Friedel en ander frappanthede

I had the pleasure of meeting Claude Paroz, the maintainer of Damned Lies (the system for localisation statistics and workflow at GNOME). We started thinking about how Damned Lies might be able to provide guidance for smaller teams on prioritisation, and statistics that might provide better help with problems people want to address.

From time to time, GNOME localisers discuss the localisation statistics, and people often mention the arbitrary method used for calculating "supported" status. Of course, there is a reason why Damned Lies has the name it has, and teams should probably not give too much attention to the numbers. However, Damned Lies also functions as a place where decisions are taken (which module should be translated next), and where people get their motivation (progress between releases).

People came up with more specific suggestions after the release of the GNOME i18n survey in October 2010, with lots of discussion in this thread started by Gil Forcada. People were discussing the idea of ignoring some possibly less important strings for translation for better statistics (lies), better prioritisation, and better motivation. GConf schema strings are always mentioned, as well as the developer tools, and there was also talk of error messages and things that are only printed on the console. Identifying error messages and console messages is not trivial, but it is mostly trivial to identify GConf schema strings. Gil also published the numbers to show what portion of the total work is taken up by GConf strings:

• 3158 strings out of 45785 (6.8974%)
• 35074 words out of 205155 (17.096%)

That is a humongous amount of work. Not translating GConf strings means that your language has almost no chance of being "supported". Extra reasons for leaving these out could be that they are not really aimed at the end-user as much as other strings, and that they tend to be more technical and therefore more demotivating. A lot of teams also give a lower prioritisation to the developer tools like Anjuta.

I wrote earlier about trimming a localisation task to get it to a manageable size. Not everyone can easily do this kind of thing, but with our work in Africa, this is probably what most languages should do before they move onto doing more. If we could provide this kind of help through Damned Lies, we could hopefully help teams to put their time into what is probably most important, and hopefully provide better motivation that might help us to retain more translators and build more successful teams. In the end, I hope that users will receive the main benefit, since GNOME can be available in more languages, and that partial translations are more likely to have translations where it matters to the average user.

Lots of the details still need to be worked out, but it feels as if this is within reach. Claude already has ideas of how he wants to approach the changes to Damned Lies (probably the biggest job), and I have some ideas about trimming modules to a smaller size. For a start, just ignoring GConf schema strings can already help a lot.