Home > Articles > Programming

Unicode Architecture: Not Just a Pile of Code Charts

  • Print
  • + Share This
Richard Gillam examines the three rules of Unicode standard: If you receive text from the outside world and pass it back to the outside world, don't mess it up; follow the rules in the Unicode standard that are relevant to the particular character you claim to support; and if you produce output that purports to be Unicode text, another Unicode-conformant process should be able to interpret it properly.
This chapter is from the book

If you're used to working with ASCII or other similar encodings designed for European languages, you'll find Unicode noticeably different from those other standards. You'll also find that when you're dealing with Unicode text, various assumptions you may have made in the past about how you deal with text don't hold. If you've worked with encodings for other languages, at least some characteristics of Unicode will be familiar to you, but even then, some pieces of Unicode will be unfamiliar.

Unicode is more than just a big pile of code charts. To be sure, it includes a big pile of code charts, but Unicode goes much further. It doesn't just take a bunch of character forms and assign numbers to them; it adds a wealth of information on what those characters mean and how they are used.

Unlike virtually all other character encoding standards, Unicode isn't designed for the encoding of a single language or a family of closely related languages. Rather, Unicode is designed for the encoding of all written languages. The current version doesn't give you a way to encode all written languages (and in fact, this concept is such a slippery thing to define that it probably never will), but it does provide a way to encode an extremely wide variety of languages. The languages vary tremendously in how they are written, so Unicode must be flexible enough to accommodate all of them. This fact necessitates rules on how Unicode is to be used with each language. Also, because the same encoding standard can be used for so many different languages, there's a higher likelihood that they will be mixed in the same document, requiring rules on how text in the different languages should interact. The sheer number of characters requires special attention, as does the fact that Unicode often provides multiple ways of representing the same thing.

The idea behind all the rules is simple: to ensure that a particular sequence of code points will get drawn and interpreted the same way (or in semantically equivalent ways) by all systems that handle Unicode text. In other words, it's not so important that there should be only one way to encode " bient?," but rather a particular sequence of code points that represents " bient?" on one system will also represent it on any other system that purports to understand the code points used. Not every system has to handle that sequence of code points in exactly the same way, but merely must interpret it as meaning the same thing. In English, for example, you can follow tons of different typographical conventions when you draw the word "carburetor," but someone who reads English would still interpret all of them as the word "carburetor." Any Unicode-based system has wide latitude in how it deals with a sequence of code points representing the word "carburetor," as long as it still treats it as the word "carburetor."

As a result of these requirements, a lot more goes into supporting Unicode text than supplying a font with the appropriate character forms for all the characters. The purpose of this book is to explain all these other things you have to be aware of (or at least might have to be aware of). This chapter will highlight the things that are special about Unicode and attempt to tie them together into a coherent architecture.

The Unicode Character–Glyph Model

The first and most important thing to understand about Unicode is what is known as the character–glyph model. Until the introduction of the Macintosh in 1984, text was usually displayed on computer screens in a fairly simple fashion. The screen would be divided up into a number of equally sized display cells. The most common video mode on the old IBM PCs, for example, had 25 rows of 80 display cells each. A video buffer in memory consisted of 2,000 bytes, one for each display cell. The video hardware contained a character generator chip that contained a bitmap for each possible byte value, and this chip was used to map from the character codes in memory to a particular set of lit pixels on the screen.

Handling text was simple. There were 2,000 possible locations on the screen, and 256 possible characters to put in them. All the characters were the same size, and were laid out regularly from left to right across the screen. There was a one-to-one correspondence between character codes stored in memory and visible characters on the screen, and there was a one-to-one correspondence between keystrokes and characters.

We don't live in that world anymore. One reason is the rise of the WYSIWYG ("what you see is what you get") text editor, where you can see on the screen exactly what you want to see on paper. With such a system, video displays have to be able to handle proportionally spaced fonts, the mixing of different typefaces, sizes, and styles, and the mixing of text with pictures and other pieces of data. The other reason is that the old world of simple video display terminals can't handle many languages, which are more complicated to write than the Latin alphabet is.

As a consequence, there's been a shift away from translating character codes to pixels in hardware and toward doing it in software. And the software for doing this has become considerably more sophisticated.

On modern computer systems (Unicode or no), there is no longer always a nice, simple, one-to-one relationship between character codes stored in your computer's memory and actual shapes drawn on your computer's screen. This point is important to understand because Unicode requires this flexibility—a Unicode-compatible system cannot be designed to assume a one-to-one correspondence between code points in the backing store and marks on the screen (or on paper), or between code points in the backing store and keystrokes in the input.1

Let's start by defining two concepts: character and glyph. A character is an atomic unit of text with some semantic identity; a glyph is a visual representation of that character.

Consider the following examples:

These forms are 11 different visual representations of the number 13. The underlying semantic is the same in every case: the concept "thirteen." These examples are just different ways of depicting the concept of "thirteen."

Now consider the following:

Each of these examples is a different presentation of the Latin lowercase letter g. To go back to our terms, these are all the same character (the lowercase letter g), but four different glyphs.

Of course, these four glyphs were produced by taking the small g out of four different typefaces. That's because there's generally only one glyph per character in a Latin typeface. In other writing systems, however, that isn't true. The Arabic alphabet, for example, joins cursively even when printed. This isn't an optional feature, as it is with the Latin alphabet; it's the way the Arabic alphabet is always written.

These are four different forms of the Arabic letter heh. The first shows how the letter looks in isolation. The second depicts how it looks when it joins only to a letter on its right (usually at the end of a word). The third is how it looks when it joins to letters on both sides in the middle of a word. The last form illustrates how the letter looks when it joins to a letter on its left (usually at the beginning of a word).

Unicode provides only one character code for this letter,2 and it's up to the code that draws it on the screen (the text rendering process) to select the appropriate glyph depending on context. The process of selecting from among a set of glyphs for a character depending on the surrounding characters is called contextual shaping, and it's required to draw many writing systems correctly.

There's also not always a one-to-one mapping between character and glyph. Consider the following example:

This, of course, is the letter f followed by the letter i, but it's a single glyph. In many typefaces, if you put a lowercase f next to a lowercase i, the top of the f tends to run into the dot on the i, so the typeface often includes a special glyph called a ligature that represents this particular pair of letters. The dot on the i is incorporated into the overhanging arch of the f, and the crossbar of the f connects to the serif on the top of the base of the i. Some desktop-publishing software and some high-end fonts will automatically substitute this ligature for the plain f and i.

In fact, some typefaces include additional ligatures. Other forms involving the lowercase f are common, for example. You'll often see ligatures for a–e and o–e pairs (useful for looking erudite when using words like "arch?ology" or "?nophile"), although software rarely forms these automatically (? and ? are actually separate letters in some languages, rather than combinations of letters), and some fonts include other ligatures for decorative use.

Again, though, ligature formation isn't just a gimmick. Consider the Arabic letter lam () and the Arabic letter alef (). When they occur next to each other, you'd expect them to appear like this if they followed normal shaping rules:

Actually, they don't. Instead of forming a U shape, the vertical strokes of the lam and the alef actually cross, forming a loop at the bottom:

Unlike the f and i in English, these two letters always combine this way when they occur together. It's not optional. The form that looks like a U is just plain wrong. So ligature formation is a required behavior for writing many languages.

A single character may also split into more than one glyph. This happens in some Indian languages, such as Tamil. It's very roughly analogous to the use of the silent e in English. The e at the end of "bite," for example, doesn't have a sound of its own; it merely changes the way the i is pronounced. Since the i and the e are being used together to represent a single vowel sound, you could think of them as two halves of a single vowel character. Something similar happens in languages like Tamil. Here's an example of a Tamil split vowel:

This looks like three letters, but it's really only two. The middle glyph is a consonant, the letter . The vowel is shown with a mark on either side of the . This kind of thing is required for the display of a number of languages.

As we see, there's not always a simple, straightforward, one-to-one mapping between characters and glyphs. Unicode assumes the presence of a character rendering process capable of handling the sometimes complex mapping from characters to glyphs. It doesn't provide separate character codes for different glyphs that represent the same character, or for ligatures representing multiple characters.

Exactly how this process works varies from writing system to writing system (and, to a lesser degree, from language to language within a writing system). For the details on just how Unicode deals with the peculiar characteristics of each writing system it encodes, see Part II (Chapters 7 to 12).

  • + Share This
  • 🔖 Save To Your Account