A Quick Primer On Unicode and Software Internationalization Under Linux and UNIX
The Unicode (R) Consortium is a registered trademark, and Unicode (TM) is a trademark of Unicode, Inc. Linux is a registered trademark of Linus Torvalds. UNIX is a registered trademark of The Open Group. Solaris is a trademark of Sun Microsystems. Windows is a registered trademark of Microsoft Corporation.
by Ed Trager. Last updated: 2013.03.08.ET
This page provides a quick summary of information with links to other URLs regarding using Unicode for multilingual internationalization projects on Linux and other UNIX-based operating systems. If you would like to be able to use more than one language on your Linux/UNIX computer but haven't completely figured out how to do that yet, then you should read this page. I have tested the software and setup configurations mentioned in this document primarily on Linux (SuSE 7.2, 7.3, 8.1, 8.2, 9.0beta) and, to a lesser extent, on OpenBSD (3.2, 3.3) and Solaris 8.
The goals of this document are 1) to introduce you to some indispensible Open Source software for using Unicode in a Linux or other UNIX environment, and 2) to highlight key aspects of setting up that software. Other Unicode web resources cover some of the topics below in much greater depth than here. Instead of being comprehensive, I have tried to focus on a few key pieces of software and key configuration issues that will allow you to quickly become productive on your multilingual or internationalization projects on Linux or other UNIX-based operating systems today. Pointers to more comprehensive treatments of various topics are provided throughout the document.
Computers assign numbers (code points) to represent letters. There are hundreds of national and ISO standards in existence for computer encoding of modern language scripts. Many of these legacy encodings are limited to 256 ( i.e., 28 ) code points. This results in numerous problems. A major problem is that 256 code points is often not enough even for a single language, much less multiple languages. A second rather obvious problem is that code points set aside to represent letters in one national or ISO encoding will unavoidably be re-used to represent completely different letters in some other national or ISO encoding (For example, LATIN SMALL LETTER U WITH GRAVE, "ù" in the Western European ISO-8859-1 encoding becomes LATIN SMALL LETTER U WITH RING ABOVE "ů" in the Central and Eastern European ISO-8859-2 encoding, GREEK SMALL LETTER OMEGA "ω" in ISO-8859-7, HEBREW LETTER SHIN "ש" in ISO-8859-8 ... and so on! For all the gory details, read this). This can easily result in garbled emails, web pages, or databases, among other things.
For an illustration of the same problem in a slightly different domain, consider a familiar language like English. The language can be written with just 26 letters, but publishers of English-language scientific and mathematical documents require many additional symbols --and 256 code points are simply not enough! Imagine how much more problematic electronic information exchange can be for languages like Chinese where multiple, incompatible encodings exist.
Unicode solves the problems of multiple encodings by assigning unique code points to the letters and ideographs of all of the world's modern language scripts and commonly used symbols. The Unicode Consortium's page is at www.unicode.org.
UTF-8 is a serialization method for Unicode that is the de facto standard for encoding Unicode on UNIX-based operating systems, notably Linux. UTF-8 is also the preferred encoding for multi-lingual web pages. In this method, ASCII code points occupy one byte. That is, the ASCII subset of Unicode serialized in UTF-8 is identical to ASCII. Unicode code points in the Basic Multilingual Plane above the ASCII range are serialized to two or three bytes (additional planes exist in Unicode, which can produce serializations of up to six bytes).
When characters are serialized to multiple bytes, the most significant bit is always set, and thus these bytes never fall in the ASCII range. Also, the first byte of a multibyte sequence representing a non-ASCII character always reserves some bits that indicate how many bytes are used for the serialization of this character (Fig. 1).
|Fig. 1. UTF-8. When Unicode characters are serialized to multiple bytes in UTF-8, the high bits of the first serialized byte indicate how many bytes are used for the serialization of that character. The bits represented by "n"s hold the unicode character code value.|
This results in a stateless encoding in which missing bytes will be evident. UTF-8 provides a simple and elegant solution for internationalizing UNIX-based, byte-oriented operating systems and software. For all of the details, read Markus Kuhn's excellent FAQ, UTF-8 and Unicode FAQ for Unix/Linux. All of the software mentioned below supports UTF-8 well.
In order to take full advantage of Unicode on your Linux or other UNIX system, you will need to set your locale to
a UTF-8 locale. Some recent distributions of Linux now default to using a UTF-8 locale by default.
However, unless you are using a very recent Linux distribution, you are still very likely using a legacy locale based
on ISO-8859 or other national encoding.
If you are using some UNIX-based OS other than Linux, it is even less likely that you are already using a UTF-8 locale.
To determine your current locale settings,
locale. Here are some results from Linux and Solaris:
All UTF-8-based locale settings end in "UTF-8", so it is evident that neither user_a nor user_b in
the examples above is using a UTF-8 locale. To determine what other locale settings are available to you, type
It's evident that the Linux distribution (SuSE 7.3 was used for the example) has many UTF-8 locales installed by default (not all are shown), while the Solaris box has none. Solaris does provide UTF-8 locales, but they must be installed as optional packages: see the Solaris Internationalization Guide.
To change your locale setting in Linux, just set the
LANG environment variable in
Note that the output from
locale -a on the Linux box shown above shows
in lower case without a hyphen: this is a BUG. When you set the
be sure to type
UTF-8 in UPPER CASE and with a hyphen:
When you log back in using the new
you should now see that many of the other "
LC_" locale environment
variables have been updated automatically:
Under a UTF-8 locale, you can now take full advantage of Unicode on your machine. Note however, that some Unicode software can be used quite effectively even if you cannot, or are not yet ready, to change to a UTF-8 locale. For example, Yudit, described below, will work just fine even on systems such as OpenBSD which currently do not support different locales.
Even in these days of KDE and Gnome, no Linux or UNIX aficionado would want to live without a good terminal emulator. Several Unicode-enabled terminal emulators are described below.
|Fig. 2. Mlterm. A GUI configurator makes it easy to set up mlterm. An HTML file encoded in UTF-8 is being viewed in vim under mlterm.|
By default, Mlterm uses a bitmap font, usually GNU Unifont which is pre-installed on most Linux distributions and other free Unices. GNU Unifont is the bitmap font shown in Fig. 2 above.
If you want, you can also have Mlterm use an anti-aliased TrueType font.
In this case, a monospaced font like Bitstream Vera Sans Mono or
Everson Mono Unicode is best. You will need (probably as root) to modify Mlterm's
$PREFIX/etc/mlterm/aafont configuration file to indicate which fonts you want to use
to display normal and double-width CJK characters (
$PREFIX depends on where Mlterm is
installed. If you installed it yourself, it is probably
/usr/local/. If Mlterm came
preinstalled, it is probably just
/etc. As an example, here is
aafont file looks like:
ISO10646_UCS4_1=Everson Mono Unicode-iso10646-1; ISO10646_UCS4_1_BIWIDTH=Bitstream Cyberbit-iso10646-1;
This specifies that Everson's sans-serif Everson Mono Unicode font be used
for normal width characters while the serif Bitstream Cyberbit font be used
for double-width CJK characters. In order to toggle anti-aliased fonts, I need to
start Mlterm with the
-A flag, like this:
mlterm -A &
Here is what the result looks like on a Mandrake Linux box:If you want to use a variable-width font, then after modifying
aafontappropriately, start Mlterm like this instead:
mlterm -A -V &
A second alternative is to use xterm (Fig. 3) which is supplied with XFree86. Xterm does not support right-to-left languages like Arabic or Hebrew. I don't think it supports most Indic scripts. It does support Thai though, as shown in Fig. 3 below.
|Fig. 3. Xterm in UTF-8 mode. Xterm supports UTF-8, including Thai, but not right-to-left languages like Arabic.|
For both mlterm and xterm, you will need to set your locale to a UTF-8 locale. When I use an account or a machine where the locale is not a UTF-8 locale, I use the following "mini scripts" for starting mlterm and xterm for multilingual work:
Although I like the features of KDE's Konsole (KDE 3.x), I have noticed annoying bugs when rendering Thai and Arabic scripts under a UTF-8 locale, so I refrain from recommending it for multilingual work at this time.
A number of good unicode editors are now available for Linux/Unix, but here I am only going to describe three:
Yudit (Fig. 4) is an indispensible unicode text editor for the X Window System. Yudit can be used under any locale setting. It can even be used on OpenBSD which lacks locales. The program is extremely easy to use and comes with a large number of keyboard maps -- and even handwriting recognition for Kanji and Hanzi. Handwriting recognition is an excellent idea, but in practice it only works well for fairly simple characters with few strokes, like "人" or "水", as drawing more complex characters with a mouse is too tedious. For serious Chinese, Japanese, or Korean (CJK) typing tasks, an input method engine such as SCIM (discussed below) is required.
Besides the editor itself, the program distribution includes two absolutely indispensible utilities:
uniprintprovides Postscript-based printing.
Uniprintcan be used for printing from within Yudit, or from the shell.
uniconvseamlessly converts files between a large number of Unicode and legacy national and ISO standard encodings.
Information on using
provided below under Utilities.
Program menus are available in many languages. The program has few external dependencies (other than X Windows itself) and there is no need for any pre-installed multi-lingual locale environment. For example, Yudit works perfectly on OpenBSD 3.2 which lacks locale support, where vim fails.
|Fig. 4. Yudit comes with pre-installed keymaps for numerous languages. The program and its accompanying utilities, uniprint and uniconv, are must-have tools in your Unicode toolkit.|
|Fig. 5. Vim running in mlterm with color syntax highlighting for C/C++. In the code shown here, the static C-style strings directly contain UTF-8 encoded locale data.|
There are two keys to making console-mode vim useful for multilingual work. First, you must run vim in a UTF-8-capable terminal emulator like mlterm. Secondly, you are going to need keyboard maps for inputting languages of your choice. Unlike Yudit, numerous standard keymaps do not appear to be distributed with vim.
To determine what keymaps are available, enter the following vim command:
:echo globpath(&rtp, "keymap/*.vim")
This tells you the location of the globally available keymaps, as well as the path where you'll want to place any keymaps that you create if you want to make those maps available to all users.
Setting up and using a keyboard map for vim is not difficult. An excerpt from a Thai keyboard map is shown below. The conventions for naming a keyboard map file are:
So, in this case, the file will be called:
Here's an excerpt from the file:
Comment lines in a keyboard map begin with a quotation mark, '"'. The line,
let b:keymap_name = "thai"" provides a short name for the
map so that we can issue a command in vim to use this map like this:
Everything on the lines following the word "
the keyboard mapping. One or multiple keys can be specified in the first column as the keys to type. One
or more bytes can be specified as the result in the second column.
For example, as shown in the excerpt above, the first six key mappings from the beginning of the top row of a QWERTY keyboard are mapped directly to the Thai characters which have been serialized as UTF-8. Each of these characters actually requires three bytes, but they appear as the Thai characters in your web browser. The fastest way to create a keymap like this is to use Yudit, which is exactly what I did.
The next two entries show an alternate approach: here the unicode code points are entered directly in hexadecimal which can be typed in simple ASCII using any editor you like. The unicode code points for any script can be obtained online in portable document format (PDF) from www.unicode.org/charts/.
After you have created a keyboard map and placed it in the appropriate location (for example,
/usr/share/vim/current/keymap), from within vim simply
...to enable this alternate keymap. When you are in insertion mode, you can toggle between the standard and alternate key maps using CTRL-^.
Finally, just to give you an idea of what else you can do, here a short excerpt from a custom keymap file which uses pinyin romanization to specify some Chinese characters. This example simply demonstrates how a series of multiple keystrokes can be mapped to unicode characters:
Note that you can specify a keymap to use in your .vimrc file, as the following example shows:
For a complete treatment of using Unicode and keymaps, from within vim type:
Mined is a console-mode unicode editor with an intuitive user interface, pull-down menus, extensive Unicode support, including double-width and combining characters, Arabic ligature joining, keyboard mapping, syntax highlighting, and many other features. Mined can be used on UNIX and DOS/Windows platforms.
|Fig. 6. Mined is another unicode editor.|
I have not personally used mined, but it appears to have a nice feature set.
For a more extensive review of Unicode editors, see Alan Wood's summary, Unicode and Multilingual Editors and Word Processors for Unix and Linux.
Keyboard maps are insufficient for typing Chinese, Japanese, and Korean (commonly referred to as "CJK"), as well as other languages such as Tibetan. These languages require sophisticated software input methods (IMs) that operate through XIM. Mike Fabian of SuSE Linux provides an excellent set of pages describing how to set up a CJK computing environment on your Linux box in which he includes descriptions and setup details for a number of IM engines. One of the best among the set of Open Source IM engines is Smart Common Input Method (SCIM) , which I describe below.
|James Su's Smart Common Input Method (SCIM) is a Unicode-based IM platform written in C++. For users, SCIM is an excellent choice because it is simple to set up and use in a UTF-8 or legacy locale. For software developers, it is also nice because it abstracts input method interfaces into a set of simple, independent classes so developers can write their own input methods easily in a few lines of code.|
|Fig. 7. SCIM is an excellent IM application with support for a number of CJK input methods, including 自然码zìránmǎ which is shown being used to enter Chinese for a Google search in Mozilla.|
SCIM currently provides input tables for at least the following methods:
Of the numerous Chinese input methods available, intelligent pinyin and ziranma are the easiest to use. The keyboard layout and a description of how to use the 自然碼 zìránmǎ, or 自然双拼 zìrán shuāngpīn, method can be found here. Note that the intelligent pinyin method is closed-source software, but you can install a binary RPM version for use with SCIM free of charge. If you are compiling from source, I think you will find the supplied 自然碼 zìránmǎ method quite satisfactory.
SCIM requires atk-1.0+, glib 2.0+, pango-1.0+, and gtk+2.0+. These libraries will be present in newer Linux distributions, or you can download them from the GTK+ site here.
After compiling SCIM, you will need to add the following lines to your
.xinitrc file in order to have SCIM start
whenever you start X windows:
If you are using an older version of SCIM (prior to version 0.8.0) and are not already running in a Chinese, Japanese, or
Korean locale, then you will need to set the
variable to refer to a Chinese, Japanese, or Korean locale in your
Note that you can do this even if your
environment variable is set to another (UTF-8) locale, such as English, as shown in the
example below. Versions of SCIM after v. 0.8.0 will work fine with
set to any UTF-8 locale.
|Mutt is an excellent email agent with good UTF-8 Unicode support. Mutt can also be extensively customized to meet your individual needs. For example, a very simple customization is to have emails from certain people, domains, or mailing lists highlighted in special colors in the message index. An example of this is shown in the image on the left below (Fig. 8). Another feature I like about Mutt is that you can use any editor you want to compose your emails. I have Mutt set to use Yudit for composing email in UTF-8. A more common choice is to use Vim as the editor. Once you have used Mutt a few times and started to play around with its various configuration options, you'll never go back to using another email agent again!|
|Fig. 8. Mutt is an excellent email agent. Once you have used Mutt a few times and customized it to your liking, you'll never want to use any other email agent! Left side: Message index displayed in Mutt with color customizations running in a KDE konsole terminal. Right side: Mutt displaying a UTF-8 encoded message in mlterm.|
This section lists conversion and printing utilities.
For converting a file from one encoding to another, three utilities are worth mentioning here:
iconvwhich is part of GNU libc (and hence probably already on your system).
uniconvwhich is distributed with Yudit
Usage of these utilities is described below.
GNU iconv has knowledge of a huge number of encodings, although the authors warn that "this does not necessarily mean that all combinations of these names can be used for the FROM and TO command line parameters". To get a list of all known encodings, type:
Usage is as follows:
Uniconv is distributed with Yudit and understands a useful range of internal and external encodings. For a complete list, type:
Usage is shown below:
Perl 5.8+ is distributed with piconv. Usage is quite similar to (but in some ways more intelligent than)
GNU iconv. The man page for piconv provides details.
provides a listing of over 120 recognized encodings, some of which are aliases of one another.
Convmv is a utility for converting file names in directory trees from one encoding to another (for example, from a legacy encoding into a UTF-8 encoding). Only the file names are converted: file contents remain unchanged. The man page for convmv may be found here.
For printing Unicode text or data files, there is
which is distributed with the Yudit package. Another, lesser-known utility is called
paps. If you are a developer, also take a look at
LASi, a C++ stream-based library for printing Postscript.
Typical usage is shown below:
Paps is a UTF-8-to-Postscript converter available from http://imagic.weizmann.ac.il/~dov/freesw/paps/. Paps has options for printing in landscape, and for printing multiple columns on one page which could be quite useful.
Also see the LASi Postscript printing library discussed below.
Because of the rapid maturation of Open Source software in the last few years, it is now possible to enjoy a Linux desktop experience complete with high-quality, anti-aliased fonts rivaling what is achievable on a Mac or Windows PC. Unicode fonts are naturally an important part of creating an internationalized, multilingual desktop. To create such an environment, you need to know:
These topics are covered below.
If you are running KDE in a recent distribution of Linux, the easiest way to install fonts is to use KDE's graphical font installer which is part of the Control Center. From the KDE menu, select Control Center, System Administration, and then Font Installer. To install fonts for everyone, click on Adminstrative Mode and enter the root password. Then simply click the Add button to add the fonts that you want (Fig. 9). When you are done, select Apply. KDE will run all the scripts necessary to update your machine's font configuration.
|Fig. 9. KDE's Font Installer makes it easy to install Unicode TrueType fonts in Linux.|
Modern Linux distributions use the fontconfig system for font management. Fontconfig periodically
scans font directories that are listed in the global
/etc/fonts/fonts.conf and user-specific
~/.fonts.conf configuration files.
All users have to do is put new font files in one of the known directories and the fonts become automagically available for use.
This is much easier than it was in the old days!
For example, on SuSE Linux and many other systems that use the X windowing system, TrueType and OpenType fonts are stored in
/usr/X11R6/lib/X11/fonts/truetype. As this path is longer than I wish to remember, on systems that
I manage I take advantage of a symbolic link to simplify my life:
~> su password: ****** ~> cd / ~> ln -s /usr/X11R6/lib/X11/fonts/truetype/ /fonts
Now installing a new font for system-wide use from my home directory is nothing more than:
~> su -c "mv my_new_font.otf /fonts" password: ******
And installing a font for personal use only is just:
~> mv my_new_font.otf .fonts
The Unicode Font Guide For Free/Libre Open Source Operating Systems is a concise guide to free and legally downloadable vector (TrueType and OpenType) fonts appropriate for use on Open Source operating systems. The Font Guide provides links to a selection of high-quality Unicode fonts for most of the world's scripts encoded in Unicode.
For a more comprehensive treatment of fonts, including many commercial fonts which are not included in the Unicode Font Guide mentioned above, see Alan Wood's font pages.
Pfaedit is an Open Source font editor for Linux/UNIX which allows you to create your own Postscript, TrueType, OpenType, cid-keyed and bitmap (bdf) fonts. You can also edit existing fonts and convert font files to different formats.
Even if you don't plan on creating your own fonts, Pfaedit is extremely useful for determining which Unicode blocks, and hence which scripts, are covered in a Unicode font. It is also indispensible for viewing the forms of glyphs before installing a font. The font previews provided in KDE's font installer and other programs are usually completely inadequate since generally only the ASCII or Latin blocks are shown. For non-Latin scripts, Pfaedit is much more useful as a font preview program. For a quick preview of all glyphs in a font, be sure to select View --> Compacted View. The default Encoded View will display cells for all Unicode code points, many of which may be undefined in a given font.
|Fig. 10. Pfaedit is an Open Source font editor for Linux.|
|The recently released Open Office version 1.1 has excellent internationalization features, including support for complex text layout (CTL), right-to-left (RTL), and bidirectional algorithm (BiDi) support. Open Office also provides excellent support for Microsoft document formats, export to PDF, XML support, and many other features. Open Office localizations exist for many languages and locales, including Czech, Dansk, Deutsch, Español, Français, Hindi, Italiano, Japanese, Nederlands, Português do Brasil, Limba Romana, Suomi, Thai and Türkçe.|
|IBM's Open Source International Components for Unicode libraries provide robust Unicode services, including text handling, Unicode regular expression analysis, language-sensitive collation, and data and formatting rules for over 200 locales. The libraries are available in C, C++, and Java. Even if you don't plan to use the ICU libraries directly, the locale data by itself is an extremely useful resource. This is an incredible resource that should not be overlooked!|
Pango provides an Open Source library for the layout and rendering of internationalized text. Pango was started by Owen Taylor of Redhat in order to provide multilingual, Unicode-based layout services for GTK+ and GNOME. Pango is not however dependent on GTK+ or GNOME and can be used in other projects as well.
FreeType 2 is an open source font engine. It provides an easy-to-use API to access font content in a uniform way, independent of the file format. Client programs can access glyph outline data, or make use of the FreeType rasterizer to produce monochrome or anti-aliased bitmaps.
An extremely interesting feature of the FreeType 2 library is that it contains an "autohinter" in order to avoid infringing three patents owned by Apple Computer, Inc. If you have paid for a license from Apple, you can compile FreeType so that it will use the patented font hinting technology. Fortunately for Linux, and for everyone else who doesn't want to purchase a license from Apple, FreeType can also be compiled to use the "autohinter" which has been specifically crafted to avoid infringing the patents while still producing acceptable bitmaps of glyphs at small pixel sizes.
LASi is a new Postscript printing library initially developed by a friend of mine, Larry Siden and now being maintained here on the eyegene web site. It uses Pango and Freetype2 for Unicode text layout and glyph rendering.
It was designed primarily for use in scientific and other programs (such as my program Madeline) which need the ability to print scientific and other non-Latin symbols and scripts in Postscript documents without being tied to large GUI application framework libraries like QT, KDE, or GTK+/Gnome.
As shown in figure to the right and in the example below,
LASi provides a simple C++ stream-based API which
isolates the developer from the complex
details of printing Unicode text, including right-to-left scripts
(Hebrew, Arabic) and scripts which have
complex shaping rules (Arabic, Thai, Devenagari). In addition
to world scripts, users can of course also take advantage of
the numerous scientific and mathematical symbols defined in
Unicode by simply supplying UTF-8-encoded strings to the
You can now find out more about LASi and obtain Up-to-date versions of the library from http://eyegene.ophthy.med.umich.edu/lasi/.