Turn translation disagreements into reproducible acceptance decisions

Agree the review question and coverage

Define the content’s purpose, intended audience, approved source and relevant style instructions before reviewing. Record whether the reviewer examines every segment or a selected sample, and explain how the sample covers different content types and risks. A result from a small collection of headings should not be reported as acceptance of an entire translated help center.

Phrase’s LQA documentation describes configurable error categories, severity and weighting by project needs. It also explicitly distinguishes its earlier DQF-MQM-based categories from current MQM Core. Use the selected profile as a documented local review method. Do not describe a tool’s default categories or score as an independent certificate of translation quality, and do not adopt example weights without considering the actual content.

Write findings another reviewer can assess

For each finding, preserve the source and target segment references, surrounding context, proposed correction and reason. Explain the consequence: a changed instruction, lost condition, wrong object, misleading tone or a stylistic preference with no demonstrated loss of meaning. A comment saying “bad translation” cannot be resolved consistently because it does not identify what needs to change.

Separate an approved-source defect from a translation defect. If the original instruction is ambiguous, the translator may have made a reasonable choice rather than an error. Route the ambiguity to the source owner and identify the other locales it may affect. Where terminology is contested, reference the relevant concept decision rather than rewarding whichever reviewer prefers a more formal-sounding synonym.

Resolve contested findings through qualified arbitration

Let the provider respond with linguistic reasoning and context. Have an appropriately qualified reviewer resolve genuine disagreements, using the agreed audience and purpose rather than personal authority alone. Record whether the finding is upheld, reclassified, withdrawn or requires a source clarification. Keep the original observation and final decision distinguishable so the review history remains understandable.

Define how repeated instances are counted and how a corrected pattern is retested. A repeated mistake may expose a broad systematic problem without representing independent evidence each time. Preserve consequential blockers separately from the aggregate score. A favorable average should not make an incorrect instruction acceptable, while an inflated count of cosmetic preferences should not obscure a small number of meaningful defects.

Accept repaired content with honest limits

Check the agreed corrections in their actual destination and confirm they did not introduce new omissions, broken variables or awkward surrounding grammar. Retest other occurrences affected by the same fix. Mark a finding closed only when the repaired version is identified, not when somebody says the comment was addressed. Preserve unresolved questions and the responsible decision owner.

Communicate the result with its scope: reviewed locales, content version, coverage, open blockers and accepted limitations. Explain changes to the review criteria before using them in a later project comparison. A lower score under a different profile does not establish deterioration. Feed confirmed recurring issues back into source instructions and reusable translations, while retaining legitimate local differences.

Make the reasoning concrete.

A fictional educational publisher receives a translated guide. One reviewer flags an informal phrase as unacceptable, while another finds a missing condition in an important instruction. A numerical summary treats both comments alike and suggests the delivery passes.

The publisher verifies the restored condition, records the style decision and reports the reviewed scope. It does not describe the result as universal language quality certification.

Take the questions into your next brief.

Use these prompts in your own notes. Nothing is submitted from this page.

1. Scope

   Which content version, locales and sample or full population were reviewed?

2. Finding

   What exact meaning or reader consequence supports each alleged error?

3. Arbitration

   Who resolves contested language judgments using which evidence?

4. Blockers

   Which consequential issues remain regardless of aggregate score?

5. Closure

   Where is the repaired version verified and the final decision recorded?

Watch for the wrong conclusion.

• Calling every stylistic preference a mistranslation.
• Blaming the translator for an unresolved source ambiguity.
• Allowing an average score to cancel a consequential open defect.

Sources & scope

Workflows, worksheets and fictional examples are original editorial analysis. Sources support only specifically attributed principles. Platform guidance can change; legal requirements depend on jurisdiction and context. No measured results or named human reviewer are claimed.

• Phrase: Language Quality Assessment →Primary platform documentation · Supports: Phrase documents configurable LQA criteria and explicitly notes the difference between its DQF-MQM-derived typology and current MQM Core; no certification or universal weighting is claimed.