I've seen some videos of people doing vocab cards where they will see the word, think for several seconds trying to remember it, drag the mouse to the review button, and then click. The whole process probably takes 6-10 seconds per card.
How fast should you actually do SRS cards? This is asked occasionally but rarely discussed with any strong definitive answer. I'll just cut to the chase: the people doing it the way I described are doing it wrong, the faster you answer the better.
But, maybe you've read about retrieval effort - don't you retain something better if you try hard to remember it? Isn't active recall better - you learn better actively retrieving something than just passively recognizing it?
Yes, and that's well supported by research. But you are always doing active retrieval when you do SRS flashcards. You see the card, attempt to remember the meaning, and fail or succeed.
The question is whether longer deliberated retrieval is more efficient than just making a quick guess and immediately moving on. We don't have to guess, there's clear stats on this that show a clear answer.
No benefits in retention on longer retrieval time
- Open Spaced Repetition Anki Stats: Across 10,000 collections and 519 million reviews, differences in answer time barely changed later-recall predictions.
- Vaughn, Hausman & Kornell (2017): 5-, 10- and 30-second attempts produced no significant difference two days later.
- McGregor (2006) and Mathews (2010): Seven multiplication-card learners and three time-telling learners showed no consistent retention advantage for either pace.
- Péladeau, Forget & Gagné (2003): In a 168-student flashcard study, speed predicted no difference in retention among 83 students tested 5-6 months later.
So let's think about the stats as time saved: 8 seconds to 3 is 62.5% less time. Let's assume the guy from earlier takes 8 seconds per card, on average. That's a reasonable estimate since some cards take longer and there are some delay periods. Say another guy is really fast does 3 seconds per card. That's around 20 cards per minute vs 7 cards per minute. Both users do 40 minutes per day, one goes through 800 cards while the other goes through 280. It will take user 1 around 3 years to learn as much as user 2 did in 1 year.
So speed is a given. But how much time should you spend in SRS in total? A lot of advice you'll hear is that 20 minutes or 30 minutes a day is fine - perhaps you'll get diminishing returns if you do too much. But let's take a look at our own Interlign stats:
How much time you should spend per card?
First we tracked the association between reviews/day and pass rate
| Reviews that day | Pass rate |
|---|---|
| 1–500 | 88.3% |
| 501–1000 | 88.6% |
| 1001–1500 | 89.8% |
| 1501–2000 | 88.0% |
| 2000+ | 84.8% |
We also tracked whether cards reviewed on high review days had a lower pass rate when reviewed again days later, to see if they were memorized less effectively:
| Reviews on the earlier day | Later pass rate |
|---|---|
| 1–500 | 87.9% |
| 501–1000 | 88.2% |
| 1001–1500 | 87.2% |
| 1501–2000 | 88.0% |
| 2000+ | 87.3% |
So up to 2k reviews a day - no significant drop. We did find diminishing returns well after this point, for people doing really extremely high reviews - 3k or more - but hardly anyone does that much so this is irrelevant for the average person.
Now let's look at the stats again, comparing our two users.
One is the average user, doing 30 minutes a day at 8 cards per minute. They go through 240 reviews per day.
The other is the power user, doing 2 hours a day at 20 cards per minute. This is 2400 reviews per day.
And again - our data finds no strong diminishing returns or meaningful difference in retention between the two. The second user simply moves around 10x faster. It would take the casual user 10 years to cover as much SRS ground as the power user does in one. So, do your anki reps (or Interlign reps).
What we do find is that session length (as opposed to length per day) is associated with retention. In a workplace-training study, participants doing three 20 minute blocks maintained 94.1% over a month, while those that did one whole hour managed 80.5%
Sentence cards or vocab cards?
Depends on if the sentence cards are pre-made or mined. If they're pre-made cards, just forget it and use the vocab cards. For mined sentences vs vocab cards - we're not sure, but here's what the stats say.
Mined sentence cards appear to have a much higher retention rate than isolated vocab. This makes sense, you know the context, the voice, etc. Sentence cards also help reinforce grammar and sentence structure that you don't get while going through vocab cards. The problem is that sentence cards partially rely on your memory of the context, so there can be an issue with actually remembering a word from a sentence card in other contexts. The other problem is that throughput is much slower - obviously you can move way faster with single vocab cards than with whole sentences.
As far as we can tell, the best method seems to be to use mined sentence cards whilst automatically extracting unknown words from the sentences into your vocab deck. Cross-context learning is really good for retention, and this increases the retention of both decks simultaneously.