Technical Summary: Beyond One-Size-Fits-All Engagement in Generative AI-Supported Speaking Practice

1. Problem Statement

While generative artificial intelligence (generative AI) has established low-threat, personalized environments for English as a Foreign Language (EFL) speaking practice, existing literature largely relies on quantitative approaches to report group-level trends or mediating effects. These studies often treat emotional, cognitive, and behavioural engagement as separate variables or assume they develop synchronously. Consequently, there is a significant gap in understanding how individual learners subjectively configure these dimensions. Specifically, it remains unclear how different learners' emotional experiences shape their cognitive appraisals and behavioural participation, and whether a "one-size-fits-all" approach to AI-supported instruction is sufficient given the heterogeneity of learner experiences.

2. Methodology

This study employs Q methodology, a person-centred approach designed to investigate individual subjectivity and identify shared viewpoints within a group.

• Participants: 27 first-year undergraduate students (22 male, 5 female; average age 19.04) from a science and engineering college in northern mainland China. All participants had at least one semester of experience using AI tools (e.g., Doubao, Chaoxing Learning Platform, Liulishuo English) for speaking practice.
• Instrument Development (Q-Set): A Q-set of 47 representative statements was constructed from three sources: (1) a review of literature on learner engagement and emotional experiences in AI-supported learning; (2) open-ended responses from participants regarding their real experiences; and (3) user comments from popular AI speaking platforms in China. Statements were refined to focus on single dimensions (emotional, cognitive, or behavioural) to avoid ambiguity.
• Data Collection:
  • Q-Sorting: Participants ranked the 47 statements on a forced-distribution grid from "strongly disagree" (-5) to "strongly agree" (+5).
  • Post-Sort Interviews: Semi-structured interviews were conducted with 10 participants exhibiting high factor loadings to explore the reasoning behind their rankings.
• Data Analysis: Q-sort data were analyzed using KADE (Version 1.3.1) via Principal Component Analysis (PCA) with Varimax rotation. Factors were retained based on eigenvalues > 1.00 and factor loadings ≥ 0.38. Four factors were extracted, explaining 51% of the total variance.

3. Key Results: Four Learner Engagement Profiles

The factor analysis identified four distinct learner engagement profiles, each characterized by a unique configuration of emotional, cognitive, and behavioural engagement:

1. Relaxed Regulators (Factor 1):

   • Emotional: High sense of relaxation and security; low anxiety regarding evaluation or mistakes.
   • Behavioural: Dynamic adjustment of interaction styles to suit personal needs (e.g., asking AI to simplify language).
   • Cognitive: No salient distinguishing cognitive engagement; processing appears superficial or focused on practice rather than deep reflection.
   • Profile: Learners who feel safe and in control but may not engage deeply cognitively.

2. Detached Doubters (Factor 2):

   • Emotional: Detached and low emotional engagement; skepticism regarding the "human" quality of AI feedback.
   • Cognitive: Critical and skeptical stance; questioning the accuracy of AI feedback and its ability to assess real proficiency.
   • Behavioural: Passive use or limited participation (e.g., using AI only when mandated by teachers).
   • Profile: Learners who lack trust in the tool, leading to disengagement.

3. Anxious Improvers (Factor 3):

   • Emotional: High sensitivity to scores and feedback; experience anxiety and emotional fluctuation.
   • Cognitive: Clear recognition of AI's value for improvement, but a rigid focus on specific weaknesses.
   • Behavioural: Targeted, repetitive practice of identified weaknesses; reluctance to explore other AI functions or features.
   • Profile: Learners driven by extrinsic motivation to avoid failure, resulting in rigid, non-exploratory practice.

4. Curious Explorers (Factor 4):

   • Emotional: Positive, activating emotions driven by intrinsic interest, curiosity, and novelty.
   • Cognitive: Positive appraisal of AI's evaluative and instructional value.
   • Behavioural: High initiative; active experimentation with different AI tools and features.
   • Profile: Learners who are intrinsically motivated, exploring broadly and engaging deeply across all dimensions.

4. Key Contributions and Findings

The study challenges the assumption that emotional, cognitive, and behavioural engagement occur synchronously. Instead, it reveals that these dimensions combine in heterogeneous, configurational patterns.

• The Centrality of Emotion: Emotional engagement is the central dimension across all profiles. However, the type of emotion (activating vs. deactivating, positive vs. negative) dictates the subsequent cognitive and behavioural outcomes.
• Control-Value Theory Application: The study applies Control-Value Theory (Pekrun, 2006) to explain these configurations:
  • Relaxed Regulators: High perceived control and positive value appraisal lead to deactivating positive emotions (relaxation), which may inhibit deep cognitive processing.
  • Curious Explorers: High control and positive value appraisal lead to activating positive emotions (enjoyment/novelty), fostering deep cognitive and exploratory behavioural engagement.
  • Anxious Improvers: High value appraisal but low perceived control leads to activating negative emotions (anxiety), which drives targeted behavioural effort but inhibits exploration.
  • Detached Doubters: Low control and negative value appraisal lead to deactivating negative emotions (boredom/detachment), undermining both cognitive appraisal and behavioural engagement.

5. Significance and Pedagogical Implications

The paper argues that a "one-size-fits-all" approach to generative AI-supported speaking instruction is insufficient. The significance of this work lies in its demonstration that learner engagement is heterogeneous and configurational.

Pedagogical Implications:

• Emotional Support Priority: Instruction should prioritize emotional support to leverage the low-threat nature of AI.
• Tailored Interventions:
  • For Relaxed Regulators: Maintain the low-stress environment but introduce reflective journals to deepen cognitive engagement.
  • For Detached Doubters: Use engaging, interactive games to build positive emotional experiences and gradually introduce tasks with increasing cognitive demands to build a sense of control.
  • For Anxious Improvers: Guide learners to set incremental goals and critically evaluate AI feedback to reduce anxiety and shift from extrinsic to more sustainable motivation.
  • For Curious Explorers: Assign open-ended, creative tasks to maintain intrinsic motivation and encourage peer sharing to reinforce engagement.

6. Limitations

The author acknowledges two primary limitations:

1. Sample Representativeness: The study was conducted with a single cohort from one university in China, limiting the generalizability of findings across different ages, proficiency levels, and geographical regions.
2. Static Perspective: The study captures engagement profiles at a single point in time and does not explore how these profiles might evolve longitudinally during sustained learning.

The paper concludes that teacher guidance is crucial in AI-integrated environments to address these distinct emotional-cognitive-behavioural configurations and promote sustained engagement.