The flaws of thinking "Just use Unicode"(itworld.com)
itworld.com
The flaws of thinking "Just use Unicode"
http://www.itworld.com/print/58558
6 comments
Great information; thanks!!! The ICU library looks very comprehensive, but I hope it's not too bureaucratic to use. :)
This may work well for 35 languages. Not 350 or 3500.
Why doesn't this scale to any number of languages? Assuming you are using the existing practice of not hard coding strings in your software but rather referring to them with some sort of IDs then I see no reason not to use 3000 languages with this approach.
We are using IDs on our website instead of hard-coded variables. It causes no end of problems. A non-exhaustive list:
1. Using variables to put numbers in the string. Singular and plural are not always only two options. Rules in this regard can get quite complex. gettext has this solved by using rules: n%2 == 0? Use A. n==0? Use B. n%2 == 1? Use C. etc. That should give you an idea about complexity.
2. Replacing words into the string is nearly impossible due to declensions.
3. Names. When using a different script, will the name be transliterated? Do the names need to be inflected? Will they be handled correctly when changing writing direction?
3a. Names have a whole load of problems anyway. The "given name" isn't always the first name. The surname isn't always the most important one. Some people don't have family names, some people have two or more. Some people have double-barrelled given names. Etc.
4. Ordinals differ greatly from language to language.
5. String length can mess up your layout. A good translator can help by using different idiom, but sometimes it's just not possible.
6. Layout expectations are different when the writing system is right-to-left. You also need to learn the details of Unicode's direction algorithm, because you will encounter situations where it doesn't work properly without guidance.
These are just some problems we've actually encountered when using multiple languages.
1. Using variables to put numbers in the string. Singular and plural are not always only two options. Rules in this regard can get quite complex. gettext has this solved by using rules: n%2 == 0? Use A. n==0? Use B. n%2 == 1? Use C. etc. That should give you an idea about complexity.
2. Replacing words into the string is nearly impossible due to declensions.
3. Names. When using a different script, will the name be transliterated? Do the names need to be inflected? Will they be handled correctly when changing writing direction?
3a. Names have a whole load of problems anyway. The "given name" isn't always the first name. The surname isn't always the most important one. Some people don't have family names, some people have two or more. Some people have double-barrelled given names. Etc.
4. Ordinals differ greatly from language to language.
5. String length can mess up your layout. A good translator can help by using different idiom, but sometimes it's just not possible.
6. Layout expectations are different when the writing system is right-to-left. You also need to learn the details of Unicode's direction algorithm, because you will encounter situations where it doesn't work properly without guidance.
These are just some problems we've actually encountered when using multiple languages.
You don't even need to go outside English for some of these to be a problem,
What letter does the name Van der Waals sort under?
What about D'Arcy
Does McKay come before MacKay ?
Is 7-4-09 an American holiday or a week after April fools?
What letter does the name Van der Waals sort under?
What about D'Arcy
Does McKay come before MacKay ?
Is 7-4-09 an American holiday or a week after April fools?
Disclaimer: I'm don't do any web software at all, I work on embedded devices.
We use IDs and I've not seen too many of the problems you mention. I will, however, agree that the string length can cause serious problems for a layout. We tend towards using very flexible UI designs where the UI elements resize and reposition themselves a long with the strings. However, UI design isn't as stylish for embedded devices so we get let off easy here. One of the biggest problems we have with our newer developers is that they don't know how to write their UIs so that they can easily re-arrange themselves for our various languages. Also, the Left and Right layout changes are all part of our look and feel engine.
I wonder if some of these problems are specific to web applications and some of the 'fancier' UI designs that we don't use? I don't actually know enough about the differences between the two to say.
We use IDs and I've not seen too many of the problems you mention. I will, however, agree that the string length can cause serious problems for a layout. We tend towards using very flexible UI designs where the UI elements resize and reposition themselves a long with the strings. However, UI design isn't as stylish for embedded devices so we get let off easy here. One of the biggest problems we have with our newer developers is that they don't know how to write their UIs so that they can easily re-arrange themselves for our various languages. Also, the Left and Right layout changes are all part of our look and feel engine.
I wonder if some of these problems are specific to web applications and some of the 'fancier' UI designs that we don't use? I don't actually know enough about the differences between the two to say.
There's more to it than language replacement - there are cultural differences and representation of numbers, quantities etc., RTL support, allowing users to input their data in more than one language and have it behave in a reasonable manner...
But honestly, out of all the problems you can have, that's a good one indicative of success.
But honestly, out of all the problems you can have, that's a good one indicative of success.
I did a quick scan of the ICU library mentioned above (thanks for that!!) and it does handle these things. http://userguide.icu-project.org/i18n
Besides several of the reasons mentioned in the original article (cultural differences of color and symbols, left to right / top to bottom writing systems, etc.)
In many non-Roman scripts the letters change form drastically based on if they are at the beginning, middle, or end. In some, the letters literally switch positions with other letters visually. Current operating system font support, Unicode standards, and fonts simply don't exist to support all of this.
Graphite (http://scripts.sil.org/cms/scripts/page.php?site_id=nrsi&...) is one great initiative to get the ball moving here and is gaining some support (being included as part of OpenOffice 3), however there are HUGE steps that have not yet been taken before applications can even think about scaling to 3000 languages. Unicode is great, but it's not a simple solution/answer/cure-all.
Hopefully we'll be able to scale that large one day, but luckily for now we only need to target < 1% of the world's languages.
In many non-Roman scripts the letters change form drastically based on if they are at the beginning, middle, or end. In some, the letters literally switch positions with other letters visually. Current operating system font support, Unicode standards, and fonts simply don't exist to support all of this.
Graphite (http://scripts.sil.org/cms/scripts/page.php?site_id=nrsi&...) is one great initiative to get the ball moving here and is gaining some support (being included as part of OpenOffice 3), however there are HUGE steps that have not yet been taken before applications can even think about scaling to 3000 languages. Unicode is great, but it's not a simple solution/answer/cure-all.
Hopefully we'll be able to scale that large one day, but luckily for now we only need to target < 1% of the world's languages.
As is typical in the software universe, you must target which subset of folks are central to the profits of your business.
How many customers will you have past your particular and local top 35 languages? Enough to warrant the translation and support and testing work inherent in adding another language?
And yes, the target market for and the list of languages used by your customer base(s) may or will differ from mine, too.
How many customers will you have past your particular and local top 35 languages? Enough to warrant the translation and support and testing work inherent in adding another language?
And yes, the target market for and the list of languages used by your customer base(s) may or will differ from mine, too.
I use three languages commonly : French, English and Russian. I can set the language for non-Unicode programs only once, which means I can't display all programs written with a different encoding properly.
Please just use Unicode.
Please just use Unicode.
It's sort of interesting how unicode solved your problem, but totally ignored the exact same problem in east asia. It's still not possible to have a single unicode text file with japanese and chinese kanji in it. The official unicode answer is to not use plain text; use a rich text format that allows the specification of fonts instead. I sort of think unicode would be a much better global standard if it hadn't tried to merge the different kanji used by over a billion people.
The article doesn't say you shouldn't use Unicode. But it explains that only using Unicode just isn't enough. You need more. You shouldn't stop thinking at Unicode. Don't JUST use Unicode, use Unicode and more.
A small example: many programs do not fold accents. So when I start typing "lo" in the to: field of most email apps, it suggests "Lorena Foo" but not "Fulanito López".
I stopped taking this post seriously after I read this:
> Further east we hit ideographics. The concept of a "letter" has just flew out of the occidental window. Not only that but the text is laid out top to bottom.
Chinese and Japanese are both read from left to right (although in the past characters were written top to bottom and right to left). The only language that I know of that is presently laid out from top to bottom is Mongolian. I'll be damned if I ever need to write software for people in Mongolia.
> Further east we hit ideographics. The concept of a "letter" has just flew out of the occidental window. Not only that but the text is laid out top to bottom.
Chinese and Japanese are both read from left to right (although in the past characters were written top to bottom and right to left). The only language that I know of that is presently laid out from top to bottom is Mongolian. I'll be damned if I ever need to write software for people in Mongolia.
For me it was:
> Oh, and there are languages with unbounded sets of "characters" such as Chinese which literally cannot be fully described in Unicode.
I read Chinese just fine in Unicode. Is there something I'm missing? You have 50,000 commonly used characters, 100,000 rare ones - this certainly doesn't bust any limitations within unicode. Makes fonts rather large, but that's another issue altogether.
The rest of the article seems to be confused about what unicode is - I don't think anybody seriously goes into unicode expecting it to practically translate for them. It's an encoding, that's it. It won't convert "," to "." depending on locale, it certainly won't rephrase your messages for you to be sensitive to local cultures. That's your responsibility, not the purview of a form of encoding.
I think the author's point at the end was: "Unicode" as a concept won't localize your programs for you. Well, duh?
> Oh, and there are languages with unbounded sets of "characters" such as Chinese which literally cannot be fully described in Unicode.
I read Chinese just fine in Unicode. Is there something I'm missing? You have 50,000 commonly used characters, 100,000 rare ones - this certainly doesn't bust any limitations within unicode. Makes fonts rather large, but that's another issue altogether.
The rest of the article seems to be confused about what unicode is - I don't think anybody seriously goes into unicode expecting it to practically translate for them. It's an encoding, that's it. It won't convert "," to "." depending on locale, it certainly won't rephrase your messages for you to be sensitive to local cultures. That's your responsibility, not the purview of a form of encoding.
I think the author's point at the end was: "Unicode" as a concept won't localize your programs for you. Well, duh?
> I think the author's point at the end was: "Unicode" as a concept won't localize your programs for you. Well, duh?
That's his point as I understand it. But I disagree with the "duh" part. It's rare to see software that goes beyond simple translation.
That's his point as I understand it. But I disagree with the "duh" part. It's rare to see software that goes beyond simple translation.
The Chinese newspaper I usually read is top to bottom, left to right. So are the vast majority of the other newspapers and magazines on the rack, the last piece of fiction I bought and my book on go. I also encounter a good number of MS word documents, including contracts printed that way.
That said, top to bottom is just too much of a pain to deal with on the web until CSS supports it.
That said, top to bottom is just too much of a pain to deal with on the web until CSS supports it.
Sure, there are some, but vast majority? That's stretching it, a lot, unless you're in the historical section of a bookstore. This is my experience in Beijing, Hangzhou, and Shanghai.
Is it too much to ask the Mongolians to turn their monitors 90 degrees clockwise and look at them in a mirror?
There is no such thing as a definitive list of languages
This is true. But SIL has what is considered the "standard" for this list, and there are almost 7000 of them. Many of them are unwritten. See http://www.ethnologue.com/ and ISO 639-3.
This is true. But SIL has what is considered the "standard" for this list, and there are almost 7000 of them. Many of them are unwritten. See http://www.ethnologue.com/ and ISO 639-3.
>>What would our software need to do to operate in, say, Europe? Well, apart from selecting a subset of the myriad of languages (France alone has 30+ )...
Who said that you need to support every language in a region to have your software operate there? If someone living in France speaks an obscure dialect then its a pretty safe bet they ALSO speak French. And if they only speak a single obscure language they probably aren't going to be your customer anyway.
If you hit the top 15-20 most used languages in the world then you'll be fine.
Just use unicode.
Who said that you need to support every language in a region to have your software operate there? If someone living in France speaks an obscure dialect then its a pretty safe bet they ALSO speak French. And if they only speak a single obscure language they probably aren't going to be your customer anyway.
If you hit the top 15-20 most used languages in the world then you'll be fine.
Just use unicode.
I really agree with just supporting the top 15-20 most used languages, but the problem with "just" using unicode is that #1 on that list of languages isn't well served by some unicode formats.
In fact, the PRC requires that all language-related products introduced into the Chinese marketplace must be able to function in GB 18030.
In fact, the PRC requires that all language-related products introduced into the Chinese marketplace must be able to function in GB 18030.
Most people in France (or Europe) learn English in school, don't they? IMHO supporting anything but US ASCII is presumptuous.
The whole premise of the link is flawed. It's an article about "localization is hard". Well, duh.
I don't see anything in it, at all, about character set and encoding choice. And in that realm, yes: you should just be using UTF-8. Not even "unicode" -- you want to be in UTF-8, period. Convert at the edges of your system if you absolutely must handle other encodings. But wherever you can, you should just be using UTF-8.
I don't see anything in it, at all, about character set and encoding choice. And in that realm, yes: you should just be using UTF-8. Not even "unicode" -- you want to be in UTF-8, period. Convert at the edges of your system if you absolutely must handle other encodings. But wherever you can, you should just be using UTF-8.
The premise of the article wasn't that you shouldn't use Unicode. It was that you need to care about more than just Unicode. Thinking that the answer to localisation is to "just use Unicode" is a flaw in people's thinking that this article points out and corrects.
I'm curious as to why you believe UTF-8 is the only encoding you should use? I know of at least one person that says you should use UTF-16, because then it's immediately obvious when a certain piece of data is in the wrong encoding, whereas with UTF-8 you need to look for non-Latin characters (or accents) to distinguish it from, for example, ISO-8859-1. That line of thinking holds some value to me. Why do you believe otherwise?
I'm curious as to why you believe UTF-8 is the only encoding you should use? I know of at least one person that says you should use UTF-16, because then it's immediately obvious when a certain piece of data is in the wrong encoding, whereas with UTF-8 you need to look for non-Latin characters (or accents) to distinguish it from, for example, ISO-8859-1. That line of thinking holds some value to me. Why do you believe otherwise?
UTF-16 can't be used with traditional tools. No grepping, no strings, no scanning quickly with a text editor. It locks you into whatever oddball toolchain you cooked up in development. Old code that passes the string to a strcat() will fail in mysterious ways. And your old libraries probably expect 1-byte characters anyway, which means you're going to be constantly converting
The inability to distinguish between UTF-8 and ASCII is a feature, not a bug. If you want to know if it's a valid string, just validate it. UTF-8 has extraordinarily strong validation properties, it's almost impossible for strings in other encodings to parse successfully.
The inability to distinguish between UTF-8 and ASCII is a feature, not a bug. If you want to know if it's a valid string, just validate it. UTF-8 has extraordinarily strong validation properties, it's almost impossible for strings in other encodings to parse successfully.
> Old code that passes the string to a strcat() will fail in mysterious ways. And your old libraries probably expect 1-byte characters anyway, which means you're going to be constantly converting
But they will have that same problem with UTF-8, right? Just as soon as you start using non-Latin characters? And if you aren't, there's no point in going beyond ASCII.
But they will have that same problem with UTF-8, right? Just as soon as you start using non-Latin characters? And if you aren't, there's no point in going beyond ASCII.
Uh, no, not at all. The ANSI strXXX() functions work just fine with UTF-8, that was kind of the point. Anything treating strings with nul termination or counted byte length works without change. Any work involving only ASCII substrings (e.g. parsing of computer-readable data) works without change. No non-ASCII character is represented using ASCII bytes in the string. Everything Just Works. Which is why you should just use it.
The sole exception are algorithms that truly need random access to characters within the string (e.g. "give me the 32nd character"), which can't interoperate with any variable-length encoding. But there are precious few of these. All anyone cares about is substrings.
The sole exception are algorithms that truly need random access to characters within the string (e.g. "give me the 32nd character"), which can't interoperate with any variable-length encoding. But there are precious few of these. All anyone cares about is substrings.
We use the IBM ICU library for our product and it handles the vast majority of the issues presented in the article. There is no reason for me to be writing the code to handle the thousands of permutations of language, country, and culture when someone else has already done the heavy lifting here. ICU solves problems such as: sorting, number representation, etc.
The other problems should be things unrelated to Unicode - things your product / business already need.
1)Pick a target market - If you haven't picked a target market worrying about using Unicode to support some languages is putting the cart before the horse.
2)Get a translator(s) that know what they are doing - You already needed a translator - now you need to pick one competent enough to know the difference between Mexican Spanish and the Spanish spoken in Spain.
3) Write software that can switch look and feel on the fly during run time - This has already be done and should be a good philosophy for companies writing GUI intensive applications anyway.
Really - The correct sentence should be, "Just use Unicode, with your pre-existing good business practices."