Spaced repetition memory prompts should be written to discourage shallow “pattern matching”

One challenge for the efficacy of a Spaced repetition memory system is that sometimes users memorize answers to questions shallowly, through “pattern matching,” rather than by integrating the knowledge more deeply or actually thinking about the question. But Spaced repetition memory prompts should ensure reviewers must retrieve answers from memory.

For example, in QCVC, there’s only one “who”-type question, i.e. one which asks for a person’s name. After a few repetitions, readers may learn that whenever they see a “who”-type question about quantum computing, the answer is John Preskill. But that shallow knowledge won’t serve them well in real creative work.

One approach to solving this problem is to construct variations on the question (with the same answer), making pattern matching more difficult.

A different but related approach is to encode the information from many different angles: not just different questions with the same answer, but different question/answer pairs illustrating how the information fits into a broader context (i.e. Spaced repetition memory prompts should connect and relate ideas). For instance, if you added some questions about the specifics of what John Preskill has discovered, why it matters, and how his work relates to others’, his name would be attached to a richer set of knowledge. Arguably, the John Preskill example in QCVC is an “orphan prompt” (Avoid orphan spaced repetition memory prompts). Such prompts seem more susceptible to this phenomenon; adding more connections would make it no longer an orphan.

Another approach is to prompt users to use their knowledge flexibly in novel contexts (e.g. The mnemonic medium can help readers apply what they’ve learned through simple application prompts). I’m not quite sure how you could do this for the John Preskill example, but it could work better for memorized constants.

I’ve noticed in my own practice that shorter prompts are somewhat less susceptible to this problem, perhaps because there’s less detail to pattern match on. This is another reason Spaced repetition memory prompts should be concise.

Is there a way to detect spaced repetition memory prompts which are likely to be answered through pattern matching?


Q. How does SRS prompt connectivity relate to pattern matching?
A. Low-connectivity prompts (especially “orphans”) are more likely to be pattern matched.

Q. How might application prompts help with SRS pattern matching?
A. They make you use your knowledge in new contexts.

Q. How does SRS prompt length relate to pattern matching?
A. Shorter prompts seem harder to pattern match.


References

We’ve received many comments about this phenomenon from Quantum Country readers. e.g.:

For some of the review questions in this essay, i know the answers by rote now due to the number of times I’ve seen those questions but I can’t say I understand the algorithm to the extent I would prefer (such that I could explain it to somebody else). I’m not sure if I need to read this essay many more times to fully grasp the algorithm.

— Srinath K R Re: Hello from Quantum Country!

Last updated 2023-07-13.