Article’s

Artificial intelligence (AI) has transformed second language (L2) pronunciation training by delivering real-time, segment-specific acoustic feedback yet theoretical models in L2 phonetics have not kept pace. This paper proposes the AI-Mediated Phonetic Automatization Theory (AIPAT), a domain-specific conceptual framework explaining how AI-driven feedback accelerates the shift from effortful to automatic speech production in adult EFL learners. Developed through empirical work with 50 Arabic-speaking English teachers from diverse Iraqi institutions, AIPAT defines true phonetic automatization as the convergence of two independent markers: (1) reduced speech onset latency (a cognitive indicator of processing efficiency) and (2) stabilized acoustic parameters, such as consistent frication duration for /θ/. The model rests on four core assumptions, including the Temporal Precedence Principle, which holds that reaction time improvements precede acoustic stabilization a reversal of traditional pedagogical sequencing. AIPAT outlines a three-stage developmental trajectory and generates falsifiable predictions about phoneme markedness, adult plasticity, and non-linear learning curves. While deliberately narrow in scope, this micro-theory offers a testable scaffold for experimental phonetics, intelligent tutoring systems, and L2 teacher education.