i18n: translation pipeline (all languages)#18273
Conversation
Co-Authored-By: Gemini <gemini@google.com>
Co-Authored-By: Gemini <gemini@google.com>
Co-Authored-By: Gemini <gemini@google.com>
Co-Authored-By: Gemini <gemini@google.com>
Co-Authored-By: Gemini <gemini@google.com>
Co-Authored-By: Gemini <gemini@google.com>
Co-Authored-By: Gemini <gemini@google.com>
Co-Authored-By: Gemini <gemini@google.com>
Co-Authored-By: Gemini <gemini@google.com>
Co-Authored-By: Gemini <gemini@google.com>
Co-Authored-By: Gemini <gemini@google.com>
Co-Authored-By: Gemini <gemini@google.com>
Co-Authored-By: Gemini <gemini@google.com>
Co-Authored-By: Gemini <gemini@google.com>
Co-Authored-By: Gemini <gemini@google.com>
Co-Authored-By: Gemini <gemini@google.com>
Co-Authored-By: Gemini <gemini@google.com>
Co-Authored-By: Gemini <gemini@google.com>
Co-Authored-By: Gemini <gemini@google.com>
Co-Authored-By: Gemini <gemini@google.com>
Co-Authored-By: Gemini <gemini@google.com>
Co-Authored-By: Gemini <gemini@google.com>
Co-Authored-By: Gemini <gemini@google.com>
Co-Authored-By: Gemini <gemini@google.com>
✅ Deploy Preview for ethereumorg ready!
To edit notification comments on pull requests, go to your Netlify project configuration. |
github-actions
Bot
added
the
translation 🌍
|
🌐 Translation review started. View progress |
![claude[bot]](https://avatars.githubusercontent.com/in/1236702?s=60&v=4){kind=small}
There was a problem hiding this comment.
Translation Quality Review
PR: #18273
Branch HEAD: bca78df5c5
Languages: 24 (ar, bn, cs, de, es, fr, hi, id, it, ja, ko, mr, pl, pt-br, ru, sw, ta, te, tr, uk, ur, vi, zh-tw, zh)
Files reviewed: 143 JSON files (6 per language across 24 langs, minus 1 for ko which lacks app-subcategories.json)
Scope: Full PR diff (no prior LLM review)
Date: 2026-05-25
Fixes: No fixes applied (review-only)
Summary
This PR updates 24 language JSON files (src/intl/{lang}/{app-subcategories,component-whitepaper,page-bug-bounty,page-start,table,template-usecase}.json). Reviewed via 24 parallel per-language sub-agents comparing each translation against the English source.
Overall average score: ~9.3/10. Most languages are ship-ready. 11 critical issues total, concentrated in 6 languages.
Summary by Language
| Language | Files | Quality Score | Issues |
|---|---|---|---|
| ar | 6 | 9.2/10 | 0 critical, 5 warnings |
| bn | 6 | 9.2/10 | 1 critical, 3 warnings |
| cs | 6 | 9.8/10 | 0 critical, 4 warnings |
| de | 6 | 9.0/10 | 0 critical, 4 warnings |
| es | 6 | 9.2/10 | 0 critical, 3 warnings |
| fr | 6 | 9.6/10 | 0 critical, 5 warnings |
| hi | 6 | 9.4/10 | 0 critical, 3 warnings |
| id | 6 | 9.8/10 | 0 critical, 5 warnings |
| it | 6 | 9.4/10 | 0 critical, 7 warnings |
| ja | 6 | 10.0/10 | 0 critical, 0 warnings |
| ko | 5 | 9.4/10 | 0 critical, 1 warning |
| mr | 6 | 8.6/10 | 3 critical, 3 warnings |
| pl | 6 | 9.8/10 | 0 critical, 3 warnings |
| pt-br | 6 | 9.7/10 | 0 critical, 4 warnings |
| ru | 6 | 9.4/10 | 0 critical, 3 warnings |
| sw | 6 | 8.8/10 | 2 critical, 4 warnings |
| ta | 6 | 7.4/10 | 3 critical, 5 warnings |
| te | 6 | 8.6/10 | 1 critical, 5 warnings |
| tr | 6 | 9.8/10 | 0 critical, 2 warnings |
| uk | 6 | 9.0/10 | 1 critical, 4 warnings |
| ur | 6 | 9.6/10 | 0 critical, 4 warnings |
| vi | 6 | 9.4/10 | 0 critical, 3 warnings |
| zh-tw | 6 | 9.2/10 | 0 critical, 3 warnings |
| zh | 6 | 9.4/10 | 0 critical, 2 warnings |
Critical Issues (Must Fix)
| Language | File | Key | Issue |
|---|---|---|---|
| bn | src/intl/bn/page-bug-bounty.json |
page-upgrades-bug-bounty-not-included-li-4 |
Acronym JSON-RPC transliterated to Bengali (জেসন-আরপিসি); must stay Latin |
| mr | src/intl/mr/page-bug-bounty.json |
page-upgrades-bug-bounty-not-included-li-4 |
Acronym JSON-RPC transliterated to Devanagari (जेसॉन-आरपीसी); must stay Latin |
| mr | src/intl/mr/template-usecase.json |
template-usecase-dropdown-desci |
Acronym DeSci transliterated (डीसाय); must stay Latin |
| mr | src/intl/mr/template-usecase.json |
template-usecase-dropdown-refi |
Acronym ReFi transliterated (रेफाय); must stay Latin |
| sw | src/intl/sw/page-bug-bounty.json |
bug-bounty-faq-q1-content-4 |
"client" mistranslated as mteja (customer) instead of computing klijenti |
| sw | src/intl/sw/page-bug-bounty.json |
page-upgrades-bug-bounty-severity-critical-li-5 |
"clients" mistranslated as wateja (customers) instead of klijenti |
| ta | src/intl/ta/page-bug-bounty.json |
page-upgrades-bug-bounty-not-included-li-4 |
Acronym JSON-RPC transliterated to Tamil (ஜேசன்-ஆர்பிசி); must stay Latin |
| ta | src/intl/ta/template-usecase.json |
template-usecase-dropdown-desci |
Acronym DeSci transliterated (டெஸ்சி); must stay Latin |
| ta | src/intl/ta/template-usecase.json |
template-usecase-dropdown-refi |
Acronym ReFi transliterated (ரெஃபை); must stay Latin |
| te | src/intl/te/page-bug-bounty.json |
page-upgrades-bug-bounty-not-included-li-4 |
Acronym JSON-RPC transliterated to Telugu (జేసన్-ఆర్పీసీ); must stay Latin |
| uk | src/intl/uk/page-bug-bounty.json |
page-upgrades-bug-bounty-client-bugs-desc-2 |
Inconsistent brand handling. Client names transliterated to Cyrillic but Geth left Latin in same sentence; align by reverting all client names to Latin |
Pattern: The dominant critical class (8 of 11) is acronym transliteration into non-Latin scripts in Indic and Bengali languages, specifically JSON-RPC in page-bug-bounty.json and DeSci/ReFi in template-usecase.json. This looks like a Crowdin TM artifact that the sanitizer acronym-protection step likely missed for these specific tokens. Worth investigating as a sanitizer fix.
Recurring Warnings (Cross-Language)
- Acronym
AIlocalized toIAin de, fr, it, pt-br (KIin de). Standard local convention in those languages but technically deviates from the "must stay Latin" rule. Worth a project policy decision. - Client brand names transliterated for execution/consensus clients (Besu, Erigon, Nethermind, Reth, Lighthouse, Lodestar, Nimbus, Teku, Prysm) in non-Latin-script languages. Permitted by spec, but several languages mix Latin
Gethwith transliterated siblings inside the same sentence (page-upgrades-bug-bounty-client-bugs-desc-2). Consistent treatment recommended. - Tone/register inconsistency in
page-bug-bounty.jsonfor de (Sie vs du) and es (usted vs tú). Mixed within the same file. - Plural acronym markers dropped (NFTs to NFT, DAOs to DAO, RWAs to RWA) in cs. Idiomatic but worth noting.
Twitterrendered phonetically (推特, ट्विटर, ٹوئٹر, ツイッター). Acceptable per transliteration policy; flagged only because the platform rebranded to X.
Notable per-language observations
- ja: 10.0/10. Zero critical, zero warnings. All Katakana transliterations correct, no semantic inversions, polite register consistent.
- ta: 7.4/10. Lowest score. Three different spellings of "Ethereum" inside
page-bug-bounty.json(எத்திரியம் 8x, எத்தீரியம் 7x, எத்தேரியம் 1x). Plus the JSON-RPC/DeSci/ReFi transliterations. - sw. The two
mteja/wateja("customer") slip-ups are particularly damaging because the rest of the file correctly usesklijentifor the computing sense; an easy mechanical fix. - bn, mr, ta, te. Share an identical pattern of
JSON-RPCbeing Devanagari/Bengali/Tamil/Telugu-fied at the same key (page-upgrades-bug-bounty-not-included-li-4). Suggests the Crowdin TM source for that single string leaked across these locales. - No semantic inversions (PoW vs PoS, validator vs miner, mainnet vs testnet) found in any language. No domain transliterations (
ethereum.orgis intact everywhere). No placeholder corruption.
Auto-fix status
This review ran in GitHub Actions and is not configured to commit fixes back to the PR branch (the workflow does not check out the PR worktree). The above critical issues were not auto-fixed.
To apply fixes, run locally:
/review-translations --pr=18273 --fix
That will reopen the worktree, apply the critical fixes, and stage them for a follow-up commit.
Reviewed by Claude Code. 24 parallel per-language agents, full PR diff scope.
myelinated-wackerow
left a comment
myelinated-wackerow
left a comment
There was a problem hiding this comment.
Translation Quality Review (re-evaluated against ETHGlossary)
PR: #18273
Branch HEAD: bca78df5c5
Languages: 24 (ar, bn, cs, de, es, fr, hi, id, it, ja, ko, mr, pl, pt-br, ru, sw, ta, te, tr, uk, ur, vi, zh-tw, zh)
Files reviewed: 143 JSON files (6 per language across 24 langs, minus 1 for ko)
Scope: Full PR diff, ETHGlossary-driven re-evaluation
Date: 2026-05-25
Fixes: No fixes applied (review-only)
Summary
This review re-evaluates the prior automated review against ETHGlossary as the authoritative source for terminology. The prior review's findings were heuristic and did not query ETHGlossary, leading to several false-positive criticals.
Net result: 7 of the prior review's 11 "criticals" are false positives. Two real semantic concerns remain (Swahili) and four consistency warnings (DeSci/ReFi transliteration in Indic languages).
Overall: approving for merge. No glossary-violation criticals remain. The Swahili semantic issue is documented as a warning for native-speaker review post-merge.
Re-evaluation of prior critical findings
ETHGlossary endpoints used: GET /api/v1/style-guide/{term}, GET /api/v1/translations/{lang}/{term}, POST /api/v1/filter.
| # | Prior critical | Re-evaluation | Verdict |
|---|---|---|---|
| 1 | bn JSON-RPC transliterated to জেসন-আরপিসি |
ETHGlossary bn entry: "term": "জেসন-আরপিসি" (prose context) — exact match |
False positive |
| 2 | mr JSON-RPC transliterated to जेसॉन-आरपीसी |
ETHGlossary mr entry: "term": "जेसॉन-आरपीसी" (prose context) — exact match |
False positive |
| 3 | mr DeSci transliterated to डीसाय |
ETHGlossary has English entry with scriptRule: "translate" but no per-language mr entry; 6/24 languages transliterate, 18/24 keep Latin |
Warning (consistency drift, not glossary violation) |
| 4 | mr ReFi transliterated to रेफाय |
Same as DeSci | Warning |
| 5 | sw "client" → mteja (customer) |
ETHGlossary has no standalone "client" entry; mteja is the everyday word for "customer" in Swahili. Computing-client sense uncertain without native review |
Valid concern — kept as warning per author's call (no confident Swahili replacement available) |
| 6 | sw "clients" → wateja |
Same as #5 | Valid concern (warning) |
| 7 | ta JSON-RPC transliterated to ஜேசன்-ஆர்பிசி |
ETHGlossary ta entry: "term": "ஜேசன்-ஆர்பிசி" — exact match |
False positive |
| 8 | ta DeSci transliterated to டெஸ்சி |
Same as #3 | Warning |
| 9 | ta ReFi transliterated to ரெஃபை |
Same as #3 | Warning |
| 10 | te JSON-RPC transliterated to జేసన్-ఆర్పీసీ |
ETHGlossary te entry: "term": "జేసన్-ఆర్పీసీ" (confidence: medium, note: "Phonetic transliteration used as per rules.") — exact match |
False positive |
| 11 | uk inconsistent client-name handling (Geth Latin while siblings Cyrillic) |
ETHGlossary uk entries: Besu→Бесу, Erigon→Ерігон, Nethermind→Незермайнд, Reth→Рет, Lighthouse→Лайтхаус, Lodestar→Лодстар, Nimbus→Німбус, Teku→Теку, Prysm→Призм — all Cyrillic, but Go Ethereum (Geth)→Go Ethereum (Geth) — explicitly Latin. The file matches glossary 10/10. |
False positive |
Additional finding (missed by prior review)
| Language | File | Key | Issue |
|---|---|---|---|
| sw | src/intl/sw/page-bug-bounty.json |
page-upgrades-bug-bounty-severity-critical-li-2 |
Third "client→mteja" instance — same semantic concern as the two found previously. Kept as warning. |
Bonus: Tamil "Ethereum" three-spelling pattern (called out as observation in prior review)
The prior review noted three spellings of Ethereum in ta/page-bug-bounty.json. After re-checking ETHGlossary, this matches glossary policy:
- Standalone
Ethereum:எத்திரியம்(8 occurrences in file) Ethereum Foundation:எத்தீரியம் அறக்கட்டளை(8 occurrences, all paired with Foundation)Ethereum Mainnet:எத்தேரியம் முதன்மை வலைப்பின்னல்(1 occurrence, paired with Mainnet)
The translation correctly uses each contextual spelling. (Worth a follow-up to verify ETHGlossary's intentional choice of three Tamil spellings; if unintentional on glossary side, that's a glossary fix, not a translation fix.)
Consistency warnings (DeSci / ReFi)
template-usecase-dropdown-desci and template-usecase-dropdown-refi across 24 languages:
- 18 languages keep the parenthetical acronym in Latin:
(DeSci)/(ReFi) - 6 languages transliterate: bn, hi (ReFi only), mr, ta, te, ur
ETHGlossary has English entries (decentralized-science-desci, regenerative-finance-refi) with scriptRule: "translate" but no per-language entries. Both forms are permissible; the consistent majority pattern would be to keep the parenthetical in Latin for brand recognition. Not a blocking issue — worth a future ETHGlossary entry to make policy unambiguous.
Semantic spot-checks (PoS / PoW / mainnet / validator)
Spot-checked PoS/PoW translations across 11 high-coverage languages (de, fr, it, ja, ko, pl, ru, tr, vi, zh, zh-tw) in page-bug-bounty.json. No semantic inversions found. Proof-of-stake and proof-of-work are correctly distinguished in every language checked, including languages that translate the full phrase (e.g., ko 지분 증명 (PoS) / 작업증명 (PoW), ja プルーフ・オブ・ステーク (PoS) / プルーフ・オブ・ワーク (PoW)).
Updated quality scores
Removing false-positive criticals raises most scores. Scores below reflect ETHGlossary-grounded evaluation.
| Language | Files | Quality | Issues (after re-eval) |
|---|---|---|---|
| ar | 6 | 9.4/10 | 0 critical, 5 warnings (unchanged) |
| bn | 6 | 9.6/10 | 0 critical, 4 warnings (JSON-RPC reclassified) |
| cs | 6 | 9.8/10 | 0 critical, 4 warnings (unchanged) |
| de | 6 | 9.0/10 | 0 critical, 4 warnings (unchanged) |
| es | 6 | 9.2/10 | 0 critical, 3 warnings (unchanged) |
| fr | 6 | 9.6/10 | 0 critical, 5 warnings (unchanged) |
| hi | 6 | 9.4/10 | 0 critical, 3 warnings (unchanged) |
| id | 6 | 9.8/10 | 0 critical, 5 warnings (unchanged) |
| it | 6 | 9.4/10 | 0 critical, 7 warnings (unchanged) |
| ja | 6 | 10.0/10 | 0 critical, 0 warnings (unchanged) |
| ko | 5 | 9.4/10 | 0 critical, 1 warning (unchanged) |
| mr | 6 | 9.4/10 | 0 critical, 5 warnings (JSON-RPC reclassified; DeSci/ReFi remain warnings) |
| pl | 6 | 9.8/10 | 0 critical, 3 warnings (unchanged) |
| pt-br | 6 | 9.7/10 | 0 critical, 4 warnings (unchanged) |
| ru | 6 | 9.4/10 | 0 critical, 3 warnings (unchanged) |
| sw | 6 | 8.6/10 | 0 critical, 6 warnings (mteja×3 reclassified to warning pending native review; previously flagged as critical) |
| ta | 6 | 9.0/10 | 0 critical, 7 warnings (JSON-RPC and "3 spellings" both glossary-compliant; DeSci/ReFi remain warnings) |
| te | 6 | 9.0/10 | 0 critical, 5 warnings (JSON-RPC reclassified) |
| tr | 6 | 9.8/10 | 0 critical, 2 warnings (unchanged) |
| uk | 6 | 9.6/10 | 0 critical, 4 warnings (Geth pattern reclassified — glossary-compliant) |
| ur | 6 | 9.6/10 | 0 critical, 4 warnings (unchanged) |
| vi | 6 | 9.4/10 | 0 critical, 3 warnings (unchanged) |
| zh-tw | 6 | 9.2/10 | 0 critical, 3 warnings (unchanged) |
| zh | 6 | 9.4/10 | 0 critical, 2 warnings (unchanged) |
Overall average: ~9.4/10. No glossary-violation criticals across any language.
Recommendations beyond this PR
- ETHGlossary coverage — Add per-language entries for
DeSci,ReFi, andclient(standalone) so future reviews have unambiguous policy. /review-translationsslash-command fix — The doc saysPOST /filteracceptstext:but the API actually expectscontent:. Worth a follow-up doc fix so future agent-run reviews don't silently fail to fetch glossary terms.- Native sw review —
page-bug-bounty.jsonhas three occurrences ofmteja/watejafor computing "client" that warrant a Swahili-fluent reviewer's call before the next translation push.
Re-reviewed by Claude Code with ETHGlossary as authoritative source. The original review's claude[bot] run did not query ETHGlossary; this review fixes that.
2 participants
Automated Translations
This PR contains translations managed by the intl pipeline.
Each run appends a summary below.
Run: 2026-05-25 10:50:36 UTC