There’s a number in applied linguistics that should be the first thing anyone learning English for work is told. It almost never is, probably because it makes obvious how short the minimum path actually is.
The number concerns lexical coverage: what percentage of the words you meet in a text or a conversation is covered by the words you already know. And the relationship between the two quantities isn’t linear.
How many words you actually need
In 2003 Svenja Adolphs and Norbert Schmitt published an analysis of CANCODE, a corpus of authentic English conversation, in Applied Linguistics. The result: the 2,000 most frequent word families cover roughly 95% of tokens in everyday spoken discourse. Going up to 3,000 adds a little over two percentage points.
Paul Nation measured the same thing in 2006 against the more demanding 98% threshold, the one his 2000 work with Marcella Hu identified as necessary to understand a text without outside help. There the numbers climb: around 6,000-7,000 word families for listening, 8,000-9,000 to read novels and newspapers without a dictionary.
| Goal | Word families |
|---|---|
| Understand 95% of everyday speech | ~2,000 |
| Understand 98% of speech (unassisted) | ~6,000-7,000 |
| Read fiction and newspapers without a dictionary | ~8,000-9,000 |
The curve is brutal in one direction and generous in the other. The first two thousand words give you almost all of spoken language. The next six thousand give you three more percentage points. If your goal is to function in a meeting, not to read Faulkner, you’re working on the first step, not the last.
One important detail: a “word family” isn’t a “word”. Manage, manager, management, managed, managing, unmanageable are one family. The actual count of forms you’ll recognise is far higher than the number in the table.
Age matters, but less than you were told
The “it’s too late, I should have started as a child” argument got its most precise measurement in 2018, when Joshua Hartshorne, Joshua Tenenbaum and Steven Pinker published a study in Cognition on 669,498 people, native and non-native, recruited through a grammar quiz that went viral.
The result: the ability to learn a second language’s grammar stays high until around 17-18, then begins to decline. But what the study measures is the probability of reaching competence indistinguishable from a native speaker’s. It doesn’t measure the probability of reaching perfectly functional English at work, which is a different and much more reachable goal.
In fact, in the short run adults start with an advantage. Carmen Muñoz documented in 2008, in studies of classroom learners, that those who start later progress faster in the early stages, because they bring study strategies, metalinguistic awareness and world knowledge that a child lacks.
The right question isn’t “can I still become native-like”. It’s “can I run a meeting”. Those are two goals with completely different curves.
How to study: two findings worth more than any method
In 2006 Nicholas Cepeda, Harold Pashler and colleagues published a quantitative synthesis in Psychological Bulletin of 254 studies covering over 14,000 participants on distributed practice. The conclusion is among the most robust in the whole psychology of learning: for the same total time, spreading study out over time produces markedly better retention than concentrating it. The advantage grows the longer the interval over which you need to remember.
The second finding is the testing effect. Henry Roediger and Jeffrey Karpicke, in experiments published in Psychological Science in 2006, compared students who reread material with students who tested themselves. In the immediate term, rereading seemed to work better. A week later, the self-testing group remembered far more. Attempting retrieval from memory, even when it fails, strengthens the trace more than passive re-exposure.
Together, these two results explain why spaced-repetition flashcards work and why rereading your notes doesn’t. It isn’t a question of apps: one applies both principles and the other applies neither.
The honest hour count
The Council of Europe and the certification bodies publish estimates of guided study hours needed to move between CEFR levels. They’re rough and vary a lot by person, but they give an order of magnitude:
| Level | Indicative cumulative hours |
|---|---|
| A2 (elementary) | ~180-200 |
| B1 (intermediate) | ~350-400 |
| B2 (upper intermediate) | ~500-600 |
| C1 (advanced) | ~700-800 |
B2 is the threshold most positions asking for “good English” require. That’s about five hundred hours: an hour a day for a year and a half, starting from zero. Starting from school-level A2, the condition of most adults in southern Europe, we’re talking about three hundred hours, roughly ten months at an hour a day.
That isn’t nothing. But it’s a finite number, and knowing it beats imagining an endless path.
Where to start, and in what order
- Start from frequency lists, not the textbook. The first 2,000 word families (the General Service List, or corpus-based frequency lists) have the highest return per unit of time. It’s the step almost no course takes explicitly.
- Add your field’s vocabulary, which is small. Academic and technical domain lists run to a few hundred items. Two thousand general words plus three hundred technical ones cover almost your entire working day.
- Use spaced repetition and self-testing. These are the two principles with the strongest experimental basis. Any tool implementing them is fine; no tool that doesn’t implement them is a good idea.
- Learn chunks, not single words. Not meeting but let’s push the meeting to Thursday. Research on formulaic sequences shows perceived fluency depends heavily on having these blocks pre-assembled.
- Listen more than you study, but with the text alongside. Listening without support below the coverage threshold produces frustration and little learning. With English subtitles it works.
- Schedule production, not just input. One conversation a week with someone, even a cheaply paid one, is worth more than three hours of app. It’s the same graduated-exposure principle I wrote about for public speaking.
- Structure the path around verifiable goals. I described the general method in how to learn a skill in 30 days.
My take: the block is fear of mistakes, not the language
I think the main obstacle, for adults who did English at school, isn’t linguistic but social: the fear of making mistakes in front of colleagues. It’s the same mechanism as the illusion of transparency, where the mistake feels far more visible than it is. In an international professional setting most of the people you’re talking to aren’t native speakers either, and are making the same effort.
I also think machine translation has changed the calculation in a way the popular advice hasn’t caught up with. For reading and writing, the marginal value of English has dropped a lot: a translation tool does an acceptable job on an email. For speaking in real time it has barely dropped at all, because in a meeting there’s no time to route through a translator. With limited time, today I’d put nearly all of it on the spoken side.
And a note on courses: many spend a lot of time on advanced grammar and little on high-frequency vocabulary. From the coverage data, the order should be the opposite.
Frequently asked questions
Are 2,000 words really enough? They’re enough to cover about 95% of what you hear in everyday conversation. They’re not enough to read a novel without a dictionary. These are different things and worth not confusing.
Course or self-study? Depends what the course gives you. If it gives you guided conversation and correction, it’s worth it. If it gives you lecture-style grammar, the two principles in the study section say you can do better alone for less money.
Do apps work? The ones implementing spaced repetition and self-testing exploit two real effects. What no app solves is spoken production under pressure, which is the part you actually need.
Do I need a certificate? Depends on the sector. It has value where it’s a formal requirement (public competitions, universities, visas). In the private sector a practical test in an interview counts for far more. I’ve written on the topic more generally in which certifications are worth it.
I’m 45, is it too late? To become indistinguishable from a native speaker, Hartshorne’s data says it’s unlikely. To reach a solid B2, there’s nothing in the literature that makes it unlikely.
Sources
- Adolphs, S., & Schmitt, N. (2003). Lexical Coverage of Spoken Discourse. Applied Linguistics, 24(4), 425-438. The CANCODE corpus analysis behind the 2,000 word families figure.
- Nation, I. S. P. (2006). How Large a Vocabulary Is Needed for Reading and Listening? Canadian Modern Language Review, 63(1), 59-82. The thresholds for 98% coverage.
- Hu, M., & Nation, I. S. P. (2000). Unknown Vocabulary Density and Reading Comprehension. Reading in a Foreign Language, 13(1), 403-430. The origin of the 98% threshold.
- Laufer, B. (1989). What Percentage of Text-Lexis Is Essential for Comprehension? In Laurén & Nordman (eds.), Special Language: From Humans Thinking to Thinking Machines. The 95% threshold.
- Hartshorne, J. K., Tenenbaum, J. B., & Pinker, S. (2018). A Critical Period for Second Language Acquisition: Evidence from 2/3 Million English Speakers. Cognition, 177, 263-277. The study of 669,498 participants.
- Muñoz, C. (2008). Symmetries and Asymmetries of Age Effects in Naturalistic and Instructed L2 Learning. Applied Linguistics, 29(4), 578-596. The early advantage of adult starters.
- Cepeda, N. J., Pashler, H., Vul, E., Wixted, J. T., & Rohrer, D. (2006). Distributed Practice in Verbal Recall Tasks: A Review and Quantitative Synthesis. Psychological Bulletin, 132(3), 354-380. The 254 studies on distributed practice.
- Roediger, H. L., & Karpicke, J. D. (2006). Test-Enhanced Learning: Taking Memory Tests Improves Long-Term Retention. Psychological Science, 17(3), 249-255. The testing effect.
- Boers, F., Eyckmans, J., Kappel, J., Stengers, H., & Demecheleer, M. (2006). Formulaic Sequences and Perceived Oral Proficiency: Putting a Lexical Approach to the Test. Language Teaching Research, 10(3), 245-261. The evidence on lexical chunks.
- Council of Europe (2020). Common European Framework of Reference for Languages: Companion Volume. The CEFR level framework; guided-hour estimates are published by certification bodies and should be read as orders of magnitude.
This article was written following the process described in how we write. If you find a figure reported incorrectly, write to me: I’ll correct it and flag it at the bottom of the article.



