400 days in a row of spaced repetition

Posted on 2022 / 02 / 17


1 Introduction

For nearly the past 2 years, I’ve been using Anki to learn several things, primarily Japanese vocabulary. I’ve also used it temporarily for a few memorization-oriented exams, and recently I’ve begun to study Mandarin. While I started to use Anki seriously around May 2020, my current streak only began at the top of 2021. In any sense, I’ve had a pretty long relationship with the program, and I feel like I’ve learned some valuable things other than a lot of Japanese words.

2 Timeline

2.1 2020

I first found out about Anki and the spaced repetition idea around March of 2020. To be honest though, I was still really toying around with the idea of learning Japanese, and I didn’t have any real sense of the amount of commitment it would require of me for the next 2 years. I picked up the 2k/6k Core deck in May of 2020, and began to regularly do repetitions. Further, I’d just entered summer break off my junior year of high school, and I had a lot of time to sink into this stuff. I started off doing ~100 cards a day for the first few weeks, but I couldn’t tell you how many new cards I was doing at first. I do remember quite clearly trying to write every character as I went, but I quickly gave up on it after I realized how much time that would take up.

There must have been a certain point over that summer when I realized the scope of what I was getting into, because i started to get impatient and set my new card/day(NC/D) to 50. Of course, this bumped up the number of reviews I was doing to around 400-500 a day. This period of time probably involved the most struggle out of anything I’ve had to do while learning Japanese. I think a lot of people who are already deep into the CJ learning process forget how it feels to have to parse and memorize these random lines and squiggles, before your brain develops the pattern recognition machinery to break down characters into their constituent radicals. I distinctly remember thinking to myself that Anki felt like trying to carve your brain into a very specific shape with a baseball bat. Needless to say, I got extremely burnt out of Anki, like I can imagine a lot of eager learners do with this program. The day I hit 2000 cards young/mature on the Core deck, I quit. Well, not really, I thought I’d just take a a break, maybe a day or two… and then the reviews pile up, and it turns into a week to recover my usual number of reviews, and then a month, and then you lose it entirely.

During this time, I started to immerse really heavily in anime, movies, manga, youtube, and so on. I really bought into the idea that after I memorized 2000 words, all I needed to do was understand grammar and I’d be able to understand the general idea of mostly everything. As you might imagine, I was extremely wrong, and it took nearly two months for me to realize this. I don’t think that those two months were wasted, because the immersion certainly must have helped somehow, but 2000 words simply isn’t enough to get to where I could start effectively “mining” words (adding words to anki from text that you find in the wild). I realized this in december, and decided to set myself against the Core deck again. I set all the cards I had finished earlier to be infinitely mature, and began starting at 2000, and continued until 2021.

2.2 2021

Unfortunately, I broke my streak right at the new year by getting really into the Witcher 3. I still love the game, but at the time I enjoyed it so much my Anki reps barely mattered to me. After putting in over a hundred hours into the game on a single playthrough, I got back on Anki to catch up on reviews.

This time, I would be in it for the long haul. I eventually finished another 2000 words off the core deck, but I made a small but significant change to the Core deck around the time I restarted it. I made the cards sentence-first instead of word-first. Essentially, i would have to read the whole sentence instead of just one word before flipping the card. This cost me a lot more time, but I realized that I needed to absolutely maximize the amount of coherent Japanese I was reading. If you’re interested in doing this, you can do so pretty easily by introducing some new css:

#hiddenfuri ruby rt  {visibility:hidden;}

And then the front of your card becomes:

<span style="font-size: 70px; font-family:IPAGothic;" id="hiddenfuri">{{furigana:Reading}}</span>

Maybe there was a better way to do this, but this worked for me.

At around 4.1k words off of Core, I began to notice that I was indeed learning more valuable words from reading actual text than doing more deck words. At this point, I turned off new cards from the core deck for the last time. This time though, I wasn’t really burnt out. I had pretty much settled in for the grind, because I fully understood how much more there was and had complete acceptance of it. For the rest of the year and until now, my primary routine would be to read and enjoy Japanese content, and add words I didn’t know into anki and review every day.

I believe I ended up adding around 3-4k words during the second half of my senior year of high school and through the summer. While these words were certainly important, I believe that what was more valuable during this time was actually immersing in a lot of content, but I think that’s a topic for another post.

Over the summer, I began to get interested in another topics, and supplemented Anki with a few miscellaneous decks. For instance, i memorized counters, food, prefectures, poems and some other decks that might have totaled to maybe 1k extra cards. I didn’t keep a lot of these cards, because I mostly matured them all and didn’t want to have clutter every day. I also got really interested in pitch accent around this time, which led to me adding pitch information retroactively to my mining deck (using this script I wrote), and producing a new visual sentence-level pitch accent deck you can find here. I might discuss the rationale behind the sentence-level pitch accent deck in a future post.

After the summer of 2021, I moved into university to study computer science, which of course led to me putting Japanese on the back burner for a while. In December, I began the whole process anew, with Mandarin using a HSK Chinese vocab deck.

2.3 2022

I feel that using Anki for Mandarin for the first time was decidedly easier than using it for Japanese back in 2020. I think this attributable to several factors. First, I am very comfortable with review sessions, Anki, and the prospect of lots and lots of cards. However, I also already have the sinograph pattern recognition and an instant recall of the meaning of various radicals and symbols. This is helped a lot by the fact that I’m currently memorizing traditional characters rather than simplified ones.

3 Statistics

Total young+mature cards
Total reviews
Average reviews/day
Average time
Total time
255 hours
Current streak

4 Lessons learned

4.1 Don’t get scared by how much left there is

Anki is a fundamentally boring and repetitive activity. Pretty much nothing you can do will change this. The repetition is literally necessary to how the program is intended to work.

It’s really, really, really easy to start out and look at the 5,000 cards you’ll have to memorize and lose all hope. I’d know, because I’ve had this happen to me, several times over the past two years. I’m not going to say something cliche like “live in the moment,” but it’s essential that you learn to tune out the sense of impending review pileup while you’re actually reviewing. You have to selectively forget that you still have X000 cards left to learn, or else you might never get to think about that number in hindsight.

I’m not entirely sure what the best way to develop this kind of “resilience against large numbers” is, and I’d be disappointed but unsurprised if the only way is to actually crunch through them yourself.

4.2 Spaced repetition isn’t magic, maximize patterns in content

What I mean by this is that your performance while doing spaced repetition is significantly affected by many things. While this seems, obvious, it’s too easy to fall into the trap of thinking “if i put it into SRS, i’ll remember it!” (even if it might be true for a lot of things).

A few months ago, I tried to memorize binary numbers – a simple task of going between decimal to binary and binary to decimal. After around a month, I had a true retention rate of around 50%. I think the problem was that there wasn’t enough consistent pattern in the cards for my brain to effectively pick out and learn.

This is why card design is so critically important, and why you should almost always customize cards how you like them, and even better, make cards yourself whenever you can. Learn to recognize when the brute-force approach is not going to work and where you can optimize for your brain’s subconscious pattern recognition, which is almost always more powerful than you realize. The key to this kind of optimization is to ensure that cards have consistent, meaningful patterns that are ideally connected to other cards, emotional experiences, stories, or strong memories.

4.3 Tips for good card design

4.3.1 Don’t underestimate visuals

Pictures the brain finds cute,thrilling, exciting, or otherwise emotionally engaging work wonders for memorizing cards

4.3.2 Use context, and force context if you must

Context is the most powerful way to optimize a card. Create context arbitrarily if you really have to, while mining a card. Take a screenshot of where you found the word, add a few lines about what it meant in context, etc.

4.3.3 Minimize clues on the front

Cards with too much information on the front are bad because you’ll start to remember the back from irrelevant clues, rather than ones you find in the real world when you need to recall. Think of this like avoiding overfitting your brain.

4.3.4 Don’t quiz yourself on too much

It’s fine to include lots of information on the back of a card, like pitch or frequency, but don’t try to quiz yourself for correctness on too much. For example, don’t test yourself on recalling reading, meaning, and pitch of word – pick maybe one or two. You’ll still absorb into memory a lot of the other content anyway.