Hacker News Reader

Comment by _-___________-_ | Hacker News Reader

Could you expand on what you mean? Is there some trend of writing English in Cyrillic script that I've completely missed?

View on HN

Not sure if HN approves (it removes emojies at least, this went fine it seems), but there are glyphs that look like the alphabet/normal letters, but are not really the same characters, like these: 𝒕𝒉𝒊𝒔, 𝖙𝖍𝖎𝖘, ⓣⓗⓘⓢ, 𝕥𝕙𝕚𝕤, ㄒ卄丨丂, ｲんﾉ丂, ꙅiʜƚ, 🅃🄷🄸🅂, sıɥʇ, 𝓉𝒽𝒾𝓈

My guess is screen readers can have trouble telling what it really says.

My guess is screen readers can have trouble telling what it really says.

It's actually the humans who don't know what it really says. We see 'some weird looking glyphs' and mentally translate them to letters we know. This quirk of the human brain is how we can read a random jumble of letters from different alphabets as English words.

Screenreaders on the other hand know exactly what it says and they read it out precisely how it's written.

I agree, and I can only imagine how aggravating it must be to be a user with a screen reader having to (mentally/programmatically) parse the very "clever" strike-through-simulating special-purpose Unicode text some user decided would suffice for "real" strike-through text, just because HN is, uh, persnickety, with its allowed characters...

I've seen it many times here at least. Just add an addendum!

What it "says" is whatever the person that wrote it intended to convey.

That's obviously not true. Communication is a two-way relationship. You literally can't be an effective communicator on your own. If you write something in a way that can't be understood (with a little effort) by the reader then you failed to communicate well. If the reader is using a screenreader and you wrote in silly characters then you failed to communicate effectively, and you didn't convey anything at all regardless of what you might have meant.

Someone posted a video on Twitter about this. What happens is that it reads out the meaning (similar to the unicode names) of the symbols.

E.g. 𝖙𝖍𝖎𝖘 get read as "Mathematical bold fraktur small t, mathematical bold fraktur small h, mathematical bold fraktur small I, mathematical bold fraktur small s"

I wonder if a screen reader could be smart enough to use OCR when it comes across an abrupt sequence of unusual character combinations like this.

I don't think you'd need to use OCR for that. Just* use transliteration tables to map those characters to the base characters.

(* I usually hate when 'just' is used this way, I'm sure it's a minefield in some respects. But still more robust than trying to optically interperet what amounts to captchas in some situations :)

There are quite a few libraries for converting arbitrary Unicode to some subset while preserving as much meaning as possible (presumably by embedding such tables), and they generally work pretty well on this kind of text.

Yeah, I've used

    iconv('UTF-8', 'ASCII//TRANSLIT', $text)

in my work to make sure I could accept arbitrary Unicode from users, and also send that same info as ASCII to legacy systems that mangled UTF. It worked really well.

When I wrote the parent comment, I tried it real quick with a REPL on my own machine, but apparently I don't have the right combination of environment variables or locales or God knows what. Didn't feel like chasing that rabbit into his hole.

IMHO, the onus is on the screen readers to improve.

It should aboslutely be "smart" enough to simply say "fraktur this" -- to recognize a string of letters of the same style and specify that once, and to interpret the text as a word.

It is the job of screen readers to convert the visual meaning to an auditory meaning as clearly and concisely as possible. If it isn't doing it concisely, then the screenreader is failing and needs some better engineering.

This can be seen quite a bit on Twitter, both in usernames and the tweets themselves. Personally, since I think human behaviour cannot be completely fixed but can be nudged in the correct direction, I'd like to see services implement something like alt text for text, especially usernames. This way people can use different characters to be as creative as they want, but provide an accessible version for screen readers. Something like this example:

    <span aria-label="like this example"><span aria-hidden="true">𝕝𝕚𝕜𝕖 𝕥𝕙𝕚𝕤 𝕖𝕩𝕒𝕞𝕡𝕝𝕖</span></span>

The alternative would be restricting the use of these characters, but this would kill many valid use cases and you can't just forbid non-latin characters on an international service.

I've written a post[1] about this and thought I'd implement it in one of my services as an example. I haven't seen any example of it live yet.

[1] https://blog.nytsoi.net/2019/12/12/alt-text-for-text

I like the idea, but the de-Unicodifier would have to use the current user's locale - de-Unicodifying certain characters (e.g. Cyrillic that look like English) would have unintended consequences if you just assumed they're meant to be English.

That wouldn't work either. What if I was born with a foreign name, but was raised in a totally different country?

My idea was that this would be something provided manually by the users, just like you can specify alt text for images on Twitter right now. I know many wouldn't bother. But at least some would, and it would make using an upside down nickname possible for me again. :)

The hosting service can solve this million times as effectively as the users can.

It should be easy to detect. You usually don't mix different alphabets in a single word.

I imagine most of these "stylised" effects are generated by services like https://yaytext.com

Getting those services to include such markup in the copy/paste version would probably solve this problem for most occurences. Note that (a) the original text is known to the service and (b) copy/paste content doesn't have to match what is displayed (e.g. "pastejacking")

I don't think many services allow pasting HTML to any fields.

Here is a video demonstrating how a screen reader reads these pseudo-fonts:

https://twitter.com/jypnati0n_/status/1303012163766743041

Seems like the screen reader could have a "say what they mean" mode to pronounce "𝒕𝒉𝒊𝒔" as "this". I assume the legitimate use of Unicode mathematical symbols like "𝒕𝒉𝒊𝒔" is much less than the pseudo-font usage.

I once used Unicode to set off some fixed width text inline with my comment. (Or, I think it was italicized? Don't remember exactly).

I was doing it to work around the limited formatting available here, but I noticed a couple hours later that my comment was transliterated into ASCII. Given the time delay, I assume it was done manually. (Dang, was that you?)

It would seem to me that this would be a simple problem to solve for screenreaders as they are just 1:1 mappings, and in many cases even consecutive blocks of Unicode characters.

Understandably though, screenreader software probably isn't as fast moving as other industries. If there was an open source screen reader I think I could easily make a pull request to solve this particular problem, if it is a problem.

It is a problem, and:

https://github.com/nvaccess/nvda

They were hiring on a "Who's Hiring" thread recently as well

> the alphabet/normal letters,

ASCII English "Latin" letters. Not "the" alphabet or "normal" letters.

Not Cyrillic in particular, all scripts(I just happen to have Cyrillic layout hence the reason I used it for my example), but if you go around blogs, twitter or other websites, there are tons and tons of people mixing those up to look fancy or whatever it is they are trying to do. It's a nightmare for those of us who know several alphabets and an even bigger nightmare for those who need to use screen readers.

Ah, OK. I haven't seen this much outside of screen names in some games.

If people are going to do strange things like this en masse, perhaps the screen readers could implement some normalisation rules.

Something of the sorts could be implemented but this would be an attempt to make something idiot-proof. And from my experience, nothing can ever be idiot-proof. To quote Mark Twain: "they will drag you down to their level and beat you with experience".

It can be found on Twitter when people want to tweet with fancy formatting.

>> this in particular

I assume he's talking about using unicode characters that visually similar to english letters, but are not actually english letters.

I guess your username has the same properties.

They're chosen because of some visual elements, and are hard, if not impossible to pronounce for text-to-speech.

Reply on news.ycombinator.com