Given that we can't expect everyone to use aria attributes, shouldn't screen readers just have a list of all widely used smileys with descriptive names (if they don't have it already)?
Compiled all predefined textual emoticons offered by Gboard app in Android [1] and let the Narrate speech synthesis in Reader view (F9) read them aloud (uses presumably the same synthesis as example) - quite predictably most of them weren't heard at all, but those simple ones that were have surprised me.
So most probably there is no problem with that particular Unicode bear in screen readers after all.
[0] https://mdn.github.io/web-speech-api/speak-easy-synthesis/ [1] https://gist.githubusercontent.com/myfonj/f6b0ed1c783d16a79d...