Wired Article on Spaced Repetition Software

[giraffe]

Some people might be interested in this article in Wired magazine about the guy who developed SuperMemo an early SRS program:

http://www.wired.com/medtech/health/magazine/16-05/ff_wozniak?currentPage=all

It's a bit of an odd article in that it completely ignores all the other SRS programs out there but there's some interesting information on research into memory (and forgetting) which is relevant to learning languages.

[Pengyou]

I have been watching this program for a couple of years and have subscribed to a yahoo group about it. As I renew my interest in Chinese studies I am considering purchasing it. What I have learned about it is:

- it has a big learning curve, i.e. takes quite a while longer to learn than most people are comfortable with

- once learned, it is quite effective. I have read posts from people who swear by the program.

- while pretty stable, it is still a little buggy and does not have much support.

What are the other SRS's?

[OneEye]

I wouldn't recommend SuperMemo. It's old, clunky, and confusing. Anki is much more user-friendly, IMO, while still being very customizable, while Mnemosyne is about as bare bones as it gets.

One of the coolest features Anki has is the ability to import Smart.FM courses as flashcards, including audio. For instance, I'm going through the Chinese Media series right now. I imported every sentence from the course into Anki as a fact (containing the Chinese sentence, audio, pinyin, and English translation), and I've made both reading and listening comprehension cards in my deck. Once you've imported the deck, you just put it on autopilot. For me, I'm currently learning 20 new sentences per day from the deck, and I'll probably increase that soon. It's a pretty painless way to acquire new vocab and grammar constructs, as long as you show up every day and do your reviews.

[imron]

Anki and Mnemosyne seem to be the two most popular. They are both significantly easier to use than SuperMemo and are free.

[OneEye]

Oh yeah, I forgot to mention the "free" part. Big "selling" point.

[Scoobyqueen]

They are both significantly easier to use than SuperMemo and are free.

I didnt think Anki was free. You can download it for free but arent you supposed to pay the fee?

Incidentally where do you think the originator of Anki is from? He does an intro on the webpage and I cant work out the accent.

[renzhe]

Anki is most certainly free in every sense of the word.

They encourage people to donate a small amount to help with the server costs and other costs the developers incur, but this is in no way binding. Many free projects do this.

The decision between Anki and Mnemosyne is generally made based on preference. they have somewhat different philosophies, and this manifests itself in some of the design decisions.

When it comes to learning things, both programs are very good, and the authors of both programs agree on this.

[Pengyou]

Thanks! You just save me $50 ....more than pays for my subscription to Chinese-forums.com ;)

[PhilipLean]

Oneye

Could you explain how to load Smart.FM lessons into Anki, I can't seem to get ti to work.

I can't capture the Smart.FM lessons.

One of the coolest features Anki has is the ability to import Smart.FM courses as flashcards, including audio.

thanks

[OneEye]

Ah, I forgot to mention that.

You need to install the Smart.FM importer plugin. Go to File-->Download-->Shared Plugin, and select the Smart.FM - Improved Importing plugin. Once it's installed (you may have to quit and restart Anki), click Tools-->Smart.FM Importer and copy the url of the list into the url box. Select whichever preferences you wish and click Import. It should compile the deck automatically from there.

[PhilipLean]

thanks

I had installed the plugin but hadn't figured out how to use it.

[nwinter]

Smart.fm itself includes spaced repetition, by the way, if you wanted to use it directly. Pimsleur audio courses use it on a small scale. PlecoDict has spaced repetition options for its flashcards.

There are a ton of other SRS programs, but I can't think of any that compare to Anki, Mnemosyne, SuperMemo, Smart.fm, and PlecoDict.

There's also Skritter, for Chinese and Japanese, which I would recommend over the others since it's my baby. Its SRS is designed specifically for Chinese, unlike the others. It's not free, though.

[OneEye]

To be blunt I think the Smart.FM SRS sucks. It's too complicated and takes too long to get through 10 flashcards. The other programs are far superior, IMO, but I can't comment on Skritter because I haven't used it.

[nwinter]

I actually just did a calculation yesterday about that. Smart.fm posts stats on how many items were reviewed the previous day and how many hours were spent practicing. Their site is much larger than Skritter: their users had reviewed 10 times as many items--in 100 times as much time. So each item takes ten times as long on their site to practice.

Their items may be bigger than ours, though: we're operating on individual characters are words; I don't know if they're counting sentences in there.

[PhilipLean]

I think CHINESE-COURSE.com should be counted seriously among those SRS sites. It is steadily building up more sentences and is is very flexible. The presentation(display) is very good plus it gives regular emails with new sentences to learn.

However, I do like having the cards and audio stored on my computer/s. The download of SmartFM cards is proving useful to me already, the audio is very good.

[imron]

Its SRS is designed specifically for Chinese, unlike the others.

Out of curiosity, how does an SRS designed "specifically for Chinese" differ from any other SRS? I always figured an SRS is all about tracking a given person's forgetting index. Does the forgetting index for Chinese characters change compared to that of other facts?

[nwinter]

Yes, I think that Chinese characters are typically harder to remember than many pieces of knowledge, at least at first. I recall seeing somewhere assumptions in the SuperMemo material about starting off cards with long initial intervals (was it some hours, or a day?). I can also recall how impossible it was for me to remember a new character for more than ten seconds when I first started learning Chinese, let alone a day. That isn't too important, though, as if you're only studying Chinese (and not a bunch of easy Spanish words or basic facts alongside), your factors will eventually catch up.

Mostly the differences between the two spaced repetition approaches lie in taking advantage of the structure of the language. Assuming familiarity with an Anki-style SRS, I'll develop an example: the word 打喷嚏 (dǎpēntì: "sneeze").

In most SRS, you can enter this word and have a couple cards created: one prompting you with 打喷嚏 to produce "dǎpēntì: sneeze," one prompting you with "sneeze" to produce "dǎpēntì: 打喷嚏", and maybe some others, like prompting with the reading or including pictures. Then these cards can be scheduled intelligently with respect to each other. So you have two or three items.

If we design specifically for Chinese, though, we can try to be more efficient by tracking more items. We can create an item for the writing of 打喷嚏, one for its reading (dǎpēntì), one for its tones, and one for its definition, as well as the same items for each of the component characters 打, 喷, and 嚏. When certain parts are due close to each other, we can prompt for them together instead of having to space them apart. When they're not, we can prompt for only the part that's due.

So if I add 打喷嚏, I'll initially study all the parts (in two promptings, as you can't do all four at once). I'll get the writing and tone 打 right after I realize that it's that dǎ in this word, which will affect its scheduling by itself and in my other words. After that, I can automatically skip writing out 打 in this word since the system knows it's 喷 and 嚏 that I don't know yet. If I learn 喷 but still keep missing 嚏, it can prompt just 嚏 until I get it.

Also, since 打喷嚏 is only hard for writing, I'll quickly learn the definition and reading. But when I'm doing the reading, I'm also doing the tones, and I sometimes mix them up and think 喷 is fourth tone. Next time I'll just get the tone by itself, which I can review very quickly.

I hope you can see how this would be much faster than producing a full "dǎpēntì: 打喷嚏" every time I see the "sneeze" card when all I need to review then is the "pēn" or 嚏 parts. The other thing being Chinese-specific lets us do is forgo the grading-yourself step, since it can determine the correctness of your input for you in most cases: write the character correctly, draw or type the right tone, type the pinyin, type a significant word from the definition. If you prefer, you can still grade yourself to be more fine-grained about certain prompts, but it often helps to skip that so you can go faster.

One other advantage is that you can easily combine lots of different vocab sources whenever you want, since the items exist independently of any deck. We know which words and characters you already have, so there are no duplications. It's also even easier than in other SRS to input vocab, since the only thing you need to specify is the characters. We can be smart about which characters are different between simplified and traditional, so you don't duplicate there, either, if you're learning both.

It does make it harder to stop studying from certain collections of words, though. We're not yet tracking every place an item was added from, so you can't just turn off a textbook and only stop studying those words that were only added from that textbook. That is to say, you only have one deck.

Further efficiencies we can take advantage of are sharing item difficulty data across users. Normal SRS can only realize 打喷嚏 is really hard to write after you miss it a bunch of times. Since a bunch of other Skritter users have already missed 打喷嚏 over and over before you, though, we can start it off with a high initial difficulty. And there are a ton of really easy characters that, since we already know they're easy, we can very quickly schedule out to long intervals, which means you can add everything indiscriminately and not worry about wasting time on stuff you already know.

Anyway, I got excited. There's a bunch of other stuff but I'll stop here.

Disadvantage: our items aren't as flexible as arbitrary cards (you can't do sentence practice yet, or things other than Chinese or Japanese words).

Advantages: you hardly have to do anything to get those items, and we can be way more efficient when prompting for and scheduling them.

Note that I'm still working on the reading and definition practice, the heuristics for skipping well-known characters in words, and the initial item difficulty estimates, so this sort of describes as Skritter's SRS as it will be soon rather than as it is.

[imron]

Thanks for the great writeup.

[Artem]

I really like the overall idea of that, but I think it can lead to some problems (atleast in the way I learn chinese).

I think to me it would be counterproductive to split up the word. In the way I study, I learned words as words, rather than a collection of characters. By splitting up the tones and characters, I feel like I would have to piece together the word every time I want to use it. I do believe learning characters is important, so I would like the idea that when I add a word, the characters would also be added as additional cards. That's a time saver.

However, take 强调 for example, and let's say I already know 调 well. if I was learning it, I would prefer to learn it as qiángdiào rather than concentrating just on 强 since I already know 调. I feel that if I concentrate on one of them, they won't flow as a word, but rather every time I have to think "okay it's 强 first, and then 调."

I feel like it would lead awkward sounding sentences because I take pauses often after characters instead of after words (to my hearing it's the equivalent of someone saying half a word in english, taking a pause and saying the rest of the word)

Plus I think it would confuse me to have 强qiáng in my review without the 调, since 强 by itself would be read as qiǎng. I would probably develop the habit of saying qiáng instead of qiǎng.

I really like the idea though, I think it definitely has potential, depending on your learning style.

[PhilipLean]

nwinter

I was posting a response here, the Admin have moved it to the Skritter feedback discussion.

Mostly a few suggestions/ comments on Skritter so that is the best place.