Problem with intervals [To BP admins]

I know you will probably be reluctant to listen, but at least please read what I’m about to say.

It is about current intervals for SRS. I think they are not optimal.

Currently, as stated in the FAQ: 4hrs, 8hrs, 24hrs, 2 days, 4 days, 8 days, 2 weeks, 1 month, 2 months, 4 months, and 6 months for finished

That is
1-5: 4hrs, 8hrs, 24hrs, 2 days, 4 days
6: 8 days
7: 2 weeks
8-12: 1 month, 2 months, 4 months, and 6 months for finished

And the problem lies in 6, 7 and 8th intervals. This is the time when the grammar point transitions from your current memory into your long term memory and there has to be more reviews in that period.

My experience is that when the 8th level review comes up - I no longer remember a lot of the grammar points. Unlike vocabulary, where you just link a concept with a word - for grammar there has to be more reviews.

Ideally there should be 15 reviews spread out across the same intervals (i.e. up to 6 months), but I understand that you are unlikely to add 3 new sentences for each of the grammar points, but please at least consider adjusting the existing intervals such that there are more reviews in the initial phase.

My suggestion would be:
4hrs, 8hrs, 24hrs, 2 days, 4 days, 7 days, 12 days, 20 days, 1 months, 3 months, and 6 months
As you can see, I didn’t touch initial intervals (they are perfect). But slightly moved the intervals in mid and late sections to put more emphasis on reviewing while you still practice a particular grammar point.

Please, I hope you can seriously consider this. I think it would significantly improve overall effectiveness of learning on BP.

14 Likes

I agree that the intervals are not optimal and that there should be more than 12 reviews in total before the grammar points disappears forever.

But I’m having most problems with reviews on SRS 9. So if I’m not mistaken reviews on SRS 9 were seen the last time 2 months before. That seems to be too much for me since I get a lot of them wrong. So contrary to your suggestion I would not increase the 2 months to 3 months but decrease the interval to maybe 1.5 months and then 2 months for the next SRS.

Would be helpful to have the data about the SRS levels on which most users get wrong answers.

5 Likes

This suggestion would actually address this. By the time you hit the 2 or 3 months delay you would have had 2 more additional reviews. Which would ensure higher chances for retention. And the last 3 intervals are there mostly to just ensure that you didn’t forget anything. There would MORE learning happening initially.

6 Likes

So if I see this correctly, from 12 to 20 day is an 8 day increase, then followed 20 to 30 days (10 day increase). If correct, the increased day spread as 3/5/8/10/60/90 as opposed to 4/6/15/30/60 established. I know you mentioned 15 examples as ideal to even it out but I have to say the 60 to 90 day spread is much further as I look at it. Sorry I don’t have the awesome charts the that @Pep95 shares.

I have no ideal what are considered optimal intervals so I have no strong opinion. I’m no expert, but seems the science isn’t clear either (https://www.dartmouth.edu/~cogedlab/pubs/Kang(2016,PIBBS).pdf) but this is just one publication. It may be worth exploring this a bit further however BunPro is NOT a traditional SRS format so the waters are uncharted.

I think this gets convoluted due to each example being 1) Different from the last 2) Grammar points have multiple translations. So if one is getting stuck on later examples, how do we separate time interval issues opposed to seeing a completely new sentence with a new perspective on usage (which may be the underlying issue). If the later, interval changes will have a limited effect and may not be the problem as you headlined.

The way I see it, if I’m streaking on the long-term, then SRS is working (regardless of the interval). If I’m get sentences wrong mid-way, SRS is still working but all these different examples are adding a new dimension. And I’m perfectly fine with putting in wrong answers as clearing grammer points is a product of my learning rather than the goal in itself. It also may be sign that I’m not using the grammar point in the wild as I should (read/writing/conversation)

I WOULD like the flexibility to review completed grammar as I suggested the other day (feel free to read) to help with the longer term intervals in case I don’t have the opportunity in real life. This wouldn’t be an overhaul of the SRS design that effects all users but rather a secondary optional feature that is rather simple to implement and use.

6 Likes

How much time are you spending on your reading outside of Bunpro? I know I read a lot and still struggle with certain grammar points that don’t show up often (ものの and ならまだしも are two that just never stick for me), but I know for a fact that it helps immensely. Reading acts as a “natural” SRS and ensures that you’re encountering grammar in the wild where it’s more likely to stick.

SRS is great for helping get you up to speed but it’s not a be-all-end-all. Maybe the grammar you just pushed into level 6 will show up in your reading and you’ll have an “aha!” moment that will cause your brain to sit up and take notice.

9 Likes

This is precisely why I end up adding lots of extra sentences to Bunpro, the timing gets too far apart with only 12 sentences per topic. I’d gladly take an increased frequency in reviews (and more content - 20 sentences per grammar point! 40!) to really pound them into my head. As many problems as I have with AJATT, the core facet - familiarity breeds fluency - rings true. The more often you see something, the less likely you are to be surprised by it - I’d love to get so good at grammar and/or see tons of sentences so many times that Bunpro is boring. :smiley:

6 Likes

Following up on this: while watching 月がきれい the other day I finally ran across ならまだしも in the wild and it felt like the neurons finally connected. The next two times it came up in reviews I got it immediately.

7 Likes

@Anthropos888, @s1212z @jacobalbano @monkeytunes

Thank you guys for the insightful discussion about the intervals. I think we are getting to the point where we have enough review data to pinpoint trouble intervals. We will take your suggestions into consideration as we make adjustments going forward.

7 Likes

Was there a follow up to this thread? I’m curious what the data says (if anything). :nerd_face:

1 Like

I only just saw this, but I think it is a good idea. If I had any suggestion to add, it would be to make the SRS more like anki, as in you can choose if you still remember it well, if it’s ok, or if it’s a little hazy, and have the intervals adjust accordingly. When you have so many things on SRS at the same time, being able to dynamically control the interval with more memory level options helps to spread things out better.

2 Likes

Adding… Since there’s basically two mechanisms for dealing with reviews, 1) get it “wrong” and trigger Ghosts or 2) get it “right” and don’t see it again for a longer time period, when I just “sorta” know something, I purposely mark it wrong to give myself more reviews. I like the idea of a “hard” and “easy” button in addition to the “next” button, so that if you keep using the site like we currently do, it continues with the same behavior.
That said, it’d also be nice if the data shows some trends that helps tailor the standard schedule for all users. :nerd_face:

I’d appreciate anki like good/difficult/etc. buttons too.

About adding more reviews in general though - I think that’s the wrong approach. If a grammar point isn’t making it into long term memory then doing more reviews at the time you expect it to transition just means you’re allowing yourself to keep it in short term memory by constantly reviewing it, without the SRS stage reflecting that you haven’t actually learned it yet. If failing a review at that stage doesn’t result in enough reviews then the problem is that failing doesn’t drop it back far enough.

But of course everybody learns differently so it would definitely make sense to have all this configurable too…

3 Likes