Stop studying and you will lose vocabulary, but not the way most people imagine. The loss is front-loaded into the first few years and then flattens out for decades. And the words that seem to vanish are mostly still there: relearning them is dramatically faster than learning them was.
The loss is front-loaded, then it mostly stops
The best evidence here comes from an unusual study. Harry Bahrick tested 733 people on the Spanish they had learned in school, across a span of 50 years since their last class, and compared their scores against how much training they originally had.
Standard forgetting curves predicted that after decades of disuse there would be essentially nothing left. That is not what he found. The decline was steep at first, concentrated in roughly the first three to five years after instruction stopped. After that, the curve went nearly flat. Bahrick reported virtually no further loss between five and twenty-five years, which is a strange and rather cheering result for anyone who took a language in school and has not touched it since.
He called the surviving layer permastore. It is not that these people kept practising, either. Most of them reported so little rehearsal that his data showed no meaningful rehearsal effects at all. The knowledge simply stayed.
| Years since you stopped | What the retention data shows |
|---|---|
| 0–3 | The steepest decline. Most of what is going to go, goes here. |
| 3–5 | Loss continues but slows noticeably. |
| 5–25 | Close to flat. Bahrick found virtually no further loss across this span. |
| 25–50 | A gentle decline resumes, nothing like the early drop. |
The practical reading is that the first couple of years after you stop are the dangerous window. Survive those with a little contact and you are likely holding a core that will sit quietly for a very long time.
”Forgotten” usually means “unreachable”
The second finding is the one that changes how the first should feel. When researchers take words a learner can no longer produce and teach them again, the relearning goes much faster than the original learning did. This is the savings effect, and the size of the saving is a measure of what is still in storage.
Hansen, Umeda and McKinney published a large study of this in 2002, testing over three thousand returnees who had learned Japanese or Korean abroad and then come home to an environment where they barely used it. They reported the strongest savings effect recorded up to that point in research on reactivating lost vocabulary.
The measure itself is simple, which is part of why it is convincing. You compare the effort it takes to relearn a forgotten item against the effort the original learning took, and the difference is the saving. No self-report is involved, and the learner’s own sense of whether they “still know” the word does not enter into it.
The interpretation matters. If a word were genuinely erased, relearning it would cost exactly what learning a brand-new word costs. It does not. So attrition is better described as an access problem than a deletion problem, which is why a word you could not produce yesterday can come back the moment you meet it in a sentence.
I find this reframes the whole anxiety. The nagging sense that your studying has drained away is mostly wrong. What has drained away is the speed of retrieval, not the entry.
What that looks like when you come back
In ordinary terms, returning to a language you have let lapse tends to feel like this. The first days are humiliating, because retrieval is slow and you reach for words that will not come. Then things move quickly, far more quickly than they did the first time around, and a fair amount arrives back within weeks rather than months.
That asymmetry between how it feels and how fast it actually returns is worth knowing in advance, because the early humiliation is what makes people quit. The slowness at the start is not evidence that you have to begin again.
It also suggests a specific way to restart. Go back to material you have already worked through rather than buying something new at your old level, and move fast, accepting that a lot of it will feel half-familiar rather than known. You are not trying to learn these words; you are trying to find which ones still have a thread attached. The ones that come back with a tug are cheap. The ones that stay silent are the real work, and there will be fewer of them than the first day suggests.
The exam problem
There is an uncomfortable implication here for anyone who has crammed for a test. Vocabulary assembled quickly, in massed sessions, in the weeks before an exam has exactly the profile the retention research says will not last: shallow original training, no spacing, and very little production practice. It carries you through the paper and then drains at close to the fastest rate the curve allows.
That is not an argument against preparing for exams, and a deadline is a genuinely useful thing to have. It is an argument for starting earlier than the deadline strictly demands, because the same total hours spread across months instead of weeks buy you a comparable score and a vocabulary that survives the following year. The cost of cramming is not paid on test day. It is paid months later, when you sit down to actually use the language and find the list you drilled has gone quiet.
How well you learned it decides how much survives
Bahrick’s second result is less comforting and more useful: retention across the whole fifty years was predictable from the level of original training. People who had studied more, and studied to a higher level, kept more, and the gap did not close.
This is the strongest argument I know for learning things properly the first time instead of racing through a list. A word you met once and half-recognised is not in the same category as a word you drilled to the point of easy production. Only the second kind is a candidate for the plateau. This is also why the number of encounters a word gets before you stop matters so much, a point covered in more detail in how many times you need to see a word.
The same logic explains why passive vocabulary is more fragile than active vocabulary. Words you have only ever recognised sit closer to the edge, and the distinction between the two is worth understanding on its own terms in active versus passive vocabulary.
Maintenance costs far less than acquisition
Here is the part that should change your schedule. Keeping vocabulary is much cheaper than getting it, and the cheapest version involves reviewing less often, not more.
The Bahrick family ran a nine-year experiment on themselves, training on foreign vocabulary across either 13 or 26 sessions spaced 14, 28 or 56 days apart, then testing retention up to five years after training ended. Thirteen sessions spaced 56 days apart produced retention comparable to twenty-six sessions spaced 14 days apart.
Read that again, because it is a remarkable trade. Half the total work, same result, purely from letting more time pass between sessions. Wider spacing felt harder in the moment and made acquisition slightly slower, and it won anyway over the long run.
The caveat is real and I would rather state it than bury it: this study had four participants, and they were the authors and their family. That is a tiny sample by any standard. It happens to point the same direction as the broader spacing literature, which is why it gets cited so often, but treat it as a strong hint rather than a settled number.
The mechanism is the same one behind spaced repetition. A review that lands while the memory is still fresh is easy and does almost nothing. A review that lands when you have half-forgotten is uncomfortable and does most of the work. Longer gaps buy you more of the second kind.
What this changes about how you study
- Treat the first two years after stopping as the risk window. That is where the decline concentrates. A small amount of contact during that period protects a lot.
- Learn fewer words further. Depth of original training predicts fifty-year retention; the size of the list you skimmed does not.
- Do not panic-restart from zero. If you are returning after a gap, run through the old material rather than replacing it. The savings effect means that pass will be fast.
- Let intervals get long. The instinct to review everything weekly is expensive and, on the evidence above, counterproductive.
- Judge yourself on recall, not recognition. Recognition survives on its own and flatters you. Production is what needs the maintenance.
The part the research does not promise
Two honest limits. First, most of this evidence concerns receptive knowledge — recognising and understanding words. Fluent production under time pressure is a harder skill and there is less reason to assume it sits as comfortably in permastore. Speaking rustier than reading, after a long gap, is the normal experience.
Second, none of this makes maintenance optional. The plateau in Bahrick’s data describes what remains after the early drop, not a guarantee against the drop. The people in that study lost a real portion of their Spanish in the first few years. What the research offers is a much cheaper maintenance plan than most learners assume they need, not a way to skip it.
How Vocaby fits
The thing a review system genuinely controls is when a word comes back, and that is exactly the lever this research points at.
Vocaby uses FSRS spaced repetition, which schedules each word against your own forgetting curve rather than a fixed calendar. As a word proves durable its interval stretches, so the maintenance cost of everything you already know keeps falling while your attention goes to the words that are still shaky. That is the wide-spacing finding turned into a default rather than something you have to plan.
Every entry carries IPA and real audio, so a word is stored as a sound as well as a shape, and the library runs past 29,000 words if you want to keep going rather than only keep what you have. You can browse it in the word library.
The words you already learned are mostly still in there. The work is keeping the door open, and that is cheaper than the first climb was.
Download Vocaby on the App Store to learn words with audio, IPA, examples, and spaced repetition on every entry.
Frequently asked questions
- How much vocabulary do you forget if you stop studying?
- Less than most people expect, and the timing matters more than the total. Harry Bahrick's 50-year study of school Spanish found the steepest losses in the first three to five years after instruction ended, followed by a long stretch where retention barely moved at all. What survives that first drop tends to stay.
- Are forgotten words gone permanently?
- Usually not. Relearning words you once knew is markedly faster than learning them the first time, which is called the savings effect. That speed advantage is evidence the original memory trace is still there and the problem is access rather than loss. Hansen and colleagues reported the strongest savings effect yet measured in a 2002 study of over three thousand returnees.
- How often do I need to review to keep vocabulary I already have?
- Far less often than you needed to review to learn it. Maintenance reviews can sit weeks or months apart once a word is established, and wider spacing actually produces better long-term retention than tighter spacing. A review system that schedules by your own forgetting curve handles this automatically.
Keep reading
Why Some Words Are Harder Than Others
When a word will not stick, the usual assumption is that something is wrong with your method or your memory. Often the word is simply more expensive than the ones around it, and the cost is measurable in advance.
Should You Finish One Deck Before Starting the Next?
The tidy answer is to mix everything, because interleaving beats blocking in most of the research. The honest answer has conditions attached, and two of them are likely to apply to you.
Can You Learn Vocabulary While You Sleep?
Not the way the apps imply. You can't absorb new words overnight, but sleep is when the words you studied while awake actually stick. Here's what the research really found and how to use it.