← Back to Research

Dissertation · In Progress

A cross-sectional latent profile analysis with exploratory longitudinal validation examining how global language learners differ in their AI-mediated self-directed learning and how those patterns evolve over time.

Latent profile analysis Person-centered analytics Longitudinal mixed methods N = 693

Foundation

YouTube studies → Interview studies → Systematic reviews → AI-SDL Framework → PA-SDA Scale

Dissertation

Core Study 1 — Learner Profiling & Longitudinal Tracking → Core Study 2 — Learning in the Loop
693 Wave 1 participants (global, multilingual)
29 Wave 2 follow-up with qualitative interviews
3 latent profiles identified at baseline
4 transition mechanisms from qualitative integration

Theoretical Foundation — AI-SDL Framework

Built on Song and Hill's (2007) SDL model, the AI-Integrated Self-Directed Learning Framework (Li et al., 2024) recast learner attributes as five AI-contextualized dimensions — self-efficacy, attitude, motivation, strategy use, and resource use — and recast the learning process as autonomous-adaptive rather than linear. The framework operates across local and global levels, situating learner–AI interaction within sociocultural, institutional, and design contexts.

AI-SDL Framework: global and local levels
Global and local level structure of the AI-SDL Framework.
Autonomous-adaptive process cycle
The autonomous-adaptive process: planning, monitoring, adapting, evaluating.

Core Study 1 — Learner Profiling & Longitudinal Tracking

Research design: Wave 1 cross-sectional LPA, Wave 2 fixed-reference follow-up, mixed-methods integration
Study design: Wave 1 cross-sectional LPA (N = 693, late 2024) → Wave 2 fixed-reference longitudinal validation (n = 29, mid-2025) → convergent mixed-methods integration.
Sample flow from 973 starts to three profiles and Wave 2 follow-up cohort
Sample flow: 973 starts → 693 complete cases retained → three profiles (Lower-standing 4.78%, Mid-range 71.43%, Higher-standing 22.81%) → 29 Wave 2 follow-up participants.
Profile comparison across five PA-SDA dimensions: Wave 1 and Wave 2 mean scores
Mean PA-SDA scores across five dimensions by profile, comparing Wave 1 baseline (solid lines) to Wave 2 follow-up (dashed lines) under the fixed reference model.
Fixed-reference transitions across profiles from Wave 1 to Wave 2
Fixed-reference transition diagram. Four pathways observed: stable mid-range (48.3%), higher-to-mid recalibration (20.7%), mid-to-higher growth (13.8%), stable higher (17.2%). Zero movement into or out of the lower-standing profile.

Core Study 1 Finding

AI-mediated self-directed learning is heterogeneous at baseline and largely stable over time. Apparent decline often reflects strategic recalibration — not regression. The GenAI learner landscape is dominated by a mid-range majority capable yet not yet fully strategically integrated.

Four transition mechanisms: bounded augmentation, iterative refinement, depth-first recalibration, strategic partnership
Four qualitative mechanisms explain the human logic behind profile transitions, with quantitative signals and exemplar voices from interview data.
PCA space showing observed movement across four transition pathways
Observed movement through the fixed-reference personal-attribute space. Panels share the same PCA coordinate system, showing only the four observed transition pathways.

Core Study 2 — Learning in the Loop

Design

A phenomenologically informed qualitative study with the same Wave 2 cohort (n = 29). Semi-structured interviews captured learning moments, meaning-making, and boundary-setting as they occurred in lived AI-mediated self-directed learning practice.

Key Finding

AI entered the execution of self-directed learning through an iterative loop of prompting, evaluating, verifying, and adapting. Agency showed up through verification routines, attempt-first rules, and selective trust — not through frequency of use.

Learning Moment Types

Conceptual moments (AI surfaced new frameworks), expressive moments (AI offered linguistic alternatives), reflective moments (AI prompted metacognitive awareness), and generative moments (AI co-produced new ideas). Moment types mapped onto learners' existing strategic profiles.

Constraint Layer

Policy ambiguity, cost, ethical concerns, and authentic voice shaped what learners were willing to do with AI — functioning as a global-level layer that bounded local learner–AI interaction regardless of individual attributes.

Core Study 2 Finding

AI in self-directed learning behaved less like a passive tool and more like a co-regulatory partner — entering the execution of learning rather than merely surrounding it. What mattered most was not how often learners used AI, but whether they retained interpretive authority over its outputs.

Related Evidence Base

This dissertation builds on a connected line of published work spanning early empirical studies, framework development, and scale validation.

Reconceptualizing SDL × GenAI PA-SDA Scale (System) SDL with GenAI: YouTube Creators Multilingual ChatGPT Applications