How to best remember new words

David Calinski   Tuesday, October 05, 2004, 22:02 GMT
and please don't accuse me of misquoting (I copied&pasted it, how I can misquote then?)
David Calinski   Wednesday, October 06, 2004, 12:48 GMT
By-the-way: if you don't see any difference between weights in ANN (thus ANN) and parameters in algorithms (thus algorithms), maybe you will be so genius to code algorithm that does OCR, huh?

Bear in mind that human memory (and human brain) is VERY complicated, there is MANY MANY MANY factors to consider - we probably don't even know yet about most of the factors (when optimizing repetitions), as we still know very little about human brain.

ANN is a sophisticated modeling technique, capable of modeling extremely complex functions that we even don't understand, and scheduling optimal repetitions is a perfect example of it.

Nobody yet understood ALL factors contributing to forgetting in human memory - nobody yet understood human brain totally.
ANN allows us to perfectly model such complex functions with huge number of variables we cannot track.

ANN is able to learn itself - without any limits. Just like our brain is able to learn.
ANN in FullRecall, as I said, could learn, memory model of a monkey, dolphin, ... or learn itself a completely different thing (e.g. solving a problem from medicine, engineering).
I could bet 100000$ that algorithm in SuperMemo (nor any algorithm made purely by human being so far) is NOT so elastic.


All-in-all: I need to stress it: I don't claim that FullRecall (or scheduling of repetitions by FullRecall) is better (although I PERSONALLY think it is).
ANN has its own simplifications, algorithm in SM has its own simplifications.
I could prove that APPROACH to the problem of forgetting in FullRecall is better (as ANN is better approach to the problem), but this is in theory.
Keep in mind that better approach not necesarrily result in better... results.

If you are swimming a better style, but you swim since 1 year - you still can swim slower than somebody who trained swimming for 15 years, uses worse method of swimming, yet more trained. Right?
The same here!

(I'm happy though to have valid solution to the problem of scheduling repetitions by ANN in FullRecall and I could improve ANN with the time - remind me 10 years later: I bet SuperMemo will use ANN too.)
Quolta   Wednesday, October 06, 2004, 12:58 GMT
Yeah, looks alike as if Supermemo does not like critisise. But If FullRecal is realy better? Many people would maybe buy from David then I think so! Would Wozniak and Tom like that? Don't think so! Supermemo is big player but David not. Looks alike David against Goliat. Keep up David!!
David Calinski   Wednesday, October 06, 2004, 13:09 GMT
Hey! it's not like that! in fact - I do cooperation with SuperMemo and Dr Piotr Wozniak is my friend, we even live in the same country :)

There are plans that my FullRecall will be soon official "SuperMemo for Linux", and later maybe cross-platform SuperMemo-light version.

One way or another - thanks to XML format soon FullRecall and SuperMemo will be even more compatible with themselves.

And I don't like to say which one is better - SM and FR has its own pros&cons.
I personally like FR, because I like what I made, I like simplicity, speed and ANN... but it's just me. Believe me or not - I still also use SuperMemo on Windows (i still have some collections in SM format), and Dr Piotr Wozniak will soon use my FullRecall/"SuperMemo on Linux" on beta-tests on Linux platform.
There is no clear winner - it all depends on what one wants from the program. And some people (e.g. using Linux/FreeBSD) doesn't have much choice...
Steve K   Wednesday, October 06, 2004, 17:29 GMT
I have come to the conclusion that since words and phrases are best learned in context, and that content has to be read and listened to over and over in order to get a feel for the language, the whole argument about which algorithm to use is really not that important to the language learner.
TomD   Wednesday, October 06, 2004, 20:34 GMT
Steve K:

Argument about which algorithm to use is maybe not important to any language learner, but it might be important to a flashcard software user. (Ie.: "in my current flashcard software, the job gets done in 20 minutes, but what if there is something else that will do the same job in 10 minutes?")

Most experienced/advanced flashcard software users do NOT use it to learn stuff, they use it to keep words and phrases in memory.

(Eg.: I learn programming language, similar to learning languages. It would be almost impossible for me to learn directly from the flashcard software database, I have to learn from context, by reading/using. However, after it is learnt, I put it into the flashcard software, to keep it in memory.)
TomD   Wednesday, October 06, 2004, 21:46 GMT
sorry: ...not important to every language learner....
Tom   Wednesday, October 06, 2004, 22:49 GMT
David -- Just to clarify, I said DaVinci was misquoting the SuperMemo website, not you!
You're right, Piotr Wozniak is listed as one of the authors of that article. I did not notice his name there.

As for the NN controversy, it's quite obvious to me that a NN is a type of algorithm. Neither of us has studied the SM algorithm, so we don't know what assumptions it makes. It probably makes more assumptions that your software, but that's not necessarily a bad thing. After all, few people are interested in training dolphins.

Steve -- I completely agree with you. This discussion is rather pointless. Nobody has performed any real tests to compare the available solutions. I guess none of the companies/individuals who develop such software are really interested in verifying its efficiency.
In any case, the results of such tests would be of marginal importance to the language learner.
Quolta   Thursday, October 07, 2004, 10:02 GMT
Hey Tom,
...I guess none of the companies/individuals who develop such software are really interested in verifying its efficiency...

U Must be joking, man! Look again what TomD sayd: "in my current flashcard software, the job gets done in 20 minutes, but what if there is something else that will do the same job in 10 minutes?")

Tests with Flashcard software are very intersting and good for individuls!! Ofcours if SM then makes only second place or third. Who will buy SM then? Looks more alike as SM, the big player, is anxius of tests with no-name Flashcard software!!
Achab   Thursday, October 07, 2004, 10:53 GMT
Quolta,

it is not true that Supermemo's developers avoid making comparisons with other programs or dislike to discuss the capability of their software.

When I asked them to make a comparison between their software and VTrain, they answered, and published their reply in a message board. And their reply was honest, they stated what they thought Supermemo was better at, and what they thought Supermemo was worse at.
TomD   Thursday, October 07, 2004, 13:56 GMT
Quolta: What for a joke did you find in my example? :-). Maybe I shall have expressed it more clearly: I meant 10 or 20 minutes per day of average daily work with the software. Such a difference is highly possible between various flashcard software! It is enough if the algorithm schedules items for repetition on much tighter intervals.

I agree with Achab, there is no one anxious to make some tests. The problem is that it would take a long time (year, two? - who will devote so much time? and who will persist till the end of testing, maybe just 20 or 30%?), lots of volunteers, a very good comparison methodics, and there still would be no guarantee such a comparison would make sense. For example, it could be distorted by different levels of knowledge of volunteers, different amount of additional work done besides working with the software (class hours, reading, watchin TV,...), different learning types/styles of volunteers, and many other issues.

Though, it could still be very interesting to see such a comparison. However, I think, from practical reasons, it is not very probable someone will do it. About few months ago, I started thinking about such an experiment myself, quite seriously, but I have dropped it - I have found I have better things to do with my time :).
Quolta   Friday, October 08, 2004, 13:35 GMT
TomD,
What for a joke did you find in my example? :-)

Hey man, this was other Tom joking, not U!

Achab, where is SM reply found in a message board?

David,
There are plans that my FullRecall will be soon official "SuperMemo for Linux", and later maybe cross-platform SuperMemo-light version.

What is basis then for this: your ANN or SM algorithm?
David Calinski   Friday, October 08, 2004, 14:58 GMT
Quolta: ANN

by-the-way: I'm not officially (yet ;P) affiliated with SuperMemo, but if you want to hear my opinion (from 3-4 years SM user): I do think SM is the best and most reliable software when it comes to remembering knowledge (counting all SM's features and capabilities). but it's just my personal opinion. I still like my FR, I like speed, simplicity, simple Q/As, using software also on Linux, I prefer the ANN approach, yet I recognize SM as superior (not only to FR, but all similar flash-cards software I have tried) on Windows.
SM has been developed and improved since about 17 years.
sure, SM has its cons: is too complicated (this is a price for so many features), too bloated, too much relied on Microsoft-Windows, yet... it's still SuperMemo.

"...I guess none of the companies/individuals who develop such software are really interested in verifying its efficiency... "

I assure you that we are interested. I'm for sure.
I bet SuperMemo World is interested too - we all are interested in improving efficiency, and to improve it, we need to verify it.
unfortunatelly, it's not so easy to verify it :-\

"Tests with Flashcard software are very intersting and good for individuls!! Ofcours if SM then makes only second place or third. Who will buy SM then? Looks more alike as SM, the big player, is anxius of tests with no-name Flashcard software!! "

Even IF SM is not the best when it comes to scheduling most optimal repetitions: remember that SM is still a big player: SM is very powerful tool when it comes to learning (e.g. incremental reading!); scheduling optimal repetitions is just a fraction of what SM does.

EVEN IF SM is not the best at scheduling repetitions - it's all features (multimedia, scripts, HTML-formatting, templates, incremental reading, ...)
make it great software.
And SM has a huge number of databases and users! And they are still improving their software.
Achab   Friday, October 08, 2004, 17:00 GMT
Quolta,

the message board I was referring to is the one on Yahoo! Groups devoted to Supermemo.
Achab   Friday, October 08, 2004, 17:06 GMT
Hey David, let's assume that a version of Supermemo without its many features (incremental reading, HTML formatting, multimedia, etc.) is released. What would you prefer? SM or FullRecall? Or they would be at the same level? I mean: do you think that Supermemo is better than FullRecall due to its many features only?

Achab