Design and Evaluation of an Automatic Text Simplification Prototype with Deaf and Hard-of-hearing Readers
Oliver Alonzo, Sooyeon Lee, Akhter Al Amin, Mounica Maddela, Wei Xu, Matt Huenerfauth · 2024 · ASSETS '24: Proceedings of the 26th International ACM SIGACCESS Conference on Computers and Accessibility · doi:10.1145/3663548.3675645
Summary
This paper investigates the design preferences and interactions of Deaf and Hard-of-Hearing (DHH) readers with automatic text simplification (ATS) tools. Many DHH individuals who use sign language as their primary language experience English as a second language with different grammar and syntax, creating reading comprehension challenges with complex written text. ATS approaches can simplify text at two levels: lexical simplification (replacing difficult words with simpler synonyms) and syntactic simplification (breaking complex sentences into shorter, simpler structures). The research was conducted in two phases. In the formative phase, the authors explored the design space of ATS tools with DHH participants through semi-structured interviews and design activities, identifying preferred configurations including how simplifications should be presented (inline replacement, pop-ups, side-by-side), what level of simplification to apply (word-level, sentence-level, or both), and how much user control to provide. This phase revealed core user values: maintaining reading fluency and efficiency, retaining control over the tool, and avoiding disruption to the reading flow. In the evaluation phase, the authors built a prototype based on popular design choices — a browser extension that highlights simplifiable words and sentences, providing pop-up definitions for lexical simplifications and inline replacements for syntactic simplifications — and evaluated it with DHH readers who used it to read news articles. The study involved participants reading passages with and without the tool, measuring reading time, comprehension, and subjective experience.
Key findings
The formative study identified a tension between what DHH readers said they wanted and what actually supported their reading. Participants expressed strong preference for word-level lexical simplifications delivered through pop-ups — clicking on a highlighted word to see a simpler synonym or definition — because this felt least disruptive and gave them control. Sentence-level syntactic simplification was more divisive: some participants appreciated simpler sentence structures, while others found the rewritten sentences jarring or felt they changed the meaning. The evaluation revealed a key conflict between user values: the pop-up interactions that participants preferred for maintaining control actually slowed their reading speed compared to reading without the tool, as each pop-up interaction interrupted the reading flow. Participants also reported a perceived need to reread simplified text to verify that meaning was preserved, adding cognitive overhead. Despite these trade-offs, participants rated the tool as useful and expressed desire to use it in their daily reading. Word-level simplifications were consistently preferred over sentence-level ones — participants valued learning individual word meanings they could apply to future reading, whereas sentence-level rewrites felt like a crutch that did not build vocabulary. The design space exploration identified important configuration dimensions: granularity of simplification (word vs. sentence vs. paragraph), presentation mode (inline, pop-up, side-by-side, tooltip), activation mode (automatic, on-hover, on-click), and degree of user control (fully automatic, user-triggered, customisable thresholds).
Relevance
This research addresses a significant but often overlooked accessibility need — reading comprehension support for DHH individuals for whom English is a second language after sign language. Text simplification is typically discussed in the context of cognitive accessibility or plain language initiatives, but this work demonstrates its specific relevance to the deaf community. For accessibility practitioners, the finding that user preference and user benefit can conflict — participants preferred pop-ups that actually slowed their reading — is an important reminder that design decisions should be informed by both subjective preference and objective performance data. The design space framework (granularity, presentation mode, activation, control) provides a structured vocabulary for discussing ATS design choices applicable to any population that benefits from simplified text, including people with cognitive disabilities, non-native speakers, and older adults. The preference for word-level learning over sentence-level replacement has implications for how ATS tools are framed: as learning aids that build vocabulary rather than as permanent crutches that replace complex text. The work highlights the importance of evaluating assistive tools with their intended users rather than relying solely on NLP metrics like readability scores.
Tags: deaf and hard of hearing · text simplification · reading accessibility · natural language processing · lexical simplification · syntactic simplification · plain language · user study