📄 social_science

Effects of Human Centered AI Supported Instruction on EFL Listening With Adaptive Feedback Gamified Engagement and Motivational Scaffolding

This mixed-methods study involving 53 Iranian high school students demonstrates that a gamified, AI-driven listening instruction system incorporating adaptive feedback and motivational scaffolding significantly improves EFL learners' listening proficiency, self-efficacy, and engagement compared to a non-gamified conversational AI approach.

Original authors: Aliakbar Tajik, Atefeh Karkhaneh

Published 2026-06-24

📖 5 min read🧠 Deep dive

CC BY 4.0

Original authors: Aliakbar Tajik, Atefeh Karkhaneh

Original paper licensed under CC BY 4.0 (https://creativecommons.org/licenses/by/4.0/). ✨ This is an AI-generated explanation of the paper below. It is not written or endorsed by the authors. For technical accuracy, refer to the original paper. Read full disclaimer

The Big Picture: Two Different Ways to Learn Listening

Imagine you are trying to learn how to listen to a foreign language (English) in a classroom in Iran. The researchers wanted to see if using Artificial Intelligence (AI) could help students get better at this. But they didn't just use any AI; they compared two very different "coaches."

Coach A (Duolingo): This is like a video game. It gives you points, badges, levels, and a "streak" counter if you don't miss a day. It feels like playing a game where you are constantly rewarded for getting things right.
Coach B (Replika): This is like a friendly pen pal. It's a chatbot that talks to you. It's helpful and listens to you, but it doesn't give you points, leaderboards, or game levels. It's just a conversation.

The study asked: Which coach helps teenagers (aged 14–15) get better at understanding spoken English faster?

The Experiment: A 12-Week Race

The researchers gathered 53 students and split them into two teams. Both teams practiced for 12 weeks (24 sessions total).

Team Duolingo used the game-like app.
Team Replika used the chatbot app.

Before they started, both teams took a listening test to make sure they were on the same starting line. They were. Then, they practiced, and at the end, they took the test again.

The Results: The Game Coach Won

Here is what happened:

Both teams got better. Just like running a race, both groups improved their listening skills over the 12 weeks.
Team Duolingo got much better. The students using the game-like app scored significantly higher on the final test than the students using the chatbot.

The Analogy:
Think of learning to listen like climbing a mountain.

Replika was like having a guide who walks with you and talks to you. It was nice, but you had to find your own way up the mountain.
Duolingo was like a guided tour with a fun map. Every time you took a step, you got a "ding!" and a gold star. The map showed you exactly how high you were climbing, and the game made you want to keep going just to see what was at the next level. The "game" part kept the students motivated to climb higher and faster.

Why Did the Game Work Better?

The researchers looked at why the game worked so well. They found that the "game" elements weren't just for fun; they acted like training wheels for the brain.

The "Just Right" Challenge: The game automatically adjusted the difficulty. If a student was struggling, it made the task easier. If they were doing well, it made it harder. This kept them in a "sweet spot" where they weren't bored, but also weren't overwhelmed.
Instant Feedback: When a student made a mistake, the game told them immediately. It was like a coach standing right next to you saying, "Try that again," rather than waiting until the end of the week.
The "Streak" Effect: The game tracked how many days in a row they practiced. This created a "don't break the chain" feeling that kept students coming back every day.
Confidence Boost: Because the game gave them small wins (badges, points), the students felt more confident. They weren't afraid to make mistakes because the game made it feel like part of the fun.

What the Students Said

The researchers also talked to the students.

They said the game made them feel less stressed. In a normal class, if you don't understand something immediately, you feel embarrassed. With the app, they could pause, rewind, and try again as many times as they wanted.
They loved the progress bars. Seeing a bar fill up as they learned gave them a clear sense of how far they had come.
They felt like they were in control. They could practice whenever they wanted, not just when the teacher said so.

The "Secret Sauce" (The Models)

The researchers created some fancy names for their findings, but here is what they actually mean in plain English:

Tajik's Scaffolded Motivator Model (T-SMM): This is a fancy way of saying, "If you build a learning system that gives you the right amount of help (scaffolding) AND makes you feel good about your progress (motivation), you will learn faster."
The AI-Enhanced Language Learning Matrix: This is just a map showing how the computer's smart features (like adjusting difficulty) mix with human feelings (like wanting to win a game) to create a super-learning environment.

The Bottom Line

The study concludes that for teenagers learning to listen to English, turning learning into a game works better than just having a chat.

It's not that the chatbot (Replika) was bad; it was helpful. But the game (Duolingo) had extra tools—points, levels, and instant rewards—that kept students engaged, motivated, and practicing more consistently. The "game" didn't just make learning fun; it made the brain work harder and smarter by keeping the students in a state of flow where they were challenged but supported.

In short: If you want to help a teenager learn to listen to a new language, give them a game that rewards them for every small step they take, rather than just a conversation partner. The points and badges act like fuel for their learning engine.

Technical Summary: Effects of Human-Centered AI Supported Instruction on EFL Listening

Problem Statement
Listening comprehension is a cognitively demanding yet frequently underemphasized component of English as a Foreign Language (EFL) instruction, particularly in adolescent classrooms within the Iranian educational context. Traditional instruction often prioritizes grammar and test preparation over communicative competence, resulting in minimal exposure to authentic listening materials, scarce structured feedback, and limited practice opportunities. While AI-supported learning platforms have emerged to supplement instruction, there is a lack of empirical research comparing how distinct instructional designs—specifically gamified, adaptive systems versus non-gamified conversational AI—impact listening development, learner engagement, and motivation in adolescent EFL learners. Existing studies often focus on general proficiency or isolate gamification effects without systematically contrasting these architectures in controlled classroom settings.

Methodology
The study employed a mixed-methods concurrent triangulation design involving 53 Iranian high school learners (aged 14–15) from public schools in Varamin. Participants were randomly assigned to two instructional conditions over a 12-week period (24 sessions):

Experimental Group (Duolingo): Utilized a gamified, AI-driven personalized learning system featuring progression tracking, leaderboards, points, badges, and structured, adaptive listening tasks.
Comparison Group (Replika): Utilized a non-gamified conversational AI platform designed for personalized dialogue practice without game-based reward mechanisms.

Data Collection & Instruments:

Quantitative: Pre- and post-intervention listening comprehension tests (aligned with Prospect 2 and Prospect 3 curricula) covering five subskills (Main Conversation, Specific Information, Classroom Instructions, True/False, Dialogue Completion). A researcher-developed 18-item questionnaire measured perceptions of engagement, usability, and motivation.
Qualitative: Semi-structured interviews with 15 participants and classroom observations using a structured checklist across 20 sessions.
Analysis: Quantitative data were analyzed using SPSS (paired/independent t-tests, ANOVA). Qualitative data underwent thematic analysis (Braun & Clarke, 2006) with hybrid coding (deductive/inductive).

Key Results

Listening Proficiency: Both groups improved, but the Duolingo group demonstrated significantly higher gains across all listening subskills compared to the Replika group.
- Total Post-test Scores: Duolingo ($M=31.32, SD=1.75$) vs. Replika ($M=26.42, SD=3.62 $);$ t(51)=6.085, p<.001$, with a large effect size ( $d=1.73$ ).
- Significant differences were found in Main Conversation Comprehension ( $p<.001$ ), Specific Information Identification ( $p<.001$ ), Classroom Instruction Understanding ( $p<.001$ ), True/False Analysis ( $p=.026$ ), and Dialogue Completion ( $p<.001$ ).
- One-way ANOVA indicated that Duolingo's improvements were evenly distributed across subskills ( $F(4, 125)=2.317, p=.078$ ), suggesting holistic development.
Engagement and Motivation: Questionnaire results indicated high engagement with Duolingo's gamified elements (Mean=4.32) and adaptive scaffolding (Mean=4.21).
Qualitative Findings: Thematic analysis revealed six key themes for the Duolingo group: adaptive difficulty progression, learner autonomy, engagement through achievement, comprehensible input, situated learning, and interactional learning. Students reported that the self-paced nature and gradual difficulty increase reduced anxiety and sustained motivation.
Triangulation: Classroom observations corroborated self-reported data, showing high rates of autonomous task adjustment (79%), use of replay features, and sustained participation during optimally challenging tasks.

Key Contributions
The paper makes three primary contributions to the field of AI-enhanced language learning:

Theoretical Synthesis: It integrates Schema Theory, Self-Regulation Theory, Dynamic Assessment (Sociocultural Theory), and Gamification Theory to explain how AI-mediated listening instruction functions. This synthesis moves beyond isolated theoretical applications to a multi-layered conceptual lens.
Model Development: Based on empirical findings, the authors propose two integrative frameworks:
- AI-Enhanced Language Learning Matrix (AELLM): Maps the interaction between algorithmic personalization and sociocultural scaffolding.
- Intelligent Learning Environment Design (ILED): Synthesizes adaptive progression, intrinsic motivation triggers, cognitive load management, and feedback loops.
- Tajik's Scaffolded Motivator Model (T-SMM): A context-specific explanatory model describing how feedback and motivational scaffolding support engagement and listening development over time.
Empirical Evidence: The study provides controlled comparative evidence that structured, gamified feedback combined with adaptive algorithms yields superior listening outcomes compared to open-ended conversational AI for adolescent EFL learners, challenging the assumption that AI integration alone ensures better results.

Significance and Claims
The authors claim that the study offers a research-validated blueprint for designing intelligent learning platforms. The significance lies in demonstrating that the synergy of motivational design (gamification), adaptive scaffolding, and repeated structured practice creates a "transformative, context-sensitive ecosystem" for language learning.

The paper explicitly states that its findings suggest:

Gamified AI environments can drive balanced progress in listening comprehension more effectively than non-gamified conversational tools in this specific demographic.
The success of such platforms relies on the alignment of machine-driven adaptability with research-validated pedagogical principles (e.g., Zone of Proximal Development, comprehensible input).
These design principles offer scalable solutions for EFL contexts with limited access to authentic listening materials.

The authors maintain a modest tone regarding causality, noting that while the results are robust, future research with larger samples and longitudinal designs (e.g., using ANCOVA to control for baseline variations) is necessary to fully isolate the causal influence of specific design features. They emphasize that the study interprets constructs like "emotional adaptivity" at a pedagogical and perceptual level rather than claiming the platforms utilize advanced affective computing or neurocognitive algorithms.