Spaced Repetition for Turkish Vocabulary: What the Evidence Shows and How SM-2 Schedules Your Reviews

What Ebbinghaus measured

The forgetting curve comes from one person testing himself. Hermann Ebbinghaus spent several years learning lists of nonsense syllables, short made-up strings chosen so that nothing he already knew could help. He published the results in 1885, and an English translation followed in 1913.

He did not count how many syllables he could still recall. He measured savings. After a delay he learned the same list again and noted how much less time the second learning took. A saving of 28% means relearning took 28% less time than the first learning.

His savings fell quickly and then levelled off. They were about 58% after 20 minutes, 34% after one day and 28% after two days. After 31 days they were still about 21%. The claim you often read, that 70% of new words are gone within two days, stretches the two-day figure. It measured the cost of relearning meaningless syllables, not recall of real words.

In 2015 Jaap Murre and Joeri Dros repeated the experiment and published it in PLOS ONE. Dros was the only subject, and he relearned lists after the same seven delays Ebbinghaus used. His curve had the same shape. They also found a small rise in savings at one day, which they suggest may come from sleep.

Turkish words are not nonsense syllables. Ebbinghaus himself found that stanzas of a poem took about a tenth of the repetitions that a syllable list of the same length needed. So his percentages are not a forecast for your vocabulary. What carries over is the shape. Most forgetting happens soon after learning, and it slows down afterwards.

The spacing effect, the part with strong evidence

The better-supported finding is about timing. When you study the same material more than once, spreading the sessions out works better than packing them together, even with the same total study time.

Nicholas Cepeda and colleagues reviewed this research in Psychological Bulletin in 2006. They pooled 839 comparisons from 317 experiments in 184 articles, all on verbal recall tasks such as learning word lists. Spaced study came out ahead of massed study.

Their second finding matters for scheduling. The best gap between two sessions was not fixed. It grew with the time until the final test, so remembering something for a week called for a shorter gap than remembering it for a year.

That is the usual argument for a scheduler whose gaps widen as a word gets older. The review did not settle whether widening gaps beat equal ones. It is evidence for spacing in general. It does not show that SM-2, or any other particular scheduler, sets the gaps in the best way.

What is SM-2?

SM-2 is a widely used review scheduler. Piotr Woźniak described it in his 1990 master’s thesis at the Poznań University of Technology, and its name comes from his SuperMemo program. It keeps three numbers for each item: an ease factor, the current interval in days, and a count of correct reviews in a row.

In the original you grade each answer from 0 to 5. After a correct answer the first interval is 1 day and the second is 6 days. Each later interval is the previous one times the ease factor. The ease factor starts at 2.5, a formula moves it after every grade, and it never falls below 1.3. A grade below 3 sends the item back to the start of the sequence.

The constants are fixed. Apart from each item’s ease factor, SM-2 does not adapt to how a particular learner forgets. It also schedules each item on its own, so it cannot know that two words share a root.

How does Dilsever schedule a review?

Dilsever gives you four buttons instead of six grades: Again, Hard, Good and Easy. Under each button it shows the interval that pressing it will set, so you see the effect before you choose. The table below is what the app’s scheduler does after you rate a word you have already learned.

A word you rate Good every time, at the starting ease of 2.5, comes back after 1, 6, 15, 38 and 95 days. Each step after the second is the previous interval times 2.5, rounded to the nearest day. A single Hard lowers the ease, so every later step grows a little more slowly.

Intervals of 3 days or more are then shifted at random by up to one day, or by about 5% for longer intervals. Without that spread, words learned together would keep falling due on the same day, and the pile would never break up.

New words first go through short steps inside the session. In vocabulary practice, Again or Hard on a new word brings it back in about a minute. Good brings it back in 10 minutes, and a second Good moves it to the day schedule at 1 day. Easy skips the steps. When you miss a word you had learned, Again resets it and lowers its ease at once. It then returns after 10 minutes, and once you rate it Good or Easy it goes back to a 1-day interval.

The same numbers set the labels on your words. A word moves from Learning to Reviewing after 3 correct reviews in a row. It counts as Mastered once its interval reaches 21 days.

The Conjugation Lab, the Suffix Lab and the numbers drill use the same day schedule without the minute steps. There you type an answer and Dilsever picks the rating for you. A wrong answer counts as Again. A right answer counts as Easy if it came quickly, Good if it took longer, and Hard if it was slow.

Where Dilsever departs from classic SM-2

The core is SM-2. The first two intervals of 1 and 6 days, the starting ease of 2.5, the floor of 1.3 and the multiply-by-ease rule all come from the original. The rest differs, and here is how.

Four buttons replace the six grades, and fixed steps replace the ease formula. Good leaves the ease unchanged, as a grade of 4 does in the original. Easy raises it by 0.15, where a grade of 5 raises it by 0.10, and Easy also multiplies the interval by 1.3.

Hard is not a pass in the usual sense. In the original, a grade of 3 still counts as correct and moves the item along the sequence. In Dilsever, Hard stretches the current interval by 1.2 and leaves the count of correct reviews where it was.

Again lowers the ease by a fixed 0.20. In classic SM-2 the formula sets the change, and Woźniak’s description can be read two ways for a failed item. Either way, the item restarts at 1 day. We keep a penalty because a word you keep failing would otherwise return at full-ease intervals every time you relearned it.

The minute steps take the place of SM-2’s rule of repeating weak items at the end of a session. The random spread is not part of SM-2 at all. Dilsever also rounds intervals to the nearest day, where the original rounds up.

Newer schedulers, and what we have not measured

SM-2 is more than 35 years old, and it is not the last word on scheduling. Newer schedulers fit a model of memory to review history instead of relying on fixed constants. FSRS is an open-source example. It grew out of work by Junyao Ye, Jingyong Su and Yilong Cao, presented at the KDD conference in 2022, which fitted a memory model to 220 million review logs.

We use SM-2 because it is simple and predictable, and because its rules fit in one table. We have not measured how Dilsever’s scheduler compares with FSRS or any other, and we publish no retention figure for Dilsever learners.

Spaced repetition and Turkish words

Turkish builds many words from one root. TDK’s Güncel Türkçe Sözlük lists yazmak (to write), yazar (writer) and yazı (writing), all built on the root yaz-. Knowing the root can help you recognise a new word the first time you meet it.

The scheduler does not use that link. Each word keeps its own ease and interval, so reviewing yazmak does not move the date for yazar. If you already know yazar well, you will rate it Easy, and its own interval will grow faster.

Suffixes need practice of their own, since a vocabulary card shows a word in its dictionary form. That is why the Suffix Lab keeps a separate schedule for each suffix family, and the Conjugation Lab keeps one for each verb.

Using the buttons well

Rate what happened, not what you hoped. Press Hard only when you got the word right but slowly. Keep Easy for answers that came at once, because Easy lengthens the next gap more than Good does.

Review words on the day they fall due when you can, since the schedule assumes you do. Add new words at a pace your reviews can absorb. Each new word brings reviews of its own over the days that follow.