The Design Space of Nonvisual Word Completion
Hugo Nicolau, Andre Rodrigues, Andre Santos, Tiago Guerreiro, Kyle Montague, Joao Guerreiro · 2019 · Proceedings of the 21st International ACM SIGACCESS Conference on Computers and Accessibility (ASSETS 2019) · doi:10.1145/3308561.3353786
Summary
This paper presents the first design space for nonvisual word completion on mobile devices, addressing a gap where word completion interfaces — ubiquitous in sighted mobile typing — remain poorly designed for screen reader users. While sighted users can glance at three suggestions displayed above the keyboard and quickly accept one, blind users experience word completion through a fundamentally different, one-dimensional, sequential auditory channel. Current screen readers (TalkBack and VoiceOver) take a conservative approach: they only notify users of the auto-complete suggestion when the prediction engine has high confidence, and they present just one suggestion at a time. The design space covers seven categories: notification (when to alert users about suggestions — always, threshold-dependent, or input-dependent), output (how suggestions are communicated — implicit earcon vs. explicit speech), confidence (whether confidence levels are represented statically or dynamically through volume/pitch), cardinality (how many suggestions to present — one, two, or three), concurrency (whether multiple suggestions are read sequentially or simultaneously using spatialized audio), interruption (whether suggestion feedback stops when the user taps a new key or continues), and selection (shortcuts for accepting suggestions). The authors built a customizable prototype by modifying the Android Open Source Project keyboard to let users configure these parameters independently.
Key findings
A participatory design study with 11 legally blind participants (ages 38-57, all smartphone owners for 1-4 years) revealed that users consistently created interfaces that were less conservative than mainstream screen reader solutions. All 11 participants preferred "always" notifications — receiving word suggestions after every keystroke rather than waiting for a confidence threshold. This was a unanimous departure from TalkBack and VoiceOver defaults. Most participants (9/11) preferred multiple suggestions over just one, though the ideal number varied by individual — driven by a cost-benefit analysis weighing cognitive load against potential keystroke savings. Of those choosing multiple suggestions, 6 preferred sequential reading while only 3 preferred concurrent spatialized audio, as participants found it difficult to discriminate similar-sounding words played simultaneously. Eight participants preferred static confidence representation over dynamic volume scaling, finding uniform volume easier to process. All but two preferred interruptible feedback, wanting to keep typing without waiting for suggestions to finish reading. Four distinct interface patterns emerged: True Talkback (single suggestion, always notified), Sequence Announcer (2-3 suggestions read sequentially), Nonstop Typist (sequential suggestions that continue reading while user explores the keyboard), and Concurrent Announcer (all suggestions read simultaneously via spatialized audio). Context of use was a significant factor — participants noted that multiple suggestions work in quiet environments but become impractical outdoors with ambient noise.
Relevance
This work fills an important research gap by formalizing how word completion should work for screen reader users — a topic with zero prior research despite word completion being standard on every smartphone. The finding that all participants wanted more frequent notifications than current screen readers provide suggests that TalkBack and VoiceOver are being overly cautious, potentially costing blind users significant typing efficiency. The design space framework is valuable for any developer or researcher working on accessible text input, providing a structured way to think about the trade-offs between information richness and cognitive load in auditory interfaces. The participatory design approach is particularly noteworthy — rather than testing fixed designs, the researchers let blind users compose their own interfaces from design space parameters, revealing preferences that researchers might not have anticipated. The context-dependent nature of preferences (quiet room vs. noisy street) argues strongly for adaptive interfaces that adjust word completion behavior based on environment. For accessibility practitioners, this paper demonstrates that sighted interface patterns cannot simply be converted to audio — they must be redesigned from the ground up for the constraints and possibilities of nonvisual interaction.
Tags: blindness · text entry · mobile accessibility · screen readers · word prediction · touchscreen · design space · participatory design