7 Hours of TV, One Word Ahead of the Group That Watched Nothing

The twenty minutes with the subtitles off

Pick an episode you have watched at least three times. Turn the subtitles off and play the first twenty minutes.

Then try to say it in English: what the main character asked, and how the other person answered.

Most people cannot manage more than a sentence or two. Some cannot recall what the question was — in an episode they have, by their own account, seen three times.

This is not a memory problem. In 1991 a group of Belgian researchers did something very plain with an eye tracker: they sat people in front of subtitled television and recorded where the eyes went. Reading the subtitles, they found, is close to compulsory. You do not decide whether to read them. They appear, and your eyes are already there. It happens when the soundtrack is perfectly easy to follow, and it happens when you have told yourself not to look.

The sound is not being ignored. The same researchers later found that adults do make an effort to follow the speech.

But the meaning is not arriving through the speech. It has already been delivered by that line of text, and the listening is running alongside. Your ears were never switched off. They were just no longer needed.

Admission to one episode costs five to seven thousand word families

So how much vocabulary does it take to follow an episode without that line of text?

Applied linguistics has counted this more than once. A 2009 study broke down 88 television programmes — more than 260,000 running words of dialogue. Knowing the most frequent 3,000 word families, plus the proper nouns in the show, covers 95% of what is said. Reaching 98% takes 7,000 word families.

In 2022 someone redid the count on a far larger corpus — soap operas, TV programmes and movies, more than 600 million words between them — and landed a little lower: 3,000 word families for 95% of TV dialogue, 5,000 for 98%. An earlier study of spoken English in general put the figure at 6,000 to 7,000.

Three studies, three corpora, and the 98% line lands in the same band each time: five to seven thousand word families.

Fairness demands a caveat. The authors of the 2009 study preferred the 95% line — they concluded that 3,000 word families may be enough for adequate comprehension of television. The reason for using 98% here is a much-cited reading experiment which found that at 95% coverage, more readers failed to understand the text than understood it, and inferred a threshold somewhere around 98%.

An inference is all it is. That experiment tested 80%, 90%, 95% and 100% — never 98% — and later attempts to replicate it have not fully reproduced the result. Taking 98% is the strict reading, not the settled one.

The unit matters too. A word family here is a stem with all its inflections and regular derivations: add, added, adding, addition, additional, additionally — those six count as one. China’s college entrance exam expects around 3,500 words and CET-6, the higher of the two national English tests most Chinese undergraduates sit, around 5,500. Both count dictionary entries, and those lists already contain plenty of words that are derivations of each other.

Convert 5,500 entries into word families and the number can only shrink. The gap is wider than it looks.

Put that back on one episode. Those 88 programmes averaged a little over three thousand running words each. At 95% coverage, roughly 150 words in an episode are ones you do not know. At 98%, there are still about 60.

And those 60 do not wait. They go past at conversational speed, buried in linked speech, swallowed syllables and a music cue.

Seven hours, six words — and the people who watched nothing got five

You do not have to understand it all, people say. You pick it up as you go.

A 2020 study measured exactly that. One hundred and eighty-seven Japanese university students watched ten episodes of the same American drama, 42 minutes each, more than seven hours in total. The researchers picked 60 low-frequency words out of the show and tested them before and after.

The average gain was 6.4 words.

The group that matters is the other one. Seventy-three students at the same university watched no episodes at all. They simply sat the same test again after the same interval. They “learned” 5.3 words.

The net return on seven-plus hours was about one word.

Where did that 5.3 come from? The test was multiple choice with seven options. On a second pass, elimination and luck move the score up a little on their own. That portion has nothing to do with watching anything, and it swallows almost the entire gain.

The one word left over is a flattered one. Those 60 targets were hand-picked, appearing an average of ten times across the ten episodes, and the students already knew more than half of them going in — one of the targets was known by 182 of the 187 beforehand. The words you actually collide with while watching mostly appear once or twice, and nobody offers you seven options.

Run the generous figure forward — the following is a hypothetical, not a statistic. Say you comfortably know 3,000 word families and want 6,000. Using 6.4 words per seven hours, not the net one, that is 3,300 hours, or about 4,700 episodes. One episode a day, thirteen years.

In fairness, the researcher’s own conclusion is less bleak than that: he calculates that picking up vocabulary by watching is about as efficient as reading a novel, maybe slightly better. He is right, and what it means is that both routes are too slow to be the main one — not that watching is uniquely bad.

And at the end of thirteen years you can follow a television show. Not speak.

The Canadian children who listened for a decade and still could not say it

The hardest evidence that input alone is not enough did not come from a lab. It came from Canadian schools.

From 1965 onward, Canada ran French immersion: English-speaking children taking maths, science and social studies in French from their first years of school. That is about as much input as a person can realistically absorb — six hours a day, for more than a decade.

The researcher Merrill Swain assessed these students in the 1980s. What she found became a turning point in the field. Their listening and reading had come close to native level, and yet the French they produced still carried errors a native speaker would not make — and kept carrying them, year after year.

Swain’s explanation was that what they lacked was not input but occasions to be pushed into producing language. Listening lets you ride on context, tone and the picture; you can recover the meaning without ever resolving the grammar. Only when it is your turn to say the sentence are you forced to decide which form that verb takes and how the clause attaches.

Children with a decade of immersion did not cross that line by listening. An adult watching two episodes a week is not going to cross it either.

“I learned English from TV” has the causation backwards

And yet people really do learn this way. Every school has a few of them, and they are not lying.

The problem is which direction the arrow points.

To get anything out of a show you have to roughly follow it — which means your vocabulary is already up in the thousands of word families. Watching television did not make their English good. Their English got good enough first, and then television started paying off. Everyone below that line is reading subtitles, and nobody in that group ever writes “I learned English from TV,” because they did not. We only hear from one side.

There is a second layer that tends to go missing: what those people were doing was often not watching.

Replaying the same episode, pausing to repeat a line out loud, copying the intonation, memorising a whole exchange and performing it — that still looks like watching a show, but it is production practice. That part is what worked. It is also the part that drops out of the story, leaving “watching TV” to collect the credit on its own.

The idea has lasted because it has all three things it needs: it dresses study up as entertainment, so it costs nothing to start; it cannot be falsified, since anyone who did not learn simply did not watch enough; and it supports a business.

In early 2021, police in Shanghai shut down the best-known Chinese fan-subtitling operation for American shows; the announcement listed more than eight million registered members. Eight million people queuing for the one thing that turns listening back into reading.

Back to those twenty minutes.

Not being able to say it with the subtitles off does not mean the last decade was wasted. It means one thing only: you practised recognition, and you did not practise listening or speaking, and the three grow separately.

Television is good. It deserves to be enjoyed, and once you have a base it really will push your vocabulary up a little. What it will not do is the hard part — producing a whole sentence while somebody waits for it.

There is only one way to practise that: actually saying things, with someone on the other end. That is why we built Unscript — fix the conversation, then pull the script away one level at a time, so that speaking does not require five thousand word families first, or a supply of nerve.

Tonight, leave the subtitles on. Watch it as a show.