Machine based translation of KDE using translate-po

Initial one-time preparations

• Install https://github.com/zcribe/translate-po according to the instructions given there (pip install)

• Install the latest googletrans,

     pip install googletrans==3.1.0a0

  to address translate-po issue 12

• Download perl module+script Text::PO::SimpleParser + pomerge and unpack it in your $HOME/bin/

• Check out your required parts from svn.kde.org/home/kde/trunk/l10n-kf5. For Esperanto, this would be

     svn co -N svn+ssh://svn@svn.kde.org/home/kde/trunk/l10n-kf5
     cd l10n-kf5
     svn up --set-depth=empty templates eo
     svn up templates/messages eo/messages
     cd eo/messages

• Save the following script to $HOME/bin/postproc_xlpo.sh and give it execute permission (chmod +x)

     #!/bin/bash
     # postproc_xlpo.sh
     # Restore line breaks between po entries
     perl -p -e 'BEGIN { $hadcomment = 1; }
             my $l = $_;
             if ($l =~ /^#/) {
               unless ($hadcomment) {
                 print "\n";
                 $hadcomment = 1;
               }
             } else {
               $hadcomment = 0;
             }
             ' -i $*
     # Remove space inserted between % and argument number
     perl -p -e 's/% (\d)/%$1/g' -i $*

• Create working directory in which the machine translation shall be generated
  (inside the messages subdir of the target language, e.g. eo/messages) :

     mkdir xlpo

• Make Python script $HOME/bin/tr.py which calls translate_po and give it execute permission (chmod +x); also, adjust the to= argument to your target language:

     #!/usr/bin/python3
     import translate_po
     from translate_po.main import run
     run(fro="en", to="eo", src="../../../templates/messages/kmuddy", dest=".")

• Note that in this script, the name "kmuddy" will have to be changed for each project to translate.

Example of translation procedure: kmuddy

• Create new directory in $L/messages

     svn mkdir $L/messages/kmuddy

• Copy the templates messages *.pot to the $L messages *.po
  (the exact number of files and filenames are project specific)

     svn copy templates/messages/kmuddy/kmuddy._desktop_.pot $L/messages/kmuddy/kmuddy._desktop_.po
     svn copy templates/messages/kmuddy/kmuddy.pot $L/messages/kmuddy/kmuddy.po

• In the new $L/messages po-files manually insert the proper PO header lines.

• translate_po requires input filenames *.po instead of *.pot, thus prepare symbolic links:

     cd templates/messages/kmuddy
     ln -s kmuddy._desktop_.pot kmuddy._desktop_.po
     ln -s kmuddy.pot kmuddy.po
     cd ../../.. 

  Note that this only works for pot-files with less than 1000 phrases (msgid).
  If a file has more phrases, it is necessary to split it into multiple po-files where each has less than 1000 phrases.
  You will also have to copy the POT header lines to the split files such that they are valid as po-files.
  After the machine translation, you will have to rejoin the split files into a complete po-file (also removing the additional PO-headers).

• Start the machine translation

     cd $L/messages/xlpo
     tr.py

• After completion, in xlpo there will be the translated po-files; in this example: kmuddy._desktop_.po, kmuddy.po

• Apply required postprocessing to the generated files:

     postproc_xlpo.sh *.po

• Open the generated *.po files in a text editor and

  • delete empty first line
  • manually proofread/fix the translations (sometimes words will be mistranslated, see sample table below)

• Merge the generated content in xlpo onto the final header+skeleton in $L/messages

     pomerge.pl ../kmuddy/kmuddy._desktop_.po kmuddy._desktop_.po > k.po
     # Make sure that the merged k.po is in good shape (run msgfmt --statistics --check-header), then move it to the target:
     mv k.po ../kmuddy/kmuddy._desktop_.po
     pomerge.pl ../kmuddy/kmuddy.po kmuddy.po > k.po
     # Make sure that the merged k.po is in good shape (run msgfmt --statistics --check-header), then move it to the target:
     mv k.po ../kmuddy/kmuddy.po

• Update the datestamps "PO-Revision-Date" in the target files

• Check new directory into Subversion

     cd ../
     svn ci -m"l10n-kf5/$L/messages : add kmuddy" kmuddy

Common mistranslations of Google Translate with target Esperanto

Note that this is not intended as criticism of Google, they do an amazing job given the many ambiguities of English