Stylometry in Action: Famous Cases and Forensic Applications

In July 2013, the Sunday Times received an anonymous tip: the debut crime novel The Cuckoo's Calling, published under the name Robert Galbraith, was actually written by J.K. Rowling. The newspaper needed proof. They turned to Patrick Juola, a computational linguist at Duquesne University, and asked him to run a stylometric analysis.

Juola compared the novel against works by Rowling and several other female British crime writers. His software measured function word frequencies, word length distributions, character n-grams, and hundreds of other features the authors never consciously controlled. The result was unambiguous: across every dimension of analysis, the Galbraith novel clustered with Rowling's known writing, not with any of the comparison authors. Within hours of the analysis becoming public, Rowling confirmed it. Her cover was blown by the statistical fingerprints she didn't know she was leaving.

The Rowling case made headlines because the author was famous. But it was far from the first time stylometry settled a real-world question of authorship. The field has a long history of practical application, from criminal investigations to literary scholarship to intelligence work. The foundational 1964 Mosteller-Wallace study on the Federalist Papers established that statistical analysis of writing patterns could succeed where conventional scholarship had failed for over a century. (For a full treatment of the underlying science, mathematics, and key researchers, see our deep dive on voice fingerprint analysis.)

This article covers the other side of the story: what happens when stylometry leaves the lab and enters the real world.

Famous Cases: Stylometry's Greatest Hits

The Unabomber Manifesto

In 1995, the FBI faced a dilemma. Ted Kaczynski, not yet identified, had been sending increasingly sophisticated mail bombs for seventeen years. He offered to stop if major newspapers published his 35,000-word manifesto, Industrial Society and Its Future. The FBI agreed, and the Washington Post printed it in September 1995.

The FBI's hope was that someone would recognize the writing. That's exactly what happened, but not through stylometry alone. David Kaczynski, Ted's brother, read the manifesto and noticed phrases and arguments that reminded him of letters Ted had written over the years. He contacted the FBI. Investigators then brought in forensic linguists to compare the manifesto against Ted Kaczynski's known writings. The linguistic analysis identified shared idiosyncratic phrases ("cool-headed logicians"), unusual spelling preferences, distinctive argument structures, and consistent patterns in how Kaczynski used qualifiers and built paragraphs.

The case illustrates a pattern that repeats across forensic stylometry: human intuition identifies a suspect, and statistical analysis provides the evidence to confirm or refute the match. The FBI's forensic linguist, James Fitzgerald, later described the Kaczynski case as the investigation that put forensic linguistics on the map for American law enforcement.

Primary Colors

In 1996, a political novel called Primary Colors appeared with "Anonymous" on the cover. The book was a thinly veiled account of Bill Clinton's 1992 presidential campaign, and Washington was consumed with guessing who wrote it. Suspicion centered on several journalists and political insiders.

Donald Foster, a Vassar English professor who had made his reputation with a stylometric analysis of a newly discovered Shakespeare poem, was brought in to analyze the text. Foster compared Primary Colors against the published work of several suspected authors. His analysis pointed to Joe Klein, a columnist for Newsweek. Klein denied it. Repeatedly. On camera. In print. For months.

He was lying. When the publisher's records eventually confirmed Klein's authorship, Foster's stylometric analysis was vindicated. The features that gave Klein away were precisely the kind of unconscious habits that make stylometry work: his characteristic use of metaphor, his sentence rhythm, his habitual qualifiers, and the way he structured anecdotes. Klein later admitted he'd tried to disguise his style while writing the book. It wasn't enough.

The Elena Ferrante Speculation

The Italian novelist Elena Ferrante, author of the Neapolitan novels, has maintained strict anonymity since her first publication in 1992. Multiple attempts have been made to identify her through stylometric analysis.

In 2016, Italian journalist Claudio Gatti published an investigation that used financial records, not stylometry, to identify a suspect. But several research teams have also applied computational stylometry to Ferrante's work, comparing it against other Italian authors. The results have been mixed and contested. Some analyses have pointed to specific candidates; others have been inconclusive.

The Ferrante case is instructive because it shows both the power and the limits of the method. Stylometry works best with a closed set of candidates and sufficient comparison texts. When the candidate pool is open and the analyst doesn't know who to compare against, the problem becomes much harder. It also raises the ethical questions I'll return to later: just because you can identify an anonymous writer doesn't mean you should.

Shakespeare's Collaborators

For centuries, scholars debated which plays in the Shakespeare canon were written solely by Shakespeare and which involved collaborators. The arguments were based on close reading, historical records, and subjective judgment. Computational stylometry has begun to settle these disputes with data.

In 2016, Gary Taylor and the editors of the New Oxford Shakespeare used large-scale stylometric analysis to attribute portions of the canon to co-authors. Their analysis identified Christopher Marlowe as a collaborator on the three Henry VI plays, a finding that generated considerable controversy but was supported by multiple independent computational methods. Function word frequencies, vocabulary distributions, and syntactic patterns in specific scenes diverged from Shakespeare's baseline and matched Marlowe's known solo works.

More recently, Santiago Segarra and colleagues at MIT applied computational stylometry to The Two Noble Kinsmen and confirmed the long-suspected division of labor between Shakespeare and John Fletcher. Their analysis could pinpoint the likely author of individual scenes with high confidence, based on features neither playwright would have consciously controlled.

These literary applications matter beyond academia. They demonstrate that stylometric methods can distinguish between authors even when those authors wrote in the same genre, in the same era, for the same audience. That's a rigorous test of the method's discriminating power.

Online Threat Investigation

I'll keep this one brief and non-specific. Law enforcement agencies routinely use forensic stylometry to investigate online threats, harassment campaigns, and extremist communications. When someone posts threatening messages under a pseudonym, investigators can compare those messages against known writing samples from suspects.

The challenges are real: online writing tends to be short, informal, and inconsistent. But when an individual produces enough text across enough messages, the statistical signal accumulates. Several successful prosecutions in the U.S. and U.K. have included forensic linguistic evidence as part of the case. I'll discuss the courtroom implications below.

How Forensic Stylometry Works in Practice

The famous cases make good stories. But what does the actual process look like when a forensic stylometrist takes on a case? Here's the practitioner's workflow.

Step 1: Collect reference samples. The analyst gathers known writing samples from the suspected author (or authors). These samples need to be verified, meaning there's no doubt about who wrote them. The more samples, the better. Ideally, the reference texts are from the same genre and time period as the disputed text, though cross-genre analysis is possible with the right methods.

Step 2: Prepare the disputed text. The questioned document is cleaned and formatted for analysis. Headers, footers, quoted material, and boilerplate are removed. The analyst needs to ensure they're measuring the author's writing, not template language or copied passages.

Step 3: Choose features and methods. This is where expertise matters. The analyst selects which features to measure (function words, n-grams, sentence structure, vocabulary richness) and which statistical methods to apply. Good practitioners use multiple independent methods. If three different approaches all point to the same author, the conclusion is much stronger than if only one method was used.

Step 4: Run the analysis. The software computes feature profiles for each reference sample and the disputed text, then measures the statistical distance between them. Tools like JGAAP, Stylo, and custom scripts built on standard statistical libraries are common in the field. The analyst examines not just the overall similarity score but the per-feature breakdown, looking for which features match strongly and which diverge.

Step 5: Interpret the results. This is the step that separates a forensic expert from someone who just runs software. The analyst must evaluate whether the results are statistically significant, whether confounding factors (genre mismatch, editing, co-authorship) could explain the patterns, and what the error rates are for the methods used. A responsible analyst also considers alternative explanations.

Step 6: Write the report. The forensic report documents the methods, the data, the results, and the conclusions. It must be detailed enough for another expert to reproduce the analysis. The conclusions are typically stated in terms of likelihood rather than certainty: "The linguistic evidence strongly supports the proposition that Author X wrote the disputed text" rather than "Author X definitely wrote this."

Step 7: Testify (if the case goes to court). The analyst may be called as an expert witness. This means explaining the methodology to a judge and jury, defending it under cross-examination, and making statistical concepts accessible to non-specialists.

The Courtroom Reality

Forensic stylometry occupies an interesting position in the legal system. It's not as established as DNA analysis or fingerprint evidence, but it's not novel science either. Courts in the U.S., U.K., and several other jurisdictions have admitted stylometric evidence, though the standards vary.

In the United States, expert testimony must pass the Daubert standard (or the older Frye standard in some states). Under Daubert, the judge acts as gatekeeper and evaluates whether the methodology is testable, has known error rates, has been peer-reviewed, and is generally accepted in the relevant scientific community. Stylometry meets most of these criteria: the methods are published, reproducible, and have documented error rates from controlled studies. The annual PAN competitions provide standardized benchmarks.

Where stylometric evidence has been challenged, the attacks usually focus on a few pressure points. Opposing counsel may argue that the sample size is too small, that the method hasn't been validated for the specific language or genre at issue, that the analyst cherry-picked features, or that the results don't exclude other possible authors. These are legitimate concerns, and a well-prepared expert will have anticipated them.

The field has also faced skepticism because of high-profile missteps. Donald Foster, the professor who correctly identified Joe Klein as the author of Primary Colors, later incorrectly attributed a poem to Shakespeare. The poem was eventually shown to be by someone else entirely. Foster's error didn't invalidate the method, but it illustrated that analyst judgment still matters and that overconfidence is a real risk.

The general trend in courts is toward greater acceptance, particularly when the analysis uses multiple methods, documents its error rates, and is presented by a qualified expert. But stylometric evidence is almost never the sole basis for a verdict. It's one piece of a larger evidentiary puzzle.

Digital-Age Applications

The cases I've described so far mostly involve books, letters, and manifestos. But the fastest-growing applications of stylometry are digital.

Sockpuppet detection. On platforms like Wikipedia, Reddit, and online forums, users sometimes create multiple accounts to manipulate discussions, inflate support for their positions, or evade bans. Stylometric analysis can detect sockpuppets by comparing the writing patterns across accounts. Wikipedia's CheckUser system incorporates some of these techniques. Several academic studies have demonstrated high accuracy in linking accounts based on writing style alone.

Cybersecurity attribution. When an anonymous actor publishes a ransom note, a data breach announcement, or a hacktivist manifesto, attribution analysts apply stylometric methods alongside technical forensics. The writing in these communications, even when short, can narrow the list of suspects or link the communication to known threat actors. Intelligence agencies have acknowledged using linguistic analysis as part of their attribution toolkit, though they rarely discuss the methods in detail.

Social media forensics. Identifying the author behind anonymous or pseudonymous social media accounts is an active area of research. The challenge is that social media text tends to be short, informal, and influenced by platform-specific conventions (hashtags, character limits, emoji). Researchers have developed specialized methods for this domain, including character-level models that can work with shorter texts and approaches that account for platform-specific writing conventions.

Fraud detection. In some financial and insurance investigations, documents purportedly written by different people are compared to determine whether they were actually authored by the same person. Stylometric analysis can flag cases where a single individual has fabricated multiple "independent" statements or reports.

The Ethics of De-anonymization

Every capability creates a responsibility. Stylometry's power to identify anonymous writers raises serious ethical questions that the field hasn't fully resolved.

When identification is appropriate. Few people object to using stylometry to identify the author of a bomb threat, to catch a serial harasser, or to verify that a student wrote their own dissertation. In these cases, there's a clear victim, a clear harm, and a legitimate interest in knowing who wrote the text.

When it gets complicated. The Ferrante case sits at the boundary. Ferrante chose anonymity as an artistic and personal decision. She wasn't hurting anyone. The attempts to unmask her, whether through financial records or stylometric analysis, were driven by curiosity, not by any need to prevent harm. Many in the literary community argued that the de-anonymization attempts were a violation of her autonomy, regardless of whether they succeeded.

Whistleblower protection. Stylometric de-anonymization poses a genuine threat to whistleblowers, dissidents, and journalists' sources. If an authoritarian government can analyze an anonymous leak and match it to a known critic's writing patterns, the consequences could be severe. Researchers in adversarial stylometry, notably Michael Brennan and Rachel Greenstadt at Drexel University, have studied how writers can alter their style to resist identification. Their work on "anonymouth" tools represents the flip side of the attribution coin: using the same science to protect anonymity rather than to break it.

Journalistic source protection. When stylometry is used to identify the source of a leak, it can undermine press freedom. Several legal scholars have argued that stylometric de-anonymization of journalistic sources should face the same legal protections as other forms of source identification.

The consent problem. Stylometric analysis can be performed on any text, without the writer's knowledge or consent. Unlike DNA, which requires a physical sample, a writing sample is available to anyone who can read the author's published or posted text. This asymmetry, where the subject of analysis has no ability to opt out, makes the ethical stakes higher than they might first appear.

The field needs clearer norms. My view: stylometric analysis is a legitimate investigative tool when there's a credible allegation of wrongdoing or a genuine dispute about authorship. Using it to satisfy curiosity about an anonymous writer who isn't harming anyone crosses a line. The technology is neutral; the ethics depend entirely on the application.

Where This Goes Next

Stylometry started with scholars counting word frequencies by hand. It now involves machine learning pipelines processing millions of features per second. But the core insight hasn't changed: writers leave unconscious patterns in their text, and those patterns are measurable.

What's changing is the scope of application. As more communication moves online and as AI-generated text complicates the question of authorship, the ability to verify who wrote what becomes more valuable. Forensic stylometry is evolving from a niche academic discipline into a practical tool for law enforcement, cybersecurity, and authorship verification.

WritersLogic's voice fingerprinting builds on these same principles. Rather than waiting for a dispute to arise, it measures your writing patterns continuously, building a baseline that can serve as evidence if your authorship is ever questioned. The goal isn't to de-anonymize anyone. It's to give writers a way to prove they are who they say they are.

For the technical foundations, including the mathematics of cosine similarity and Burrows' Delta, the five dimensions of voice fingerprinting, and detailed accuracy benchmarks, see our deep dive on voice fingerprint analysis. For how this evidence holds up in formal settings, see legally defensible metrics. And for the broader context of writing provenance, see the rise of provenance verification.