RNineDo opened a new issue, #42536:
URL: https://github.com/apache/superset/issues/42536
Hi @rusackas — following #42099 and #41827, which backfilled several catalogs
with AI-generated translations explicitly marked *needs review*, I'd like to
do
that review pass for French. I'm a native French speaker.
Before opening anything, here is what I measured on
`superset/translations/fr/LC_MESSAGES/messages.po` at `master`, so we can
agree
on the scope first.
**Current state — 5039 entries**
| | count |
|---|---|
| `fuzzy` entries | 1124 |
| … of which carry the `backfill_po.py (claude-sonnet-4-6)` comment | 219 |
| … the remaining ones, re-fuzzied by `pybabel update` on changed sources |
905 |
| entries with an empty `msgstr` (not fuzzy) | 93 |
**Proposed scope for a first PR** — one file, `fr/LC_MESSAGES/messages.po`,
same shape as #42137 for `pt_PT`:
1. **89 untranslated strings translated.** The other 4 are deliberately left
empty: `schema1,schema2`, `superset.example.com`, `your-project-1234-a1`
are
listed in `do-not-translate.txt`, and `%(name)s.%(extension)s` is a
filename
template.
2. **38 of the 219 AI-generated entries corrected**, on objective grounds
only:
- *Terminology drift.* The backfill picked terms that contradict the
already-validated part of the catalogue. Counting only entries where the
English term appears exactly once, the validated entries use
`dataset` → « ensemble de données » (43 vs 23 for « jeu de données »),
`metric` → « mesure » (77 vs 3 for « métrique »),
`tooltip` → « infobulle » (12 vs 5), `layer` → « couche » (45 vs 3).
The AI entries use the minority term throughout.
- *French typography.* French takes a non-breaking space before a colon.
The validated entries do this 121 times (U+00A0) against 10 plain
spaces;
the AI entries use a plain space in all 10 of their occurrences.
- *Two entries* are flagged as translated but have an empty `msgstr`
(`Added 1 new column to the virtual dataset`, `Delayed (missed %s
refresh)`).
The `fuzzy` flag and the `Machine-translated via backfill_po.py` comment
are
removed **only** on entries actually reviewed. The rest keep them.
I checked the whole batch for placeholder integrity and
`do-not-translate.txt`
violations first — it is clean on both counts, so this PR is about wording
and
typography, not breakage.
The remaining ~1086 `fuzzy` entries (the `pybabel update` ones) are a larger
job;
I'd rather keep them out of this PR and propose follow-up batches if this one
lands well.
Does this scope work for you, and is a single-file PR on the French
catalogue the
shape you want? Happy to split it differently — I'd just rather agree before
opening it so we don't end up doing the work twice.
--
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
To unsubscribe, e-mail: [email protected]
For queries about this service, please contact Infrastructure at:
[email protected]
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
