9

I found that some words hard to recall after several minutes or on next day when learned them from flash cards.

With longer period of time between recalls (5/10/60 min) and after working with another cards I see larger effect of forgetting. Therefore during studying knowledge retains in short memory and doesn't move to long memory.

How long/often should I practice on card within a day to effectively acquire knowledge? Note that I like to preserve card quality and only interested in timing patterns - how many times, how long each time and how long delays between practice.

Is it right to intensively learn card in single day and practicing recall on next day?

What about alternative strategy - to learn cards during several days without requiring 100% recall within certain period?

TL&DR

I have a cards in AnkiDroid made from publicly available VOA Simple English wordlist with definitions. Cards are of moderate quality and explanations sometimes ambiguous. Better quality of study materials improve learning but I interested in repetition patterns itself.

There are a lot of studies on SRS. Common approach is to use geometric progression on repetition intervals: I(n) = I(n-1)*Coef where Coef usually is 2.5 and can usually vary within 1.1 - 3.0 (because of adjustment depending on card difficulty).

Many users of SRS technique missed that SRS only improves comprehension / memorizing. For details see:

I heard a lots of experts said that nearly 30 or 40 repetition of word/idiom/construct (in clear/well defined context) is required in order to memorize word (so you are able to understand specific meaning and able to use word 100% correct).

I found that the article Learning, made joyful nicely mix quizzes with new cards and ask for repetition at point near that I start to forget.

Putting recalling of already learnt words/phrases/grammatic/etc aside how often should I repeat SRS cards or make exercise on specific grammatic/vocabulary in order to remember?


I believe that SRS cards can be used to learn new topics if cards have high quality. I make cards by myself from vocabulary in text form and constantly improve cards if I found them difficult or unclear.

On other hand I think it is more useful to make textbook exercises instead of drilling SRS cards with clozes (sentences/phrases with missing words).

Of course I interested in optimal frequency of exercise repetition on given grammatic/vocabulary/topic. I found that one study at class + homework + exam (3 times in total) is not sufficient to acquire knowledge.

Anki SRS software has two modes: learning and recalling but default learning intervals set to 1 and 10 minutes which I think have no any sense if you learn new material. This and this discussion suggest that you should have between 4 and 6 repetition on the first day and you should have some intervals several hour large.


After private communication with author of Spaced repetition article I understand that I am looking for efficient way to move data to long term memory.

Paul Pimsleur's graduated-interval recall from 1967:

5 seconds, 25 seconds, 2 minutes, 10 minutes, 1 hour, 5 hours, 1 day

Of course these intervals are for his audio-lingual learning method (not for general learning) and given that he died in 1976 may not represent cutting edge in science.

As say advertisement:

Dr. Pimsleur’s research on memory was perhaps one of his most revolutionary achievements. He discovered that if learners were reminded of new words at gradually increasing intervals, each time they would remember longer than the time before. He documented the optimal spacing for information to move from short-term into long-term, or permanent, memory.

Quotation from his original article shows many guessing in his observation:

Still, the evidence both from my programming experience and from the findings of experimental psychology seems to indicate that there is an "ideal" schedule one can keep in mind and adapt to fit the circumstances. This schedule is exponential in form. That is, if the first interval (between the original presentation and the first recall) is, say, five seconds then next interval may need to come 5² = 25 seconds later, the next one 5³ = 125 seconds (2:05) later, the next one 5⁴ = 625 seconds (10:25) after that, and so on. The first interval can roughly be defined as the time that elapses before the student's probability of remembering the item drops to some arbitrary level, say 60%.

For long runs SuperMemo study suggests E-factor in between 1.3 and 3 with 2.5 as starting point, and definitely not a 5!

  • So you're asking for an algorithm to learn something, as opposed to recall it? I don't think such a thing exists. But I think the question makes more sense now. – Flimzy Oct 20 '16 at 6:29
  • It's very unclear what kind of answer you are looking for. Do you want a specific set of intervals for initial learning? Or something else? There is surely no single schedule that would work for every type of knowledge. If I learn something difficult (say, writing Chinese characters), I may need to set the initial interval to just a few minutes, as I forget what I initially learnt almost immediately. But if I learn new words in a language that is similar to the ones I already know, the initial interval may be measured in days. – michau Oct 23 '16 at 22:04
  • 1
    Of course quality of knowledge representation, past learner experience biased absolute values. Do you know scientifically approved relative values for effective repetition to mass (move from short term memory to long term) knowledge? I saw that short term memory duration is near 30 sec. So at least one repetition should be done after 2 or greater min to ensure that knowledge moved to long term memory. Also I think that number of repetition during a day (that worth it) should have a specific value. – gavenkoa Oct 24 '16 at 8:26
  • 1
    Another short memory phenomenon is rehearsal. If you keep thinking on something it re-enter short memory for another 10-20 sec to live there. So you should switch focus and repeat knowledge after some time. I found some fragments but can't find complete research on remembering and repetition of data. – gavenkoa Oct 24 '16 at 8:36
4

The question is still very unclear. I looks like you're asking about the value of E-Factor for initial learning. Is that what you want? Is the E-Factor mentioned at the end the same as Coef mentioned at the beginning?

I don't know why you're surprised by the initial factor of 5. In the SM-2 algorithm we have I(1)=1 and I(2)=6, which corresponds to the E-Factor of 6.

But the most important thing is that E-Factor is an approximation of the easiness of an item. As you make more and more repetitions, the value of E-Factor for each items will become more and more proportional to its easiness. But the initial value is a pure guess. How can you expect someone to provide you the right initial value from the start? The right value depends on the easiness of your items, and nobody can know in advance what items you are going to study.

For the same reasons, scientific studies aren't going to be of much help: the goal of SRS algorithms is to personalise the parameters such as E-Factor for each learner and each item, so the parameters personalised for one learner aren't going to be useful for another one.

Another point is that the initial values don't matter too much, unless they are too high. If you feel you're repeating an item too often, all you need is to give the highest grade to your answer, by pressing "Very Easy" or "Bright", or whatever that button is called in your SRS program. The E-Factor will get increased, and it will approach the optimal value after a few repetitions.

So, when you create a new deck, the most important factor that can affect your efficiency is your guess of the initial interval. Obviously, you'll get many more repetitions when you set it to 2 minutes than when you set it to 5 days. But the this value will change from deck to deck, depending on the type of your items, so it's hard to give a fixed recipe. If in you're unsure about your estimation of the initial interval, err on the lower side. If you find yourself pressing "Very Easy" very often, you'll know that next time you create a deck with similar items you need to use slightly higher values.

  • I updated question. I am interested in initial learning within a day and have no any problem with E-Factor when cards repeated between several days/months. – gavenkoa Nov 6 '16 at 21:35
  • @gavenkoa Simply use the standard SuperMemo algorithm. For example, you may try the standard E-Factor of 2.5 and an initial interval of 5 minutes. In Anki Deck Options you can set "Steps (in minutes)" to "5 12 30 75 188 470 1175" (with every new interval about 2.5 times longer). If it's too easy, try increasing the initial interval and/or E-Factor, and calculate the intervals accordingly. If it's too hard, try decreasing them. – michau Nov 6 '16 at 22:16
  • @gavenkoa By the way, you really should have example sentences on your cards. It will help you remember the words and learn how to use them. Definitions are simply not enough. – michau Nov 6 '16 at 22:19
2

Jeffrey Karpicke, Althea Bauernschmidt - Spaced Retrieval. Absolute Spacing Enhances Learning Regardless of Relative Spacing, 2011 shows that retrieval pattern (traditionally expanding and equal/contracting) has no influence on resulted score. With equal number of repetition the longer total period of practice - the better result of retention in long term memory.

Difference between massed studying and studying with spaced repetition show the difference in 200% (experiment subjects recall ~75% of material instead of ~25%).

So for short term learning best strategy is to expand learning period.

Having in mind original question I should spread repetitions across typical learning session time.

With my current AnkiDroid settings I spend near 60 minutes per session (repetition of old cards, learning and repeating 3 times new cards). By keeping same repetition count I should choose 20/20/20 minute delay pattern (which gives an hour). Previously I had 2+5+10 = 17 total minutes...

How many times to recall within a day is still open question...


Another research Jeffrey Karpicke, HenryRoediger - Is expanding retrieval a superior method for learning text materials, 2010 found:

The present experiments showed four important findings.

  • First, there was a testing effect: Taking a single free recall test produced better long-term retention than did reading the text and not recalling it — even when there was no feedback after the test.

  • Second, repeated retrieval practice enhanced retention relative to practicing retrieval once on a single test.

  • Third, rereading the texts after recalling them (and thus receiving feedback after the tests) improved learning and long-term retention.

  • Finally, and most important for the present purposes, whether repeated retrieval attempts were expanding or equally spaced did not matter for long-term retention.

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.