Type, Then Correct: Intelligent Text Correction Techniques for Mobile Text Entry Using Neural ...
summary
Three mobile correction interactions are built and measured: Drag-n-Drop, Drag-n-Throw, and Magic Key. A recurrent neural network runs on a server (trained on ~5 million injected-error phrases from CoNLL Yelp Amazon and tested on ~3k Wikipedia/CoNLL cases) to map a typed correction to an error position. Implemented in a modified Android keyboard on a Pixel 2 with 30 participants, results show faster corrections than cursor/backspace and accuracies of ~87% for Drag-n-Throw and ~97% for Magic Key.
AI-written summary · about
chapters (7)
- 0:00Start
- 0:53CURRENT PRACTICE
- 1:56THE CONCEPT
- 2:52THREE INTERACTIONS
- 4:46Magic Key
- 6:10THE CORRECTION ALGORITHM
- 10:51USER STUDY
related talks
08mEfficient Typing on a Visually Occluded Physical Keyboard
05mKeeping it Simple
20mAn Approach to Generate Text-Based IDEs for Syntax Correction based on Syntax Specification
09mCursorless: keyboards and mice are sooo last year!!
21mMultitasking with Play Write, a Mobile Microproductivity Writing Tool
27mDefault Disambiguation in Online Parsers
16mKeynote
38mCreate the next generation of user experiences with text generation