Flashcard Automatic scoring Algorithm

[sthubbar]

Is it possible to obtain more information about the flashcard automatic scoring algorithm or is that considered proprietary?

According to the documentation an aggressiveness setting of 5 is similar to Supermemo et al. Why then is 3 the default setting for Plecodict?

I have done some serious tweaking in Pleco 1.0.3 with the repetition spacing and those settings. It appears that 2.0 is going to be even better, it would just be nice to really understand what is going.

Thanks.

EDIT: I found the answer here http://www.pleco.com/manual/flash.html#scorealg

Thanks.

 

[sthubbar]

After reviewing the above algorithm, I am concerned about step number 4 where incorrect answers are reset to the default value (100).

I used to use this technique when I was using Pleco 1.0 and I found that it ended up not making sense and causes undue burden upon the user by forcing many cards back to the beginning where they shouldn't be.

Think about this, after using this program for a year or so, there will be cards that might have a repetition interval of 2 - 3 years. It is 100% sure that some of these items with 2 - 3 year intervals will be recalled incorrectly. Maybe the student was just having a bad day, or responded quickly, or maybe the interval was just a little to long.

Does it really make sense to change the interval on an item from 2 - 3 years down to 1 day?! This is ridiculous. The interval should be proportionally be reduced based upon the difficulty of the item, so that the next review would happen say in 1 year or 1.5 years.

I'm hoping that this example proves the point, though the same is true for items that have a 1 week or 1 month interval already and then are being reset to a 1 day interval. It doesn't make sense and is burdening the learning.

The better solution is to adjust the interval(score) down based upon the difficulty. In other words, there should be a way to decrease scores, as well as increase them.

Does that make sense?

 

[mikelove]

There's an option for that in the Scoring panel of Advanced Settings in 2.0.1 - "scale decreases by difficulty". The reason we lower scores to 100 by default (and I imagine the reason SuperMemo does the same thing) is that, particularly with repetition-spaced study, if you get a card wrong because you completely forgot the word (and I imagine almost everyone here has had words just drop out of memory like that - I certainly have) you could have to wait a very long time before you see it again. Dropping the interval from 2 years to 1 year or even 1 month isn't really acceptable in that case, it's a word you're supposed to know and now you've forgotten it and now you need to start learning it again.

Now there are two areas where this seems to give people problems. One is with cards that were accidentally answered incorrectly, tapping the wrong button when you're sure you actually knew the correct answer; the "Skip Card" button can help in most of those cases, since tapping it before advancing to the next card will avoid the score change, but if you like really rapidfire flashcard testing and aren't sure you'd even notice you'd tapped the wrong button then "scale decreases" might be a useful option. The other problem area is cards that you almost but didn't quite remember, but remember as soon as you see the correct answer; if you have a lot of those then "scale difficulty" also might make sense. In the future we might want to consider tying "scale difficulty" to which incorrect score you choose - the "completely incorrect" score of 1 could drop the score back to 100 while the "almost correct" score of 3 would scale it by difficulty (and 2 perhaps more aggressively by difficulty).

As for the aggressiveness of 3, that was chosen because SuperMemo's algorithm seemed a bit too aggressive (at least in our testing) and because with frequency-adjusted testing it seems better not to have quite such a dramatic gap between new and old cards.

 

[sthubbar]

OK. I don't understand "scale decreases by difficulty" so I'll just have to wait for the documentation to come out and figure that one out.

Thanks.