An Invisible Keyguard
Shari Trewin · 2002 · Proceedings of the Fifth International ACM Conference on Assistive Technologies (Assets '02) · doi:10.1145/638249.638275
Summary
This paper from IBM T.J. Watson Research Center presents OverlapKeys, a software-based alternative to physical keyguards that filters overlap typing errors — instances where two keys are pressed simultaneously when only one was intended. Overlap errors are common among keyboard users with motor disabilities, occurring when a finger accidentally presses an adjacent key while moving to the intended key, or when the user leans on a key unintentionally. Physical keyguards (thin boards with holes for each key that sit over the keyboard) are the standard hardware solution, but they are unpopular because they slow down typing, make key labels hard to see, must be purchased separately for each keyboard model, and are physically cumbersome. OverlapKeys aims to provide the same error reduction benefit invisibly through software. The paper evaluates three identification techniques — basic filtering (suppressing all overlapping keystrokes), timing-based filtering (using keystroke duration and overlap timing to distinguish deliberate from accidental presses), and language-based filtering (using character bigram frequencies to choose the most likely intended character) — and three correction techniques for determining which of two overlapping keys was intended.
Key findings
Using a corpus of typing data from 14 participants with motor disabilities (including stroke, cerebral palsy, multiple sclerosis, nerve damage, and arthritis), the timing-based approach performed best overall, correctly identifying 97% of overlap errors with a 100ms lock time threshold. The basic technique identified all overlaps but also flagged 28 deliberate keystrokes as errors (100% recall, poor precision). The timing-based filter with 100ms lock time achieved the best balance, identifying 96% of errors. For correction, the timing-based filter correctly resolved 80% of overlap errors, compared to 48% for the basic filter and 50% for the language-based filter. For the three participants with the highest error rates, combining identification and correction techniques gave a 50-75% reduction in total errors. The paper also derived a predictive formula (B = 0.7836 - 0.5cD, where c is the proportion of adjacent overlapping key presses and D is the proportion of deliberate overlapping presses) that can estimate the benefit of the timing-based filter for any individual user from a sample of their typing, without needing to actually run the filter. In 16 of 19 typing samples, the number of overlap errors exceeded the number of deliberate overlapping keystrokes, confirming that filtering would improve rather than degrade typing for most users in the study.
Relevance
This research addressed a practical, everyday accessibility problem with an elegant software solution that avoids the drawbacks of physical hardware. The approach of analysing keystroke timing patterns to distinguish intentional from accidental key presses is foundational to modern keyboard accessibility features. The finding that overlap errors are predominantly positioning errors (adjacent keys) rather than random provides a clear design target for filtering algorithms. The predictive formula for estimating individual benefit is particularly practical — it allows clinicians or users to assess whether OverlapKeys would help by simply analysing a typing sample, without committing to using the filter. The study also revealed important individual variation: participants with slower, more deliberate typing styles benefited most from timing-based filtering, while faster typists who intentionally overlapped keys (a normal typing technique called rollover) needed different approaches. This highlights a recurring theme in motor accessibility: solutions must be adaptable to individual typing styles rather than assuming a single model of "disabled typing."
Tags: keyboard accessibility · motor disability · typing errors · input filtering · assistive technology · keyguard · cerebral palsy · alternative input