The research behind 95%

Read95 is built on one finding from reading research: you can follow a text when you already know most of its words, and the words you do not know are the ones you are in a position to learn. This page sets out what the studies found, where they disagree, and what Read95 does with it.

How much of a text do you need to know?

Researchers call the share of a text's words that a reader knows "lexical coverage". Several studies have measured how comprehension changes as coverage falls.

Batia Laufer (1989) compared learners' comprehension scores with the share of a text's words they knew, and found that about 95% coverage was the point at which reasonable comprehension became likely. Below it, readers more often failed to understand the text.

Marcella Hu and Paul Nation (2000) went further with fiction. They replaced some of a short story's words with invented ones, so that readers met the same text at 80%, 90%, 95% or 100% coverage. At 80%, nobody understood it adequately. At 95%, some readers did. Their estimate for reading fiction comfortably without help was about 98%.

Laufer and Geke Ravenhorst-Kalovski (2010) reconciled the two figures by proposing two thresholds: 95% as a minimum for comprehension, and 98% as the optimum.

Norbert Schmitt, Xiangying Jiang and William Grabe (2011) found no sudden threshold at all. In their study, comprehension rose steadily as coverage rose. That is a useful correction: 95% is not a wall but a point on a slope, where most readers understand most of what they read.

Why 95% and not 98%?

Because the goal is to learn while reading, not only to understand. At 98% or more, reading is comfortable, but there is little left to learn from the page. At 80%, there is plenty to learn, but too little context to work out what any of it means.

At about 95%, around 5 words in every 100 are new to you, unevenly spread through the text. The rest give you enough context to follow the story or the argument, and to guess, check or look up the new words without losing the thread.

How words are learned from reading

People learn many words from reading without studying them, by meeting them in context. Researchers call this incidental vocabulary learning, and it is slow: one encounter rarely teaches a word.

Rob Waring and Misako Takaki (2003) followed learners reading a graded reader. Words met only a few times were rarely remembered three months later; words met more often were far more likely to stay.

Maria Pigada and Norbert Schmitt (2006) followed one learner reading extensively in French for a month. Knowledge of the words grew in spelling, meaning and grammatical use, and the gains were larger for words met more often.

Reviews of extensive reading, the practice of reading a lot of text at a comfortable level, come to a similar conclusion. Richard Day and Julian Bamford (1998), and later Paul Nation and Rob Waring (2020), describe gains in vocabulary, reading speed and comprehension, on condition that learners read a great deal of text they can understand.

The amount matters because the target is large. Nation (2006) estimated that reading a wide range of written English with 98% coverage takes knowledge of about 8,000 to 9,000 word families, a word family being a word with its forms and close relatives.

Why looked-up words come back

Meeting a word again is what fixes it in memory, and when you meet it again matters too. A review of hundreds of experiments by Nicholas Cepeda and colleagues (2006) found that practice spread out over time is remembered better than the same practice done all at once.

What Read95 does with it

• Each text is fitted to your own vocabulary so that you know about 95% of its words. A text is never served if you would know less than 80% of them.
• Read95 estimates your vocabulary from a short calibration at the start, then refines it with every text you read and every word you look up, in both directions.
• You can press and hold any word to see what it means in that sentence, so a new word never stops the reading.
• The words you look up come back for review at widening intervals, until you know them.
• If a text is too hard or too easy, you can say so, and the next ones adjust.

The studies above measured readers in their own conditions, not Read95. What they give Read95 is a target and a direction: keep the text understandable, keep a small share of it new, and make sure new words come back.