Standard

Posted by

Posted on

April 8, 2012

Posted under

Tips

Anki and the Region of Proximal Learning

Many people are familiar with Anki’s Study Options for reviews, and tend to fluctuate between them for a variety of reasons. Here I will recommend a particular option based on scientific research.

Background

The Region of Proximal Learning (RPL) model is the perfect accompaniment to the idea of desirable difficulties, where learning methods such as spaced retrieval increase initial difficulty, but in a good way, as they vastly improve long-term retention of studied materials.

DD is process-oriented (e.g. the cognitive processes involved), whereas RPL focuses on the content of materials and the strategies used to approach them. In particular, RPL deals with optimal study time allocation.

According to RPL, in a given study session, such as for an impending exam, the best method, giving superior performance on subsequent tests when study choices are honoured, is for the learner to prioritize the easiest of not-yet-mastered items first. What they study isn’t too easy, nor too difficult. If it’s too easy, it gets boring and discouraging fast, and is less efficient, as there are more urgent targets that remain to be dealt with. If it’s too difficult, it’s frustrating and impels the learner to quit. Instead it needs to be desirably difficult, in a proximal region somewhat resembling Vygotsky’s zone of proximal development.

Learners, to learn the most in the least amount of time, should study the easiest of the items first, then move on to the progressively more difficult items. This applies for any given study session, but in particular for sessions with greater time constraints, perhaps with an exam the next day, the most learnable items are thus taken care of, leaving more time for study of the more difficult items, and barring that, you’ll at least have learned what you could in a limited timeframe. If you focus on the most difficult items first instead, you may fail to learn anything at all.

RPL and DD rely on, and with practice increase, metacognitive awareness and control. For example, your judgment of the rate of learning (jROL) helps determine whether you’re studying items of excessive difficulty, where the rate of learning is too low, causing learners to quit rather than persevere. You can also underlearn items by assuming you know them well enough and removing them from the study schedule too early.

This qualitative reliance on the individual user’s cognition is ameliorated somewhat by the more quantitative SRS (spaced retrieval is now finally recognized as the best ‘study’ method), even as SRS techniques increase metacognitive awareness as part of the self-regulated DD/RPL learning process, such as through the improved diagnostics of balancing confidence against corrective feedback.

Spaced retrieval, the automaticity of the algorithm modulated by self-grading in a mixed qualitative/quantitative procedure, and thusly reviewing cards as they expire in Anki ensures they aren’t too easy or too difficult in a more macro way. But there’s also the micro level of ordering items for any given study session on any given day that you have a group of due cards.

Time constraint optimization noted above applies during such reviews as well as the initial criterial study/test phase of my method for new and failed cards, including during the use of self-imposed timeboxing strictures. To increase motivation and perseverance it becomes important to use the RPL in these study sessions, studying the easiest yet-to-be-learned items before the most difficult.

火
火火火
火火火火火
火火火火火火火
火火火火火火火火火
火火火火火火火火火火火
火火火火火火火火火火火火火
火火火火火火火火火火火火火火火
火火火火火火火火火火火火火火火火火
火火火火火火火火火火火火火火火火火火火
火火火火火火火火火火火火火火火火火火火火火
火火火火火火火火火火火火火火火火火火火火火火火
火火火火火火火火火火火火火火火火火火火火火火火火火
火火火火火火火火火火火火火火火火火火火火火火火火火火火
火火火火火火火火火火火火火火火火火火火火火火火火火火火火火
火火火火火火火火火火火火火火火火火火火火火火火火火火火火火火火
火火火火火火火火火火火火火火火火火火火火火火火火火火火火火火火火火
火火火火火火火火火火火火火火火火火火火火火火火火火火火火火火火火火火火
火火火火火火火火火火火火火火火火火火火火火火火火火火火火火火火火火火火火火

The Technique

This is where we come to a very simple solution for achieving an RPL-like study time allocation in Anki: in Study Options → Display Order, selecting Review cards from largest interval is optimal, as this presents you with the easiest of the due cards, the ones you know best, first. The option‘s ‘positive note’ mentioned by Damien Elmes, Anki’s brilliant author, takes on a new, important weight in light of RPL research.

Update: For Anki 2, see this add-on.

Some prefer selecting the shortest interval, fearful that these cards are the ones most sensitive to forgetting, and planning ahead for an inability to clear their expired cards, they want to take care of these. By planning for this failure, one mistakenly preempts the strength of the RPL method, which increases your motivation, perseverance, clearance rate, and efficacy of study time. Don’t plan for failure, making a habit of it, and possibly making a habit of pushing due cards further back as you take on too many new cards.

But setting this side, research from Rohrer and Pashler (2007) shows that for the spacing effect—which applies to intervals as small as seconds and minutes, hours and days—gaps that are too brief are actually more harmful to retention than gaps that are too long. It’s intuitive to err on the side of reviewing too soon, but the spacing and testing effects, desirable difficulty is counterintuitive, even though it’s more effective. So don’t be tempted to review hardest cards first because you’re afraid that a longer wait is the worst thing that can happen. Plan for success, don’t bite off more than you can chew, stay motivated and persevere by reviewing easiest cards first.

Speaking of this bias towards reviewing too early, you’ll notice the Good button is emphasized in Anki, rather than Easy, with smaller intervals than Easy. For non-mature cards, this is fine, because it reflects the aim of Desirable Difficulty, wanting the card to be just right, not too easy, nor too hard. But once the card is mature, it becomes easier and easier and we want this to continue. We don’t punish the ease of cards at this point by reducing the intervals. The emphasis on the Good button, and fear of reviewing too late rather than too soon might tempt us to continue hitting Good even for easy, mature cards. When we should be acknowledging and promoting the ease instead. So perhaps for mature cards, Anki’s grading interface should acknowledge this.

In the future, perhaps other RPL-like study session options related to Morph Man’s iPlusN and other ease/difficulty ranking factors might be possible.

In my aforelinked tutorial for new and failed cards, I have the recommended Display Order for cards as … order due; this is mostly moot, as the cards are repeatedly reset, but for the sake of consistency it might be marginally useful to keep it that way during that first 24 hours, as at one point within the phase you no longer reset the cards. The interval differences won’t be significant enough for Anki to quantify an effective RPL-like ordering.

Therefore it’s also worth noting, as there is no automated schedule at that point or with its application to failed card restudy, with the cards reset, the ease/difficulty judgment will be subjective, though still useful, particularly with more complex non-vocabulary cards, and of course it’s a skill for outside Anki as well.

I will reiterate a simplified version of that new/failed tutorial here: First you study a set of cards, focusing on elaborative encoding, then you test (retrieve) them minutes later. What you failed you restudy and test again minutes later, repeating the cycle till all cards have reached criterion level and been passed once. You then repeat the study/test-to-criterion cycle twice more, at intervals a few hours then slightly less than a day later (so all cycles are complete within ~24 hours). This is done with new and failed cards, though with subsequent failed cards during the course of later reviews (without manual rescheduling, where Anki takes care of resetting intricacies) only the initial encode/retrieve-to-criterion cycle is needed; no need to fuss with the other two cycles, because as found in psychology and neuroscience, relearning materials by then tends to be easier and accelerated.

I have Show new cards in random order selected, in order to aid the clearing of immediate memory (the latter through waiting minutes between study/testing) by preventing the learner from memorizing the order of cards. However, if using a different technique for new cards and using Morph Man, the Show new cards in order added option might achieve a quantitative RPL-like effect because Morph Man seems to attempt to modify creation dates based on its own standards for estimating difficulty.

Interleaving

While we’re on the topic of study time allocation and timeboxing, don’t forget the benefits of task-switching and interleaving (contextual interference) as motivation for such session-based strategies. For both shifting cognitive load to other areas of the brain with different types of tasks, and/or introducing desirable difficulty by forcing the learner to discriminate between them. Keep in mind we want the components to be complementary in some way, both systemically and cognitively, taking advantage of multimodality and ensuring that tasks, in the case of interleaving, aren’t too different and thus too easy to discriminate between: ideal possibilities include using various senses and language goals for different decks and card types, switching from history to psychology to statistics as study topics, shuffling different types of math problems, et cetera.

Both inter-deck and intra-deck possibilities exist, when speaking of interleaving in Anki, though the shift in types from card to card might problematize the RPL-like ordering by difficulty using maturity as a factor (long to short intervals) as content will vary. But as long as you’re attending to the above moderation of relatedness, it shouldn’t be a real problem. It’s also worth noting that attempting to create contextual interference for target items while spacing their study can help, but can also hinder learning if done too much: specifically, varying a card’s formatting between intersession intervals (ISI) to the point that retrieval fails.

References