Research on English-language TV and film estimates that you need about 3,000 word families to know roughly 95% of the words in a typical program, and between 5,000 and 7,000 to reach 98%, depending on the study and the word list (Webb & Rodgers, 2009; Ha, 2022). At 95% you meet an unknown word about once every 20 words. At 98%, about once every 50.
Those figures come with fine print. They are for English, they count names and fillers such as "hmm" as known, and they measure vocabulary, not whether you will understand every scene. This guide shows what each study measured, why the numbers disagree, and how to turn them into a plan. If you prefer Spanish, you can read the Spanish version of this guide.
The Numbers: How Many Word Families for 95% and 98%
Two kinds of studies matter here. Corpus studies take large collections of scripts and ask how many word families you must know to cover a given share of the running words. Comprehension studies test real people at different coverage levels. Start with the corpus studies, all of them on English:
| Study | English corpus | Word lists | 95% coverage | 98% coverage |
|---|---|---|---|---|
| Webb & Rodgers (2009), TV | 88 TV programs, 264,384 running words | BNC families, proper nouns, marginal words | 3,000 families (95.45%) | 7,000 families (98.27%) |
| Webb & Rodgers (2009), movies | 318 movies, 2,841,887 running words | BNC families, proper nouns, marginal words | 3,000 families | 6,000 families (7,000 for British films) |
| Ha (2022), TV | 75,000 episode scripts, 326 million words | BNC/COCA families, proper nouns, marginal words, compounds, acronyms | 3,000 families | 5,000 families |
| Ha (2022), movies | 25,000 scripts, 199 million words | Same | 2,000 families | 5,000 families |
| Ha (2022), soap operas | 22,000 scripts, 101 million words | Same | 2,000 families | 4,000 families |
In both Webb and Rodgers papers, proper nouns and marginal words (hmm, ah, oh) count as known. Genre matters: in the 88-program TV corpus, 95% needed between 2,000 and 4,000 word families and 98% needed between 5,000 and 9,000, depending on the genre. In their movie study, the same 3,000-family mark reaches 95%. Durbahn and colleagues, citing Webb and Rodgers, note that the 1,000 most frequent families already cover more than 80% of the words in movies and TV.
What "Word Family" and "Coverage" Mean
A word family groups a base word with its inflections and regular derivations: inform, informs, informed, informing, information, informative. Knowing 3,000 families therefore means recognizing many more than 3,000 forms. In Nation's lists built from the British National Corpus, the first 1,000 families average about six forms each, falling to about three forms at the 9,000 level (as reported by Schmitt, Jiang and Grabe, 2011).
Coverage is the share of running words you know, counting every repetition. The word "the" counts each time it appears. That is why a few thousand frequent families cover most of any script, and why each extra family buys less: in the TV corpus, going from 3,000 to 7,000 families added about three percentage points (95.45% to 98.27%).
Here is what each coverage level means in unknown words:
| Coverage | Unknown words | In a 3,000-word script |
|---|---|---|
| 80% | 1 in 5 | about 600 |
| 90% | 1 in 10 | about 300 |
| 95% | 1 in 20 | about 150 |
| 98% | 1 in 50 | about 60 |
| 99% | 1 in 100 | about 30 |
The last column is arithmetic on a hypothetical script of 3,000 running words, not a measured episode.
Why the Numbers Differ Between Studies
Webb and Rodgers put 98% of TV at 7,000 families. Ha puts it at 5,000. Three differences explain most of the gap:
- Word lists. Webb and Rodgers used lists based on the British National Corpus. Ha used Nation's BNC/COCA lists, which add American English, and analyzed far larger corpora: about 626 million words from 122,000 TV, movie and soap opera scripts.
- What counts as known. Ha counted proper nouns, marginal words, transparent compounds and acronyms as known. Without them, none of the genres reached 95% at the 5,000-family level.
- Genre. Ha found soap operas less demanding than TV programs and movies (2,000 families for 95% and 4,000 for 98%). Webb and Rodgers found a wide spread across TV genres.
Treat the results as a range: about 2,000 to 4,000 families for 95%, and about 4,000 to 9,000 for 98%, with 3,000 and 5,000 as the central estimates in the most recent study.
Learn Languages with YouTube
Watch videos in your target language, auto-capture words into flashcards, and track your progress — all in one place.
Start Free →95% vs 98%: What Each Level Does to Understanding
The thresholds come from experiments in which people read or listened to texts with a controlled share of words they could not know. The pattern is a curve, not a cliff.
| Study | Material and method | What it found |
|---|---|---|
| Hu & Nation (2000) | A written story; 66 learners in four groups; 80%, 90%, 95% or 100% of words known | No learner reached adequate comprehension at 80%; 3 of 16 did at 90%; 35% to 41% did at 95%. The authors concluded that 98% to 99% is needed to read for pleasure without help |
| Schmitt, Jiang & Grabe (2011) | Two texts; 661 learners from 8 countries | Comprehension rose in a roughly straight line with the share of words known, with no sharp threshold. 98% is a reasonable target for academic texts |
| van Zeeland & Schmitt (2013) | Four spoken informal narratives; native and non-native speakers; 100%, 98%, 95% or 90% known | No statistical difference between 95% and 90%, but non-native listeners varied more at 90%, so 95% is likely needed. 98% suits detailed comprehension |
| Durbahn et al. (2024) | Four 2-minute documentaries (video); native and non-native English speakers; 100%, 95%, 90% or 80% known | Scores fell as coverage fell, and 100% beat 90% and 80%. The authors suggest 95% for optimal and 80% for minimal adequate comprehension |
Hu and Nation never tested 98% directly. They tested four levels and concluded from the trend that 98% to 99% would be needed. Schmitt and colleagues later found no sharp threshold at all, only steady improvement as the share of known words rose.
A reasonable reading of these studies, which is an interpretation and not a finding: below about 90%, following a story is hard work; around 95%, you can follow the main line with frequent gaps; around 98%, the gaps are small enough to guess from context. The same logic sits behind the 98% rule for reading, which we cover in extensive vs intensive reading.
Video is the one case with its own data. In Durbahn et al. (2024), average comprehension scores stayed above 70% in both groups even at 80% coverage, and the authors credit images with helping viewers fill gaps. Treat this with care: the clips were four two-minute documentaries, not forty-minute episodes of a fast comedy.
What These Numbers Do Not Tell You
- They are for English. Other languages need their own corpus counts; these figures do not transfer.
- They count words on paper. The corpora are scripts. Speech speed, accents and overlapping dialogue are not part of the count.
- Knowing every word is not understanding every scene. Webb (2021), cited by Durbahn and colleagues, stresses that 100% coverage does not guarantee full comprehension.
- They ignore your first language. The corpus counts are not adjusted for it, and some words are easier for you than their frequency suggests.
How to Use This: A Practical Plan
- Measure a real scene. Open the subtitles of a scene from a show you want to watch. Count 100 consecutive words and mark the ones you could not explain. Five or fewer means you are near 95%; two or fewer, near 98%. Repeat with three scenes, because coverage varies by scene. This is a rough self-check, not a test.
- Pick material near your range. If you find more than ten unknown words per 100, start with a less demanding genre. Webb and Rodgers found that the vocabulary load changes a lot between TV genres.
- Learn frequent words first. The first 3,000 families carry most of the coverage, so frequent words repay study fastest. Rare words appear too seldom to deserve a flashcard unless you keep meeting them.
- Keep lookups fast. Pausing for a dictionary breaks the scene. The Metheus browser extension is free for Chrome, Firefox and Edge. On YouTube, Netflix, Amazon Prime Video, Disney+ and Viki it adds interactive subtitles and word mining; on other pages you start it from the extension popup. Our guide to learning languages with Netflix for free walks through the setup.
- Review what you mine. Words that blocked you in a scene become cards with spaced repetition (FSRS in Metheus), so the vocabulary you meet in shows is reviewed before you forget it.
Vocabulary is one input among several. Whether input alone is enough is a separate question, and we look at the evidence in is pure comprehensible input enough for fluency.
Sources
- Webb & Rodgers (2009), "Vocabulary demands of television programs," Language Learning 59(2), 335-366.
- Webb & Rodgers (2009), "The lexical coverage of movies," Applied Linguistics 30(3), 407-427.
- Ha (2022), Frontiers in Psychology 13, 831684.
- Hu & Nation (2000), Reading in a Foreign Language 13(1), 403-430.
- Schmitt, Jiang & Grabe (2011), The Modern Language Journal 95(1), 26-43.
- van Zeeland & Schmitt (2013), Applied Linguistics 34(4), 457-479.
- Durbahn, Rodgers, Macis & Peters (2024), Studies in Second Language Acquisition 46(4), 1045-1068.
Create your free Metheus account to mine words from the shows you watch and review them with spaced repetition. The free plan includes immersion, mining, spaced repetition, your own decks, the full extension and module 1 of every course.
FAQ
How many words do you need to understand English TV shows?
Research estimates about 3,000 word families for roughly 95% of the words in TV dialogue and about 5,000 to 7,000 for 98%, depending on the study and word list (Webb & Rodgers, 2009; Ha, 2022). These figures count proper nouns and filler words as known.
Is a word family the same as a word?
No. A word family groups a base word with its inflections and regular derivations, such as inform, informs, informed, informing, information and informative. Knowing 3,000 families means recognizing many more than 3,000 individual forms.
Is 95% coverage enough to follow a show?
It depends on what you need. Studies find comprehension rising steadily with coverage rather than jumping at one line. In a video study, 95% was suggested for optimal comprehension and 80% as a minimum for short documentaries (Durbahn et al., 2024), but longer, faster shows will be harder.
Do these numbers apply to languages other than English?
No. The studies analyzed English scripts and English word lists. Other languages need their own corpus counts, so these figures should not be copied over.
2026 Immersion Report
Language learning in 2026 isn't about finishing daily streaks. The fastest path to fluency combines three elements: comprehensible input from real media you enjoy, active vocabulary capture while consuming that content, and spaced repetition to lock words into long-term memory. Gamified apps build the habit. Immersion tools build the fluency. The most effective learners use both—in that order.
Read the Full 2026 Comparison GuideStart with Metheus
Watch videos in your target language, auto-capture words into flashcards, and track your progress — all in one place.
Get Started Free →
