There's a large number of prebuilt Anki decks available here as well if this is useful for anyone exploring the space - https://ankiweb.net/shared/decks?search
It sucks though, it's also the one thing that makes me constantly not be consistent using Anki, I get tired of creating cards and stop for a while.
People are just repeating this advice about making your own decks, and it's based in nothing but having had it repeated to them. Spaced repetition is boiling in pseudoscience and ancient studies that don't say much other than that there's a forgetting curve.
Most people are just parroting stuff they read on the Supermemo wiki (or somebody read off the Supermemo wiki and repeated to them like they came up with it), and all of that is just thoughts off the top of one guy's head. His innovation is that he wrote a program to do Leitner boxes before he had ever heard of a Leitner box, but people treat every word like gospel.
The only five things I can say for language learning is to go really hard on systems in a new language that are completely unknown to you (like Romance conjugations for an English speaker); only drill sentences, not individual words; always say your Anki answers out loud, and read out loud as much as you can; comic books have pictures, too; and once you get comfortable in an L2->L2 dictionary, you're a more comfortable reader than a lot of natives.
-----
* Random Anki decks for a few European languages: https://sookocheff.com/post/language/cloze-deletions/
(Edit: the lovely thing about 10K algorithmically generated clozes is that they're utterly disposable, unlike cards that you make yourself. If one is a leech, forget about it. You'll see another one just like it when you get to the point that it won't be a leech for you.)
* Instructions on how to generate your own in other languages, for developers: https://sookocheff.com/post/language/bulk-generating-cloze-d...
(You could probably point out the above URL to an LLM and it would generate the code for you.)
* Anki to learn Romance conjugations first: https://www.asiteaboutnothing.net/w_ultimate_spanish_conjuga...
I feel even better than natives sometimes because they learn conjugations in order at school, and when asked to recall them out of order (or hop from form to form) get confused. Once you have conjugations, you can read anything with a dictionary (and the online dle is the best dictionary I've ever used.)
I'm about to start again with KOFI French, but I had to do a lot of work to get my mouth and ears adjusted to hearing French as anything other then murmurs, and to be able to read (luckily for me, French is the opposite of perl and read-only instead of write-only.) There's a lot of stuff in French I want to read; reading all of the stuff translated into Spanish from French (but never into English) has got my beak wet.
Also, Spanish-language comic books will make you forget about English-language comic books. And they are very online, examples: http://columberos.blogspot.com/ and https://comicsmexicanosdejediskater.blogspot.com/
Also lots of other good material for vos: https://ahira.com.ar/ including https://ahira.com.ar/revistas/skorpio/ and https://ahira.com.ar/revistas/hora-cero-suplemento-semanal/ which is a landmark of literature that is too good for us (El Eternauta spans the entire length of the series.)
For hearing French, I went through the old FSI French phonology course at the beginning, as well as all the grouped (A1/A2/etc) comprehensible input from this channel: https://www.youtube.com/@FrenchComprehensibleInput/playlists. Oh, and I did a bunch of the French listening comprehension on Yabla for a few months: https://french.yabla.com/.
I did the same. Got me interested in the old programmatic methods and how they could be adapted to LLMs.
Also, pssst... https://seulementbd.blogspot.com/
> using preselected word lists makes sense
I have never used Anki for word to word translation, because there is no Spanish word that means any English word. You've restricted yourself to a particular paradigm from the beginning. I mean, do whatever you need to do to get a foothold, but you want to get away from L1 as much as possible as soon as you can.
I know you consider it pseudoscience to force creating your own cards, but I do find premade decks result in a lot more leeches. How do you avoid them when relying on these thousands of auto-generated cards?
Also, one of the many parts of the spaced repetition lore that I do agree with is that if you keep getting the same card wrong in the same way over and over again, you're building up a weird habit that is going to be tough to break. Better to trash it. [edit: you can't do this with the conjugation cards, though. If you keep failing a particular conjugation card, you need to stop, write it down, and spend time with it individually. All of those are important, except maybe the unique conjugations of europeizar.]
> if you're still in the earlier stages it's much more fun to learn a variety of words and focus on the most common ones.
I believe of course in the have fun rule above all others, because this is an ultramarathon, not a sprint. You're going to have to get enjoyment from the process if you're going to stick to it at all. I get a "dopamine hit" every time I get a card right.
If you wanted to sprint, the best way is probably doing the full old-school* Glossika method where you go over the day's sentences, you listen and repeat, you listen and transcribe, then you repeat on your own and record. The next day you start by listening to your recordings and figuring out how to improve them, rerecord, go over your new sentences, rinse and repeat for 3-4 hours a day. You can certainly hammer a language into your head that way, but you probably need a tiger mom threatening to withhold food or something to keep you doing that for 6 months.
About creating your own cards - I've done thousands (not language related), and I've learned and remembered things with them, but there's no science behind writing a good card. Everybody is on their own and flailing, and asking themselves "what would Woźniak do?" rather than coming up with formal rules and testing them. I've got ideas, and there are a few datasets (of people doing spaced repetition sessions over time) available, but I think that treating cards as a black box, discovering the relationships between them over time (through finding which cards are passed and failed together), and creating some sort of internal market of cards between adversarial LLMs could generate good decks, and generate mappings that would allow failing or passing one card to affect many other cards rather than the single card alone.
The best thing about FSRS to me is the decision to stop asking the user about process, and just ask them about goals. Some of the Anki community seem on the verge of giving up on the self-grading, too, and just moving to pass/fail, which I couldn't support more. I never used anything but pass/fail, with the settings from refold (https://refold.la/simplified/) which have been obsoleted by FSRS, which works better with pass/fail only. As I said before, a lot of how these things were designed from the beginning was simply cargo-culting Woźniak.
----
[*] I haven't tried the new school online version, but I'm sure they have varieties in the same mode.
I think what's key is that I'm taking the words and conjugation rules I'm learning and using them relatively quickly, often that day or that week. I.E., I'll come across words in Anki, then hear them in a baseball broadcast or see them in a news article. Or I'll recognize what tense something is because of the rules.
So it's supplemental, and maybe that's why it's sticking better. I don't think I'd want to create decks constantly, I created one 140 card deck and that was enough.
Finally, I do frequently use memory tricks to create associations so maybe my experience with memory castles, mnemonics, and other techniques (which I use on cards I forget frequently to create links I'm unable to create quickly on my own (or to differentiate similar words (or words that are the same but in different tenses))).
Yea, it's _work_!