When to remove a word from Anki

Question

Anyone familiar with Anki, or Spaced repetition in general, should know that the algorithm assumes that you will never remove a word from the system. You'll always review learned words, but after a while, reviews will be years apart.

At this point, I've been using Anki long enough, that my longest review interval is 3.7 years, and a large number are 2 years or longer.

According to SRS literature, these infrequent reviews are still helpful, as they ensure you don't forget these words.

But my question is: Is this really necessary when learning a language?

Languages are learned to be used, so if I learn a word sufficiently well that Anki waits 3.7 years to show me again, I'll either be using that word throughout the review interval, while using the language, or it's such an obscure word that remembering it 3.7 years later likely isn't that important.

Is there any cognitive/memory evidence that a review delay above a certain threshold is generally not useful? Further, is there any evidence from others who have used Anki, or similar systems, for multiple years, that some personal experience can be shed on the situation? When have you found that further reviews were no longer useful?

Answer 1

You start with a false premise. An SRS like Anki does remove words. It removes words when they are too hard to learn by marking them as leeches. The default condition for a leech is answering a card wrong at eight separate days.

The basic premise is that the amount of time spent on a card that gets correctly answered is tiny.

Let's take an example. I'm a native German speaker and I have an Anki card for the English word veal. The card spent some time in a deck that I paused reviewing. I think before Anki 2 there was no review history, so you don't see what the card did before that point.

Date                 Type   Rating  Interval    Ease    Time
2016-04-14 @ 10:42   Review 3       4.5y        206%    1.9s
2013-09-15 @ 23:18   Review 3       1.6y        206%    2.0s
2012-05-05 @ 18:39   Review 3       2.7mo       206%    2.5s
2012-03-21 @ 23:30   Review 3       1.0mo       206%    4.3s
2012-03-07 @ 20:12   Review 3       13d         206%    2.2s

If I had found the card too trivial at any point I would have pressed 4. I happen to use most of my English online and don't often come into contact with the word veal.

The next card I looked at is tin. It has the following review history:

Date                Type    Rating  Interval    Ease    Time
2016-05-08 @ 11:46  Review  3       1.2mo       260%    1.8s
2016-04-23 @ 08:38  Review  3       14d         260%    2.4s
2016-04-17 @ 10:52  Review  4       6d          260%    2.3s
2016-04-15 @ 08:35  Review  2       2d          245%    1.3s
2016-04-14 @ 10:44  Relearn 2       1d          260%    2.7s
2016-04-14 @ 10:34  Review  1       10m         260%    3.4s
2013-01-30 @ 14:57  Review  3       2.7y        280%    2.3s

I think I didn't have a problem at any time to recognize tin as meaning Zinn while reading a text but I still forgot the direct translation. While I likely did come into contact with the word while reading, I think it was still useful to learn the word again.

The amount of time I spent reviewing either of those is tiny. 16.0s for tin and 22.0s for veal (in both cases not counting reviews before 2012 because of data availability issues).

Let's look at a card on which I spent 69 seconds, i.e. daisy:

Date                Type    Rating  Interval    Ease    Time
2016-04-14 @ 10:18  Review  3       2.3y        130%    2.3s
2013-04-12 @ 17:36  Review  3       7.2mo       130%    3.2s
2012-05-31 @ 00:18  Review  3       16d         130%    5.8s
2012-05-17 @ 17:40  Review  3       12d         130%    3.3s
2012-05-05 @ 17:28  Review  3       5d          130%    3.3s
2012-05-01 @ 13:54  Review  3       4d          130%    4.3s
2012-04-30 @ 15:29  Relearn 2       1d          130%    3.6s
2012-04-30 @ 13:52  Review  1       0d          130%    3.5s
2012-04-16 @ 14:19  Review  3       4d          130%    2.5s
2012-04-15 @ 20:33  Relearn 2       1d          130%    2.5s
2012-04-15 @ 20:25  Review  1       0d          130%    3.2s
2012-04-06 @ 20:18  Review  3       7d          130%    3.5s
2012-03-31 @ 16:19  Review  3       5d          130%    2.8s
2012-03-27 @ 16:46  Review  3       4d          130%    2.6s
2012-03-26 @ 14:29  Relearn 2       1d          130%    2.1s
2012-03-26 @ 14:25  Review  1       0d          130%    3.8s
2012-03-16 @ 11:52  Review  3       9d          130%    6.3s
2012-03-08 @ 23:46  Review  3       7d          130%    4.5s
2012-03-02 @ 12:24  Review  3       4d          130%    3.6s
2012-03-01 @ 20:09  Relearn 2       1d          130%    2.8s

This card needed a lot of effort but I wouldn't have saved much effort by simply blocking reviews over two years. I don't have any problem seeing the card for daisy again in two years. For the purpose of writing this post I only looked at the statistics of those three cards.

Answer 2

It depends on your goal. After 10 years you may not need those words altogether.

If you have 20000 cards for review with intervals of 3 years and greater and quickly review them it takes:

20000 * 10 sec / 3600 sec = 55.5 hours

Do you mind spending 50 hours over a 3-5 year period just to review well-known cards?