Beat the Forgetting Curve: Why Recall Beats Re-Reading

What Ebbinghaus found

In 1885, Hermann Ebbinghaus memorized lists of nonsense syllables and tested himself at intervals. The pattern he uncovered, which modern researchers have replicated almost exactly, is harsh:

• After 20 minutes: ~58% retained
• After 1 hour: ~44% retained
• After 9 hours: ~36% retained
• After 1 day: ~33% retained
• After 6 days: ~25% retained

Without intervention, you keep about a quarter of new material a week later.

The retrieval-practice escape hatch

What Ebbinghaus didn't test, but Karpicke & Roediger demonstrated definitively in 2008: the act of retrieving a memory dramatically strengthens it. Reading something again is much weaker than recalling it.

Their experiment had students read a passage and then either re-read it 4 times or read it once and recall it 3 times. A week later, the recall group remembered 50% more.

This is why "looking through your flashcards" is not the same as "doing your flashcards." Reading is recognition. Recall is the workout.

What this means in practice

Three operating principles:

1. Test yourself, don't review. Recall before you re-read.
2. Space the recall. A failed recall followed by re-encoding is worth more than five passive re-reads.
3. Sleep on it. Memory consolidation during sleep is itself a form of free retrieval practice (your hippocampus replays the day).

The compound rule for vocabulary

If you encounter a new word today, you should re-encounter it (in retrieval, not recognition):

• Within 24 hours
• Within 3–4 days
• Within 9–10 days
• Within 3 weeks
• Within 2 months

If those touches happen, the word is yours for the long haul. If even one is missed by a week, you'll need to start partly over.

This is a practical adaptation of the graduated-interval pattern (Pimsleur 1967 used finer intervals starting at 5 seconds; modern SRS implementations like SuperMemo and Anki tune their schedules adaptively from the same idea).

Dia's vocabulary engine works alongside that pattern from a different angle: words you've already used in past chats keep coming back into Dia's context the next time you talk, so retrieval keeps getting tested in real conversation rather than on a card. (The full forgetting-curve schedule, 1d / 3d / 10d / 3w / 2mo, is on the roadmap; today the engine prioritizes by usage and recency, which is why it pairs so well with a dedicated SRS app on the side.)

You don't need to be disciplined about it. You need a system that's disciplined for you.