Filteryedping: Design Challenges and User Performance of Dwell-Free Eye Typing
Diogo Pedrosa, Maria da Graça Pimentel, Amy Wright, Khai N. Truong · 2015 · ACM Transactions on Accessible Computing · doi:10.1145/2724728
Summary
This paper introduces Filteryedping, a dwell-free eye typing technique designed to help people with severe motor disabilities communicate through eye gaze. Traditional dwell-based eye typing requires users to fixate on a key for a specific duration (typically 400-1000ms) to select it, which creates two problems: the Midas touch problem (unintended selections) and slow typing rates. Filteryedping addresses this by allowing users to simply look at letters sequentially without dwelling, then using algorithmic filtering to identify the intended word. The technique works by capturing all letters the user gazes at, filtering out accidentally viewed letters, and ranking possible words based on length and frequency from a 133,000-word corpus. When the user looks at the candidate list area, the system presents ranked word suggestions. The researchers compared Filteryedping against a shape-based approach (similar to gesture typing) and found the filtering method significantly outperformed it. The evaluation proceeded in two phases. Phase 1 involved 10 participants without motor disabilities comparing Filteryedping to AltTyping, the fastest dwell-based eye typing tool in the literature. Phase 2 was an iterative design study with six participants who had ALS or Duchenne muscular dystrophy, conducted in participants' homes. This second phase was crucial for identifying real-world challenges that don't appear in lab studies with non-disabled users.
Key findings
Participants without disabilities achieved an average of 15.95 words per minute with Filteryedping after approximately 100 minutes of practice, compared to 11.71 WPM with AltTyping—a 37% improvement. Error rates were also lower with Filteryedping (0.64% vs 3.2% MSD error rate). Subjective workload assessments showed Filteryedping required less effort than dwell-based typing. For participants with ALS and DMD, results were more nuanced. Average typing rates were 7.60 WPM with Filteryedping versus 6.36 WPM with AltTyping. The study revealed critical challenges: eye tracker calibration quality varied dramatically between users (particularly problematic for glasses wearers), and some ALS participants exhibited slower saccade velocities that caused the system to capture many unintended letters during eye movements. The researchers developed two key adaptive parameters: a "focus dwell" time (100ms) to prevent accidental activation of the candidate list, and a "slow movement threshold" to filter out letters captured during slow saccades. After implementing these parameters, the participant with slow saccades improved from 2.47 WPM to 3.24 WPM with reduced error rates. Eleven of twelve Phase 1 participants preferred Filteryedping over dwell-based typing.
Relevance
This research has significant implications for designing eye-gaze AAC systems. The finding that dwell-free typing can be substantially faster than dwell-based methods challenges the dominant paradigm in commercial eye tracking systems. For practitioners selecting or configuring AAC devices, this suggests exploring dwell-free options, especially for users frustrated by slow dwell-based input. The iterative design with actual users with motor disabilities reveals critical lessons: laboratory studies with non-disabled participants miss important issues like calibration difficulties with corrective lenses, saccade velocity impairments, and the progressive nature of conditions like ALS. The adaptive parameters (focus dwell, slow movement threshold) demonstrate that eye typing systems need customization options to accommodate individual differences in eye movement characteristics. For developers building eye-gaze interfaces, the technical solutions are transferable: filtering algorithms that tolerate input noise, adaptive thresholds based on user calibration quality, and visual/auditory feedback to confirm selections. The research also highlights that participants using head mice expressed interest in applying Filteryedping's approach to their existing input method.
Tags: eye tracking · eye typing · dwell-free · motor disability · ALS · Duchenne muscular dystrophy · AAC · text entry · gaze interaction