Accessible Graphs - Exploring TikZ Code in Alt Text Generation
Supervisor(s): Sarah Ebling, Chuqiao Yan
Summary
Scientific papers increasingly contain complex figures and diagrams, but the accompanying alternative text (alt text) is often missing, incomplete, or insufficient for people who rely on screen reader software. This project investigates the current quality of alt text in the real-world scientific publications and explores whether the underlying source code (TikZ code) used to create figures can provide useful information for generating better alt text. This is suitable for Programming Task or Internal Internship.
Tasks:
1. Analyze and gain an overview of alt text quality from real scientific publications.
2. Understand how TikZ code can help with alt text generation and assist to build a dataset by leveraging LLMs.
Requirements
- Curious / passionate about Accessibility, AI topics
- Good Python Skill