In recent years, artificial intelligence (AI) has surged to the forefront of educational innovation, promising transformative impacts across diverse learning contexts. Yet, the potential of AI to specifically support students with learning differences, such as dyslexia, remains largely underexplored. A pioneering study published in ECNU Review of Education sheds light on this critical gap by rigorously investigating a novel AI-assisted writing platform named CHATTING, designed to aid secondary school students in both Chinese and English writing. The research provides a nuanced understanding of AI’s dual role in enhancing learner engagement while simultaneously raising complex challenges related to writing quality and academic integrity.
Dyslexia, a neurodevelopmental condition affecting approximately 5–10% of the global population, fundamentally disrupts reading and writing processes. Students with dyslexia often grapple with limited vocabulary retrieval, difficulties in organizing coherent ideas, and mechanical errors that undermine writing fluency. These challenges are further exacerbated in crowded classroom settings, such as those typically found in Hong Kong, where individualized instructional support is scarce. Against this backdrop, the development of remediative technologies holds particular promise, but the specificities required for effective AI design tailored to dyslexia have remained largely unaddressed—until now.
The research team, led by Fung K.Y. and colleagues, formulated CHATTING, a ChatGPT-powered system augmented with accessibility features including adjustable speech rates, integrated speech-to-text functions, and multilingual support stretching over Traditional Chinese, Cantonese, and English. These technical adaptations sought not only to accommodate diverse linguistic preferences but also to tailor interaction modalities in ways that reduce cognitive load. Unlike generic AI writing assistants, CHATTING was purpose-built with inclusivity as a core design principle, poised to adapt to student needs and foster autonomy in the writing process.
Central to the investigation was a controlled study involving 101 secondary school students, comprising both dyslexic and non-dyslexic learners. The participants were randomly divided into two groups: the experimental cohort received writing instruction supplemented by CHATTING, while the control group continued with conventional writing pedagogy. The intervention spanned four days, incorporating pre- and post-intervention writing tasks in both Chinese and English, thus enabling comparative analysis of performance and motivational shifts attributable to AI integration.
The researchers deployed an array of evaluative tools to capture the multifaceted dimensions of engagement. Drawing on Self-Determination Theory, they assessed behavioral, emotional, cognitive engagement, and intrinsic motivation through standardized questionnaires. Writing outputs were scrutinized for content richness, linguistic accuracy, and organizational coherence. Additionally, qualitative data were obtained through open-ended interviews and surveys, while Copyleaks plagiarism detection software was employed to identify unoriginal text, a crucial dimension seldom foregrounded in AI-education research.
Intriguingly, the findings articulate a complex interplay between enhanced learner engagement and unexpected declines in writing quality. Dyslexic students demonstrated remarkable improvements in emotional engagement (+16.57%) and intrinsic motivation (+8.71%) after using CHATTING, far surpassing their peers without dyslexia who exhibited more modest gains. Interview feedback underscored that dyslexic learners found CHATTING exceptionally helpful for idea generation and boosting confidence, benefiting from its interactive question-and-answer design—a departure from traditional didactic learning models.
However, these motivational advances were tempered by a paradoxical dip in writing performance across both groups. Post-intervention evaluations revealed reductions in overall writing scores for both Chinese and English tasks despite increased word counts. This suggests that while CHATTING facilitated greater writing volume, depth and quality suffered. The prevalence of plagiarism emerged as a significant concern, with dyslexic students primarily copying English texts and non-dyslexic students exhibiting similar tendencies in Chinese, indicating that authority and language complexity shaped dishonest practices.
A key insight unearthed by the study is the critical role of students’ question-asking proficiency in maximizing AI benefits. Participants who crafted specific, open-ended inquiries elicited more targeted, relevant AI-generated responses, enhancing their writing output. Conversely, students with limited questioning skills struggled to interpret AI feedback and resorted to direct text replication, underscoring the necessity of scaffolding critical thinking and information literacy alongside AI tool deployment.
The authors emphasize a balanced view on AI integration in education, underscoring that while platforms like CHATTING can substantially elevate engagement and motivation, they risk undermining essential writing competencies if implemented without structured guidance. The research advocates for teacher-facilitated AI incorporation strategies, wherein educators actively mediate AI use to support, rather than supplant, core learning processes such as ideation, drafting, and iterative revision.
Technical limitations surface prominently in the findings. The brevity of the intervention—limited to two days of AI-assisted writing—restricts conclusions about long-term impacts. Moreover, some AI outputs generated by CHATTING were excessively verbose or culturally misaligned, posing interpretative challenges for students navigating second-language acquisition. Such issues illuminate the imperative for developing refined AI algorithms featuring adaptive complexity control and culturally sensitive content moderation.
The study’s implications ripple beyond technology design to educational policy and ethics. Incorporating robust plagiarism detection mechanisms, embedding adaptive difficulty settings, and prioritizing learner comprehension are proposed as critical developmental priorities for future AI tools targeting special education. Simultaneously, policymakers are urged to establish clear ethical guidelines surrounding AI use, geared toward fostering academic honesty, preventing over-reliance, and ensuring inclusivity.
This pioneering research positions CHATTING as both a beacon of promise and a cautionary tale. It concretely demonstrates that AI-enabled writing systems can disrupt entrenched barriers faced by dyslexic learners, engendering heightened engagement and self-efficacy. Yet, it also unmasks the intricate challenges at the intersection of technology and pedagogy, where motivation gains must be carefully balanced against retention of fundamental writing skill development and academic integrity cultivation.
In sum, this investigation pioneers new frontiers in understanding AI’s role in differential learning contexts, bridging technical innovation with human-centered educational design. It underscores the necessity of integrating AI as a complementary scaffold—deliberately embedded within pedagogical frameworks, under active teacher supervision—to unlock its full potential without compromising the holistic development of writing skills.
Subject of Research: Not applicable
Article Title: A study on using ChatGPT to help students with dyslexia learn Chinese and English writing
News Publication Date: 10-Aug-2025
Web References: http://dx.doi.org/10.1177/20965311251358269
References:
Fung K. Y., Fung K. C., Lee L. H., Lui R. T. L., Qu H., Song S., and Sin K. F. (2025). A study on using ChatGPT to help students with dyslexia learn Chinese and English writing. ECNU Review of Education. DOI: 10.1177/20965311251358269
Keywords: Education, special education, dyslexia, AI in education, generative AI, ChatGPT, writing assistance, language learning, intrinsic motivation, engagement, plagiarism, educational technology
