Converting a manual translation to po4a

Internally, each po4a parser reports the syntactical type of each extracted strings. This is how desynchronization are detected during the gettextization. For example, if the files have the following structure, it is very unlikely that the 4th string in translation (of type 'chapter') is the translation of the 4th string in original (of type 'paragraph'). It is more likely that a new paragraph was added to the original, or that two original paragraphs were merged together in the translation.

    Original         Translation

  chapter            chapter
    paragraph          paragraph
    paragraph          paragraph
    paragraph        chapter
  chapter              paragraph
    paragraph          paragraph

po4a-gettextize will verbosely diagnose any detected structure desynchronization. When this happens, you should manually edit the files (this probably requires that you have some notions of the target language). You must add fake paragraphs or remove some content in one of the documents (or both) to fix the reported disparities, until the structure of both documents perfectly match. Some tricks are given in the next section.

Even when the document is successfully processed, undetected disparities and silent errors are still possible. That is why any translation associated automatically by po4a-gettextize is marked as fuzzy to require an manual inspection by humans. One has to check that each retrieved msgstr is actually the translation of the associated msgid, and not the string before or after.

As you can see, the key here is to have the exact same structure in the translated document and in the original one. The best is to do the gettextization on the exact version of master.doc that was used for the translation, and only update the PO file against the latest master file once the gettextization was successful.

If you are lucky enough to have a a perfect match in the file structures, building a correct PO file is a matter of seconds. Otherwise, you will soon understand why this process has such an ugly name :) But remember that this grunt work is the price to pay to get the comfort of po4a afterward. Once converted, the synchronization between master documents and translations will always be fully automatic.

Even when things go wrong, gettextization often remains faster than translating everything again. I was able to gettextize the existing French translation of the whole Perl documentation in one day, even though the structure of many documents were desynchronized. That was more than two megabytes of original text (2 millions of characters): restarting the translation from scratch would have required several months of work.

Hints and tricks for the gettextization process

When this happens, the whole game comes down to the alignment of these damn files' structures again through manual edits. po4a-gettextize is rather verbose about what went wrong when it happens. It reports the strings that don't match, their positions in the text, and the type of each of them. Moreover, the PO file generated so far is dumped as gettextization.failed.po for further inspection.

Here are some other tricks to help you in this tedious process:

• Remove all extra content of the translations, such as the section giving credits to the translators. You can add them back in po4a afterward, using an addenda (see po4a(7)).
• If you need to edit the files to align their structures, you should prefer editing the translation if possible. Indeed, if the changes to the original are too intrusive, the old and new versions will not be matched during the PO update, and the corresponding translation will be dumped anyway. But do not hesitate to also edit the original document if required: the important thing is to get a first PO file to start with.
• Do not hesitate to kill any original content that would not exist in the translated version. This content will be automatically reintroduced afterward, when synchronizing the PO file with the document.
• You should probably inform the original author of any structural change in the translation that seems justified. Issues in the original document should reported to the author. Fixing them in your translation only fixes them for a part of the community. Plus, it is impossible to do so when using po4a ;)
• Sometimes, the paragraph content does match, but not their types. Fixing it is rather format-dependent. In POD and man, it often comes from the fact that one of them contains a line beginning with a white space while the other does not. In those formats, such paragraph cannot be wrapped and thus become a different type. Just remove the space and you are fine. It may also be a typo in the tag name in XML.

  Likewise, two paragraphs may get merged together in POD when the separating line contains some spaces, or when there is no empty line between the =item line and the content of the item.

• Sometimes, the desynchronization message seems odd because the translation is attached to the wrong original paragraph. It is the sign of an undetected issue earlier in the process. Search for the actual desynchronization point by inspecting gettextization.failed.po, and fix the problem where it really is.
• In some unfortunate settings, you will get the feeling that po4a ate some parts of the text, either the original or the translation. gettextization.failed.po indicates that both files matched as expected up to the paragraph N. But then, an (unsuccessful) attempt is made to match the N+1 paragraph in the original file not with the N+1 paragraph in the translation as it should, but with the N+2 paragraph. Just as if the N+1 paragraph that you see in the document simply disappeared from the file during the process.

  This unfortunate situation happens when the same paragraph is repeated over the document. In that case, no new entry is created in the PO file, but a new reference is added to the existing one instead.

  So, the previous situation occurs when two similar but different paragraphs are translated in the exact same way. This will apparently remove a paragraph of the translation. To fix the problem, it is sufficient to slightly alter one of the translations in the document. You can also prefer to kill the second paragraph in the original document.

  To the opposite, if the same paragraph appearing twice in the original document is not translated in the exact same way at both locations, you will get the feeling that one paragraph of the original document just vanished. Just copy the best translation over the other one in the translated document to fix the problem.

• As a final note, do not be too surprised if the first synchronization of your PO file takes a long time. This is because most of the msgid of the PO file resulting from the gettextization don't match exactly any element of the POT file built from the recent master files. This forces gettext to search for the closest one using a costly string proximity algorithm.

  For example, the first po4a-updatepo of the Perl documentation's French translation (5.5 MB PO file) took about 48 hours (yes, two days) while the subsequent ones only take a dozen of seconds.