Back-to-back in reviews; am I harming my learning?

Probably makes you more Burnt though :eyes:

2 Likes

I think you answered your own question there. :slight_smile:

To add to the informal poll : I also do back-to-back but it’s random which comes first. I figure my brain ideally pops up meaning and reading when I see the kanji, so this reinforces both. It also corrects me right away if I have one of them wrong!

I agree with @schtitt
I’ve tried using them and found that it greatly increases the review clearance speed.
However, when making mistakes, it is harder to correct and sink them into memory. I suspect because there is not enough time to correct the wrong in my head, unlike random order.

2 Likes

As someone who’s fairly close to finishing WaniKani after a long journey, I will say this: the main goal of your learning should be to do whatever it takes to continue to learn.

There’s an adage in software development that goes something like “the last 10% of the work takes 90% of the time”. In the context of learning a language and depending on your goal, seeking absolute perfection may actually be a detriment to your overall learning. Switching my reviews from “scrambled” to back-to-back reading/meaning was one of the best decisions I made because it significantly decreased the amount of time I was spending on reviewing at the detriment of maybe failing to recall 10% more items. In the long run that doesn’t matter to me, because eventually if I need to learn something, I’ll learn it in the context of an article, or a game, or a show, or a conversation.

For about half my time using WaniKani, I used it without any scripts or modifications, and found myself getting derailed and discouraged by vocab I’ve basically never had to encounter in the wild again. Not to mention, WaniKani is an isolated flash card system, meaning you’re recalling meaning and pronunciation without any context. I found myself spending an excessive amount of time working on reviews in a way that I knew I wouldn’t be asked of when I wanted to read or play a game or watch something. And so using scripts or modifying the way I used WaniKani allowed me to learn MORE, even if I still get stuck on a vocab I learned at level 20. Because in the end, who cares? Especially if in the time it would’ve taken me to learn that, I actually solidly learned 50 other words and kanji that I encounter much more frequently in the wild.

Any way that’s my 2 cents. I’ve spent years learning Japanese and I kind of wished that I didn’t stress so much early on about “how” I was learning, and just allowed myself to make mistakes and immerse more.

7 Likes

There’s a reason why WK uses random order. The developer definitely cares. What works for you might not for everyone. Learning is better using random order than systematically group them. Grouping is easier but less effective. Sure it doesn’t make sense and harder for the brain, but it’s more effective for learning. If you feel wasting time using WK, why use it in the first place? Just use another system that uses contextual SRS instead, for example.

Here’s a reference I found: https://link.springer.com/article/10.1007/s11251-007-9015-8

4 Likes

I can’t read the full article because it’s behind a pay wall, but from reading the abstract, I’m going to respectfully disagree. I don’t think this study is supporting your claim. This study is evidence spaced repitition works, in that reviewing old material with new material is more likely to retain memory than if you were to bunch all the reviews from a single lesson at once.

Experiment 1, college students learned to solve one kind of problem, and subsequent practice problems were either massed in a single session (as in the standard format) or spaced across multiple sessions (as in the shuffled format).

The way Wanikani works isn’t analogous to a Math textbook. In the standard math textbook this study is critiquing, you would learn something in lesson 1, review it all at once at the end of lesson 1 in the form of X number of questions, and then never return to those reviews as you progress through the book. From what I can gather, they’re arguing if there are 10 questions at the end of Lesson 1, the student should study 1-2 of them, and then the other 8 will appear at the end of other lessons. While not exactly spaced repetition, this works for similar reasons. Over time as they progress through the book, they review old material. They’re not cramming, they’re spacing their reviews.

Wanikani doesn’t present you all of Lesson 10’s items and force you to review them once and only once. By design you’re reviewing them over time, as the reviews become available again. It’s just not analogous to this study at all. Similarly, reviewing the meaning and reading at the same time isn’t the same as “bunching” your reviews, because you’re still reviewing the context multiple times over time, exactly what this study is arguing for.

In Experiment 2, students first learned to solve multiple types of problems, and practice problems were either blocked by type (as in the standard format) or randomly mixed (as in the shuffled format).

This experiment might support your claim, but I don’t know what they mean by “types of problems” and so I really have no idea what the equivalent would be in Wanikani. Does this experiment imply a type of problem is subtraction, and that grouping a bunch of subtraction problems together is detrimental? In the case of Wanikani is “reading” in general a type of problem? Or is the reading of a particular word a “type of problem”. If it’s the former, even if you’re reviewing meaning and reading of a word back to back, you’re still shuffling both those “types” of problems over the course of a full wanikani review. You’re not just doing reading over and over, and then moving on to meaning. If it’s the latter, a math problem is structured completely differently than recalling vocab of a particular word. I just can’t see how you could try and get this granular with this particular study and try to apply it to learning a language without actually conducting the study with language learning. Many other SRS systems provide reading and meaning back to back and yet people overwhelmingly report success using something like Anki.

To your other point, I hope I didn’t imply that the developers of WaniKani haven’t thought through their system. But I also don’t think WaniKani is a perfect system. There’s no such thing. For example, Wanikani uses spaced repetition, but on reviews, doesn’t present you with the review that is most important to review first, it just gives you them in a random order. Here’s a user script that attempts to fix this: WaniKani SRS Reorder Button. There’s plenty of evidence that timing matters in SRS. If you learned something 3 hours ago, and something 6 months ago, it’s more important you review the one from 3 hours ago sooner, because you’re approaching the time when your brain is most likely to forget it. A mature review can wait another day and the difference in recall will be negligible. The reality is everyone is limited in their time. Maybe you have 200 reviews and only have time for 100 of them. In that case it’s in your best interest to prioritize accordingly.

All this being said, I use (and LOVE) WaniKani because it offers a structured way to learn kanji and vocab, and provides built in reviews in the form of spaced repetition, which is one of the most effective ways to commit something to memory. In other words, it’s an amazing, incredible tool in my language learning repertoire. To my original point, just because I might think forcing yourself to agonize over perfecting learning specific words is a waste of time doesn’t mean I think WaniKani itself is a waste of time. Leeches are a common concept in SRS. Many other SRS systems will just bury a leech item for this exact reason, because it’s not a good use of your time to force yourself to memorize something using flashcards if you’re spending too much time on it. More likely than not, if you eventually need to learn a word, you will do so in another context. This is not an argument against WaniKani by any means, and offering critique of it doesn’t imply it’s not a good resource.

In the grand scheme of things, if your goal is to learn a language, and because of the design of a particular system, that person is discouraged and eventually drops off learning, then this is all for nothing. That’s what I was trying address in my reply to the OP. Literally none of this matters if they don’t continue their learning journey. And if modifying WaniKani so that you choose to do your meaning and reading reviews back to back enables you to learn more and experience more of WaniKani, and in turn learn more Japanese, then that’s a no brainer to me. People get hung up on perfection and optimization. None of that matters if you give up in 3 months.

8 Likes

The main advantage I’d expect is just the chance to correct a reading or meaning mistake in the same session by attempting to recall the correct version when seeing the same item the second time, albeit for a different purpose (e.g. to enter the meaning first in the review dialogue, then to enter the reading). However, for all the random order advocates here… you do realise that it’s a known fact that associative memory is a very powerful tool, right? Associating something new with something you already know is a very helpful way to remember it, and I think that structuring knowledge and linking pieces of knowledge is similarly helpful. Now, perhaps none of this matters if you have no intention to link the meaning to the reading, or to link the mnemonics for the two, and that’s certainly quite possible, but if you are trying to do so, then I don’t see why you wouldn’t want to group reading and meaning reviews for a given kanji or vocabulary item. If nothing else, at the very least, reviewing the two together should help you create links purely due to repeated appearances in the same context.

Case in point: today, I decided to start learning the lyrics for 残酷な天使のテーゼ (The Cruel Angel’s Thesis, the OP for Evangelion), and one thing I did was to sing it while watching a karaoke video. Just because of the sheer number of times I repeated the video, I now remember that there was a segment where the background video looked like it had been filmed from a bicycle or moving vehicle just around the point where「あなただけが夢の使者に呼ばれる朝がくる」is sung, so now everything I sing the song (at least until the lyrics themselves are perfectly memorised), that video will be triggered in my mind and help me to remember the words, or at least give me a context from which to recall what I need to sing. (That’s also helpful because, as is the case with many songs, The Cruel Angel’s Thesis contains many repetitions of similar melodic elements, which could lead me to confuse parts of the song if I don’t have other markers.) I don’t see how doing something similar in the context of an SRS would be disadvantageous, especially because one’s ultimate goal is to be able to look at a kanji, pronounce it and remember what it means immediately. That’s all the more reason to associate these three elements as closely as possible, in my opinion.

2 Likes

That is what was pointed out to me in this thread, and which I didn’t consider in my first reply. Which led me to the half/half position you’re replying to. But I have a hard time understanding why it’s so controversial to say that trying to recall something twice, no matter whether you are linking reading/meaning or not, will somehow have a beneficial effect. And this can be true without diminishing what you’re saying about associative memory. It seems to me there are possible benefits to both approaches.

Just by observing my own mind while I’m doing reviews this seems pretty evident - having to dig up the reading/meaning pair for a given kanji twice is more hard work than doing it once per review session. That hard work, I imagine, must have some effect on the neural pathways - again, exactly how big that effect is is total speculation. But that it is >0 seems pretty clear to me, and I wonder why so many are reacting to that statement, because we’ve already seen that there are definite benefits to the back-to-back process as well :man_shrugging:

Edit: And in the end, neither of these distinctions matters more than learning in a way that keeps you learning, as also many people have mentioned here.

1 Like

Well, I can’t speak for everyone else, but in my case, it’s the idea that random order is necessarily better/more effective that bugs me. I know that’s not necessarily your position, and you’ve certainly qualified your stance to allow for the possibility that each approach has its benefits, but you’re also not the only person who responded to this question, so my comment wasn’t only directed at you, and I’m sorry if it seemed that way.

That aside, I guess some people feel that it only makes sense to choose the more efficient method. Now, I’m not sure myself that random order is necessarily less efficient, but intuitively, putting readings and meanings together seems more structured, and therefore perhaps more efficient, which might lead some of us to ask, ‘Does it make sense to keep the other order just because streamlining things makes them “too easy”?’ I mean, I know this isn’t your intention, but that does smack a little of how some teachers teach horribly/make their lessons really hard to follow, only to tell students that they’ll learn better because they get to experience the ‘real difficulty level’ of the material. Maybe that’s what’s fuelling these reactions.

Finally, just on a personal level (and this is entirely me and my personal preferences), I’m a little irked by the idea of treating readings and meanings as separate things. (To be clear, I don’t use the WK SRS at all, so if anyone’s tempted to tell me to ‘stop using a system [I] don’t like then!’, there’s no need to do so.) I understand that many of the meaning and reading mnemonics are connected somehow, so it wouldn’t be fair for me to say that WK encourages users to separate them entirely, but my personal belief is that it’s better to remember meanings via readings where possible, possibly by using a mnemonic to force the reading to fit the meaning, especially since at some point, one would like to be able to hear the reading and automatically remember the meaning. If using linking the meaning directly to the reading isn’t possible, then remembering via the kanji itself will probably be necessary.

True enough, and I think studies have shown that more or less any sort of effort required for recall strengthens future memory, and that’s why testing oneself is one of the most effective ways of improving one’s learning outcomes.

Well, yes, and so it’s ultimately up to people’s personal preferences beyond a certain point, but I just felt the need to defend the idea that taking readings and meanings together is not necessarily harmful/less effective. (Your original post – not the one after you adjusted your original position – is not the only reason for this; another post included a study about mixed practice for maths problems, and while I’ve read a possible explanation of the effectiveness of such methods, I’m not sure if the situation with kanji meanings and readings is necessarily analogous. My post was meant as a response to the idea of dismissing structured reviews as necessarily less effective.) Everyone is of course free to choose and is probably best served by picking the method that they most enjoy, but I felt I should raise the importance of associative memory.

After some reflection, this is what I think: I’d like to advocate learning meanings and readings together for both the initial acquisition phase and any correction steps (i.e. when someone makes a mistake, both the reading and the meaning should be reviewed, though perhaps only at the end of a testing session in order to avoid giving away the answer for the other element). However, I suppose that it might not be bad to review everything in a random order, and perhaps the resultant increased difficulty will strengthen memory further, so using random order for reviews themselves might not be bad. Perhaps combining things this way will allow people to reap both the benefits of associative memory and those of making additional mental effort.

2 Likes

You’ve gotten some good responses so far. Personally, I tried it that way for a few levels and found that my recall after Guru went way down so I switched it back to random order.

You may have to just try it and see how it works for you in the long term.

Anecdotally, this is what I’ve found to be true. Mentally saying both meaning and reading during a review regardless of which one was asked, and having to do that twice with some time separation in between, greatly improved my recall once the SRS intervals got to a week or more.

One caveat is that I show the full answer after a review, so if I missed either but typed in the correct answer for that specific review, I would intentionally mark the same item wrong later on to preserve the integrity of the system.

2 Likes

I’ve been doing this for the past 4ish levels, and I’ve seen a big boost in actually connecting the reading of a word to the meaning. I don’t know if this is just me, but I tended to look at onyomi compounds more as two kanji connected rather than one whole word if that makes sense. Every time I would see certain words I would individually remember the readings for the kanji instead of actually learning the reading for the word. Because of that I had a lot of words where I couldn’t connect the reading to a meaning and vice versa. I would often find words in the wild that sounded vaguely familiar, and when I searched the meaning I had already seen them in WaniKani.

Judging by the other replies on this post, this seems to be a pretty divisive topic. For me personally back to back has been very beneficial. I’m seeing increased review accuracy, and outside of WaniKani I can recognize the words I’m learning.

2 Likes

That is fair, and I think that is partially what’s going on as well. I certainly know that I tend in this direction myself, i.e. believing that being exposed to difficulty will make you somehow more robust in mysterious ways. I am trying to separate this pull (fetish?) in myself from the subject at hand in this case, but of course it’s hard to completely account for all the subconscious things one might believe.

As for meanings/readings being connected I am actually a bit curious what you mean. I can certainly see that it’s very beneficial to do this with words, compound kanji, since they actually make up a quite unique reading. But when you have about a hundred kanji (lol) that reads ‘kou’, is this still something that is possible? In the case of kanji, to me it makes more sense to connect the reading/meaning/actual shape together in a sort of trinity - the reading without the actual shape of the kanji seems not so useful to me.

For example what @weirdconnie is talking about here - is that kanji and vocabulary, or vocabulary only?

1 Like

There’s plenty of research about where having shorter SRS intervals doesn’t prevent leeches or lead to better recognition in the end. I think it’s pretty comparable to this comparison. So you are more or less just wasting your time drilling that same item twice while it’s still in your short term memory. It doesn’t lead to better long term retention. I did WK normally, though.

Anki tests both together, as well. In the end SRS only gets you maybe half-way to ‘knowing’ the word, so doing it back-to-back doesn’t really have a significant effect, IMO.

4 Likes

Yeah, I agree with the idea of a ‘trinity’ approach, actually, but what I meant is that there ought to be a way to connect the reading fairly directly to the meaning or the kanji, or both at the same time, if possible. Since you raised the example of ‘kou’, here’s how I do it with 口, whose on’yomi is こう (and is pronounced kǒu in Mandarin):

As a result, I remember that it means ‘mouth’, and the reading, meaning and kanji are all connected.

Of course, this is easier to do with simpler kanji that literally represent what they mean, but you can seek out a unique way to associate each reading with the kanji and meaning at once.

Here’s another pair of examples to demonstrate how you can do it even when two characters share the same reading:

I don’t know exactly where all these ideas come from. A lot of it is etymology-based, or at least runs on pop etymology (i.e. stuff that seems to make sense that I make up myself even though I know it’s probably historically false), and much of the time, it’s something my brain comes up with spontaneously. Also, in some (perhaps most?) other cases, I don’t use mnemonics at all and just focus on learning how to write each kanji, but that might just be because I’m a Chinese speaker and I’ve been writing these for years. Nonetheless, I do tend to find that my brain automatically ‘assigns’ the sounds I need to pronounce to parts of kanji when I learn them, so I’m just expressing roughly what goes through my head in the mnemonics above. Either way though, I’m just saying that there is a way to do this for almost everything, even for something as common as 今日(きょう), which I think of as being linked to ‘today’ via the O that each word has: ‘kyOu’ and ‘tOday’. I see a curved tunnel running between them, and I visualise the kanji as I pronounce the sound きょう. Very little of what I do relies entirely on logic, however: my aim is to make all this vivid, emotional and visceral. I prefer to feel kanji and sounds as opposed to thinking about them, because emotions are more memorable than chains of words or sentences.

If you want even more examples, here’s my mnemonics thread, which I haven’t updated in a while:
https://community.wanikani.com/t/non-wk-mnemonics/49513?u=jonapedia

1 Like

Wow, that makes sense. Wish I had come across this a bit earlier than level 59 :slight_smile: I like that approach, seems to tie things together in a more tight-knit way.

1 Like

It has benefits. I read this recently and the study shown 150% improvement with reviewing twice for each word. So one word must be correct two times for each session, randomly ordered.
Ref: Optimizing distributed practice: theoretical analysis and practical implications - PubMed

I can’t find anything like that in the study. It’s comparing the different gaps between learning sessions, so I fail to see how it is at all relevant.

Read the design study; I am saying reviewing twice has benefits as it has been done in a study. You claimed that it does not lead to better long-term retention. Anecdotal, or is there any study based on your claim?

That study doesn’t measure the effect of reviewing twice vs. once in a session. All the groups had the same study sessions with different gaps, so it doesn’t really say anything about reviewing the item twice in the same session.

I definitely agree with @morteASD , I don’t think we’ve seen any relevant studies to this questions posed in this topic. The whitepaper’s conclusion is simply that longer study intervals are better for long term retention, which is basically the concept behind SRS, nothing new here.

In the first study, for the first review sessions, they did require 2 correct answers for each item as part of the successive testing review cycle before it was removed from the list, but when they got to the 10 day interval, they stopped doing that. The paper doesn’t quite explain the intent, but i assume based on the short interval it had some rational to initially learn the material.

The overall takeaway from the paper is that re-reviewing something on intervals in the order of seconds and minutes is not helpful, so you could argue that it agrees with the 1x1 method except for the fact that none of the tests in the paper separated out related information as part of the testing process in any method (e.g., meaning and reading) so i don’t understand how its relevant at all.

I personally switching to 1x1 because like others have indicated, i feel like seperated out the reviews actually prevents the correct connections from being formed as its follows the meaning of one work with the reading of another instead of reviewing in pairs. I think this causes alot of problems for me. I don’t believe its equivalent to reviewing twice unless you’re doing it wrong, since you are only tested on half at once. Another problem is that we are tested on multiple related things, so sometimes i get one side of a card right (maybe meaning) but i forget the reading, and it isnt until i do another related card that i remember the reading of the first card. This means i’m getting some cards correct that i would have failed if they were back to back.

At the end, I’m not going to pretend to know what’s right, but given an equal amount of time. I’ve started going 1x1 for faster reviews, and then rereviewing failed cards from the last 24 hours using the self study scripts. This seems to use time best to rereview the items that haven’t been retained.

2 Likes