The Fossilized Rhotic: Acoustic Rigidity and AI-Detected Entrenchment in Tikriti Arabic Speakers’ English /r/

AbstractThis study investigates the fossilization of English /r/ among native speakers of the Tikriti dialect of North Mesopotamian Arabic a phonetically conservative variety underrepresented in second language (L2) phonetics research. Drawing on experimental phonetics and self-supervised artificial intelligence (AI), we examine whether persistent rhotic substitution reflects acoustic rigidity and neural entrenchment. Twenty Tikriti Arabic–English bilinguals and ten native English controls produced English words containing /r/ across initial, post-consonantal, and vocalic contexts. Acoustic measurements (F1–F3, duration) were extracted using Praat, while contextualized speech representations were obtained via Wav2Vec 2.0. Results reveal that Tikriti speakers consistently produce /r/ with elevated F3 (>2200 Hz), characteristic of alveolar trills, contrasting sharply with native English /r/ (F3 < 1800 Hz). Critically, within-speaker variability in F3 was significantly lower among Tikriti participants (M = 45 Hz) than controls (M = 95 Hz), indicating phonetic rigidity. A Random Forest classifier achieved 92% accuracy in distinguishing fossilized tokens using combined acoustic and AI features. These findings support the AI-Mediated Automatization Theory (Al-Kasab, 2025) fossilization arises not from learning failure but from over-automatized sensorimotor routines that resist re calibration. The convergence of human acoustics and AI embeddings offers a novel framework for detecting entrenched pronunciation patterns. Implications for L2 phonology, cross-linguistic transfer, and AI-augmented phonetics are discussed.