As someone who has been forced to contribute due to someone using an LLM
to make ridiculously bad translations, this is an absolutely appalling idea.
For minority languages these things are a disaster as they will
inevitably feed on their own output eventually.
Ar 30/9/26 8:44 i.n., scríobh Neil Roberts:
Hi,
It might help make the proposal more convincing if you could find some
examples where one these AIs finds some errors.
I think one of the major sources of translation mistakes results from
translating short strings without being able to look at the context. For
example, sometimes it’s hard to tell if an English word is being used as
a noun or a verb, or if it’s the imperative or the infinitive, or
sometimes a single English word should be translated to several
different words in other languages depending on the context. I find it
hard to imagine the AI would be able to help in this area because it
doesn’t have access to the context either.
Otherwise, for detecting spelling mistakes or grammatical errors, maybe
it would make more sense to use normal spelling and grammar checkers
without having to involve AI.
Personally I think it would be kind of a shame for TDF to give donaters’
money to these AI companies with questionable ethics. I would prefer to
spend the resources on improving Weblate, for example automatically
adding screenshots for groups of strings so we can easily see where the
string is used.
Regards,
– Neil
Valter Mura<[email protected]> writes:
Hi All,
It is currently possible to perform quality checks on translations, on
UI and Online Help, using AI (I'm thinking of Claude, ChatGPT, etc).
These kind of checks are quite fast (for AI) and would detect
inconsistencies, errors, typos, etc.
However, using AI for this kind of work usually requires a subscription,
the cost of which only the Foundation can cover.
Would it be possible to implement such a system? I believe it could
rapidly improve the quality of translations across all supported
languages. I am thinking of Italian, of course, but I believe all
languages would benefit.
With "implementation", I mean something like an area with integrated AI
where the user or translator can connect and provide instructions to run
the check — for example, a set of public folders containing the files to
be verified. Or simply an access granted to the "team leaders" to the AI
interface.
The translation process itself obviously would remain unchanged; this
check would simply identify errors that often escape the human eye,
especially given the large volume of translated files involved.
What do you think about it?
Ciao!
--
Valter
Free Software is better!
LibreOffice:www.libreoffice.org
KDE:www.kde.org
La bici non consuma e non produce emissioni, non ingombra e non fa
rumore: è un mezzo ecologico per definizione e ad alta efficienza
energetica.
La bicicletta migliora il traffico, l'ambiente, la salute e l'umore.
--
To unsubscribe e-mail to:[email protected]
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
--
`~:.,.:'^`~:.,.:'^`~:.,.:'^`~:.,.:'^`~:.,.:'^`~:.,.:'^`~:.,.:'^`~:.,.:'^`
S e á n Ó S é a g h d h a sean.anseo [at] gmail.com
Is trom cearc i bhfad. Seanfhocal.
--
To unsubscribe e-mail to: [email protected]
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
