Stylometry

Also called: computational stylometry, stylistic analysis

Stylometry is the statistical measurement of writing style — sentence length, function-word frequency, punctuation habits, hedging and vocabulary distribution — used to characterise or identify an author from text alone.

It is a century-old method from authorship attribution, now used to build machine-readable descriptions of how one person writes.

Stylometry predates language models by decades. Its classical use is authorship attribution: deciding which of several candidate authors wrote a disputed text, by comparing distributions of features the author does not consciously control. Function words carry most of that signal precisely because nobody chooses them deliberately.

The same measurements serve a newer purpose. If a set of features characterises an author well enough to identify them, the same features can be handed to a language model as a target to write toward. That is the bridge between stylometry as forensics and stylometry as personalisation.

Feature familyExampleWhy it carries authorial signal
Function wordsthe, and, of, but, ratherChosen unconsciously; frequency is stable per author
Sentence length distributionmean and variance of words per sentenceRhythm is habitual and hard to fake deliberately
Hedges and boostersperhaps, might / clearly, certainlyEncodes how strongly an author commits to claims
Punctuation habitssemicolon rate, dash rate, comma densityLearned early, rarely revised
Vocabulary richnesstype-token ratio and variantsReflects reading history and domain
Feature families used in stylometric analysis

Frequently asked questions

Is stylometry accurate enough to identify an author?
In controlled authorship-attribution settings with enough text per candidate, function-word-based methods perform well; the open stylometric tooling built by Eder and colleagues is the standard reference implementation. Accuracy falls as the candidate pool grows and the text per candidate shrinks.
What is the difference between stylometry and AI detection?
Stylometry asks which author wrote this; AI detection asks whether any human wrote it. They share some features, such as sentence-length variance, but answer different questions and fail in different ways.
Can stylometry work across languages?
The method transfers, but the word lists do not. Function words, hedges and boosters are language-specific, so a stylometric system covering several languages needs a separate list per language rather than a translated English one.

Read further

Sources and how to cite this page

Cite this page

MyWritingTwin. “Stylometry.” Last updated 2026-09-15. https://www.mywritingtwin.com/glossary/stylometry