Show HN: Kimchi Reader – Immersive Korean Learning with a Popup Dictionary (kimchi-reader.app)
I was learning and still am learning Korean (now using my own tool to learn!). I initially made the tool for myself because none of the tools out there could correctly figure out what a word was in the text. And that's where the biggest challenge was: recognizing the lemma.
A lemma is the dictionary form of a word. An example in English would be "break, breaks, broke, broken and breaking" all becoming "break". In Korean a word cannot always accurately be returned to the lemma; this might lead to several possibilities. Then the reader, with the context can understand which one ends up being the correct one.
Zero IA was used, pure rules based bruteforcing. I do perhaps intend later to use AI, but as a layer on top. I wanted to make sure I can parse massive amounts with cheap computation first. This open me doors to more crazy ideas for later.
You can see a live example on the landing page. Any feedback is appreciated!
45 comments
[ 2.5 ms ] story [ 98.3 ms ] threadI tried authenticating with Google to try it out on mobile, but I got a 500 when redirected back to the site. I’ll be sure to check back later though to try again :)
https://mirinae.io/
But, I would like to see some more about what it offers. Right now it looks like it is a video caption replacer with a dictionary. There are free tools like that that exist. Anki is a plugin and not built in, why? What about those starting learning? Do you have lessons or small references?
What's your plan for ML? Honestly, the largest benefit I see is for pronunciation feedback. This is a thing a lot of apps get wrong, because it is important to learn how to pronounce things from the beginning. This usually isn't as big of a problem for someone learning English but tones are a bit more important in Korean though not as much as say Chinese (this is part of why it's typically easier to understand someone in English with a very heavy accent than understanding Chinese with a very heavy accent). But a self learner is going to have an incredibly difficult time getting verbal feedback while a native or more traditional learner has this feature built in because there's another person there to give them feedback.
Yes there are free tools that does the same job (I have noted that on the description of the post). The main difference is that Kimchi is capable of recognizing words correctly (ofc not 100%, but better than other tools. If you think there are better tools out there, I would love to see it!).
As for Anki, I don't want to rebuild a whole SRS on my side if that what you were thinking. I think Anki is an awesome software and does its job extremely well. So using it instead of recreating the wheel sound much more interesting.
For the content, I don't have anything right now indeed. I would love to build a customized content recommendation later matching with the knowledge you marked in Kimchi. But I don't plan producing material myself and I don't think I have level to do so anyway.
And finally for ML. I think I should use it where it make the most sense. I definitively want to use it! But I also don't want to use for the sake of using it. I wanted the foundation to be without, then ML can always be slapped on top of it. As you said, there's plenty of scenario where I could use it, be pronunciation, ocr for webtoon, translation with context for when you understand each word and grammar but still don't get the sentence, whisper, tts for making audiobook on the fly, etc.
I'd probably focus less on the content recommendation. But I'm not seeing the big utility as specifically a language learning app. Unless it is purely about language learning content, not just youtube and netflix videos.
For ML, as an ML person I'm glad to hear you say this. I hate when it's hamfisted into bullshit things. The ideas you have seem good though but most are probably handled better by other products which you could integrate. But I am not aware of a good product that helps with pronunciation (at least to any passable degree), though I am aware of ones that do this for singing. I'd see this as a killer feature but it'll cost you some compute to get there.
[1] https://ankiweb.net/shared/info/4066961604
Not op, but…
Start with memorized words and phrases — salutations, meal time commentary, celebration terms, etc.
Then move to short, simple sentences — emphasis on short and simple. You might not understand any replies, but that’s ok.
Over time, these might progress into short and simple conversations as you become better able to understand their responses (and they probably learn how to simplify for you), and your vocabulary grows.
Once you get this far, you will have more specific questions to ask regarding gaps you want to address.
Korean is one of the harder languages for native speakers of English to learn, so don’t expect a quick ramp up time like you might have with Spanish or French.
One of the challenges with Korean is the S-O-V order. This is like Japanese, so Korean is quite a bit easier to learn if you already know Japanese.
It's not just a dictionary but a grammar parser, see also mirinae (https://mirinae.io/), the grammar parsing tool for Korean. Korean and English are so distant that a good grammatical parsing is infinitely more helpful than a dictionary and/or translator alone.
There actually isn't any other option, free or otherwise that combines all this "easy reading and watching" tools with a good parser.
Mirinae is great and better than the tool used here but is standalone and no longer lets you view grammar explanations for free.
GPT-4 also gives good parsings but it's even more cumbersome to use individually at scale (copy/paste, ask etc)
2. another big thing here is convenience. I'm not exaggerating when i say you can cut studying time in half at least having everything in one place.
If you're talking about content to try directly after making an account, you're right. There's none for now. Something definitively can be improved there. Noted.
[0]: https://foosoft.net/projects/yomichan/ [1]: https://chrome.google.com/webstore/detail/10ten-japanese-rea..., https://apps.apple.com/us/app/10ten-japanese-reader/id157354..., https://addons.mozilla.org/en-US/firefox/addon/10ten-ja-read... [2]: https://github.com/kha-white/mokuro [3]: https://chrome.google.com/webstore/detail/subadub/jamiekdimm..., https://addons.mozilla.org/en-US/firefox/addon/subadub/
What I would dream about is the ability to read Manwha with it. That's probably a bit hard as the text is inside images, but maybe with some image recognition it might be possible...?
There is a kind of anti-pattern in tools of this kind where the people building them already know some of the basics and so leave them out. It shows up in tools for Japanese, Chinese, Hebrew... I've studied quite a few languages and so have been the beginner over and over.
A one-page alphabet reference chart would be enough to remind the reader which letter is which without relying on the romanization crutch.
Normally I don't like to make argumentative internet comments but I really passionately think romanization is a detriment to a learning tool.
As someone that knows and can read Hindi, Gujarat, Cyrillic, Greek, Arabic and beginner at Hangul and Hiragana, romanization absolutely helps anchor the sound. The actual sound when speaking is going to change anyways as your converse with more people. But at least, romanization helps recall and focus on the actual word that you are learning.
Using romanization is a lot worse than in a language like Japanese. In Japanese the romanization somewhat maps well to Japanese. In Korean, the various romanization methods are horrendous and don't come close to Korean at all.
Not only that-- many grammar forms require understanding the Hangul vowels to understand how to conjugate them. If you're using romanization they don't exist.
Even worse, Korean sound change rules make it hard to read Hangul without a lot of practice. If you're using romanization you're completely doomed.
For folks who are learning Korean, hangul is maybe two days to get the basics and maybe a week or less to be able to get comfortable with it.
For super-casuals who just want a ballpark representation, you might be right, but I don’t think it’s reasonable to design around these super-casual folks — when they decide to get serious about learning Korean, Hangul will come quickly.
Hangul, as a phonetic representation, has about the same difficulty as learning hiragana, which takes about the same amount of time to learn.
Maybe if you are referring to older Korean texts that had Chinese characters in them, then I could understand.
> … It shows up in tools for Japanese, Chinese, Hebrew
Kanji (not a syllabary or alphabet), hanzi (not a syllabary or alphabet), and no vowels respectively.
These languages need some romanization support much more than Hangul, imho.
I don't think you'll be able to find empirical support for the idea that romanization is unhelpful after the first week of someone's taking up Hangul in a serious way; or for the idea that Hangul is as simple as Hiragana.
Many people say things like this -- "two days to get the basics" and "a week or less to be...comfortable" -- but I doubt there is any empirical support for figures like that. I suspect that a review of the literature will show that people are still referring back to the romanization for the first two years of their studies, across a wide array of language families, if their first language uses the Roman alphabet.
Hebrew has vowel marks. They are little used but they are very helpful, as well.
Hangul is definitely one of the most logically organized, consistent and systematic systems of writing in the world -- and it may actually be number one in all those categories.
[1] https://github.com/FreeLanguageTools/vocabsieve
[1] https://github.com/FreeLanguageTools/vocabsieve
a few general remarks:
the landing page above-the-fold is just ugly, the color gradient is horrendous, and the top screenshot of a video is... blank. at first i tried to click the play button, but nothing happened. then i realized it's more of an abstract illustration of what you app does. makes sense, but a real video or animation would be even straight to the point. please, wow me!
what is mining? at first, i thought you've created a cryptocurrency. i see it's explained later down the page.
after login, the grammar page doesn't respect the light mode theme. more examples for grammar would be helpful. in fact, what i want is, extract relevant grammar from the CC of YouTube videos in Korean, and then link me to the EXACT TIMESTAMP, so i can hear it in actual use. i would pay for this.
I get the frustration with the landing page and I think you are right. I was not exactly perfectly happy with this above the fold section either. I just made what I could do the best for now. 100% this will be changed in the future once I can get a full fledged trailer.
As for the remaining points, they're are all good points, all noted. Thanks again for the feedback.
I am completely self taught.
My wife is Korean and I am not. We lived in Korea for 5+ years (but in the USA now). We have two children that speak English and Korean. My Korean comprehension is quite good. We are close to the Korean side of our family and we're surrounded by Korean all the time (even in the USA).
All of that is to give some context for how I learned Korean and what has worked (and didn't work) for me.
What has worked wonderfully:
1. Learn Hangul (Korean alphabet) immediately. You simply cannot learn or study Korean with romanization of any kind.
2. Deliberate textbook study with an emphasis on grammar (for example, Korean Grammar in Use by Darakwon).
3. Learning vocabulary in context of the grammar from #2. For example, take a noun/verb out of a new grammar structure and replace with new ones. Always study in the context of grammar and structure.
4. Listening comprehension through transcription. Given some Korean audio, listen and transcribe what is said in Korean. This certainly helps train your ear to pick up native Korean. Korean you hear from textbook audio is very easy and doesn't help in real life.
What did not work for me:
1. Using Anki to cram words in my head. Sure-- everyone can grab a 2k or 6k deck of words and memorize them, but you won't know how or when to use anything. You also won't be able to recall much of it.
2. Watching drama or reading webtoons. It's tempting to call this "studying" but it's really low bandwidth. What you get out of it is in no way at all related to the amount of time it takes. You'd be better off with 10 minutes and a textbook than a 50 minute drama that takes you 600 minutes to get through looking up words and grammar.
3. The cult/following of "total immersion/no output" which was born out of the Ajatt method of learning Japanese. It's baloney. Sorry if this includes your tool :)
Some feedback: - The hero image is a bit vague, not sure what the product is, however, scrolling down a bit I see a great image of a K-Drama playing with the app in action, which was really impressive and informative - The tool tip background is transparent so hard to read the text in it as it gets mixed with the background