properly, even when you see both at the same time. Learning kana is a small effort for a lot of gain.
But this feature seems to be a fairly common requirement; we can consider add a switcher to implementing it.
Edit: Decided to make my own firefox addon to do it, no worry about daily limits and I can simply update a json file with more words when I feel like I'm remembering things.
I've just whipped one together and am currently testing it out, I won't have time tonight but check back tomorrow and I should have it up by then. (I'll post a reply to my own comment here)
In short - I'm definitely now in the "browser for everything" group, I don't use apps if I can avoid it, and definitely not on phones.
is there a possibility of using local llm endpoints for this?
We will seriourly consider the point of support local llm, this will also allow more users to utilize our basic functions.
Translating everything into your native language is pretty universally considered a very bad habit in language pedagogy.
I’ve been experimenting with monolingual vocab this month but it is too soon to say if I like it or not: https://rickcarlino.com/notes/korean-language/monolingual-vo...
https://github.com/rowan-walshe/vocabify
https://github.com/GoodPraxis/oumu
https://migaku.com/
https://jointoucan.com/
BTW, is there a way to remove the romaji? Having both hiragana and romaji together is not good.
Edit: And for some words there's _only_ romaji.. :-(
As a struggling lifelong English learner I had an exactly same idea, but for English.
Yes, we will prioritize support for Safari, Opera, and Arc. Support for other browsers will be added as needed.
Is that just my Debian/Firefox system? Or is "AI slop" the reason here?
I tried the above words in Chrome, and got the same problem. sorry about that, our tool is far from perfect. this is a bug in the extension, we will fix it asap.