Jayeless.net

Low-Key Anki

The guy who came up with the Mass Immersion Approach also came up with an alternative system for using Anki, which he called Low-Key Anki. Some of the points he made were:

  • SRS is linked with the concept of the “forgetting curve”; basically, every word you haven’t mastered is going to be forgotten by you at some (indeterminate) time after you last saw it. The idea of SRS is to find the sweet spot where you haven’t yet forgotten the card, but are as close to that “forgetting point” as possible (so you’re not inefficiently practising words more often than you need to). In practice, this means you have to make a judgement call about what percentage of mature cards being recalled is ideal — 85%? 90%? — with the cards you don’t remember being called lapses.
  • By default, Anki resets the interval on lapsed cards to zero. This is inefficient, because most of the time when you’ve forgotten a card it’s not like you feel you’ve never ever seen it before in your life — usually it was on the tip of your tongue, or you were just slightly off in what you thought it was, or you came up with a synonym but you did actually know the word, etc.. Plus, you’ve now just seen the word, so your place on the “forgetting curve” has reset. The MIA guy recommends, rather than setting the interval to zero, setting it to 70% of what it was before — sooner than last time (because last time was too late and you forgot), but not that much sooner. And of course, if you really do feel like you’ve never seen a word in its life, you can always reset the card to an interval of zero manually.
  • The other alteration he recommends is to use a pass/fail system for grading your recall of cards, rather than Anki’s default again/hard/good/easy. He laid out a number of reasons for this. One is that if you feel like you need to grade a card “hard”, then you probably didn’t really remember it, and you should really be grading it “again”. People grade it “hard” instead because they don’t want the harsh penalty of the interval being set to zero, but if you make the above change, that won’t happen anyway. Then as for “good” and “easy”, the problem there is that just because you remembered a card now, doesn’t mean you know how many days into the future that that recall will continue to be strong. You might rate it “easy” and then push the card so many days into the future that you no longer remember it. Plus, having all those options can put you in a state of “decision fatigue”, where you no longer just have to worry about whether you remembered the card or not, but also about whether remembering it was hard, normal or easy. That fatigue can put you off practising your cards at all.
  • Then there was a last bit about regularly adjusting the interval multiplier on your deck so as to reach your aimed-for recall ratio. Like if you’ve decided your ideal ratio is remembering 90% of mature cards, then bump up that modifier when your ratio creeps above 90%, and lower it if the ratio drops below 90%.
  • This strategy also presumes that the cards in your deck are all of roughly equal difficulty. If they’re not, he recommends first examining the difficult cards — why are they difficult? are the sentences poor-quality? are you confusing it with another word (like a synonym), and if so, do you really need both? etc. — and then, if you’re sure the cards are all good and necessary, splitting the deck into multiple so the more difficult cards can be in their own sub-deck with unique settings (i.e. a lower interval multiplier).

You can certainly adjust Anki’s settings in order to create this kind of system for yourself. However, I came across another flashcards app which, it seems to me, already incorporates many of the principles of Low-Key Anki: Mochi Cards. In Mochi, grading cards is pass/fail only, and if you fail, the card doesn’t reset to zero. The old interval will instead be multiplied by the “fail” multiplier — 0.5 by default, but I set mine to 0.7. Because, oh yeah, you can tinker with the multipliers. Even on a per-deck level. The app is also more intuitive than Anki, although there is a USD$5 subscription fee if you want to sync your decks across devices, which is bleh. Nonetheless, the approach itself seems good, regardless of what app you use. I really like that by removing the harsh penalty of resetting the card to zero, it frees you up to hit “fail” on those cards that you very nearly but didn’t quite remember.

See Also / References