Multimodal and Agentic AI for Cognitive Accessibility and Linguistic Preservation

Abstract

Artificial Intelligence (AI) has become a transformative tool in enhancing cognitive accessibility and preserving linguistic heritage. This thesis investigates the convergence of Generative AI (Gen AI), multimodal learning, and agentic AI to address challenges such as language learning and communication deficits faced by neurodivergent individuals, particularly those with Autism Spectrum Disorder (ASD), while simultaneously contributing to the preservation of endangered and lost languages in minority communities.

This research adopts a multi-faceted approach. First, it conducts a comprehensive literature review and technical analysis of multimodal AI applications for ASD clinical practices. Second, it designs and tests Gen AI tools, including Large Language Models (LLMs), for autism therapies and interactive social learning. Third, it examines the linguistic, cultural, and technical challenges of preserving endangered languages and proposes the Revitalization Framework of Linguistic Heritage (RFLH)—a Gen AI approach for reconstructing lost languages. Finally, it develops the Auto DB Agent, an agentic self-improving system capable of interpreting natural languages and generating SQL queries in dynamic database environments, lowering the barriers for non-technical users in healthcare and cultural data analysis.

By synthesizing these interconnected fields, this research contributes to human-centered AI design, expanding the accessibility of information and computational systems. The outcomes of this study have implications for education, historical linguistics, human-computer interaction, and AI-driven cognitive support, paving the way for more inclusive and intelligent AI systems that bridge the gap between digital information and diverse human needs.