Skip to main content

ASCII transliterations of Unicode text

Project description

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).

In most of these examples 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: 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.

Note that this module generally 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.

This is a Python port of Text::Unidecode Perl module by Sean M. Burke <sburke@cpan.org>.

Module content

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(u'ko\u017eu\u0161\u010dek')
'kozuscek'
>>> unidecode(u'30 \U0001d5c4\U0001d5c6/\U0001d5c1')
'30 km/h'
>>> unidecode(u"\u5317\u4EB0")
'Bei Jing '

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.

Requirements

Nothing except Python itself.

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.

Installation

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

Frequently asked questions

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.

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 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.

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 use unidecode to generate URLs for your website, either generate the URL slug once and store it in the database or lock your dependency to unidecode to one specific version.

Some of the issues in this section are discussed in more detail in this blog post.

Performance notes

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.

Source

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

Contact

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 tomaz.solc@tablix.org.

Alternatively, you can also open a ticket or pull request at https://github.com/avian2/unidecode

Project details


Download files

Download the file for your platform. If you're not sure which to choose, learn more about installing packages.

Source Distribution

Unidecode-1.0.23.tar.gz (210.7 kB view hashes)

Uploaded Source

Built Distribution

Unidecode-1.0.23-py2.py3-none-any.whl (237.4 kB view hashes)

Uploaded Python 2 Python 3

Supported by

AWS AWS Cloud computing and Security Sponsor Datadog Datadog Monitoring Fastly Fastly CDN Google Google Download Analytics Microsoft Microsoft PSF Sponsor Pingdom Pingdom Monitoring Sentry Sentry Error logging StatusPage StatusPage Status page