People say reality TV is linguistically simple.

It is, and you can put a number on exactly how simple.

A corpus study of every episode of Love Island USA season 4. The 100 most common words are 61% of everything anyone says. Word frequency is near-perfectly Zipfian at R² = 0.98. And the practically useful number: knowing about 1,250 word families gets you 95% of the show, against roughly 3,000 for everyday English. For a language learner, that gap is the whole point.

Read the paper (PDF) or the code and corpus

61%

of all speech comes from just the 100 most common words.

1,250

word families for 95% coverage. Everyday English needs about 3,000 for the same.

3rd–5th

grade reading level, with 97.9% of words among the 3,000 most familiar in English.

R² = 0.98

fit to Zipf's law. The vocabulary is not unusual in shape, only in how little of it gets used.

Start · why bother

"Simple" is a compliment if you are learning the language

The obvious framing is mockery, and it is the least interesting one. The genuinely useful question is the one a language learner asks: how much English do I need before I can follow this? Native-speaker television is the standard advice for intermediate learners and most of it is far too hard.

A corpus answers that directly. Build the vocabulary coverage curve and the answer comes out as a number of words rather than an opinion.

↓ the useful number

1,250 words gets you 95% of it

1,250

word families for 95% coverage. About 2,550 gets you to 98%.

Standard estimates put everyday English at roughly 3,000 word families for the same 95% threshold. This show needs well under half of that. For someone at an intermediate level, that difference is the difference between following along and constantly reaching for a dictionary.

Vocabulary coverage curve showing the percentage of all speech covered as more word families are known, crossing 95 percent at about 1,250 families. — The coverage curve. The 95% line is the one that matters.

Reading-level scores for the corpus placing it around a third to fifth grade level across several readability measures. — Reading level across several standard measures.

↓ the distribution

Textbook Zipf, unusually steep in effect

Word frequency follows Zipf's law almost exactly, at R² = 0.98. That in itself is unremarkable — nearly all natural language does. What is notable is the consequence here: the top 100 words carry 61% of all speech, so the long tail is doing very little work.

Log-log plot of word rank against frequency forming a near-straight line, the signature of a Zipfian distribution. — **Zipf** — a straight line on log-log axes.

Bar chart of the fifty most frequent words in the corpus, dominated by function words and a few show-specific terms. — **Top 50** — function words, and a few the show made its own.

Word cloud of the most frequent content words once stopwords are removed, with relationship and feeling vocabulary dominating. — With stopwords removed, what the season is actually about.

↓ texture

Flat, positive, and full of filler

Lexical diversity sits at a mean MTLD of 59.96 with no trend across the season — episode 1 and the finale are linguistically indistinguishable. Sentiment is uniformly positive and highly subjective, and adjectives make up only 8.4% of speech, which is low for such emotionally saturated content.

Filler words are their own chart, and they are not incidental. This is unscripted speech, which is exactly why it is easy: unscripted speakers reuse a small stock of words and lean on filler to buy time.

Lexical diversity measured per episode across the season, showing a flat line with no upward or downward trend. — **Diversity per episode** — no trend at all.

Frequency of common filler words across the corpus, with like and just among the most frequent. — **Filler** — the signature of unscripted speech.

Sentiment polarity and subjectivity per episode, staying positive and highly subjective throughout the season. — **Sentiment** — positive throughout, and very subjective.

Part-of-speech distribution across the corpus, with adjectives making up about 8.4 percent. — **Parts of speech** — only 8.4% adjectives.

↓ what this cannot tell you

One season, and transcripts are not speech

This is a single season of a single show, so nothing here generalises to "reality TV" as a genre without more seasons to check against.

More importantly, transcripts flatten things that matter. Overlapping speech, prosody, and the pace of delivery are all invisible to a word-frequency analysis, and all three affect how hard something actually is to follow. A show can be lexically easy and still hard to parse at speed. The coverage number is a floor on difficulty, not the whole of it.

Reading-level measures were built for written prose and applied here to conversational speech. They are reported because they are comparable across studies, not because a grade level is a natural unit for dialogue.

Finish · how it was built

Where the numbers come from

The repository ships the cleaned corpus, one .txt per episode plus a combined CSV, so the analysis is checkable rather than merely described. The paper is written in Markdown, converted to LaTeX by a script in the repo, and compiled to PDF.

Corpus: Every episode of Love Island USA season 4, scraped and cleaned, one row per episode
Coverage: Word-family coverage curve against a standard frequency list
Diversity: MTLD, which is stable across differing text lengths in a way type-token ratio is not
Built with: Python, pandas, nltk, textblob, textstat, wordcloud, matplotlib
Figures: Twelve, at 400 DPI, regenerated by two scripts writing to LI_OUTPUT_DIR
Related: The same lens on pop lyrics in Are song lyrics getting simpler?, and turned into a tool anyone can point at a book in the EPUB reading-level analyzer