You probably already thought about this in case not here is some ideas:
When training words slowly increase the mastery level of the components and characters in that word. The same applies to the components when learning characters.
Right now I have lessons were I first do a bunch of words and then get the component and character lessons that make up those words. I know the characters because I already succeeded the words.
It seems to me that a word can be treated as a test for the characters and components. Or at least contribute to the mastery upon successfully guessing the pronunciation. We would skip them from the current lesson or perhaps the SRS could slowly increase mastery of components/characters that were indirectly trained during every session.
As much as I’d love to reduce the daily review load (pls implement FSRS), I don’t think this it. I’ve been using HH <2 months and already have around 300 of the ~500 components with less and less of them showing up in my reviews each day. Just stick with it for a little while and you’ll stop seeing them altogether.
Edit: just realized you brought up words for characters as well. Thats a really good idea. My review mistakes consist almost entirely of words so if I get the word its pretty much a given Ill get the characters
Yeah concretely it could consist of those usecases:
have characters contribute to the component learning mastery
have characters contribute to the sounds
have words contribute to characters
Contribution can be either:
A direct removal from the current daily desk,
or just add mastery points so they will be slowly pushed back in the backlog.
The same would apply in reverse where mistakes contribute to lower mastery.
Perhaps this solution can be implemented after a certain tasks reaches a certain level of mastery. E.g in the beginning HH may need to reinforce the item but over time the above contributions could be applied
I was thinking about how the issue this could cause with components, because eventually you memorize the word itself and can easily forget components by the time another character uses it. Theres a few different ways that could be handled. (like asking if you remembered the component, handling more abstract components differently, etc)
Ultimately I do think the review process could be significantly more efficient. I started after SRS implementation though so Im sure theyve made great progress in this regard. I just hope it continues to be a priority bc theres definitely room for improvement and review load has such a large impact on the user experience
Yeah, review load is definitely an area we would want to help out with somehow, whether that be making it more painless or less volume.
Grading items in a linked-together manner sounds like an interesting way to go about reducing the volume
I can think of a few questions/edgecases, like what if you review the character and then later get the word correct in a later session.
I guess the calculation to add onto the character “mastery points” could be something like todays_date - character.last_reviewed
What makes this “card-link” idea interesting is that not a lot of SRS system account for the relationships between cards. Usually every card is its own data point, even if it’s building off of previous cards
Yeah, to me the mastery points meant the input into the SRS system. Items can still be their own, you just have to fetch the individual characters and then add/remove to/from the score like you usually do when guessing correct/wrong. You can start of conservatively, like giving it 1/5th of the score of the current correct/wrong guess. So if you have 5 words with that character you will level up to the next level with that character. If you can have some stats on the review load one can observe the effect this has, if no big effect one can increase it or decrease it when it is to drastic. It requires some data analysis to find the right value for this, but perhaps could be a setting to so ppl can change it if they feel comfortable, but I think the main thing here is that we want to preserve the UX and decent learning rates, if we increase the learning rate to much may be negative.
