The short answer
If the goal is speaking, the numbers are smaller than most people expect. A few hundred words will carry you through simple transactions — ordering food, buying a ticket, asking directions. Everyday conversation takes something like 1,500 to 3,000. Speaking needs fewer words than people expect because you choose them: when one will not come to mind, you can say it another way.
Understanding other people takes more, because they choose the words and you have to follow. To follow television comfortably you want around 6,000 to 7,000 word families, and to read a newspaper without reaching for a dictionary, 8,000 to 9,000. Those two figures come from Paul Nation's 2006 study and are the most widely used benchmarks in the field.
Before going further, one piece of vocabulary about vocabulary, because it explains most of the confusion in this subject. Researchers rarely count words one by one. They usually count word families: a base word together with its obvious relatives. Run, runs, running, ran and runner are one word family, not five words, on the reasoning that once you know run you can work the others out. Some studies use a slightly tighter grouping called a lemma, which keeps run, runs, running and ran together but treats runner as separate. And some counts simply tally every distinct form.
This matters enormously. The same person can be described as knowing 11,000 word families or 42,000 individual words — both true, both measuring the same knowledge. Whenever you see two vocabulary figures that seem to contradict each other, this is almost always why. In this article, numbers are word families unless stated otherwise.
The reason nobody can give you one number is that the real range is enormous. Toki Pona, a constructed language, has about 120 words, and people hold real conversations in it. At the other end, Marc Brysbaert and colleagues estimated in 2016 that an average 20-year-old American English speaker knows around 42,000 lemmas — and keeps adding roughly one new word every two days until they are 60.
That is a span of more than two orders of magnitude, so the useful question is not 'how many words does a language need' but 'how many do I need for what I want to do'. The rest of this article is about answering that for yourself.
What beginner, intermediate and advanced mean in numbers
Courses and exams describe levels with CEFR codes — A1 through C2, from the Common European Framework of Reference, the scale used across Europe and by most exam boards — while everyone else says beginner, intermediate or advanced. They refer to the same thing, and both can be roughly translated into vocabulary size using James Milton's research.
A beginner (A1) is working with a few hundred to a thousand words: greetings, numbers, set phrases, enough to meet basic needs. An elementary learner (A2) has perhaps 1,500 to 2,500 and can handle shopping, directions and familiar routines. Intermediate (B1) sits around 2,750 to 3,250 words, which is enough to hold a conversation about familiar topics without preparing it in advance. Upper intermediate (B2), roughly 3,250 to 3,750, means following most speech and discussing ideas rather than just facts. Advanced (C1) is around 5,000, the level at which people work or study in the language. Near-native (C2) is 7,000 to 9,000 and up.
Two things are worth noticing in those numbers. The first is how much of the range sits below 4,000 words — the entire journey from absolute beginner to comfortably conversational. The second is that vocabulary size accounts for about 60 to 70% of the difference between CEFR levels, according to Milton and Alexiou. Grammar, listening speed and confidence make up the rest, which is why two people with the same word count can perform very differently.
Why the first thousand words do most of the work
Words are not evenly distributed. A small number of them — the, and, have, go, good — turn up constantly, while most words in a dictionary appear rarely. This pattern, described by George Zipf in the 1930s and 40s, holds across every language that has been measured, and it has two practical consequences.
The first is good news. The most frequent thousand word families cover the large majority of everything you will hear or read, which is why beginners make such visible progress: almost every word learned early is one you will meet again within days.
The second is the part people find discouraging. Each additional thousand words adds less coverage than the thousand before it. By the time you are learning your fifth or sixth thousand, you are working just as hard while noticing far less difference. This is the arithmetic behind the intermediate plateau — the long stretch where you know you are improving but cannot feel it. Nothing has gone wrong; you have simply moved onto the flat part of the curve.
The chart above shows where that leads. Researchers talk about coverage, meaning the share of the words in front of you that you already know. At 90% coverage roughly one word in ten is unfamiliar, which is exhausting; at 98% it is one in fifty, which is comfortable. To reach 98% Nation found you need 6,000 to 7,000 word families for speech and 8,000 to 9,000 for writing. The dashed lines mark the two best-known teaching lists, which sit well below both. They are excellent places to start and were never intended to be finish lines.
One caveat about that 98% figure: it traces back to a study by Hu and Nation in 2000 with 66 university students, and a 2023 replication with a different group could not fully reproduce it. It remains the best benchmark available and is used throughout the field, but treat it as a well-supported guide rather than a precise threshold.
Speaking needs fewer words than reading — and different ones
Two distinctions change the number a great deal, and both work in the learner's favour.
The first is producing versus understanding. You need far fewer words to speak than to read, because when you speak you choose the words. If you do not know 'windscreen wipers', you can say 'the things that clean the window when it rains' and be understood. When you are listening or reading, you do not get that choice — the other person picks the words. This is exactly the trick that constructed languages use on purpose: Toki Pona and Ogden's Basic English work by making circumlocution the normal way to speak rather than a fallback.
The second is speech versus writing. Everyday conversation uses a smaller and partly different set of words than books do. Sandra Adolphs and Norbert Schmitt measured this against a modern corpus — a large searchable collection of real recorded speech and found that 2,000 word families covered slightly under 95% of conversation, with about 5,000 words needed to reach 96%. An earlier study of Australian oral English in the 1950s had put 2,000 families at around 99% — the gap between those results comes from different corpora and different decades, which is a reminder that these numbers describe a particular sample of language rather than the language itself.
That matters when choosing what to study. Most published frequency lists are compiled from written material, so they describe what is common in books, articles and academic writing. If your aim is reading, that is the right target. If your aim is talking with family, dealing with shopkeepers or following a conversation at work, lists built from recorded speech, film subtitles or spoken corpora will point you at more useful words sooner. It is worth checking what a course, deck or word list was built from before you commit weeks to it.
Why the numbers you read never agree
Search for this question and you will find figures that appear to contradict each other. Usually they do not — they are counting different things.
The biggest factor is the counting unit. Brysbaert's participants knew about 42,000 lemmas, which came from roughly 11,100 word families. Same people, same knowledge, numbers differing nearly fourfold. A word family groups run, runs, running, ran and runner together; a lemma treats fewer of those as one item; a raw word count treats them all separately. Whenever two sources disagree wildly, check this first.
The language matters too. Word families behave differently in Finnish or Hungarian, where a single root generates dozens of forms, than in Spanish or English. Chinese is normally counted in characters rather than words, which makes direct comparison meaningless.
So does the person. Brysbaert found a range from about 27,000 lemmas at the fifth percentile to 52,000 at the ninety-fifth among native speakers of the same age. Averages hide a lot of variation.
Finally, there is what counts as knowing a word. Recognising it when you see it, understanding it correctly in context, and being able to produce it accurately in speech are three different thresholds, and studies measure different ones. Receptive vocabulary is always larger than productive vocabulary, often by a wide margin.
Getting there
Three things follow from all of the above.
Learn frequent words first. That is where the curve is steepest, and every major app and shared deck is already ordered roughly this way. Anki's shared library, for instance, includes a Spanish deck built from a published frequency dictionary with 5,000 entries — enough to carry you past conversational level if you finish it.
Match your input to your purpose. If you want to talk to people, spend your time on spoken language: conversation, television with subtitles, recorded dialogue. If you want to read, read. The vocabulary you meet is the vocabulary you will learn, so the material you choose quietly decides which words you end up with.
Review on a schedule rather than by mood. Vocabulary is lost as reliably as it is gained, and spaced repetition — reviewing an item just before you would forget it — is the best-supported technique for keeping it. Our guide to teaching methods covers how the different apps handle this.
And a final point of proportion. Nobody sets out to learn two thousand words; they learn words they keep meeting, and the count is a by-product. Frequency lists are useful as a route — do the common things first — and misleading as a destination. The person who knows 800 words and uses them daily is further along than the person who has reviewed 3,000 and spoken to no one.
References
- Nation, I. S. P. (2006). How Large a Vocabulary Is Needed For Reading and Listening? The Canadian Modern Language Review, 63(1), 59–82. — the 6–7,000 and 8–9,000 word family thresholds for 98% coverage
- Hu, M. & Nation, P. (2000). Unknown vocabulary density and reading comprehension. Reading in a Foreign Language, 13(1), 403–430. — the original experiment behind the 98% figure
- Kremmel, B. et al. (2023). Unknown Vocabulary Density and Reading Comprehension: Replicating Hu and Nation (2000). Language Learning. — the replication that could not fully reproduce the original result
- Adolphs, S. & Schmitt, N. (2003). Lexical Coverage of Spoken Discourse. Applied Linguistics, 24(4), 425–438. — how far 2,000 word families go in real conversation
- Brysbaert, M., Stevens, M., Mandera, P. & Keuleers, E. (2016). How Many Words Do We Know? Frontiers in Psychology, 7:1116. — 42,000 lemmas for an average 20-year-old, and the range between individuals
- Milton, J. (2010). The development of vocabulary breadth across the CEFR levels. EuroSLA Monographs 1, 211–232. — the vocabulary sizes attached to each CEFR level
- West, M. (1953). A General Service List of English Words. — the 1953 list of about 2,000 words, still widely used
- Browne, C., Culligan, B. & Phillips, J. (2013). The New General Service List. — the modern replacement, 2,809 words, free and open
- Ogden, C. K. (1930). Basic English: A General Introduction with Rules and Grammar. — English reduced to 850 words
- Lang, S. (2014). Toki Pona: The Language of Good. — a working language of about 120 words
- Zipf, G. K. (1949). Human Behavior and the Principle of Least Effort. — why a few words account for most of what we say
Frequently asked questions
How many words do I need to have a conversation?
Roughly 1,500 to 3,000 word families for everyday conversation, and a few hundred for simple transactions like ordering food or asking directions. You need fewer words to speak than to listen, because when you speak you can work around the words you do not know.
How many words is intermediate level?
Intermediate (CEFR B1) is around 2,750 to 3,250 words, and upper intermediate (B2) around 3,250 to 3,750, based on James Milton's research. Advanced (C1) is roughly 5,000.
How many words does a native speaker know?
An average 20-year-old native English speaker knows about 42,000 lemmas, which corresponds to roughly 11,100 word families, with wide variation between individuals. People continue adding words throughout adult life at about one every two days.
Why do different sources give completely different numbers?
Mostly because they count differently. A word family groups run, runs, running and ran as one item; counting lemmas or individual words gives a much larger figure for exactly the same knowledge. Sources also differ on whether they measure words you recognise or words you can use.
Should I study a frequency list?
Frequency order is a good route, since common words are the ones you will meet again soonest. Check what the list was built from: most are compiled from written material, so if your goal is conversation, a list drawn from speech or subtitles will be a closer match.
