Notice

This page show a previous version of the article

Adding a New Language to the Corpus (for Developers)

Introduction

These are the instructions for adding a language to the Tatoeba corpus. Instructions for adding a language in which the Tatoeba UI will be displayed are found elsewhere.

These instructions were modified from Assembla.

FAQ for users

The FAQ for users who want to add a new language: How to request a new language.

Language icon

Create the icon for the language. The icon should be a PNG file of dimension 30x20. On each icon there is (in theory) a 1px line of color #dcdcdc on the border bottom and right. Most of the icons also have gone through a luminosity change, so that they are a bit more pale than the original image. Anyway, most of it doesn't matter right now. The most important is that the icon is a PNG file of dimension 30x20.
Commit the image to the repository. The icons are stored in the app/webroot/img/flags folder. Ask one of the people with repository access if you don't have it yourself and don't want to obtain it.
Update the app/webroot/img/flags directory on the server, to retrieve the new images for the new languages.

Source code

There is a script to add the new language code to the appropriate files. It takes the following parameters:

three-letter ISO 639-code (e.g., "epo" for Esperanto")
the English name of the language (e.g., "Esperanto")
the ID of a list containing at least 5 sentences in the given language (to find the ID of the language, search the list of lists for the appropriate list, then roll over it with your mouse to see the URL, which contains the list ID)
the string "dev" (on a development machine) or "prod" (on the server)
the username for the database
the password for the database
the database name

It should be executed from the directory in which the script is located.

Once this script is executed, the new languages will be available. You should execute it locally and verify that it works correctly. (See the "Tests" section below.) The sentences that were in the list with id <list_id> should have their language set to the new language (instead of being set to language unknown). This script edits the following files:

app/model/sentence.php Adds the language ISO code to the $validate array. Languages that are not part of this array are not allowed.
app/views/helpers/languages.php Adds the language ISO code and the name to the languagesArray() method.
docs/generate_sphinx_conf.php Adds the language ISO code and name to the $languages array. Also adds the ISO code to the $cjkLanguages array if the language uses Chinese, Japanese or Korean characters.

After you make your changes and test them according to the next section, commit your code to the repository, or have someone do it for you. See Repositories. Refer to the appropriate issue ticket(s) in your comment and indicate the languages that were added.

Configure Sphinx

Sphinx have a list of characters that are considered as used in words. Every character not in this list is ignored (they are used as word separators). So you need to make sure that the characters used by the language are in this list. If you’re sure they are, you can skip reading this section.

You need to have a list of the unicode code points used by the language and make sure they are defined into the charset_table option that is defined in generate_sphinx_conf.php. Include case folding rules if appropriate. (gillux can help with that.)

Refresh the Sphinx indexes

Add some real sentences in the language, so that you can check if searching works after reindexing. Then, regenerate the Sphinx config using the generate_sphinx_conf.php script, and refresh the Sphinx index for that new language (indexer index_XXX --rotate).

Tests

Test that the language detection works (or can work) by adding a sentence with 'auto-detect'. There should be on Tatoeba a list of sentences in the language in question (named after the language in question).
Test that you can change the language of a sentence into the language in question.
Check that the count displays properly in the languages stats.
Check that search is working by looking up a word used in sentences that has been added before reindexing.
Check if the random sentence feature is working.

On the server

If everything is fine so far, go to the server.
Execute "htop" to verify that the load is below 2.
Update the repository if necessary
Connect to the mysql database of the prod version.
CALL add_new_language(iso_code, list_id, tag_name);
exit
Check that the sentences that were in the list and tags have now the appropriate icon.
Check that the language appears in the languages stats.
cp /usr/local/etc/sphinx.conf /usr/local/etc/sphinx.conf.old
php generate_sphinx_conf.php > /usr/local/etc/sphinx.conf
Change the necessary things in the new config file (user, password, database and port). Look at the old conf file for reference.
indexer --all --rotate & disown

Historical information only

We used to follow this procedure for adding a new language (example: French):

Create folder : /app/locale/fre/LC_MESSAGES
Copy-paste default.pot into this folder
Change it into default.po
Open default.po with PoEdit (http://www.poedit.net/) and translate.
Save. It will generate a *.mo file, which is used when replacing strings at runtime.

and when new strings were added:

Follow the cake i18n instructions to generate the up-to-date POT file.
Open the PO file (PO, not POT).
In the menu : Catalog > Update from POT file…
Choose the POT file that was newly generated

The language of the page is set through the URL. Example: http://localhost/tatoeba2/fre/sentences/index

Language detection

In the past, we used to edit this:

app/controllers/components/google_language_api.php Adds the corresponding case to the google2TatoebaCode() method, if Google supports the detection for the language. See the Language enum.

but now tatodetect takes care of language detection.

Resources:

http://blog.jaysalvat.com/articles/choix-des-langues-par-url-dans-cakephp.php
http://www.formation-cakephp.com/41/multilingue-18n-l10n