Unlocking the Past: How Stylometry Solves the Mysteries of Anonymous Poetry

For centuries, scholars have grappled with a tantalizing puzzle: who really wrote those haunting lines of verse? Countless historical poems survive without a clear author, their origins lost to time, deliberate anonymity, or the simple decay of parchment. From the epic Beowulf to the sonnets of Shakespeare’s rivals, the question of authorship is not merely an academic curiosity—it reshapes our understanding of literary history, cultural influence, and even political thought. Traditional methods—handwriting analysis, historical records, and close reading—are invaluable, but they often hit a wall when faced with unsigned works. Enter stylometry: a data-driven, quantitative approach that applies statistical analysis to writing style. By measuring subtle, often unconscious linguistic fingerprints, stylometry offers a powerful, scientific lens to attribute uncertain authorship, sometimes settling debates that have simmered for centuries.

What Is Stylometry? The Science of Linguistic Fingerprints

Stylometry is the systematic, statistical study of written language. The core premise is simple yet profound: every author has a unique, stable pattern of stylistic habits that they cannot easily disguise. Just as a fingerprint distinguishes one person from another, so do certain textual features—word choice, sentence length, punctuation, and even the frequency of common “function words” like the, and, of, or because. These elements operate below the level of conscious authorial control, making them reliable markers of identity.

The discipline emerged in the 19th century but truly flourished with the rise of computers, which can crunch millions of data points in seconds. Modern stylometry doesn’t just look at word counts. It employs sophisticated techniques including principal component analysis, cluster analysis, and machine learning classifiers. For poetry, the analysis is particularly nuanced because poets often deliberately break rules of grammar and meter—but their idiosyncrasies remain. Stylometry asks: do two poems use the same ratio of nouns to verbs? Do they favor enjambments in the same way? Do they share a preference for certain rhyme schemes or vowel patterns? By answering these questions statistically, researchers can build a profile for an author and compare it to unknown works.

Key Metrics in Stylometric Analysis

While early stylometrists counted word lengths, modern analysis uses a richer set of features. For poetry, some of the most powerful include:

  • Function word frequencies: Words like a, an, the, in, on, that—they are nearly invisible to readers but statistically stable across an author’s body of work. These are often the most reliable markers.
  • Morphological patterns: The use of past tense versus present tense, specific suffixes, or contractions.
  • N-gram sequences: Short chains of letters, words, or parts of speech (e.g., “the-king-of” or “noun-verb-article”).
  • Meter and rhythm: In formal poetry, scans of iambic pentameter or variations in stress patterns can be statistically modeled. Stylometry can detect an author’s signature deviations from a strict meter.
  • Vocabulary richness and hapax legomena: The number of unique words used, and particularly the count of words that appear only once in a text, can differentiate writers.
  • Punctuation habits: Use of semicolons, dashes, or parentheses—often a deeply personal stylistic tic.

How Stylometry Works in Practice: From Raw Text to Statistical Inference

The workflow of a stylometric study is methodical. First, researchers assemble a corpus of known works by the candidate author and a reference corpus of comparative authors from the same period and genre. The poem in question—the “test text”—is cleaned: hyphens are regularized, archaic spellings are sometimes normalized or preserved depending on the analysis, and line breaks are noted. Then, software extracts the chosen stylometric features. The critical step is statistical testing. A common technique is to compute a distance metric—for example, how far apart the test text’s function-word profile is from each known author’s profile. Another powerful tool is Delta, introduced by John Burrows, which measures the absolute difference in standardized word frequencies between texts. Smaller Delta scores suggest closer stylistic proximity.

Machine learning has added even more precision. Algorithms such as support vector machines or random forests are trained on labeled texts from known authors. They learn patterns that discriminate between writers, then classify the anonymous poem. Of course, confidence levels are always reported—stylometry rarely claims 100% certainty, but probabilities above 95% are often considered strong evidence.

Special Considerations for Historical Poetry

Poetry from earlier eras presents unique challenges. Language changes over time: a word common in 1600 might be rare in 1700. Dialect and regional spelling can introduce noise. Moreover, many poems survive only in scribal copies—how much of the “style” is the original author, and how much is the copyist’s tinkerings? Stylometricians address these issues by focusing on features that are robust to transmission error, such as syntax-level patterns or content function words (e.g., prepositions and conjunctions that are less likely to be altered by scribes). They also use “rolling stylometry”—analyzing small, sequential chunks of a long poem—to see if stylistic consistency holds throughout, which would argue against multiple authors.

Famous Case Studies: Stylometry in Action

The success of stylometric analysis is best illustrated through landmark cases that have reshaped literary history.

Shakespeare and the Apocrypha

The most celebrated application is the attribution of works to William Shakespeare. For decades, scholars debated authorship of plays like Edward III and the poem “A Funeral Elegy” attributed to “W.S.” Using function-word frequencies and rare word distributions, stylometric analyses in the 1990s and 2000s convincingly attributed Edward III to Shakespeare (with possible collaboration) and rejected “A Funeral Elegy” as non-Shakespearean. More recently, stylometry has helped untangle the collaboration between Shakespeare and his contemporaries like John Fletcher in Henry VIII and The Two Noble Kinsmen. These studies show that even the Bard’s style is quantifiable and distinct.

The Federalist Papers: A Classic Authority

Though prose, not poetry, the Federalist Papers remain a foundational proof-of-concept for stylometry. For years, attribution of the 85 essays to Alexander Hamilton, James Madison, or John Jay was disputed. In 1964, statisticians Frederick Mosteller and David Wallace used Bayesian analysis of function words (like upon, also, enough) to assign authorship with high accuracy, confirming Madison’s hand in many disputed essays. This case established that stylometric conclusions could hold up even under intense scholarly scrutiny.

Medieval Mysteries: The Pearl Poet and the Homeric Hymns

In historical poetry, stylometry has shed light on the anonymous 14th-century poem Sir Gawain and the Green Knight. For centuries, it was grouped with three other poems (Pearl, Patience, Cleanness) based on dialect and manuscript, but no author knew. Stylometric analysis of vocabulary richness, alliterative patterns, and syntactical structures has provided strong evidence that the same unknown author wrote all four works—the “Pearl Poet”—settling a long-standing debate.

Similarly, the Homeric Hymns, a collection of ancient Greek poems traditionally attributed to Homer, have been scrutinized. Stylometric studies that analyze the use of epithets and line-initial formulas reveal that some hymns are stylistically distinct from the Iliad and Odyssey, likely composed by different poets in different centuries. This helps classicists map the evolution of epic tradition more accurately.

Modern Controversies: The Poems of Thomas Hardy and Others

Even into the modern era, stylometry is used to verify or question attributions. For instance, disputed poems from the estate of Thomas Hardy were analyzed against his known canon. Stylometric evidence showed that some posthumously published poems had inconsistent patterns, leading scholars to re-evaluate their inclusion. In the digital age, the technique is also used to identify anonymous online poetry—though that raises its own ethical questions.

Limitations and Challenges: When Stylometry Holds Its Tongue

Despite its successes, stylometry is not a silver bullet. Practitioners openly acknowledge significant limitations:

Sample Size and Quality

Stylometry requires a large enough sample of known, reliable texts to build a statistically robust profile. For poets who left only a handful of poems—or whose works survive only in fragmentary form—the method struggles. A common rule of thumb is at least 5,000 words of known text, but the more the better. For very short poems (haikus, epigrams) the analysis becomes noise-prone. In such cases, researchers may combine multiple short pieces into a single aggregate profile, but this dilutes precision.

Temporal and Dialectal Variation

An author’s style can shift over a lifetime. Shakespeare’s later plays use different vocabulary and sentence structures than his early comedies. A stylometric model trained on early works may fail to recognize his late style. Similarly, poets who wrote in multiple dialects (e.g., Robert Burns in Scots and English) appear as different “authors” to the algorithm. These variations must be accounted for by building period-specific reference corpora.

Intentional Imitation and Collaboration

Some forgers are skilled. In the Donation of Constantine forgery, the imitation was so good that stylometry alone might not have exposed it—modern forgers of poetry (like Thomas Chatterton) also created convincing pseudo-medieval works. Stylometry detects unconscious patterns; a determined imitator can sometimes mimic surface-level features, though deeper function-word patterns are harder to fake. Collaboration is even trickier. Many historical poems were written by multiple hands (e.g., medieval ballads). Stylometric tools can identify shifts in style within a single text, but teasing apart who wrote which line often requires human judgment.

Transmission and Editing

Historical poems suffer from scribal errors, editorial emendations, and translation. A poem that survives in a unique manuscript may have been “improved” by a later editor. Stylometric analysis must be applied to the original text—not a modernized version—and researchers often create a “diplomatic” edition (preserving original spelling and punctuation) to minimize introduced noise. Even then, the scribe’s own style can contaminate the data.

Integrating Stylometry with Traditional Scholarship

The most robust authorship studies do not rely on stylometry alone. Rather, they integrate it with historical, paleographical, and philological evidence. For example, a stylometric attribution that places a poem in the circle of a certain poet becomes much stronger when documentary evidence shows that the poet lived in the right region, used the same paper stock, or had known connections to the manuscript’s commissioner. Stylometric evidence can confirm or refute a hypothesis formed by close reading—it rarely replaces it.

This interdisciplinary approach has yielded triumphs. A case in point is the disputed poem The Soul’s Errand (attributed to Sir Walter Raleigh). Traditionalists argued on internal evidence that it was by Raleigh. Stylometric analysis of function words and syntactic patterns later supported that attribution, and historical research found a letter referencing Raleigh writing those very lines. The convergence of methods gave the literary community confidence.

For scholars today, the best practice is to treat stylometry as a tool in a larger forensic toolkit. It offers objective, reproducible data that can counterbalance subjective interpretive biases. When multiple independent stylometric tests—using different feature sets and statistical methods—point to the same author, the case becomes powerful. And when they conflict, it signals the need for deeper investigation.

Future Directions: Poetry in the Age of AI

Stylometry is evolving rapidly. With the increasing availability of digital archives (like the Early English Books Online or Perseus Digital Library), researchers can access vast corpora. Machine learning models, particularly those based on deep learning, can now analyze not just words but semantic patterns and emotional arcs. For poetry, this opens up new possibilities: could we distinguish a poem by Emily Dickinson from a modern imitator writing in her style? Early research suggests yes—the underlying “stylistic DNA” is captured even in subtle phraseology.

However, AI-generated poetry (like that from large language models) poses a new challenge. If a machine mimics a human poet perfectly, stylometry may temporarily be fooled. But current AI still has recognizable quirks, such as overusing certain transition words or falling into rhythmic plateaus. Stylometric models trained specifically to detect AI text are being developed, maintaining stylometry’s relevance as a tool for authentication.

Conclusion

Stylometry provides a rigorous, scientific approach to one of the most human of questions: who wrote this poem? By measuring the invisible fingerprints of style—word choices, syntactic rhythms, and unconscious patterns—it offers clarity where tradition and intuition only guess. While no statistical method can replace the nuanced judgment of a trained literary historian, stylometry supplies an indispensable check and a means to test hypotheses objectively. As digital archives grow and analytical techniques improve, we can expect more authorship mysteries to be solved, giving voice to the anonymous poets of the past and deepening our appreciation of our shared literary heritage. The next time you encounter a poem of uncertain origin, consider that its author might already have been identified—by a computer counting the very words you thought were just poetry.