7 ms·

I find that writing spaced-repetition prompts in the way the article describes to have too small of a return on investment: Reviewing them is quick, but it take

by kd5bjo 6y ago

I find that writing spaced-repetition prompts in the way the article describes to have too small of a return on investment: Reviewing them is quick, but it takes a lot of effort to craft good ones from a text. Good authors take care with their prose and use nuance to get their point across, which is often lost in the translation to a few bullet points. This subtlety is often the meat of the text, and the high-level points are the bones that hold it all together: essential, but not where the real value is. My Anki deck is mostly cloze deletions automatically extracted from ebooks, where I've imported the entire text ~20 lines at a time. I've deliberately set up the review schedule so that it's hard for me to memorize particular phrases: 1 week for the first review and 4 weeks for the second, with a 1 day delay for misses. This forces me to remember what role the passage plays in the wider narrative (argument for non-fiction) to successfully predict the blanked-out word. With this setup, I'm noticing new connections between different sections of the same book, and between different books, every time a card comes up for review: My understanding of the world has changed since I last saw it, and so I notice different subtleties than I did the first time around. Instead of viewing them as a chore, I'm now actually excited to do my Anki reviews every morning-- where they used to make me mentally exhausted, they're now a warm-up that gets my brain started for the day.

hyperpallium2 6y ago
> it takes a lot of effort to craft good ones Though this engagement itself aids memorization... I'm not a fan of spaced repetition, instead preferring to understand material and find connections (associations) integrating it within my existing knowledge. This makes it usable, not just recallable. It's holographic, in that I can also recover it from other things I know, like the shape of the missing jigsaw piece, or a theorem or proof. However, I've come to realize that the very act of engagement to understand is, in itself, aiding memorization.
- kd5bjo 6y ago
  A key insight for me was when I realized that spaced repetition isn't really about memorization, it's about skill maintenance. The skill most people use formal spaced-repetition methods for is factual recall, to be sure, but the underlying principles can be applied to any human endeavour. My Anki deck, for example, is designed to exercise reading comprehension. For cooking, I don't try to memorize recipes like the article demonstrates; I do, however, try to revisit and make the dishes in my repertoire semi-regularly to reinforce the operational cues that are impossible to put into words. It's similar for all of my other hobbies: I naturally tend to have a period of intense engagement with them at first that trails off as my confidence builds. I seem to revisit most of them on the same sort of exponential schedule without really trying, as if my subconscious understands when it's time to have a refresher.
edylemond 6y ago
Really interesting how you extract cloze deletions from ebooks! It sounds very similar to what I'm building with Traverse.link, a social spaced repetition app with connected cards. You can import an article (working on ebook support), and it'll turn the paragraphs into cards, to which you can then add cloze deletions and free form prompts.
andymatuschak 6y ago
(author of parent article here) This is very interesting! Can you say more about the task you perform when you see these prompts? How do you decide how to "grade" your response? Presumably it's not just whether or not you can remember the deleted word(s)? Also: are you blanking out individual words? Phrases? Sentences? An example would be interesting if you're willing to share!
- kd5bjo 6y ago
  Edit: Though my examples below are all fiction, that's half or less of my current reading list. I also have philosophy, research papers, long-form academic books, popular science books, and the like. ----- For reference, here's the next new card for one of the books I'm reading: Bleak House Charles Dickens CHAPTER III --------------------------------------------------------------------- coach gave me a terrible start. It said, "What the de-vil are you crying for?" I was so frightened that I lost my voice and could only answer in a whisper, "Me, sir?" For of course I knew it must have been the gentleman in the quantity of wrappings, though he was still looking out of his window. "Yes, you," he said, turning round. "I didn't know I was crying, sir," I faltered. "But you are!" said the gentleman. "Look here!" He came quite opposite to me from the other corner of the coach, brushed one of his [...] furry cuffs across my eyes (but without hurting me), and showed me that it was wet. "There! Now you know you are," he said. "Don't you?" "Yes, sir," I said. "And what are you crying for?" said the gentleman, "Don't you want to go there?" (The prior cards include the text up to and including the line starting "But you are!") My response, having never seen the passage before, was "coat's"; the correct word is "large". The substitution doesn't meaningfully alter the passage: The only thing we know about the gentleman at this point is that he's a stranger whom Esther has just met in the coach, wearing a large coat (cloak?). This is the first time he speaks in the book. I selected "again" because this is a new card; if this had been a review, I might have deemed it acceptable. The bar for similarity varies depending on the book the passage came from, my mood on any particular day, and how much I care about the contents of the passage. The Rime of the Ancient Mariner, being poetry, doesn't get any latitude: I must remember the word exactly. Some of the tedious battle descriptions in Le Morte d'Arthur get a pass no matter how badly they go wrong. If a card is causing me trouble, there are a few corrective actions I can take: In some instances I'll edit the card to blank out a more meaningful word than the one randomly selected, which is an easier and more engaging review task. If I can't figure out the blanked word from context, I'll look it up in a dictionary; this happens most often with my second-language texts. And in extreme cases, I have enough coverage of each book that permanently dropping card or two won't cause problems. The idea for this came from the paper[1] that introduced the concept of Cloze deletions. They were originally envisioned as a readability score. Instead of calculating statistics about the words used, like traditional methods, Taylor proposed to blank out every Nth word and measure how many could be guessed correctly by a reader. Readability and comprehension are two sides of the same coin. They both rely on observing the relationship between text and reader. If used to evaluate the text, this is called readability; if to evaluate the reader, comprehension. [1] https://www.gwern.net/docs/psychology/writing/1953-taylor.pdf https://www.gwern.net/docs/psychology/writing/1953-taylor.pd...

by kd5bjo 6y ago

hyperpallium2 6y ago

> it takes a lot of effort to craft good ones Though this engagement itself aids memorization... I'm not a fan of spaced repetition, instead preferring to understand material and find connections (associations) integrating it within my existing knowledge. This makes it usable, not just recallable. It's holographic, in that I can also recover it from other things I know, like the shape of the missing jigsaw piece, or a theorem or proof. However, I've come to realize that the very act of engagement to understand is, in itself, aiding memorization.

kd5bjo 6y ago
A key insight for me was when I realized that spaced repetition isn't really about memorization, it's about skill maintenance. The skill most people use formal spaced-repetition methods for is factual recall, to be sure, but the underlying principles can be applied to any human endeavour. My Anki deck, for example, is designed to exercise reading comprehension. For cooking, I don't try to memorize recipes like the article demonstrates; I do, however, try to revisit and make the dishes in my repertoire semi-regularly to reinforce the operational cues that are impossible to put into words. It's similar for all of my other hobbies: I naturally tend to have a period of intense engagement with them at first that trails off as my confidence builds. I seem to revisit most of them on the same sort of exponential schedule without really trying, as if my subconscious understands when it's time to have a refresher.

edylemond 6y ago

Really interesting how you extract cloze deletions from ebooks! It sounds very similar to what I'm building with Traverse.link, a social spaced repetition app with connected cards. You can import an article (working on ebook support), and it'll turn the paragraphs into cards, to which you can then add cloze deletions and free form prompts.

andymatuschak 6y ago

(author of parent article here) This is very interesting! Can you say more about the task you perform when you see these prompts? How do you decide how to "grade" your response? Presumably it's not just whether or not you can remember the deleted word(s)? Also: are you blanking out individual words? Phrases? Sentences? An example would be interesting if you're willing to share!

kd5bjo 6y ago
Edit: Though my examples below are all fiction, that's half or less of my current reading list. I also have philosophy, research papers, long-form academic books, popular science books, and the like. ----- For reference, here's the next new card for one of the books I'm reading: Bleak House Charles Dickens CHAPTER III --------------------------------------------------------------------- coach gave me a terrible start. It said, "What the de-vil are you crying for?" I was so frightened that I lost my voice and could only answer in a whisper, "Me, sir?" For of course I knew it must have been the gentleman in the quantity of wrappings, though he was still looking out of his window. "Yes, you," he said, turning round. "I didn't know I was crying, sir," I faltered. "But you are!" said the gentleman. "Look here!" He came quite opposite to me from the other corner of the coach, brushed one of his [...] furry cuffs across my eyes (but without hurting me), and showed me that it was wet. "There! Now you know you are," he said. "Don't you?" "Yes, sir," I said. "And what are you crying for?" said the gentleman, "Don't you want to go there?" (The prior cards include the text up to and including the line starting "But you are!") My response, having never seen the passage before, was "coat's"; the correct word is "large". The substitution doesn't meaningfully alter the passage: The only thing we know about the gentleman at this point is that he's a stranger whom Esther has just met in the coach, wearing a large coat (cloak?). This is the first time he speaks in the book. I selected "again" because this is a new card; if this had been a review, I might have deemed it acceptable. The bar for similarity varies depending on the book the passage came from, my mood on any particular day, and how much I care about the contents of the passage. The Rime of the Ancient Mariner, being poetry, doesn't get any latitude: I must remember the word exactly. Some of the tedious battle descriptions in Le Morte d'Arthur get a pass no matter how badly they go wrong. If a card is causing me trouble, there are a few corrective actions I can take: In some instances I'll edit the card to blank out a more meaningful word than the one randomly selected, which is an easier and more engaging review task. If I can't figure out the blanked word from context, I'll look it up in a dictionary; this happens most often with my second-language texts. And in extreme cases, I have enough coverage of each book that permanently dropping card or two won't cause problems. The idea for this came from the paper[1] that introduced the concept of Cloze deletions. They were originally envisioned as a readability score. Instead of calculating statistics about the words used, like traditional methods, Taylor proposed to blank out every Nth word and measure how many could be guessed correctly by a reader. Readability and comprehension are two sides of the same coin. They both rely on observing the relationship between text and reader. If used to evaluate the text, this is called readability; if to evaluate the reader, comprehension. [1] https://www.gwern.net/docs/psychology/writing/1953-taylor.pdf https://www.gwern.net/docs/psychology/writing/1953-taylor.pd...