The author of this article mentions he'll post an article tackling this issue, looking forward to that!
I later wrote my own flashcard app[1] with spaced repetition and ended up hiding the details of what was overdue and instead just let you set up a daily goal of number of cards reviewed. When you do a lesson, it picks up the overdue cards first, but if there aren't any, it pulls in new cards instead.
It's funny because I originally thought telling you how many were due each day would be a great motivator (at least for me), but I'm finding I actually don't care. If you miss a few days, the cards still remain overdue, but when you return it doesn't feel bad.
I think for me, it's more important when learning to aim for the long-term commitment to it than to be perfect each day in your studies. If you aim for perfection, you end up not meeting your high standards and may end up quitting. Also, I think the "deadline" for when to review in spaced repetition isn't exactly accurate to the day or hour that something is due, so there's a little bit of malleability there, and I think that's okay.
If there's enough interest, I might try to port it to Android next year as a fun project.
Those are features of AnkiDroid, and probably the desktop Anki also.
There is a daily review limit, and you can configure to see review cards before new cards. Anki does not tell you how much is overdue beyond the review limit.
If you're set on writing your own app, you will tend to overlook the configuration details of the original, though.
I think Anki is powerful and has a lot of configuration settings and add-ons to support new scenarios, but I think for a lot of average non-technical users, they may be overwhelmed by that and not interested in learning it. My goal was to write an app for those people (and for me).
I found with Anki that it was sometimes difficult to self report precisely how well I remembered something. It also added an extra cognitive step on each card to have to consider it, which over time was an annoyance.
I don't have any scientific proof that judging response time is better than self-reporting, but my thinking is self-reporting accuracy is low enough and forgetting curve is approximate enough for each card that in aggregate it doesn't make a huge difference. It would be interesting to have real scientific tests, but I think it would be difficult to set up.
I considered adding a way to self-report your score, but it didn't make the MVP and I'm really trying to target less technical users and not people who are already familiar with Anki. I might add it one day as an option since the underlying design of app can handle it.
My strategy for this was to dump the entire text of some good literature into Anki decks that give me one new paragraph per day. The steady drip of unseen material removes the stigma of pressing the “again” button and gives me a small reward of story progression from each session.
In a separate deck, I use cloze deletion on sheet music to help learn songs. That deck only gets reviewed by actively playing music, which is an inherently enjoyable activity on its own.
Brilliant strategy.
This is still a fairly new strategy for me, so I don’t yet know how many parallel threads I can keep straight at the same time— So far, it doesn’t feel like I’m anywhere near the limit.
This idea was actually born to combat review starvation: I haven’t been very active at adding new cards lately, and there was a risk that my daily review count would hit 0 for long enough that I’d stop checking regularly— I needed a source for lots of interesting but low-priority cards to keep the pump primed.
The original paper on cloze deletion [1] uses them as a readability measure: Readers are given an unfamiliar text with blanks and are asked to guess the omitted words; the percentage correct is then a measure of the text’s quality rather than the reader’s knowledge.
Instead of a knowledge quiz, which is how clozes are usually treated in the SRS world, this is an automated reading comprehension test— Just the thing for capturing the intangible benefits of reading literature. In theory, as you become more familiar with the book’s style and subject matter, you should be able to pass the first review of an unseen passage most of the time.
I settled on 25 lines of text per card with one omitted word about 2/3 through the passage. Successive cards contain some duplicated lines (5) to provide a sense of continuity between the cards, which are presented in a disjoint manner.
I import each book into its own deck set to show 1 new card per day (in the order added)- I want whatever mental connections are necessary to understand the next passage to end up in long-term rather than short-term memory. The reviews give me an opportunity to spot details that seemed unimportant on a first reading but that foreshadow something that happens later.
Beyond that, it’s driven by Anki’s normal scheduling algorithm; these are subdecks of my general-review deck, so Anki will autumatically mix the new cards and reviews with any other reviews I have due.
Most of the books I’ve imported are 300-400 cards, so it’ll take about a year to work through each of them. I wouldn’t be surprised if I end up doubling the number of books I’m reading at once, which would bring the average to around 1/month.
[1] https://www.gwern.net/docs/psychology/writing/1953-taylor.pd...
I did the first one manually. In addition to being a lot of work, there were major spoiler hazards— If I wasn’t already familiar with the general shape of the plot, the process would have revealed secrets too early.
The script will ask you for a text file to process and the pop up a Tkinter window with a single button. Each time you press the button, it copies the next card's body text to the clipboard, ready to be pasted into Anki's Add window.
The text file needs to be already split into reasonable-length lines, like the ones you get from gutenberg.org
Card template: https://pastebin.com/BFJxW1jC
Cloze splitting script: https://pastebin.com/4Jek0G7u
Because paragraphs are so variable, my Project Gutenberg import script makes 25-line cards with a 5-line overlap between adjacent cards (for context). It also blanks out a random word on the 15th line as a cloze deletion test, to ensure I have some understanding of what’s going on.
Maybe you only review a few of those hundreds of cards that are due. So what? The algorithm still achieves its purpose. You are still focused on cards you are least likely to have permanently learned. No matter how you use or abuse space repetition algorithms they ensure your study time is well spent (assuming flashcards are the right thing to be studying at all).
I learned this from experience. I have some math concepts I'd like to have memorized. I made some flashcards in a space repetition program. I haven't review them for about 6 months. It is what it is -- at least I know all my progress wont be gone when I finally do return to those flashcards. I will have forgotten some cards, those cards will be shown to me frequently. I will have remembered some cards, those cards will get a boost because I've remembered them for 6-months without review, and will rarely be shown to me. The spaced repetition algorithms still work; they are robust.
I will update the post with screenshots or something tomorrow, for credibility.
1. Never set a maximum number of reviews. It will create an unstoppable backlog no matter what you do.
2. If you have a huge backlog of like 500 cards then just hit "HARD" on 50% of them until the backlog is spread out over multiple days and you have a manageable number reviews.
3. Don't have too many young cards at once. Stop adding new cards after you have 500 young cards (exact number depends on how easy the cards are).
I've been able to use it about 30-45 minutes per day, every single day for five years. Even when Im sick or traveling. If I know I am going to be busy, I just turn down the number of new cards in advance and by the time I'm busy my review count has usually responded.
(I don't use the desktop Anki, by the way; I used it briefly for doing some editing on decks.)
Tha said, AnkiDroid could use a vacation feature. How vacation mode could work would be simply by rescheduling all cards, delaying them by the vacation amount (14 days or whatever). Thus, no cards are due for the next 14 days, and on day 15, the same cards are due that would have been due day 1 (no "snowplow" accumulation). It would probably help if the app generated a notification a day before vacation ends.
It's incredible that neither the author of the original Anki, nor the authors of the AnkiDroid clone, recognize the value of a vacation.
But the vacation idea is great, I'm sure someone could make an add-on to implement it.
A simple toggle: "pause Anki". Here are my requirements:
- in the paused state, the review UI is entirely disabled. You cannot review anything.
- when you toggle out of the paused state, Anki calculates the number of days since the pause, and delays all cards in all decks by that many days.
- pausing and resuming on the same day has no effect since the days delta is zero; the UI just becomes enabled.
- For the purposes of the delta calculation, a day is the study day (e.g. 7 a.m to 7 a.m), not the midnight-to-midnight calendar day. If you pause at 11:45 p.m. and resume at 1:15 a.m., that's a no-op since that's the same study day.
- since resuming is potentially a time-consuming operation that destructively manipulates the database, unpausing comes with some yes/cancel prompt, except in the no-op case.
I've started to think of it like node: A slim core with an expansive userland in the form of add-ons.
It happens. Every time it was a slog but I just got through it.
The one thing I can say for certain is given enough time it will happen. But whether you can overcome it or it causes you to quit is down to how intrinsically motivated you are (Anki or no Anki) to achieve the goal you are using Anki to help you accomplish. If you would continue to pursue the goal even without Anki then you will most likely persevere. If the goal is simply to remember everything and not forget stuff and you're only using Anki to use Anki then you will almost certainly quit when you hit this wall.
>I feel like Anki is not designed to help you succeed. Eventually, you'll start missing your reviews. One day you'll open Anki, see that you have hundreds of cards scheduled, and just quit the app.
You could also just do it every day, if it's important to you. You're not destined for failure, it's a choice you make.
You will forget stuff over time. What was the leading actor's name in that movie? Aaargh, it was practically a household word in the early 90's ....
> You could also just do it every day.
Even people who stick with it 7 days a week for 50 weeks could use a 2 week vacation.
>Even people who stick with it 7 days a week for 50 weeks could use a 2 week vacation.
Then you'll need the discipline to do ~10x your normal daily volume when you get back (not doing new cards ever day makes it not a clean 14x).
>That is not a given. Just several days ago I lapsed on a 4.5 year card. The interval went down to 2.2 years or so; hopefully I will get it that time around.
The default settings set it to 10% of the interval after an again, so you must have messed with that. You can also configure that to be 0% if it bothers you.
I've seen curious comments like this in various past forum discussions on the topic of Anki vacation ideas.
A vacation is not a period of rest followed by double the amount of work to catch up.
A vacation is a pause in work, which delays all subsequent work by that much time.
> not doing new cards ever day makes it not a clean 14x
You can't do new cards everyday; eventually you will have seen all new cards of a deck. I'm still working decks whose new cards ran out years ago. The presence and scheduling of new cards is a temporary condition with little long-term significance.
My new cards comment was to explain why it's not 14x when you go on vacation.
>I've seen curious comments like this in various past forum discussions on the topic of Anki vacation ideas.
>A vacation is not a period of rest followed by double the amount of work to catch up.
>A vacation is a pause in work, which delays all subsequent work by that much time.
This is simply cheating the SRS. Some SRS programs do allow you to do this, wanikani does it for instance, but it is quite literally cheating. Instead of seeing it after 30 days, you see it after 44, but the program pretends it's only been 30. It's not a good feature, besides to make people feel better about themselves.
>I've seen curious comments like this in various past forum discussions on the topic of Anki vacation ideas.
It's cheating. And in this case, the only person you are cheating, is yourself.
One particular comment I remember was about the Anki feature of reviewing ahead. The poster said that you shouldn't do it because it will "mess up your stats". But, think about how idiotic that is. According to that reasoning, you should not read any native language material. Because if you read, you will encounter some of the words which are scheduled in your Anki, and recall them prematurely. And that will cause your Anki progress to deviate from what the SRS algorithm predicts, making it seem like you're doing better. Hence, reading native material is cheating: sticking to the algorithm is the goal, not learning the language!
SRS is just a tool; it's not a master to be served, but the servant. It's based on soft science, and is self-correcting. Any user behavior which leads to increased forgetting will result in more lapses, and an increased workload to make up for it.
Mature grown-ups with busy lives adopt the tool to best suit them, rather than to adopt themselves to serve the tool.
If the only way you can use the tool is less optimal, so that learning takes longer, well maybe that's the best it can be, in relation to everything else you have going on in your life!
The SRS implementation in Anki has a large number of parameters. The defaults are poor and don't work well for people. Good parameters are a matter of opinion. Someone taking a vacation according to a regular pattern, like one week every four months, is effectively just tuning another parameter.
You're making a mistake of inference. Someone getting a card right after 44 days, but getting only 30 days of "credit" for it in he program does not have a reason to feel good, other than to feel good about remembering the card after 44 days. You raise a good point there though. In fact, a correct implementation of vacation mode should not literally just make time stop; it should somehow give the user credit for recalling cards beyond their scheduled interval. The recalled cards should perhaps get a bit of a boost in their next interval.
The point of a vacation feature is simply to feel rested, not to feel gleeful about somehow "cheating". If someone misuses the feature to take a one week vacation after every week of studying, that's their problem, and not a reason why a reasonable user shouldn't be able to take a break.
It's just a tool, and if you miss some lessons, it's not like those cards immediately vanish from your head. If you fall behind because you weren't disciplined, no big deal, just pick it up again and carry on. Learning is a marathon, not a sprint.
For me, SRS algorithms try to model how memory works, it's not an exact match to it, so even missing a card deadline by a day or two or even a week isn't the end of the world.
I suspect that the reason is that language-learning forums are full of children.
Jesus christ, calling someone a child because they disagree about adding a cheat vacation mode to an SRS program? You gotta take a step back and look at your life.