JUnidecode - A Unicode to ASCII Java Library

If you already know what is Unicode and you are sure that you know that you want to convert it to 7-bit ASCII and of the downside that it has, you can skip the chit-chat and jump to JUnidecode description or download section.

This page contains Unicode characters. If the font that you are using doesn’t contain all the characters you may get a garbled output. In this case try to install and use a font with a good Unicode support like DejaVu.

What is Unicode

The characters that you are reading right now, i.e. the latin alphabet’s characters, are not as common as one may think. Summing up the number of native speakers of the top 20 most spoken langueges of the world it comes up that almost 3100 million people (source) use a language that doesn’t contain even a single latin character; for example Chinese, Hindi, Arabic, Bengali, Russian and so on. Except for English all the languages that use the latin alphabet “enrich” it by using diacritic marks.

Unicode was invented to represent and manipulate all the different characters not included in the traditional 7-bit ASCII encoding. Unicode assigns to each character a unique so called “code point”. For example the letter “a” has as code point U+0061, while “Я”’s code point is U+042F.

Unicode’s code points are just a standardized way to say: “I mean that letter”, but Unicode doesn’t say how you should encode the code point. Result is, of course, that there are many different ways to encode Unicode like UTF-8, UTF-7 or UCS2, the most common being probably UTF-8.

For a nice article about what you should know about Unicode as programmer read this article by Joel Spolsky: The Absolute Minimum Every Software Developer Absolutely, Positively Must Know About Unicode and Character Sets (No Excuses!).

Java’s String implementation internally use UTF-16, but we can get the encoding for many other charsets using the method getBytes(String charsetName). Most important, we can ask: “what is the code point of the character at index x?” (codePointAt(int index)).

How to strip diacritic marks in Java

Unicode is nice and good. Why should one ever want to strip diacritic marks? There are some situations where it’s sensible to do so.

Let’s assume you’re writing a software for a multinational industry to manage its employees. Since it’s a multinational it has employees from all around the world with exotic (invented) names like “Franco Lorè” or “Stjepan Bebić”. The management will be very displeased when it will discover that if they look for “Bebic” they won’t find Stjepan.

Another example: you’re managing a travel agency web site. A customer logs in, looks for travel offers for “cote d’Azur” and then goes away because your web site knows nothing about “cote d’Azur”, it just knows “côte d’Azur”.

So: you should use Unicode whenever it’s possible, but you should also know when “dumb it down”.

If you use Java 6 or above you can use the java.text.Normalizer class:

String normalized = java.text.Normalizer.normalize(originalstring, java.text.Normalizer.Form.NFD)
String accentsgone = normalized.replaceAll("\\p{InCombiningDiacriticalMarks}+", "");

The first line applies the so called “canonical decomposition”, which takes a string and recursively replaces composite characters using the Unicode canonical decomposition mappings. For example the character “à” is decomposed in a (a, `). For more informations refer to Unicode Standard Annex 15.

The second line uses a regular expression that matches all the diacritic marks that can be combined to characters (like a grave accent that combined with “a” creates “à”) and replaces them with an empty string. Being able to specify a Unicode block as pattern in a regular expression is a rarely used feature, but it can come pretty handy when working with i18n and l10n. You can find more details in Java’s java.util.regex.Pattern documentation. The list of Unicode blocks is available at the address http://www.unicode.org/Public/UNIDATA/Blocks.txt.

Applying this code to “Bebić” returns “Bebic”.

Another option is to use IBM’s ICU library, which offers a larger support to Unicode functionality than Java. The downside is that the JAR of this library is about 6MiB, and can be quite cumbersome if one needs only some basic functionality. However if you are using this library you can remove diacritic marks in this way:

Transliterator accentsconverter = com.ibm.icu.text.Transliterator.getInstance("NFD; [:Nonspacing Mark:] Remove; NFC.");
String accentsgone = accentsconverter.transliterate(originalstring);

This code is more refined than the previous one. It first applies a canonical decomposition, then removes the diacritic marks and then applies a canonical composition.

Both techniques remove diacritic marks and transform, for example, “côté” in “cote” and “Ελληνικά” in “Ελληνικα”. Wait, what’s that? If you don’t have a greek keyboard you should use a lot of alt+[numeric code] to write that word.

Unicode to ASCII: why and how

Let’s say that you’ve got an e-mail from Mr. まさゆき たけだ. Assuming that you do not speak japanese and you don’t know what “hiragana” is, how are you going to add this person to your address book? And, even more important: how are you going to retrieve him?. Sometimes it’s worse: a lot of software doesn’t know how to handle characters that do not belong to the usual blocks, so they mangle the output and show just a bunch of question marks, so the e-mail is from Mr. ���� ���, which is even less useful than “weird-stuff-that-I-can’t-read”. Sometimes they fail gracefully and show something like “[307E][3055][3086][304D] [305F][3051][3060]”.

Going back to our employees example: if you strip diacritic marks from Łukasiński you get “Łukasinski” which is slightly better, but still your boss won’t be able to find your polish colleague.

If you are indexing text documents and you are from a western country, you’ll hardly find the documents about China’s capital 北亰.

It turns out that sometimes just stripping out characters is not enough: sometimes you need to transliterate words from an alphabet to another. When the destination alphabet is the latin one this process is called romanization. So in the first case the e-mail sender is Mr. Masayuki Takeda, your boss will find Mr. Łukasiński even if she looks for Lukasinski and your documents about Bei Jing will be easier to retrieve.

IBM’s ICU library offers a great transliteration functionality (ICU’s transform documentation), but has some downsides:

For a little project I needed a simple and fast function to transliterate arbitrary Unicode strings to ASCII without hassle, and so JUnidecode was born. The philosophy it follows it’s that for transliteration “something, even wrong, is better than nothing”.

JUnidecode

JUnidecode is a Java port of Text::Unidecode perl module by Sean M. Burke, who deserves all the credit. Ça va sans dire that any bug in JUnicode is all my fault.

JUnidecode official web site is http://junidecode.sf.net

JUnidecode has only one class, net.sf.junidecode.Junidecode, which, as of version 0.1, has only one method: public String unidecode(String s). It takes a String and transliterates it to a valid 7-bit ASCII String (obviously it also strips diacritic marks). For example:

import net.sef.junidecode;

// ...

// s = "résumé"
String s = "r\u00E9sum\u00E9";
System.out.print(Junidecode.unidecode(s));
// prints "resume"

This code can be made less verbose using the static import feature available on Java 5.0 and higher:

import static net.sf.junidecode.Junidecode.*;

// ...

// s = "résumé"
String s = "r\u00E9sum\u00E9";
System.out.print(unidecode(s));
// prints "resume"

This is a little example of JUnidecode capabilities:

Unicode block Original text Transliteration Notes
Basic Latin JUnidecode. JUnidecode.
Latin-1 Supplement résumé resume All diacritic marks are always correctly stripped.
Latin Extended-A čeština cestina
Armenian Հայաստան Hayastan
Cyrillic Москвa Moskva
Greek and Coptic Ελληνικά Ellenika
Ethiopic ኣዲስ ኣበባ aadise aababaa Ethiopia’s capital city Addis Ababa isn’t correctly transliterated
Arabic هاتف htf Expected: hatif. Arabic is extremely hard to transliterate without an intelligent knowledge of the language
Hebrew לשון הקודש lshvn hqvdsh Missing vowels, as expected. The correct transliteration is “Leshon HaKodesh”.
Syriac ܠܫܢܐ ܣܘܪܝܝܐ lshn’ swryy’ Expected: lessana suryaya
Bengali বাংলা baaNlaa Expected: bangla
Devanagari देवनागरी devnaagrii Expected: devanagari
Gurmukhi ਗੁਰਮੁਖੀ gurmukhii Expected: gurmukhi
Malayalam മലയാളം mlyaallN Expected: malayalam
Tamil தமிழ் tamilll Expected: tamil
Thai ราชอาณาจักรไทย raach`aanaacchakraithy Expected: ratcha anachak thai.
CJK Unified Ideographs 北亰 Bei Jing
Unified Canadian Aboriginal Syllabics ᐌᐘᑯ wewako Expected: Ewako (Êwako)
Currency Symbols ₤ ₧ ₨ L Pts Rs
Braille Patterns ⠏⠗⠑⠍⠊⠑⠗ premier Braille patterns are substituted with matching latin characters.
Number Forms ⅔ Ⅷ 2/3 VIII

Unidecode mapping tables are largely based on Text::Unidecode’s tables. However was released in 2001, and meanwhile Unicode was extended. Most notably JUnidecode supports the transliteration of most characters from Unicode’s letterlike symbols block, which includes among the others ℕ, ℝ, Ω (which represents Ohm and is different from Ω, which is the greek capital letter omega), ℏ and other lovely nerdy characters.

Download

Several download packages are available. JUnidecode is self-contained and doesn’t need any additional library to work (except for the unit tests, which are based on JUnit).

Please visit the download page of JUnidecode.

License

JUnidecode is licensed under BSD license, which is fairly permissive, compatible with both commercial and free (as in speech software) approved by OSI. You aren’t forced, buy if you modify JUnidecode I’d be happy to hear about it.

Copyright (c) 2009-2014, Giuseppe Cardone. All rights reserved.

Redistribution and use in source and binary forms, with or without modification, are permitted provided that the following conditions are met:

THIS SOFTWARE IS PROVIDED BY GIUSEPPE CARDONE ‘’AS IS’’ AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL GIUSEPPE CARDONE BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.

Credits

JUnidecode is a Java port of Text::Unidecode perl module by Sean M. Burke, who deserves all the credit (while I deserve all the blame for any error in JUnidecode).

There is a beautiful article by Burke himself that explains the design choices behind “unidecoding”: http://interglacial.com/~sburke/tpj/as_html/tpj22.html. Basically there are two idea behind this software:

  1. Writing a context aware transliteration library is an almost infeasible task. This is due to the complexity of many languages, for example Thai writing runs all togheter, in Arab a character should be transliterated in different ways based on the gender or other characteristics of the preceding word. So the first choice is: when a character can be transliterated in different ways, transliterate it to the way that is right most of the time.
  2. The same character can be transliterated in different ways in different languages, for example the same ideogram has a different transliteration in Mandarin, Japanese and Korean. The choice in this case is: try to please most of the people (in the case of ideogram Chinese transliteration is always preferred since more than 1 billion people speak it and it is usually the root from which other languages derived).