All posts Signals

Intent: knowing when the pen is really writing

MKMalli KosuriJun 20265 min read
In short

Much of a pen's motion is travel, not writing, lifting between letters, gliding to the next word. Intent detection decides, moment to moment, whether the pen is down and inking or up and moving, using pressure and motion cues together with a learned classifier.

Getting it right prevents phantom lines and is essential for marks that sit apart from the main stroke: the dot on an i, and the stacked vowel signs and dots that hover above and below the aksharas of Indic scripts.

The short version. The full post has the detail and the why.

In this post

  1. Half the motion isn't writing
  2. Detecting pen-down vs pen-up
  3. Why a classifier, not just a threshold
  4. Intent and stacked Indic signs

Between every two strokes you write, the pen does something invisible: it lifts, travels, and comes back down. The page never records those journeys, but the sensor feels all of them. Telling writing from travelling is the job of intent.

Half the motion isn't writing

Across a normal page, a large share of the pen's movement is not ink at all, it's the hops between letters, the slide to the next word, the pause while you think. An inertial pen records this travel indiscriminately. If the system treats every motion as a stroke, the reconstructed page fills with phantom lines joining marks that were never connected. Intent detection is the filter that keeps only the real ink.

Detecting pen-down vs pen-up

The strongest cue is pressure: when the tip is on the paper, the force sensor reads a clear, sustained load; when it lifts, that load falls away. But pressure alone is fragile, a light writer barely presses, and a resting hand can fool a simple threshold. So intent combines several signals: tip force, the character of the motion (writing strokes are short and constrained; travel moves are longer and freer), and brief context from just before and after the moment in question.

, ink (pen down) - - travel (pen up)
Intent detection labels each instant as ink or travel. The dot, written as a separate pen-down event, only survives as its own mark if the lifts around it are correctly recognised.

Why a classifier, not just a threshold

A fixed pressure cutoff is brittle: it fails light writers, mistakes a resting tip for writing, and flickers on the boundary. So intent is usually a small learned classifier that weighs pressure, motion features and timing together, and tends to respect continuity (writing doesn't flicker on and off every millisecond). This is the same touch-versus-hover problem that motion reconstruction depends on, solved well, it gives crisp, correctly separated strokes; solved poorly, it smears the whole page.

Intent and stacked Indic signs

Intent matters most where meaningful marks sit apart from the main stroke. In English that's the dot on an i or j. In Indic scripts it's pervasive and load-bearing: vowel signs (matras), dots (anusvara, the nukta) and other diacritics attach above, below or beside the base akshara, each often written as a separate pen-down after a lift. Telugu and Kannada stack signs above and below; Malayalam and Tamil add marks around the base; Devanagari (Hindi) places matras above and below the head-line.

If intent detection misreads the lift before one of these marks, the diacritic either fuses into the wrong stroke or vanishes, and the akshara can change meaning. Reading intent accurately is what lets Vahini keep each of these small but crucial marks intact, and score how cleanly they were placed.

Key takeaways

  • A large fraction of pen motion is travel, not writing.
  • Intent detection labels each instant as pen-down (ink) or pen-up (travel).
  • It blends pressure, motion character and timing in a learned classifier, not a brittle threshold.
  • Accurate intent keeps detached marks, dots, matras and diacritics, intact in Indic scripts, where they carry meaning.

Read next