Exploring using MQM Translation Error analysis on AI Translation MR GITTECHA-83
Test
Downloaded the EN file from GITTECHA-83 as well as the JA translated file from the AI Translation MR review. Asked Claude to review the translations following the MQM quality framework and I instruct them to be as precise as precise as possible.
The objective is to reduce the AI Translation MR review while having a system that will give us more confidence in merging the AI MR.
Prompt used: You're a JA translation reviewer that previously worked as a developer. Review these EN-JA file pair and find translation errors. Please follow the Multidimensional Quality framework. Be meticulous.
Translation Errors and Issues
1. Awkward Word Choice ⚠️
Location: Introduction paragraph Japanese: "リストされた言語とパッケージ" Issue: "リストされた" sounds unnatural in Japanese Better: "掲載された言語とパッケージ" or "記載された言語とパッケージ"
2. Inconsistent Verb Forms ⚠️
Location: Multiple JavaScript sections Examples:
- "Jestを使用してテストレポートを生成しています"
- "Karmaを使用してテストレポートを生成しています" Issue: Using progressive form (-ています) inconsistently with other sections that use simple form Fix: "生成します" for consistency
3. Unnatural Expression ⚠️
Location: Jest section English: "To make the job pass when there are no .test.js files" Japanese: "ジョブをパスするには" Issue: "パスする" is unnatural in this context Better: "ジョブを成功させる" or "ジョブを通す"
See comment thread below for assessment and conclusions.
4. Word Order Awkwardness ⚠️
Location: Flutter/Dart section Japanese: "flutter test出力をJUnitレポートXML形式に変換するために、[JUnit Report]パッケージを使用します" Issue: Purpose clause placement creates awkward flow Better: "[JUnit Report]パッケージを使用して、flutter testの出力をJUnitレポートXML形式に変換します"
5. Case Inconsistency ⚠️
Location: Mocha section
English: "NPM package" Japanese: "npmパッケージ" Issue: Should match - typically "npm" (lowercase) is correct, but should be consistent
Not applicable, inconsistency was present on the source, not on the target.
6. Technical Flag Inconsistency ⚠️
Location: Helm section Code example: -t JUnit (uppercase) Japanese text: "-t junitフラグ" (lowercase) Issue: Text should match the code example exactly
Not a translation issue. The translation is as inconsistent as the source. Translations should follow up if the source gets edited.
7. Redundant Phrase ⚠️
Location: GoogleTest section Japanese: "結果はその後、集約されます" Issue: "その後" is unnecessary Better: "結果は集約されます" or "その後に集約されます"
Emi confirmed that this review item can be disregarded.
8. Convoluted Structure ⚠️
Location: CUnit section Japanese: "CUnitCI.hマクロを使用して実行すると、XMLファイルを自動的に生成するように設定できます" Issue: Complex conditional structure is confusing Better: "CUnitCI.hマクロを使用して実行すると、XMLファイルを自動的に生成できます"
This seems to be a stylistic preference and shouldn't be a blocker for determining if an AI Translation MR should be merged or not.
9. Terminology Inconsistency ⚠️
Location: Python section vs .NET section Python: "整形します" (format) NET: "書式設定引数" (formatting arguments) Issue: Should use consistent terminology - prefer "フォーマット" for technical contexts
Emi confirmed that this review item can be disregarded.
10. Missing Article Context ⚠️
Location: Flutter section Japanese: "この.gitlab-ci.ymlファイルは" Issue: Missing "例の" for clarity Better: "この例の.gitlab-ci.ymlファイルは"
11. Particle Usage ⚠️
Location: Various sections Example: "異なるアーキテクチャ（x86、x64、またはarm）用に" Issue: Could be more natural with different particle Better: "異なるアーキテクチャ（x86、x64、arm）向けに"
Emi confirmed that this review item can be disregarded.
12. Subtle Meaning Shift ⚠️
Location: .NET section English: "improve the readability of test data in the test widget" Japanese: "テストウィジェットのテストデータの可読性を向上させる" Issue: Accurate but could be more natural Better: "テストウィジェット内のテストデータを読みやすくする"
Nice to have, this shouldn't be a blocker merging AI Translation MRs.
Quality Re-Assessment
Additional Observations
- Colon usage: Consistent use of ":" vs "。" after introductory phrases
- Code comment preservation: All English comments in YAML preserved correctly
- Link formatting: All maintained properly
- Table formatting: N/A (no tables in this document)
- Bullet point consistency: Good formatting throughout
Overall: Still a very good translation that would serve users well, but these refinements would make it exceptional and more naturally flowing in Japanese.