I wanted to make a mini-announcement about a couple of little changes to Lessons that we’re releasing today. The changes are related in that they’re both aimed at evening out the level of difficulty of the Lesson quizzes across all items.
Lesson Batch Rounding
We’re tweaking the way Lesson batches work, so that instead of having a straggler or two in your final batch of Lessons, those extras will be adding into your final batch. That means that you’ll sometimes have final batches of Lessons with a couple more Lessons than usual. This should help ever so slightly improve retention, because those small batches don’t stretch your memory quite as much and make the quiz a little too easy.
Semi-random Quiz Shuffle
We’re also slightly changing the way the ordering of Lesson Quizzes work, to make sure you won’t see a Lesson and Quiz for the same item back-to-back. Again, this should give a little extra boost for retention, because it eliminates the possibility of answering the quiz right after the Lesson, which is just a bit too easy and so forms a weaker memory.
So there you have it. Not a big change, but hopefully these little tweaks will make remembering items that little bit easier in Reviews. Happy studying
If you have any feedback or questions, let us know below or send us an email at hello@wanikani.com.
This is interesting. I understand if you can’t share, but do you have data behind this? I really prefer doing my lessons in batches of 3, but I guess I can live with this.
I’m going to guess that if you’re used to doing them 10 at a time for example, 2 would feel excessively easy. Some people do jump to the quiz fast (e.g. one of Kumirei’s scripts jumps directly to the quiz without even seeing the lesson at all)
I agree with the concern though. Hopefully WK has accommodated this change to not be excessive with small batch sizes like 3 or with large batch sizes like 10.
They did not change the batch option, so you can still have 3 lessons at a time which will make it 4-5 lessons at times.
A premade system always forces it’s structure on it’s users, that’s the whole point.
And it never satisfies everyone,
As for the lesson picker - it was soft launched, and there’s something to say about the way they did it… hiding it in a reply on a third party api thread and asking specific users to spread the word to people that can’t use their broken scripts anymore. Guilt and growing pain, I guess. At least they’re trying.
I did e mail them about the lesson picker and they told me “It will be a permanent addition that’s more formally announced once we put the finishing touches on it.” That is their official answer.
But yeah, people should know about it officially even though it was soft launched.
This would be the research that SRS is based on. The closer you are to forgetting something (i.e., time marching further forward), if you recall it correctly, you will generally build a stronger memory and it will take an even longer interval to to get closer to forgetting - eventually getting something into long-term memory. Conversely, the closer you are to having just learned something, then recalling it based on your short-term memory, while still better than nothing, is not going to lengthen your next forgetting interval as much.
I believe the data, if you dig it out, shows you need fewer interactions to achieve the same memory strength if those interactions are spaced out more. In other words, put to practical use, the interval before you will forget something depends to some extent on the interval of your previous interactions.
I guess you’d have to look up the data behind SRS and the algorithms used for tools like Anki, it’s definitely out there. But if what you’re currently doing is working for you… just keep going and don’t worry about the theory!
Like I pointed out, I understand the hypothesis, I’d like to see or hear about the data.
To make it clear, I have an opposing hypothesis, at least when it comes to the learning stage.
This hypothesis would go something like “the longer you try to focus, the less focused you are”, which would indicate that learning in smaller batches would be more efficient, especially given that the difference between the timings of learning in batches of 3 and 5 is so small (you will see items about 2-3 minutes later, then 4 hours later, rather than 1-2 minutes later, then 4 hours later).
Anki’s “learning steps” are 1 minute then 10 minutes, afaik, but they’ve been pretty clear from my recollection that the learning steps are mostly a preference thing and might vary with what kind of thing you’re studying. If we were basing things off of Anki, we should ideally be seeing the card ASAP, and then 10 minutes later (perhaps after a regular review), but this is not realistic with any batch setting and not part of WKs modus operandi.
I’m sorry if this comes off crass, but I’ve looked at the data available to me, and it does not suggest this change. On the other hand, it really doesn’t matter too much, WK uses a very suboptimal SRS as is, and it works just fine.
And from there you read the linked papers and thesis, search for others, etc. If you’re lucky, some are on researchgate and not behind paywalls. It’s really time consuming to find relevant research, then sift out enough that is high quality to decide it’s even worth looking at the data to draw relevant conclusions. But yeah, it’s possible, and you could seek research related to your hypothesis as well. As far as I can tell, the two things you’re talking about aren’t necessarily mutually exclusive.
That’s all just to say, it’s not as easy as “show me the data” unless a bored grad student reads your post who just happens to be reading about exactly that research right now when they should really be working on their thesis proposal.
I think @finnra was asking specifically if Tofugu have any data from user’s use of WaniKani that would back up the claim – they weren’t asking you to look for it
I have searched (and read quite a lot). My searches come up with either nonconclusive results or results that partially contradict the hypothesis in OP, see for example this study. SRS and modern Spaced Repetition Scheduling studies such as the FSRS focus mainly on long-term retention, and has little or no information about the “learning phase”, from all I’ve been able to gather.
I am asking WK for their data (EDIT: Or a simple confirmation that there is data). They made this change and I’m assuming they know why they did it.
I understand you’re trying to be helpful. You are not, in this case.
