I have a feeling that you’d find various parts of “CJKV Information Processing” Second Edition, due out next month, to be very helpful. See: http://oreilly.com/catalog/9780596514471/

If you send to me an email, I can send back its current TOC.

Dr. Ken Lunde
Senior Computer Scientist, CJKV Type Development
Adobe Systems Incorporated
lunde@adobe.com

Qiu Xigui’s Chinese Writing (which I set into type) is a detailed and comprehensive account of the subject. It emphasizes historical evolution of a writing system that has been in continuous use since the Bronze Age and the subject of scholarly study for two millennia; it is not an easy read for the non-specialist. (I keep my copy handy, shelved near “CJKV Information Processing.”)

Chinese writing of course is not alphabetic. Nor is it “ideographic,” despite the Unicode Consortium’s unfortunate adoption of that term: a 12th-century survey of 24,235 different characters counts only 7% ideographic forms. The sheer number of characters is a formidable obstacle when it comes to type; Chinese and Koreans may argue for precedence in inventing movable type many centuries ago, but the difficulty of creating large enough fonts was a major factor in delaying serious use of the technology for CJK languages until well into the Industrial Age.

Font size does not go way in the Information Age, and continues to constrain artistic innovation. So does geography: the same writing system covers a host of dialects that have evolved separately so long as to be mutually unintelligible. This is compounded by the relative paucity of sounds compared to other languages, so that meaning leans more heavily on context. It is not uncommon to see Chinese resolving doubts in conversation by “writing” a character with a finger to show what they mean. All in all, tradition weighs very heavily on Chinese writing.

On the other, Chinese writing is also an ancient art form. Beautiful (brush) writing goes back many, many centuries, with long-established conventions and the artistic freedom to re-interpret and vary from same.

One area where Opentype-type “context” could count is traditional vs. simplified Chinese, two separate “languages” (or attributes) in computerdom. Anyone well imbued with the both forms will know how hard it can be to unify a given piece of Chinese text to one set or the other: the two overlap not only in the concrete world but also as one reads them. A few recent Opentype fonts implement traditional and simplified characters as alternate forms that InDesign’s glyph palette allows one to swap. I read somewhere that one of the Windows system fonts for Chinese briefly could use a language attribute to select between but I’ve mislaid the reference.

I don’t mean to be as obscurantist as this may sound, but occasionally I recall the closing words in the movie “Chinatown,” where someone explains to Jack Nicholson’s character that some things Chinese defy simple explanation.

Good luck!
David

I think the first thing you need to clear is what you mean by “contextual features”. The choices are:

1. Context between print characters within a line.
2. Context between strokes and radicals within a print character.
3. Context between strokes, radicals or characters in handwritten script.

To 1: There is basically none. Each character occupies an equal rectangle space (it’s why Chinese is nicknamed the “square script”) and does not interact with neighboring characters. The only contextual variation happens to the spacing of punctuations, and to some punctuations themselves depending on the direction of the text flow (horizontal vs. vertical punctuation).

To 2: As far as I know, there are no clear-cut rules, and all common-sense rules are derived from calligraphic traditions, see below.

To 3: Well, Chinese calligraphy is a self-contained discipline of visual art, as complex as any other discipline. Although I’m somewhat familiar with the topic (some in my family are professionals and semi-professionals practicing and/or teaching Chinese typography and its history), I don’t even know where to begin. In any case, I’d say this is something you’re most likely not able to learn in a couple of weeks. And though the same could be said about Latin and Arabic calligraphy, the difference is that there isn’t a Chinese Gutenberg who’s formalized a somewhat authoritative set of rules for Chinese (micro-)typography. I’ve seen one book about Chinese type design (written in Chinese, of course), and it’s basically a collection of a few hundred examples. The whole field is far from being systematically developed (in the scientific sense) yet.

With regard to ideographs (I agree that this name is unfortunate, but I now prefer it over “Chinese characters”) being confined to a square or otherwise uniform box, that is merely a convention. We have developed a typeface design, called Kazuraki, which breaks that assumption. (It also broke some Adobe applications in the process. ;-)

Each ideograph (and kana) in Kazuraki is genuinely proportional. Some are wider than they are tall, and vice versa. It depends on the ideograph. The typeface design even includes two- and three-character vertical hiragana ligatures. I crafted this just now:

Note the two-character hiragana ligatures in the first, third, and fourth lines.

Anyway, you can read about the production aspects of Kazuraki in Adobe Tech Note #5901, which I authored. See: http://www.adobe.com/devnet/font/pdfs/5901.Kazuraki_Tutorial.pdf

Dr. Ken Lunde
Senior Computer Scientist, CJKV Type Development
Adobe Systems Incorporated
lunde@adobe.com

Fascinating, could you provide an example in horizontal, please?

Edit: OK, I just read “All glyphs needed to be replicated for vertical use, due
to limitations in the ability to shift glyphs in both X- and Y-axis directions in the OpenType ‘vmtx’ table”. That seems to be an very unreasonable limitation, does the last version of OpenType spec get rid of that?

@lunde - wow! That is impressive!

Here is the same text, but set horizontally:

Although the glyphs themselves are the same as their vertical counterparts, they have different GIDs, and the ’vert’ GSUB feature is performing the context-sensitive substitution (the context is vertical writing mode).

In terms of the need to replicate the glyphs for vertical writing, the OpenType specification has not change to alleviate the need for this. Given the fact that subroutinization effectively takes care of any filesize-related issues, and the fact that users need not be exposed to the inner workings of the font, it is a non-problem. Even if the OpenType specification were to change, there is then the ripple effect of ensuring that all of the clients and applications that use the ’vmtx’ table recognize the new version correctly.

Dr. Ken Lunde
Senior Computer Scientist, CJKV Type Development
Adobe Systems Incorporated
lunde@adobe.com

Henry, thanks for your question.

I mean “2. Context between strokes and radicals within a print character.” and also
“3. Context between strokes, radicals or characters in handwritten script.”

Ken Kunde is absolutely right to point out that running hand style produces ligatres and therefore a kind of contextual impact of one glyph on the other. In fact in the ra/no ligature ( hiragana) the one blends into the other completely.

What I am after though is PhD or other works that analyse what goes on within a chinese glyph/kanji when a radical has to fit with other strokes. There are many examples of exceptions and even reinventions of shapes but there also seems to be a bit of a pattern especially when the radical is essentially intact.

For example

If you take the radical 夕 (evening) and then look at it in 死 (death) you can see the angle become more acute and to preserve clarity all the strokes become less dense. If you look at 多 (frequent) the angle becomes less acute and again there is stroke thinning which I willstop mentioning now. A more extreme axample is 夢 where the angle is flattened even more.

The reason I picked the first one is that it’s angle makes it’s transformations fairly interesting.

The example of 日 (sun) is contrasting because it is a boxy shape and also because it is used a great deal - 603 entries in the dictionary I am using just now. It can appear on the side 晤 (clear),曚 darkness; top 昇 (rise up), and the vertical and horizontal compression looks almost like it could be done mechanically even if it can’t be.

Then there is 尸 (flag) which interestingly shrinks just part of itself. A very exteme axample is 屬 ( how is that for dense with strokes? - genus, subordinate official, belong, affiliated)

Then there is 川 (River, stream) which can be found in 順 (which means several things; docility, obey, occasion, order, right, turn) and which is a horizontal squish. But also can be interleaved like this: 州 (province or state) or shot through: 卅 (thirty).

So that’s the behavior, or some of it, and what I am wondering is if any body recalls work done that describes the systems at work. I mean something beyond “common sense” and is fairly analytical - especially in visual terms.

My next step is to go trolling in the PhDs and see what that might turn up.

Eben, I suspect that rather than looking into PhDs, you may get more just by asking a classically trained Chinese or Japanese calligrapher. I’m sure there are training manuals also, but they are likely to be in Chinese or Japanese. Keith Tam I think would be a good person for you to contact.

Since even with Roman letters design balance is such a complex matter, I doubt you are going to get beyond a handful of rules. That handful will no doubt be very significant, though.

Finally, as you probably know, some forms, like the characters for heart and water, are different on their own and in combination with other radicals.

The technology that we developed, and which we continue to use as a design tool, does this in a very unique way. We first disclosed it in late 2005, when we held OpenType Developer Day events in Beijing and Seoul. Basically, an element library is developed that is used to construct the glyph. What makes this element library unique is that each element is a multiple master glyph, with up to four design axes. Some simpler elements may have only two design axes. This allows the shapes to contort within a defined scope. Height and width are obvious parameters, but stroke thickness and curvature are others. We also implemented a quasi-fifth axis, for rotation.

BTW, I found the document I was referring to. It is one of the presentations at the OpenType Developer Day in Beijing (11/2005). I will send the PDF to you in a few minutes.

In terms of the design tool, it is proprietary, and we do not sell or license it. This is mainly due to the extreme nature of its learning curve. Documentation and support would be a nightmare. But, it works. Our Kozuka Mincho and Kozuka Gothic typeface families were designed with this technology.

Dr. Ken Lunde
Senior Computer Scientist, CJKV Type Development
Adobe Systems Incorporated
lunde@adobe.com

I see from Lunde and James that this subject is already very far advanced. Thanks.

@Eben: Since we’re talking about micro-typography and type design, you should also pay attention to the differences between vector and bitmap fonts, because they sometimes deal with areal compression in very different ways — sometimes with extreme measures in case of bitmap font. Take a look at the following example of Microsoft SimSun at three hinted sizes plus the anti-aliased rendering:

Notice how the 10 pt version simply omits strokes and converts complex radicals into abstract stand-ins? Also interesting, how the 16 pt version suddenly gets some serifs, which the 10 pt and 12 pt versions don’t have?

@William: On the subject of areal compression, I wouldn’t concentration too much on free calligraphy because calligraphers are by and large free to break out of the square box while type designers usually can’t. Also, calligraphers don’t have to deal with limited resolution, which is another highly relevant aspect for the type designer.

@Lunde:

This it a little off-topic, but since you actually work for Adobe Type, I’d like to ask you if Adobe has any plans to systematically improve and expand the hinting of its fonts (not just CJKV, but generally), especially because with the “web revolution” (and Adobe now on the forefront of online DTP), print publishing is no longer the only major focus as it once was.

Or do you believe with the dominance of LCD screens nowadays, font hinting has been made more or less obsolete by subpixel rendering? (I personally don’t like subpixel rendering very much and still use traditional font smoothing on LCD screens.)

James,

That is a loaded question, because there are several ways to approach the issue of web (and mobile) fonts. One way is to start off with a typeface design that is display-friendly. A lighter sans serif is what we’re favoring these days. The design can be further optimized for lower-resolution devices.

In terms of hinting, we’re constantly improving our rasterizers. In terms of the hinting parameters that we specify in our fonts, and the hinting that is applied via our “autohint” tool, they work in unison. Poorly-chosen hinting parameters effectively means poor hinting. It’s one of those “garbage in, garbage out” sort of things. It is better to not hint a font than to apply poor hints. As their name suggests, hints are designed to guide the rasterizer. When the hints are poor, our rasterizers tend to do a better job without them.

Dr. Ken Lunde
Senior Computer Scientist, CJKV Type Development
Adobe Systems Incorporated
lunde@adobe.com

It’s just I’ve read quite a few articles on how Microsoft (or rather, the people they’ve contracted) spent extra time to fine-tune the hinting of fonts like Tahoma — sometime manually re-create them — in order to achieve perfect low-resolution rendering. The screen optimization of the new MS YaHei has created quite some buzz in the Chinese IT press, too (also, how much MS has allegedly paid for it). And looking at my SimSun example above — that kind of “hinting” *has* to be done manually, right?

And the effort really pays off: Adobe Myriad looks stunning in print, but it doesn’t even come close to Tahoma or Verdana at 10 pt. Given that Adobe doesn’t really have many expanded sans-serif families for “business” usage (why is that actually — because Slimbach mainly does serifs?), I’ve always wondered why Adobe wouldn’t commission a manually hinted 10 pt version of Myriad — or better yet, Myriad SemiCond, as it saves more screen real estate. Am I really the only person who wants to have Myriad SemiCond as his screen font?

Henry, It may be that they expect screens to get better. But I don’t think it’s fair to expect Ken to spill the beans. Also for my purposes the printed text is what’s relevant although you certainly make a good point!

William, you are absolutely right. Keith would be just the guy to ask.

Ken thanks very much indeed!

I unfortunately have never seen a reference work that describes the contextual features of the radicals of Chinese characters in the kind of systematic detail that Eben is looking for.

I did briefly learn Chinese calligraphy when I was young, in the regular script (楷書) style. In this style, each character needs to fit into a square—not as extremely as in the Song/Ming typefaces used for print, but certainly a lot more than in cursive styles where the calligrapher is freer to break out of the box (Ken’s sample of Kazuraki beautifully shows how the freer calligraphic styles break out of the confinement of the square).

Learning Chinese calligraphy consisted entirely of slavishly reproducing the characters written by my teacher. I never learned any theory or even how to write other characters without a sample in front of me to imitate. The idea was to copy lots and lots of characters written by masters to get a sense of how to adjust the shape and proportion of the radicals in order to fit each character into a square box.

I have a feeling most Chinese calligraphers achieve this instinctive sense of harmony and proportion of the character components through practice and looking at hundreds of examples, and I doubt they analyze to any detailed extent what they are doing when they shape each character. Type designers might be more systematic (I frankly have no idea), but whatever system they use would be derived from an instinctive sense of proportion, just like calligraphers.

Eben, is it true that someone at Reading is gearing up to do a CJK font? He or she would need to find a way to make it doable in a year. I would say that without something like Adobe’s proprietary design tool, selecting a subset of the 100 most frequently used characters will be enough of a challenge to design and implement in a font by one person (who has no prior CJK font design experience) in one year.

Or what about tackling hangul, the Korean alphabet? Although hangul glyphs are built up from components and require a sense for shaping these components as in Chinese characters, it is possible to apply a more modular approach as the variation won’t be as extreme as in Chinese characters. If one is aiming for a more innovative design, one can minimize the variation in the shapes of the individual alphabetic components and break free of the square box. I’ve seen at least one person take that approach to create a free hangul font for Linux. Plus, I may be able to provide design feedback for hangul as a native reader of Korean. In any case, good luck to whoever is willing to take up the CJK challenge and to Eben and all the Reading students.

Now you’ve mentioned it, I think creating a Hangul font is a perfect starting exercise for a Unihan font, especially regarding the contextual placement of strokes and radicals.

@Jongseong:

You’re right that almost all Chinese calligraphers learn “placement” (that’s the proper English term, I believe) intuitively rather than systematically. Still, there have been some (rather rudimentary) attempts to scientifically analyze the common aesthetics of Chinese calligraphy, most notably by the later grand master Qi Gong.

He analyzed a large number of famous calligraphical works and determined that the weight of a character (center of gravity) should ideally fall to the four intersection points which divide the bounding square at the Golden Ratio, and most prominently, the upper left of the four should be the main center of gravity (among others, it means statistically, the density of strokes should be higher around that point than anywhere else) — which, of course, is very much in line with the well known rule of “top denser than bottom, left denser than right, inside denser than outside”.

I don’t know if anyone has tested the Golden Ratio Hypothesis on standard computer fonts, though it’s not very hard to write an algorithm for it: Rasterize a font to a fixed size, then add all characters on top of each other, and look where the “blackest” point is within the square.

I don’t have the programming skills to write such an algorithm, but for fun I manually stacked the first 32 characters of the Thousand Character Classic in the default Windows XP font SimSun. I had to stop at 32 characters because it was taking forever.
天地玄黃　宇宙洪荒　日月盈昃　辰宿列張
寒來暑往　秋收冬藏　閏餘成歲　律呂調陽

I’m not sure I understand Qi Gong’s theory correctly, though. So the four points he describes are the four intersections below? How is the bounding square measured?

@Jongseong:

You’re absolutely right that the densest point of concentration is not the same as the center of gravity — measuring the latter requires significantly more sophistication. The algorithm I suggested above was just meant to be a starting point for any systematical analysis.

As for the compound box, I’d say it’s determined optically — how else could it be done for free-hand calligraphy?

Song is probably not the ideal representative, it’d much better to start with Kai. This is the example my father used in his “Introduction to Chinese Calligraphy” seminar (for people who don’t know Chinese):

The typeface is Adobe Kaiti Std here, while my father showcased free-hand samples of the character in different calligraphical styles (Li, Cao, Xing, Kai).

Basically, an element library is developed that is used to construct the glyph. What makes this element library unique is that each element is a multiple master glyph, with up to four design axes.

Very interesting. Although there is also another solution: An Improved Representation for Stroke-based Fonts [PDF]. I wonder which is the best way to design parametric fonts and preserve every character unique…

“Book culture” has been a hot topic in American sinology for a decade, resulting in a number of publications on printing history in China that touch on type. Joseph P. McDermott’s “A Social History of the Chinese Book” discusses issues such as the value of hand-copies vs. wood-block prints, and consequent influence. He also provides details of the transition between: the artist who wrote out the masters for wood-block pages was necessarily literate, but carvers were not; experienced carvers handled the tricky parts of individual characters, leaving the easier strokes for apprentices. In other words, Chinese “type” was industrialized and formalized centuries before becoming movable on a significant scale. The emphasis on practicality is summed up in Lucille Chia’s title, “Printing for Profit.” Nothing too startling in all this — Gutenberg imitated manuscripts, and market factors have long colored printing in the west.

One difference is that the act of writing has retained its value in China, Korea and Japan through its traditional association with scholarship. Like my parents before me, I was taught penmanship in elementary school (in the 1950s), but that disappeared long before my children reached school age. It continues in China. One indication of the separation between type and writing-as-art (“calligraphy”) in the Chinese context surfaces in the common criticism of simplified characters, which often borrow short-hand forms from hand-writing that look inelegant re-cast in type. One of the ironies here is that when typing Chinese you no longer have to draw all the strokes — computerized type addresses the challenging aspect of using traditional characters.

I am very impressed by the technical virtuosity of Ken’s examples, but when I want to see beautiful Chinese writing I look to the calligraphy of Zhao Mengfu — centuries old but still very much alive and capturing the moment of its creation. Different strokes for different purposes.

Computers have obviously been a huge boon to Chinese type — one has only to contemplate mechanical Chinese typewriters. The complexity of CJK characters makes screen resolution a big deal, as James noted already. The less-than-ideal quality of computer displays may ultimately be more influential than any other factor in the acceptance of simplified characters outside of the People’s Republic. But although PRC citizens now comprise the single largest group on the internet, the real action may be happening on telephones. Google “chinese text messaging” for an idea of the importance of tiny-screen type: an account from a couple of years ago claimed 1.2 billion text messages per day. How might this effect the course of Chinese writing?