Support mathematical alphanumeric symbols

[sukiletxe]

Is your feature request related to a problem? Please describe.

Currently, mathematical alphanumeric symbols are gaining traction, both in social media (in nicknames for example), and when viewing LaTeX documents converted to docx (I have encountered this twice).

Describe the solution you'd like

It would be useful to have the Mathematical alphanumeric symbols included in NVDA.

Describe alternatives you've considered

This could go to eSpeak NG too, like they do with cyrillic letters. I'm currently considdering both options as equally valid, but I'm biased towards eSpeak NG mainly because that is the only synth I use, so OneCore users wouldn't benefit from this.

Additional context

Unfortunately I do not know how many of these symbols are widely used (I believe not all of them are, but I don't have any evidence for or against it). Also, to make matters worse, they work on similarity. In this Agora discussion thread it says that "Judicial Jocularity Act" is "𝒥𝓊𝒹𝒾𝒸𝒾𝒶𝓁 𝒥𝑜𝒸𝓊𝓁𝒶𝓇𝒾𝓉𝓎 𝒜𝒸𝓉" are in fact the same, but I find some letter missmatches with the Character Information Addon. I believe, however, that just putting the characters here, in eSpeak NG or wherever will improve the situation greatly.

[sukiletxe]

Any opinions on this? How would one tackle this? Would it be necessary (and enough) to edit symbols.dic and then let the translators translate it? I'm asking because there have been complaints about other screen readers spelling them (because of its misuse as decorative symbols, other screen readers pronounce "math bold / italic / whatever" before each symbol and spell them, just like EspeakNG does with cyrillic letters). In my opinion however spelling them is better than letting the synth choose (most of the time, silence or unicode representation), as NVDA does.

[sukiletxe]

I've solved this with an addon which uses Unidecode. Here is the addon, I hope I can submit it for review for the official addons soon.

I don't know if I should close this now.

[LeonarddeR]

I'm not sure whether the symbol dictionary would work for this, likely the speech dictionary might be better here.
Having said that, I really acknowledge this problem, so great to see you're looking into this.

[sukiletxe]

I don't know how the speech dictionary would work in this case. Can we create an entry for each letter? How would that be different from a symbols dictionary?

Also, I believe the addon solution is more elegant now. Sure, you have to install an addon, select or copy the text which contains the symbols and press a keystroke, but at least it's less verbose than other screen readers which read the symbols one by one with their attributes.

[atnbueno]

Hi. I just found out about NVDA, but I'm familiar with the issue of abusing Math symbols.

Wouldn't it be better to detect this kind of "Unicode abuse" at a word-level instead of character by character? I'm sure there's all kinds of "fake font" texts, and probably more than mathematical ones, but in my experience the mathematical ones use these characters isolated. Would it be possible/enough to treat them as a "fake font" if there is a sequence of them? A lonely "𝕴" may be a Lie algebra, but "𝕴 𝖆𝖒 𝕲𝖗𝖔𝖔𝖙" is clearly not Math 😅

Unfortunately I don't have the programming chops nor the time to do a PR to the NVDA repo, but if I can help in any other way...

[seanbudd]

As a start, these symbols should be being reported with their Mathematical names by adding them to the Symbols.dic.

We could eventually add an option to read them as how they are presented visually e.g. "𝕴 𝖆𝖒 𝕲𝖗𝖔𝖔𝖙" as "I am Groot" (in a cursive font).

[seanbudd]

@atnbueno I don't think a sequence is sufficient to differentiate between how these symbols should reported. For example, a sequence of legitimate math would not be announced in a useful way. Do you have an alternative mechanism for deciding how these should be reported?

[ultrasound1372]

I think the fake font would be the best way, although it would cause issues in systems where a font doesn't exist, and would make browse mode handling difficult to do. Presumably renderers are calling up some font to handle these symbols anyway as math writing looks very specific. If someone doesn't have font reporting on, the characters would be converted to their unmodified letter equivalents transparently, whereas if they have font reporting on it would say some dummy font name like that from the character "mathematical italic" but only when entering the block, at least in browse mode and perhaps review cursor. Sukil's add-on solution works well enough but to my knowledge is using a bigger hammer than necessary, also downconverting, say, Greek and Cyrillic letters that may look like Latin letters. This has massive ramifications for people who actually read in those languages. My proposal would be based on the information available in unicodedata, which you already ship with and make use of. If the character is within this block, extract the font specifying part of the name for the fake font, and then perform a lookup for the corresponding unmodified character. For example 𝑥 (MATHEMATICAL ITALIC SMALL X) would correspond to x (LATIN SMALL LETTER X). There would be some conditional branching and perhaps regular expression substitution involved, so this might be used to build up an internal dictionary which is used for performance reasons, but it seems like a viable approach for these blocks in particular. There's not much we can do about people using Cyrillic and Greek letters that look kind of like Latin ones for decorative purposes without alienating anyone actually needing to read that, and the language of the synth can't really be relied on either as some synths might internally be able to handle multilingual text in different scripts, see vocalizer's unicode-based automatic language detection. And these blocks contain Greek letters as well which should be converted to their corresponding unmodified Greek counterparts, not similar-looking Latin equivalents! This would also keep equation reading intact for the most part, it might only get slightly odd when people stick multiple letters side by side to indicate a series of multiplication. I think at that point it's better to manually read by character anyway, and you can't automatically break series up because of functions.

[HiEv]

It looks like the solution would involve two things:

1. Add an option to prefer reading mathematical and other symbols as if they were the closest standard alphabetical character or characters. This way, people who are far more likely to encounter these symbols' use as fancy text, will be able to get NVDA to read those characters as though they're normal letters, instead of merely skipping them or misinterpreting them. Unless you're a mathematician, you're unlikely to need the mathematical symbol names, but adding a way to toggle this option via keyboard would probably also be good. (Perhaps add an optional tone to denote when this is happening?)
2. Add a new dictionary type for each language to translate these special characters into the language's standard characters. This way, any language can have their own translation dictionary from a special symbol or group of symbols into a character or group of characters. This dictionary would allow you to add a symbol like U+2469 (⑩; the number 10 inside of a circle) and have that character be translated as "10". Any characters that are not in this symbol translation dictionary would fall back to their pronunciation in any of the other given dictionaries.

Additionally, I'll note the following website as an example which translates standard ASCII characters into these special symbols:
https://yaytext.com/bold-italic/

If I could assist with this, please let me know.

[Adriani90]

See #14781. I've created a template that we can use to integrate the symbols in the symbols.dic along with the way they should be pronounced to make equations readable, then we would only need a way to retrieve the full unicode name on demand if needed with either nvda+comma or by pressing nvda+dot four times.
We could call the unicode name from the CLDR data base or the unicodedata database but the unicodedata database does not have localizations.
However, I think it would be a big step forward to have the english unicode names on demand first and then think on how to retrieve localizations. There is I think a localized unicode data base in Windows itself, so there should be a way to get the unicode names according to Windows locale.

[HiEv]

@Adriani90 , just to be clear, this particular issue is not about making equations readable. This is about uses of mathematical symbols that look like alphanumeric characters for things other than equations, such as for creating fancy looking text in social media posts, and making NVDA able to read them as though the text is made up of the equivalent standard uppercase and lowercase letters for whatever standard alphabet is appropriate for the user's particular language settings.

As such, the symbols.dic won't help with this, as far as I've been able to determine.