On Suggesting Phrases vs. Predicting Words for Mobile Text Composition
summary
Extends a mobile keyboard to show phrase previews under the three next-word slots and accept them with repeated taps; built a prototype that pulls phrases from Yelp and ran a within-subjects restaurant-review study. Phrase visibility yields ~15% more words coming from suggestions and 24% of successive same-slot taps occur within 300 ms (0% in the word-only condition), while overall typing speed and error rate stay unchanged. Users report phrases spark ideas but can feel trite or overly prescriptive, with examples like the best thing about this place and lunch menu is very reasonably priced.
AI-written summary · about
chapters (10)
- 0:00Intro
- 0:19Predicting Words
- 2:58Openended Composition Task
- 3:40In Lab Experiment
- 5:36Results
- 6:08Time between successive acceptances
- 7:03Text entry metrics
- 8:17More than efficiency
- 9:04Audience questions
- 10:26Inspiration
related talks
07mScreen2Words: Automatic Mobile UI Summarization with Multimodal Learning
05mExploring Smartphone Keyboard Interactions for Experience Sampling Method driven Probe Generation
21mThe Impact of Word, Multiple Word, and Sentence Input on Virtual Keyboard Decoding Performance
37mA self learning word prediction module for handling multilingual input
03mInterview with Ben Medlock
22mImage Retrieval using Short Texts
19mPhrazIt : Tool for automatic text summarization
39mWould You Let Your Phone Explain the World to You?