Generative Artificial Intelligence and L2 Oral Proficiency in Higher Education: Effects on Fluency, Interactional Competence, and Learner Autonomy

Abstract

The integration of language models based on Generative Artificial Intelligence (AGI) has transformed the teaching and learning of second languages (L2) in the university environment. This study analyzes the pedagogical impact of conversational voice agents driven by IAGen on three core dimensions of orality: fluency (temporal and formulation), interactional competence (turntaking, co-construction of meaning and repair), and learner autonomy (selfregulation and reflective agency). Through an interdisciplinary analytical framework that combines Kormos' L2 model of speech production, the Interaction Hypothesis and the Self-Determination Theory, we examine how immediate feedback and practice in environments devoid of social judgment mitigate linguistic anxiety and accelerate the articulatory rhythm. However, the findings reveal a marked asymmetry: while fluency and self-regulation show significant quantitative improvements, interactional competence presents critical limitations derived from discursive condescension and the absence of authentic sociocultural contingency in algorithmic models. A hybrid pedagogical framework is proposed to articulate IAGen with teacher mediation.