Stylometry

Stylometry is the study of measurable linguistic patterns — word choice distributions, sentence length profiles, function word frequencies, punctuation habits, syntactic structures — that characterize an individual author's writing. The field predates modern AI by decades; it has been used in literary forensics (the Federalist Papers authorship analysis), legal proceedings, and historical research to attribute anonymous texts to specific authors.

In the context of AI detection, stylometry provides the theoretical foundation for distinguishing human from machine writing. Human authors exhibit idiosyncratic stylistic fingerprints that are stable across documents but vary between individuals. Language models, in contrast, tend to produce a narrower band of stylistic output — a kind of averaged style drawn from their training corpus. Stylometric features such as type-token ratio, hapax legomenon frequency, and part-of-speech distribution often reveal this difference.

Modern AI detectors do not rely purely on classical stylometry. They typically combine stylometric features with perplexity measurements from reference language models and learned neural classifiers. But the underlying intuition — that writing carries measurable stylistic signatures — remains central. Understanding stylometry helps explain both why detection is possible in principle and why it is hard in practice: skilled editing can shift a text's stylometric fingerprint, which is essentially what humanization tools attempt to do systematically.