Internationalization is a complex problem. Natural languages differ in so many ways (eg. in pluralization rules) that it is hard to provide tools for solving all problems at once. For that reason the Rails I18n API focuses on:

As part of this solution, every static string in the Rails framework — eg. Active Record validation messages, time and date formats — has been internationalized, so localization of a Rails application means "over-riding" these defaults.

Thus, the Ruby I18n gem is split into two parts:

As a user you should always only access the public methods on the I18n module, but it is useful to know about the capabilities of the backend.

The most important methods of the I18n API are:

These have the aliases #t and #l so you can use them like this:

There are also attribute readers and writers for the following attributes:

So, let’s internationalize a simple Rails application from the ground up in the next chapters!

There are just a few, simple steps to get up and running with I18n support for your application.

Following the convention over configuration philosophy, Rails will set-up your application with reasonable defaults. If you need different settings, you can overwrite them easily.

Rails adds all .rb and .yml files from config/locales directory to your translations load path, automatically.

See the default en.yml locale in this directory, containing a sample pair of translation strings:

This means, that in the :en locale, the key hello will map to Hello world string. Every string inside Rails is internationalized in this way, see for instance Active Record validation messages in the activerecord/lib/active_record/locale/en.yml file or time and date formats in the activesupport/lib/active_support/locale/en.yml file. You can use YAML or standard Ruby Hashes to store translations in the default (Simple) backend.

The I18n library will use English as a default locale, ie. if you don’t set a different locale, :en will be used for looking up translations.

The translations load path (I18n.load_path) is just a Ruby Array of paths to your translation files that will be loaded automatically and available in your application. You can pick whatever directory and translation file naming scheme makes sense for you.

The default environment.rb files has instruction how to add locales from another directory and how to set different default locale. Just uncomment and edit the specific lines.

For the sake of completeness, let’s mention that if you do not want to use the environment.rb file for some reason, you can always wire up things manually, too.

To tell the I18n library where it can find your custom translation files you can specify the load path anywhere in your application - just make sure it gets run before any translations are actually looked up. You might also want to change the default locale. The simplest thing possible is to put the following into an initializer:

If you want to translate your Rails application to a single language other than English (the default locale), you can set I18n.default_locale to your locale in environment.rb or an initializer as shown above, and it will persist through the requests.

However, you would probably like to provide support for more locales in your application. In such case, you need to set and pass the locale between requests.

The setting part is easy. You can set locale in a before_filter in the ApplicationController like this:

This requires you to pass the locale as a URL query parameter as in http://example.com/books?locale=pt. (This is eg. Google’s approach). So http://localhost:3000?locale=pt will load the Portugese localization, whereas http://localhost:3000?locale=de would load the German localization, and so on. You may skip the next section and head over to the Internationalize your application section, if you want to try things out by manually placing locale in the URL and reloading the page.

Of course, you probably don’t want to manually include locale in every URL all over your application, or want the URLs look differently, eg. the usual http://example.com/pt/books versus http://example.com/en/books. Let’s discuss the different options you have.

So, for having available locales easily available in Rails 2.2, we have to include this support manually in an initializer, like this:

You can then wrap the constant for easy access in ApplicationController:

One option you have is to set the locale from the domain name where your application runs. For example, we want www.example.com to load English (or default) locale, and www.example.es to load Spanish locale. Thus the top-level domain name is used for locale setting. This has several advantages:

You can implement it like this in your ApplicationController:

We can also set the locale from the subdomain in very similar way:

If your application includes a locale switching menu, you would then have something like this in it:

assuming you would set APP_CONFIG[:deutsch_website_url] to some value like http://www.application.de.

This solution has aforementioned advantages, however, you may not be able or may not want to provide different localizations ("language versions") on different domains. The most obvious solution would be to include locale code in the URL params (or request path).

Most usual way of setting (and passing) the locale would be to include it in URL params, as we did in the I18n.locale = params[:locale] before_filter in the first example. We would like to have URLs like www.example.com/books?locale=ja or www.example.com/ja/books in this case.

This approach has almost the same set of advantages as setting the locale from domain name: namely that it’s RESTful and in accord with rest of the World Wide Web. It does require a little bit more work to implement, though.

Getting the locale from params and setting it accordingly is not hard; including it in every URL and thus passing it through the requests is. To include an explicit option in every URL (eg. link_to( books_url(:locale => I18n.locale) )) would be tedious and probably impossible, of course.

Rails contains infrastructure for "centralizing dynamic decisions about the URLs" in its ApplicationController#default_url_options, which is useful precisely in this scenario: it enables us to set "defaults" for url_for and helper methods dependent on it (by implementing/overriding this method).

We can include something like this in our ApplicationController then:

Every helper method dependent on url_for (eg. helpers for named routes like root_path or root_url, resource routes like books_path or books_url, etc.) will now automatically include the locale in the query string, like this: http://localhost:3001/?locale=ja.

You may be satisfied with this. It does impact the readability of URLs, though, when the locale "hangs" at the end of every URL in your application. Moreover, from the architectural standpoint, locale is usually hierarchically above the other parts of application domain: and URLs should reflect this.

You probably want URLs look like this: www.example.com/en/books (which loads English locale) and www.example.com/nl/books (which loads Netherlands locale). This is achievable with the "over-riding default_url_options" strategy from above: you just have to set up your routes with path_prefix option in this way:

Now, when you call books_path method you should get "/en/books" (for the default locale). An URL like http://localhost:3001/nl/books should load the Netherlands locale, then, and following calls to books_path should return "/nl/books" (because the locale changed).

Of course, you need to take special care of root URL (usually "homepage" or "dashboard") of your application. An URL like http://localhost:3001/nl will not work automatically, because the map.root :controller => "dashboard" declaration in your routes.rb doesn’t take locale into account. (And rightly so. There’s only one "root" URL.)

You would probably need to map URLs like these:

Do take special care about the order of your routes, so this route declaration does not "eat" other ones. (You may want to add it directly before the map.root declaration.)

In specific cases, it would make sense to set locale from client supplied information, ie. not from URL. This information may come for example from users' preffered language (set in their browser), can be based on users' geographical location inferred from their IP, or users can provide it simply by choosing locale in your application interface and saving it to their profile. This approach is more suitable for web-based applications or services, not for websites — see the box about sessions, cookies and RESTful architecture above.

One source of client supplied information would be an Accept-Language HTTP header. People may set this in their browser or other clients (such as curl).

A trivial implementation of using Accept-Language header would be:

Of course, in production environment you would need much robust code, and could use a plugin such as Iaian Hecker’s http_accept_language.

Another way of choosing the locale from client’s information would be to use a database for mapping client IP to region, such as GeoIP Lite Country. The mechanics of the code would be very similar to the code above — you would need to query database for user’s IP, and lookup your preffered locale for the country/region/city returned.

You can also provide users of your application with means to set (and possibly over-ride) locale in your application interface, as well. Again, mechanics for this approach would be very similar to the code above — you’d probably let users choose a locale from a dropdown list and save it to their profile in database. Then you’d set the locale to this value.

OK! Now you’ve initialized I18n support for your Ruby on Rails application and told it which locale should be used and how to preserve it between requests. With that in place, you’re now ready for the really interesting stuff.

Let’s internationalize our application, ie. abstract every locale-specific parts, and that localize it, ie. provide neccessary translations for these abstracts.

You most probably have something like this in one of your applications:

Obviously there are two strings that are localized to English. In order to internationalize this code, replace these strings with calls to Rails' #t helper with a key that makes sense for the translation:

When you now render this view, it will show an error message which tells you that the translations for the keys :hello_world and :hello_flash are missing.

So let’s add the missing translations into the dictionary files (i.e. do the "localization" part):

There you go. Because you haven’t changed the default_locale, I18n will use English. Your application now shows:

And when you change the URL to pass the pirate locale (http://localhost:3000?locale=pirate), you’ll get:

OK! Now let’s add a timestamp to the view, so we can demo the date/time localization feature as well. To localize the time format you pass the Time object to I18n.l or (preferably) use Rails' #l helper. You can pick a format by passing the :format option — by default the :default format is used.

And in our pirate translations file let’s add a time format (it’s already there in Rails' defaults for English):

So that would give you:

When you are using the default SimpleStore, shipped with the i18n library, you store dictionaries in plain-text files on the disc. Putting translations for all parts of your application in one file per locale could be hard to manage. You can store these files in a hierarchy which makes sense to you.

For example, your config/locale directory could look like this:

This way, you can separate model and model attribute names from text inside views, and all of this from the "defaults" (eg. date and time formats).

Other stores for the i18n library could provide different means of such separation.

Do check the Rails i18n Wiki for list of tools available for managing translations.

You should have good understanding of using the i18n library now, knowing all neccessary aspects of internationalizing a basic Rails application. In the following chapters, we’ll cover it’s features in more depth.

Covered are features like these:

Translations are looked up by keys which can be both Symbols or Strings, so these calls are equivalent:

translate also takes a :scope option which can contain one or many additional keys that will be used to specify a “namespace” or scope for a translation key:

This looks up the :invalid message in the Active Record error messages.

Additionally, both the key and scopes can be specified as dot separated keys as in:

Thus the following calls are equivalent:

When a default option is given its value will be returned if the translation is missing:

If the default value is a Symbol it will be used as a key and translated. One can provide multiple values as default. The first one that results in a value will be returned.

E.g. the following first tries to translate the key :missing and then the key :also_missing. As both do not yield a result the string "Not here" will be returned:

To lookup multiple translations at once an array of keys can be passed:

Also, a key can translate to a (potentially nested) hash as grouped translations. E.g. one can receive all Active Record error messages as a Hash with:

In many cases you want to abstract your translations so that variables can be interpolated into the translation. For this reason the I18n API provides an interpolation feature.

All options besides :default and :scope that are passed to #translate will be interpolated to the translation:

If a translation uses :default or :scope as a interpolation variable an I+18n::ReservedInterpolationKey+ exception is raised. If a translation expects an interpolation variable but it has not been passed to #translate an I18n::MissingInterpolationArgument exception is raised.

In English there’s only a singular and a plural form for a given string, e.g. "1 message" and "2 messages". Other languages (Arabic, Japanese, Russian and many more) have different grammars that have additional or less plural forms. Thus, the I18n API provides a flexible pluralization feature.

The :count interpolation variable has a special role in that it both is interpolated to the translation and used to pick a pluralization from the translations according to the pluralization rules defined by CLDR:

The algorithm for pluralizations in :en is as simple as:

I.e. the translation denoted as :one is regarded as singular, the other is used as plural (including the count being zero).

If the lookup for the key does not return an Hash suitable for pluralization an 18n::InvalidPluralizationData exception is raised.

The locale can be either set pseudo-globally to I18n.locale (which uses Thread.current like, e.g., Time.zone) or can be passed as an option to #translate and #localize.

If no locale is passed I18n.locale is used:

Explicitely passing a locale:

I18n.locale defaults to I18n.default_locale which defaults to :en. The default locale can be set like this:

The shipped Simple backend allows you to store translations in both plain Ruby and YAML format. [2]

For example a Ruby Hash providing translations can look like this:

The equivalent YAML file would look like this:

As you see in both cases the toplevel key is the locale. :foo is a namespace key and :bar is the key for the translation "baz".

Here is a "real" example from the ActiveSupport en.yml translations YAML file:

So, all of the following equivalent lookups will return the :short date format "%B %d":

Generally we recommend using YAML as a format for storing translations. There are cases though where you want to store Ruby lambdas as part of your locale data, e.g. for special date.

You can use the methods Model.human_name and Model.human_attribute_name(attribute) to transparently lookup translations for your model and attribute names.

For example when you add the following translations:

Then User.human_name will return "Dude" and User.human_attribute_name(:login) will return "Handle".

Active Record validation error messages can also be translated easily. Active Record gives you a couple of namespaces where you can place your message translations in order to provide different messages and translation for certain models, attributes and/or validations. It also transparently takes single table inheritance into account.

This gives you quite powerful means to flexibly adjust your messages to your application’s needs.

Consider a User model with a validates_presence_of validation for the name attribute like this:

The key for the error message in this case is :blank. Active Record will lookup this key in the namespaces:

Thus, in our example it will try the following keys in this order and return the first result:

When your models are additionally using inheritance then the messages are looked up for the inherited model class names are looked up.

For example, you might have an Admin model inheriting from User:

Then Active Record will look for messages in this order:

This way you can provide special translations for various error messages at different points in your models inheritance chain and in the attributes, models or default scopes.

The translated model name, translated attribute name, and value are always available for interpolation.

count, where available, can be used for pluralization if present:

If you are using the Active Record error_messages_for helper you will want to add translations for it.

Rails ships with the following translations:

Rails uses fixed strings and other localizations, such as format strings and other format information in a couple of helpers. Here’s a brief overview.

* ActiveRecord::Errors#full_messages prepends the attribute name to the error message using a separator that will be looked up from activerecord.errors.format.separator (and defaults to ' ').

For several reasons the shipped Simple backend only does the "simplest thing that ever could work" for Ruby on Rails [3] ... which means that it is only guaranteed to work for English and, as a side effect, languages that are very similar to English. Also, the simple backend is only capable of reading translations but can not dynamically store them to any format.

That does not mean you’re stuck with these limitations though. The Ruby I18n gem makes it very easy to exchange the Simple backend implementation with something else that fits better for your needs. E.g. you could exchange it with Globalize’s Static backend:

The I18n API defines the following exceptions that will be raised by backends when the corresponding unexpected conditions occur:

The I18n API will catch all of these exceptions when they were thrown in the backend and pass them to the default_exception_handler method. This method will re-raise all exceptions except for MissingTranslationData exceptions. When a MissingTranslationData exception has been caught it will return the exception’s error message string containing the missing key/scope.

The reason for this is that during development you’d usually want your views to still render even though a translation is missing.

In other contexts you might want to change this behaviour though. E.g. the default exception handling does not allow to catch missing translations during automated tests easily. For this purpose a different exception handler can be specified. The specified exception handler must be a method on the I18n module:

This would re-raise all caught exceptions including MissingTranslationData.

Another example where the default behaviour is less desirable is the Rails TranslationHelper which provides the method #t (as well as #translate). When a MissingTranslationData exception occurs in this context the helper wraps the message into a span with the CSS class translation_missing.

To do so the helper forces I18n#translate to raise exceptions no matter what exception handler is defined by setting the :raise option:

At this point you hopefully have a good overview about how I18n support in Ruby on Rails works and are ready to start translating your project.

If you find anything missing or wrong in this guide please file a ticket on our issue tracker. If you want to discuss certain portions or have questions please sign up to our mailinglist.

I18n support in Ruby on Rails was introduced in the release 2.2 and is still evolving. The project follows the good Ruby on Rails development tradition of evolving solutions in plugins and real applications first and then cherry-picking the best bread of most widely useful features second for inclusion to the core.

Thus we encourage everybody to experiment with new ideas and features in plugins or other libraries and make them available to the community. (Don’t forget to announce your work on our mailinglist!)

If you find your own locale (language) missing from our example translations data repository for Ruby on Rails, please fork the repository, add your data and send a pull request.

If you found this guide useful please consider recommending its authors on workingwithrails.

[1] Or, to quote Wikipedia: "Internationalization is the process of designing a software application so that it can be adapted to various languages and regions without engineering changes. Localization is the process of adapting software for a specific region or language by adding locale-specific components and translating text."

[2] Other backends might allow or require to use other formats, e.g. a GetText backend might allow to read GetText files.

[3] One of these reasons is that we don’t want to any unnecessary load for applications that do not need any I18n capabilities, so we need to keep the I18n library as simple as possible for English. Another reason is that it is virtually impossible to implement a one-fits-all solution for all problems related to I18n for all existing languages. So a solution that allows us to exchange the entire implementation easily is appropriate anyway. This also makes it much easier to experiment with custom features and extensions.

Lighthouse ticket