A century of science, pointed at one classroom question
Stylometry — identifying authors by measurable writing habits — predates computers. Authentext applies it to the question teachers ask now: does this sound like my student?
Where stylometry comes from
In 1887, T.C. Mendenhall counted word lengths to distinguish authors' "characteristic curves." In 1964, Mosteller and Wallace settled the disputed authorship of the Federalist Papers using the frequencies of small function words — upon, while, whilst — words writers use without thinking. In 2002, John Burrows introduced Burrows' Delta, still the field's benchmark method: an author's signature measured across dozens of function-word frequencies at once. Stylometric analysis has since become an established method in literary scholarship and forensic linguistics.
The consistent finding across a century: writers carry measurable habits they don't notice and rarely change — especially in the small structural machinery of language rather than its visible vocabulary.
Why "compared to themselves" beats "compared to AI"
Generic AI detectors ask whether text resembles AI output in general — a question with a documented false-positive problem, which falls hardest on multilingual students and distinctive young writers, because their authentic writing diverges from the detector's idea of "normal."
Authentext asks a narrower, older, more answerable question: does this text match this specific writer's established patterns? A multilingual student isn't compared to native-speaker norms — they're compared to themselves. Consistency is the standard, and every student gets their own.
What Authentext measures
Two complementary layers, 30+ markers.
Structural: function-word distributions (the Burrows' Delta tradition), syntactic patterns, sentence and clause architecture, punctuation signatures — the deep habits most resistant to change.
Surface: vocabulary diversity, lexical maturity, word-length and readability patterns — the visible layer that grows naturally over time and shifts sharply when authorship changes.
Structural markers anchor identity; surface markers track development. Reading both is what lets the system tell a student who grew from a submission that isn't theirs.
Where the field is going
Recent academic work has converged on exactly this architecture: per-student stylometric baselines built from verified writing, longitudinal tracking, and authenticity evaluated against the student's own profile rather than a population model. Educators are independently arriving at the same practice — supervised in-class writing as a fair reference point for each student's voice. Authentext is that practice, automated.
The method is a century old. The classroom problem is brand new. We connected them.