An Original Generative AI Framework for Automated Accessibility Content Adaptation for Nigerians with Visual Impairments

The majority of digital content produced in Nigeria is inaccessible to visually impaired users due to the absence of alt text, screen-reader-compatible layouts, and audio description, representing a serious equity and human rights concern. Existing automated accessibility tools are designed for English language content without accommodation for Nigerian English lexical patterns or Nigerian contextual references. This study develops an original generative AI framework for automated accessibility content adaptation targeting Nigerian users with visual impairments. The framework, designated NigeriaAccessGen, introduces four original contributions: a multimodal image description generator fine-tuned on a dataset of 45,000 Nigerian contextual images (markets, festivals, government services, medical settings) paired with professionally written alt-text descriptions annotated for cultural specificity; a Nigerian English simplification model based on a fine-tuned LLaMA-2 architecture that translates complex digital content into plain language maintaining Nigerian English conventions; an automated WCAG 2.1 compliance remediation pipeline that detects and corrects the 12 most common Nigerian government website accessibility violations; and a user evaluation methodology involving 80 visually impaired Nigerians using screen readers across six geopolitical zones. NigeriaAccessGen image descriptions were rated 34 percent more culturally relevant and 21 percent more contextually accurate than Google Vision API descriptions for Nigerian images. Automated WCAG remediation improved target site compliance from 31 percent to 74 percent. The framework constitutes an original contribution to accessibility AI for African digital contexts.

Keywords: generative AI, digital accessibility, visual impairment, Nigeria, WCAG compliance