Tuesday, October 20, 2009

MediaWiki, localisation and scrum

For all kinds of reasons I am learning about agile and scrum. The one thing that will be important to me is how to combine this with internationalisation and localisation. I just read a blog about this by Colin Cooper.

Colin writes his observations from the perspective of a commercial company. This is quite different from what we do at translatewiki.net for Wikipedia and all the other MediaWiki implementations. However, even when there is no "official" scrum or agile project management in place, I find that it is possible to describe what we do in these terms.

The first key thing is the recognition that the language communities themselves are responsible for the quality of their localisation. When a language is not able or willing to keep up the localisation, it is essentially their problem. We do have in the developers at translatewiki.net a group of people who spend a lot of time on the Internationalisation and the quality of the English that is used. This is one essential requirement that Colin mentions for successful localisation. We add to this the annotation of the context of the messages in our "qqq" messages.

MediaWiki knows in effect three stages for the development of software;
  1. Software that is being developed, maintained
  2. Software that is in production on WMF projects
  3. Software that is available in a stable release of MediaWiki
There used to be a waterfall of localisations; the localisations were entered in the first stage. When software was implemented, the localisations available at that time became available. In general step 1 was a continuous process, step 2 happened every few weeks and step 3 happened once every 3 or 4 months. To improve the available localisations, it happened that patches or "language packs" were made available for the stable releases.

With the implementation of LocalisationUpdate localisations became available on a daily basis everywhere after the technical translatewiki.net Q&A. This availability depends on the stability of the message. Given that both the development and Internationalisation happen in step 1, the messages are typically quite stable when the software goes into production.

Colin writes that the "waterfall" development method is problematic. The model for MediaWiki is one where the localisation is separate from the internationalisation. The provision of code with localisation is still essential but from a localisation process point of view it is no longer the moment of truth.

One thing a commercial i18n/L10n environment may want to implement differently is how the provisioning of the messages happens. There is no review of the language used in the localised messages. We assume good faith and are rarely disappointed.
Post a Comment