SRS Review Intervals: Theoretical Interval Length vs. Real Interval Length / When should you take cards out of review because you already “know” them?

This is something that’s been bugging me for a very long time:

Theoretically an SRS algorithm like the one Anki uses (SM-2) presents cards to you around the time you might struggle to remember the card. Over time those intervals become longer and longer. In theory, extending the review intervals extends how long you will remember the fact tested. For vocabulary cards, though, odds are you will also encounter words *outside* of Anki. Thus your actual review intervals are shorter than Anki’s interval.

For most vocabulary it’s useful to review when you start learning a given word; except for the very most frequently used words, the word won’t naturally occur frequently enough to get reviewed before there’s a significant chance of forgetting. So early on, there’s a clear boost to new vocabulary retention by using some kind-of review method, SRS or otherwise.

However, in theory, at some point the natural frequency of a words occurrence will be greater than the cards interval in the SRS. At that point, there’s theoretically no need for SRS review and, for the sake of streamlining review, you might as well take those cards out of active review. I actually never put the first 500 or so words I learned into SRS and while I did eventually forget many of them and later make cards for them, many of them I didn’t. (E.g. I never had need to make a card for 日本語) They occur so often there was no need for review.

However, in practice, I haven’t regularly taken cards out of review so I’ve noticed that sometimes I *do* fail cards with long intervals, sometimes even year+ long intervals. I think often the failure comes from slight mistakes, e.g. instead of 順調（じゅんちょう;favorable）remembering じゅんちょ as the reading, maybe I made this mistake once encountering a word outside of Anki and not bothering to check I read it correctly. If I made a small mistake like that when the Anki cards interval was already pretty long, I probably practiced and reinforced that mistake for some time before the card finally came back around; I was convinced I knew the word, but actually I was mistaken.

Natural occurrence intervals for vocabulary won’t nicely expand like Anki intervals: there are likely to be bursts where you see a word often for a period of time, then rarely see it for a while afterwards. Those breaks between frequently seeing a word may be punctuated by some Anki reviews as well. However, if you have a card with an interval of 356 days, but you happened to see the word in the news a month ago, the real effective interval is a month, not a year. Your next Anki interval will be a year and a half or so, but if you don’t see that card outside of Anki, your review after a one month gap isn’t really preparing you for the next year+ long interval between reviews.

But what should I do in a situation like that? Let’s say I actually *failed* to recall that word when I read it in the news, but when it comes up in Anki I can recall it because I looked it up after I saw it in the news a month ago. Should I fail the card to decrease the interval? If it’s been a month since I looked the word up, though, I may not even remember that I’d looked the word up, though. Either way, I remember it now, but I may not be likely to remember it in over a year. Even in the case I remembered the word at the last encounter, the Anki review interval has becoming greatly disconnected from the actual practical review interval at which I encounter the word.

When I started writing this post, I thought I was going to be talking about how actual review intervals become erratic because of the interplay between Anki reviews and seeing a word outside of Anki. But maybe the actual conclusion I should draw is that real world exposure to vocabulary is sporadic and after the very first round of reviews (say interval is less than or equal to a month or so) the disconnect between Anki review intervals and the actual interval between encountering the word, whether inside or outside of Anki, becomes so great that I should just suspend those cards and add a new card if I see the word again and fail to recall it. On the other hand, having long interval cards around, even for words I regularly see outside of Anki does seem to be useful in providing a more rigorous test of knowledge than I might apply when I’m just reading or listening to Japanese. So maybe there’s value to keeping them longer, though clearly at some point as the review interval becomes multiple years long, it’s definately going to become redundant.

I am unable to draw good conclusions. Even though SRS flashcard intervals eventually become disconnected from actual exposure intervals, I certainly can’t think of any better, practical way to review vocabulary. Even though theoretically it seems like I should be able to suspend long-interval cards without effect on my retention rate, I still catch some mistakes by reviewing long-interval cards. I wonder if anybody else has ever thought about these kinds of things and has reached a more tangible conclusion that I have been able to.

Edit:

Let me add one practical piece of advice: if you read this and decide you want to suspend some Anki cards the search filter to use to find cards with an interval larger than 356 days is:

prop:ivl>365

​