Other than that, I use globally the same strategy as you do. I coincidentally also started using marking to kill leeches once I’m back home, and it has proven to be very efficient.

Excellent point again. I have mentioned this in other articles about SRS, but perhaps I should have mentioned it here as well. I spend quite a lot of time with Anki, but I spend much, much more time listening to the radio, watching various broadcasts and reading books. This is also review. More importantly, over time, exposure to audio and text will generate knowledge about how words are used, something I think is tricky to incorporate into SRS practice (although of course it can be done with sentences and so on).

Interesting comment! I have a couple of things to say. First, the numbers you give for your experiment are roughly similar to what they would be in Anki and I did some calculations prior to writing this article and ended up with the same figures, but with a completely different conclusion. Decreasing the review interval with 38% (using your numbers), which means that you reduce review time by the same amount. For all cards in a year of studying 40 minutes per day, this would save 244 hours. Naturally, this isn’t relevant for all cards. I mentioned 10 hours, which is because the difference was less than 38% in my case and I only assumed that it was relevant for 5% of the cards, which ends up close to 10 hours. I think this is quite a lot of time. In ten hours, I could learn several hundred characters. Perhaps our perspectives are different.

Second, I don’t think you’re observation about better recall is weird or that your approach is unorthodox. What you’re doing is aiming for more than 90%, which is something I wouldn’t do, but which is perfectly fine to do if you know what you’re doing. Remember that the goal with SRS isn’t to learn all the words, it’s to learn the words as efficiently as possible given a 90% recall rate. So, it’s fairly obvious that you will get better results from reviewing more often (I haven’t heard or read about what you say about memory being reinforced when trying to recall something which is close to being forgotten, do you have a reference for this?). The question isn’t if it works or not, the question is if it’s worth the extra time.

Finally, I think it’s great that you conduct these experiments! This is what everyone should be doing. I’d be delighted to read about other experiments you’ve tried.

SRS should not be the only time you review vocabulary/grammar/whatever. In my opinion, it should not even be the main way to review vocab etc.

The main way I review language is actually reading/listening to stuff in Mandarin. Everytime I understand a word in context, it’s a review. I consider SRS to merely be a support. If I relied on reading/listening alone for review, then there would be words where the intervals would be to long between encounters. I use SRS to fill most of this gap (as you say in the article, trying to completely fill in the gap is inefficient), and to me, SRS is just that – a fill-in. That might be why I found the press-2-only method to be inefficient – I am actually reviewing most words far more often than they appear in my SRS, so increasing the frequency of SRS reviews for words which are not actually hard/rare is excessively redundant.

I actually have tried this before, and for me, it is definitely not the most efficient method. While I was able to keep my queues to zero, it took a lot of time. I can get the almost the same results (as far as remembering stuff) by choosing the interval based on difficulty as Olle describes in this post with a lot less time spend on reviews.

Recently, I tried an experiment by creating a separate category of cards within my deck. The only difference with the other cards is that I would systematically rate my response a “2” (out of 5), regardless of the strength of my recollection. With Mnemosyne, a “1” will almost make you start anew with a card, while a “2” will simply reduce the interval before the card comes up again.

With Mnemosyne’s algorithm at least, I have found that the interval between reviews is not dramatically affected by choosing “2” instead of say “4”. I am making up an example here, but rating a card “2” might make it show up say 21 days later, whereas a “4” would give you a 34 day interval. With each review, regardless of the rating, the individual intervals get longer and longer – but cards rated 2 are simply scheduled to come back earlier.

Now, I know the theory behind SRSing is that for maximum reinforcement, you should only review a card when you are on the cusp of forgetting it – this is not only to reduce the total number of reps over time, but also because the extra effort involved in successfully retrieving an answer from your memory will have a reinforcing effect. (This is my understanding of it anyway.)

However, I have personally found that my special category of “always-rate-them-2” cards have impressed themselves on my memory more thoroughly. The added repetition is not so much that they ever get boring or crop up too often for my taste ; and the whole process is not more time-consuming because my recall is quicker and more effortless. I can go through a deck faster, and tend to avoid my daily reviews less often (not at all in fact).

So, although this is unorthodox, I have decided to use this same method with all my cards. As I mentioned before, this does not have the effect of freezing my reviews in time, since the intervals do get longer – it just happens more gradually.

Perhaps I am an anomaly, but I would suggest trying “systematic two-ing” as an experiment to those who are intrigued by the idea.

Excellent point, Sara. I do this as well and can think of some situations:

1) Characer parts (i.e. learning how to pronounce some radicals is useless, I hit “hard” rather than “again)
2) Classical Chinese (I don’t have much of this, but there is still some; I don’t need to read this aloud)
3) Very formal vocabulary (your example, but the problem is that it’s hard to know what’s formal enough, so to speak)
4) Very rare vocabulary (enough to just understand, perhaps similar as the previous situation, although rare and formal aren’t necessarily the same)

I have some grammar patterns in there as well as well as some phrases I find interesting or difficult. I also have a separate deck for memorising 道德經. I don’t have a deck based on sentences and cloze deletions, though. The reason isn’t that I disapprove of other methods, indeed I think using sentences might work very well (I’m trying it out with 成語), but there is simply no way I can turn my current deck into sentences, much less MCDs, ajatt style. I have around 21 000 cards, so any systematic change would be impossible.

I think the main advantage of having just words is that it’s much, much quicker, both when reviewing and when adding (finding the right sentences is sometimes very hard, especially if you don’t know the key words you’re supposed to learn. How do you know the sentence you’re using is good? I prefer to rely on massive volumes of reading and listening to pick up context. This is perhaps not ideal, but it seems to work pretty well.

What I could do is try using sentences from now on with new words I learn. That would enable me to form an opinion about what it’s like. That won’t mean I change my main method to this, but that’s mostly a question of momentum. In short, I think all methods mentioned are sound, at least in theory. The reason I haven’t written about this in more detail is because I haven’t formed an opinion about the question yet!