[Bug 95274] Wrong editing languages offered

bugzilla-daemon at bugs.documentfoundation.org bugzilla-daemon at bugs.documentfoundation.org
Wed May 29 12:08:19 UTC 2024


https://bugs.documentfoundation.org/show_bug.cgi?id=95274

--- Comment #58 from Jonathan Clark <jonathan at libreoffice.org> ---
(In reply to Heiko Tietze from comment #57)
> Jonathan, do you have an opinion on this?

My vote is for the following, in this order:

1.) Languages that already exist in the current document. If a language has
already been used in a document, I think it's likely that it will be used
again. Those languages should be positioned prominently.

2.) Languages that the user has *explicitly* specified as languages they
understand and intend to use with LibreOffice. The user is the best judge of
this, so we should give them the opportunity to tell us. I don't think it will
ever be possible to guess this right every time for every user.

3.) Languages derived heuristically from LO configuration: default languages
for documents, user interface language.

I think installed spellcheck dictionaries is a weak signal. My LO Snap install
was bundled with dictionaries for many languages I don't know, and I'd rather
not see them in this list.

4.) Languages derived heuristically from system configuration. This is a
reasonable starting point, but is only a rough guess. There are many reasons
why a user's system configuration might not reflect all of their languages.

For an extreme example, consider Linux: English-primary users can't set a
second language on Linux. If you try, it will break localization in most
gettext programs. An English-locale Linux user with a US international keyboard
might need to regularly work in dozens of languages, but we'd have no way of
knowing which ones.


Regarding libexttextcat: Based on the above discussion, am I correct that this
is being used on individual words? Given cognates and loanwords, I don't expect
classifying individual words can ever be reliable. The docs say they expect
"hundreds of bytes", which seems more reasonable.

Instead of using this to generate confusing recommendations, perhaps this could
be used somehow for recommending the best match from the high-signal
candidates?

-- 
You are receiving this mail because:
You are on the CC list for the bug.


More information about the Libreoffice-ux-advise mailing list