In order to make a Django project translatable, you have to add a minimal amount of hooks to your Python code and templates. These hooks are called translation strings. They tell Django: “This text should be translated into the end user’s language, if a translation for this text is available in that language.” It’s your responsibility to mark translatable strings; the system can only translate strings it knows about.

Django then provides utilities to extract the translation strings into a message file. This file is a convenient way for translators to provide the equivalent of the translation strings in the target language. Once the translators have filled in the message file, it must be compiled. This process relies on the GNU gettext toolset.

Once this is done, Django takes care of translating Web apps on the fly in each available language, according to users’ language preferences.

Django’s internationalization hooks are on by default, and that means there’s a bit of i18n-related overhead in certain places of the framework. If you don’t use internationalization, you should take the two seconds to set USE_I18N = False in your settings file. Then Django will make some optimizations so as not to load the internationalization machinery. You’ll probably also want to remove 'django.core.context_processors.i18n' from your TEMPLATE_CONTEXT_PROCESSORS setting.


There is also an independent but related USE_L10N setting that controls if Django should implement format localization. See Format localization for more details.

Internationalization: in Python code

Standard translation

Specify a translation string by using the function ugettext(). It’s convention to import this as a shorter alias, _, to save typing.


Python’s standard library gettext module installs _() into the global namespace, as an alias for gettext(). In Django, we have chosen not to follow this practice, for a couple of reasons:

  1. For international character set (Unicode) support, ugettext() is more useful than gettext(). Sometimes, you should be using ugettext_lazy() as the default translation method for a particular file. Without _() in the global namespace, the developer has to think about which is the most appropriate translation function.
  2. The underscore character (_) is used to represent “the previous result” in Python’s interactive shell and doctest tests. Installing a global _() function causes interference. Explicitly importing ugettext() as _() avoids this problem.

In this example, the text "Welcome to my site." is marked as a translation string:

from django.utils.translation import ugettext as _

def my_view(request):
    output = _("Welcome to my site.")
    return HttpResponse(output)

Obviously, you could code this without using the alias. This example is identical to the previous one:

from django.utils.translation import ugettext

def my_view(request):
    output = ugettext("Welcome to my site.")
    return HttpResponse(output)

Translation works on computed values. This example is identical to the previous two:

def my_view(request):
    words = ['Welcome', 'to', 'my', 'site.']
    output = _(' '.join(words))
    return HttpResponse(output)

Translation works on variables. Again, here’s an identical example:

def my_view(request):
    sentence = 'Welcome to my site.'
    output = _(sentence)
    return HttpResponse(output)

(The caveat with using variables or computed values, as in the previous two examples, is that Django’s translation-string-detecting utility, makemessages, won’t be able to find these strings. More on makemessages later.)

The strings you pass to _() or ugettext() can take placeholders, specified with Python’s standard named-string interpolation syntax. Example:

def my_view(request, m, d):
    output = _('Today is %(month)s %(day)s.') % {'month': m, 'day': d}
    return HttpResponse(output)

This technique lets language-specific translations reorder the placeholder text. For example, an English translation may be "Today is November 26.", while a Spanish translation may be "Hoy es 26 de Noviembre." – with the the month and the day placeholders swapped.

For this reason, you should use named-string interpolation (e.g., %(day)s) instead of positional interpolation (e.g., %s or %d) whenever you have more than a single parameter. If you used positional interpolation, translations wouldn’t be able to reorder placeholder text.

Comments for translators

If you would like to give translators hints about a translatable string, you can add a comment prefixed with the Translators keyword on the line preceding the string, e.g.:

def my_view(request):
    # Translators: This message appears on the home page only
    output = ugettext("Welcome to my site.")

This also works in templates with the comment tag:

{% comment %}Translators: This is a text of the base template {% endcomment %}

The comment will then appear in the resulting .po file and should also be displayed by most translation tools.

Marking strings as no-op

Use the function django.utils.translation.ugettext_noop() to mark a string as a translation string without translating it. The string is later translated from a variable.

Use this if you have constant strings that should be stored in the source language because they are exchanged over systems or users – such as strings in a database – but should be translated at the last possible point in time, such as when the string is presented to the user.


Use the function django.utils.translation.ungettext() to specify pluralized messages.

ungettext takes three arguments: the singular translation string, the plural translation string and the number of objects.

This function is useful when you need your Django application to be localizable to languages where the number and complexity of plural forms is greater than the two forms used in English (‘object’ for the singular and ‘objects’ for all the cases where count is different from one, irrespective of its value.)

For example:

from django.utils.translation import ungettext

def hello_world(request, count):
    page = ungettext(
        'there is %(count)d object',
        'there are %(count)d objects',
    count) % {
        'count': count,
    return HttpResponse(page)

In this example the number of objects is passed to the translation languages as the count variable.

Lets see a slightly more complex usage example:

from django.utils.translation import ungettext

count = Report.objects.count()
if count == 1:
    name = Report._meta.verbose_name
    name = Report._meta.verbose_name_plural

text = ungettext(
        'There is %(count)d %(name)s available.',
        'There are %(count)d %(name)s available.',
) % {
    'count': count,
    'name': name

Here we reuse localizable, hopefully already translated literals (contained in the verbose_name and verbose_name_plural model Meta options) for other parts of the sentence so all of it is consistently based on the cardinality of the elements at play.


When using this technique, make sure you use a single name for every extrapolated variable included in the literal. In the example above note how we used the name Python variable in both translation strings. This example would fail:

from django.utils.translation import ungettext
from myapp.models import Report

count = Report.objects.count()
d = {
    'count': count,
    'name': Report._meta.verbose_name,
    'plural_name': Report._meta.verbose_name_plural
text = ungettext(
        'There is %(count)d %(name)s available.',
        'There are %(count)d %(plural_name)s available.',
) % d

You would get an error when running compilemessages:

a format specification for argument 'name', as in 'msgstr[0]', doesn't exist in 'msgid'

Contextual markers

Sometimes words have several meanings, such as "May" in English, which refers to a month name and to a verb. To enable translators to translate these words correctly in different contexts, you can use the django.utils.translation.pgettext() function, or the django.utils.translation.npgettext() function if the string needs pluralization. Both take a context string as the first variable.

In the resulting .po file, the string will then appear as often as there are different contextual markers for the same string (the context will appear on the msgctxt line), allowing the translator to give a different translation for each of them.

For example:

from django.utils.translation import pgettext

month = pgettext("month name", "May")


from django.utils.translation import pgettext_lazy

class MyThing(models.Model):
    name = models.CharField(help_text=pgettext_lazy(
        'help text for MyThing model', 'This is the help text'))

will appear in the .po file as:

msgctxt "month name"
msgid "May"
msgstr ""

Contextual markers are also supported by the trans and blocktrans template tags.

Lazy translation

Use the lazy versions of translation functions in django.utils.translation (easily recognizable by the lazy suffix in their names) to translate strings lazily – when the value is accessed rather than when they’re called.

These functions store a lazy reference to the string – not the actual translation. The translation itself will be done when the string is used in a string context, such as in template rendering.

This is essential when calls to these functions are located in code paths that are executed at module load time.

This is something that can easily happen when defining models, forms and model forms, because Django implements these such that their fields are actually class-level attributes. For that reason, make sure to use lazy translations in the following cases:

Model fields and relationships verbose_name and help_text option values

For example, to translate the help text of the name field in the following model, do the following:

from django.utils.translation import ugettext_lazy

class MyThing(models.Model):
    name = models.CharField(help_text=ugettext_lazy('This is the help text'))

You can mark names of ForeignKey, ManyTomanyField or OneToOneField relationship as translatable by using their verbose_name options:

from django.utils.translation import ugettext_lazy as _

class MyThing(models.Model):
    kind = models.ForeignKey(ThingKind, related_name='kinds',

Just like you would do in verbose_name you should provide a lowercase verbose name text for the relation as Django will automatically titlecase it when required.

Model verbose names values

It is recommended to always provide explicit verbose_name and verbose_name_plural options rather than relying on the fallback English-centric and somewhat naïve determination of verbose names Django performs bu looking at the model’s class name:

from django.utils.translation import ugettext_lazy

class MyThing(models.Model):
    name = models.CharField(_('name'), help_text=ugettext_lazy('This is the help text'))

    class Meta:
        verbose_name = ugettext_lazy('my thing')
        verbose_name_plural = ugettext_lazy('my things')

Model methods short_description attribute values

For model methods, you can provide translations to Django and the admin site with the short_description attribute:

from django.utils.translation import ugettext_lazy as _

class MyThing(models.Model):
    kind = models.ForeignKey(ThingKind, related_name='kinds',

    def is_mouse(self):
        return self.kind.type == MOUSE_TYPE
    is_mouse.short_description = _('Is it a mouse?')

Working with lazy translation objects

The result of a ugettext_lazy() call can be used wherever you would use a unicode string (an object with type unicode) in Python. If you try to use it where a bytestring (a str object) is expected, things will not work as expected, since a ugettext_lazy() object doesn’t know how to convert itself to a bytestring. You can’t use a unicode string inside a bytestring, either, so this is consistent with normal Python behavior. For example:

# This is fine: putting a unicode proxy into a unicode string.
u"Hello %s" % ugettext_lazy("people")

# This will not work, since you cannot insert a unicode object
# into a bytestring (nor can you insert our unicode proxy there)
"Hello %s" % ugettext_lazy("people")

If you ever see output that looks like "hello <django.utils.functional...>", you have tried to insert the result of ugettext_lazy() into a bytestring. That’s a bug in your code.

If you don’t like the long ugettext_lazy name, you can just alias it as _ (underscore), like so:

from django.utils.translation import ugettext_lazy as _

class MyThing(models.Model):
    name = models.CharField(help_text=_('This is the help text'))

Using ugettext_lazy() and ungettext_lazy() to mark strings in models and utility functions is a common operation. When you’re working with these objects elsewhere in your code, you should ensure that you don’t accidentally convert them to strings, because they should be converted as late as possible (so that the correct locale is in effect). This necessitates the use of the helper function described next.

Joining strings: string_concat()

Standard Python string joins (''.join([...])) will not work on lists containing lazy translation objects. Instead, you can use django.utils.translation.string_concat(), which creates a lazy object that concatenates its contents and converts them to strings only when the result is included in a string. For example:

from django.utils.translation import string_concat
name = ugettext_lazy(u'John Lennon')
instrument = ugettext_lazy(u'guitar')
result = string_concat(name, ': ', instrument)

In this case, the lazy translations in result will only be converted to strings when result itself is used in a string (usually at template rendering time).

Localized names of languages


The get_language_info() function provides detailed information about languages:

>>> from django.utils.translation import get_language_info
>>> li = get_language_info('de')
>>> print li['name'], li['name_local'], li['bidi']
German Deutsch False

The name and name_local attributes of the dictionary contain the name of the language in English and in the language itself, respectively. The bidi attribute is True only for bi-directional languages.

The source of the language information is the django.conf.locale module. Similar access to this information is available for template code. See below.

Internationalization: in template code

Translations in Django templates uses two template tags and a slightly different syntax than in Python code. To give your template access to these tags, put {% load i18n %} toward the top of your template.

trans template tag

The {% trans %} template tag translates either a constant string (enclosed in single or double quotes) or variable content:

<title>{% trans "This is the title." %}</title>
<title>{% trans myvar %}</title>

If the noop option is present, variable lookup still takes place but the translation is skipped. This is useful when “stubbing out” content that will require translation in the future:

<title>{% trans "myvar" noop %}</title>

Internally, inline translations use an ugettext() call.

In case a template var (myvar above) is passed to the tag, the tag will first resolve such variable to a string at run-time and then look up that string in the message catalogs.

It’s not possible to mix a template variable inside a string within {% trans %}. If your translations require strings with variables (placeholders), use {% blocktrans %} instead.

If you’d like to retrieve a translated string without displaying it, you can use the following syntax:

{% trans "This is the title" as the_title %}

<title>{{ the_title }}</title>
<meta name="description" content="{{ the_title }}">

In practice you’ll use this to get strings that are used in multiple places or should be used as arguments for other template tags or filters:

{% trans "starting point" as start %}
{% trans "end point" as end %}
{% trans "La Grande Boucle" as race %}

  <a href="/" title="{% blocktrans %}Back to '{{ race }}' homepage{% endblocktrans %}">{{ race }}</a>
{% for stage in tour_stages %}
    {% cycle start end %}: {{ stage }}{% if forloop.counter|divisibleby:2 %}<br />{% else %}, {% endif %}
{% endfor %}

{% trans %} also supports contextual markers using the context keyword:

{% trans "May" context "month name" %}

blocktrans template tag

New keyword argument format.

Contrarily to the trans tag, the blocktrans tag allows you to mark complex sentences consisting of literals and variable content for translation by making use of placeholders:

{% blocktrans %}This string will have {{ value }} inside.{% endblocktrans %}

To translate a template expression – say, accessing object attributes or using template filters – you need to bind the expression to a local variable for use within the translation block. Examples:

{% blocktrans with amount=article.price %}
That will cost $ {{ amount }}.
{% endblocktrans %}

{% blocktrans with myvar=value|filter %}
This will have {{ myvar }} inside.
{% endblocktrans %}

You can use multiple expressions inside a single blocktrans tag:

{% blocktrans with book_t=book|title author_t=author|title %}
This is {{ book_t }} by {{ author_t }}
{% endblocktrans %}


The previous more verbose format is still supported: {% blocktrans with book|title as book_t and author|title as author_t %}

If resolving one of the block arguments fails, blocktrans will fall back to the default language by deactivating the currently active language temporarily with the deactivate_all() function.

This tag also provides for pluralization. To use it:

  • Designate and bind a counter value with the name count. This value will be the one used to select the right plural form.
  • Specify both the singular and plural forms separating them with the {% plural %} tag within the {% blocktrans %} and {% endblocktrans %} tags.

An example:

{% blocktrans count counter=list|length %}
There is only one {{ name }} object.
{% plural %}
There are {{ counter }} {{ name }} objects.
{% endblocktrans %}

A more complex example:

{% blocktrans with amount=article.price count years=i.length %}
That will cost $ {{ amount }} per year.
{% plural %}
That will cost $ {{ amount }} per {{ years }} years.
{% endblocktrans %}

When you use both the pluralization feature and bind values to local variables in addition to the counter value, keep in mind that the blocktrans construct is internally converted to an ungettext call. This means the same notes regarding ungettext variables apply.

Reverse URL lookups cannot be carried out within the blocktrans and should be retrieved (and stored) beforehand:

{% url arg arg2 as the_url %}
{% blocktrans %}
This is a URL: {{ the_url }}
{% endblocktrans %}

{% blocktrans %} also supports contextual markers using the context keyword:

{% blocktrans with name=user.username context "greeting" %}Hi {{ name }}{% endblocktrans %}

Other tags

Each RequestContext has access to three translation-specific variables:

  • LANGUAGES is a list of tuples in which the first element is the language code and the second is the language name (translated into the currently active locale).
  • LANGUAGE_CODE is the current user’s preferred language, as a string. Example: en-us. (See How Django discovers language preference.)
  • LANGUAGE_BIDI is the current locale’s direction. If True, it’s a right-to-left language, e.g.: Hebrew, Arabic. If False it’s a left-to-right language, e.g.: English, French, German etc.

If you don’t use the RequestContext extension, you can get those values with three tags:

{% get_current_language as LANGUAGE_CODE %}
{% get_available_languages as LANGUAGES %}
{% get_current_language_bidi as LANGUAGE_BIDI %}

These tags also require a {% load i18n %}.

Translation hooks are also available within any template block tag that accepts constant strings. In those cases, just use _() syntax to specify a translation string:

{% some_special_tag _("Page not found") value|yesno:_("yes,no") %}

In this case, both the tag and the filter will see the already-translated string, so they don’t need to be aware of translations.


In this example, the translation infrastructure will be passed the string "yes,no", not the individual strings "yes" and "no". The translated string will need to contain the comma so that the filter parsing code knows how to split up the arguments. For example, a German translator might translate the string "yes,no" as "ja,nein" (keeping the comma intact).

You can also retrieve information about any of the available languages using provided template tags and filters. To get information about a single language, use the {% get_language_info %} tag:

{% get_language_info for LANGUAGE_CODE as lang %}
{% get_language_info for "pl" as lang %}

You can then access the information:

Language code: {{ lang.code }}<br />
Name of language: {{ lang.name_local }}<br />
Name in English: {{ }}<br />
Bi-directional: {{ lang.bidi }}

You can also use the {% get_language_info_list %} template tag to retrieve information for a list of languages (e.g. active languages as specified in LANGUAGES). See the section about the set_language redirect view for an example of how to display a language selector using {% get_language_info_list %}.

In addition to LANGUAGES style nested tuples, {% get_language_info_list %} supports simple lists of language codes. If you do this in your view:

return render_to_response('mytemplate.html', {
    'available_languages': ['en', 'es', 'fr'],
}, RequestContext(request))

you can iterate over those languages in the template:

{% get_language_info_list for available_languages as langs %}
{% for lang in langs %} ... {% endfor %}

There are also simple filters available for convenience:

  • {{ LANGUAGE_CODE|language_name }} (“German”)
  • {{ LANGUAGE_CODE|language_name_local }} (“Deutsch”)
  • {{ LANGUAGE_CODE|bidi }} (False)

Internationalization: in JavaScript code

Adding translations to JavaScript poses some problems:

  • JavaScript code doesn’t have access to a gettext implementation.
  • JavaScript code doesn’t have access to .po or .mo files; they need to be delivered by the server.
  • The translation catalogs for JavaScript should be kept as small as possible.

Django provides an integrated solution for these problems: It passes the translations into JavaScript, so you can call gettext, etc., from within JavaScript.

The javascript_catalog view

javascript_catalog(request, domain='djangojs', packages=None)

The main solution to these problems is the django.views.i18n.javascript_catalog() view, which sends out a JavaScript code library with functions that mimic the gettext interface, plus an array of translation strings. Those translation strings are taken from applications or Django core, according to what you specify in either the info_dict or the URL. Paths listed in LOCALE_PATHS are also included.

You hook it up like this:

js_info_dict = {
    'packages': ('',),

urlpatterns = patterns('',
    (r'^jsi18n/$', 'django.views.i18n.javascript_catalog', js_info_dict),

Each string in packages should be in Python dotted-package syntax (the same format as the strings in INSTALLED_APPS) and should refer to a package that contains a locale directory. If you specify multiple packages, all those catalogs are merged into one catalog. This is useful if you have JavaScript that uses strings from different applications.

The precedence of translations is such that the packages appearing later in the packages argument have higher precedence than the ones appearing at the beginning, this is important in the case of clashing translations for the same literal.

By default, the view uses the djangojs gettext domain. This can be changed by altering the domain argument.

You can make the view dynamic by putting the packages into the URL pattern:

urlpatterns = patterns('',
    (r'^jsi18n/(?P<packages>\S+?)/$', 'django.views.i18n.javascript_catalog'),

With this, you specify the packages as a list of package names delimited by ‘+’ signs in the URL. This is especially useful if your pages use code from different apps and this changes often and you don’t want to pull in one big catalog file. As a security measure, these values can only be either django.conf or any package from the INSTALLED_APPS setting.

The JavaScript translations found in the paths listed in the LOCALE_PATHS setting are also always included. To keep consistency with the translations lookup order algorithm used for Python and templates, the directories listed in LOCALE_PATHS have the highest precedence with the ones appearing first having higher precedence than the ones appearing later.

Directories listed in LOCALE_PATHS weren’t included in the lookup algorithm until version 1.3.

Using the JavaScript translation catalog

To use the catalog, just pull in the dynamically generated script like this:

<script type="text/javascript" src="{% url django.views.i18n.javascript_catalog %}"></script>

This uses reverse URL lookup to find the URL of the JavaScript catalog view. When the catalog is loaded, your JavaScript code can use the standard gettext interface to access it:

document.write(gettext('this is to be translated'));

There is also an ngettext interface:

var object_cnt = 1 // or 0, or 2, or 3, ...
s = ngettext('literal for the singular case',
        'literal for the plural case', object_cnt);

and even a string interpolation function:

function interpolate(fmt, obj, named);

The interpolation syntax is borrowed from Python, so the interpolate function supports both positional and named interpolation:

  • Positional interpolation: obj contains a JavaScript Array object whose elements values are then sequentially interpolated in their corresponding fmt placeholders in the same order they appear. For example:

    fmts = ngettext('There is %s object. Remaining: %s',
            'There are %s objects. Remaining: %s', 11);
    s = interpolate(fmts, [11, 20]);
    // s is 'There are 11 objects. Remaining: 20'
  • Named interpolation: This mode is selected by passing the optional boolean named parameter as true. obj contains a JavaScript object or associative array. For example:

    d = {
        count: 10,
        total: 50
    fmts = ngettext('Total: %(total)s, there is %(count)s object',
    'there are %(count)s of a total of %(total)s objects', d.count);
    s = interpolate(fmts, d, true);

You shouldn’t go over the top with string interpolation, though: this is still JavaScript, so the code has to make repeated regular-expression substitutions. This isn’t as fast as string interpolation in Python, so keep it to those cases where you really need it (for example, in conjunction with ngettext to produce proper pluralizations).

Internationalization: in URL patterns

Django provides two mechanisms to internationalize URL patterns:


Using either one of these features requires that an active language be set for each request; in other words, you need to have django.middleware.locale.LocaleMiddleware in your MIDDLEWARE_CLASSES setting.

Language prefix in URL patterns

i18n_patterns(prefix, pattern_description, ...)

This function can be used in your root URLconf as a replacement for the normal django.conf.urls.patterns() function. Django will automatically prepend the current active language code to all url patterns defined within i18n_patterns(). Example URL patterns:

from django.conf.urls import patterns, include, url
from django.conf.urls.i18n import i18n_patterns

urlpatterns = patterns(''
    url(r'^sitemap\.xml$', 'sitemap.view', name='sitemap_xml'),

news_patterns = patterns(''
    url(r'^$', 'news.views.index', name='index'),
    url(r'^category/(?P<slug>[\w-]+)/$', 'news.views.category', name='category'),
    url(r'^(?P<slug>[\w-]+)/$', 'news.views.details', name='detail'),

urlpatterns += i18n_patterns('',
    url(r'^about/$', 'about.view', name='about'),
    url(r'^news/$', include(news_patterns, namespace='news')),

After defining these URL patterns, Django will automatically add the language prefix to the URL patterns that were added by the i18n_patterns function. Example:

from django.core.urlresolvers import reverse
from django.utils.translation import activate

>>> activate('en')
>>> reverse('sitemap_xml')
>>> reverse('news:index')

>>> activate('nl')
>>> reverse('news:detail', kwargs={'slug': 'news-slug'})


i18n_patterns() is only allowed in your root URLconf. Using it within an included URLconf will throw an ImproperlyConfigured exception.


Ensure that you don’t have non-prefixed URL patterns that might collide with an automatically-added language prefix.

Translating URL patterns

URL patterns can also be marked translatable using the ugettext_lazy() function. Example:

from django.conf.urls import patterns, include, url
from django.conf.urls.i18n import i18n_patterns
from django.utils.translation import ugettext_lazy as _

urlpatterns = patterns(''
    url(r'^sitemap\.xml$', 'sitemap.view', name='sitemap_xml'),

news_patterns = patterns(''
    url(r'^$', 'news.views.index', name='index'),
    url(_(r'^category/(?P<slug>[\w-]+)/$'), 'news.views.category', name='category'),
    url(r'^(?P<slug>[\w-]+)/$', 'news.views.details', name='detail'),

urlpatterns += i18n_patterns('',
    url(_(r'^about/$'), 'about.view', name='about'),
    url(_(r'^news/$'), include(news_patterns, namespace='news')),

After you’ve created the translations, the reverse() function will return the URL in the active language. Example:

from django.core.urlresolvers import reverse
from django.utils.translation import activate

>>> activate('en')
>>> reverse('news:category', kwargs={'slug': 'recent'})

>>> activate('nl')
>>> reverse('news:category', kwargs={'slug': 'recent'})


In most cases, it’s best to use translated URLs only within a language-code-prefixed block of patterns (using i18n_patterns()), to avoid the possibility that a carelessly translated URL causes a collision with a non-translated URL pattern.

Reversing in templates

If localized URLs get reversed in templates they always use the current language. To link to a URL in another language use the language template tag. It enables the given language in the enclosed template section:

{% load i18n %}

{% get_available_languages as languages %}

{% trans "View this category in:" %}
{% for lang_code, lang_name in languages %}
    {% language lang_code %}
    <a href="{% url category slug=category.slug %}">{{ lang_name }}</a>
    {% endlanguage %}
{% endfor %}

The language tag expects the language code as the only argument.

Localization: how to create language files

Once the string literals of an application have been tagged for later translation, the translation themselves need to be written (or obtained). Here’s how that works.

Locale restrictions

Django does not support localizing your application into a locale for which Django itself has not been translated. In this case, it will ignore your translation files. If you were to try this and Django supported it, you would inevitably see a mixture of translated strings (from your application) and English strings (from Django itself). If you want to support a locale for your application that is not already part of Django, you’ll need to make at least a minimal translation of the Django core.

A good starting point is to copy the Django English .po file and to translate at least some translation strings.

Message files

The first step is to create a message file for a new language. A message file is a plain-text file, representing a single language, that contains all available translation strings and how they should be represented in the given language. Message files have a .po file extension.

Django comes with a tool, makemessages, that automates the creation and upkeep of these files.

Gettext utilities

The makemessages command (and compilemessages discussed later) use commands from the GNU gettext toolset: xgettext, msgfmt, msgmerge and msguniq.

The minimum version of the gettext utilities supported is 0.15.

To create or update a message file, run this command: makemessages -l de

...where de is the language code for the message file you want to create. The language code, in this case, is in locale format. For example, it’s pt_BR for Brazilian Portuguese and de_AT for Austrian German.

The script should be run from one of two places:

  • The root directory of your Django project.
  • The root directory of your Django app.

The script runs over your project source tree or your application source tree and pulls out all strings marked for translation. It creates (or updates) a message file in the directory locale/LANG/LC_MESSAGES. In the de example, the file will be locale/de/LC_MESSAGES/django.po.

By default makemessages examines every file that has the .html or .txt file extension. In case you want to override that default, use the --extension or -e option to specify the file extensions to examine: makemessages -l de -e txt

Separate multiple extensions with commas and/or use -e or --extension multiple times: makemessages -l de -e html,txt -e xml


When creating message files from JavaScript source code you need to use the special ‘djangojs’ domain, not -e js.

No gettext?

If you don’t have the gettext utilities installed, makemessages will create empty files. If that’s the case, either install the gettext utilities or just copy the English message file (locale/en/LC_MESSAGES/django.po) if available and use it as a starting point; it’s just an empty translation file.

Working on Windows?

If you’re using Windows and need to install the GNU gettext utilities so makemessages works, see gettext on Windows for more information.

The format of .po files is straightforward. Each .po file contains a small bit of metadata, such as the translation maintainer’s contact information, but the bulk of the file is a list of messages – simple mappings between translation strings and the actual translated text for the particular language.

For example, if your Django app contained a translation string for the text "Welcome to my site.", like so:

_("Welcome to my site.")

...then makemessages will have created a .po file containing the following snippet – a message:

#: path/to/python/
msgid "Welcome to my site."
msgstr ""

A quick explanation:

  • msgid is the translation string, which appears in the source. Don’t change it.
  • msgstr is where you put the language-specific translation. It starts out empty, so it’s your responsibility to change it. Make sure you keep the quotes around your translation.
  • As a convenience, each message includes, in the form of a comment line prefixed with # and located above the msgid line, the filename and line number from which the translation string was gleaned.

Long messages are a special case. There, the first string directly after the msgstr (or msgid) is an empty string. Then the content itself will be written over the next few lines as one string per line. Those strings are directly concatenated. Don’t forget trailing spaces within the strings; otherwise, they’ll be tacked together without whitespace!

Mind your charset

When creating a PO file with your favorite text editor, first edit the charset line (search for "CHARSET") and set it to the charset you’ll be using to edit the content. Due to the way the gettext tools work internally and because we want to allow non-ASCII source strings in Django’s core and your applications, you must use UTF-8 as the encoding for your PO file. This means that everybody will be using the same encoding, which is important when Django processes the PO files.

To reexamine all source code and templates for new translation strings and update all message files for all languages, run this: makemessages -a

Compiling message files

After you create your message file – and each time you make changes to it – you’ll need to compile it into a more efficient form, for use by gettext. Do this with the compilemessages utility.

This tool runs over all available .po files and creates .mo files, which are binary files optimized for use by gettext. In the same directory from which you ran makemessages, run compilemessages like this: compilemessages

That’s it. Your translations are ready for use.

Working on Windows?

If you’re using Windows and need to install the GNU gettext utilities so compilemessages works see gettext on Windows for more information.

.po files: Encoding and BOM usage.

Django only supports .po files encoded in UTF-8 and without any BOM (Byte Order Mark) so if your text editor adds such marks to the beginning of files by default then you will need to reconfigure it.

Creating message files from JavaScript source code

You create and update the message files the same way as the other Django message files – with the makemessages tool. The only difference is you need to explicitly specify what in gettext parlance is known as a domain in this case the djangojs domain, by providing a -d djangojs parameter, like this: makemessages -d djangojs -l de

This would create or update the message file for JavaScript for German. After updating message files, just run compilemessages the same way as you do with normal Django message files.

gettext on Windows

This is only needed for people who either want to extract message IDs or compile message files (.po). Translation work itself just involves editing existing files of this type, but if you want to create your own message files, or want to test or compile a changed message file, you will need the gettext utilities:

  • Download the following zip files from the GNOME servers or from one of its mirrors


    X is the version number, we are requiring 0.15 or higher.

  • Extract the contents of the bin\ directories in both files to the same folder on your system (i.e. C:\Program Files\gettext-utils)

  • Update the system PATH:

    • Control Panel > System > Advanced > Environment Variables.
    • In the System variables list, click Path, click Edit.
    • Add ;C:\Program Files\gettext-utils\bin at the end of the Variable value field.

You may also use gettext binaries you have obtained elsewhere, so long as the xgettext --version command works properly. Do not attempt to use Django translation utilities with a gettext package if the command xgettext --version entered at a Windows command prompt causes a popup window saying “xgettext.exe has generated errors and will be closed by Windows”.


The set_language redirect view


As a convenience, Django comes with a view, django.views.i18n.set_language(), that sets a user’s language preference and redirects to a given URL or, by default, back to the previous page.

Activate this view by adding the following line to your URLconf:

(r'^i18n/', include('django.conf.urls.i18n')),

(Note that this example makes the view available at /i18n/setlang/.)

The view expects to be called via the POST method, with a language parameter set in request. If session support is enabled, the view saves the language choice in the user’s session. Otherwise, it saves the language choice in a cookie that is by default named django_language. (The name can be changed through the LANGUAGE_COOKIE_NAME setting.)

After setting the language choice, Django redirects the user, following this algorithm:

  • Django looks for a next parameter in the POST data.
  • If that doesn’t exist, or is empty, Django tries the URL in the Referrer header.
  • If that’s empty – say, if a user’s browser suppresses that header – then the user will be redirected to / (the site root) as a fallback.

Here’s example HTML template code:

<form action="/i18n/setlang/" method="post">
{% csrf_token %}
<input name="next" type="hidden" value="{{ redirect_to }}" />
<select name="language">
{% get_language_info_list for LANGUAGES as languages %}
{% for language in languages %}
<option value="{{ language.code }}">{{ language.name_local }} ({{ language.code }})</option>
{% endfor %}
<input type="submit" value="Go" />

In this example, Django looks up the URL of the page to which the user will be redirected in the redirect_to context variable.

Using translations outside views and templates

While Django provides a rich set of i18n tools for use in views and templates, it does not restrict the usage to Django-specific code. The Django translation mechanisms can be used to translate arbitrary texts to any language that is supported by Django (as long as an appropriate translation catalog exists, of course). You can load a translation catalog, activate it and translate text to language of your choice, but remember to switch back to original language, as activating a translation catalog is done on per-thread basis and such change will affect code running in the same thread.

For example:

from django.utils import translation
def welcome_translated(language):
    cur_language = translation.get_language()
        text = translation.ugettext('welcome')
    return text

Calling this function with the value ‘de’ will give you "Willkommen", regardless of LANGUAGE_CODE and language set by middleware.

Functions of particular interest are django.utils.translation.get_language() which returns the language used in the current thread, django.utils.translation.activate() which activates a translation catalog for the current thread, and django.utils.translation.check_for_language() which checks if the given language is supported by Django.

Implementation notes

Specialties of Django translation

Django’s translation machinery uses the standard gettext module that comes with Python. If you know gettext, you might note these specialties in the way Django does translation:

  • The string domain is django or djangojs. This string domain is used to differentiate between different programs that store their data in a common message-file library (usually /usr/share/locale/). The django domain is used for python and template translation strings and is loaded into the global translation catalogs. The djangojs domain is only used for JavaScript translation catalogs to make sure that those are as small as possible.
  • Django doesn’t use xgettext alone. It uses Python wrappers around xgettext and msgfmt. This is mostly for convenience.

How Django discovers language preference

Once you’ve prepared your translations – or, if you just want to use the translations that come with Django – you’ll just need to activate translation for your app.

Behind the scenes, Django has a very flexible model of deciding which language should be used – installation-wide, for a particular user, or both.

To set an installation-wide language preference, set LANGUAGE_CODE. Django uses this language as the default translation – the final attempt if no other translator finds a translation.

If all you want to do is run Django with your native language, and a language file is available for it, all you need to do is set LANGUAGE_CODE.

If you want to let each individual user specify which language he or she prefers, use LocaleMiddleware. LocaleMiddleware enables language selection based on data from the request. It customizes content for each user.

To use LocaleMiddleware, add 'django.middleware.locale.LocaleMiddleware' to your MIDDLEWARE_CLASSES setting. Because middleware order matters, you should follow these guidelines:

  • Make sure it’s one of the first middlewares installed.
  • It should come after SessionMiddleware, because LocaleMiddleware makes use of session data. And it should come before CommonMiddleware because CommonMiddleware needs an activated language in order to resolve the requested URL.
  • If you use CacheMiddleware, put LocaleMiddleware after it.

For example, your MIDDLEWARE_CLASSES might look like this:


(For more on middleware, see the middleware documentation.)

LocaleMiddleware tries to determine the user’s language preference by following this algorithm:

  • First, it looks for the language prefix in the requested URL. This is only performed when you are using the i18n_patterns function in your root URLconf. See Internationalization: in URL patterns for more information about the language prefix and how to internationalize URL patterns.

  • Failing that, it looks for a django_language key in the current user’s session.

  • Failing that, it looks for a cookie.

    The name of the cookie used is set by the LANGUAGE_COOKIE_NAME setting. (The default name is django_language.)

  • Failing that, it looks at the Accept-Language HTTP header. This header is sent by your browser and tells the server which language(s) you prefer, in order by priority. Django tries each language in the header until it finds one with available translations.

  • Failing that, it uses the global LANGUAGE_CODE setting.


  • In each of these places, the language preference is expected to be in the standard language format, as a string. For example, Brazilian Portuguese is pt-br.

  • If a base language is available but the sublanguage specified is not, Django uses the base language. For example, if a user specifies de-at (Austrian German) but Django only has de available, Django uses de.

  • Only languages listed in the LANGUAGES setting can be selected. If you want to restrict the language selection to a subset of provided languages (because your application doesn’t provide all those languages), set LANGUAGES to a list of languages. For example:

      ('de', _('German')),
      ('en', _('English')),

    This example restricts languages that are available for automatic selection to German and English (and any sublanguage, like de-ch or en-us).

  • If you define a custom LANGUAGES setting, as explained in the previous bullet, it’s OK to mark the languages as translation strings – but use a “dummy” ugettext() function, not the one in django.utils.translation. You should never import django.utils.translation from within your settings file, because that module in itself depends on the settings, and that would cause a circular import.

    The solution is to use a “dummy” ugettext() function. Here’s a sample settings file:

    ugettext = lambda s: s
        ('de', ugettext('German')),
        ('en', ugettext('English')),

    With this arrangement, makemessages will still find and mark these strings for translation, but the translation won’t happen at runtime – so you’ll have to remember to wrap the languages in the real ugettext() in any code that uses LANGUAGES at runtime.

  • The LocaleMiddleware can only select languages for which there is a Django-provided base translation. If you want to provide translations for your application that aren’t already in the set of translations in Django’s source tree, you’ll want to provide at least a basic one as described in the Locale restrictions note.

Once LocaleMiddleware determines the user’s preference, it makes this preference available as request.LANGUAGE_CODE for each HttpRequest. Feel free to read this value in your view code. Here’s a simple example:

def hello_world(request, count):
    if request.LANGUAGE_CODE == 'de-at':
        return HttpResponse("You prefer to read Austrian German.")
        return HttpResponse("You prefer to read another language.")

Note that, with static (middleware-less) translation, the language is in settings.LANGUAGE_CODE, while with dynamic (middleware) translation, it’s in request.LANGUAGE_CODE.

How Django discovers translations

At runtime, Django builds an in-memory unified catalog of literals-translations. To achieve this it looks for translations by following this algorithm regarding the order in which it examines the different file paths to load the compiled message files (.mo) and the precedence of multiple translations for the same literal:

  1. The directories listed in LOCALE_PATHS have the highest precedence, with the ones appearing first having higher precedence than the ones appearing later.
  2. Then, it looks for and uses if it exists a locale directory in each of the installed apps listed in INSTALLED_APPS. The ones appearing first have higher precedence than the ones appearing later.
  3. Then, it looks for a locale directory in the project directory, or more accurately, in the directory containing your settings file.
  4. Finally, the Django-provided base translation in django/conf/locale is used as a fallback.

Deprecated since version 1.3: Lookup in the locale subdirectory of the directory containing your settings file (item 3 above) is deprecated since the 1.3 release and will be removed in Django 1.5. You can use the LOCALE_PATHS setting instead, by listing the absolute filesystem path of such locale directory in the setting value.

See also

The translations for literals included in JavaScript assets are looked up following a similar but not identical algorithm. See the javascript_catalog view documentation for more details.

In all cases the name of the directory containing the translation is expected to be named using locale name notation. E.g. de, pt_BR, es_AR, etc.

This way, you can write applications that include their own translations, and you can override base translations in your project path. Or, you can just build a big project out of several apps and put all translations into one big common message file specific to the project you are composing. The choice is yours.


If you’re using manually configured settings, as described in 无DJANGO_SETTINGS_MODULE下使用配置, the locale directory in the project directory will not be examined, since Django loses the ability to work out the location of the project directory. (Django normally uses the location of the settings file to determine this, and a settings file doesn’t exist if you’re manually configuring your settings.)

All message file repositories are structured the same way. They are:

  • All paths listed in LOCALE_PATHS in your settings file are searched for <language>/LC_MESSAGES/django.(po|mo)
  • $PROJECTPATH/locale/<language>/LC_MESSAGES/django.(po|mo) – deprecated, see above.
  • $APPPATH/locale/<language>/LC_MESSAGES/django.(po|mo)
  • $PYTHONPATH/django/conf/locale/<language>/LC_MESSAGES/django.(po|mo)

To create message files, you use the makemessages tool. You only need to be in the same directory where the locale/ directory is located. And you use compilemessages to produce the binary .mo files that are used by gettext.

You can also run compilemessages to make the compiler process all the directories in your LOCALE_PATHS setting.

Finally, you should give some thought to the structure of your translation files. If your applications need to be delivered to other users and will be used in other projects, you might want to use app-specific translations. But using app-specific translations and project-specific translations could produce weird problems with makemessages: it will traverse all directories below the current path and so might put message IDs into a unified, common message file for the current project that are already in application message files.

The easiest way out is to store applications that are not part of the project (and so carry their own translations) outside the project tree. That way, makemessages, when ran on a project level will only extract strings that are connected to your explicit project and not strings that are distributed independently.