Translator attribution for Arabic using machine learning.

This article focuses on the task of translator attribution for Arabic translations of world-famous books, investigating whether individual translators leave identifiable stylistic fingerprints despite the notion of "Translator Invisibility." Using a newly compiled corpus of 49 Arabic-translated books by six translators, the study applies preprocessing techniques including devocalization and morphological segmentation, then extracts stylometric features such as the top 100 most frequent words and function words. Supervised machine-learning algorithms, particularly tree-based classifiers like CatBoost, achieved up to 87% accuracy in distinguishing translators, indicating that translators' styles are detectable; morphological analysis did not significantly improve results over simple frequent word features. Unsupervised clustering further supported these findings, suggesting some translators have more visible stylistic signatures than others, and word-based features outperformed character-based ones in attribution accuracy.