The 98% rule

Have you ever opened a novel in the language you’re learning, recognized most of the words on the first page, and still felt like you were wading through mud? You’re not imagining it. “Most of the words” turns out to be a lot less than you need.

Researchers have a name for the share of words in a text that you know: lexical coverage. And one of the most useful findings in reading research is how sharply the experience of reading changes as coverage climbs from 90% to 95% to 98%.

What coverage means

Coverage is counted over running words, every word on the page in order, not over the dictionary. If a 200-word passage contains four words you don’t know, your coverage is 98%. If it’s twenty, it’s 90%.

That sounds like a small difference. On the page, it isn’t.

At 98%, an unknown word is an occasional bump. At 90%, there’s one in nearly every line, and you spend more energy decoding than reading.

What the research found

In 1989, Batia Laufer proposed that around 95% coverage was needed for reasonable reading comprehension. A decade later, Hu Hsueh-chao and Paul Nation (2000) tested this more directly. They took a short fiction text and replaced different proportions of its words with made-up words, creating versions at different coverage levels, and then measured how well learners understood each version. Their results suggested that most learners need about 98% coverage to understand a fiction text adequately without help, and that at 95% only some managed it.

Later work refined the picture. Laufer and Tamar Ravenhorst-Kalovski (2010) proposed two thresholds: around 95% as a minimal level for reading with some support, and 98% as an optimal level. Norbert Schmitt, Xiangying Jiang and William Grabe (2011) found that comprehension rises fairly steadily with coverage rather than switching on at a single magic number. Every extra percentage point helps.

So “the 98% rule” is best understood as a guideline, not a law: around 98% is where unassisted reading becomes comfortable for most people, and comprehension falls away gradually below it.

How many words is that?

In a well-known 2006 paper, Paul Nation used large collections of real text to estimate how big a vocabulary you need to reach 98% coverage. His estimate: roughly 8,000 to 9,000 word families for a wide range of written text, and around 6,000 to 7,000 for spoken language such as conversation and films.

A word family is a base word plus its related forms: read, reads, reading, reader, readable and so on. So the number of individual forms you’d recognize is larger, but you don’t learn each one from scratch.

That might sound daunting. But you don’t need 98% coverage of everything to enjoy reading something. Coverage depends on the text: a blog about a hobby you know inside out might hit 98% long before a literary novel does. And it grows with every page you read.

Why the last few percent matter so much

Three things make high coverage disproportionately valuable:

1. Context needs company. You can usually guess an unknown word from the words around it, but only if you know those words. At 98% the surroundings are solid; at 90% the clues themselves are often unknown.
2. Attention is limited. Every unknown word takes a little working memory. A few per page is fine. Thirty per page means you lose the thread of the sentence you’re in.
3. Enjoyment drives volume. At high coverage, reading is pleasant, so you read more, and reading more is how coverage rises. At low coverage it’s exhausting, so you stop.

Caveats worth knowing

• Background knowledge counts. Knowing the topic, the genre or the plot makes lower coverage much more manageable.
• Not all unknown words are equal. A name or a word whose meaning is obvious from the picture costs less than a key verb.
• Assisted reading changes the maths. If looking up a word takes a second, you can read comfortably at somewhat lower coverage. You’ll be slower, but you’ll understand.
• Listening is different. Research on listening is thinner, and some of it suggests that slightly lower coverage can be enough for informal speech.

Putting it to work

A practical approach is to match coverage to the kind of session you want:

• 95–98%+ for relaxed, extensive reading: lots of pages, few lookups, maximum enjoyment.
• 90–95% for focused sessions with a look-up tool, where you’re happy to slow down and learn new words.
• Below 90% only for short texts you’re very motivated to read.

How Lumo uses coverage

Lumo keeps track of the words you know, and for every text in your library it works out what share of the words on the page are already known. That becomes a simple label: Easy at 98% and above, Just right from 90% to 98%, Challenging from 80% to 90%, and Hard below that. When Lumo recommends what to read next, it aims for the sweet spot around 95%.

Because turning the page marks the words you didn’t tap as known, the estimate grows with you. Texts that came up Hard a few months ago quietly become Just right.

That’s the hopeful part of the 98% rule. The way to get to 98% is to keep reading at 95%.