Not happy at all

[Ask Hjorth Larsen]

On Thu, Dec 23, 2010 at 1:24 PM, Danilo Šegan <danilo op canonical.com> wrote:
> Hi Ask,
>
> У сре, 22. 12 2010. у 13:01 +0100, Ask Hjorth Larsen пише:
>
>> It's unfortunate that the `fuzzy' feature of gettext is not supported
>> in LP, and there have been previous discussions in the manual team
>> about what to do, but there's no really good solution as it is.
>
> Perhaps we should really just turn fuzzy translations into suggestions.
> It wouldn't be a big deal for us, but:
>
>  1. people *will* mistakenly approve suggestions which look almost the
> same (imagine someone fixing a typo of a missing "not" in a long
> documentation message: I will bet you that most people will not notice
> the "not" and will just approve the existing translation); this is a
> problem even with offline translation, but Launchpad makes it very easy
> to make this mistake for reviewers
>  2. if msgmerge was not done prior to the import, there won't be any
> benefit
>  3. people will expect this to actually work, and yet it wouldn't (iow,
> we'll get questions why was there not a suggestion for one string or
> another, and those questions are nothing but a distraction); today we
> just get feature requests about this :)
>
> Those are the reasons why it's not behaving like that today.  We do,
> however, want to have a smart "similarity matching" feature for
> suggestions.  That's a lot more work.  I'd be happy to guide anyone with
> some time on their hands into implementing that.
>
> Cheers,
> Danilo

Hi Danilo

I agree that fuzzies should not be labeled as "suggestions" for the
reasons you mention.  But it would be great the fuzzies could be
displayed with a different label like "Fuzzy", although possibly more
descriptive like "This string may be similar to...".  Then the
reviewer will know that this was not a suggestion as such, and will be
more careful.  (If the project expects high-quality translations, the
reviewer will presumably be a member of a translation team, and can be
expected to at least follow some basic instructions.)

If this is too much of an interface change, the text accompanying the
suggestion could instead be changed from the present form:

  Suggested in coreutils in Ubuntu Intrepid package "coreutils" by Ask
Hjorth Larsen on 2010-03-08

to something like

  This suggestion was chosen by a computer based on existing similar strings

Wouldn't this mostly solve the fuzzy issue, and require less work than
a new similarity matching feature?  Sorry to further derail the
discussion.  (It's unrealistic for me to help with implementation for
the next 10 months or so, although I would love to if I had more time)

Best regards
Ask