ubiquity-slideshow-ubuntu: contemplating some string changes

Danilo Šegan danilo at canonical.com
Thu Apr 1 16:03:13 BST 2010


Hi Khaled,

I am sorry you find this missing feature of Launchpad a deal-breaker.
Here's a few explanations on what's the case with fuzzy matching, and
other ways you can work-around it.

У пет, 26. 03 2010. у 00:00 +0200, Khaled Hosny пише:

> What I can't really understand how a very simple and basic feature 
> like marking old translation as fuzzy when merging new templates
> doesn't yet exist! some thing that gettext tools had years before my birth. 

I am sorry to hear you feel this way. gettext fuzzy matching works very
well for cases like typos. Unfortunately, it also fails for anything
else because it uses an algorithm based on character counting.  I've
seen numerous cases were translations in Ubuntu were 'approved' from
incorrect fuzzy suggestions.  At the same time, majority of messages in
Ubuntu are "short" where it does more harm than good.  And with long
messages, it's even harder to notice wrong translation if it differs in
something like "not".

Also, all the fuzzy flag handling stuff made code very complex and UI
even more so: we had two concepts of "unreviewed" messages, and people
never understood what's really going on. So, we did away with the fuzzy
handling.

I'll attach two PO small PO files which demonstrate how it's not really
perfect algorithm: i.e. it recognizes "winds" as similar to "windows".
I am pretty sure most languages translate these differently.

To try it out, do "msgmerge windows.po new.pot" and see how it works
with fuzzy matching. I've seen even worse examples where the two short
words might even have completely opposite and not just unrelated
meanings.

I'd like us to have more time to implement an excellent tool for doing
similarity matching that doesn't suck. Word-based at least, but perhaps
trigram (three connected words) based would be even better. We can't do
it trivially because of performance considerations, but there is a clear
path forward.  If you have some time and skill to help us get there
faster, I'd be more than happy to discuss it with you.  There is only so
much we can do at any given time, and as David mentioned, we are focused
on getting direct upstream translation imports into Launchpad and
Ubuntu.

The old blueprint describing this feature is:


https://blueprints.edge.launchpad.net/rosetta/+spec/rosetta-fuzzy-merge

As you can see from the "related bugs" section there: we get bug reports
about it all the time. And we've only recently came to a point where we
can spend time on implementing new features (such as bazaar integration,
and translations sharing between series, direct upstream import).
Eventually, we'll get to this as well, and to things like glossaries,
full APIs, discussions on translations and so forth.

> I'm really annoyed by this, my time is more valuable to waste it redoing
> translations over and over again. The fact that launchpad is annoying me
> more and more every day, that I might just abandon doing any Ubuntu
> translation and save my valuable time to something else.

I am really sorry Launchpad causes so much waste of your time. However,
apart from the workaround Gabor mentioned, there is another: download
latest POT/PO file and do your merging locally using gettext tools, and
you'll get exactly the same benefit msgmerge (from gettext) provides you
otherwise.

You'll have to go through all the fuzzy messages in an offline PO editor
(like KBabel, GTranslator, POEdit...), but once you are done you can
re-upload it (don't forget to strip fuzzy flags off).

I don't see how this is different than what happens anywhere else (like
GNOME, KDE, or anywhere else at all). It's just that Launchpad otherwise
hides all the complexity so doing this step that is painful everywhere
is as painful with Launchpad, instead of being simple as everything else
is simple in Launchpad. But then again, I *am* biased.

Cheers,
Danilo
-------------- next part --------------
A non-text attachment was scrubbed...
Name: new.pot
Type: text/x-gettext-translation-template
Size: 262 bytes
Desc: not available
Url : https://lists.ubuntu.com/archives/ubuntu-translators/attachments/20100401/445b82d4/attachment.bin 
-------------- next part --------------
A non-text attachment was scrubbed...
Name: windows.po
Type: text/x-gettext-translation
Size: 286 bytes
Desc: not available
Url : https://lists.ubuntu.com/archives/ubuntu-translators/attachments/20100401/445b82d4/attachment-0001.bin 


More information about the ubuntu-translators mailing list