Illustration of a brain with the accessibility icon in the center

How Deque Furthers Digital Accessibility Through Machine Learning

Working to achieve digital equality is not an easy feat. My colleagues and I at Deque believe that practicing digital accessibility is a team sport. One that requires a combination of automated and manual processes to maximize your testing coverage as efficiently as possible. Automation in this sense is of course referring to automatically detecting accessibility barriers – not automatically fixing them. By working as a team, relying on automation to do the bulk of the work, an organization then has a fighting chance at sustaining ongoing digital accessibility defect prevention.

Automatically detecting 57% of total accessibility issues and 80% of total accessibility issues when supplemented with semi-automated tests is no easy feat either. Our approach to accessibility is to empower the development team and this means pushing the boundaries of what can be tested fully automatically and then providing tools that speed the testing that needs human input. It is our vision that AI and ML should be leveraged to empower the human and this has been the cornerstone of our ability to get to 80% coverage today and to constantly push it higher.

Deque has a unique position in that we have been able to gather years of data about comprehensive accessibility audits for thousands of organizations and applications. This gives us insights into the frequency with which developers make specific types of mistakes which in turn drives our focus about which problems to solve first. We also have a vast number of users of our free and paid tools that provide a wealth of data and we leverage this data to drive machine learning (ML) based and additional heuristic advances.

As with any strategically vital function, we have internal ML capabilities that are leading the charge but we also have to take advantage of the best of breed technologies to make us as efficient as possible in important but not strategic areas. One of our key technology partners is Labelbox, an ML provider we use to help us turn raw data into a11y analysis gold.

Using ML to improve digital accessibility

Before integrating Labelbox into our ML pipeline, our datasets were unorganized and our data collection efforts lagged behind, spending time labeling unnecessary data that didn’t improve our models. Labelbox, firstly provided us with a platform to speed up the basic manual work of labeling data, but has, over time added features like its Model Diagnostics and Catalog features, which streamlines how quickly and effectively we can look for weaknesses in the prediction, inconsistencies in the data and also target specific types of features for model performance improvement. Now, we’re automating, or semi-automating many of our pipeline operations — reducing costs and saving time and resources across the board in model improvement.

With better and faster modeling processes, we’ve also improved our data collection and quality assessment practices. In fact, by evaluating and visualizing model performance and errors within Labelbox’s platform, our team found a portion of our dataset was noisy and hampering the efficacy of our overall model. Model performance increased 5% when we dropped a third of the data points which Labelbox helped us identify as low quality. We also were able to target data collection more precisely, reducing the amount of data that needed to be collected, labeled and tested.

AI and the road ahead for digital accessibility

While we’ve already seen significant improvements in our automated digital accessibility testing thanks to ML integration, this is just the tip of the iceberg. ML — and AI as a whole — show immense potential to transform digital accessibility and any organization’s ability to create inclusive digital spaces.

We often hear that time is a major roadblock to more widespread and proactive digital accessibility. For large global corporations or organizations at immature stages of digital accessibility, manual testing is rarely feasible as a standard approach. In these cases, AI brings hope for a future with more efficient and compliant automated detection. Sure, some level of manual and semi-automated testing is still necessary for comprehensive accessibility coverage, but AI reduces burdens — and is often enough of a support to help companies get started on their own digital accessibility journey.

“We were thrilled to partner with Deque around the mission of making the digital world more accessible for all. This is an important initiative, and Deque’s application of Labelbox as a core part of their data engines is quite mature and impressive. Our goal was to empower them to translate tedious — albeit extremely important — manual processes into automated practices that save time and effort. It was a pleasure to combine our technologies to reinforce Deque’s reputation as an industry leader and enhance abilities around speed and accuracy.” – Manu Sharma, CEO, Labelbox

As part of our commitment to the pursuit of detecting 100% of accessibility issues through automation, we’ll continue to lead efforts to build stronger, more sophisticated AI and ML systems. The data engine built with support from Labelbox will be vital to these efforts.

As the axe DevTools application feeds data to our ML system, we gather better, more comprehensive information that helps us identify opportunities to enhance automated detection moving forward. We’re better equipped than ever to stay at the cutting edge of emerging technology trends and forge new paths toward fully inclusive digital environments.

Onward and upward

Labelbox provided our teams with the resources and knowledge to more effectively integrate ML into automated accessibility testing for faster, more accurate results — and there’s no looking back.

Although AI and ML are still relatively new tools in the world of digital accessibility, huge advances in these technologies within other industries point toward the future of our industry as well. Enhanced, automated accessibility testing with quick, accurate, comprehensive detection will become the standard for digital accessibility conversations and needs. Deque customers can trust their organizations are helping make the web a more inclusive place.

Contact our team to learn more about our automated digital accessibility tools and the fight for more inclusive digital spaces.

Photo of Dylan Barrell

About Dylan Barrell

Dylan is Deque's CTO and leads product development initiatives. He works to help to build a barrier-free web by making it really easy for developers, quality assurance engineers and content writers to create accessible applications and content. Dylan has an MBA from the University of Michigan and a BS from the University of the Witwatersrand.
update selasa

Leave a Reply

Your email address will not be published. Required fields are marked *