Spaced Repetition + Conversation: Why Both Beat Either Alone

The forgetting curve, revisited

Hermann Ebbinghaus's original 1885 self-experiment found that without re-encountering material, we forget most of it within a day. His exact 70% figure was for one subject (himself) memorizing nonsense syllables, but Murre & Dros (2015) replicated the curve almost exactly with modern controls. The shape — fast loss in the first 24 hours, then a long tail — has held up.

The fix is spaced repetition: reviewing material at expanding intervals (1 day, 3 days, 7 days, 21 days). Cepeda et al.'s 2006 meta-analysis put the effect size of spaced over massed practice at roughly Cohen's d ≈ 0.6 over long retention intervals — in plain English, students who spaced their study remembered roughly half again to twice as much a month later compared to students who crammed the same total time. Their 2008 follow-up sharpened the recipe: the optimal gap between reviews is roughly 10–20% of the time until the test. (For words you'll need next month, that means review them every 5–10 days. For words you'll need in a year, every month or so. Most apps default to whatever Anki ships with and don't tune for this.)

Why SRS alone is not enough

The reason most people quit flashcards isn't that the method doesn't work. It's that it's lonely. Words divorced from any conversation you ever had attach to nothing in your life, and the loneliness eventually beats the discipline.

Beyond that, there's a structural limit. Spaced-repetition systems (SRS) like Anki and Memrise solve recognition: shown a word, you recall its meaning. They do not, by themselves, train production — pulling the word out of your head while you're trying to make a point. That gap is what SLA researchers call automaticity (Segalowitz 2010): the difference between knowing a word and being able to say it without thinking. Pimsleur's 1967 graduated-interval method tried to train both at once; most modern flashcard apps quietly dropped the production half.

The compound effect: SRS + conversation

Conversation does three things flashcards cannot:

1. Trains automatic retrieval under pressure. This is the difference between "knowing" a word and being able to say it in the half-second a real conversation gives you.
2. Encodes contextual meaning. You learn that "kind of" hedges and "rather" intensifies — nuance flashcards can't capture (lexical priming, Hoey 2005).
3. Generates emotional valence. Words used in real moments are tagged with feelings, which improves recall in line with the broader literature on affect and memory (Kensinger 2009).

The cleanest evidence for retrieval-plus-use vocabulary methods comes from Nakata (2008) and Nakata & Webb (2016): tasks that combine spaced retrieval with active production produce stronger long-term retention than spaced retrieval alone. (Intensive immersion programs like Middlebury and DLI also show fast B2 attainment, but they're mostly evidence for high contact-hours and feedback, not for SRS specifically — different mechanism, often misattributed.)

What this means for TalkToDia users

When you talk to Dia, every word you and Dia use is tracked in your personal language model. The system ranks them by your engagement and threads the most active ones back into your next conversation, so the words you've already met keep getting retested in real dialogue. That's the loop that closes the gap between "I recognize this word" and "I can use this word." (Calendar-style spaced-repetition scheduling — the kind Anki does — is on the roadmap; today the engine prioritizes by usage and recency.)

You don't have to grind 200 cards a day. You have to converse — and let the system keep the words alive.