Spaced Repetition for Travel, Not for Life

Hermann Ebbinghaus spent the 1880s memorizing nonsense syllables and timing how fast he forgot them, and the curve he drew is still the one every spaced-repetition app is quietly built around: you forget fast at first, then slower, and a review at the right moment resets the clock and flattens the curve a little more each time. It’s a genuinely good idea. It is also, almost universally, implemented for the wrong horizon.

The algorithm doesn’t know you have a flight

Open a mainstream flashcard app and add “La cuenta, por favor.” It shows you the card, you get it right, and it schedules the next review for tomorrow. Get it right again, and the gap grows — three days, then a week, then two weeks, then a month. This is correct behavior if your goal is remembering the phrase in a year. It is close to useless if your trip to Spain is in sixteen days, because the schedule is optimizing for a kind of durability you don’t need yet and won’t need again until your next trip, if ever.

The tell is what happens to a phrase you’ll actually use on day one of the trip — say, “¿Cómo está usted?” — once the app decides you know it. The review gets pushed a month out. You board your flight in the gap, arrive with a phrase your own app has scheduled to reintroduce after you’re already home. The algorithm did its job. It just had the wrong deadline.

What the schedule should actually look like

A travel-shaped review schedule doesn’t stretch outward toward infinity — it compresses toward a date. Take the same phrase, but built for a countdown instead of a calendar:

Day	What happens
Day 1	First exposure — see it, hear it, produce it once
Day 2	Review — the fastest-forgetting window is in the first 24-48 hours
Day 4	Review, now mixed with two other phrases from the same scenario
Day 7	Review inside a short dialogue, not as an isolated card
Day 12	Review under a slight twist — a different name, a different order
Day 18-19	Final pass, two to three days before departure
Day 20-21	Trip. The phrase needs to survive contact with a real stranger, not a screen.

Notice the intervals shrink rather than stretch as the trip gets closer, and the last two reviews happen right before the deadline that actually matters, not on a schedule indifferent to it. That’s the entire adjustment: same forgetting curve, same underlying math, but the target isn’t “remember this indefinitely” — it’s “have this ready on the specific Tuesday you land in Athens.” Nineteen days after that, if you forget “τον λογαριασμό παρακαλώ” entirely, the schedule did exactly what it was supposed to do.

Recognition isn’t the skill you’re testing for

There’s a second problem, and it’s arguably the bigger one: most spaced-repetition apps test recognition, not production. You see “Πώς είστε;” and tap the English translation from a list of four. That’s a real memory operation, but it isn’t the one a waiter is going to ask of you. Nobody abroad hands you a multiple-choice card. They ask a question out loud, at a normal pace, once, and you either produce a response or you don’t.

This is why a phrase can feel “known” — green checkmark, long interval, app satisfied — and still fall apart the moment it’s needed. Recognition and production draw on overlapping but different retrieval paths, and only one of them gets exercised by tapping the right box. A schedule built for travel has to test the harder one: say the line out loud, from nothing, in response to something unpredictable, the way “Da dove vieni?” actually arrives — as a question from a stranger who wants an answer, not as a card waiting for your tap.

That’s also where the emotional stakes of travel do something a semester-long course doesn’t have to account for: a mistake in a classroom costs you nothing, but a mistake with a border officer, a pharmacist, or a taxi driver at 11pm costs something real. Phrases tied to higher-stakes scenarios — “Aiuto!”, “Βοήθεια!”, “助けて！” — deserve tighter, more frequent review than a phrase like “where are you from,” not because they’re harder to remember, but because the cost of a blank moment is so much higher when it happens.

Weighting by what a mistake actually costs

A generic algorithm treats every card as equidistant from disaster. A travel-shaped one shouldn’t. Getting “grazie mille” wrong costs you an awkward half-second. Getting the emergency phrase wrong costs you the one moment you actually needed it. The review schedule can — and should — spend more of its limited budget on the second kind of phrase, even if you’ve technically “known” it since day one, because the cost of forgetting isn’t symmetric across a phrase list the way most flashcard decks assume it is.

The same logic applies to register, not just vocabulary. The seed material behind this site’s Japanese phrases stays consistently in the polite -masu/desu form — “お会計をお願いします”, not a casual shortcut — because getting formality wrong in Japan reads very differently than getting it wrong in, say, a Zagreb café where “Bok” to a stranger barely registers as informal. A spaced-repetition schedule that doesn’t know this treats every phrase as equally forgivable. One that does spaces the higher-stakes register choices more tightly and lets the low-stakes ones drift.

What this actually asks of a language app

None of this requires new math. Ebbinghaus’s curve is still the right curve. What it requires is building the schedule around a date instead of a mission to remember forever, testing production instead of recognition, and weighting review frequency by consequence instead of treating a coffee order and a call for help as the same kind of card. A three-week trip doesn’t need a system that would serve you well in year four. It needs one that gets “La cuenta, por favor” out of your mouth correctly on a specific Tuesday, in a specific restaurant, without a pause long enough for the waiter to switch to English first.