Anki and Comprehensible Input: Incompatible?

On a strict reading of the input hypothesis, flashcards are a waste of an evening. Krashen’s argument separates acquisition from conscious learning and holds that the second does not turn into the first; a deck of cards is conscious learning in its purest form, stripped of context, of story and of anything resembling a message. If you believe the theory as written, spaced repetition is at best a distraction.

The objection is worth taking seriously rather than waving away, not least because there is a real failure behind it. Anyone who has spent time in language forums has met the learner who does forty-five minutes of reviews a day, has a deck of nine thousand cards, tracks a streak measured in years, and has never finished a book. That person is not learning a language. They are maintaining a database.

And yet the answer to the question in the title is no, they are not incompatible — provided the deck is subordinate to the reading in a way that can be stated precisely.

What spaced repetition is actually for

It helps to be modest about what an SRS does. It does not teach you a word. It does not give you collocation, register, or the feel for which of two synonyms belongs in a sentence, and nobody has ever acquired a language from a deck.

What it does, and does better than anything else available, is keep a form-meaning link alive across an interval in which you would otherwise have forgotten it. That is a narrow function, and it happens to be the exact shape of the main weakness of learning from text.

Words are acquired from reading over a series of encounters spread across many pages. For frequent words the gaps between encounters are short and nothing leaks. For the mid-frequency words — the ones you meet once a fortnight — the gap is long enough that each meeting starts nearly from zero, and the series never completes. A review card does not replace those encounters. It carries the word from one to the next, so that the fourth meeting builds on the third instead of repeating it.

Stated that way, the deck is not a rival to the input. It is scaffolding between two pieces of input.

The rule that resolves the argument

Nothing goes into the deck that you did not meet in something you were reading.

Almost every objection to flashcards is really an objection to decks assembled from somewhere else: a shared deck of the commonest six thousand words, a textbook’s unit list, a word that looked interesting in a dictionary. Those cards are upstream of the reading and they compete with it. A card made from a sentence in the story you are reading is downstream of it, and it competes with nothing.

The rule also acts as a filter you do not have to think about. Words you meet are words that occur; words that occur are worth keeping; and the ones you keep meeting will arrive at the deck on their own.

Card design

This is where most of the difference between a useful deck and a bad one lies.

Use the sentence, not the word. A card that shows a bare word and asks for a translation trains a link between two dictionary entries. A card that shows the sentence you actually met — with the target word removed, or emphasised — asks you to retrieve the word in the place it lives, and carries its grammar, its company and the right one of its several meanings along with it. It is also, in miniature, comprehensible input: a short passage you understand, read attentively, several times over weeks.

One unknown per card. The sentence on the front should contain exactly one thing you do not know. A card with two unfamiliar words is a card you will fail for reasons you cannot diagnose, and it will become a leech. If the sentence you met is too hard, shorten it or take a different one.

Keep it short. Ten to fifteen words. Long sentences are pleasant to make and punishing to review at eight o’clock on a Wednesday.

Put the formal detail on the card. This is where a deck earns its place, because it covers precisely what reading does not make salient: the gender of a noun, the aspect partner of a verb, the stress, an irregular plural, the case or preposition a verb governs. If the language has those, the card should show them; the page never will.

Add audio if you have it. One line of a narration attached to the card fixes the pronunciation of a word you would otherwise only ever have seen.

Note where it came from. The story, the author, the page. It costs a second and it turns an item into a memory, which is worth more for retention than any amount of algorithm.

The ten-word rule

Cap new cards at about ten a day, and treat the cap as the most important setting in the program.

The reason is arithmetic. In a spaced system, every new card generates a long tail of reviews — commonly something like ten or twelve over the following months. Ten new cards a day settles at roughly a hundred to a hundred and fifty reviews a day, which is fifteen to twenty minutes. Thirty new cards a day settles at four hundred or more, which is the better part of an hour, every day, indefinitely, and it arrives about six weeks after you decided to be ambitious. The people who abandon their decks are almost always people who set the number too high in the first fortnight.

Ten a day is also more than enough. It is over three thousand words a year of deliberate work, sitting on top of everything the reading is doing anyway. Nobody needs to add words faster than that, and most weeks you will not have ten worth adding.

The time budget

One rule, and it decides every borderline case:

Review time must always be the smaller half of your language time.

If you read for forty minutes, ten on cards is right. If you have twenty minutes today, read for twenty and skip the cards. A deck exists to make reading more productive, so a deck that consumes reading time has failed at its only purpose, and the correct response is to cut it — fewer new cards, or a smaller deck — rather than to find more time.

When the two are in genuine conflict, read. This is not a moral position; it is simply which of the two activities produces language.

What to add, and what to leave out

• Add: the word that has resisted. Three encounters and it still will not stick. This is the classic case and the strongest one: the deck is finishing a job the reading started.
• Add: words carrying the book you are reading now. Recurring in this text, useful this month, and you will meet them again within days.
• Add: the formal detail. Gender, aspect, stress, government — the things the page shows and does not teach.
• Leave out: words you met once and did not care about. If it matters it will return.
• Leave out: anything you could have guessed. A card for a transparent cognate is a card that will interrupt you eleven times over two years.
• Leave out: proper nouns, and words that are rare in the language. The interesting archaic term you enjoyed is not worth a place in a queue.
• Leave out: whole sentences you did not understand. Reviewing an unparsed sentence teaches an unparsed sentence.

Maintenance

A deck needs pruning more than it needs additions, and the two habits worth having are both about deletion.

Suspend the leeches. A card you have failed eight times is telling you something true: that word is not ready, and no number of further reviews will make it ready. Suspend it and let it come back through the reading, where it will arrive with a context attached. Learners treat this as a defeat; it is ordinary maintenance.

Declare bankruptcy after a break. Two weeks away and you return to nine hundred due cards. Grinding through them is the surest way to stop reading for a month. Reset the deck, or delete the backlog outright and keep the cards you can name from memory. Nothing of value is lost — words that mattered will be met again.

The lighter alternatives

Spaced repetition does not require software, and for many readers something smaller is a better fit.

• A notebook with sentences in it. Copy the sentence you met, with the date. Reread the last two pages of the notebook once a week. This is a crude spacing schedule and it captures a surprising proportion of the benefit.
• The margin dot. Mark rather than look up, and deal with the marks at the end of the chapter. Most of them will have resolved themselves.
• Rereading the story. The purest version: a second pass over a text you know is spaced repetition with the context still attached, and it reviews everything in the story rather than the dozen words you happened to write down.

Which words are worth this treatment in the first place is the substance of what reading does and does not teach, and the related question of whether to start from a ranked list at all is answered separately. For where the review sits inside a week of reading, see the complete method.