What research says about how adults learn to understand, speak and remember a new language. In plain words, with every study linked, and what we built because of it.
You learn most from what you can almost follow.
We pick up a language by understanding messages in it: things said to us that we can just about follow. Stephen Krashen called this comprehensible input.
Too easy, and there is little to learn. Too hard, and it is only noise. The useful zone sits just past what you already know.
In a classic paper, Fergus Craik and Robert Lockhart made a simple case. How well we remember depends on how deeply we process something, not on how often we see it.
A word you used to say something you meant has been processed for meaning. A word on a list often has not.
In a controlled study, adults learned short phrases in Hungarian, a language none of them knew. The group that learned by listening and singing recalled and produced more of the phrases than the groups that listened and spoke.
The effect was on remembering exact phrases. It is one tool among several, not a shortcut.
Saying it out loud shows you what you can’t say yet.
Merrill Swain studied children in Canadian French immersion schools. After years of understanding French all day, their spoken and written French still lagged well behind native speakers of their age.
Her conclusion: understanding is not enough. You also have to produce the language, because trying to say something shows you exactly what you cannot say yet.
A review of 33 studies found that correcting learners while they speak has a real, medium-sized effect on learning.
Direct corrections worked faster. Subtler ones, like hearing your own sentence said back the right way, held up better over time.
Students learning Swahili words remembered far more a week later when they kept testing themselves than when they kept studying the list.
Even a wrong guess helps, as long as the right answer follows straight after.
What you come back to, you keep.
A review of hundreds of experiments found that practice spread over time beats the same practice crammed together.
The longer you want to remember something, the wider the gaps between sessions should be.
Language anxiety is real and measurable, and speaking is where learners feel it most.
A worldwide survey of 1,746 learners found that enjoyment is not just the absence of anxiety. It is its own thing, and learners reported more of it than anxiety.
The largest study of its kind tested grammar in around two-thirds of a million people. The ability to learn grammar stayed strong until about age 17, then declined slowly.
Starting later makes sounding exactly like a native speaker less likely. It does not stop you from learning to speak well.
Practice can come from a machine. Some things still take a person.
When students practised a language task with a chatbot, their interest fell after the first session. With a human partner, interest held and carried into the course.
A 2024 review of generative AI in language learning reaches a similar line. It is useful as a tutor and for feedback. It is weaker at the social side of language: knowing what to say, to whom, and when.
Studies measure groups, usually in classrooms or labs, often over a few weeks. None of them can promise how fast one person will learn.
We kept only findings we could trace to the original study, and we linked every one so you can read it yourself.