CHATWELL: an AI-enabled adaptive tutoring system for improving mandarin composition skills in L2 students with learning difficulties.

Background: Conventional automated writing evaluation systems typically provide insufficient support for students with special needs, especially in tonal language acquisition such as Chinese, primarily because of rigid feedback mechanisms and limited customisation.

Objective: This research develops context-aware Hierarchical AI Tutor for Writing Enhancement(CHATWELL), an intelligent tutoring platform that incorporates optimised large language models to deliver instantaneous, customised, and multi-dimensional writing assistance for Chinese language learners, with special consideration for those with cognitive learning barriers.

Methods: CHATWELL employs a hierarchical AI framework with a four-tier feedback mechanism designed to accommodate diverse learning needs. In a 12-week quasi-experiment involving 100 HSK4 learners (including participants with dyslexia and other learning challenges), the system was compared against a conventional Bi-LSTM-based AES, assessing improvements in writing scores, error reduction, and accessibility metrics.

Results: The AI-powered CHATWELL group showed significantly greater score gains (14.2 vs. 6.8 points, t(98) = 7.85,  < 0.001), with marked reductions in grammatical (72.1%) and logical (75.3%) errors. Learners with disabilities particularly benefitted from its scaffolded feedback, with 89% rating the system as "highly accessible and effective." Engagement data revealed increased use of deeper feedback tiers, suggesting improved self-regulated learning.

Conclusions: The CHATWELL system substantiates the effectiveness of AI-augmented language instruction in advancing writing capabilities and promoting inclusive pedagogy for second language learners, including those with special educational requirements. These results emphasise the transformative potential of intelligent tutoring systems in addressing linguistic learning challenges.