Markov Keyboard: keyboard layout that changes by Markov frequency (2019)
A prototype “Markov Keyboard” that dynamically rearranges keys based on next-letter probabilities has sparked broader examination of how predictive models could reshape text input. Commenters weigh potential speed and accessibility gains—especially for assistive technologies and on-screen keyboards—against the heavy cost to muscle memory, usability, and error rates when layouts shift too often. The thread branches into related ideas such as randomized keypads for security, hardware with reconfigurable key legends, alternative keyboard layouts, and whether modern language models could drive more adaptive but still learnable designs.
Dynamic / Markov-Based Layouts
- Many find the idea of a layout that reorders keys by Markov (or similar) prediction intellectually appealing and “rabbit hole”–worthy.
- Some compare it to autocomplete: if probabilities reset at each word and use a static table, users might learn the pattern, with most letters reachable from the home row.
- Others argue that constantly moving keys destroys muscle memory, adds visual search time, and is impractical for everyday touch typing.
Prediction, Personalization & Assistive Tech
- Several comments suggest using more advanced or personalized language models (e.g., PPM, recurrent models) trained on the user’s own text to improve predictions.
- There is interest in using next-letter prediction for scanning or switch-based access, where efficiency gains could be large. Ideas include:
- Scanning letters in order of likelihood rather than fixed order.
- Showing word predictions directly on keys.
- Highlighting or lighting only likely next letters or classes (e.g., vowels).
Security-Oriented Randomization
- Randomized on‑screen or physical keypads are already used for PIN entry (ATMs, doors, online banking) to defeat “clean key,” shoulder-surfing, and thermal attacks.
- Discussion includes combinatorics of inferring codes from which keys are dirty/clean, and notes that random layouts cut the search space but don’t eliminate it.
- A possible “timing attack” is noted: longer seek time to displaced digits could leak information. Proposed mitigation: enforce fixed pacing between layout reveal and input.
Ergonomics, Muscle Memory & Alternative Layouts
- Multiple commenters discuss long, difficult transitions to static alternative layouts (Dvorak, Colemak, Neo, custom designs) and RSI motivations.
- Opinions differ: some say layout switching is extremely hard and productivity-killing; others report relatively quick adaptation and easy switching back to QWERTY.
- Dynamic layouts are criticized as even worse for muscle memory than static alternatives.
Beyond Traditional Keyboards & Hardware Ideas
- Suggestions include analog keypress levels (e.g., half-press vs full-press), multi-function thumb clusters, chorded devices, handwriting or shorthand on touchpads, and systems like Dasher or chorded input via musical instruments.
- Dynamic key legends (LCD/OLED/e‑ink per key or screen-under-keys) are seen as enabling technology for Markov keyboards.
- Some explore DIY keycap production (injection molding, 3D printing) to support custom layouts.