A Model of Keyboard Configuration Requirements
Shari Trewin, Helen Pain · 1998 · Proceedings of the Third International ACM Conference on Assistive Technologies (Assets '98) · doi:10.1145/274497.274530
Summary
This paper from the University of Edinburgh presents a user model — a computer program that observes a person's typing behaviour and automatically identifies keyboard difficulties, then recommends appropriate configuration settings. The model addresses four common keyboard problems experienced by people with motor disabilities: long key press errors (unintentionally holding a key long enough to trigger repeats), difficulty using modifier keys (inability to press two keys simultaneously), additional key errors (accidentally pressing adjacent keys), and bounce errors (unintentionally pressing a key multiple times). For each problem, the model recommends settings for existing operating system accessibility features: Repeat Keys (adjusting or disabling key repeat delay), Sticky Keys (latching modifier keys so they don't need to be held), and Bounce Keys (introducing a delay to prevent duplicate key presses). The model operates unobtrusively by intercepting keyboard events during normal word processing, requiring no special test tasks. It was evaluated with 30 participants — 20 with motor disabilities (including cerebral palsy, tetraplegia, muscular dystrophy, stroke effects, multiple sclerosis, and rheumatoid arthritis) and 10 without disabilities. Participants typed passages of approximately 625 keystrokes each on Macintosh computers, and the model's recommendations were compared against observed errors and user opinions.
Key findings
The model proved highly accurate in identifying users' keyboard difficulties. For Repeat Keys, the model's recommended delay setting eliminated all long key press errors for 14 of 23 participants who tried it, dramatically reducing error rates — one participant went from 74.4% errors under default settings to far fewer with the recommended delay. The model typically stabilised its recommendation within just 20 keystrokes. For Sticky Keys, the model's evidence scores showed statistically significant correlation with users' own opinions of the utility's usefulness (Spearman Rho = 0.472, p < 0.05). All five one-handed typists rated Sticky Keys at least "useful" and received a "yes" recommendation. The model detected 55% of additional key errors with strong correlation to actual error rates (Spearman Rho = 0.947, p < 0.01). For Bounce Keys, the model correctly identified the four participants with the highest bounce error rates without ever recommending it for users who didn't need it. Crucially, the model made no irrelevant positive recommendations for either the disabled or non-disabled groups — it never suggested a facility to someone who didn't need it. Only 35% of participants with disabilities had access to a computer teacher, highlighting the need for automated configuration support.
Relevance
This paper is foundational for the concept of automated accessibility configuration — the idea that a computer can observe a user's behaviour and proactively suggest accessibility settings, rather than requiring users to discover and configure these features themselves. The finding that only 35% of participants had expert help available remains painfully relevant today: most users with motor disabilities still lack guidance in configuring keyboard accessibility features that could dramatically reduce their errors. The user modelling approach demonstrated here — unobtrusive observation, evidence-based recommendation, dynamic adaptation — has influenced modern accessibility features in operating systems, including the accessibility setup assistants in macOS and Windows. For practitioners, the paper's taxonomy of four keyboard error types (long press, modifier, additional, bounce) provides a useful framework for assessing keyboard accessibility needs. The work also demonstrates that configuration is not a one-time task: individual needs can vary day to day, reinforcing the need for adaptive systems rather than static settings.
Tags: keyboard accessibility · motor impairment · user modelling · assistive technology · input devices · adaptive interfaces · empirical studies