Automated text categorization in a dead language. The detection of genres in Late Egyptian

This paper is a first step in applying machine learning methods typical of Automated Text Catego-rization (ATC) for Automatic Genre Identification (AGI) in Late Egyptian, a language written in either hieroglyphic or hieratic scripts that is found in documents from Ancient Egypt dating from ca. 1350-700 BCE. The study is divided into three parts. After a general intro¬duction on AGI (§1), we introduce the levels of annotation that are integrated in the Ramses corpus and can be used when performing AGI on Late Egyptian (§2). In the following section (§3) we offer a brief survey of the types of features that have been discussed in the literature on AGI, before proceeding with three case studies where we apply supervised machine learning methods — namely the naïve Bayes classifier (§4.1), the Support Vector Machine (§4.2), and the Segment and Combine approach (§4.3) — to a selection of texts in the corpus. Their respective performances are tested using lexical, part-of-speech and inflectional features.