This is adapted from a presentation I gave at the Berlin Quantified Self Meetup in November 2012. There are two parts:
- What is spaced repetition software?
- Why don’t more people use it, and what can we do about that?
Part 1: What is Spaced Repetition Software?
Spaced repetition software is a simple studying tool that works incredibly well, especially for learning languages. But it’s been around for over twenty years and – despite some encouraging signs in recent years – still hasn’t spread beyond a small group of nerds like myself.
It’s basically just flashcards with rating buttons. After you see each card, you rate your knowledge of it on a numeric scale. The example below is for someone learning German; the German word appears on the top of the screen and you try to think of the English word. When you think you know it (or when you’ve given up) you hit Enter, and the answer appears on the bottom of the screen, along with these numbered rating buttons. And once you’ve rated the card, you see the next one.
Now, this particular scale goes from one to four. If you knew the answer right away, you’d rate it a four. If you had no idea, it would be a one. If you were close – maybe you thought “chair” – you might rate it a two. These are self-ratings, remember, so you can decide what they mean, as long as you’re consistent and they form a continuous scale.
As you may have guessed, there’s an algorithm in the background that uses these ratings to decide when to schedule the card again. So if you rate it one, you might see it again ten cards later. If you rate it four, you might not see it again for a week or two.
This is not rocket science – in fact, it’s the same thing you’d do with paper flashcards, taking out the ones you know best and reviewing the others more frequently. The basic algorithm, which was written in 1987 by a Polish graduate student named Piotr Wozniak, is short and simple and any competent programmer could probably write their own rough implementation of it in a pretty short time. In fact, several of the programmers I know who study this way have done just that.
Similar „spacing“ techniques had been implemented with older analog study methods, like the Pimsleur tapes of the ’70s, but Wozniak realized that using a computer to manage this process — and introducing user-specific feedback, rather than relying on standard spacing intervals — would result in a quantum leap in efficiency. And it does. I used this method two years ago when I was learning German, and I’d say I learned just as much this way as I did in five months of intensive language classes. And that’s comparing three and a half hours a day in class to just twenty minutes a day with the cards. But for this article, I did a more self-contained experiment, to show you how the whole process works from start to finish.
A Self-Experiment: October 2012
So I decided to read a book in German, look up all the words I didn’t know, and learn as many as possible over a one-month period. Here’s the book I chose, a translation of an American crime novel from the 1940s:
Step one was to mark the unfamiliar words and look them up:
Step two was to enter them into the software:
And step three was to review and rate the scheduled cards every day:
The rating scale here was zero to five, and you can’t finish a session without having rated every card a two or better. Here’s how I defined the rating scale:
1: got it wrong
2: got it right, but only because I just saw it
3: got it right, but had to think about it
4: knew it right away
5: actually getting a bit annoying
And finally, here are the results:
The stacked bars show the total cards in my “deck,” broken down by rating. As you can see, I was adding 20-80 cards every day for twenty days, about a chapter a day, and when I finished the book, the last ten days were just spent reviewing cards.
You can see at the end that according to my own target of 4 or better, I had learned well over 500 new German words by the end of the month. And I knew these words outside the context of these drills, recognized many of them in other settings, and even managed to use a few in everyday conversation.
So, how long did it take? The following graph shows, in minutes how much time I spent on entering and rating cards. The average was about twelve minutes a day in total:
Now, I didn’t track the time I spent reading and looking up the words, because where the content comes from is a whole separate thing. You could even just skip this step entirely and buy a vocabulary list, which is what I did when I first started using this software as a beginner:
But of course, the process of encountering the words in context and looking them up was also helping me learn them, and without that, I bet that some words would have required a few extra repetitions. So if you’re starting totally cold, you may want to allow for a bit more time to get results like this.
Even if you do, I don’t know any other method that even comes close to this kind of efficiency. For example, here’s what that Hueber book has to say…
I can imagine spending twice as much time doing that and not making the same kind of progress. But even if it worked, it would be much harder to stick with it without this kind of quantitative feedback showing daily improvement. It may sound silly, but the best part of every card review session for me was typing in the new scores and seeing another bar show up on this chart.
Want to try it yourself? Here are the „big three“ existing spaced rep programs:
- Supermemo (old versions free, new versions $40-60)
- Mnemosyne (free and open source)
- Anki (Desktop version free, iOS client $25)
Supermemo is the oldest, developed by Wozniak himself. It’s by far the most powerful, with a huge feature set, though it’s consequently harder to learn. Mnemosyne and Anki are two more recent open-source projects that are much simpler and easier to get started with.
Mnemosyne is my personal favorite, and the source of the screenshots that you just saw. But I’ve tried all three and they all work well; it’s largely a matter of taste.
In Part 2, I’ll talk about why these apps haven’t become more widespread, and discuss some newer alternatives that take different approaches. Continue to Part 2 –>