How many words you need before the news stops hurting

It is the most asked question in language learning and it usually gets a useless answer. Somebody says a thousand words is enough for a conversation, somebody else says three thousand, and neither of them says what “enough” means or enough for what.

There is a real answer, and it comes from counting. Paul Nation ran the numbers across novels, newspapers and spoken corpora, and if you accept 98% coverage as the point where you can read comfortably without help, the figures are:

• 8,000 to 9,000 word families for written text
• 6,000 to 7,000 word families for spoken text

Both are larger than most learners expect. The second is smaller than the first, which is the surprising part, and we will come back to it.

What a word family is, because it changes the number

Nobody counts individual words, because it would be misleading. Research counts word families: a base word together with its inflections and the derivations you can work out once you know the base.

So govern, governs, governed, governing, government, governments and governor are one family, not seven. If you know the base and the pattern, the rest come nearly free.

This matters twice. It makes the number smaller than a raw word count would suggest, and it makes those “learn 5,000 words” figures on the internet incomparable to these, because they usually count something else. When you see a vocabulary target quoted, the first question is what the units are, and most of the time nobody can tell you.

The shape of the problem

Here is the coverage you get for a given vocabulary size, from Nation’s counts of Lady Chatterley’s Lover and of newspaper text.

The same numbers, since a chart is no use to anyone who cannot see it:

The cruel arithmetic

Look at what the curve does.

Your first 2,000 word families buy you 83 to 88% of everything. That is an extraordinary return, and it is why beginners feel fast progress: a few months of work and most of the words on any page are already familiar.

Then it stops. Going from 2,000 to 4,000 adds about seven points. Going from 4,000 to 9,000, which is more than doubling your vocabulary, adds about three.

So the last stretch is slow in a way the first is not, and it is exactly the stretch that decides whether reading is enjoyable. At 4,000 families you are at 95%, which is one unknown word in twenty, roughly one per two lines. That is the level where many readers struggle. At 9,000 you are at 98%, one in fifty. The difference between those two experiences is enormous and the vocabulary difference is 5,000 families of increasingly rare words.

This is the real reason people give up on reading in another language. Not laziness. A genuinely flat part of the curve, right where it matters.

Speech needs fewer words than print

The spoken figure is 6,000 to 7,000 against 8,000 to 9,000 for writing, and that gap is worth acting on.

Speech recycles a smaller set of words. People writing choose from a wider vocabulary, take more time over it, and are not constrained by whether a listener can follow in real time. So the pool you need for a podcast is meaningfully smaller than for a novel.

Which means audio is often the easier on-ramp, not the harder one, even though most learners assume the reverse. If listening feels harder than reading, that is usually about speed and about not recognising words you already know when spoken, rather than about vocabulary you lack. Those are different problems with different fixes.

What to do instead of waiting

The numbers are not a target to hit before you start. Treat them as a map.

Frequency order is the whole game early on. The first 2,000 families are worth more than the next 6,000 combined for coverage. Anything that gets you those quickly, graded material, a frequency list, a beginner course, is time well spent. Past that point, frequency lists stop paying and reading takes over.

After about 3,000, volume beats study. The words that take you from 95% to 98% are too numerous and too infrequent to drill. You meet them by reading a lot, which is also the only way you learn which sense each one carries.

Do not wait for 9,000. That figure is for reading unassisted, with no way to resolve a word you do not know. With instant lookup the coverage you need for a text to be workable drops a long way, and the reading itself is what carries you up the curve.

Narrow the topic and the number falls. These figures are for general text. Within one subject the recurring vocabulary is far smaller, which is why people can often read about their own field long before they can read a novel.

The short version

Eight to nine thousand word families to read comfortably without help. Six to seven for speech. Your first two thousand do most of the work, the rest arrives slowly, and the slow part is the part that makes reading pleasant.

Nobody gets there by waiting. They get there by reading things they could not quite read yet.

---

Sources: Nation, I.S.P. (2006), How large a vocabulary is needed for reading and listening?, Canadian Modern Language Review 63(1), including the coverage figures for Lady Chatterley’s Lover and newspaper text; Hu, M. and Nation, P. (2000), on the 98% threshold.