When to introduce Characters + Words

[kaysik]

Is there any data from HH on what’s better between unlock when learnt or familiar?

When I’ve been using the unlock when learned (so get related items straight away), you get loads of reinforcement right off the bad for any new thing. This works great up front and seems to get things to stick in my short term memory really well since everything is showing up together (components, characters, words all on the same day). Then for the first week or two they keep showing up together and it works great. But once the interval expands to longer than say a week or two, they reviews all tightly packed time wise so instead of getting a slow stream of reinforcement, it’s famine or floor. You might see nothing about a character for a month, then like 5 directly related reviews all show up at once. Maybe you fail the first one, but that jogs your memory so you get the other 4 right. And there is still a pretty big cluster of almost identical reviews that come around which I don’t like.

The flip side is the unlock when familiar. Initially this is much harder to get things to stick for me since items are learnt in isolation - learn a component but have no reason to remember it. Then a while later learn a character that relates to nothing, then later still learn a word etc. I seem to forget these items much more readily. However if I power through, force it, then once I’m actually feeling ok about it the reviews are all really evenly spread out, potentially over quite a wide time span so there is no big clumb of almost identical reviews turning up on the same day which I love. The spread out nature of introducing new things only when the related item is “learnt” leads to a very nice tail.

So I want the initial burst of relatedness from unlock when familiar, but the tail of unlock when learnt. Essentially I can’t decide what works best, so wondering if there is any data to back up one side or the other.

Has anyone ever looked over the review history and seen which version work better overall? Just curious if there is a “known good” setting.

[phil]

@kaysik

Is there any data from HH on what’s better between unlock when learnt or familiar?

This is something we were hoping to explore There are strong arguments on both sides.

I’m personally a fan of getting everything right away, as the cardinal rule in memory is the more associations = the more likely you’ll remember.

@kaysik

But once the interval expands to longer than say a week or two, they reviews all tightly packed time wise so instead of getting a slow stream of reinforcement, it’s famine or floor.

We could probably bump the “fuzz factor” of the algorithm even more to prevent these sort of cases. Do you think that’d help?

[kaysik]

@phil

We could probably bump the “fuzz factor” of the algorithm even more to prevent these sort of cases. Do you think that’d help?

That would probably help yes.

The other thing I’ve thought about is the penalty for getting stuff wrong could be stronger and a non-factor, something that really splits groups of related items up. Currently if you get one of a group wrong it halves the interval, but then get it right and it’s still likely to get back in sync quite often - if the penalty was something obscure like 23% of original interval instead of half, then it’s not a nice factor of the original interval. As you start getting it right again doubling it doesn’t regularly put it back in sync with the main group. With this, any time you get something wrong it’s permanently separated from any card around it, never to nicely sync up regularly.