The Impact of Word, Multiple Word, and Sentence Input on Virtual Keyboard Decoding Performance
summary
A full QWERTY keyboard running on a Sony SmartWatch 3 uses a probabilistic decoder with Gaussian touch models and n‑gram language models; recognition is triggered by a right-swipe recognition gesture. Allowing sentence‑at‑a‑time input raises measured entry rates from 26→32 wpm while keeping character error rates under 4% on‑device and yielding 2.0% offline error with more compute. On‑watch accuracy suffers from a constrained search beam and compact LM; if the decoder knows word boundaries instead of inferring spaces, longer input gives clear accuracy gains.
AI-written summary · about
chapters (10)
- 0:00Introduction and motivation
- 1:30System and input method
- 4:10First experiment setup
- 5:56Results on entry rate and error
- 8:44User behavior and preferences
- 9:50Second experiment on composition
- 12:46Offline analysis and error trends
- 14:31Computational resource impact
- 15:39Key takeaways and design implications
- 17:04Q&A session
related talks
21mDualKey: Miniature Screen Text Entry via Finger Identification
12mEvaluation of Korean Text Entry Methods for Smartwatches
20mTipText: Eyes-Free Text Entry on a Fingertip Keyboard
06mWrite, Attend and Spell: Streaming End-to-end Free-style Handwriting Recognition Using Smartwatches
13mUniKey: Enabling Surface-Based Typing with Commodity Smartwatches via Cross-Modal Learning
18mTyping on a Smartwatch for Smart Glasses
18mImproving Virtual Keyboards When All Finger Positions Are Known
23mIJQwerty: What Difference Does One Key Change Make? Gesture Typing Keyboard Optimization Bounded ...