Comprehension runs on coverage, not on knowing every word
The useful way to think about understanding a text is lexical coverage: the share of the running words on the page that you already know. Comprehension tracks coverage closely, and vocabulary researchers have spent decades pinning down how much you need. Two figures come up again and again. About 98% coverage is the level at which most readers can follow a text on their own, for pleasure, without help. Around 95% is the floor for what researchers call minimal or reasonable comprehension.
It helps to feel what those percentages mean. At 95% coverage you still meet an unfamiliar word roughly once every twenty words, which is often enough to slow you down; at 98% it is about one in fifty. The gap sounds small but changes the experience of reading a great deal, which is why difficulty matching matters so much when you are learning from real content.
Reading takes more words than listening
How big a vocabulary do those coverage levels require? For written text, the most cited estimate is that reaching 98% coverage of novels and newspapers takes a vocabulary of about 8,000 to 9,000 word families. A word family is a base word plus its inflections and common derivations, so it bundles help, helps, helped, helping, and helper into one; that makes a family count much smaller than a raw count of the words in the same text.
Speech is gentler. The same research puts 98% coverage of spoken English at about 6,000 to 7,000 word families, noticeably fewer than reading. Conversation recycles the most common words more heavily and reaches for rare ones less often, so a given vocabulary carries you further when you are listening than when you are reading. That is part of why understandable television and conversation are such efficient input.
| Category | Word families |
|---|---|
| 80 to 87% of writing | 2,000 |
| 98% of speech | 6,000 to 7,000 |
| 98% of writing | 8,000 to 9,000 |
The first few thousand words do most of the work
The reason frequency matters is that coverage does not climb evenly. The most common words are common everywhere, so they pay off enormously. The first 1,000 word families alone cover around 85% of everyday speech and roughly 80% of written text. The first 2,000 families take written coverage to somewhere between 80 and 87%, depending on the genre and how you count.
After that, each additional thousand words adds less and less, because you are into progressively rarer vocabulary. This is the whole case for learning words in rough frequency order rather than at random: a few thousand of the right words unlock most of what you read and hear, and the long tail is where the effort stops paying back quickly. It is also why a show pitched at your level, where you already know most of the words, is worth far more than one where you do not.
Nation, 2006
What this means for CEFR levels
It is tempting to convert this into a target per level, and some researchers have tried. Using a 5,000-word recognition test, one influential line of work measured the average vocabularies of learners sitting at each CEFR level: very roughly, under 1,500 words at A1, about 1,500 to 2,500 at A2, 2,750 to 3,250 at B1, 3,250 to 3,750 at B2, 3,750 to 4,500 at C1, and 4,500 to 5,000 or more at C2.
Two cautions matter here, and we state them plainly. These are measured averages of real learners, not official requirements: the Common European Framework deliberately defines its levels by what you can do, not by any word count. And because the test tops out at 5,000 words, it cannot see past that ceiling, so the advanced figures are best read as at least. One technical wrinkle is worth naming too: this test counts individual words, while the coverage figures above count word families that bundle a word with its inflections, so the two scales are related but not directly interchangeable. Treat the numbers as a rough map, not a finish line.
Nation, 2006