From Letters to Words: Efficient Stroke-based Word Completion for Trackball Text Entry
Jacob O. Wobbrock, Brad A. Myers · 2006 · Proceedings of the 8th International ACM SIGACCESS Conference on Computers and Accessibility (Assets '06) · doi:10.1145/1168987.1168990
Summary
This paper presents a significant extension to Trackball EdgeWrite, a unistroke text entry method designed for people who use trackballs as their primary pointing device. The original Trackball EdgeWrite allowed users to enter text one character at a time by rolling a trackball toward the corners of a virtual square, with each corner sequence mapping to a letter. This new version elevates the system from character-level to word-level entry through a technique called stroke-based word completion. The key innovation is how word completions are selected: rather than requiring users to pick from a list (as most word prediction systems do), completions appear at the four corners of the EdgeWrite square and users select them through a fluid crossing gesture — a simple pulse of the trackball toward the desired corner. This design draws on goal crossing theory, which requires less precision than pointing, making it well-suited for users with motor impairments. The system uses frequency-based language models built from 850MB of news text, offering the four most likely completions for the current prefix at each corner. Context-dependent predictions using trigram models are also provided after word boundaries. The authors conducted two evaluations with a single participant ("Jim"), a 15-year trackball veteran with spinal cord injury, comparing Trackball EdgeWrite to the commercial on-screen keyboard WiViK, both with word prediction enabled.
Key findings
A theoretical model predicted that stroke-based word completion would be 45% faster than character-only entry, and empirical results closely matched this. In a controlled comparison, Trackball EdgeWrite with word completion achieved 12.09 words per minute versus 11.82 WPM for WiViK with word prediction — making the two methods statistically comparable in speed. Total error rates were 3.95% for EdgeWrite versus 2.21% for WiViK, also not significantly different. Critically, word-level Trackball EdgeWrite was 46.5% faster and 36.7% more accurate than the participant's prior peak performance with character-level EdgeWrite, and 75.2% faster and 40.2% more accurate than his prior peak with his preferred on-screen keyboard. Qualitative feedback revealed that the participant found WiViK visually tedious — requiring constant scanning between the keyboard, word list, and document — while EdgeWrite allowed him to stay focused on what he was writing. A two-month field deployment showed a 43.9% reduction in total unistrokes needed due to word completion, with an average of 3.11 characters saved per completion. Over 897 hours of intermittent use, the participant made 13,288 strokes, of which 2,201 were word-selection strokes.
Relevance
This research demonstrates that alternative text entry methods can match or rival commercial on-screen keyboards when augmented with intelligent word completion, while being less visually and cognitively demanding. The stroke-based word completion approach is particularly relevant for users with motor impairments who rely on trackballs — including people with spinal cord injuries, repetitive stress injuries, arthritis, or neuromuscular disorders. The crossing-based selection mechanism, which requires less precision than pointing at targets, represents an important design principle for accessible interfaces more broadly. The finding that reduced visual demand was the participant's primary reason for preferring EdgeWrite underscores that speed and accuracy alone do not capture the full user experience of assistive text entry — cognitive and visual load matter significantly. The field deployment data provides rare longitudinal evidence of real-world assistive technology adoption and usage patterns.
Tags: text entry · trackball · motor impairment · word prediction · word completion · assistive technology · spinal cord injury · alternative input