i18n: AI translations — ar, as, bn, cs, de, es, fr, hi, hr, id, it, ja, ko, pl, pt, ro, ru, sc, te, tl, tr, uk, zh (2026-09-02)

[openlibrary-bot]

Languages

• Arabic (ar)
• Assamese (as)
• Bengali (bn)
• Czech (cs)
• German (de)
• Spanish (es)
• French (fr)
• Hindi (hi)
• Croatian (hr)
• Indonesian (id)
• Italian (it)
• Japanese (ja)
• Korean (ko)
• Polish (pl)
• Portuguese (pt)
• Romanian (ro)
• Russian (ru)
• Sardinian (sc)
• Telugu (te)
• Tagalog (tl)
• Turkish (tr)
• Ukrainian (uk)
• Chinese (zh)

Summary

AI-translated untranslated strings. Format strings, HTML attributes, and proper nouns
preserved verbatim. Fuzzy entries skipped.

Review guidance

• Format strings (%(name)s, %s, {page}) must match the English source exactly
• HTML tag structure must be identical to the English source
• If any translation looks wrong, open an issue — we can retranslate

AI attribution

Translations generated by AI translation API.

[openlibrary-bot]

Deviation note — branch creation

At session start, git status on main already showed uncommitted modifications to messages.pot and all 23 locale files in this batch (HEAD matched origin/main exactly, so these weren't leftover from a prior branch). This looks like the result of an extract/sync step that ran before this agent was invoked, but was never committed.

Per instructions, .po files are "already synced" and I should not run pull/sync myself — so I treated this pre-existing working tree diff as the synced baseline, branched from it (i18n/ai-batch-20260902 off origin/main), and committed it as a first commit ("i18n: sync messages.pot and locale files with latest source strings") instead of the documented empty commit, since the branch already had real diff from origin/main at creation time.

Flagging in case this indicates a workflow step (extract/update) is running but not committing its output before handing off to the translation agent.

[openlibrary-bot]

Run audit — ar, as, bn, cs, de, es, fr, hi, hr, id, it, ja, ko, pl, pt, ro, ru, sc, te, tl, tr, uk, zh — 2026-09-02

Stats

Every language in this batch had the same small set of newly-added source strings (5–6 entries, all related to TestingEnvironment.html.jinja deploy-status messages plus one admin/spamwords.html string).

Lang	Translated	Skipped	Cleared by fix	validate	test
ar	5	1	0	OK	2641 passed
as	5	1	0	OK	790 passed
bn	5	1	0	OK	120 passed
cs	5	1	0	OK	138 passed
de	5	1	0	OK	491 passed
es	5	0	0	OK	2641 passed
fr	5	1	0	OK	337 passed
hi	5	1	0	OK	736 passed
hr	5	0	0	OK	1440 passed
id	5	1	0	OK	313 passed
it	5	1	0	OK	1062 passed
ja	5	1	0	OK	159 passed
ko	5	0	0	OK	113 passed
pl	5	1	0	OK	761 passed
pt	5	1	0	OK	234 passed
ro	5	1	0	OK	1112 passed
ru	5	1	0	OK	168 passed
sc	5	0	0	OK	211 passed
te	5	1	0	OK	2641 passed
tl	5	0	0	OK	595 passed
tr	5	1	0	OK	2641 passed
uk	5	1	0	OK	120 passed
zh	5	1	0	OK	118 passed

The 5 translated strings per language are the %(user)s-suffixed variants of TestingEnvironment.html.jinja's deploy-status messages (Deploying, started %(time)s by %(user)s, its — %(stage)s variant, Deploy succeeded, Deploy failed, Last deploy). I cross-referenced each language's existing translation of the non-by %(user)s counterpart string already in the file so phrasing/register stayed consistent (e.g. de → "von %(user)s", fr → "par %(user)s", ja/ko/zh → parenthetical （%(user)s） since inserting an agent clause mid-sentence would have required reordering around the existing %(time)s placeholder).

Skipped entries

1 entry skipped in every language except es, hr, ko, sc, tl (which don't have it in their untranslated set): the admin/spamwords.html string "For example, if you want to prevent edits containing the domain <code>t.co</code>, add <code>(\x08|https://|http://)t\.co</code> to the spamwords list." — contains a literal \x08 backspace control character (source .pot bug; it should be the regex \b but got corrupted to an actual backspace byte). Per instructions, left msgstr empty. Not filed as a new issue — already tracked as #21.

Issues encountered

None blocking. No format errors were cleared by ./i18n fix for any language (all 0 in the table above).

Commands that failed

None.

Improvisations / deviations from instructions

• Branch was not zero-diff at creation. At session start, main (== origin/main, verified via git rev-parse) already had uncommitted modifications to messages.pot and all 23 batch locale files — an apparent extract/sync step had run before this agent was invoked but its output was never committed. Since instructions say the .po files are "already synced" and not to run pull/sync myself, I branched from this working-tree state (i18n/ai-batch-20260902) and committed it as a first commit ("i18n: sync messages.pot and locale files with latest source strings") instead of the documented empty commit, since the branch already had real diff from origin/main. Posted as a separate comment on this PR immediately when it happened.
• The sync commit's line-wrap formatting was partially undone by ./i18n fix/compile. The ambient sync diff had rewrapped some #: location comments and long msgid/msgstr lines to a narrower width; running this batch's apply/fix/compile on top reformatted those same lines back to the wider, single-line style the ./i18n toolbox normally produces. Net effect: per-language commits in this PR contain some cosmetic rewrap-reversion alongside the real new translations. This is the same phenomenon already documented in Checkout starts with uncommitted reformatting diff across messages.pot and all locale .po files #44 (Babel version/config mismatch between whatever produced the checked-in files and the Actions environment's Babel 2.18.0) — this run is corroborating evidence, no new issue filed.

Suggested workflow improvements

• Per Checkout starts with uncommitted reformatting diff across messages.pot and all locale .po files #44: pin the .po/.pot wrap width in the toolbox (or regenerate+commit a one-time reformat) so sync steps and ./i18n subcommands agree on formatting and stop bouncing diffs back and forth across runs.
• Consider having whatever step performs the pre-agent .pot/.po sync commit its output directly (or document that the agent should), rather than leaving it as ambient uncommitted state at checkout — this has now shown up across at least two independent runs (this one and Checkout starts with uncommitted reformatting diff across messages.pot and all locale .po files #44's).