Register  Login  Active Topics  Maps  

The "low hanging fruit" method for Anki

  Tags: Anki
 Language Learning Forum : Learning Techniques, Methods & Strategies Post Reply
16 messages over 2 pages: 1


emk
Diglot
Moderator
United States
Joined 5538 days ago

2615 posts - 8806 votes 
Speaks: English*, FrenchB2
Studies: Spanish, Ancient Egyptian
Personal Language Map

 
 Message 9 of 16
28 October 2014 at 1:55pm | IP Logged 
I'm really not a fan of pre-made decks, because they always seem to involve far more suffering for less payoff than decks I make myself. But then again, I have much-better-than-average card creation tools, and I can easily capture words in context and add definitions.

But I definitely agree with your advice:

Improbably wrote:
The method is really just a simple change of settings in Anki:

Set the leech threshold to 2, and leech action to suspend card. The "inspiration" for this, is this simple statement in the article "The 20 rules of formulating knowledge in learning" on the SuperMemo website: "Remember that usually you spend 50% of your time repeating just 3-5% of the learned material!" We want to eliminate all the difficult cards as soon as possible.

I've kept my leech threshold at 4 for a good while now, and I'm beginning to suspect that it's worth lowering it even further. If I don't get a card the first few times I try to learn it, it's going to be a continual nuisance. But there's a good chance it will be easier later on.

For example, back in my Memnosyne days, when I had an L1<->L2 vocabulary deck (*shudder* *twitch*), I spent a ridiculous amount of time agonizing over a particularly evil group of cards:

- ainsi que "as well as"
- alors que "while", "when"
- tant que "as long as"
- autant que "as much as", "as many as"
- en tant que "as"

But later on, I learned all of these distinctions easily, most of them from context. It's possible for a card to agonizingly difficult at one stage, and ridiculously easy later on. I've seen much the same thing learning hieroglyphs: Some of them are very difficult to learn on their own, but once I see them in the right word, they stick easily.

This is one of the major reasons why I'm such a big fan of deleting or suspending cards. As rdearman mentions, I've configured AnkiDroid to suspend a card when I swipe my thumb to the right. And I discard any card which makes me feel stressed or depressed or ignored or bored.

As always, if you actually need to know something, you'll see it again in your input soon enough. And maybe next time, it will appear in a better context, or you'll know a few more related words, or your brain will simply be ready to learn it.
6 persons have voted this message useful



Ezy Ryder
Diglot
Senior Member
Poland
youtube.com/user/Kat
Joined 4355 days ago

284 posts - 387 votes 
Speaks: Polish*, English
Studies: Mandarin, Japanese

 
 Message 10 of 16
28 October 2014 at 2:41pm | IP Logged 
patrickwilken wrote:
However, given the exponential (?) drop-off in word frequencies, I doubt
you would seem many words past the 10k range in the Super Challenge ever, which doesn't mean
they aren't useful, just that they only occur only once in a million words or so.

I thought the drop-off was logarithmic (if that's the right word). And 10k won't give you that high of
a coverage, meaning, if you want to read 5k pages (~1,25kk words), you're gonna see plenty of
rare words (though, it depends what you'll read).
For example, I've recently been reading a light novel in Japanese (so, aimed at teenagers), and
across six pages I've seen probably some 9 or more words which were in the 20k, or even 30k
range on the frequency list I use (although it lists なお and 尚 as separate entries, so maybe they
would be a bit higher on the list of word families). So, as the drop-off is logarithmic, each
individual word may occur very rarely, but the sheer number of such words makes them contribute
to a fair part of an average novel.
1 person has voted this message useful



rdearman
Senior Member
United Kingdom
rdearman.orgRegistered users can see my Skype Name
Joined 5242 days ago

881 posts - 1812 votes 
Speaks: English*
Studies: Italian, French, Mandarin

 
 Message 11 of 16
28 October 2014 at 3:25pm | IP Logged 
patrickwilken wrote:
rdearman wrote:

I figure these decks are only for review or to see a new word a couple of times. If it is a word I need then I'm bound to see it in my extensive reading in the super challenge.


You are working through the 20k most common words? Wow. That's impressive. However, given the exponential (?) drop-off in word frequencies, I doubt you would seem many words past the 10k range in the Super Challenge ever, which doesn't mean they aren't useful, just that they only occur only once in a million words or so.


Actually in Italian is is slightly more than 20k and French slightly less, but don't be to impressed, I've seen a little more than 10% of the words so far, so I've only done 2000+ words. The advantage of this deck however is I think I'll quickly burn through the first 10k words. To reach C1/C2 I'd need to expand my vocabulary well over the 10k mark. So once I've gone through the first 1/2 of the deck, I bring the leech setting back to normal and bring down the new cards in order to use this as an advanced vocabulary deck.

That is the plan anyway.
1 person has voted this message useful



Ari
Heptaglot
Senior Member
Norway
Joined 6588 days ago

2314 posts - 5695 votes 
Speaks: Swedish*, English, French, Spanish, Portuguese, Mandarin, Cantonese
Studies: Czech, Latin, German

 
 Message 12 of 16
29 October 2014 at 8:22am | IP Logged 
emk wrote:
As always, if you actually need to know something, you'll see it again in
your input soon enough. And maybe next time, it will appear in a better context, or
you'll know a few more related words, or your brain will simply be ready to learn it.


But what about all those words you don't need to know but want to? I know lots of
Cantonese characters that most natives don't know. I don't need them, but I love
knowing them, and they're great discussion topics for interested natives.

Some characters I'd probably benefit from dropping, though, but I do very little
reading in Chinese, so I'm afraid to drop things I might encounter at some point. Also
it's really nice to have an exact number of the amount of characters I know (5073).

I think I might follow this advice in other languages, though. At the moment, I don't
have a leech threshold at all, and I rarely delete cards.
3 persons have voted this message useful



patrickwilken
Senior Member
Germany
radiant-flux.net
Joined 4539 days ago

1546 posts - 3200 votes 
Studies: German

 
 Message 13 of 16
29 October 2014 at 9:46am | IP Logged 
Ezy Ryder wrote:
patrickwilken wrote:
However, given the exponential (?) drop-off in word frequencies, I doubt
you would seem many words past the 10k range in the Super Challenge ever, which doesn't mean
they aren't useful, just that they only occur only once in a million words or so.

I thought the drop-off was logarithmic (if that's the right word). And 10k won't give you that high of
a coverage, meaning, if you want to read 5k pages (~1,25kk words), you're gonna see plenty of
rare words (though, it depends what you'll read).
For example, I've recently been reading a light novel in Japanese (so, aimed at teenagers), and
across six pages I've seen probably some 9 or more words which were in the 20k, or even 30k
range on the frequency list I use (although it lists なお and 尚 as separate entries, so maybe they
would be a bit higher on the list of word families). So, as the drop-off is logarithmic, each
individual word may occur very rarely, but the sheer number of such words makes them contribute
to a fair part of an average novel.


For English if you know the most common 8000-9000 words you get about 98% coverage of text. So 2% or 1:50 words in an adult novel should fall in less frequent categories.

The problem is that these words only appear once or perhaps twice in an individual book and won't appear in the next book you read. So yes, you can learn them, but once you've learnt them you may not see them again for a long time, certainly longer than the 5000 pages (1.25 million words) of the Super Challenge.
1 person has voted this message useful



Improbably
Diglot
Newbie
Norway
Joined 4942 days ago

34 posts - 87 votes 
Speaks: Norwegian*, English

 
 Message 14 of 16
04 November 2014 at 2:14pm | IP Logged 
In the article about leeches at supermemo.com it is suggested that leeches be exported to a mobile device to be worked on separately, and that knowing that every card you're studying is a leech might sharpen one's focus. So I guess an additional tip would be to move all your worst leeches from all your decks to a separate deck in Anki, and review that deck, say, at another time of day. That way, you can give the leeches the attention they deserve.

Edited by Improbably on 04 November 2014 at 2:15pm

1 person has voted this message useful



yogert909
Newbie
United States
jonkenpo.net
Joined 3798 days ago

5 posts - 7 votes
Speaks: English*
Studies: Japanese

 
 Message 15 of 16
15 November 2014 at 5:51am | IP Logged 
I've been obsessed with leeches for some time based on the same information quoted from supermemo.
However, there are a few things I've
been thinking about that don't necessarily contradict setting a low leech threshold - but might:

1. From my understanding, anki doesn't count a lapse until a card has transitioned out of learning. By that time,
you've already done a good
number of reps before even your first lapse. Therefore, perhaps you may have already sunk a large proportion of
time you will ever spend on
that particular card.

2. Once a non-learning card lapses, I don't believe it typically takes very many additional reps to either 1. get to
a long interval, or 2. lapse a few
more times and get suspended with threshold of 5 or 7 for that matter. In other words, the difference in number
of reps between my cards with
2 lapses and 5 lapses is not that great. I'm wondering if it would be more efficient to spend the extra reps to
"see" if it is actually a leech, or if 2
lapses were a sort of fluke. Of course, a few cards will eventually lapse many more times unless they are
suspended, but I'm just not sure that a
lot of cards that I've already invested time in and happened to lapse twice aren't actually low hanging fruit
themselves.

3. Another way of looking at #2 is after a card lapses a few times, I have a choice. I can trade it for a fresh card
that I've never studied before, or
I can keep studying it, hoping that it will stick. Which card is more likely to become mature with the least amount
of reps? I know there's a
threshold somewhere, and maybe 2 lapses is where that threshold is, or not..I haven't decided.

4. The next thing I've been thinking about is partially psychological one. I've had my threshold set to 5 for a
while now and I get one or two
leeches almost every day. I'm only adding 10-20 cards per day, so for every, say 10 cards I add and spend time
learning, 1 of them will end up
getting suspended. Some days I get a leech and I'm like "later...good riddance!" but some days it really bums me
out because I thought I actually
know the card and remember getting it right a bunch of times. Maybe it even had an interval of 6 months and
something wasn't clicking in my
brain today and now it's gone and I'll forget it for sure... And it bums me out. I'll probably fail the next card
because Im now in a foul mood.
Anyway, maybe I'm making too much of this, but maybe you felt a bit of this too..?

5. The supermemo developers seem to have crunched a lot of data and figured out somewhat optimal settings. I
haven't checked, but I'm
thinking Damien just copied supermemo's SM2 algorithm almost verbatim including the optimal leech threshold.
So perhaps 7 is actually the
magic number if your goal is optimizing quantity of facts per unit of time.

6. There's probably better ways of catching leeches before so many reps are put in.   Filtering by lapses requires
quite a few reps before the card lapses beyond whatever threshold you set. I would bet you could identify
leeches by identifying the cards that stay in learning the longest and suspending them before you've spent too
much time on them.

Anyway, am I over thinking this? The best thing to do would be to go through a long mature deck and crunch
some learning data to see where
the actual threshold should be set. Someday... But until then let me know what anybody actually thinks about
any of this.

Edited by yogert909 on 17 November 2014 at 7:55pm

1 person has voted this message useful



chiara-sai
Triglot
Groupie
United Kingdom
Joined 3714 days ago

54 posts - 146 votes 
Speaks: Italian*, EnglishC2, French
Studies: German, Japanese

 
 Message 16 of 16
15 November 2014 at 10:37am | IP Logged 
emk wrote:
For example, back in my Memnosyne days, when I had an L1<->L2 vocabulary deck
(*shudder*
*twitch*
), I spent a ridiculous amount of time agonizing over a particularly evil group of cards:

- ainsi que "as well as"
- alors que "while", "when"
- tant que "as long as"
- autant que "as much as", "as many as"
- en tant que "as"
.


I’ve also found conjunctions and adverbs to be the hardest words to remember, probably because it’s hard to
picture
them in your mind. I second the idea of suspending/deleting them very liberally, but if one doesn’t like deleting
cards I
found that it helps to put those words into context, such as creating cards with short phrases that contain the
preposition in question: instead of having a card that is [‘as well as’ -> ‘ainsi que’], have one like [‘I really like
the name
as well as the presentation’ -> ‘J’aime beaucoup le nom ainsi que la présentation’]*. Personally I try to use this
latter
method, although I still suspend cards after a 5 lapses.

*I found this sentence with Goole, using the mask: "J'aime beaucoup * * ainsi que"
(including the quotation marks). Although Google doesn’t guarantee that it was written by a native
speaker,
coming up
with my own would guarantee that it isn’t, so it’s still an improvement.

Edited by chiara-sai on 15 November 2014 at 10:39am



1 person has voted this message useful



This discussion contains 16 messages over 2 pages: << Prev 1

If you wish to post a reply to this topic you must first login. If you are not already registered you must first register


Post ReplyPost New Topic Printable version Printable version

You cannot post new topics in this forum - You cannot reply to topics in this forum - You cannot delete your posts in this forum
You cannot edit your posts in this forum - You cannot create polls in this forum - You cannot vote in polls in this forum


This page was generated in 0.2969 seconds.


DHTML Menu By Milonic JavaScript
Copyright 2024 FX Micheloud - All rights reserved
No part of this website may be copied by any means without my written authorization.