한국어 Contact us

Home Services Localization QA

Korean localization QA: UI, documents, AI output

We check Korean (ko-KR) UI text, documents and model output against the source and Korean writing conventions. Each issue we log has a category, a reason and a suggested fix.

Contact us See examples

What you receive

  • 01

    Issue log

    One row per issue: where it is, the source and target text, category, severity, a suggested fix and a short reason.

  • 02

    Corrected text

    Fixes made in the file you sent, or left as suggestions if your team prefers to apply them.

  • 03

    Quality summary

    Recurring patterns across the batch, such as speech level drift or date handling, grouped by type and severity.

  • 04

    Term and style decisions

    Terms, speech level and format choices settled during review, written down so later batches follow the same standard.

How a project runs

  1. 1

    Sample and materials

    You send a sample, the source and target locales, and any glossary or style guide you already use.

  2. 2

    Agree on the standard

    We agree on scope, error categories, severity levels and who settles unclear cases.

  3. 3

    Pilot review

    We review a small set first and go through the findings with you, so both sides judge the same way.

  4. 4

    Full review

    Every item is checked against the agreed standard. Unclear source text and conflicting rules go to your team as questions.

  5. 5

    Final check and delivery

    Before delivery, we re-check the batch for consistency and run format checks scripted for the project, then send the files and summary.

What we check

Meaning and terminology

  • Mistranslations, omissions and additions
  • Numbers, names and links carried over intact
  • Glossary terms used the same way throughout
  • Button and menu names match the actual screen
  • Unclear source raised as a question, not guessed

Honorifics and tone

  • One speech level throughout: formal, polite or casual
  • Honorific verb forms when addressing the customer
  • Address terms like 고객님 (customer), not 당신 (you)
  • Tone that suits errors, payments and apologies
  • Model replies keep their speech level across turns

Translationese

  • Literal pronouns such as 그것 (it) and 그들 (they)
  • English word order carried into Korean
  • Doubled passives such as 되어지다
  • Calques like 회의를 가지다 (to have a meeting)
  • Chains of nouns joined by repeated 의 (of)

Spacing, particles and formats

  • Word spacing by standard Korean rules
  • Particles that fit the word before them
  • Particles after placeholders, where the form depends on the value
  • Dates read in the source locale, written ko-KR style
  • Times with 오전/오후 (AM/PM) before the hour
  • Numbers, currency and units in Korean conventions

Worked examples

Synthetic data, made for this page.

Example

Calendar reminder with a misread date

Synthetic example: a calendar app's reminder, translated from en-US to ko-KR.

  1. Source (en-US)

    Reminder: Team meeting on Thu, 11/05/2026 at 3:00 PM

  2. Machine translation

    알림: 2026. 5. 11.(목) 3:00 PM 팀 회의

    Reminder: Team meeting on Thu, May 11, 2026 at 3:00 PM

  3. Issue: meaning

    The date was read as day/month. In an en-US source, 11/05/2026 is November 5, and May 11, 2026 is a Monday, not a Thursday.

  4. Issue: format

    3:00 PM keeps the English order. In Korean, the PM marker comes before the hour.

  5. After QA

    알림: 2026. 11. 5.(목) 오후 3시 팀 회의

    Reminder: Team meeting on Thu, Nov 5, 2026 at 3 PM

VerdictFix before release: wrong date

The source locale is en-US, so 11/05/2026 can only mean November 5. A reminder with the wrong date sends people to a meeting on the wrong day, so this is fixed before any style issue.

Example

Delivery notice that reads like a translation

Synthetic example: an online shop's order notification, translated from en-US to ko-KR.

  1. Source (en-US)

    Your order has shipped. Track your package in the app.

  2. Machine translation

    당신의 주문이 배송되었습니다. 앱에서 당신의 패키지를 추적하세요.

    Your order has been delivered. Track your package in the app.

  3. Issue: meaning

    The verb reads as 'has been delivered', but the source means the order has left, not arrived.

  4. Issue: translationese

    'Your' is translated literally, twice. Korean notices leave the pronoun out and use honorific verbs instead.

  5. Issue: terminology

    'Track your package' is rendered word for word. Korean shopping apps use a set term for delivery tracking.

  6. After QA

    주문하신 상품이 발송되었습니다. 앱에서 배송 조회를 하실 수 있습니다.

    The item you ordered has been sent out. You can track the delivery in the app.

VerdictRevise: misleading and literal

Customers may think the parcel has already arrived and contact support. The literal pronouns and wording also make the shop's notice read like a machine translation.

Data and formats

  • App and web UI strings
  • Notifications and emails
  • Help articles and documents
  • Chatbot and LLM responses
  • Search queries with result lists
  • XLSX, CSV and shared spreadsheets
  • JSON string files
  • Markdown and plain text

Questions

Which languages do you work with?

We review Korean (ko-KR). Source text is usually English. We work in professional working English, so questions, issue logs and reports can be written in English or Korean.

How is translation review different from proofreading?

Proofreading looks at the Korean text alone. A translation review compares it with the source, so it also catches changed meaning, omissions and locale errors, such as a misread date. Each issue gets a type and a severity.

Do we need a glossary or style guide first?

It helps, but it is not required. Without one, we propose terms and a speech level based on your existing Korean text, confirm them with you, and record them so every batch follows them.

Do you also check Korean output from AI models?

Yes. We review chatbot and LLM responses with the same categories, and also check that the speech level holds across turns and that the answer fits a Korean user's situation.

How do you judge search result relevance?

For each query and result, we judge how well the result matches what a Korean user was looking for, following your guidelines. Where the call is close, we add a short reason.

Other services

LLM and agent evaluation RLHF and post-training data Red teaming and safety evaluation In-house QA tools

Contact

Tell us about your project

Write in English or Korean, and include these four points if you can.

  1. 1Model, agent or product
  2. 2Type of work
  3. 3Rough volume and timeline
  4. 4Existing guidelines, if any
contact@marudata.io

The address opens a draft in your mail app. We then agree scope and schedule with you.