Stylometric fingerprint

A stylometric fingerprint, also called a writeprint, is the measurable pattern of a writer's unconscious habits, from word choice to punctuation, that stays stable enough across texts to identify authorship, much as a physical fingerprint identifies a person.

The numbers

What a stylometric fingerprint is

A stylometric fingerprint, known in forensic linguistics as a writeprint, is the distinctive pattern left by a writer's unconscious habits: which words they reach for, how they build sentences, how they punctuate, what they consistently get wrong. The premise of stylometry, the statistical study of linguistic style, is that these habits are stable enough within one writer, and variable enough between writers, to serve as identification evidence.

The term writeprint is a deliberate play on fingerprint. It was coined by Li, Zheng and Chen in a 2006 Communications of the ACM article titled 'From Fingerprint to Writeprint', which described authorship identification for digital text such as emails, blog posts and forum messages.

What the fingerprint is made of

Writeprint research groups the measurable features into five families: lexical (vocabulary, word length, letter frequencies), syntactic (sentence structure and punctuation), structural (paragraphing and document layout), content-specific (slang, acronyms, niche terminology) and idiosyncratic (habitual errors and nonstandard usages, which achieve high accuracy when combined with the other families).

Counterintuitively, the strongest signal comes from the most boring words. Analysts deliberately set aside content words, which track topic rather than author, and lean on function words such as 'of', 'upon' and 'though', whose frequencies a writer neither notices nor controls.

Evidence that it works

The classic demonstration is Mosteller and Wallace's 1960s statistical resolution of the 12 disputed Federalist Papers. A 1999 study reported a neural network reaching 70% accuracy on Dutch poems it had not analysed before, and modern systems combine dozens of features rather than trusting any single one.

The best-known forensic case is the Unabomber. FBI linguist James Fitzgerald and sociolinguist Roger Shuy analysed the 35,000-word manifesto published in 1995 and assembled a profile from its quirks: spellings such as 'wilfully' and 'clew', promoted by The Chicago Tribune's mid-century spelling reforms; the Northern American phrase 'rearing children' rather than 'raising children'; and learned words such as 'anomic' and 'chimerical'. When Ted Kaczynski's brother recognised characteristic expressions like 'cool-headed logicians', the linguistic profile matched, contributing directly to Kaczynski's 1996 arrest.

What it cannot do

A writeprint is probabilistic evidence, not certainty. An early method by A. Q. Morton attributed James Joyce's Ulysses to five different authors, and style genuinely shifts with genre, audience and heavy editing. Serious analysis therefore uses many features, sufficient text and honest error bars, and reports likelihoods rather than verdicts.

Worked example

Two writers answer the same question, 'Can you get the report to me by Friday?'. Writer A: 'Should be doable, I'll aim for Thursday night so there's a buffer.' Writer B: 'It is anticipated that the report will be completed prior to Friday.'

The content is identical; the markers are not. Writer A: two contractions, first person, an informal hedge ('should be doable') and the everyday connective 'so'. Writer B: zero contractions, a passive construction ('will be completed'), an impersonal opener ('it is anticipated') and the formal preposition 'prior to'. Five separating habits appear before a single topic word is considered.

A fingerprint is never one marker. It is the accumulation of many small, unconscious choices, which is why stylometric analysis counts dozens of attributes rather than hunting for a single tell.

Sources

More terms

ScriptGrain · Why ScriptGrain · Sample voice profiles · The Journal · Support