Artificial Intelligence as a Catalyst for Nonlinear Skill Development: Evaluating Adaptive and Collaborative AI Tools for Transforming Academic Speaking Proficiency Among EFL Learners

Postgraduate EFL learners often struggle to attain academic speaking proficiency because traditional instruction offers limited opportunities for individualized practice, delayed feedback, and inadequate support for anxiety regulation. This quasi-experimental study evaluated whether adaptive and collaborative AI can address these constraints by comparing three AI-driven interventions—personalized learning pathways (PLP), conversational AI tutors, and AI-mediated collaborative platforms—with traditional computer-assisted language learning (CALL) in 414 postgraduate EFL students. A pretest–posttest design with analysis of covariance, rather than raw-score comparison, statistically controlled baseline speaking performance. Outcomes targeted fluency (speech rate, pause frequency), accuracy (error density), and syntactic complexity (clauses per T-unit), as well as plus metacognitive awareness and speaking anxiety. PLP’s adaptive algorithms produced the largest and most educationally meaningful gains: fluency improved by 3.21 points; accuracy improved through an approximately 37.5% reduction in errors; and complexity increased by 1.8 clauses per T-unit. These improvements were statistically reliable, exceeded those of the other AI conditions and CALL, and corresponded to very large effects sizes. PLP learners also reported enhanced metacognitive self-monitoring (approximately 88%) and an approximately 44% reduction in speaking anxiety. Findings indicate that well-designed AI can disrupt fossilized errors, accelerate nonlinear skill integration, and create responsive, learner-centered speaking environments for advanced EFL learners.