The moment for fair gaze‑typing tests has arrived
Eye‑tracking is no longer a bolt‑on. Windows includes Eye Control with a built‑in on‑screen gaze keyboard, dwell selection, word prediction, and a toggle for shape‑writing (Windows’ term for swipe/gesture typing). You can also tune separate dwell times for typing keys versus function keys. In short: gaze‑to‑text is officially part of the OS, not an experiment. (support.microsoft.com)
Apple, meanwhile, has shipped system‑level Eye Tracking for iPhone and iPad that uses the front‑facing camera—no extra hardware—and supports Dwell Control to “tap.” It works across iOS and iPadOS apps, integrating with features like AssistiveTouch and Voice Control. That means millions of devices can now navigate and select purely by gaze. (apple.com)
If the platforms have done their part, typing tests should catch up. Traditional WPM‑only leaderboards miss what makes gaze entry hard (and impressive): dwell‑time tuning, accurate target acquisition, error recovery, and smart candidate selection. Let’s fix that.
Why WPM isn’t enough for gaze keyboards
- Dwell timing is a speed/accuracy tradeoff. Research surveys and experiments commonly find “comfortable” static dwell times around 500–600 ms, though studies span 100–3000 ms depending on context and task. A score that ignores dwell tuning rewards risky settings and penalizes users who choose accuracy. (tandfonline.com)
- Gesture (shape) writing can change the game. In one real‑time eye‑controlled swipe‑typing study, average speed reached ~12.9 WPM versus ~3.8 WPM with a pure dwell method—over 3× faster—yet it still depends on clean starts/ends and good prediction. Benchmarks must account for this mode, not treat it as an “assist.” (sciencedirect.com)
- Interface strategies matter. Multi‑stage or region‑refinement keyboards have reported ~15 WPM with notable gains over older layouts, showing that layout and selection strategy strongly influence outcomes. Comparing users without noting the method is apples to oranges. (pmc.ncbi.nlm.nih.gov)
What a fair gaze‑typing test should measure
Keep WPM (standardized to 5‑char words) if you like—but add metrics that reflect eye‑typing skill and configuration.
- Gaze Selection Latency (GSL): median time from first fixation on a key (or UI element) to confirmed selection. Lower is better. Helps compare dwell settings fairly.
- Error Recovery Time (ERT): average time from error onset to corrected text (include backspaces, undo, or re‑select). Captures practical efficiency, not just speed.
- Prediction/Candidate Utilization Rate (PUR): percent of words entered via top‑bar predictions or gesture candidates. Higher rates often indicate effective scanning and shape paths rather than brute‑force letter entry. Windows’ Eye Control keyboard explicitly supports predictions and shape writing; measure how users capitalize on them. (support.microsoft.com)
- Dwell Efficiency Index (DEI): WPM divided by mean dwell time used for confirmations (normalized). Encourages thoughtful dwell tuning rather than reckless minimums.
- Reselection/Slip Rate (RSR): fraction of selections immediately undone or corrected—an indicator of over‑short dwell times or jitter.
- Gaze Path Efficiency (GPE): ratio of straight‑line path length (between intended keys) to actual gaze path. Especially relevant for shape‑writing.
Report these alongside WPM for transparency.
Proposed benchmark formats
Design a few short, repeatable tests instead of one monolith:
1) Dwell‑Tuning Sprint (90 seconds)
- Fixed word set of short/medium words.
- Ask users to run three passes at different dwell times (e.g., 300 ms, 500 ms, 700 ms) and auto‑plot speed vs. accuracy. Research suggests many users prefer ~600 ms for general object selection—use this to help them find their sweet spot. (sciencedirect.com)
2) Shape‑Writing Burst (2 minutes)
- Prompt only dictionary words, encouraging continuous gaze traces with dwell on first/last letters. Score WPM, PUR, and GPE. Acknowledge that shape modes can deliver multi‑fold speedups over pure dwell. (sciencedirect.com)
3) Error‑Recovery Challenge (60 seconds)
- Include confusable words (“form/from,” “there/their”) and punctuation. Score ERT and RSR to reward graceful correction strategies.
4) Mixed‑Mode Text Passage (3 minutes)
- Natural paragraph with varied lengths and punctuation; user may combine dwell, gesture, and predictions. This simulates real writing.
Accessibility‑first leaderboards that include everyone
Make the competition inclusive and honest:
- Input‑Method Tags: Require a simple declaration that’s stored with each run: “assistive input: eye tracking,” plus details like “dwell typing,” “gaze keyboard,” “shape‑writing,” or “switch + gaze.” This makes cross‑method comparisons transparent without segregating users entirely.
- Filterable Ranks: Let users filter by assistive input type, platform (Windows Eye Control vs. Apple Eye Tracking), and device class. Surface both overall and method‑specific ranks.
- Composite Scores: Alongside WPM, compute a Composite Gaze Score = f(WPM, GSL, ERT, PUR, RSR), with weights you publish. Keep it simple at first (e.g., 40/20/20/10/10) and revise as the community gives feedback.
- Fair Play Signals: Detect the input source where possible (e.g., on‑screen keyboard events vs. hardware keys) and log dwell settings used during the run.
Practical setup tips (Windows and Apple)
- Calibrate often and stabilize the scene. Small posture shifts can degrade accuracy. Microsoft’s Eye Control works with supported eye trackers; ensure proper mounting distance and avoid cable strain that can nudge the sensor. (support.microsoft.com)
- Start conservative on dwell, then optimize. Try 600 ms first, then shave down in 50–100 ms steps while watching your RSR. Evidence points to 500–600 ms as a common comfort zone, but your lighting, fatigue, and tracker quality matter. (tandfonline.com)
- Turn on predictions and try shape‑writing. On Windows, enable shape writing in Eye Control settings; it lets you glance across letters and dwell only at the ends, with alternative candidates at the top. Practice short common words to build “shape memory.” (support.microsoft.com)
- On iPhone/iPad, enable Eye Tracking and Dwell Control under Accessibility. It works system‑wide and can pair with features like AssistiveTouch and Voice Control; experiment with dwell timing until selections feel reliable. (apple.com)
Training drills to build real‑world speed
- Dwell Ladders: Type a 50‑word list three times, at 700/600/500 ms. Note WPM and errors; pick the best tradeoff.
- Candidate Snap: For 2 minutes, accept a prediction whenever it appears in the top 3—even if you’d normally keep typing. This trains fast scanning and candidate trust.
- Shape Loops: Trace 10 common words (“and, the, this, with, you, we, can, have, from, just”) in a loop, focusing on smooth, shallow arcs rather than precise letter hits.
- Recovery Reps: Purposefully make a mistake every fifth word and correct it using your fastest method (backspace vs. candidate swap). Track ERT.
The road ahead
Built‑in Eye Control on Windows and system‑level Eye Tracking on Apple platforms mean assistive input is mainstream. Typing tests should reflect that reality—measuring gaze selection, dwell tuning, prediction savvy, and recovery grace, not just raw WPM. When we do, inclusive leaderboards will finally showcase the full craft of gaze‑to‑text alongside touch and hardware keyboards.
Sources referenced: Windows Eye Control basics and shape‑writing; Apple newsroom and docs on Eye Tracking and Dwell Control; research on dwell‑time preferences and eye‑swipe performance. (support.microsoft.com)
