Project Description
My research tackles the digital divide in AI, where under-represented communities are often excluded due to systems prioritizing dominant languages and cultures. We focus on making Large Language Models (LLMs) accessible and relevant by considering the unique linguistic, cultural, and ethical frameworks of these groups.
Currently, we are adapting LLMs for diverse Arabic dialects, moving beyond standard Arabic to incorporate specific linguistic features and embed cultural sensitivities. This involves fine-tuning models with dialectal data and co-designing evaluations with native speakers to ensure the AI is accurate, appropriate, and truly beneficial from the community's perspective.
The ultimate goal is to foster digital inclusivity, empowering these communities with AI tools that resonate with their identity. We believe our work with Arabic dialects can offer a valuable framework for adapting LLMs to other under-resourced languages, contributing to a more equitable global AI landscape.
Research Classification
- Computational languages (including domain-specific languages)
Research Interests
- Neural Networks
- LLMs
- Domain Adaptation