How Fast Do You Forget a Language? What the Research Actually Shows

Most people who stop using a language assume it drains away steadily, a little every month, until nothing is left. The best long-term evidence describes a different pattern: a steep loss over the first few years, then a long plateau where much of what remains stays put for decades. Knowing that shape changes what "maintenance" needs to do.

This post goes through the main studies, what they do and do not show, and how to turn them into a realistic plan for a language you are no longer actively studying.

On this page

Fifty years of school Spanish: loss, then a plateau

The landmark study is Harry Bahrick's "Semantic memory content in permastore" (Journal of Experimental Psychology: General, 1984). Bahrick tested 733 people who had learned Spanish in high school or college, with retention intervals of up to 50 years, on reading comprehension, vocabulary and grammar, using both recall and recognition tests. It was a cross-sectional study: different people at different distances from their last class, not one group followed for 50 years. A questionnaire recorded how much Spanish they had originally studied, their grades, and any use of Spanish since.

The findings, from the abstract:

• Memory declined exponentially for the first 3 to 6 years after the end of study.
• After that, retention stayed unchanged for up to 30 years, before a final decline late in life.
• Large portions of what people learned remained accessible for over 50 years, even without use.
• How much survived depended mainly on the level of original training and the grades received, and on how it was tested: people did better at recognising than at recalling.
• The great majority of participants used Spanish so little after school that the data showed no significant effect of later practice.

Bahrick called the stable portion "permastore." Learned items seemed to fall into two groups: those that were lost within about six years, and those that lasted more than 25 years, with almost nothing in between.

What that means in practice

Three lessons follow, with one caution.

The first few years are where loss happens. If you stop using a language, the damage is concentrated early. This is the period when even light maintenance has the most to protect.

The level you reach matters a great deal. In Bahrick's data, retention across the whole 50-year span was predictable from the level of original training, and the grades people received also mattered. In practice, someone who stopped after a year of classes keeps less than someone who studied for several years and did well. Getting a language to a solid level before you move on to another one is, on this evidence, a form of insurance.

Recognition outlasts recall. Participants did better when they only had to recognise the right answer than when they had to produce it. In everyday terms, understanding tends to survive better than speaking. Many people who "lost" a language find they can still follow far more than they can say.

The caution: these participants were American classroom learners of Spanish, mostly tested decades ago, and the study measured reading, vocabulary and grammar rather than conversation. It is the best long-range data available, but it is not a direct measure of spoken fluency, and it does not tell you how a language learned through years of immersion behaves.

Children are different

Bahrick's classroom learners kept a good share of what they had learned. Young children can lose a language far more completely. A brain-imaging study by Pallier and colleagues (2003) tested adults born in Korea and adopted by French families in childhood, who reported no conscious memory of Korean. On behavioural tests of Korean they did no better than native French speakers who had never heard Korean, and their brains did not respond to Korean differently from other unfamiliar languages.

That is an extreme case: young children, complete loss of contact, and replacement by another language. It is a useful reminder that a language learned early is not automatically permanent.

Forgotten is not the same as gone

Even when you cannot retrieve a word, traces often remain, and they make relearning faster. Psychologists call this "savings." A study by Hansen, Umeda and McKinney (2002) in Language Learning drew on data from 304 returnees from Japan and Korea who had learned Japanese or Korean as a second language. They found a strong savings effect for previously known vocabulary. The advantage shrank as time passed (age was a confounding factor) and was related to how much vocabulary people still knew.

The practical message is encouraging. Returning to a language after years away is usually much faster than learning it the first time, and the earlier you return, the more you get back. Our guide on relearning a forgotten language covers how to structure that comeback.

How much maintenance is enough?

This is the question everyone wants answered with a number, and the honest answer is that no study has established a minimum weekly dose for keeping a language at a given level. Anyone who quotes an exact figure is giving you a rule of thumb.

The closest evidence is about spacing, not volume. In a nine-year study, Bahrick and colleagues (1993) had four participants (Harry Bahrick and three family members, all co-authors) learn 300 English–foreign-language word pairs and then relearn them in either 13 or 26 sessions, spaced 14, 28 or 56 days apart. Thirteen sessions spaced 56 days apart produced retention comparable to 26 sessions spaced 14 days apart. Wider gaps slowed learning slightly during training but gave substantially better retention, tested up to five years after training ended.

Two caveats: there were only four participants, all of them the researchers themselves, and it covered vocabulary, not conversation. Still, it supports a reasonable idea. Maintenance can be sparse if it is regular, and a session that makes you work to remember is doing useful work.

A maintenance plan built on that evidence

Given what the research does and does not say, here is a sensible structure. It is a heuristic, not a tested protocol:

• Protect the first few years most. If you are stepping back from a language, that is when to keep a small, regular habit, even if it is only a podcast episode or a chapter a week.
• Favour production over passive exposure. Recall holds up worse than recognition, so the most valuable maintenance makes you produce language: a monthly conversation, a short piece of writing, talking through your day out loud.
• Space it rather than cram it. A regular session every week or two is more in line with the spacing evidence than a big catch-up once a year.
• Raise the level before you move on. If you plan to switch focus to a new language, getting the current one higher first may protect more than any amount of later maintenance.
• Check yourself occasionally. Every few months, watch something unscripted or have a real conversation. If it feels markedly harder than last time, increase the dose for a while.

If you keep several languages going, our weekly framework for maintaining 3+ languages shows one way to rotate them. Because the spacing evidence comes from memory research, the principles in the science of spaced repetition apply here too.

Make the neglect visible

The early years of loss are easy to miss because nothing obvious happens. You simply stop using the language and assume it is still there. A log turns that into something you can see. In LangTrack, per-language stats show how much time each language has actually received, so a language that has quietly had nothing for months stands out. The free plan covers one language; Pro adds more, which is what you need for tracking maintenance across several.

The bottom line

The best long-term evidence says language loss is front-loaded: the decline is concentrated in the first 3 to 6 years without use, and much of what survives that period can last for decades. How far you got in the first place is the biggest predictor of what you keep, recognition outlasts recall, and relearning is faster than first learning. Nobody has measured a precise weekly minimum, but regular, spaced sessions that make you produce the language are the best bet the evidence supports.