Accelerating math accessibility with using AI


A 12 months in the past, NWEA, now a part of HMH, shared their modern method to make math extra accessible for college kids. The goal was to determine the most important challenges and gaps in arithmetic for college kids who use display screen readers and refreshable braille gadgets, as a result of classroom supplies should not all the time tailored to codecs comparable to braille or massive print, and supplies should not all the time appropriate for a screen-reader navigation, voice enter, or a mix of those designs. NWEA developed prototypes that enabled display screen readers to work together with equations in a extra intuitive means, based mostly on a technique known as course of pushed math (PDM). 

NWEA continued to innovate and construct on their earlier analysis to create other ways of presenting complicated math, particularly for math taught in grades six to 9. In addition they labored on other ways of outputting math that included screen-reader performance and refreshable braille gadgets in each UEB (Unified English Braille) and Nemeth codecs. Furthermore, they developed a prototype for a voice-activated chatbot.  

To account for the accessibility of math equations, they used two markup languages, HTML and ARIA, to separate equations into elements or areas. Every area, in addition to the entire equation, had a hidden label {that a} display screen reader would say to customers as they explored the equation or expression. When college students moved from one area to a different, they might hear a phrase that described the form of math in that area (for instance, “time period” or “fixed”). College students may then resolve to enter the area and listen to the precise math, or they may simply skip to the following area.

The usage of generative AI  

By utilizing AI, particularly GPT-4, the group was capable of enhance each the standard of the mathematics in addition to the time required to transform the equations to HTML, and to make use of code era to write down the code for the primary prototype. The mannequin solely wanted just a few examples to learn to change the preliminary check set of equations from MathML to the HTML construction that was probably the most accessible. From there, the mannequin required context to make sure that responses have been formatted in one of the best ways for the app.  

Demo of utilizing the equations with a display screen reader:


Leave a comment