Comment by raincole | Hacker News Reader

Comment by raincole | Hacker News Reader

(Off-topic) I'm convinced that the ideal language learning app should look like this:

1. A HUGH repository of raw materials, both in text and in audio. They are all written/recorded by native speakers, not non-native language teachers.

1.5. (Optional) The materials come with supplemental vocabulary lists and grammar guides.

2. You take a test.

3. It recommends materials for you to read/listen to, according your proficiency level shown in the test.

3.5. (Opt-in) it can read your YouTube history and social media to recommend materials that you might like.

4. Every month or every N hours of reading/listening, you take a new test to recalibrate your proficiency level.

That's it. However due to copyright issues, I don't expect to see such an app in near future. What a bummer.

(Not-so-off-topic) Personally I consider all the apps that don't resemble the above workflow "dictionary-like" (useful but as a reference tool, not a learning tool) or "Duolingo-like" (a healthier alternative to doomscrolling, but nothing more). The article sounds Duolingo-like.

I built something similar [1]

It's basically a podcast player where you can browse a database of podcasts filtered by spoken language, and listen with transcriptions and translations.

For each language I made a podcast to learn the most frequent words.

You can also get audible feedback on your pronunciation.

I am in the process of building a YouTube database of channels by spoken language to play youtube videos on the app.

> Every month or every N hours of reading/listening, you take a new test to recalibrate your proficiency level.

I slightly disagree with this part, I think the moment you add some sort of "test" or drills it can become tedious or dreadful to learn in the long term.

[1] https://www.langturbo.com

Yes, the huge repository of raw materials is likely the hardest part. You can try crowdsourced collections ( https://tatoeba.org , https://datacollective.mozillafoundation.org/datasets?q=comm... , https://opus.nlpl.eu/OpenSubtitles/corpus/version/OpenSubtit... ) but you'll quickly run into data quality issues. My personal solution is to do manual data curation on the fly, but I think an app that occasionally throws up garbage and asks its users to pick out the good parts is unlikely to get popular.

Maybe the free version of the app could do the collaborative filtering part. And in the paid version you'd get the high quality content.

That lets you turn the problem of figuring out which part of your content is badly machine-translated into the problem of figuring out which of your users have enough attention to detail to spot badly machine-translated content in a language they themselves are still learning. Though I guess if you show paying users only content that lots of people think is good, at least it reduces the chances that a paying user notices an issue and complains, so it could still work in that sense.

Look at the app ISSEN. The "best" would be to somehow have a hypothetical LLM that actually works. For now it's only at the beginner level. We can use raw input and spaced repetition to guage progress instead of tests.

I've learned 2 languages to fluency, and the only thing that ever works for me is immersion with comprehensible input, and conversation. I've been generally disillusioned with language learning apps that aren't "language exchange / penpal". And I've tried all of them. I don't think language learning is easily "gamified".

Immersion is important, so I'd expect AI in there, or video calls with native speakers.

Reply on news.ycombinator.com