N-gram: the word runs behind stock-phrase detection
An n-gram is a sequence of n items in a row, and in text analysis the items are usually words. "Fast-paced world" is a trigram (three words); "it is important to note" is a five-gram. Count how often each n-gram occurs across a large body of text and you have a map of which phrases are common and which are rare. Humanize 360 uses n-gram counts in its stock-phrase density signal, one of the eleven behind the Human Score, and in the simplest form of predictability measurement.
Take the sentence "In today's fast-paced world, effective communication is crucial." Its trigrams include "today's fast-paced world", "fast-paced world effective" and "communication is crucial". Two of those appear constantly in machine output and marketing copy; the middle one appears almost nowhere, because it straddles a comma. Now take "My aunt runs the only pharmacy in Jinja that stays open past nine." Its trigrams are almost all rare. The second sentence is not better. It is just not made of pre-assembled parts.
Why n-grams catch formulaic writing
Language models learn which word sequences are likely, and when they write they reproduce the likeliest ones. So the same trigrams and four-grams turn up across millions of documents: "plays a crucial role", "it is worth noting", "in the realm of", "a testament to". None is wrong. All are pre-fabricated. A detector with a list of frequent machine n-grams can measure how much of a text is built from them, and a human editor can do the same by ear.
Three sizes and what each shows
- Unigrams (single words): vocabulary spread and tell-word ratio, such as "foster" or "tapestry" appearing more than chance suggests.
- Bigrams and trigrams (two and three words): stock phrases, transition patterns and fixed openers like "Additionally, the".
- Longer n-grams (four or more): whole clichés and, in plagiarism checking, verbatim copying from a source.
How Humanize 360 uses them
The stock-phrase density signal scores how much of your text falls into frequent formulaic n-grams. The tell-word ratio is a unigram measure. Opener repetition looks at the n-grams that start sentences. The humanizer, in every mode, replaces flagged phrases with the plain thought they were standing in for, and the Changes view shows each replacement so you can veto one that mattered. Our plagiarism checker uses long n-grams for a different purpose: finding matching runs in an external index.
Limits of the method
N-gram counting knows nothing about meaning. "Plays a crucial role" may be exactly the right phrase in one sentence, and a technical term that happens to be a common trigram in your field will register as stock. That is why it is one signal among eleven and why the breakdown shows it separately: you can look at the flagged phrases and decide whether they were yours or the machine's.
Common questions
What is the difference between a bigram and a trigram?
A bigram is two consecutive words, a trigram is three. Longer runs are named by number: four-gram, five-gram. All are n-grams.
Are n-grams used by plagiarism checkers too?
Yes. Plagiarism checkers look for long matching runs of words between your text and an index of sources. AI detectors look for short runs that are common in machine output. Same tool, different question.
Does using common phrases make my writing look like AI?
A few do not matter. A high density does, because it is the pattern models produce. The stock-phrase reading in our breakdown tells you whether yours is high.
Can I see which phrases were flagged?
The detector reports the stock-phrase density signal; the humanizer's Changes view shows every replaced phrase side by side with the original.