Did Shakespeare Write His Plays? Stylometry Explained

Investigative forensic linguistics uses language as evidence to solve crimes, including determining whether suicide notes are genuine or staged, identifying anonymous authors, and analyzing coded criminal organization language. Descriptive forensic linguistics analyzes how language is used in legal contexts without making authenticity judgments, such as studying how judges address witnesses or how police communicate with citizens. These two approaches are interconnected - descriptive research provides patterns that support investigative applications, though descriptive studies alone do not provide direct evidence for specific cases.

This segment explains how stylometry, a scientific method analyzing linguistic patterns, was used to prove the speaker's innocence. The process involved comparing anonymous messages to known texts, identifying unique linguistic signatures, and achieving 97-100% match certainty. The segment demonstrates how modern forensic linguistics can be used in legal proceedings, with the speaker noting that this method has been recognized in French courts (Grégori case) and is now being applied to his own defense.

Stylistometry is a forensic linguistic technique that analyzes unique linguistic patterns (idiolect) in text to identify authors, functioning like a 'blood analysis' for literature; it measures approximately 75 unconscious linguistic parameters that distinguish each person's writing style, enabling detection of authorship in cases of plagiarism, fraud, and criminal investigations, as demonstrated by its successful application in identifying the author of 'Primary Colors' and solving cases like the Unabomber.

Forensic linguistics, also known as stylometry, is a scientific method that analyzes writing patterns to identify authors of anonymous messages. In Tarik Ramadan's case, this technology was used to prove that the accuser had written threatening messages weeks before the alleged incident, demonstrating that the entire situation was a premeditated trap. The analysis showed 97.7% to 100% probability that the accuser authored these messages, providing scientific evidence of her deception. The speaker argues that when new scientific evidence emerges that could exonerate a convicted person, the justice system should be willing to reopen cases and revise judgments based on facts rather than political considerations.

Multi-class classification in NLP encompasses diverse tasks including text classification (assigning topics to documents), textual entailment (determining relationships between sentence pairs as entailment, contradiction, or neutral), entity disambiguation (linking text mentions to real-world entities), and authorship attribution (identifying authors based on stylistic features like stop word frequencies and n-gram patterns). These tasks demonstrate how classification models can leverage different feature types—such as bag-of-words, character n-grams, and word n-grams—to solve increasingly complex problems, with modern approaches achieving high accuracy even with limited training data.

Authorship attribution can be framed as a supervised learning problem where classifiers are trained on known authorship data to predict authors of unknown texts. Two key classification paradigms exist: closed-set classification (where test authors match training authors) and open-set classification (where test authors may include new individuals). Similarity-based approaches determine if texts share authorship without identifying specific authors. Beyond traditional linguistic analysis, machine learning has reorganized how authorship is detected, with deep learning methods like word embeddings capturing semantic patterns. The field combines classical linguistic analysis with modern computational techniques, with research spanning from character-level features to sophisticated neural networks.

This comprehensive section covers the entire workflow for building a plagiarism detection system using deep learning. It begins with defining plagiarism as copying content without credit, explaining why traditional approaches (exact matching, TF-IDF, keyword matching) fail to detect paraphrasing. The project uses a dataset with source text, plagiarized text, and labels. The pipeline includes: (1) Data loading and exploration; (2) Text preprocessing - lowercasing, punctuation removal, tokenization using top 20,000 frequent words, and padding to uniform length; (3) Data splitting (80-20 train-test); (4) Building a specialized LSTM architecture with subtraction layer for comparing encoded vectors; (5) Model compilation with binary cross-entropy loss and Adam optimizer. This establishes the complete foundation for the plagiarism detection model.

Advanced machine learning includes Natural Language Processing (NLP) for understanding human communication, enabling sentiment analysis and chatbots. Deep learning with neural networks represents a more advanced approach where the model itself learns features from data, with fewer statistical restrictions. These areas require completing the basic and intermediate routes first.

Text classification categorizes documents into predefined classes. Sentiment analysis determines emotional tone (positive, negative, neutral). A classic example is the Federalist Papers authorship analysis, where statistical methods examined word frequency patterns (like 'and', 'however') to identify authors. Modern approaches use Large Language Models (LLMs) with encoder-decoder architectures. The encoder converts text to numerical representations, while the decoder transforms these back into language. LLMs can generate text, translate languages, or create images based on decoder capabilities.

Two landmark cases illustrate stylometric principles. The Federalist Papers dispute (1787-1837) involved 85 anonymous documents attributed to Hamilton, Madison, and Jay. After Hamilton's death, his claim of 63 papers conflicted with Madison's. For 150 years unresolved, statistical analysis finally vindicated Madison with 99% accuracy by exploiting Hamilton's distinctive avoidance of common words. Similarly, J.K. Rowling's 2013 pseudonym 'Robert Galbraith' was exposed when stylometric comparison against five authors revealed only Rowling matched. These cases demonstrate that distinctive writing patterns persist despite attempts at concealment.

Stylometry has a long history in authorship determination. In the 1960s, Frederick Mosteller used stylometry to determine that James Madison was the main author of the Federalist Papers, taking nearly five years. More recently, JK Rowling was found to be the author of 'Cuckoo's Calling' under the pen name Robert Galbraith. In the crypto space, stylometry has been applied to analyze Satoshi Nakamoto's emails and documents to determine the true identity of the Bitcoin creator. The method examines word frequency, sentence structure, and linguistic patterns to identify authorship.

Stylometry enables objective authorship detection through word frequency analysis. Applications include: (1) 19th-century Polish authors (Sienkiewicz, Dickens, Balzac), revealing genre distinctions and chronological development; (2) Literary forgery cases like the Elena Ferrante controversy, where analysis exposed that texts attributed to Ferrante matched her husband's writing style; (3) J.K. Rowling's pseudonyms, where analysis confirmed Galbraith texts clustered with Rowling's works. In translation studies, the author's stylistic signal 'penetrates' into target languages, creating measurable similarities between originals and translations. Machine translation can sometimes produce translations indistinguishable from human translations. Cross-linguistic analysis shows translations from English and French cluster separately from original Polish literature, while Russian translations mix more readily due to linguistic similarities. Stylometry enables objective literary classification, supplementing traditional approaches with quantitative data.

The true identity of Elena Ferrante, the acclaimed author of the Neapolitan Novels series, remains one of literature's enduring mysteries; despite extensive research, theories, and scientific analysis including a 2017 University of Padua study comparing writing styles, her identity has not been definitively revealed, with Ferrante herself stating that the author and their work should be separate entities, and that maintaining anonymity is a fundamental element of her creative process.