Skip to content
The Chat Archive
o3-2025-04-16· OpenAI LMArena Human Preference 140K

whats an interesting way to input text without using a typical keyboard setup. i am not…

  1. User

    whats an interesting way to input text without using a typical keyboard setup. i am not looking for basic ones that everyone might know like stegonography or speech recognition. as a guide, a good answer would be something like dasher.

  2. Assistant

    Here are eight text-entry techniques that are a bit off the beaten path—none of them rely on a standard QWERTY keyboard, speech recognition, or classic steganography. I’ve tried to pick systems that are (a) genuinely different in how you interact, and (b) have at least a research prototype or hobby build you could replicate.

    #Name & Core IdeaHow It WorksWhy It’s InterestingLinks / Papers
    1Quikwriting (Continuous Stylus Paths)The stylus is always in motion. A 3×3 “flower” of zones surrounds a center home cell. Each character is produced by an outward stroke through one or more zones and back to the center, forming a unique path.• Entire words can be written with one continuous squiggle—no lifting.• Much faster than Graffiti once memorized (~60 wpm documented).• Great for pen, touch, or even eye-gaze pointers.Ken Perlin, “Quikwriting: Continuous Stylus-Based Text Entry,” UIST ’98.
    2EdgeWrite (Stabilized Square)The user traces characters along the edges and corners of a square guide instead of open space. The rigid borders correct for hand tremor or joystick slop.• Works on tiny joysticks, trackpads, or wheelchair joysticks.• Very low error rate for users with motor impairments.Jacob O. Wobbrock et al., “EdgeWrite,” ASSETS ’03.
    3P300/SSVEP Brain-Computer SpellerA 6×6 grid of letters flashes in pseudo-random sequences. EEG picks up the P300 or SSVEP response when the desired letter subconsciously “pops out” to the user.• Truly hands-free; works even for locked-in patients.• 5–15 wpm in lab settings.• Open-source toolkits (e.g., OpenBCI, BCI2000) make DIY feasible.Farwell & Donchin (1988); Nicolas-Alonso & Gomez-Gil (2012).
    4Morse via EMG Bite Sensor (“Tongue-Tap”)Place a small EMG pad on the masseter jaw muscle or use a Bluetooth mouthguard. Short bites = dot, long bites = dash. Optional auto-decode with adaptive timing.• Works in noisy or silent environments.• Cheap hardware (<$50).• Your mouth stays mostly closed—discreet.Example build: Hackaday.io project “Jaw-Wink Morse.”
    5Foot-Pedal ChordingSix foot pedals (or capacitive spots in a shoe) are mapped to binary chords (2⁶ – 1 = 63 combos). Common letters get the simplest chords; Huffman coding minimizes steps.• Leaves hands free for other tasks (surgery, mixing audio, VR controllers).• Surprisingly high throughput (~35 wpm) once learned.Engelbart’s original “five-key keyset” inspired this; modern demo: “Chord Pedalboard” (SIGGRAPH ’17).
    6TiltText (Phone-Tilt Huffman)A phone’s accelerometer divides space into 8 tilt buckets. Each tilt selects branches of a Huffman tree until a unique letter is resolved.• One-handed, no on-screen buttons needed.• Works while wearing gloves or underwater (sealed bag).Kristensson, C., “Tilt Text: Accelerometer-based Text Entry,” CHI ’09.
    7AirWrite Gloves (In-Air IMU Glyphs)A finger-worn 9-DOF IMU captures 3-D motion. DTW or neural net recognizers classify the air-scribbled glyphs.• Nothing to hold—great for VR/AR spaces.• Recognizes full alphabet at ~30 wpm with practice.• Can add haptics for eyes-free use.O. Hilliges et al., “Digits: Full-Hand 3-D Motion Capture,” UIST ’12.
    8Smart-Laser “SweepSelect”A low-power laser pointer sweeps across a large projected radial menu. User briefly blocks the beam (with a finger, eyelid, or object) when the desired wedge highlights; successive sweeps zoom in hierarchically until a letter is chosen.• Works at huge scales (the side of a building) or tiny retinal-implant scales.• Can be used from several meters away—useful for sterile or hazardous environments.Conceptually similar to Gaze-based MAGNUS and dwell-select hierarchies; see Nowacka et al., “Sweep Selection,” CHI ’13.

    ================================================ Deep-Dive Pick: Quikwriting

    Small ASCII diagram (top view of zone layout):

      NW   N   NE
       \   |   /
    W —  (C)  — E
       /   |   \
      SW   S   SE
    

    Example paths
    • “hello” = center→E→SE→C→NE→C (one stroke)
    • “world” = ... well, the pattern starts to look like cursive scribbles that you learn after a couple of hours.

    Open-source implementations exist for iOS, Android, and even eye-tracker cursors—so you can try it without building hardware from scratch.

    ================================================ Choosing & Prototyping

    1. List your constraints: hands-busy? eyes-busy? mobility limitations?
    2. Pick the sensing modality that best survives those constraints (IMU, EMG, EEG, optical, foot pedals, etc.).
    3. Map characters to actions using information-theoretic coding (Huffman) to keep common letters quick.
    4. Provide progressive feedback (strokes on screen, tones for dots/dashes, dwell timers) to aid learning.
    5. Add error-tolerant decoding (e.g., phrase-level language models) to achieve real-world typing rates.

    Happy hacking!

Responses were produced in different, uncontrolled contexts (system prompts, settings, prior turns). This is not a benchmark.