[james_s_tayler]

I smashed out Hiragana in 4 hours flat. Learning the rest of the language took 3 years.

I do love the obvious wisdom that if you put a little bit of active effort as opposed to just passive consumption that you can build a skill and not just consume things. It's like those low hanging fruit are good but you're still in pain for another 180 hours until you've done your first 200 hours. That's usually when things start to get sustainably enjoyable because you know enough to not be constantly frustrated and can actually have fun no matter the remaining difficulty. Then once you hit the 2000 hour mark you're pretty much good.

[yorwba]

> you're still in pain for another 180 hours until you've done your first 200 hours. That's usually when things start to get sustainably enjoyable because you know enough to not be constantly frustrated and can actually have fun no matter the remaining difficulty.

You can get there a bit earlier by carefully adjusting your practice difficulty to be closer to your comfort zone than the "real" skill you want to tackle and then slowly raising the bar as you get better.

For example, I wouldn't be able to understand any random Japanese sentence someone could throw at me right now, but I built a system to pick sentences from https://tatoeba.org that I should be able to understand based on the words they contain, and it's been very encouraging to notice how much I can understand already. The only problem being that I haven't found a good selection criterion yet. Optimizing for highest probability that I need to refresh at least one of the words in a sentence tends to produce very long "sentences", e.g. https://tatoeba.org/sentences/show/4752008

[james_s_tayler]

I used the Kanji Odessy 2001 Anki deck back in the day when you could get your hands on it. It was 2000 sentences with audio ordered such that each sentence followed the n+1 model. Hands down the best resource.

It still takes a few hundred hours to get through the pain period.

[imtringued]

I think there is a Japanese Core 2000 deck which is of a similar nature.

[james_s_tayler]

Yeah. KO was way better tho. Such a shame it's not around anymore.

[btown]

Is the system you wrote for this open-source? Sounds really really interesting.

[yorwba]

I started working on it about 4 months ago and have been dog-fooding it since, but it's still in a pretty rough state.

Part of the problem is the quality of open-source language analysis tools. I use Open JTalk for text-to-speech on examples without available recordings (which is most of them) and sometimes it gets the pronunciation plain wrong. So I use a different program (Kuromoji) to get a written representation of the pronunciation, which it also frequently gets wrong but (hopefully) at different times than Open JTalk.

The other problem is that so far I've been the only user, and many things that should be configurable aren't (I expect the constantly looping audio to be high on the list of things people would want to change).

That said, if you're interested, I could write up a minimal readme file, slap on a license and push it to GitHub.