WearWrite: Crowd-Assisted Writing from Smartwatches
Michael Nebeling, Alexandra To, Anhong Guo, Adrian A. de Freitas, Jaime Teevan, Steven P. Dow, Jeffrey P. Bigham · 2016 · Proceedings of the 2016 CHI Conference on Human Factors in Computing Systems (CHI 2016) · doi:10.1145/2858036.2858169
Summary
WearWrite is a system that enables users to write documents from their smartwatches by leveraging crowd workers to translate ideas into text. The system addresses the fundamental limitation that smartwatches have severely constrained input/output — touch-based text input is extremely slow, speech-to-text is error-prone for long passages, and the small screen makes reviewing complex content difficult. WearWrite connects a smartwatch user (acting as the domain expert and project director) with a crowd of non-expert writers recruited from Amazon Mechanical Turk. The watch interface supports three main interactions: creating tasks via speech-to-text or audio recording, reviewing and accepting/rejecting major edits via document thumbnails, and answering worker questions. The system uses a mixed-initiative approach where minor edits (inserts under 60 characters, replace/deletes under 30 characters, and format changes) are automatically accepted, while major edits require watch user approval. Crowd workers access a desktop web interface wrapping a shared Google Doc, with a dynamic task queue that prioritizes user-specific tasks over generic writing tasks (expanding bullet points, improving sentences, editing paragraphs, and general proofreading). Workers can ask questions visible to all workers on the same task, creating an asynchronous Q&A channel between the watch user and the crowd.
Key findings
In a week-long deployment with 7 smartwatch users and 205 crowd workers (142 unique, averaging 29 workers per participant), all projects progressed from initial outlines to prose first drafts. Documents grew an average of 258% in word count (ranging from 130% to 745%). Watch users created an average of 15 tasks per project, with workers spending 7.4 hours and completing 83 tasks per project at an average cost of $27.26. The mixed-initiative edit acceptance approach worked well — users only needed to review 30% of all suggested edits on the watch, with a 96% overall acceptance rate. Five of seven participants used WearWrite during spare moments throughout their day (riding buses, waiting in line, at bars, during road trips), validating the on-the-go use case. Four of seven participants said the crowd did not write what they would have written themselves, but all seven planned to use the crowd's output as a starting point for their final drafts, with three intending to use significant portions without edits. Workers rated the tasks as clear and interesting and many expressed desire to continue with similar projects, though coordination issues arose when multiple workers edited simultaneously.
Relevance
While not explicitly framed as accessibility research, WearWrite has significant implications for people with disabilities who face input constraints similar to smartwatch limitations. Users with motor disabilities, those who cannot type effectively, or people who rely on voice-only interfaces face analogous challenges — they can articulate ideas but struggle with the physical mechanics of producing long-form text. The crowd-assisted writing model demonstrates that meaningful content creation is possible from severely constrained interfaces when human intelligence fills the gap. The system's mixed-initiative approach (automatically handling minor edits, only escalating major changes) is a design pattern directly applicable to assistive writing tools. The research also connects to Bigham's broader theme of using crowdsourcing to overcome interface limitations for accessibility — the same principle underlying Scribe (captioning from limited input), VizWiz (visual understanding from a camera), and WearMail (email search from a watch). The finding that users can effectively manage a crowd writing process through brief, intermittent interactions suggests potential for assistive technologies that support people who can only interact in short bursts due to fatigue, pain, or attention constraints.
Tags: crowdsourcing · wearable technology · human computation · collaborative writing · mobile accessibility · conversational interface