Re: [libreoffice-l10n] Unusual behavior of Weblate

Hi Mihail,

On Mon, Feb 6, 2023 at 11:46 AM Mihail Balabanov <m.balabanov@gmail.com> wrote:
> >
> I noticed two unusual things working with Weblate lately:
> - There are sporadic 'Delete entry' buttons. Why do they appear only for a
> small percentage of the entries and what do they actually do?

I assume you refer to the entries in the "Automatic Suggestion" area
from the weblate translation memory.

Weblate has a three-tier translation memory:
* the project's translation memory, that can also optionally create
all new entries into an
* instance-wide shared translation memory
* and finally a user-specific translation memory

So you'll see delete entry buttons for the strings coming from your
user translation memory, for strings that you translated in the past
within weblate.

> I would very
> much like to delete erroneous or obsolete entries but the button never
> appears on any of them.

If you point me to those invalid entries, I can remove them.

> - Sometimes there is no similarity percentage. This hampers productivity
> noticeably because when I know that a string is a 100% match, I can just
> click through, otherwise I have to copy and compare in an external program.

Also please provide a link for that.

> This link illustrates both:
>
> https://translations.documentfoundation.org/translate/libo_help-master/textscalc01/bg/?q=state%3A
> <translated&offset=1

Those kind of links are not stable, what it points to is depending on
what you translate, better use the "copy permalink" option (the little
chain-link icon at the top-right of the translation area)

Looking at 
https://translations.documentfoundation.org/translate/libo_help-master/textscalc01/bg/?checksum=6ea269b7376bb4cb
I see it has entries from weblate's translation memory as well as from
the amagama server, the tmserver entries. That contains old strings
from pootle, from the time before weblate was used. So old strings
from tmserver, new strings from weblate's translation memory.
And I guess weblate just doesn't return amagama's quality score
(Levenshtein distance), but all results returned by amagama have at
least a 75% confidence (from amaGama's POV), but that's just comparing
against the source string, so not sure whether it is suitable to also
show it....

curl http://localhost:8888/api/v1/en/bg/unit/is%20less%20than

[{"quality": 75.0, "rank": 100.00000000000024, "source": "less than",
"target": "по-малко"}, {"quality": 75.0, "rank": 100.00000000000024,
"source": "less than", "target": "По-малко"}, {"quality": 75.0,
"rank": 100.00000000000024, "source": "less than", "target": "по-малка
от"}, {"quality": 75.0, "rank": 100.00000000000024, "source": "Is Less
Than", "target": "По-малко"}, {"quality": 71.42857142857143, "rank":
100.00000000000024, "source": "Much less than", "target": "Много
по-малко"}]

(ignore rank, that was always flagged as internal flag that can
disappear in future)

-- 
To unsubscribe e-mail to: l10n+unsubscribe@global.libreoffice.org
Problems? https://www.libreoffice.org/get-help/mailing-lists/how-to-unsubscribe/
Posting guidelines + more: https://wiki.documentfoundation.org/Netiquette
List archive: https://listarchives.libreoffice.org/global/l10n/
Privacy Policy: https://www.documentfoundation.org/privacy