They are an accessibility nightmare. I read a comment here recently from a blind person that had to give up on math education entirely due to all of the symbols not being supported by screen readers. It was pretty sad.
That's on screen readers developers. And I don't know when that comment you seem to remember dates back to, but I have had several blind students in my math lectures over the past decade. They seemed did fine.
I avoid using oddball Unicode characters for this reason,[1] but I also feel like at this point the can of worms has been unleashed, and the screen reader authors are going to have to adapt to a more hostile environment.
It seems like one approach might be something like this, assuming none of them are doing it already. If this is TLDR, the key feature relating to the topic at hand is the point about handling mathematical symbols, so skip everything except that one.
* Pre-parse the entire page and count the number of characters that would be read using a multi-word description. While doing this, create sub-counts of characters that fall into each Unicode script (Latin, Cyrillic, mathematical notation, etc.).
* If the number of characters that would be read using a multi-word description exceeds a user-specified threshold, prompt the user for how to proceed.
* First, tell the user how many different Unicode scripts (and blocks if necessary[2]) are contained on the page. For each set, prompt using the following logic if the user hasn't already set a default:
** If a script/block/whatever consists of abstract symbols, like the mathematical set, offer to read it using the closest approximation in the character set it's derived from. e.g. for 𝕏, read it as "X" instead of "mathematical double-struck capital X". Read both 𝜸 ℾ as "gamma" instead of their lengthy "mathematical..." descriptions.
** If a script/block/whatever represents a spoken language, offer to do one of the following:
** Read it as if spoken by someone fluent in the language.
** Precede it with "the following [n] [script name] characters" and read each one individually, without any leading per-letter indicators, e.g. for "ΑΒΓΔ", "the following four Greek characters: alpha, beta, gamma, delta". Optionally include per-letter capitalization indication, but only if the user has previous enabled that option.
** Replace it with a count and the script name, e.g. "five hundred characters of Chinese script". Provide a way for the user to interrupt the narration and expand that section instead of omitting it.
* Allow the user to store all of the prompt answers as defaults and not prompt them again unless the user resets the option. e.g. "always read mathematical symbols as the closest Latin letter approximation".
* Maybe remind the user that it is an option in edge cases like "this entire page would be read as 'a series of 53,198 symbols you've suppressed reading'".
[1] One of my first full-time jobs, decades ago, was doing UI development with a focus on accessibility.
[2] Pretty unbelievable that the Unicode consortium has taken the position that an apparently unlimited number of stylistic variations on Latin letters should get their own Unicode renditions, but that basically every other culture's character sets gets merged into one style of character even if there are extensive historical reasons to have different renderings, and also has the time to keep adding corner-case emojis, but can't be bothered to provide a single way to easily differentiate "Cyrillic" from "Latin" as well as differentiate "mathematical notation" from "Latin".
