## Tuesday, January 29, 2013

### Using Pimsleur's Algorithm To Remember Other Stuff

(Update: The incremental video is done, I just need to sync the recorded video and audio together. It's about 30 minutes long.)

First of all, I am very grateful to those that posted a link to the Pimsleur Method in Wikipedia, it proved most helpful. I am most interested in the intervals between reviewing newly learned information; here is what they (apparently) use at Pimsleur:

5 seconds,
25 seconds,
2 minutes,
10 minutes,
1 hour,
5 hours,
1 day,
5 days,
25 days,
4 months,
2 years.

In other words, after you first learn a new word, you are quizzed over that new word after 5 seconds. 25 seconds later, you are asked about that word again. 2 minuets later, you are asked again. etc. etc.
Once you begin discussing intervals that are days, weeks or months long, you are in long-term retention territory, which SuperMemo and Anki already do very well. But for those first six intervals, I think there is something very useful there: 5 seconds, 25 seconds, 2 minutes, 10 minutes, 1 hour, 5 hours.

As an experiment, I tried to figure out if these same increments could be used to remember other things. To do this, I first found a timer app for my iPhone that supported multiple timers. Instead of setting a timer each time I reviewed something, I would rather start six timers at once and review the information when each timer went off. To do this, I simply totaled each increment with the previous one.

By adding 5 seconds to 25 seconds, you get 30 seconds. Adding 30 seconds to 2 minutes gives you 2:30 min. Adding 10 minutes to 2:30 min gives you 12:30 min and so on. For the first few increments, I decided that I should add 5 seconds or so to allow for actual reviewing time. Thus, I ended up with a little (fairly impractical) study system: Learn a new word (or two) with a physical flashcard, then start all six timers. Each time the timer goes off I review the word(s). In between I had to do something mentally stimulating (Play a game, watch the news, etc.) otherwise I would think about the flashcards I was learning and potentially skew the results.

Here are the intervals:

0:09 - (5 seconds plus 4 seconds to review when the alarm goes off)
0:45 - (25 seconds plus the 10 seconds of the previous interval, with 10 seconds to review)
2:30 - (2 minutes plus the 30 seconds of the previous interval)
12:30 - (10 minutes plus the 2:30 from the previous interval)
1:12:30 - (12:30 plus 1 hour)
6:12:30 - (1:12:30 plus 5 hours)

I thought adding a few extra seconds for review to the first two intervals would be good because the intervals are already very short to begin with (Otherwise the intervals might overlap). After you start waiting for more than a couple of minutes, having an extra few seconds seems to matter less, especially after 12 minutes.

But the result of this experiment was this: any word I put forth effort to learn (Which usually meant making a mnemonic or Chinese character connection with), was learned. It doesn't matter if the word I tried to learn was a Hindi word (Which I have no experience in) or a Chinese word (Which I have quite a bit of experience in). When I applied this formula to learning new words, by the time I hit the fifth interval (1:12:30), I had no problems recalling the word. After putting the word into SuperMemo, it has been stable in my mind and I use it with as much ease as I do other words.

Basically, I think we have found a reliable formula that  maximizes the solid, short-term retention on desired information.

Please try this yourself, but make sure you are doing something that arrests your attention in between the intervals (Play a video game, watch a TV show, etc.), don't think about the word you are trying to learn. Otherwise your mental grip on the word doesn't have a chance to strengthen.

While it is possible to use multiple timers to pull this off, it is not extremely convenient to do. If this entire thing were packaged as an app, I think it would be most useful (And perhaps profitable if it catches on). If anybody has any experience with Objective C, please leave me your info (I won't publish the comment if it contains your contact information).

Here's how the app would work (Roughly):

Create a flashcard of something you want to learn (Vocab word, phrase, etc.) and press "LEARN." This would start a timer that goes off after 5 seconds. The alarm goes off, you review the word. If you get it correct, it goes onto the next interval (25 seconds). Incorrect, it goes back to 5 seconds. Keep repeating this pattern until you hit the 5 hour interval, and you've now learned that word. The word can be put into Anki, SuperMemo, etc. and then deleted (From the short-term flashcard app). BUT, once a word has gone onto 2 or 10 minutes, it would be very easy to add another word into the mix (Each with its own set of timers). Once that word's intervals goes to 2 or 10 minutes, add another. Like trying to keep track of spinning plates, the app would keep track of which cards need immediate review, so you don't have to fiddle with timers and physical flashcards.

Also, if you know of an App that already does this, please mention so in the comments. Thank you very much.

I'm fairly confident this will be useful to lots of people (Myself included).

1. Great idea!

Why not ask Piotr Wozniak, SuperMemo developer, what are his thoughts about such short term memory algorithm? Maybe he would consider creating additional 'timers' module in SuperMemo that would work in conjunction with main SM algorithm instead of working as another standalone program?

2. Oooh! That sounds like a great idea! I'll compose that e-mail now!

3. Wozniak has said he doesn't believe in microintervals.

"Myth: Review your material on the first day several times. Many authors suggest repeated drills on the day of the first contact with the new learning material. Others propose microspacing (i.e. using spaced repetition for intervals lasting minutes and hours). These are supposed to consolidate the newly learned knowledge.
Fact: A single effective repetition on the first day of learning is all you need. Naturally it may happen, you cannot recall a piece of information upon a single exposure. In such cases you may need to repeat the drill. It may also happen that you cannot effectively put together related pieces of information and you need some review to build the big picture. However, in the ideal case, on the day #1 you should (1) understand and (2) execute a single successful active recall (such as answering the question "When did Pangea start breaking up?"). One exposure should then suffice to begin the process of consolidating the memory trace"

from http://www.supermemo.com/articles/myths.htm

I agree with you that microintervals help you learn the material thoroughly the same day. It does come at a large upfront cost however. Maybe Wozniak's opinion has changed however.

1. I agree when it comes to information that fits into your working knowledge (You have a BASIS for remembering it), but when it comes to stuff totally foreign (Like foreign languages), you need to lock it in somehow. I think shorter intervals might be the answer.

2. Yup, I agree. I've only been seriously SRSing for 6 months, but I feel it's gotten me very acquainted with my own memory. I can remember psychology or biology facts after glancing at them only briefly, but languages are more difficult. You mentioned at one point 'priming' knowledge, which I thought was a very interesting idea, making it stickier. I think microintervals like you mention achieve that efficiently. I've also noticed that just passive exposure makes knowledge stickier. This accounts for part of the benefit of massive input for language learning I believe. For example, by the time I learned what '那麼' meant, I had heard it hundreds of times, so the sounds already were familiar to me. After all that passive exposure, attaching a meaning was easier.

The metaphor I think about is that things like passive exposure, microintervals, or procedural repetition (shadowing a language) really help to 'carve out a bookshelf' in your brain. You have managed to construct the neural circuits corresponding to words, etc. Then, attaching a meaning to that neural circuit is easier, like putting a book on your bookshelf.

Another thing I'm interested in is the issue of semantic versus acoustic knowledge. Knowledge in the short term is stored acoustically, and semantically in the long term. That is, in the short term it's easier to remember the order of a list 'big, huge, gargantuan' than it is to remember the list 'b,p,e,c.'

6 months from now though, remembering whether I used the word gargantuan or mammoth will be very difficult. And also: we remember the gist of conversations, but never the exact wording.

I'm curious how this interacts with learning, and whether it explains part of the much greater ease I have of memorizing 'scientific facts' over foreign languages.

3. Neat! I think we are very much on the trail of solving the short-term "unsticky" knowledge acquisition problem. And it will more likely have SOME KIND of transferable properties that aids in other skills.

4. I started using this & didn't find it too difficult to just use the standard timer in the Clock on my iPhone. For the 5 & 25 second intervals I just started the stop watch & did some free association (saying 1 random word per second until 5 or 25 seconds came up). For the longer intervals I wrote the question on 1 side of a small scrap of paper & the time of the alarm on the other side. When the alarm goes off, say at 11:11, I just look for the scrap of paper that says "11:11" & review it.

Seemed to work VERY well. I got all of my reviews correct. I think this is more efficient than taking several days of getting it incorrect to finally lock it in.

1. Woo hoo! That's great to hear!

I'm currently trying to memorize five new Chinese characters per "alarm set," and reviewing all five characters when that alarm goes off. I currently have five sets of timers going, and I don't think I will add anymore today. I will probably wait a week to see how well I can remember these characters when I see them in SuperMemo. But if this keeps working, I don't see why you couldn't use 20 or 30 sets of overlapping timers to memorize more than a hundred new words or characters in the course of a couple of hours... I'm excited about the implications of this, but I'll have to wait and see what the results are first.

2. Yeah, stacking the cards into groups works just as well. In the past I used to get ~85% of my Anki reviews correct. I changed the intervals so that they are shorter & brought my % up to about 93. This morning I got 97% correct, a new record for me. My new strategy is to Mark the cards I miss, make physical flash cards out of them, & study them using these Pimsleur intervals throughout the day.

5. Anki lets you set custom intervals, 1st disable advance by 20min in preferences. Then set the interval timing for new/lapsed cards in options.

Originally I was using the fib sequence of 1,1,2,3,5,8,13 for new materials as wikipedia points out that mass presentation works better for new materials. But after reading your post, I changed the intervals to 0.09,0.42,2,5,10,60,300 and saw improvements to my studies. I was able to remember more cards than I normally would and I believe this is due to the immediate repeat of the material plus the 1 hour and 5 hour pause.

I can't give you any statistical data since I only recently switched to anki from mnemosyne.

Above all, thanks for this great site. Looking forward to that vid on Inc Reading.

hpcLover.