Date: February 4, 2026 Time: 5:00 PM Venue: Laboratorio FDS (4th Floor,Building 14 “La Nave”, Via Bonardi 9, Campus Leonardo)

Abstract: This abstract presents current research and discussions on the use of multivariate continuous data, which is becoming increasingly prevalent in forensic science. As an illustrative example, attention is drawn to the field of comparative forensic handwriting examination.

Multivariate continuous data can be obtained in this field by analyzing the contour shape of loop characters using Fourier analysis. This methodology allows for a detailed description of the morphology of character contours through a set of variables. Data collected from female and male writers are presented to conduct a comparative analysis of Bayes factor-based evidence assessment procedures in both evaluative and investigative proceedings.

While the use of the Bayes factor in the former situation (evaluative) is now rather well-established – typically for discriminating between propositions of authorship of a known individual versus an unknown individual – its application in the investigative setting still remains relatively unexplored in practice. We highlight that investigative settings also represent an area of application for which the Bayes factor can offer logical support. As an example, the inference of the gender of the writer of an incriminated handwritten text is discussed. The general viewpoint that Bayes factor analyses are helpful for investigative proceedings is supported here through various simulations, which characterize the robustness of the proposed methodology.

Moreover, a question frequently arises as to whether an article or work, such as one submitted by a researcher, was actually produced by the alleged author of the questioned text. The role of artificial intelligence (AI) is increasingly debated due to the risks of its undeclared use. A current example is, undoubtedly, the undisclosed use of ChatGPT for text generation.

Here, it is presented an AI model-independent measure to support the discrimination between hypotheses on the authorship of various multilingual texts, differentiating between those written by humans and those produced by AI. The syntax of texts written by humans tends to differ from that of texts produced by AIs. This difference can be captured and quantified, even with short texts. A probabilistic approach is implemented, using the Bayes factor to offer a consistent classification criterion and meet the efficiency criteria required for the evaluation of forensic evidence.

Analyses performed on multilingual texts covering different scientific and humanities areas reveal the feasibility of successful authorship discrimination with limited misclassification rates. The issue of literary originality can also be approached in full compliance with the standards for evaluative reporting and legal jurisprudence.