Crowd Caption Correction (CCC)
Rebecca Perkins Harrington, Gregg C. Vanderheiden · 2013 · Proceedings of the 15th International ACM SIGACCESS Conference on Computers and Accessibility (Assets '13) · doi:10.1145/2513383.2513413
Summary
This short paper presents Crowd Caption Correction (CCC), a feature that allows meeting participants or authorized third parties to correct errors in real-time captions during telecollaboration sessions. Captions are critical for deaf and hard of hearing people to participate in webinars, videoconferences, and meetings, but they frequently contain errors — misinterpretations, missing words, misspellings of technical terms and proper names — caused by poor audio quality, multiple speakers, fast talking, or the captioner being unfamiliar with domain-specific terminology. Caption providers often lack the time or background knowledge to correct these errors during live sessions. CCC addresses this by enabling anyone authorized (from a few designated participants to all attendees) to click on erroneous caption text, edit it in place, and submit the correction. A collision-prevention mechanism handles simultaneous edits. Corrected text is displayed to all participants highlighted in the corrector's chosen font characteristics (font, size, color, highlight) to indicate it has been modified. The system also automatically generates a corrected transcript from the ongoing caption stream — a significant practical benefit since professional captioning services typically display only 1-2 lines at a time and often charge additional fees for post-session transcripts. CCC is part of the Open Access Tool Tray System (OATTS), a web-based suite of open-source widgets developed at the University of Wisconsin-Madison Trace Center under the Telecommunications RERC, designed to increase access to telecollaboration for people with various disabilities.
Key findings
The paper demonstrates the CCC workflow through a step-by-step example: an error where "known as" was captioned as "notice" (semantically opposite, making the caption confusing or misleading). The authorized participant clicks the incorrect word, edits it, presses Enter, and the correction appears highlighted and underlined for all viewers. The system provides dual benefits: immediate improvement to caption accuracy for current viewers, and a free, corrected transcript available during and after the meeting. The OATTS framework is designed to be modular and web-based — widgets run from any computer via a server with no software installation required, and users can select which accessibility widgets to include in their personal tool tray. The CCC widget can work independently or alongside other OATTS widgets. The paper identifies that caption errors range from confusing to embarrassing to presenting information directly opposite of what was said, and that deaf participants cannot always determine whether a confusing caption is an error or whether they simply don't understand the content — creating a particularly frustrating experience.
Relevance
CCC addresses a practical and persistent problem in caption accessibility: even when captions are provided, their quality significantly impacts comprehension and participation for deaf and hard of hearing users. The crowdsourced correction approach is notable because it leverages the knowledge of people already in the meeting — the very people most likely to know the correct technical terms, proper names, and context. This is complementary to existing crowd captioning systems like Scribe, which use multiple remote captioners whose outputs are compared, but where captioners may still lack domain knowledge. For accessibility practitioners, the paper highlights that providing captions is necessary but not sufficient — caption quality matters, and mechanisms for improving quality in real-time are essential. The automatic transcript generation addresses the often-overlooked need for post-meeting reference materials. The OATTS modular widget approach anticipates the modern trend toward accessibility overlays and browser-based accommodation tools for videoconferencing, which became especially relevant during the shift to remote work.
Tags: captioning · deaf and hard of hearing · crowdsourcing · telecollaboration · real-time captioning · universal design · open source · transcription · videoconferencing