Speakers
Description
Electronic dictionaries pose particular accessibility challenges for blind screen reader users, yet the field lacks evaluation frameworks that account for the complexity of their information structures. This paper reports on a three-phase study comparing Automated Web Accessibility Evaluation Tools and Large Language Models in evaluating three online dictionary entries from different lexicographical traditions. Phase 1 establishes a baseline through automated WAET scanning and zero-shot LLM auditing. Phase 2 compares their outputs and derives a four-category framework, reframing compliance as a design question. Phase 3 tests two enhanced prompting strategies — literature-augmented and design-oriented. The former uses specific dictionary accessibility literature while the latter combines a structured checklist derived from Phase 1 with persona-based use-type simulation, surfacing a macro-to-micro hierarchy of dictionary-specific barriers. The study argues for an iterative, multi-condition hybrid workflow integrating WAETs, LLMs, and expert human review for e-dictionaries.