AI detection in Italian essays through different text representations and adversarial robustness evaluation

Andrea Sciandra, Francesco Dal Cero, Michele A. Cortelazzo, Arjuna Tuzzi · 2026

Read AI detection in Italian essays through different text representations and adversarial robustness evaluation by Andrea Sciandra, Francesco Dal Cero, Michele A. Cortelazzo, Arjuna Tuzzi (2026) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/pmc-13350967.txt.

Description

Andrea Sciandra, Francesco Dal Cero, Michele Cortelazzo, and Arjuna Tuzzi wrote this research paper, published in 2026. It tests how well machines can tell AI-written essays from student-written ones in Italian, and how easily those detectors can be fooled.

The team assembled one thousand essays and encoded each four ways: hand-crafted text features, most frequent words, correspondence analysis, and a fine-tuned language model. Classifiers trained on the encodings scored high accuracy, simple text features especially. Then came attacks: small edits meant to disguise AI prose. Feature and word models proved fragile, while the correspondence and language model approaches held up. Correspondence analysis offered the best balance of few predictors, accuracy, and toughness. The authors discuss what the detectors see, sentence shape and word choice, and warn schools about leaning on brittle tools.

Catalog record

CREATORAndrea Sciandra, Francesco Dal Cero, Michele A. Cortelazzo, Arjuna Tuzzi
YEAR2026
COLLECTIONPAPERS
FORMATTXT
FILE SIZE54 KB
VERSION1.0
LICENSECC BY 4.0
RIGHTSCC BY
SOURCEEurope PMC

Source: https://europepmc.org/articles/PMC13350967

Browse the full library · About the Tower of Babel