Why a word comes back when it does

Gaps beat repetition

Reading a word ten times in one sitting feels like learning it. Spreading the same ten encounters over days produces far more durable memory for the same effort. This is the spacing effect, first measured by Hermann Ebbinghaus in 1885, and it is one of the most replicated results in psychology: a 2006 meta-analysis in Psychological Bulletin pooled over 250 studies and found spaced practice beat massed practice again and again. How long the gap should be depends on how long you need to keep the word — in one 2008 experiment the best gap was roughly a tenth to a fifth of the time you want to remember it for.

So intervals grow rather than repeat. A word you get right moves from minutes to a day, then to days, then weeks, then months — and one you get wrong drops back to minutes.

Recalling beats reviewing

Being shown the answer again is not the same as producing it. In a 2006 experiment, students who were tested on a passage remembered substantially more of it a week later than students who simply reread it the same number of times — even though the rereaders were more confident at the time. A 2013 review rated ten common study techniques; practice testing and distributed practice were the only two rated high utility. Highlighting and rereading, the two most popular, were rated low.

So a card asks you to produce the word before it shows you anything. Lesson tests and grammar drills are written from your own material, and the tutor makes you say it out loud, which is retrieval with nowhere to hide.

Producing beats recognising

A card that shows you the answer and asks you to rate how well you knew it is testing your judgement, not your memory — and the experiment above is precisely about how unreliable that judgement is. So Duru asks you to type the word and checks it before you see anything. The direction is deliberate too: the prompt is in the language you already speak and the answer is in the one you are learning, because recognising a word on sight and producing it when you need it are different skills, and a conversation asks for the second.

Every card is typed and checked against the word and its dictionary form, and the check decides the next interval: right sends the word further out, a near miss a little less far, a miss back to minutes. If a right answer felt too easy, you can say so and it goes further still — but you cannot tell yourself you knew it once the answer is already on screen.

Saying it out loud

Words read aloud are remembered better than the same words read silently — the production effect, delineated across nine experiments in 2010. Speaking also removes the places to hide: a conversation asks for the word at the moment it is needed, with no button to press first and no list to choose from.

The tutor builds its sentences from the vocabulary and grammar in your own course, and glosses anything new once before reusing it. What you say is graded as you go — a particle, a tense, a speech level — and those mistakes join the same list your typing and your tests feed, so what you keep getting wrong out loud comes back to you in writing.

Grammar in sentences you have not seen

A rule you can recite is not a rule you can use. Each grammar pattern is practised in fresh sentences — translate it, fill the gap, or pick the form — so you apply the pattern rather than remember one example of it.

Sentences are written from your course’s vocabulary and checked for grammar and meaning rather than exact wording, so a correct sentence phrased differently still counts. Ten right answers and the pattern counts as mastered.

Mistakes say what to practise

A wrong answer is information: it says which word, and which kind of thing, did not come out. Duru records every miss — typed, in a test or out loud — with what went wrong: the word, the particle, the tense, the speech level.

Words you keep missing are weighted up: new grammar sentences are chosen to contain them, tests prefer questions that use them, and the tutor is told to work them back into the conversation.

A test that checks the lesson, not your mood

A test is retrieval practice with a stake: it shows you, rather than your sense of it, whether a lesson has stuck. Each lesson’s test is written from that lesson and the ones before it, mixing picking the right form with filling the gap.

Answers are checked exactly, each comes with a one-line explanation, and 80% marks the lesson as done. A test does not change your card schedule — that stays with the cards.

What Duru actually runs

An SM-2 family scheduler, the same lineage Anki made familiar. A new card is shown after a minute, then ten minutes, then a day. After that every correct answer multiplies the interval by an ease factor that starts at 2.5: “Good” multiplies, “Easy” multiplies harder and raises the ease, “Hard” barely moves it and lowers the ease, and “Again” sends the card back to minutes and lowers the ease further. The ease never falls below 1.3, so no card can collapse into being asked forever.