It often happens that you have text data in Unicode, but you need to represent it in ASCII. For example when integrating with legacy code that doesn't support Unicode, or for ease of entry of non-Roman names on a US keyboard, or when constructing ASCII machine identifiers from human-readable Unicode strings that should still be somewhat intelligible. A popular example of this is when making an URL slug from an article title.
Unidecode is not a replacement for fully supporting Unicode for strings in your program. There are a number of caveats that come with its use, especially when its output is directly visible to users. Please read the rest of this README before using Unidecode in your project.
In most of examples listed above you could represent Unicode characters as
??? or \\15BA\\15A0\\1610, to mention two extreme cases. But that's
nearly useless to someone who actually wants to read what the text says.
What Unidecode provides is a middle road: the function unidecode() takes
Unicode data and tries to represent it in ASCII characters (i.e., the
universally displayable characters between 0x00 and 0x7F), where the
compromises taken when mapping between two character sets are chosen to be
near what a human with a US keyboard would choose.
The quality of resulting ASCII representation varies. For languages of western origin it should be between perfect and good. On the other hand transliteration (i.e., conveying, in Roman letters, the pronunciation expressed by the text in some other writing system) of languages like Chinese, Japanese or Korean is a very complex issue and this library does not even attempt to address it. It draws the line at context-free character-by-character mapping. So a good rule of thumb is that the further the script you are transliterating is from Latin alphabet, the worse the transliteration will be.
Generally Unidecode produces better results than simply stripping accents from characters (which can be done in Python with built-in functions). It is based on hand-tuned character mappings that for example also contain ASCII approximations for symbols and non-Latin alphabets.
Note that some people might find certain transliterations offending. Most common examples include characters that are used in multiple languages. A user expects a character to be transliterated in their language but Unidecode uses a transliteration for a different language. It's best to not use Unidecode for strings that are directly visible to users of your application. See also the Frequently Asked Questions section for more info on common problems.
This is a Python port of Text::Unidecode Perl module by Sean M. Burke
<[email protected]>.
The module exports a function that takes an Unicode object (Python 2.x) or string (Python 3.x) and returns a string (that can be encoded to ASCII bytes in Python 3.x):
>>> from unidecode import unidecode
>>> unidecode('ko\u017eu\u0161\u010dek')
'kozuscek'
>>> unidecode('30 \U0001d5c4\U0001d5c6/\U0001d5c1')
'30 km/h'
>>> unidecode('\u5317\u4EB0')
'Bei Jing '
You can also specify an errors argument to unidecode() that determines
what Unidecode does with characters that are not present in its transliteration
tables. The default is 'ignore' meaning that Unidecode will ignore those
characters (replace them with an empty string). 'strict' will raise a
UnidecodeError. The exception object will contain an index attribute that
can be used to find the offending character. 'replace' will replace them
with '?' (or another string, specified in the replace_str argument).
'preserve' will keep the original, non-ASCII character in the string. Note
that if 'preserve' is used the string returned by unidecode() will not
be ASCII-encodable!:
>>> unidecode('\ue000') # unidecode does not have replacements for Private Use Area characters
''
>>> unidecode('\ue000', errors='strict')
Traceback (most recent call last):
...
unidecode.UnidecodeError: no replacement found for character '\ue000' in position 0
A utility is also included that allows you to transliterate text from the command line in several ways. Reading from standard input:
$ echo hello | unidecode hello
from a command line argument:
$ unidecode -c hello hello
or from a file:
$ unidecode hello.txt hello
The default encoding used by the utility depends on your system locale. You can
specify another encoding with the -e argument. See unidecode --help for
a full list of available options.
Nothing except Python itself. Unidecode supports Python 2.7 and 3.4 or later.
You need a Python build with "wide" Unicode characters (also called "UCS-4 build") in order for Unidecode to work correctly with characters outside of Basic Multilingual Plane (BMP). Common characters outside BMP are bold, italic, script, etc. variants of the Latin alphabet intended for mathematical notation. Surrogate pair encoding of "narrow" builds is not supported in Unidecode.
If your Python build supports "wide" Unicode the following expression will return True:
>>> import sys >>> sys.maxunicode > 0xffff True
See PEP 261 for details regarding support for "wide" Unicode characters in Python.
To install the latest version of Unidecode from the Python package index, use these commands:
$ pip install unidecode
To install Unidecode from the source distribution and run unit tests, use:
$ python setup.py install $ python setup.py test
- German umlauts are transliterated incorrectly
- Latin letters "a", "o" and "u" with diaeresis are transliterated by
Unidecode as "a", "o", "u", not according to German rules "ae", "oe",
"ue". This is intentional and will not be changed. Rationale is that these
letters are used in languages other than German (for example, Finnish and
Turkish). German text transliterated without the extra "e" is much more
readable than other languages transliterated using German rules. A
workaround is to do your own replacements of these characters before
passing the string to
unidecode(). - Japanese Kanji is transliterated as Chinese
- Same as with Latin letters with accents discussed in the answer above, the Unicode standard encodes letters, not letters in a certain language or their meaning. With Japanese and Chinese this is even more evident because the same letter can have very different transliterations depending on the language it is used in. Since Unidecode does not do language-specific transliteration (see next question), it must decide on one. For certain characters that are used in both Japanese and Chinese the decision was to use Chinese transliterations. If you intend to transliterate Japanese, Chinese or Korean text please consider using other libraries which do language-specific transliteration, such as Unihandecode.
- Unidecode should support localization (e.g. a language or country parameter, inspecting system locale, etc.)
- Language-specific transliteration is a complicated problem and beyond the scope of this library. Changes related to this will not be accepted. Please consider using other libraries which do provide this capability, such as Unihandecode.
- Unidecode should automatically detect the language of the text being transliterated
- Language detection is a completely separate problem and beyond the scope of this library.
- Unidecode should use a permissive license such as MIT or the BSD license.
- The maintainer of Unidecode believes that providing access to source code on redistribution is a fair and reasonable request when basing products on voluntary work of many contributors. If the license is not suitable for you, please consider using other libraries, such as text-unidecode.
- Unidecode produces completely wrong results (e.g. "u" with diaeresis transliterating as "A 1/4 ")
- The strings you are passing to Unidecode have been wrongly decoded
somewhere in your program. For example, you might be decoding utf-8 encoded
strings as latin1. With a misconfigured terminal, locale and/or a text
editor this might not be immediately apparent. Inspect your strings with
repr()and consult the Unicode HOWTO. - I've upgraded Unidecode and now some URLs on my website return 404 Not Found.
- This is an issue with the software that is running your website, not
Unidecode. Occasionally, new versions of Unidecode library are released
which contain improvements to the transliteration tables. This means that
you cannot rely that
unidecode()output will not change across different versions of Unidecode library. If you useunidecode()to generate URLs for your website, either generate the URL slug once and store it in the database or lock your dependency of Unidecode to one specific version.
Some of the issues in this section are discussed in more detail in this blog post.
By default, unidecode() optimizes for the use case where most of the strings
passed to it are already ASCII-only and no transliteration is necessary (this
default might change in future versions).
For performance critical applications, two additional functions are exposed:
unidecode_expect_ascii() is optimized for ASCII-only inputs (approximately
5 times faster than unidecode_expect_nonascii() on 10 character strings,
more on longer strings), but slightly slower for non-ASCII inputs.
unidecode_expect_nonascii() takes approximately the same amount of time on
ASCII and non-ASCII inputs, but is slightly faster for non-ASCII inputs than
unidecode_expect_ascii().
Apart from differences in run time, both functions produce identical results. For most users of Unidecode, the difference in performance should be negligible.
You can get the latest development version of Unidecode with:
$ git clone https://www.tablix.org/~avian/git/unidecode.git
There is also an official mirror of this repository on GitHub at https://github.com/avian2/unidecode
Please make sure to read the Frequently asked questions section above before contacting the maintainer.
Bug reports, patches and suggestions for Unidecode can be sent to [email protected].
Alternatively, you can also open a ticket or pull request at https://github.com/avian2/unidecode
Original character transliteration tables:
Copyright 2001, Sean M. Burke <[email protected]>, all rights reserved.
Python code and later additions:
Copyright 2021, Tomaz Solc <[email protected]>
This program is free software; you can redistribute it and/or modify it under the terms of the GNU General Public License as published by the Free Software Foundation; either version 2 of the License, or (at your option) any later version.
This program is distributed in the hope that it will be useful, but WITHOUT ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for more details.
You should have received a copy of the GNU General Public License along with this program; if not, write to the Free Software Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA. The programs and documentation in this dist are distributed in the hope that they will be useful, but without any warranty; without even the implied warranty of merchantability or fitness for a particular purpose.