Optimizing Anki Settings for Japanese Study

Default anki settings are ok for learning general material, but sub-optimal for most people learning Japanese.  Many people are put off by the myriad of options in anki.  This post suggests some anki settings which should make things a lot easier and goes on explain how the settings work.  If you are studying another subject, scroll down a bit where I explain a little anki theory.

Anki Settings Quick Start

If you want the tldr and just want to plug in some settings that should work better for you, here you go.  Start with anki’s defaults and change these settings:

  • New Cards – Steps:  1 10 60 360
  • New Cards – Starting Ease: 180
  • Lapses – Steps: 5 75 360
  • Lapses – New Interval: 15

Theory

The settings above show cards to you more often than the default so that you fail cards less often.  You might be tempted to adjust the settings even more so that your accuracy is doubly improved, but there is a point of diminishing returns where more frequent study yields less additional knowledge acquisition.

Anki Settings resulting in 70-80% accuracy are most efficient

70-80% is most efficient with efficiency falling off rapidly outside this range.

The efficiency sweet spot seems to be in the 70-80% accuracy range.  The efficiency curve is roughly parabolic, so the difference between 70% or 80% is negligible, but setting your anki settings to attain 90% accuracy you would need to study the same words roughly twice as many times which is clearly inefficient.  85% is just a bit less efficient than 80%.

On the other hand, 70% accuracy might be efficient, but forgetting 30% of the words you study every day can be soul crushing.  Even 80% accuracy is dispiriting some days, and remember that this is average, so some days will be better and some days will be worse.  This is really up to you, but personally I target accuracy around 80-85% because I am willing to trade some efficiency to make studying more enjoyable.  But you should target a specific accuracy range to aid in adjusting your anki settings.

Understanding Anki’s Settings

The anki settings we are discussing today affect three different groups of cards.  The ideal settings would have your accuracy for each of these groups of cards falling in the range you’ve decided work best for your goals.

  • Learning cards are brand new cards that you just started learning.  These have never had an interval over 1 day.
  • Young and Mature cards that you’ve been reviewing for a while.  These have an interval over 1 day.
  • Lapses (aka relearn) are young or mature cards that you just failed.  These had an interval over 1 day, but are treated like a learning card again.

Learning Cards Settings

The setting to adjust learning cards is ‘Steps‘ in the new cards section.  Steps is simply a list of how many minutes to wait until you see the same card again with a space between each step.  E.g. if you go with the default of 1 10, anki will show you a brand new card and then show it to you again 10 minutes later.  If you fail, you go back to the beginning of the list.  If you answer correctly twice in a row, the card will no longer be considered new, will become young and you will see it again the next day.  I don’t know about you, but studying a foreign language word twice in 10 minutes usually isn’t sufficient for me to remember it the next day.  I need to see it a few more times.  So if you use steps of 1 10 60 360, you will see it after 1 minute, 10 minutes, one hour, and 6 hours before it becomes young and the next interval is 1 day.

The goal is to get your learning accuracy within the range you set. Take a look your anki graph called ‘Answer Buttons’ and if your learning percentage is lower than desired, you may need to add another step – something like 1 6 36 180 560.  Or if your learning accuracy is high, consider 1 10 100.

It is important to know that anki’s graph combines the learn and relearn stats.  So it is best to install this anki add-on which separates the two stats.

Young and Mature Card Settings

The setting which affects young and mature accuracy is ‘Starting Ease‘.  A card’s ease is the percentage by which the interval is increased for each successive step.  Anki’s default stating ease is 250, so if you pass a card with an interval of 10 days, anki will multiply 10 days by 250% so the new interval will become 25 days.  This setting will vary widely by person, but I find a starting ease of 250 to be a bit too high for me to remember 80% of my mature cards so I set my starting ease around 180.

Whatever starting ease you set, anki will adjust each card’s ease over time to account for the difficulty of each card.  Anki decreases each card’s ease factor every time you press ‘again’ or ‘hard’.  Anki increases each card’s ease factor every time you press easy for that card.  But ideally you set an appropriate starting ease so that you aren’t studying cards at sub-optimal intervals while anki finds it’s level.  So keep an eye on your mature accuracy and adjust accordingly.  Keep in mind that mature accuracy will take a month or more to reflect changes because the intervals are 21 days and greater.

Lapses

When you fail a card, the card will go through it’s learning steps (default 1 10) and it’s next step will be one day, just like for new cards.  But this behavior is adjustable separately from new cards.  Usually if you fail a card that has an interval of several months or years, after a relearn step or two, you will not need to reset it’s interval back to 1 day.  You can set a percentage of the old interval in ‘Lapses’ / ‘New Interval‘.  A setting of 10-15 is appropriate here.  So if you use 15% and your current interval is 6 months (180 days), the new interval will be 27 days after going through the relearning steps.

Lapses / Steps defaults to 1 10, but I prefer 5 75 360.  Again this should be adjusted based on your relearn accuracy and personal preference.

Recap

The settings at the top of the page are a decent starting point.  But to use your time most efficiently you should adjust them so that your learn, relearn, young and mature accuracy is within your target range.  The settings will vary considerably from person to person and for different types of cards.  For instance, my kanji cards need very different settings than my vocab cards and my grammar cards are different as well.  If you have any questions please leave a comment below and I will do my best to provide a thoughtful answer.

Revised Lesson Plan for Q4 2016

My last lesson plan was a complete fail due to RTK reviews taking longer than anticipated.

Long story short, about a month after finishing RTK I noticed my reviews were not going down and my accuracy was poor and not improving.  I saw no progress whatsoever for over 30days.  I made the decision to drastically crank my anki settings to study my cards more often and improve my accuracy.  Slowly over two months, my accuracy went from low 70s to high 80s, sometimes peeking into 90s but if my workload did anything it increased.

Virtually no progress but this should change soon.

Virtually no progress but this is finally starting to change.

I turned the settings back to normal a little over a month ago and workload has gone down substantially while maintaining accuracy. I should have normalized my settings much sooner so I could have a lower workload sooner, but it sure is nice to have accuracy still peeking into the 90s some days. Overall, my RTK workload was over 400 cards for several months, now it is hovering around 250 and dropping.  Since I made the change a little over a month ago, I’ve managed to increase mature cards from 500 to 750 which is amazing.  In another 45 days I expect to have 50% of my RTK cards mature and a workload in the mid 100’s which will make more time for studying other things.

Moving on

I plan to do roughly my old lesson plan set back by several months.  I’m still working on the plan so things may change. I’m still not not sure how I want to prioritize everything. One thing I’m considering adding is what I’m calling Easy Vocabulary™ because I believe I can make a list of words that I can learn much quicker than a normal vocabulary list.  Basically easy vocabulary are words whose constituent kanji give a hint to both the word’s reading and it’s meaning. I will write more about Easy Vocabulary at another time because there is more to it than this. I’ve added the WIP plan below.

Exercise Cards* Cumulative Cards* Finish by*
jpod 101 vocab 90 90 oct 23done oct 26
jpod 101 dialogs 119 209 dec 2
shirokuma vocab 500 709 jan 21
Rtk lookalikes 85 794 TBD
shirokuma manga TBD
Easy Vocabulary™ TBD

* card counts could be WAY off which will obviously effect the schedule.

Lesson Plan for the 2nd half of 2016

Update:  This plan was a total fail.  RTK reviews took WAY longer than than I anticipated and crowded out everything else.  I know RTK isn’t even on this list but it was the lingering reviews that refused to go down for several months.  Good news is they are going down now and I resumed  jpod101 vocab last week.  I’m working on a slightly revised lesson plan now.  More details to follow.

I’ve revised my lesson plan slightly for the rest of the year. The notable addition is NHKeasy vocab since I’ve started reading one article per weekday. I’ve decided to prioritize learning things that directly help me consume media that I can fit into the gaps in my day. NHKeasy I can read while at work, and jpod101 I can listen while driving. I kept perfect groups and anime on the list with a question mark. If I have time, I will consider adding them. It’s looking like N3 is a distant dream this year. But I haven’t completely written it off.

Exercise Cards* Cumulative Cards* Finish by*
jpod101 dialogs listening - ongoing
NHKeasy vocab ~400 ~400 Aug. 12 fail
NHKeasy articles read 1 each weekday - ongoing fail
jpod101 beginner s1 vocabulary ~400 ~800 Oct. 8 200 to go
?perfect groups ~400 ~1200 Dec. 3 fail
?anime vocabulary + watch ~400 ~1600 Jan. 30 fail

* card counts could be WAY off which will obviously effect the schedule.

Clearing my review backlog

I finally got my reviews down to zero last night.  It took 21 days to catch up after missing the better part of 11 days of reviews.  All of the red in the chart below shows how much skipping a few days affected my accuracy.  The first few days of catchup in particular saw my accuracy down in the low 70’s.  The low accuracy combined with a wall of 2k reviews didn’t do much for morale I can tell you that.  Screen Shot 2016-06-15 at 2.30.29 PM
The good news is that I’m caught up now.  I have a few more days until the reviews truly settle down to the same general level they were before this madness.  I’m actually hoping they settle down below the original level since I haven’t added any new cards in the last 20 days!  Ideally I will be able to add a few extra cards than normal to catch back up a little.  Unfortunately I won’t be able to match the pace I set for myself at the beginning of the year, but I’ll get close.

I’ve also started reading one NHKeasy article each weekday with much help from the new yomichan chrome extension.  Once I start adding cards again, I will probably add some of the more frequent NHKeasy vocab sourced from this site which has aggregated a frequency list of words culled from from a few years of NHKeasy articles.

2016 Lesson Plan – Progress Report

I was going pretty well on my lesson plan until around mid-April and things started getting progressively off track.

First, I realized that my estimate for jpod101 beginner vocabulary was about half of the amount I needed.  I initially budgeted 400, but the real number is likely closer to 800 once I weed out all of the words I already know.  The complete list was around 1600 as I recall, but I seem to already know half of the words.  It’s a little bit discouraging to learn that I still don’t know half of the beginner vocabulary list.

Then I started noticing that my rtk workload wasn’t going down over time despite getting decent and increasing accuracy above 80%.  I noticed my pile of mature kanji wasn’t increasing and my mature accuracy was in the low 70s, so I’m guessing I was failing enough mature cards to offset the effect of increasing intervals on younger cards.  I’ve recently decided to change my settings on my rtk deck to get my mature accuracy much higher, but that it going to slow my progress as I’m not willing to increase my study time to compensate.

The worst thing to happen to my progress however was that I got extremely busy at work and wasn’t able to finish many reviews for a little over a week.  I ended up with about 2k reviews to work though and more reviews becoming due each day.  I’ll get through them in a few weeks and I might even spam correct answers on my most mature deck.  However, catching up with the younger cards will end up putting me behind my already behind schedule by about a month.

I’ve also decided to tweek the lesson plan a bit. I’m still planning on finishing the jpod101 vocabulary so that I have something comprehensible to listen to in the car.  That will effectively add time that I am able to do japanese things that I didn’t have before.  After that, I will probably insert some reading and reading related vocabulary study.  I’m not really sure what I’ll be reading, but maybe it will be nhk easy or some easy manga.  If I can find some short things that I can read during downtime at work, I can add a little more study time that I didn’t have before.  That also fits into my desire to focus on consuming as much actual Japanese media as possible.

All this means that it is all the more unlikely that I will be studying for the N3 this year.  I will probably take a few practice tests towards the test time and decide if I should take N3 or maybe N4 if N3 looks unlikely.

Updated Lesson plan for 2016

I’ve been getting more serious about making intelligent decisions on how I spend my study time.  To that end I’ve decided to start writing down my tentative lesson plan as a way of formalizing my plans which will hopefully lead to better decisions.

I’ve pretty much given up on the idea of proceeding further with core.  Instead, I am going to be focusing on learning only the words used in a particular work and then reading, watching, or listening to the work until I it is well understood.  I’ve decided to start with japanesepod101 dialogs because transcripts in english and japanese exist, vocabulary lists also exist for each season and I can listen in my car.  Ideally longer, well written stories would be better but I don’t feel like taking time to search and cobble together transcripts in 2 languages and create vocabulary lists.  Jpod101 dialogs will do just fine.  Additionally, there are levels all the way up to advanced material, so they are a bit like graded readers with audio.  The listening in my car part is important because it’s time that would be wasted, so it’s extra study time I wouldn’t otherwise have.

Before I do all that, I want to finally finish RTK and then proceed to start learning unknown vocab from jpod101 beginner season 1 and so on.  Mastering each season until proceeding to the next until it becomes too easy and I’ll move up to intermediate level.  Concurrently to this, I want to learn all of the vocabulary in a few songs that I can listen to for listening practice.  I’d also like to pre-learn the vocab used in a movie and then watch the movie until I can follow along easily.  I’ll use subs2srs if I need help with longer sentences.

I’m still not sure, but I might study for the JLPT N3 test in december time and enthusiasm permitting.

Exercise Cards* Cumulative Cards* Finish by*
finish rtk 140 140 feb 9 done! feb 7
add rtk leeches 300 289 440 mar 10 done! mar 28
rtk lookalikes deck 180 184 620 mar 28 done! feb 27
jpod101 beginner s1 vocabulary 400 1020 may 7
jpod101 dialogs - - ongoing
song vocabulary + srs + listen 100 1120 may 17
perfect groups 400 1520 june 26
anime vocabulary + watch 400 1920 aug 5
jlpt n4 ? dec?

* Card counts are approximate which makes finish by dates approximate as well.

My 2016 lesson plan

I should finish RTK by the end of the year. I’ll be traveling to Japan for the holidays, so I’m also planning to learn/refresh some survival japanese – the kinds of things that will really be useful, like reading a menu, and asking for a larger size of pants, etc. In 2016, I’m hoping to spend a lot less time learning, and more time using Japanese. Here is my tentative 2016 lesson plan:

  • Learn a few songs in Japanese that I can listen to over and over and also sing in karaoke.
  • Pick a few easy anime and/or dramas to scan and learn unknown vocab, then subs2srs, then watch.
  • RTK 2 (or at least memorize perfect groups).
  • Un-suspend all anki leeches.
  • Convert all kana vocab cards to kanji and eventually remove all furigana from the front of all cards.
  • Start reading NHK easy.
  • Start listening to native Japanese audio(podcasts, radio, etc) in the car and at work.

Progress Report: [1013d]::[1248hr]::[2329vocab]::[1621kanji]::[2312sentences]

In my last report, I mentioned that I was still planning on taking the jlpt n3 test, but almost immediately changed course.  Instead, I decided to finish RTK and then start native media.  I just felt that not finishing RTK, I would still have too many unknown kanji which would make studying native material and vocabulary too inefficient.

I currently have 350 unseen kanji(there are some suspended leeches).  I’m adding 98/week, so I’m hoping to finish just before the holidays and my Japan trip.  I’d really like to start studying some of my survival Japanese in order to understand as much as possible while I’m there.

I’ve been spending all of my time on getting through RTK, so I haven’t made any attempts to read japanese.  I’m expecting I’ll be able to understand a lot more written Japanese once I finish RTK.