Article

Preliminary Research on AI-Generated Caption Accuracy Rate by Platforms and Variables

As a result of the global Coronavirus pandemic, there has been a rapid increase in online courses and digital educational materials, specifically video content. It is a legal requirement in the United States that these educational materials be accessible to those with disabilities. In the case of educational videos, this requires that clean and accurate captions are provided, as well as text or audio descriptions of all visual images within the video, in a format that is accessible to learners. Many educators lack an understanding of the details of the legal requirements and the skills, financial resources, staff, and technical support needed to fulfill them. As a result, many rely on Artificial Intelligence (AI)-generated captions and descriptions for their educational videos, if they provide them at all. Here, we report the results of a preliminary analysis of the accuracy rates of several popular platforms used to create AI-generated captions. Based on our results, these AI-generated captions have, on average, an accuracy rate of 89.8%, which does not meet the current legal requirements and industry standards for digital accessibility. We show the importance of manually copyediting captions and share insights from our data analysis which will assist educators to make their video content more accessible.

Items in ScholarWorks are protected by copyright, with all rights reserved, unless otherwise indicated.