Verb Conjugations: Stop carding the whole table
How to memorize verb conjugations without carding six forms at once. Card the ending pattern once, then only the forms the pattern gets wrong.
The card I made in grade 11 French looked like this: front, parler (present). Back, all six forms stacked in a little block. I made about forty of those over a weekend and felt organized. Then every review turned into the same ritual: read the block, nod, press Good, learn nothing.
Here is the short version of how to memorize verb conjugations. Card the ending pattern once, for the whole class of verbs. Then make cards only for the forms that the pattern gets wrong. For a regular verb that is zero extra cards. For Spanish tener it is two. For French être it is all six, and even then they should not sit on one card.
A little about me: I'm Daniel, third year engineering science at UofT, machine intelligence. I work on Nebulearn, a flashcard app, so treat everything I say about tools as a conflict of interest and read the honest exceptions near the bottom. The method here works fine in Anki, in a notebook, or on paper.
The reason the stacked table fails is not that it is lazy. It is that it hides two very different problems inside one object, and your scheduler cannot tell them apart.
One table on one card grades you wrong
A conjugation table is six facts wearing one coat. Je parle and nous parlons are separate retrievals: separate cue, separate answer, separate difficulty. When you put all six on one back, you are asking your memory one question and grading it on six answers at once.
Watch what happens when you miss exactly one form. You get nous tenemos right, you get the yo form wrong. Now you have two bad options. Press Again, and tomorrow you re-prove five forms you already own, which is where most of the review minutes go. Press Good, because you knew "most of it," and the one form you actually cannot produce quietly gets a three week interval. Neither button is telling the truth, and both of them are your scheduler's only input.
This is the atomic encoding rule applied to grammar, and grammar is where it bites hardest, because a table is typeset as one block, so we card it as one block. A spaced repetition system schedules per card, not per fact, so the block is the unit it gets to reason about. Six facts on one card means six facts on one interval, which is the interval of whichever form is currently your worst.
So the first cut is to stop treating the table as the thing being learned. Most of it is not a fact at all.
Card the ending pattern once, not once per verb
Almost every table you are about to card is mostly the same table you already carded. For French -er verbs in the present, the endings are -e, -es, -e, -ons, -ez, -ent, and that is true for parler, regarder, donner, aimer, chercher, and a few thousand others. Once you know the pattern and you know the stem, you do not need the table. You need to be able to run it.
So make it one card, once, per verb class and tense. Front: "French present tense, -er verbs, six endings." Back: the endings in order. That card is worth carding because the pattern genuinely is information.
Then add one card per verb that gives you the stem and the meaning, which is the vocabulary card you were going to make anyway. chercher, to look for. Nothing about cherchons needs its own card, because you can derive it in under a second from two things you already have.
Do the same thing with the predictable spelling adjustments instead of carding them verb by verb. Nous commençons and nous mangeons are not two facts about two verbs. They are one fact about French orthography: a c or g before the -ons ending has to keep its soft sound. One card, and it retires the -cer and -ger verbs forever. Spanish preterite endings for -ar verbs are one card that covers hablé, hablaste, habló, hablamos, hablasteis, hablaron and every other regular -ar verb in the language.
Once the patterns are carded, you can look at any new verb and ask a much smaller question: which of these six forms is the pattern going to miss?
Run the surprise test on every new verb
The surprise test takes about fifteen seconds and it decides your entire card count for that verb. Write out what the regular pattern predicts. Compare it to the real table. Card the differences and nothing else.
Spanish tener is the example I always use, because it looks intimidating and it is not. The regular -er present endings give -o, -es, -e, -emos, -éis, -en, so on the stem ten- the pattern predicts teno, tenes, tene, tenemos, tenéis, tenen.
| Person | Pattern predicts | Actual | Needs a card? |
|---|---|---|---|
| yo | teno | tengo | Yes, the -go group |
| tú | tenes | tienes | Yes, e to ie |
| él / ella | tene | tiene | Covered by e to ie |
| nosotros | tenemos | tenemos | No |
| vosotros | tenéis | tenéis | No |
| ellos | tenen | tienen | Covered by e to ie |
Six slots, and two facts. Two of the forms are already correct, so carding them is pure cost. The other four collapse into two patterns that are not even specific to tener: the yo form that ends in -go is shared with poner, salir, venir, traer, caer, oír, and the e to ie stem change in everything except nosotros and vosotros is shared with querer, perder, pensar, entender, sentir. Card the two patterns and you have just done fifteen verbs, not one.
French aller runs the same way and lands somewhere different. It gives vais, vas, va, allons, allez, vont. The nous and vous forms are exactly regular off the stem all-, so they get nothing. The other four are pure surprise, so they get four cards, one per form. Avoir is the same shape: avons and avez behave, while ai, as, a, ont do not.
One more free cut, if your course or your region does not use vosotros. That is a sixth of every Spanish table you will ever see, and if you are never going to be tested on it and never going to hear it in Mexico City, it is not a card. Deciding that once is worth more than any scheduler setting.
The verbs where all six forms really are information
The surprise test cuts hard, and then it runs into a wall. French être is suis, es, est, sommes, êtes, sont. There is no stem and no pattern. Spanish ser is soy, eres, es, somos, sois, son. German sein is bin, bist, ist, sind, seid, sind. Every slot is genuinely unpredictable, so every slot is genuinely a fact.
The temptation here is to shrug and go back to the block, since the whole thing has to be learned anyway. Do not. The argument was never "tables are too big." It was that one card holds one grade, and êtes being solid tells you nothing about whether sommes is. Make six cards. Front: "être, nous, present." Back: "sommes." That is it.
Two details matter on those cards. Prompt in the direction you will be tested in, which is almost always production: English cue plus a person, target form as the answer. A card that shows you sommes and asks what it means trains recognition, and recognition is the skill you already have from reading. And keep the person in the cue explicitly, because "être, present" with no person is the block again in disguise.
If you prefer cloze for this, a card like Nous {{c1::sommes}} en retard. works, and it has the nice side effect of dragging a real sentence along with the form. Just know what you are trading: cloze in a fixed sentence can start cueing off the sentence shape rather than the grammar, which is the failure I wrote about in cloze deletion versus Q and A encoding. For a table of six irregular forms I would rather have six blunt production cards and one or two sentence cards on top.
This is the same shape as carding kanji radicals before whole characters: find the reusable layer, card that once, then card only what the layer cannot generate. Genders work this way too, which is why the phonetic ending groups beat carding der and die one noun at a time.
What breaks, and what to do about it
Three things go wrong with this method, and all three are fixable without abandoning it.
The first is the pattern card you can recite but cannot run. You know the six -er endings cold, and in conversation you still stall on vous cherchez. That is a retrieval problem, not a knowledge problem, and the fix is an application card rather than a rule card: front "chercher, vous, present," back "cherchez," using three or four common verbs as your drill vehicles. You are not learning those verbs. You are training the act of applying the pattern under time pressure.
The second is that you split too far and drowned. Thirty verbs times six forms times four tenses is a number you will not review, and a deck you dread is a deck you skip. The guard is that only surprises get individual cards. If you have carded tenemos, delete it. If a form keeps failing for weeks even as its own card, it is a leech and it probably needs a mnemonic or a sentence, not another interval, which is the rule of three for deleting a card.
The third is the one that looks like success. You can produce every form on demand and you still write the wrong tense in an essay, because choosing a tense from context is a different skill than conjugating. Cards will not fix that on their own. Keep a small number of cards where the cue is a sentence with a gap and a time marker, and do the rest with writing you get corrected on.
Where to build this deck
The mechanics do not care about the app, but some apps make the split easier than others. Anki is still the strongest tool here if you are willing to build templates, because you can generate one card per person from a single note and there are add-ons built specifically for conjugation drilling. If you already have a long streak and a deck you trust, stay there. Review history does not transfer between apps, and rebuilding it is not worth the hour. Quizlet genuinely wins the Friday case: quiz on -ir verbs tomorrow, a classmate already shared the set, you need to see the forms fifty times tonight. Its free Learn mode is scoped to that sitting rather than a per card due calendar, and for tomorrow that is fine. RemNote is the one I would pick if your grammar notes and your cards need to live in the same linked document.
Nebulearn is where I work, and the specific reason it fits this page is the part nobody advertises: deleting cards. Once you run the surprise test you will find dozens of forms you already derive for free, and skip and ignore let you pull those out of the queue without hunting through a deck editor. The other half is that FSRS is on by default, so once the six forms of être are six cards, they get six honest intervals, and sommes keeps coming back while est goes quiet. The free tier is 1,000 saved questions and 40 weekly AI credits, which is more than a semester of conjugation needs once you stop carding forms you can generate. One thing to be blunt about: there is no offline mode, so if you study on a subway with no signal, Anki is the right answer and I am not going to pretend otherwise. I wrote more about that trade in language learning without living in Anki.
The whole argument is one line: a conjugation table is six facts, most of them derivable, and the ones that are not deserve their own card and their own schedule. That is what the tooling should be for. Pattern cards you write once, the derived forms pruned out of the queue instead of quietly eating your reviews, and a real per card interval on the handful of forms that are actually irregular.