NoaLingua for Chrome
Add to Chrome — Free Add to Chrome

Article

How many words do you need to know in a language?

Around 1,000 word families cover the gist of everyday speech, 3,000 handle ordinary conversation, and 8,000 to 9,000 are needed to read an unsimplified novel without constant lookups. The numbers are much smaller than people expect because word frequency is extraordinarily uneven: the top 1,000 words account for roughly 80 per cent of everything said in a normal day.

The numbers, and what they buy you

Vocabulary targets get quoted carelessly, so it is worth being precise about what each one gets you. These are word familiesrun, running and ran count once — and they assume you know them well enough to recognise them at speed, not just to remember having seen them.

~1,000 families
You follow the gist of ordinary speech and can survive a transaction. You will not follow an argument.
~3,000 families
Ordinary conversation works. Most subtitled television is followable. This is the level at which the language starts being useful rather than being homework.
~5,000 families
You read journalism without stopping often. Idiom and register are still ahead of you.
~8,000–9,000 families
You read an unsimplified novel at something close to reading speed. This is the point most learners quietly stop calling themselves learners.

Why the numbers are lopsided in your favour

Word frequency is not evenly spread; it is one of the most skewed distributions in nature. In most languages the top 100 words account for around half of everything said, and the top 1,000 for roughly 80 per cent. The next 1,000 add perhaps another 5 per cent, and it keeps flattening from there.

The practical consequence is that the beginning is steep in the useful direction. Your first thousand words buy you far more comprehension than your fifth thousand will. It also means the last stretch is long: getting from 95 per cent coverage to 98 per cent takes several thousand more words, because you are collecting the rare ones now.

This is why frequency lists are worth caring about, and why a tool that marks statistically uncommon words is more useful than one that marks every word you have not personally seen. Knowing that a word is rare tells you it can wait; knowing it is in the top 1,500 tells you it cannot.

Counting your own, roughly

You do not need a precise figure and no test will give you one. What is worth knowing is your rough band, because it determines what material you should be using.

The quickest honest estimate: open something written for adult native speakers in your target language, read 200 words, and count what you had to look up. Under 2 lookups and the material is too easy to teach you much. Between 2 and 10 is roughly the useful range. Over 20 and you are decoding, not reading, and you will be exhausted before you are educated.

A tool that highlights the words your deck has never met does this continuously rather than as a one-off test — in NoaLingua that is what the unknown-word highlighting on any web page is for. The point is not the number; it is being able to tell, before committing forty minutes to an article, whether it is at your level.

What actually moves the number

Only two things add vocabulary: meeting a word in context often enough that it sticks, and deliberately reviewing it before you forget it. Everything else is a delivery mechanism for one of those two.

Meeting words in context is the pleasant half and it works, slowly, if the volume is high. Reviewing is the unpleasant half and it is far faster per minute spent. Almost every successful self-directed learner ends up doing both, with the material supplying the words and a review system stopping them leaking away — which is the argument for spaced repetition in one sentence.