Skip to main content


Question for #screenreader users: do text emotes like kaomoji generally cause your tools to read out noise or annoying nonsense, or does it just not pronounce it? I am wondering whether it's okay to use them or whether I should go back to good old emoji (that, to my knowledge, get properly read out).

Like this one:
˚‧º·(˚ ˃̣̣̥᷄⌓˂̣̣̥᷅ )‧º·˚

#accessibility #totallyblind

in reply to Lianna

Screen Readers will try to read the punctuation, so it won't make sense. Just trying now with NVDA, you can have it set to read more or less punctuation, so with your example, it will read anywhere from:

"Hyphenation point, degree, hyphenation point, degree"
to
"Hyphenation point, degree, middle dot, left paren, right paren, hyphenation point, degree, middle dot"

This is how most screen reader users will experience Kaomoji.

1/2

in reply to NV Access

2/2
You CAN set how NVDA reads words or character strings. I just made a dictionary entry to read ˚‧º·(˚ ˃̣̣̥᷄⌓˂̣̣̥᷅ )‧º·˚ as "Cute Crying Kaomoji". It works fine, but most users won't have set that. I did reach out to Microsoft about this recently as the Kaomoji panel in the Windows emoji panel is also inaccessible. They are aware of the issue, but don't have a solution. Unicode defines a standard list of emoji descriptions: unicode.org/emoji/charts/full-… if they add kaomojis we could utilise it.
in reply to NV Access

@NVAccess That's very enlightening, thank you! I suppose it's not really a solvable problem with dictionaries, because as opposed to standard smileys like colon and uppercase D - this one :D - Kaomoji are very, very varied and can be personalized.

I am wondering whether some traditional machine learning classifier could be good at detecting what is and what isn't a smiley.

in reply to Lianna

Since Microsoft do definte a whole list of Kaomoji, if Unicode or someone even just defined descriptors for those, we could perhaps incorporate it as a starting point. Although as you say, as soon as you change one slightly, it will break that. So it may be something solveable with machine learning, perhaps as an NVDA add-on (there are several for image description and other things, so it would certainly fit) - definitely interesting!