A critique of SRSing, and a solution.
Zhaoyang
January 08, 2010, 05:09 PM posted in General Discussion(This is the first thread I've started here. I'm reposting this into this group where it belongs. )
Somewhere, between the initial memorization and the later real language usage, there is a breakdown in the logic and wisdom of flashcarding.
At the short-term end, these programs seem extremely good. At the long end, the item in effect graduates out of the daily routine- 人 at 2+ years is not a problem. In the middle is the morass of items recognizable with effort. As the deck grows and this burden increases, it becomes more obvious that some "real language" program is necessary, for example simply reading and listening to the materials from which the deck was build. Let's say this should have been done all along. So, the flashcards will graduate quickly, *but not because we've stretched the memory* a la the algorithm/theory, but because we're seeing the items periodically. I see 人 dozens of times a day = the 2+ year interval is nonsense. So, a reading/listening program renders the algorithm nonsensical.
人 eventually doesn't matter, but the thousands of reps of the partially-remembered items, when instead I could be reading/listening, is a huge problem. Should there be a cutoff/graduation/suspension of items? With enough supplementary reading the items are *eventually* effortless enough, but *until then*, they are a workload of "reading" in the form of flashcards. My opinion is that this is unnecessarily punitive, since I can actually read my texts with help from the audio I can recall as I'm reading. So, yes they should.
Someone might reply that with the whole-sentence method, one is reading all along. Yes, maybe it is a practical compromise, but it looks like just one step that suggests more, as follows:
As a brand-new newbie, dissect everything into tiny bits as necessary. As soon as something becomes readable/listenable, put it on a schedule to do exactly that- real reading/listening. What should the schedule look like? A graduated-interval scheduling would be nice!
[Begin extra credit assignment] Take some large body of text and/or audio. Parse it recursively. (The fullest solution would parse all the way down into strokes/phonemes.) Generate flashcards of the leaves. As the smallest bits are mastered, back up to the parent. Make a flashcard of the parent, suspending the leaves. Continue. In the end, the entire original work is scheduled for reading/listening. Implement this and be rich and famous. [end]
I have a practical idea too for graduated-interval scheduling of whole texts. My first try mixed such cards into the deck. They interrupted the routine. So, make a separate deck. It would have cards assigning the reading/listening of whole texts- whatever you're capable of. Then you could suspend all of its constituent bits in the main deck. Keeping strict track of the details of this would require the algorithm above, but it doesn't have to be that complicated: If in your readings you can't understand something, unsuspend the item, maybe, or just look it up and say "ah, I knew that".
Conclusion: For the initial memorization to intervals up to (a few days? 30 days?), flashcard programs are powerful tools, but the learner should be reading/listening to real language as soon as possible. Then the cards for the "bits" of the readings/listenings could be, and I argued above should be, suspended. I haven't actually done this yet. This being a new year, I might try it now. Doing more "real language" is why I joined CPod.
Zhaoyang
August 31, 2010, 05:40 PMFive months later...
Hey koujiacheng , thank you for the excellent reply. I regret not coming back here to thank you.
Are you still happy with your flashcarding of larger blocks?
matthiask
August 31, 2010, 07:37 PMi started to use sentences in the flashcards - it gives you context, reiterates characters you know effortless and teaches you the flow of the sentence as well as the grammar structure/patterns. So far, I'm quite happy with my retention (which was not nice with single words)
Zhaoyang
September 01, 2010, 12:15 AMThanks. Mine wasn't nice at times and I had to figure out what to do about it.
Moving from "minimum information" to whole sentences is one step (maybe the biggest single step) toward handling bigger blocks of whole language- very desirable- and incidentally not inconsistent with CPod's mission.
If all of your lessons were represented as sentences being drilled successfully, you'd get a tremendous drilling in the more frequent words on those cards. Maybe the cards for those words could be suspended. In most discussions about this someone will say they don't matter because they get promoted out of sight.
But...
I have a very strong argument that these flashcard programs should have a maximum interval limit, and that the limit should be surprisingly short.
IF that argument (which I've fortunately condensed now and will post if anyone is interested) is accepted, then the high frequency easy cards would build up inconveniently, giving a good reason to suspend them.
I have a thousand sentences so far, 120+ poems with audio, and all my other coordinated storybooks and courseware I can now read and listen to as real language, and that is what I'd rather do as much as possible when I'm not listening to CPod.
simonpettersson
September 01, 2010, 01:48 AMNot only am I a big fan of SRS, I much prefer single words to that whole sentence thingie. I will occationally add a phrase in order to memorize a particular pattern, but usually I use my SRS to memorize heaps of vocab. Here are my thoughts:
1: SRS is self-correcting. If you encounter the word outside of your SRS study schedule, you will be more likely to remember it next time and thus your review schedule will be adjusted for your better understanding. Reading and listening in no way interfere with SRS efficiency.
2: Using SRS to memorize such a common word as "人" is indeed a folly. You're likely to encounter it at least once a paragraph in your reading. With such frequent appearences, SRS would be completely unnecessary. However, a word such as "忖度" is much more rarely encountered. I have no idea of the actual frequency of this word, but let's for the sake of argument say it appears about once in 100 pages of text. I don't know how quickly you read Mandarin, but it takes me maybe half an hour to go through a page of text (since I look up every word I don't recognize and add it to one of my SRS decks). Spending an hour a day on reading (which is more than many students have time for) I should expect to encounter the word roughly once every 50 days. This is much too seldom for me to remember it from the first encounter to the next. I will have to relearn it every time I encounter it. Eventually I'd probably start to remember it, but it'll take me ages. With the help of SRS I can learn to passively recognize 30 new words a day on a fairly intense study schedule and 10 on a more leisurely schedule. The time spent on this is almost negligible. I do my SRS review during my morning walk to buy breakfast. 15 minutes there and 15 back gives me half an hour, more than enough to cover a day's worth of repetition.
Tal
Umm... what is SRS actually? Is it anything to do with dickometrics ?
simonpettersson
Spaced Repetition Software. Flashcard application that uses an algorithm to calculate a review schedule for you. It has very little to do with dickometry.
Curses, you found my blog. Now I will have to update it, won't I?
Tal
Ah ha! Thanks Simon, I get it now, (always did have trouble with acronyms.) In fact I've been learning to love Anki myself recently.
As for the blog... of course you must update it! The world is waiting! For more vomit and dickometrics!
hamshank
Simon,
I'm interested. If you only use SRS for learning vocab then how are you dividing your decks?
I have one for Words, one for sentences, one for writing and one for Zhuyin (which I can never seem to get the hang of for some reason!)
Also if you are getting through your flashcards walking to get breakfast are you not in danger of hitting a bus or something? ;)
simonpettersson
I've got one for Mandarin words, which is by far the biggest one. Then one for characters (I add them when I encounter unknown characters, but that's pretty rare nowadays) and one for trad characters (the answer is the corresponding simp). I also have one for phrases, but it's very small. Finally one for chengyu. That's in the case for Mandarin. I have another case for Cantonese, but that's divided differently, seeing as learning Cantonese is different from learning Mandarin.
In Foshan (and, I suspect, most of China), you're always in danger of being hit by a bus. What can I say? I'm living on the edge, man.
hamshank
Interesting thanks.
Well your a braver man then me! It's true enough for Taiwan too. The bus drivers seem to come straight out of the institute and into work.
daniel70
Simon, for your main Mandarin deck, do you just use pinyin->swedish/english recognition cards, or do you do the reverse production cards as well?
I too enjoyed your blog entry. It was hard to find due to its obscure url, but definitely worth the effort.
kimiik
Actually this forum is also a kind of SRS system.
I checked 忖度 and 猜度 yesterday in my dictionnary and when I just read this thread it reminded me my search of yesterday and the meaning.
If I read this thread again tomorrow, I'm sure that this two words will stay in my long-term memory.
koujiacheng
March 02, 2010, 10:10 AMIn my experience, SRS is an extremely powerful tool, and not just for the first 30 days. While flashcards can seem painful, is probably the most efficient means of acquiring and maintaining vocabulary. I don't advocate suspending words once they get to a sufficiently large interval, the reason being is that with an interval so large, they are not contributing significantly to your daily workload anyway. Might as well leave them in. Occasionally you will forget a word that you had gotten up to a 1+ year interval, although it will be very rare.
Granted, memorizing individual vocbulary items is not sufficient to learn a language, but it's a pretty damn good start. I think your idea of a recursive parsing is interesting, altho it might be difficult to implement efficiently, and I would be wary of eliminating the leaves (individual vocabulary items) for the following reason: Words can have many meanings, especially single character words. Comprehending a word in a specific context only means that you know that particular meaning of the word. For words that have many meanings, a single context is by far not sufficient. Going through the pain of flashcards ensures that the vocabulary words don't fall out out of our memory for too long.
Your idea of putting larger chunks of language on a graduated interval is something that I've already been doing in addition to my flashcards and I think that it is quite useful. Specifically, I am doing this for ChinesePod lessons and other Audio lessons that I have studied over the years. I simply created an Anki card that lists the lesson, I go listen to it, compare what i heard or thought i heard to the dialog if necessary, and then answer with how well I felt I understood. I've been doing this for about a month and I'm very enthusiastic about my progress.
So, while I agree with you that we should be spending time studying "real language", I don't think that we should be abandon SRS for vocabulary items beyond the first 30 days. The mature items are the ones that contribute least to your daily load anyhow.
For my studies, I see that doing my daily load of vocabulary flashcards as the baseline to insure that the things I've learned don't fall out of your memory. Consuming other material will help reduce your flashcard load by increasing your success rate.