In this guide
  1. Accuracy
  2. Consistency
  3. Fluency and Register
  4. Completeness

Receive

The Problem With Evaluating Translation Quality Translation quality is uniquely difficult to evaluate if you do not speak the target language. And the people who commission translation — procurement managers, project coordinators, legal teams, regulatory affairs specialists — frequently do not speak the target language. They receive a translated document, look at it, and have no reliable way of knowing whether it is excellent, adequate, or dangerously wrong.

This creates a structural vulnerability. A translator who produces poor work is rarely caught by the person who paid for it. Errors persist in published documentation, legal agreements, and regulatory submissions because no one in the commissioning organization was equipped to find them.

This guide gives you practical tools for evaluating translation quality even when you cannot read the target language — and explains what to look for when you can. Why "It Looks Fine" Is Not a Quality Assessment The most common quality evaluation method used by organizations that receive translation is visual inspection of the output: does it look professional, is it the right length, are there obvious formatting problems? This tells you almost nothing about quality.

A translation can look completely professional — correct fonts, correct paragraph structure, no obviously missing sections — and be substantially wrong. The terminology may be inconsistent throughout. Key technical terms may be translated incorrectly. A contractual clause may have been mistranslated in a way that changes its legal effect. A safety warning may have been softened or omitted. None of these failures are visible to someone who cannot read the target language.

Conversely, a reviewer who does speak the target language but is not a translation specialist will often flag stylistic preferences — "I would have phrased it differently" — as translation errors.

Not every difference from how a native speaker would phrase something represents a quality failure. Translation allows for multiple valid formulations of the same meaning, and a bilingual reviewer's preferred phrasing is not the benchmark.

Effective quality evaluation requires a framework — a set of criteria applied consistently — rather than an impression.

The Four Dimensions of Translation Quality

Professional translation quality is assessed across four dimensions. Understanding what each covers helps you evaluate output systematically rather than impressionistically.

1

Accuracy

Does the translation convey the meaning of the source text correctly?

Accuracy failures include: Mistranslation — the target text says something different from the source Omission — content present in the source is missing from the translation Addition — content not present in the source appears in the translation Distortion — the meaning of the source is technically present but has been changed in emphasis, implication, or nuance

Accuracy is the most fundamental quality dimension. A translation that is fluent, consistent, and appropriately registered but contains a mistranslation of a safety-critical instruction or a legal term has failed at the most basic level.

2

Consistency

Are the same concepts, terms, and structures translated the same way throughout the document — and across all documents in an ongoing relationship?
Consistency failures include: Terminology inconsistency — the same term translated differently in different parts of the document

Structural inconsistency — similar structures (numbered lists, warning boxes, headings) handled differently in different places

Cross-document inconsistency — terminology that was approved in a previous project appearing differently in a new translation

Consistency is often where long-term translation quality degrades. A provider without translation memory and terminology management may produce a translation that is accurate on first reading but inconsistent in ways that accumulate and compound across multiple projects.

3

Fluency and Register

Does the translation read naturally in the target language, and does it use the register appropriate to the document type and its intended audience?

Fluency failures include: Unnatural syntax — sentence structures that follow the source language rather than the natural patterns of the target language

Unidiomatic phrasing — technically correct words assembled in ways that a native speaker would not use

Literal translation artifacts — phrases that are translated word-for-word and produce correct but awkward target-language output

Register failures include: Wrong formality level — a legal document written in conversational register, or a customer- facing instruction written in formal bureaucratic language

Inconsistent register — formal in some sections, informal in others, without reason Wrong professional register — a pharmaceutical document using consumer health language rather than clinical register, or vice versa

Fluency and register failures are the dimension most likely to be noticed by bilingual reviewers and the dimension most frequently confused with preference rather than error. Not every awkward phrasing is wrong — but systematic awkwardness indicates either a translator working below their competence level or translation produced by AI without adequate human revision.

4

Completeness

Has all source content been translated, with nothing omitted and nothing added?

Completeness failures include: Omitted sentences or paragraphs — source content that is simply missing from the translation Omitted formatting elements — tables, lists, or structured content that has been collapsed or removed

Untranslated segments — source-language text appearing in the middle of target-language translation

Additions — content not present in the source that appears in the translation (typically an expansion of a concept that the translator felt needed clarification)

Completeness checking is one of the few quality dimensions you can assess without reading the target language. Compare the structural elements — number of paragraphs, number of list items, number of table rows, presence of headings, numbered sections, and figure captions — between source and translation. Significant structural discrepancies warrant investigation. What You Can Check Without Speaking the Target Language Even if you cannot read the target language, there are meaningful quality checks you can perform on any translation.

Structural Completeness Check

Compare the source and translation side by side: Do they have the same number of paragraphs?

Do numbered lists have the same number of items? Do tables have the same number of rows and columns? Are all headings and subheadings present? Are all footnotes, captions, and labels present? Are there any untranslated passages — source language text appearing in the translated document?

This check takes minutes and reliably identifies the most serious completeness failures. Terminology Spot-Check Identify five to ten key terms in your source document — product names, technical terms, regulatory terms, or proper nouns. Find where each appears in the translation. Ask these questions:

Is the term translated consistently throughout, or does it vary? Does the translation of proper nouns and product names follow any instruction you provided (translate, do not translate, use the approved equivalent)?

Do regulated terms — certification numbers, standards references, regulatory codes — appear identically to the source?

If you have a previously approved translation of the same term from an earlier project, does the new translation match it?

This check does not require reading the translation — it requires identifying specific strings in a document, which is possible with basic search functions. Format and Layout Check Does the translation fit its layout without overflow or truncation? Are text boxes, table cells, and heading fields properly filled? Is text direction correct for right-to-left languages (Arabic, Hebrew)? Are numbers, dates, and measurement units formatted according to target-language conventions (1.234,56 vs 1,234.56; DD/MM/YYYY vs MM/DD/YYYY)?

Are currency symbols and units of measurement appropriate for the target market? Delivery Specification Check Is the file in the format you requested? Are all files present? Is the file named correctly? Are any special instructions — specific fonts, track changes removed, comments cleared — followed?

Certification Check (for certified translations) Is the certification statement present? Is it signed? Is it stamped where a stamp was required? Does it identify the translator or agency and the date? What Bilingual Reviewers Should and Should Not Do When you have access to a bilingual reviewer — an in-country employee, a local partner, or a subject-matter expert who speaks the target language — use them for what they are good at and protect against what they are not.

What bilingual reviewers are good at: Identifying obvious mistranslations in their domain of expertise Flagging terminology that does not align with local professional usage Identifying register that is inappropriate for the local market or audience Catching cultural references that do not transfer to the target context Confirming that product names and brand terms are handled correctly

What bilingual reviewers are not good at — and should not be tasked with: Comprehensive accuracy checking (they are not professional revisers and will not catch everything)

Distinguishing translation errors from stylistic preferences (they will flag both as "wrong") Evaluating completeness systematically (they will read for content, not check structure) Assessing terminology consistency across long documents (humans are not reliable for this at scale)

The most important briefing instruction for a bilingual reviewer: Ask them to distinguish between errors (the translation says something different from the source, something is missing, a term is wrong) and preferences (they would have phrased it differently, but the translation is not incorrect). Preferences are useful feedback that can be incorporated into the glossary and style guide for future projects. They are not translation failures. The Error Severity Framework Not all translation errors are equal. A professional evaluation framework distinguishes between error severities to prioritize correction and assess overall quality.

Critical errors — errors that change meaning in ways with legal, safety, or regulatory consequences:

A mistranslated dosage in a pharmaceutical document

A missing safety warning in a machine manual A contractual obligation translated as a permission A regulatory requirement omitted from a compliance document Critical errors require correction before publication or submission, regardless of cost or timeline. Major errors — errors that significantly affect comprehension or usability, but without immediate safety or legal consequences:

A technical term consistently mistranslated throughout a user manual A procedural instruction that is confusing or misleading due to translation error An omission that leaves a section incomplete Register that is so wrong it impedes communication Major errors require correction before publication.

Minor errors — errors that affect quality but not comprehension: Occasional inconsistency in terminology that does not cause confusion Mildly unnatural phrasing that reads correctly but awkwardly Punctuation variations that follow the translator's convention rather than a specified style Formatting minor variations Minor errors should be noted and fed back into glossary and style guide refinement, but they do not necessarily require correction in the current delivery.

Preferences — not errors: A different but equally valid word choice A sentence structure that differs from the reviewer's preference but is not incorrect Regional vocabulary variation that is valid but not the reviewer's preferred variant Preferences should not be treated as errors in quality evaluation. Conflating preferences with errors inflates error counts, creates disputes with providers over non-errors, and — most expensively — trains reviewers to over-report, making their feedback progressively less useful. How to Give Useful Feedback to Your Translation Provider Quality improves over time in a provider relationship only if feedback is communicated in a way the provider can act on. Most feedback provided by organizations to translation providers fails this test.

Feedback that providers can act on: "The term [X] was translated as [Y] throughout. The approved translation is [Z]. Please update the glossary."

"Section 3.2, paragraph 2: the instruction 'do not exceed [value]' was translated as 'should not exceed [value]'. The original is a prohibition, not a recommendation. Please correct."

"The register throughout this document is more formal than required for a consumer-facing product. Our preference for this content type is [description]. Please update the style guide."

Feedback that providers cannot act on: "The translation doesn't feel right" "This doesn't sound natural" "Our local team doesn't like it" "Please improve the quality" Actionable feedback identifies the specific location, the specific error, and the specific correction. Vague feedback generates a response that is equally vague — and does not improve the next project.

Building Quality Into the Relationship, Not Just the Review The most efficient quality management approach is preventing errors rather than finding them. A review process that catches errors after delivery is valuable — but a provider relationship structured to minimize errors before delivery is more valuable.

What prevents errors before delivery: A maintained terminology glossary. Every approved term is documented, applied automatically, and updated when decisions change. Terminology errors in delivery are a symptom of inadequate glossary management.

A style guide for your content. Register, tone, formatting preferences, and common stylistic decisions documented once and applied across all projects. Style errors in delivery are a symptom of inadequate briefing.

Translation memory that is actually maintained. Approved translations from previous projects applied to current ones. Cross-project inconsistency is a symptom of TM that is not being maintained or applied.

A briefing process that communicates context. What is this document? Who will read it? What is the most important quality requirement for this content type? A translator who receives context produces better output than one who receives only a file.

Revision as a process requirement, not a premium. Every translation reviewed by a second qualified linguist before delivery. Errors that revision catches are not provider failures — they are the process working as intended. Errors that pass revision and reach you are process failures that should be investigated.

The goal of quality evaluation is not just to identify errors in what you have received. It is to feed information back into the relationship in ways that reduce errors in what you receive next.

Summary: The Quality Evaluation Checklist Without reading the target language: Structural completeness check — same paragraphs, list items, headings, table rows as source Terminology spot-check — key terms consistent throughout, proper nouns handled correctly Format and layout check — no overflow, correct number formats, correct text direction Delivery specification check — correct file format, all files present, special instructions followed Certification check if applicable — statement present, signed, stamped, dated With a bilingual reviewer: Brief them on error vs. preference distinction before they begin Ask for specific location references for all flagged items Ask them to classify each item as critical, major, minor, or preference Ask for the correct version of any flagged errors, not just the flag After reviewing: Distinguish critical and major errors from minor errors and preferences in your feedback Communicate errors to your provider with location, error description, and correction Update your glossary with any new approved terms arising from the review Update your style guide with any register or tone decisions made during review Feed the review findings back into the briefing for the next project Working With Business Team Translations Business Team Translations applies mandatory independent revision — a second qualified linguist checking every translation before delivery — to every project under our ISO 17100:2015 certified process. We maintain client-specific translation memories and terminology databases, and we treat feedback as information that improves the next project, not as a dispute to be managed.

If you have received translation from any provider that you are uncertain about, contact us. We are happy to discuss what a quality review of that translation should cover.

Request a Quote