labStylometry. An indicator, not proof
Comparing texts by manner of writing helps to check a hypothesis about an author. It cannot establish the author, and the report says so in plain words.
Stylometry is the comparison of texts by measurable features of writing: sentence length, the frequency of function words, punctuation, typos, word order. The method is old and has one task: to determine whether two texts are similar enough that they could share an author.
In an investigation stylometry checks a hypothesis that already exists: whether two public documents have one author, whether the author of a corporate channel has changed, whether a press release was written by the same hand as a filing to a registry. It is not used to search for a person.
What is measured
- Function words. Prepositions, conjunctions, particles. Their frequency depends little on the topic, and the author has little control over it.
- Punctuation. A dash instead of a colon, a comma before "and", a space before a mark. These habits persist across a change of subject.
- Persistent errors. The same typo or the same wrong agreement across texts.
- Length. The length of sentences, paragraphs and words. The distribution is compared, and an average alone is not enough.
- Formatting. Quotation marks, list markers, the case of headings, the habit of ending a heading with a full stop.
Features are counted. The sense that two texts "read the same" remains an impression, and the data is a table of frequencies.
What stylometry cannot do
- Establish the author. A match of manner speaks only of similarity. One editorial office may have a single style, and one person's manner changes with the platform and the years.
- Work on a short text. A two-line post gives no distribution, and there is nothing in it to compare.
- Tell the author from the editor. A corporate text passes through several hands, and its manner is made up of the edits of several people.
- Replace a document. Similarity of manner is no basis for a statement of authorship. The basis is a document, a testimony or an admission.
Where it sits in the chain
In the chain of an investigation a stylometric result has one place: a proposal. It enters the report as an indicator with the method and the features named. A verification task is opened on it, and it does not change a record by itself. The record "the author of the document is such-and-such" appears only with the document that confirms it.
In a report it is written like this: "texts A and B are similar in features X, Y, Z. The hypothesis of a common author was checked by document D", or "was not checked, the gap is open". The wording "stylometry showed that the author is N" is not admitted into a report.
Boundary
Stylometry is applied to public texts of persons in their public capacity: statements of organisations, documents from registries, publications made in the name of an office. Establishing the identity of a private individual by their manner of writing (deanonymisation) falls outside the subject of the work. The boundary of use is written down and does not change on request.