Hacker News Board Hacker News, browsed like a classic message board
Live data from Hacker News

Help! Is This Arabic?

isthisarabic.com

Post reply on HN
Print view View original

231–240 of 304 posts

Re: Help! Is This Arabic?

#231

Post by psychphysic »

IanCal wrote: post #223
psychphysic wrote: post #203

Earlier quoted context omitted.

Exactly if you aren't comfortable with the idea that glyphs are distinct and order can be assumed to matter. This article can't possibly be scoped for someone like that. There's a side note here, that dyslexia can become apparent in very different written languages.[0] In which case don't try and handle multilingual text pay someone else, even if that's on fiverr. [0] https://blogs.scientificamerican.com/observations…

> Exactly if you aren't comfortable with the idea that glyphs are distinct and order can be assumed to matter. But isn't that exactly what it's just telling you? The order is important and if you see this very simple pattern it's wrong. If you read any latin character based language then you surely must be OK with the idea that glyphs can be distinct? Are there many people who exclusively read languages where the ord…

Yea we are in agreement.

Re: Help! Is This Arabic?

#232

Post by IIAOPSW »

kybernetikos wrote: post #132
IIAOPSW wrote: post #52

Earlier quoted context omitted.

ah but you see, my conviction on one letter == one visual glyph context be damned is not hypocrisy nor is it cultural chauvinism. i for one am perfectly consistent. capitalization ought be abolished, tradition is not justification. language is about the encoding of information. capitalization is wasted code space. tradition is dead men telling you how to live. and after that, simplify the letters down into combinatio…

Completely seriously I'd like to see English get rid of capitals. They're almost entirely useless and confuse learners for no benefit. On top of that kids books that use a font that makes I and l look the same annoy me too.

i an conbletelu serious too

Re: Help! Is This Arabic?

#233

Post by Oxidation »

quickthrower2 wrote: post #214
Oxidation wrote: post #204

Earlier quoted context omitted.

Ah yes, the good old "your heuristic doesn't work on my carefully constructed pathological case" observation, wouldn't be the Internet without that. GP also said "probably". That's how heuristics work.

You would almost think what you described was our profession :-)

Wait, you get paid for Internet snark?

Re: Help! Is This Arabic?

#234

Post by IIAOPSW »

junon wrote: post #134
IIAOPSW wrote: post #35

Earlier quoted context omitted.

That's not entirely what I meant by my question. Let me try rephrasing. Is there a thing that works on the same principle as a 7 segment display, possibly with a more than 7 segments but still a manageable amount, which is designed to display Arabic characters with a similar amount of legibility?

I'm not an expert but I have gone down both the Arabic font creation road and the custom segmented display roads - not at the same time, of course. My understanding of Arabic is that this would be _exceedingly_ hard to generalize into individual, discrete glyphs that could not only be broken out into their own segments but then repeated in a way that makes sense. The only way I could personally think of is with matri…

What exactly are the character blending rules? When you say non-trivial, are we talking normal non-trivial or oops-I-embedded-a-Turing-machine-in-my-written-language non-trivial?

I'm going to guess the various shifts you allude to are simple transformations for a pen-plotter, perhaps accumulating some "context" offsets to the initial position or flipping some glyph etc. Not sure how to translate that to segments, unless you can electro-mechanically adjust the segments...?...

Re: Help! Is This Arabic?

#235

Post by jmopp »

AdamN wrote: post #200
zarzavat wrote: post #159

Earlier quoted context omitted.

There is a difference between literacy and fluency. Literacy is ability to understand a writing system. I am literate in the Latin alphabet. Fluency is ability to understand a language. I am fluent in English. Neither implies the other. You can be fluent but illiterate (the default until modern universal education), and literate but not fluent (I am literate in the Latin alphabet but I am not fluent in Italian). The…

2B still seems like a stretch. Can 2B people read (even with low comprehension) a basic paragraph in Arabic? My SWAG would be 1B people.

A quick Google search seems to indicate that there are 1.9 billion Muslims in the world. It makes sense to assume they know enough Arabic to recgonise at the very least common religious set-phrases like the Bismillah or the Shahada. In my book this counts as "some degree of Arabic script literacy"

Re: Help! Is This Arabic?

#236

Post by foldr »

Dylan16807 wrote: post #220
tsukikage wrote: post #172

Earlier quoted context omitted.

> pump up the numbers in a misleading way Is it misleading? You don't have to be anything like fluent to realise when text rendering is broken. The quantity that is actually relevant to the discussion is the number of people who, when they look at your UI, will know that the arabic text rendering is broken; not the number of people who are fluent in arabic.

The only reason for me to care that it's broken is on behalf of the people that can read it. So even though I know a single digit number of words, and you can count me in that two billion, nobody should care about getting it right on my behalf.

>The only reason for me to care that it's broken is on behalf of the people that can read it.

The people who can't read it, but who can see that it's broken, will form a lower opinion of your product. It's as if I went to a Polish website and the text was all right-aligned and in all caps. I can't read Polish at all, but I'd still form an opinion about the quality of the site.

Re: Help! Is This Arabic?

#237

Post by joveian »

schoen wrote: post #160
7373737373 wrote: post #152

I hope someone creates something similar for distinguishing southeast asian scripts. Would be very useful for GeoGuessr

A good start is https://en.wikipedia.org/wiki/Wikipedia:Language_recognition... although it's pretty weak on Southeast Asia in particular.

That is interesting. I happened to notice that some described ligatures under Dutch aren't appearing for me for some reason. I've been thinking about font rendering in different scripts lately since I like the DejaVu fonts but unfortunately the project was abandoned a while ago and coverage is incomplete in many areas. With other more complete fonts available these days it seems likely to be better to use a more restricted version of the fonts to allow fallback to the more complete scripts, although I'm not sure how well the unicode blocks match actual usage. It is easier than ever to mix languages but still hard to know if you are messing them up. I found out my terminal was rendering Japanese wrong when someone posted this link recently:

https://heistak.github.io/your-code-displays-japanese-wrong/

Now it displays Chinese wrong.

Re: Help! Is This Arabic?

#238

Post by thiht »

rippercushions wrote: post #5

This website is well-meaning, but will be difficult to parse if you don't have elementary Arabic and can't tell apart ال from ل ا. What it really needs is a simple reference input string, examples of how it gets broken, and what to do to fix them. The middle case, where the sentence is correctly rendered RTL but the individual words are LTR (breaking the ligatures), is particularly common and insidious because it loo…

> This website is well-meaning, but will be difficult to parse if you don't have elementary Arabic and can't tell apart ال from ل ا.

That's a really weird comment. Just Ctrl+F ل ا in your supposedly Arabic text?

Re: Help! Is This Arabic?

#239

Post by IIAOPSW »

least wrote: post #69
IIAOPSW wrote: post #41

Earlier quoted context omitted.

I could understand if this were a non-alphabet system like Chinese where the information coding is that dense per character. But for an alphabet system, why the mismatch between the bits of information per character and the bits needed to visually represent that character? The Latin alphabet mostly fits on a 7 seg display, and 26 letters requires a theoretical minimum of 5 bits to encode, so the graphical efficiency…

Because the information coding is dense. Arabic script has a lot of tiny shapes that indicate what letter it is and it also has diacritic marks, which on their own are similarly intricate and small. You can't represent one line, two lines, a dot, a circle, a hamza, etc. above, below, to the side of particular characters, without a lot of resolution.

hold up. Representing a line, two lines, a dot, a circle... in one of several fixed positions relative to the character...that's exactly what a segmented display is good at. I think you may not have got what I meant by "dense". I'm looking at it from a pure information theoretic view. There's something like 5000 Chinese characters overall. Even if I could find a system of on/off segments to represent them all, I would still have to use at least 13 segments, because it takes at least 13 bits to encode 5000 possibilities. Chinese is dense in the sense that a whole 13 bits have to fit in the space of a single character. Granted diacrtics and other markers do add one or two new bits, but adding permutations isn't the same as adding complexity. Every time you learn one diacritic, you've doubled the number of letters you know how to write. Whereas in Chinese, when you learn a new character, you've learned a new character.

Other commenter nailed the real technical hurdle. Context dependent spacing system. How do monotypes do it (or do those not exist)?

Re: Help! Is This Arabic?

#240

Post by IIAOPSW »

schoen wrote: post #77
least wrote: post #69

Earlier quoted context omitted.

Because the information coding is dense. Arabic script has a lot of tiny shapes that indicate what letter it is and it also has diacritic marks, which on their own are similarly intricate and small. You can't represent one line, two lines, a dot, a circle, a hamza, etc. above, below, to the side of particular characters, without a lot of resolution.

Another challenge for representing Arabic a display made of line segments is all the curves. If it weren't for the dots, I imagine you could write consonant-only Arabic on something close to a seven-segment display, although it would look bizarre because all the letters would be isolated and there would be so many more straight line segments than in normal Arabic script. (An absolute majority of the isolated forms of…

I don't think curves are a technological problem for a segmented display. You can easily make curved segments and dots. Its all just an LED behind a window painted over with the negative image of the curve. In principle you could make a display that is optimized for OCGDU rather than EIFHTL.

Much like latin characters, at some point a segmented display is just going to have to compromise on the tradeoff between correct an the number of parts. We see it work just fine casting the curvy latin letters to squares. So the real question I reckon is how much can you mangle the diacritics / curves / other parts and still read it? In other words, what is the bandwidth redundancy of this writing system?

Post reply on HN