Spaced Repetition for Vocabulary: What the Research Says and How to Use It

By Beeblio · October 23, 2026

You learned a German word on Monday. By Wednesday it felt hazy. By Friday it was gone. This is not a failure of effort or intelligence — it is the forgetting curve doing exactly what Hermann Ebbinghaus documented in 1885, and what every serious vocabulary researcher has confirmed since. The good news is that the curve has a known shape, and a known fix. That fix is spaced repetition — the practice of reviewing material at expanding intervals, timed to catch memories just before they decay. For language learners, understanding both the problem and the solution at a mechanistic level is what separates people who accumulate vocabulary permanently from people who re-learn the same words every few months. This article walks through the forgetting curve, the SM-2 algorithm that turns the theory into a schedule, and — critically — why spaced repetition vocabulary work is most powerful when it is anchored in real reading.

The Forgetting Curve: A 140-Year-Old Problem with Numbers

Hermann Ebbinghaus was a German psychologist best known for his pioneering work on memory and forgetting, and his forgetting curve research demonstrated that individuals tend to forget a significant portion of what they have learned relatively quickly if they do not review the material. His method was rigorous for its era: he studied the memorization of nonsense syllables by repeatedly testing himself after various time periods and recording the results, then plotted those results on a graph, creating what is now known as the "forgetting curve," which behaves like an exponential decay function.

The decay is steep. He observed that after learning new material, his subjects forgot 50% of the information within 30 minutes, and after 24 hours they had forgotten between 70% and 80%. He further found that each time individuals reinforce their learning, the rate of forgetting decreases — an insight that led to the development of the concept of spaced repetition, which involves reviewing material at increasing intervals over time, thus reinforcing learning.

Is a 19th-century self-experiment trustworthy? Largely, yes. Murre & Dros (2015) replicated Ebbinghaus's original experiment with modern methodology, and the results closely matched the 1885 data, confirming the forgetting curve's shape and timing — published in PLOS ONE, this study validated that the core pattern holds across different experimental conditions. There was one notable wrinkle: retention around the 24-hour mark came in slightly higher than the smooth curve predicted, which the researchers linked to sleep's role in memory consolidation — a detail that matters for scheduling, because a first review that lands after a night's sleep tends to hold better than one crammed in a few hours later the same day.

The mathematical picture is equally clear. The key variable is S, the stability of the memory: when S is low (new, weakly-encoded information), the curve drops steeply; when S is high (well-rehearsed material), the curve flattens. This is the fundamental insight behind spaced repetition — each review session increases S, making the memory more resistant to decay.

The scale of the spacing advantage is large. A meta-analysis by Cepeda et al. (2006) analyzed 184 articles involving 317 experiments on spacing effects and found that distributed practice consistently outperformed massed practice (cramming) by 10–30% across all study types and age groups. And Karpicke & Roediger (2008) demonstrated that active retrieval combined with spacing produced 150% better long-term retention compared to restudying — published in Science, this is one of the strongest pieces of evidence for spaced repetition.

How Much Vocabulary Do You Actually Need to Read?

Before diving into how to schedule reviews, it helps to have a precise target. The research here is unusually concrete. Nation (2006) conducted a corpus analysis and found that the first 1,000 most frequent word families in the British National Corpus (BNC) provide coverage of 78% to 81% of a written text, with the second 1,000 adding another 8% to 9%. That sounds encouraging until you realize coverage and comprehension are different things.

Hu and Nation (2000) proposed the concept of the lexical threshold, arguing that accurate reading comprehension in English requires mastery of at least 98% of the vocabulary in a given text. To put that concretely: in a text with 98% coverage, one out of every 50 words is unknown to a reader; in a text with 80% coverage, one out of five words is unfamiliar. The latter is cognitively exhausting — comprehension collapses. Nation (2001) summarizes the finding: "The probabilistic threshold is 98%. With this coverage almost all learners have a chance of gaining adequate comprehension."

It is worth acknowledging the limits of this research: Hu and Nation's (2000) study, which stipulated that L2 readers need to be familiar with 98% of lexical items for adequate text comprehension, has become highly influential, but the 98% threshold figure is based on findings from a regression analysis conducted with only 66 university students in New Zealand. Subsequent replications have produced mixed results, and Schmitt, Jiang, and Grabe (2011) found a linear relationship between lexical coverage and comprehension but did not find a single coverage figure that ensured comprehension, suggesting the level required may vary according to the degree of comprehension needed. The 98% figure is a useful planning heuristic, not a universal law.

What does 98% coverage look like for a real German learner? Here is a short passage pitched at the 1,500-word frequency band, the level Beeblio uses to calibrate this example:

At the 1,500-word band, a learner who knows those words recognizes 100% of the tokens above — the passage sits comfortably within reach. The next step is encountering words just beyond that band, the words Beeblio introduces as new targets. The research question then becomes: once you meet a new word in a text, how many times do you need to encounter it, and in what pattern, before it sticks?

How Many Encounters Does a Word Need?

The answer is more than most learners assume, and the distribution matters as much as the count. Nation (2013) argued that repeated exposure should be the foundation of vocabulary learning, and that exposure to a word once is insufficient. Reviews of the literature put the range somewhere between 7 and 20+ encounters depending on the depth of knowledge being targeted. Webb and Nation (2013) summarize that there should be at least between 7 and 16 encounters with each word, and that repetition should be spaced rather than massed.

For productive knowledge (being able to actively use a word), the bar is higher still: Teng (2016) found that EFL learners needed to encounter a word 14 times to recognize it, and at least 18 times for acquiring productive vocabulary knowledge. The uncomfortable implication is that incidental reading alone — however pleasurable — rarely delivers the encounter density that full acquisition requires. Repeated exposure to words during reading is an important factor in incidental vocabulary learning, but the statistical reality is that most words in a given text appear too infrequently to push past the acquisition threshold on their own.

This is precisely where spaced repetition review cards earn their keep. They guarantee the encounter density that reading luck cannot.

SM-2: The Algorithm Behind the Flashcard

The most widely used spaced repetition algorithm in language learning is SM-2. SuperMemo, created by Piotr Woźniak in 1987, was the first spaced repetition software — a program that computed the optimal moment to review each item, and its openly published SM-2 algorithm still schedules reviews in Anki and countless other apps.

For every flashcard, SM-2 maintains three values: a repetition count tracking how many times in a row you have successfully recalled the card, an easiness factor (EF) that adjusts the interval multiplier based on your self-rating, and the current interval length. The schedule SM-2 produces works like this:

• First successful recall: review again in 1 day
• Second successful recall: review again in 6 days
• Every subsequent recall: multiply the previous interval by the item's easiness factor (default: 2.5)
• Any failure: reset to interval 1 and drop the easiness factor

SM-2 assigns a specific level of difficulty to each piece of information and uses this analysis to select the appropriate intervals between repetitions. The effect is that easy words drift into month- or year-long intervals while difficult words cycle back quickly — and because every item's schedule is independent, an SM-2-style scheduler guarantees that no word is silently lost: every item comes back precisely when it is about to slip away, while keeping the daily review load bounded, because well-known material recedes into month- or year-long intervals.

SM-2 has real limitations. In the SM-2 algorithm, new items do not benefit from the experience of old items — the algorithm is not adaptable across the card population. Modern successors address this: FSRS, currently in version 6, uses a differential equation model of memory stability and retrievability, achieving significantly lower mean absolute error in recall prediction than SM-2. For most learners, however, SM-2 (as implemented in tools like Anki) remains effective enough that algorithm choice is rarely the bottleneck — consistency is.

Does Spacing Type Matter? Expanding vs. Equal Intervals

A common claim is that expanding intervals (1 day → 6 days → 15 days → …) are superior to equal-spaced reviews. The evidence is more equivocal than the marketing suggests. Whether the gaps expand or stay equal makes little or no difference; the total amount of spacing is what matters (Karpicke & Bauernschmidt, 2011; Nakata, 2015).

What is clearly established is that any spaced practice beats massed practice by a large margin. Nakata (2015) compared the effects of massed and spaced schedules on L2 vocabulary learning: in the massed condition, target items were repeated four times in a row; in the spaced condition, the same words were repeated four times but after 30 intervening trials (approximately 6 minutes). Posttests conducted a week after the treatment showed that spaced learning was more than twice as effective as massed learning.

A second principle matters almost as much as the spacing itself: retrieval effort. The effort of generating an answer — even a wrong one followed by correction — does more for long-term memory than simply re-exposing a learner to the material. Retrieval practice and spaced repetition work best combined: spacing decides when to test, and retrieval practice decides what happens during that review.

Why Reading Plus Review Beats Flashcards Alone

Here is the core argument of this article, stated plainly: spaced repetition vocabulary review is not a replacement for reading at your level — it is a repair mechanism for reading's inherent inefficiency.

Reading delivers words in rich, meaningful context, which aids deep encoding. New words can be learned incidentally in reading, but most words encountered this way are not learned; items occurring more frequently in the text are more likely to be acquired and are more resistant to decay. The problem is that authentic texts are not vocabulary exercises — they don't guarantee that the word you need to acquire appears at the frequency you need.

Meanwhile, flashcard-only study has the opposite flaw: it gives you encounter density but strips away context. Research on the spacing of encounters has revealed that spaced learning consistently leads to greater retention of vocabulary knowledge than massed learning, but a word drilled in isolation — L2 → L1 translation — is shallowly encoded and rarely transfers smoothly into reading fluency.

The research points firmly toward combining both strands. One of the pedagogical implications of studies on lexical coverage is that when L2 learners have the lexical coverage necessary to understand a text, they may have greater success at incidentally learning any unknown words encountered in it — and Nation (2006) reports that 98% coverage is most appropriate for incidental vocabulary learning. In other words: read at your level, so the context is comprehensible and incidental acquisition can happen; create cards for the words just above your band, so your spaced repetition system drives them past the acquisition threshold; return to reading, and notice those words appearing naturally. The loop is self-reinforcing.

The meta-analytic evidence supports a strong advantage for this integration. In a Cambridge meta-analysis of incidental vocabulary learning studies, the effect size for learning in spaced conditions (g = 1.51) was larger than in massed conditions (g = 0.97) at the first posttest — and at a delayed follow-up posttest the advantage of spaced learning (g = 1.71) relative to massed learning (g = 0.58) became even more pronounced. Words that are first encountered in context and then driven to stability through spaced review are words that last.

A Practical Framework: What This Looks Like in Practice

Translating the research into a daily habit does not require precision engineering. The following table maps out a minimal viable system for a learner working at the 1,500-word German band.

One caution from the research: keep new card intake disciplined. Introducing too many new words at once leads to shallow learning and rapid forgetting. Ten new cards a day sounds modest, but at SM-2's compounding schedule, that is 3,650 new words per year — enough to move through multiple frequency bands in 12 months.

FAQ

Isn't reading enough? Do I really need flashcards?

For high-frequency words — those in the top 1,000–2,000 — reading alone may eventually work because those words appear constantly. For mid-frequency words (2,000–9,000), the encounter frequency in natural text falls sharply, and most learners will not hit the 7–16 exposures needed for acquisition without deliberate review. There is no consensus on the exact number of exposures required, but in EFL contexts it is difficult for learners to encounter words at a sufficient rate under diverse reading conditions alone. Flashcards close that gap.

What self-rating should I use in SM-2 to get accurate scheduling?

SM-2 uses a 0–5 quality scale. Anything below 3 is a failure and resets the interval. A common mistake is rating cards 4 ("correct with hesitation") when you actually felt certain — this suppresses the interval and wastes review time on already-solid words. Be honest: if you saw the word and immediately knew the meaning and pronunciation, rate it 5. Reserve 4 for mild effort, 3 for barely correct. The algorithm is only as good as the signal you give it.

Does the order of items in a review session matter?

Interleaving — mixing different words rather than reviewing thematically grouped items in blocks — has shown benefits for vocabulary acquisition, largely because it increases retrieval difficulty in a productive way. Previous research has established the effectiveness of "desirable difficulties," such as spacing, retrieval, and interleaving, in L2 vocabulary acquisition. Most SRS apps shuffle cards by default, which achieves this automatically.

How is the 98% coverage threshold relevant to my specific language level?

The 98% figure means roughly one unknown word per 50. At the 1,500-word German band, a text written entirely within that band gives 100% coverage — ideal for fluent reading and incidental encounter with new words introduced just above the band. As you move to authentic texts, coverage drops; the gap between your known-word count and 98% coverage tells you exactly how many words still need to be acquired. Nation (2006) concluded that to read authentic non-simplified texts, learners would need a vocabulary size of about 8,000–9,000 word families to achieve 98% coverage. That is a long journey for most learners — which is why a precise known-word tracker is more motivating than vague level labels.

Are newer algorithms like FSRS worth switching to?

Possibly, if you are a heavy user reviewing hundreds of cards per day. FSRS uses a differential equation model of memory stability and retrievability, achieving significantly lower mean absolute error in recall prediction than SM-2. For most learners, however, the difference in daily retention is small. The larger gains still come from the habit itself — reading consistently at your band and reviewing cards daily — rather than from algorithm fine-tuning.

Sources

• Ebbinghaus, H. (1885/2013). Memory: A Contribution to Experimental Psychology — Wikipedia summary and formula
• Murre, J. M. J., & Dros, J. (2015). Replication and Analysis of Ebbinghaus' Forgetting Curve. PLOS ONE
• Woźniak, P. (1987–ongoing). SuperMemo Method and SM-2 Algorithm
• Woźniak, P. History of Spaced Repetition. supermemo.guru
• Nation, I. S. P. (2006). How large a vocabulary is needed for reading and listening? Canadian Modern Language Review, 63(1), 59–82.
• Hu, M., & Nation, I. S. P. (2000). Unknown vocabulary density and reading comprehension. Reading in a Foreign Language, 13(1), 403–430.
• Webb, S., & Chang, A. (2021). How effective is second language incidental vocabulary learning? A meta-analysis. Language Teaching.
• Nakata, T., & Elgort, I. (2021). Effects of spacing on contextual vocabulary learning. Second Language Research, 37(2), 227–255.
• Nakata, T. (2015). Effects of expanding and equal spacing on second language vocabulary learning. Studies in Second Language Acquisition, 37(4), 677–711.
• Nation, I. S. P. (2001). Learning Vocabulary in Another Language. Cambridge University Press. [Summary via COERLL]

In Beeblio, every word you tap in a German (or any other) passage becomes an SRS card anchored to the exact sentence where you first met it — meaning your first encounter is already rich in context. Your known-word count advances only when the card has been reviewed successfully across multiple sessions, so the number you see on your progress bar reflects the research standard for actual retention, not just recognition on a single lucky day. Reading at your measured band and reviewing your cards daily is the most direct implementation of the research described above: comprehensible input paired with guaranteed encounter density, all calibrated to where you actually are.