vhrrsrivr
Joined 1 year ago
Karma
4
Submissions
3
Member since
Aug 5
Recent submissions
Comment · 1 year ago
Comment · 1 year ago
1. yes that's quite doable. the placement test only gets you to see a mix of basic and advanced vocab. without importing learning history from another platform you do have to see everything at least a few times eventually, easy or not.
2. this is more challenging as there's very often not a 1-to-1 relationship between grammar points.
3. I have a branch on the hsrs github that changes the sampling to be prefix-order so an llm can guide it, with mixed results. There's a tension between picking common outputs, and picking the output that will maximize your increase in retention across multiple cards. That being said 色んな娘 is definitely me forgetting to tag 娘 as non-attributive (like pronouns), will fix. you can read about the mechanisms I have to keep the content as natural as possible here: https://github.com/satchelspencer/hsrs/blob/main/docs/deck-c...
Comment · 1 year ago
Rather than relying on an embedding space, my approach is to have the cards themselves be grammars that can define the relationships between concepts explicitly. Then the problem becomes what specific sampling of all the possible outputs is optimal for a learner to see at any given time, given their knowledge state.
See how it's applied to Japanese learning here: https://elldev.com/feed/grsly
however, individual grammar outputs aren't their own cards, you get a fresh example every time you see a card. this requires a very different scheduling approach, since you have to estimate how all the cards in the 'call tree' contribute to the overall result and reschedule them as well https://github.com/satchelspencer/hsrs/blob/main/docs/overvi...