I’m working on foreign vocabulary and in Mandarin, my retention rate rapidly increased from 84 to 89% in the past couple of months because I improved my method of learning. Anki keeps saying that my minimum retention rate should be around 82%.
as a comparison, in other languages, my retention rate is steadier at about 93% with a minimum retention rate at around 91%.
any thought on that ? before I ratchet up my desired retention in Mandarin, should I wait until Anki tells me to do so or is it a bit slow and should I go ahead already ?
You should set your Desired Retention where you want it.
I think you’re mistaking what minimum recommended retention is. First, it’s a minimum that you’re under no obligation to stick with. Setting your DR lower than that will mean you’re working harder to learn less, but you are always free to work harder to learn more. Deck Options - Anki Manual
You might want to re-read some of the answers you’ve gotten when you posted about this before.
there is no silver bullet and it has nothing to do with Anki.
more listening with multiple sources. absolutely critical for memorizing the tones.
better understanding of what every single character means to remember the words composed of two or more characters : 難過 / 傷心 are good examples. difficult to pass / that hurts the heart. both mean sad).
it’s a long race, don’t start too fast : if I want to learn 100 new words, I’m starting with 10 and I’m not adding another 10 new words before the first 10 are well implanted into my brain. in the end, this method comes with much better and faster results than starting with 100 from the beginning. it’s especially true with words that look / sound / which meaning is a little similar. this also works magnificently with interval ear training (I’m talking music now, not languages).
While I am sure that you are well-meaning and using the information that you have at hand, which likely until recently was the best information that we as a society had available to us, but as of right now at this second, I have severe doubts about the validity of this statement.
The CMRR feature was only available in versions that were using FSRS-5 which underestimated how much people would remember at lower DRs. As such, it was overestimating the minimum, where going below would result in more work for less gains. FSRS-6 has made improvements in this area.
On all of my decks, over the past several months, I’ve upgraded to FSRS-6 and set DR to 70%. The previously calculated CMRR was around 80%. The amount of R I’m gaining per time in the app has gone up dramatically. Accordingly my average R has dropped from around 88% to 84%, but my number of reviews per day has dropped to about 1/3 of its previous value. This means the efficiency has increased by a factor of about 3x, which is… definitely extreme. I’ve also compensated by increasing my number of new cards per day from 40 up to 126. I’ve been doing this for a bit over a month now and even with 126 new cards per day, it’s still only about an hour/day of app time.
FWIW, mine is with learning Japanese vocabulary, mixed in with massive amounts of exposure outside of the app. As such it is likely similar to OP’s case of learning Chinese vocabulary.
I’m not suggesting that CMRR is accurate or the gold standard for anything. I’m only saying that you shouldn’t try to use it for more than what it is intended to be. The user is talking about increasing their DR, which is totally unrelated to CMRR.
And yes, it’s been removed as of 25.07, because it was giving unhelpful results in too many cases. But this user isn’t using 25.07/FSRS-6, so that didn’t seem relevant.
It’s almost impossible to compare anecdotal results anymore, since you and OP don’t have the same collections or parameters.
sure but now, I don’t know what to do with my desired retention rates (there are four of them because I’m working on four different languages). I still want to find the best possible tradeoff between memorization and time spent. how do I do that now ?
would it be better to go back to a 90% desired retention for all my decks ?
The latest betas have a “Compute retention vs. time spent in app as a function of DR”, but it’s not currently displaying the accurate information, but in the near future such a feature will be available.
The optimal value, to maximize retained information per unit time in app, is very likely is somewhere around 70%, but it will depend on a gajillion factors and having DR so low causes other issues.
