|
|
|
|
|
by andymatuschak
1996 days 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! |
|
-----
For reference, here's the next new card for one of the books I'm reading:
(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.pd...