The 98% rule: how many unknown words can a page take?

Open a book in a language you’re learning and count the words you don’t know on one page. If there are five, you’ll probably enjoy it. If there are twenty-five, you probably won’t, even though the difference looks small: 98% of the words known against 90%.

That gap is one of the most useful findings in research on reading in a second language. It explains why some books teach you and others just exhaust you, and it gives you a quick way to tell them apart.

What “coverage” means

Researchers call the percentage of words in a text that a reader knows lexical coverage, or just coverage. It’s counted in running words: every word on the page, including repeats. If a 250-word page contains five words you don’t know, your coverage of that page is 98%.

Here’s what different levels of coverage mean in practice:

Coverage	Unknown words	On a 250-word page
98%	1 in 50	about 5
95%	1 in 20	about 12
90%	1 in 10	about 25
80%	1 in 5	about 50

The percentages sound close together. On the page, they are worlds apart.

Try it yourself

Here’s a 50-word paragraph we wrote for this demonstration, with nonsense words standing in for words a reader doesn’t know.

98% coverage (one unknown word in fifty):

The rain had not stopped for three days. When the old sailor finally came down to breakfast, he stood at the window and stared at the empty brennick. My mother put a bowl of hot soup in front of him. He ate it slowly, without a word, watching the door.

You don’t know what a brennick is, but you don’t need to. It’s something empty that you can see from a window: a harbor, a street, a beach. You understood the paragraph, and you might even have learned a word.

90% coverage (five unknown words):

The rain had not flenned for three days. When the old tobbler finally came down to breakfast, he stood at the window and plinted at the empty brennick. My mother put a bowl of hot soup in front of him. He ate it wanely, without a word, watching the door.

You still get the outline. But notice what happened to your reading: you slowed down, you started guessing, and the guesses depend on each other. Is a tobbler a person? Probably, since he eats soup. Did he plint at the brennick, or does plinted mean something else entirely?

80% coverage (ten unknown words):

The skarn had not flenned for three days. When the old tobbler finally came down to morrock, he gromped at the window and plinted at the empty brennick. My mother put a crull of hot soup in front of him. He ate it wanely, without a word, brinting the door.

Now the paragraph is a puzzle. You can work out that someone came down and that there was soup, but most of the scene is gone. And this is the easy version: you’re a fluent reader of English, so the grammar and the known words cost you nothing. A learner reading at 80% coverage is also working hard on the words they do know.

What the research found

Hu and Nation: about 98% for reading without help

The best-known study is by Marcella Hu and Paul Nation (2000, “Unknown vocabulary density and reading comprehension”). They worked with 66 adult learners of English on a pre-university course. Each read one version of a short story: one in which they would know every word, and others in which nonsense words replaced the rarest words so that only 95%, 90% or 80% of the running words were known, much as in the demonstration above. The learners then answered comprehension questions and wrote down what they could recall.

The results were stark. With every word known, 15 of 17 readers reached the level the researchers counted as adequate comprehension. At 95%, 6 of 17 did. At 90%, 4 of 16. At 80%, none of the 16.

No group actually read a 98% version. Projecting from the trend in their results, Hu and Nation estimated that most learners need to know about 98% of the running words, roughly one unknown word in fifty, to understand fiction adequately without help. That estimate is where the “98% rule” comes from.

Laufer: 95% as a minimum

Earlier, Batia Laufer (1989, “What percentage of text-lexis is essential for comprehension?”) had studied university students reading academic English and proposed 95% as the point at which reasonable comprehension becomes likely. Later, with a much larger group of 745 students, Laufer and Ravenhorst-Kalovski (2010, “Lexical threshold revisited: Lexical text coverage, learners’ vocabulary size and reading comprehension”) suggested that the two figures can live together: about 95% as a minimal threshold, for reading with some guidance, and about 98% as an optimal one, for reading independently.

Schmitt, Jiang and Grabe: a slope, not a cliff

Another large study, by Norbert Schmitt, Xiangying Jiang and William Grabe (2011, “The percentage of words known in a text and reading comprehension”) tested 661 learners on two academic-style texts, measuring for each reader both how many of the words they knew and how well they understood. They found no threshold at all. Between 90% and 100% coverage, comprehension rose steadily with every extra percentage point of known words:

Words known	Average comprehension score
90%	about 50%
95%	about 60%
98%	about 68%
100%	about 75%

Two lessons follow. There’s no magic number at which reading suddenly works; the right target depends on how much you need to understand. And knowing every word isn’t the same as understanding everything, which is a reminder that grammar, background knowledge and reading skill matter too.

Put together, the research gives a reasonable rule of thumb rather than a law: around 98% for comfortable, independent reading; around 95% for reading that’s possible with effort and help; below that, most readers struggle.

Why coverage decides whether reading teaches you

Coverage matters for learning, not just for understanding.

You learn new words from the words around them. At 98%, each unknown word is surrounded by dozens of known ones, so its meaning is often clear or at least guessable, as with brennick. At 90%, the unknown words start surrounding each other, and guessing becomes guesswork.

Easy reading is fast reading, and fast reading is more reading. A page you can read without stopping might take a couple of minutes. A page with 25 unknown words might take ten, most of it spent looking things up. The learner at 98% meets far more language in an hour.

Pleasure keeps you going. Reading at 98% feels like reading. Reading at 90% feels like homework, and few people keep up homework they’ve set themselves for long.

There’s a balance here. A text in which you know 100% of the words is excellent for building speed, but it has no new words to teach you. A text at about 98% gives you a few new words on every page, with enough context around each one to learn from. That makes it a good target for learning from reading.

What the 98% rule doesn’t say

It’s an estimate, not a law. Hu and Nation projected the 98% figure from the trend in their results rather than testing it directly, and the research behind it involves a small number of studies, mostly with learners of English, specific texts and specific comprehension tests.

Coverage isn’t everything. Grammar, background knowledge and interest all affect comprehension. A story you already know is easier at the same coverage than one you don’t.

Help changes the picture. Hu and Nation’s figure is for reading without assistance. With glosses or a dictionary, you can read text at lower coverage, more slowly, which is fine for short study sessions. It’s just not the kind of reading you can do for an hour.

Not all words are equal. Names are easy to spot and usually don’t count. A word that appears twenty times in a book becomes known after the first few. And knowing teach usually helps with teacher and taught, which is why researchers often count word families rather than individual forms.

How to use the rule

Use it as a quick test before you commit to a book:

1. Pick a page from the middle, not the opening, which introduces everything at once.
2. Count the words you don’t know. Count repeats once and skip names.
3. Read the result. On a page of about 250 words, five or fewer unknown words means you can read the book comfortably. Six to twelve means it’s workable with help. More than that, choose an easier level for now.

We go into more detail, with other quick tests, in how to choose your reading level. And if you’re wondering how big a vocabulary you’d need to reach 98% coverage of an unsimplified novel, see how many words you need to read a novel.

How Lexo helps

Lexo is an iPhone and iPad app that rewrites classic novels at five CEFR levels, from A2 to C2, in the language you’re learning, so you can choose an edition whose vocabulary matches yours. You pick the level per book. If a book turns out to be denser than you’d like, you can add another level of it for $1.99 at planned launch pricing.

For the 2% you don’t know, tap the word. Lexo shows what it means in that sentence, its base form, the phrase it belongs to and a translation of the whole sentence into your own language, and plays the word aloud. A sentence-by-sentence mode helps with the occasional dense passage. Lexo is in beta and launching on the App Store soon. Get notified at launch, or read about the CEFR levels.