Defense Date: 2026/17/11
Student

Reza Bozorg Aliabadi

Department / Program Literature and Humanities | English language

Teacher-driven and AI-generated Dynamic Assessment: EFL Speaking Components and Perceptions via Ecological Systems Theory

Supervisor Saman Ebadi

Abstract

This mixed-methods study, drawing on Ecological Systems Theory (EST), compared the contributions of two Dynamic Assessment (DA) modalities to speaking fluency and accuracy among Iranian EFL learners. The participants comprised 60 B2-level male and female learners who were assigned to a teacher-initiated DA group (n=30) and an AI-generated DA group (n=30). The former received teacher-driven DA based on Aljaafreh and Lantolf’s graduated mediation, while the latter received feedback via the ChatGPT interface using the same protocol. Before and after the treatment, both groups took pre-/post-test IELTS speaking tests. The results of one-way ANCOVA showed that the teacher-driven DA group outperformed the AI-generated DA group in terms of speaking fluency. On the contrary, the results of one-way ANCOVA indicated that the AI-generated DA group outscored their counterparts in terms of speaking accuracy. The results of the paired sample t-tests demonstrated improvements in speaking fluency and accuracy for both groups. The thematic analysis of transcribed data from semi-structured interviews, framed by EST, revealed learners’ positive perceptions of teacher mediation; more specifically, the results showed that learners perceived teacher mediation as offering attuned microsystemic mediation, institutional mesosystemic alignment, integrative exosystemic processes, and a culturally resonant macrosystem. In contrast, the learners in the AI-generated DA perceived DA treatment to be decontextualized; they found DA to be mesosystemically disconnected and exosystemically fragmented. AI-generated DA was also viewed by learners to be culturally dissonant. The thematic analysis for speaking accuracy revealed contrasting perceptions: teacher-driven DA was viewed as constrained by grammar-focused microsystemic mediation, rigid mesosystemic alignment, exosystemic overemphasis, and culturally resonant macrosystemic authority. Conversely, AI-generated DA elicited predominantly positive perceptions, including immediate non-judgmental microsystemic correction, supplementary mesosystemic practice, self-tracking exosystemic affordances, and culturally neutral macrosystemic acceptance. These findings suggest a complementary pedagogical model may optimally address both fluency and accuracy development. Keywords: Fluency, Accuracy, Speaking, Dynamic Assessment (DA), Artificial Intelligence (AI), Ecological Systems Theory (EST), Perceptio