Universally Designed Text on the Web: Towards Readability Criteria Based on Anti-Patterns

The readability of web texts affects accessibility. The Web Content Accessibility guidelines (WCAG) state that the recommended reading level should match that of someone who has completed basic schooling. However, WCAG does not give advice on what constitutes an appropriate reading level. Web authors need tools to help composing WCAG compliant texts, and specific criteria are needed. Classic readability metrics are generally based on lengths of words and sentences and have been criticized for being over-simplistic. Automatic measures and classifications of texts' reading levels employing more advanced constructs remain an unresolved problem. If such measures were feasible, what should these be? This work examines three language constructs not captured by current readability indices but believed to significantly affect actual readability, namely, relative clauses, garden path sentences, and left-branching structures. The goal is to see whether quantifications of these stylistic features reflect readability and how they correspond to common readability measures. Manual assessments of a set of authentic web texts for such uses were conducted. The results reveal that texts related to narratives such as children's stories, which are given the highest readability value, do not contain these constructs. The structures in question occur more frequently in expository texts that aim at educating or disseminating information such as strategy and journal articles. The results suggest that language anti-patterns hold potential for establishing a set of deeper readability criteria.