Frequently asked questions

Table of contents
  1. How are your kanji keywords better than those from Heisig's Remembering the Kanji book?
  2. In what order do you teach kanji?
  3. How is your spaced repetition algorithm better than what other SRS apps use?
  4. Which button on the bottom of the review screen should I press?
  5. What are the keyboard shortcuts?
  6. I can change the order of my decks with the arrows on the right of the deck list; what does that do?
  7. I want the cards to autoplay the audio during reviews; how do I do that?
  8. I got myself in a situation where I have hundreds of overdue cards, what do I do? Help!
  9. How are you calculating the difficulty ratings on your difficulty lists?
  10. Will you also add manga to your database?
  11. Do you have a mobile app available?
How are your kanji keywords better than those from Heisig's Remembering the Kanji book? #KanjiKeywords

Heisig's keywords are widely considered to not be the best among the more experienced learners of Japanese. The canonical — and the most extreme — example of that is probably the following:

KanjiHeisig's keywordActual meaning
townvillage
villagetown

In this case Heisig got it completely wrong and assigned the keywords the other way around!

There are many other kanji where he arguably picked the wrong keyword, even though the keyword can be considered to be somewhat related to the kanji; for example:

KanjiHeisig's keyword
exam

So how can we tell how good is his keyword for this kanji? It's not "obviously" wrong like with the town/village, but it becomes pretty obvious that we can pick a better keyword once we look at some of the most frequently used vocabulary which makes use of that kanji:

VocabularyHeisig's keywords
高校
high school
tall
+
exam
校舎
school building
exam
+
cottage

This kind of makes sense, right? A "high school" is a place where you go to take exams, and it is "tall" since you go there after you finish middle school and primary school, and also because "tall" is a synonym of "high". And a "school building" is more-or-less a kind-of a cottage where you take those exams.

We can do better though; here's what we've picked as keywords for those particular kanji:

VocabularyOur keywords
高校
high school
high
+
school
校舎
school building
school
+
building

Oh look, we don't have to do any mental gymnastics now that we've picked better keywords! The word for "high school" is simply composed of the kanji for "high" plus the kanji for "school", and the word for "school building" is made out of the kanji for "school" plus "building". It immediately makes sense!

Of course, this is an ideal example and it isn't always this nice; often it's impossible to pick a keyword that would make sense for each and every word it is used in. However, in general, our picks should be on average better than Heisig's.

Let's look at another example. Here's a pair of closely related kanji for which we've also assigned different keywords than Heisig did:

KanjiHeisig's keywordOur keyword
legfoot
shinsleg

It is true that 足 can carry the meaning of a "leg" in common usage since in Japanese both "foot" and "leg" are phonetically homonyms, however technically the 足 is a "foot", and 脚 is a "leg", and the whole point of having separate kanji for those (since, again, they're pronounced the same!) is so that you can easily distinguish between them in text. And yet Heisig assigned "leg" to 足, and then for 脚 — which is the actual kanji for "leg" — he just picked a random vaguely related word and called it day.

Another habit of his is using somewhat obscure and/or difficult vocabulary in his keywords, more than it is necessary, for example:

KanjiHeisig's keywordOur keyword
decameronten days
hankerlong for
donwear
prostratedlay face down
environslocal area

Not everyone out there is a native English speaker; and while it is sometimes necessary to use a more complex word to properly represent the nuance of a given kanji, in a lot of cases you can just use a simpler keyword that is at least as good or even better.

Of course, it's virtually impossible to pick the perfect set of keywords for kanji. You often have to compromise somewhere, and we're no exception. When picking our keywords we tried to capture the nuance of various kanji as best as we could, so that later it makes the most sense when you're learning vocabulary which uses that particular kanji. We think we did a better job than Heisig did, but nothing's perfect, so you'll have to keep that in mind.

In what order do you teach kanji? #KanjiLearningOrder

Our method has no preset learning order for kanji. The kanji you're taught is picked dynamically based on the vocabulary you want to learn, so it only teaches you exactly what's necessary, nothing more and nothing less.

For example, let's say you want to learn the word "子豚" (piglet). You will be automatically taught the following characters in roughly this order:

  1. 月 (kanji)
  2. 豕 (kanji)
  3. 豚 (kanji)
  4. 子 (kanji)
  5. 子豚 (vocabulary)

We automatically decompose each word into kanji, and those kanji into components, and teach everything to you bottom-up, so that the next thing you're learning always builds upon what you've already learned.

How is your spaced repetition algorithm better than what other SRS apps use? #SpacedRepetitionAlgorithm

Our spaced repetition algorithm is centered around the concept of the forgetting curve, and is directly based on current scientific research into how the human memory works.

In a nutshell we've experimentally came up with a model that, based on the history of your previous repetitions, attempts to model the full forgetting curve of an item, and then we used machine learning along with gigabytes of real, historical SRS data to finetune it so that it can predict the probability of you remembering a given piece of information after a certain period of time.

One big advantage that our spaced repetition algorithm has is that it's very flexible. When calculating your next review interval we ask the model for you: "Hey, when will I have to review this card so that I have an 80% probability that I'll still remember it?", and it might tell us that it's, for example, in 20 hours.

But to the model it doesn't really matter when exactly you'll do your review. You could review your card sooner, say, after only 10 hours, or you could do it much later, after 40 hours. Since we simulate the full forgetting curve the algorithm will be able to handle both cases, and it will dynamically adjust the next repetition interval based on how late or how early you were.

The later you review your card the lower the probability of you actually remembering it, but if you get it right you will be rewarded by a longer repetition interval next time. The earlier you review your card the higher the probability of you actually remembering it, but your next repetition interval won't be as long.

The beauty of this approach is not only that it allows you to review anything anytime without the algorithm breaking down, but also because it allows you to decide for yourself as to what your ideal retention rate and daily workload is.

Do you want a higher retention rate? Shorten your intervals. Are you overwhelmed, or perhaps you want to minimize the time you spend reviewing cards at the cost of slightly worse retention rate? Increase your intervals. The choice is yours.

Which button on the bottom of the review screen should I press? #ReviewGrading

When you're presented with a card for the first time you will have to tell us whenever you know it already. If it's something new then you should take a moment and try to memorize it, and then press "I don't know this". If it's something you already know then feel free to click on "I know this already".

In either case the card you're seeing will be shown to you again in the future.

Any cards which you've already seen will start hidden by default. Once you reveal a card you will have to tell us how well you have remembered it by grading yourself on a 5-point scale:

"Nothing"You didn't know the answer at all.
"Something"You knew something, but in the end you couldn't remember it.
"Hard"You knew the answer, but you struggled.
"Okay"You knew the answer, and it was neither hard nor easy.
"Easy"You knew the answer, and it was easy for you to recall it.

If you forgot the correct answer then you should also take this opportunity to try and memorize the given kanji or vocabulary before grading yourself.

What are the keyboard shortcuts? #KeyboardShortcuts
ActionKey
Show answerSpace
Grade as "Nothing"1
Grade as "Something"2
Grade as "Hard"3
Grade as "Okay"4
Grade as "Easy"5
Grade as "Never show this again"B
Grade as "Fail"1 or 2
Grade as "Pass"3, 4 or 5
Replay audioR

By default the grading shortcuts will only select a given button, after which you'll have to press space to confirm your choice. You can change this behavior in the options.

I can change the order of my decks with the arrows on the right of the deck list; what does that do? #DeckOrder
The order of your decks decides from which deck we will pull any new cards that you'll see during reviews. That is, we won't show you any new kanji or vocabulary from the second deck on your list until we've already shown you everything that comes from your first deck.
I want the cards to autoplay the audio during reviews; how do I do that? #Autoplay
That's actually the default, it's just the web browser you're using blocks audio autoplay by default. You need to go into the settings and turn it on. Here's how to do it in Firefox.
I got myself in a situation where I have hundreds of overdue cards, what do I do? Help! #OverdueOverload

First and foremost — do not panic! Our SRS algorithm is explicitly designed to gracefully handle cases like this. Even if you have a thousand overdue cards you don't have to do them immediately! You can either go into the settings and increase the length of your review intervals, or you can just simply do only as many cards per day as you'd like and ignore all the rest.

How are you calculating the difficulty ratings on your difficulty lists? #HowIsDifficultyCalculated

All of the difficulty ratings on our lists are automatically generated by a machine learning model. The model analyses the exact vocabulary that is used in each work, the kanji, grammar, sentence complexity, etc., and uses that to produce a difficulty rating based on what it has previously learned during training.

Here's a step-by-step walkthrough on how the whole thing works in a nutshell:

  1. We feed the model a bunch of training data, where the training data is of form "X is harder than Y" and is crowdsourced from people who've watched and/or read multiple works and can compare them.
  2. We feed the model a bunch of feature vectors for each work.
  3. The model automatically learns what makes one work more difficult than another, based on the training data it was fed.
  4. The model is then used to generate difficulty ratings for everything in our database, including those works which were never rated by a human.

If you disagree with any of the ratings please feel free to contact us. We're always looking for more training data for our model. The more data we have the more accurate we can make it!

Will you also add manga to your database? #MangaList

Yes, this is a planned feature, although not in the near future due to the huge amount of work necessary.

Do you have a mobile app available? #MobileApp

Not currently. However you can add our website to your home screen and it will effectively work like an app. Here's how to do in on iOS, and here's how to do in on Android.