A Framework for Context‐Aware and Adaptive Large Language Models for Autism Spectrum Disorder Therapy

Abstract

Gene mutations and exposure to toxins are two examples of environmental and genetic variables that may contribute to ASD. In addition, people with ASD may suffer with language and communication, sensory sensitivity, restricted interests, repetitive behaviors, and social relationship formation. A person's capacity to engage with the environment might be impacted by several challenges, which can vary in severity. The developmental disease known as autism spectrum disorder (ASD) has varying effects on individuals. Although early therapeutic interventions significantly improve developmental outcomes, existing therapy approaches often lack personalization, linguistic inclusivity, and real‐time adaptability—particularly for low‐resource languages such as Bangla. Recent advances in Large Language Models (LLMs) offer new opportunities to address these limitations through intelligent, adaptive, and empathetic conversational systems. This paper proposes a context‐aware and adaptive LLM‐based framework for ASD therapy, specifically designed for Bangla‐speaking children with ASD. The framework integrates Bengali‐BERT, mBERT, XLM‐R, IndicBERT, and BanglaT5 to enable emotion‐aware, personalized, and linguistically nuanced interactions. Specialized data preprocessing techniques—including data augmentation, noise filtering, text normalization, sentiment and intent labeling, syntactic‐semantic analysis, and context tagging—are employed to enhance contextual understanding and adaptability. The system is evaluated through internal performance metrics and real‐world interaction studies with ASD patients, followed by caregiver and clinician surveys. The proposed framework demonstrates the potential of multilingual LLMs to transform ASD therapy by providing scalable, inclusive, and adaptive support across clinical, educational, and home environments.