Imron on “You can learn chinese” Podcast

[imron]

1 hour ago, wibr said:

The SRS algorithm tries to show you cards that you are just about to forget,

The algorithm is a lie

 

I mean I know the theory of the algorithm, but how well does it work in practice?

 

Many SRS programs are based on the SuperMemo 2 algorithm, and the guy who invented the algorithms has written extensively about them and about memory (see his articles here) and thinks that SM2 has many flaws (currently he's up to SM18).  SM2 is decades out of date with his research, so if you are using an SRS based on SM2 (like Anki) then it's almost certainly *not* showing you cards that you are just about to forget.

 

The other reason it's a lie is because the algorithm is only as good as the data you feed it.  If you feed it good data you'll get good results and if you feed it bad data you'll get bad results.

 

If you need a second to remember the word, and tell the algorithm that you know this word, then the standard the algorithm is tracking is 'taking a second to remember the word', and this standard is inadequate for real world usage.

 

1 hour ago, wibr said:

so I think it's natural that sometimes it might take a bit longer to recall the meaning and pronunciation

And this is my entire point - most of the cards in the deck will be these long term, mature cards, and if you're not careful you'll end up with a deck full of cards that take a small amount of time to remember, which is not a good enough standard.  You might be seeing deck numbers going up, but it doesn't relate to increase in real world ability.  Set the standard higher, and you'll see the interval between reviews reduced to the point where you won't need to think for a moment before recognizing it.

 

1 hour ago, wibr said:

but I would say usually it helps.

Incidentally, the guy who created all the algorithms used by flashcard programs believes reading to be the most effective way of remembering facts, and for maximum retention and recall recommends using SRS scheduling to schedule reading of short passages of text - a process he calls 'incremental reading'.

[vellocet]

7 hours ago, imron said:

Many SRS programs are based on the SuperMemo 2 algorithm, and the guy who invented the algorithms has written extensively about them and about memory (see his articles here) and thinks that SM2 has many flaws (currently he's up to SM18). 

Any reason the programs haven't changed to SM18?  Is it copyrighted or something?  Or are developers just not aware of this?  

[大块头]

14 hours ago, DavyJonesLocker said:

the forum now knows you have been cycling around your living room on your unicycle practicing juggling 6 balls ?

 

For his safety, I hope the sword is put away while he's doing that!

[imron]

7 hours ago, vellocet said:

Any reason the programs haven't changed to SM18?

I don't know if all of the algorithms are published, but the later algorithms that are published are more complex than the earlier ones and some developers think the extra complexity isn't worth the benefit.  For what it's worth, SM2 was almost 20 years old when Anki was created.  You can read more about the history of the algorithms here.

[imron]

7 hours ago, 大块头 said:

For his safety, I hope the sword is put away while he's doing that!

Only when there are small children running about.

[Jan Finster]

Only found this now. Really liked it.

 

@Imron, please do a longer interview (like 2-3 hours)

 

 

(https://podcasts.apple.com/tt/podcast/21-how-important-is-chinese-culture-when-learning-chinese/id1451549039?i=1000454296682)

[mungouk]

Wow that supermemo.guru website really is quite something, too.

 

[imron]

On 4/2/2021 at 9:03 AM, Jan Finster said:

@Imron, please do a longer interview (like 2-3 hours)

Happy to do longer interviews, but it's really up to the host more than me.

 

On 4/2/2021 at 9:49 AM, mungouk said:

Wow that supermemo.guru website really is quite something, too.

Lots of very interesting stuff on memory there but also a lot of other stuff too - you have to sift through it for the gems.

[Jan Finster]

5 hours ago, imron said:

On 4/2/2021 at 11:03 AM, Jan Finster said:

@Imron, please do a longer interview (like 2-3 hours)

Happy to do longer interviews, but it's really up to the host more than me.

 

Unfortunately, most language interviews on Youtube are between Youtube "polyglots", who try sell their products and trade their fan bases. 

Olle Linge had a good idea when he asked several advanced learners about their method (https://www.hackingchinese.com/asking-the-experts-how-to-learn-chinese-grammar/), but he never took such interviews to the next level. Here is hoping someone will pick up on this and do some longer podcasts. Maybe Olly Richards (https://www.youtube.com/channel/UCSSLq4KYuztsj6ch2RbqoIg) (?)