> if you reverse that, you will often have to remember many different foreign words for one card.
This is one of many, many reasons why spaced-repetition systems should support dependencies - as in, "delay card X until after cards A, B, C have been solidly learned." Then you'd just gradually learn a sequence of reverse cards for that one word: most common first, then second-most common (prompted as: "<word> (2)"), third most-common and so forth. With each "depending" on the previous cases in the sequence.
There is no way I'm working with cards like discussion 7 which would require me to recall this deck's #7 way of saying "discussion" in Japanese.
There is also no way I'm working with a single card that requires recall of seven different translations.
I.e., there is no way I'm working with English -> Japanese cards, period. It's a waste of time. I don't want to speak or write Japanese by thinking of English words and then translating, so why train that way.
English is just a crutch to help connect the Japanese words to meaning. Once they are connected to meaning, the original English loses importance.
When I'm answering J -> E cards, I don't always use the exact words that appear on the back. I think of the meaning and proceed from there.
New cards I add are often J -> J. If I look up a word or phrase using a Japanese source, I just write the card in Japanese.
Also, like grandparent said, there is the active versus passive vocab distinction. Why would I learn translations of large numbers of English words, which I don't actually use? Knowing how to translate words I don't use is not going to help speak the other language even if I translate from English, because I simply won't think of the words that I don't use.
My English vocab might have, say, 25000 words in it. But actively use maybe, say, 6000 of that. If I had a list of which 6000 that is, then drilling on translating those specific ones to anther language might be helpful. In those situations when I'm stuck thinking in English, whatever I think of using a combination of my personal 6000 words would have a translation from having drilled on that.
> This is one of many, many reasons why spaced-repetition systems should support dependencies - as in, "delay card X until after cards A, B, C have been solidly learned.
I half agree, but I think it works as is. You just manage the dependencies yourself. It's kind of an art, but you get a feel for how to do it. I find in practice it works without having to have the delay, though perhaps a delay would eek out marginally better performance. It's mostly just up to you figuring out the total set of things you need to be able to recall and whether to include any bi-directional facts or any meta-facts to aid the recall.
The Wanikani (paid) spaced repetition software does exactly this for learning Kanji. It's exceptionally well thought out, both in order of dependency and frequency. I'm currently using it, and I love it. I still use Anki for vocabulary, but wish there was something as polished as Wanikani instead.
In Tankan, instead of viewing flashcards, you take batch tests, typing answers into blanks, tabbing from one field to the next. The machine grades your answers as a batch, and you can take a new test immediately over just the unanswered/incorrect items.
As you gain proficiency, a "pipeline" process takes over: your eyes can read ahead two or three kanji while your fingers basically don't stop.
I've been able to review some 1200 characters in the space of about 40 minutes of furious typing, when my accuracy on those characters hit around 96%.
Instead of spaced repetition, once you're proficient, you can just fire this up once every few months, and just review every kanji you know in one short, sweet little session and be done with it.
> ...instead of viewing flashcards, you take batch tests, typing answers into blanks, tabbing from one field to the next. The machine grades your answers as a batch, and you can take a new test immediately over just the unanswered/incorrect items.
You could integrate this with spaced repetition as well, though it would only be more efficient than the flashcard format when the answer really is incredibly easy to type in. OTOH, it introduces a verification aspect that makes this applicable to adversarial scenarios. The Khan Academy does this already, btw - once you make an account on the site, it has a "Review" feature that operates on SRS principles.
This is one of many, many reasons why spaced-repetition systems should support dependencies - as in, "delay card X until after cards A, B, C have been solidly learned." Then you'd just gradually learn a sequence of reverse cards for that one word: most common first, then second-most common (prompted as: "<word> (2)"), third most-common and so forth. With each "depending" on the previous cases in the sequence.