Disambiguating pen gestures from written content
Aliases: gesture–ink conflict · ink versus command · scratch-out
What it is
The same pen has to write and draw, and also issue direction marks, scribble-erase, and lasso commands. The system must decide, around lift, whether this is ink or a gesture. The two mistakes are asymmetric: treating a character as delete removes content; treating delete as a character leaves a stray mark. Disambiguation is a mode problem, not a prettier command set.
Why it happens
Usual cues: an explicit mode (barrel button, ink/gesture tool), scope (a mark on empty canvas is a command, a mark in a cell is writing), a shape vocabulary (Z, pigtail, back-and-forth scratch rarely meant as letters), time (a short ballistic stroke versus writing’s pauses), and object hit (a stroke from a selected object is a verb). No single cue is enough. Shape vocabularies collide with real letters and proof marks. A button mode is reliable and costs a mode switch. Timeouts wait to see whether the stroke “looks like a gesture”; wait too long and ink is late, which wrecks writing’s following feel. A stabler stack is: default to ink; intercept only high-confidence shapes in a gesture scope; when intercepting, give undoable command feedback rather than a silent delete. Scribble-erase and written strikethrough mix especially easily—writers already stroke out wrong glyphs. Scope plus “must cover existing ink and reverse enough times” can separate strikethrough from ordinary strokes.
Studying it
Have people alternate writing, drawing, and delete/lasso/direction commands on one canvas, including glyphs that resemble gestures (Z, dashes, circles). Compare shape-only classification, a button mode, and scope rules.
Independent variables: strategy, gesture vocabulary size, whether a barrel button is required, visibility of undo. Dependent measures: ink deleted as a command, commands left as ink, mode-switch count, writing delay from waiting to classify.
The corpus must include real strikethrough, not only neat copying.
Where it stops holding
A drawing-only tool can default to ink and enable gestures only under a modifier. A checkbox grid can default to gestures. Mixed notes are the hard case. Scripts whose strokes already contain hooks and loops cannot inherit a Latin gesture vocabulary unchanged. If a recognizer consumes every stroke as a character in the background, the gesture channel is preempted.
Applying it
- Default to ink; keep commands to a few shapes hard to confuse with letters, or to off-object / button mode.
- When a gesture fires, show command feedback (a flash, the object gone) and make it immediately undoable; do not silently edit the document.
- Do not disambiguate with a long timeout; it becomes inking latency. Prefer an explicit button.
- How to check: write sentences containing Z, circles, and dashes, then scribble-erase a wrong character. Count false deletes and failed deletes. Contrast a button mode: false deletes should fall; log the extra mode switches and whether people prefer to pay that cost.