Stylometry

Based on Wikipedia: Stylometry

In 1901, a researcher attempted to settle a centuries-old literary debate by counting contractions. He believed that the playwright John Fletcher had a distinct, almost compulsive habit of using the contraction "'em" for "them," a fingerprint he hoped would separate Fletcher's hand from that of his collaborator, Philip Massinger. The math seemed sound, the logic airtight, until the fatal flaw revealed itself: the edition of Massinger's works he had chosen for comparison had been edited by a Victorian scholar who had dutifully expanded every instance of "'em" back to "them." The data was not wrong; the source was. This historical stumble captures the essence of stylometry, the forensic study of linguistic style, in its rawest form: it is an attempt to find the ghost of an author in the machine of their text, a quest that is as vulnerable to human error as it is powerful in its potential to unmask the anonymous.

Stylometry is not merely the study of how people write; it is the measurement of who they are, distilled into syntax, vocabulary, and rhythm. While it has traditionally been applied to written language to attribute authorship to disputed or anonymous documents, its reach has expanded far beyond the printed page. Today, the same algorithms that seek the shadow of Shakespeare in a lost sonnet are analyzing musical compositions, the brushstrokes of paintings, the strategic moves in chess games, and even the structure of computer source code. It is a discipline that bridges the gap between the humanities and the hard sciences, offering legal, academic, and literary applications that range from the high-stakes drama of forensic linguistics to the quiet resolution of historical mysteries. Yet, as we stand in 2026, with the digital corpus of humanity growing exponentially and artificial intelligence capable of mimicking human expression with chilling accuracy, stylometry has evolved from a niche academic curiosity into a critical battleground for privacy, identity, and truth.

The roots of this discipline stretch back further than the computer age, though the tools have changed. The modern practice of stylometry received a significant boost from the study of English Renaissance drama, where researchers observed that playwrights of the era possessed distinctive, almost involuntary patterns of language preference. They sought to use these patterns to identify authors of uncertain or collaborative works, a task that was fraught with difficulty. The early efforts were not always successful, as the Fletcher and Massinger incident demonstrates. It was not until 1890 that the basics of the discipline were truly established by the Polish philosopher Wincenty Lutosławski. In his work Principes de stylométrie, Lutosławski used these methods to develop a chronology of Plato's Dialogues, arguing that the style of the texts evolved in a way that mapped onto the philosopher's intellectual development. This was a radical departure from traditional literary criticism, which relied on subjective interpretation; Lutosławski proposed that style was a quantifiable metric, a biological signature of the mind.

The true revolution, however, arrived with the computer. In the early 1960s, the Rev. A. Q. Morton turned his attention to the New Testament, producing a computer analysis of the fourteen Epistles attributed to St. Paul. The results were startling: the data suggested that six different authors had written that body of work, shattering the traditional attribution to a single apostle. Morton's methods were audacious, but they were not infallible. When a check of his method was applied to the works of James Joyce, the computer, unable to distinguish between the deliberate stylistic shifts of a genius and the presence of multiple authors, concluded that Ulysses—Joyce's multi-perspective, multi-style masterpiece—was composed by five separate individuals, none of whom apparently had any part in crafting his first novel, A Portrait of the Artist as a Young Man. This error highlighted a fundamental challenge in the field: distinguishing between the deliberate artistic evolution of a single author and the statistical noise of multiple contributors. Despite these early missteps, the field found its footing with notable successes. In a landmark study, Frederick Mosteller and David Wallace resolved the long-standing dispute over the authorship of twelve of The Federalist Papers, definitively attributing them to James Madison based on the statistical frequency of function words. Similarly, the application of textual analysis to the Fletcher canon by Cyrus Hoy and others in the late 1950s yielded clear, undeniable results, proving that the method could cut through the fog of historical collaboration.

The Mechanics of the Invisible Signature

At its core, stylometry operates on a counter-intuitive premise: to identify an author, one must often ignore what they are saying and focus entirely on how they are saying it. Authors often have strong preferences for certain topics, but these topics are the least reliable indicators of identity. If a writer is discussing politics, they will use political vocabulary regardless of who they are. If they are writing about love, they will use romantic imagery. To avoid "overfitting" their models to the subject matter rather than the author, researchers in authorship attribution mostly remove content words—nouns, adjectives, and verbs—from the feature set. Instead, they retain the structural elements of the text, the skeletal framework that remains even when the flesh of the content is stripped away.

These stylistic features are often computed as averages over a text or the entire collected works of an author. Measures such as average word length, average sentence length, and the frequency of specific punctuation marks become the data points. This enables a model to identify authors who have a clear preference for wordy, sprawling sentences versus those who favor terse, punchy declarations. However, this averaging process has a blind spot. It hides variation. An author who writes a mix of long, winding sentences and short, abrupt ones will have the same average sentence length as an author who consistently writes mid-length sentences. To capture this nuance, modern experiments have moved beyond simple averages. They now look at sequences and patterns over observations, noting that an author might show a preference for a certain stress or emphasis pattern, or that they tend to follow a sequence of long sentences with a short one. One of the first approaches to authorship identification, by Mendenhall, can be said to aggregate its observations without averaging them, preserving the texture of the writing rather than just its bulk.

The features of interest are subtle and often unconscious. On one hand, researchers compute the occurrences of idiosyncratic expressions or constructions, checking for how an author uses interpunction or how often they employ agentless passive constructions. On the other hand, they look at measures of lexical and syntactic variation similar to those used in readability analysis. These are not the grand gestures of style, the famous metaphors or the signature opening lines that literary critics celebrate. They are the micro-habits, the digital footprints left in the margins of the text. Since the 2020s, with the explosion of machine learning, these models have become increasingly sophisticated. Modern authorship attribution models use vector space models to automatically capture what is specific to an author's style, yet they still rely on judicious feature engineering for the same reasons as more traditional models: the machine needs to know what to look for before it can find it. The software systems available today, such as Signatur, can process vast corpora of text, scanning millions of words to find the statistical anomalies that point to a specific human hand.

The Adversarial Turn: Hiding in Plain Sight

As stylometry has grown more powerful, a new dynamic has emerged: the arms race between the identifier and the hidden. This is the realm of adversarial stylometry, the practice of altering one's writing style to reduce the potential for stylometric discovery. This task, also known as authorship obfuscation or authorship anonymization, is born of necessity for those who must remain anonymous. Stylometry poses a significant privacy challenge in its ability to unmask anonymous authors or to link pseudonyms to an author's other identities. For whistleblowers, activists, and even those seeking to expose fraud or protect their safety, the ability to be unmasked by an algorithm is a terrifying prospect. The privacy risk is expected to grow as machine learning techniques and text corpora develop, making the need for effective obfuscation more urgent than ever.

All adversarial stylometry shares the core idea of faithfully paraphrasing the source text so that the meaning is unchanged but the stylistic signals are obscured. Such a faithful paraphrase is an adversarial example for a stylometric classifier; it is a text that reads naturally to a human but confuses the statistical models looking for the author's fingerprint. Several broad approaches exist, with some overlap. Imitation involves substituting the author's own style for another's, effectively wearing a digital mask. Translation applies machine translation with the hope that the process of converting the text into another language and back will eliminate characteristic style in the source text. Obfuscation involves deliberately modifying a text's style to make it not resemble the author's own, perhaps by introducing random errors or altering sentence structures in unnatural ways.

Manually obscuring style is possible, but it is laborious and often fails to be consistent. In some circumstances, manual editing is preferable or necessary, but the sheer volume of text produced in the digital age makes automation essential. Automated tooling, either semi- or fully-automatic, could assist an author in this task. However, how best to perform the task and the design of such tools remains an open research question. While some approaches have been shown to be able to defeat particular stylometric analyses, particularly those that do not account for the potential of adversariality, establishing safety in the face of unknown analyses is a major issue. The fear is not just that a specific tool will fail, but that the very act of trying to hide will leave its own signature. Ensuring the faithfulness of the paraphrase is a critical challenge for automated tools; if the tool changes the meaning of the text, it has failed its primary purpose. It is uncertain if the practice of adversarial stylometry is detectable in itself. Some studies have found that particular methods produced signals in the output text, but a stylometrist who is uncertain of what methods may have been used may not be able to reliably detect them. The cat-and-mouse game is ongoing, and the outcome is far from guaranteed.

Beyond the Text: Applications and Limitations

The applications of stylometry extend far beyond the resolution of literary disputes. In the realm of legal and forensic studies, it has become a vital tool for analyzing evidence, determining the origin of threatening letters, or verifying the authorship of disputed wills. In historical studies, it has advanced long-standing debates about anonymous medieval Icelandic sagas, offering a new lens through which to view the past. In social studies, it has been used to track the evolution of language and communication patterns across different demographics. More recently, stylometry has found a surprising application in the realm of computer code. Just as a writer has a unique syntax, a programmer has a unique coding style. Intrinsic plagiarism detection, which detects plagiarism based on the writing style changes within a document, can now be applied to source code, helping to identify unauthorized copies or the theft of intellectual property.

However, the method is not without its vulnerabilities. Stylometry as a method is vulnerable to the distortion of text during revision. If an editor significantly rewrites a text, the original author's stylistic signature may be erased or altered beyond recognition. There is also the case of the author adopting different styles in the course of their career. As demonstrated in the case of Plato, who chose different stylistic policies such as those adopted for the early and middle dialogues addressing the Socratic problem, an author's voice is not static. It evolves, it shifts, and it adapts to the context. This fluidity makes the task of attribution even more difficult, as the model must account for the possibility that the author is not the same person who wrote the text a decade ago.

Furthermore, the technique can be used to predict whether someone is a native or non-native English speaker by their typing speed and other subtle markers, a capability that raises profound questions about surveillance and profiling. The ability to link a text to a demographic or a specific individual based on typing patterns or syntactic quirks is a powerful tool, but one that must be wielded with extreme caution. The line between identification and profiling is thin, and the implications for privacy are immense. As we move deeper into the 21st century, the intersection of stylometry and artificial intelligence will only become more complex. The tools that can identify an author are becoming more sophisticated, but so are the tools that can hide them. The future of stylometry will likely be defined by this tension, a continuous struggle between the desire to know and the right to remain unknown.

The story of stylometry is ultimately a story about the human desire for connection and the fear of exposure. It is a testament to the fact that even in the most digital, anonymous corners of the internet, the human hand leaves a mark. Whether it is the preference for a specific contraction, the rhythm of a sentence, or the structure of a line of code, our style is our signature. And as long as we write, we will leave traces. The question is no longer whether we can be found, but whether we want to be. For the whistleblower in a repressive regime, the answer is a resounding no. For the literary scholar trying to solve a centuries-old mystery, the answer is a resounding yes. Stylometry stands at the crossroads of these desires, a discipline that offers the power to reveal the hidden, but also the responsibility to respect the right to remain silent. As the technology advances, the stakes rise, and the need for a nuanced understanding of what we can and cannot know about the author behind the text becomes more critical than ever. The ghost in the machine is real, but it is a ghost that can be summoned, and perhaps, just perhaps, exorcised.