Full Post-Editing Quality Standards: What Publishable Really Means

When Is a Translation “Done”?¶

You’ve spent four hours on a document. Grammar’s solid, terminology’s consistent, the client’s glossary is enforced. You send it off—and the client bounces it back. “Not publication-ready.” You ask: publication-ready for what? A Reddit thread or a legal contract? A promo video or a manual for a nuclear plant? Those are very different “done” states.

Post-editing quality has no single threshold. It’s a spectrum with names like “light PE,” “full PE,” and “publishable.” And if your client or manager just says “make it publishable,” without clarity on what that means, you’re guessing.

This guide cuts through the confusion. We’ll walk through the frameworks professionals actually use (ISO 18587, ISO 17100, MQM, TAUS), show you how to translate those frameworks into a checklist, and help you spot the difference between “good enough” and “truly done.”

The Post-Editing Spectrum: Light PE vs Full PE vs Publishable¶

Post-editing isn’t one thing. It ranges from quick cosmetic fixes to a full second-language review.

Light Post-Editing (light PE): You’re aiming for readability, not perfection. Focus on grammar, obvious errors, and comprehension. Ignore minor style inconsistencies or phrasing that’s awkward but technically correct. Light PE typically takes 20-30% of the time a full human translation would take. It’s fine for internal documents, draft materials, or situations where speed matters more than polish.

Full Post-Editing (full PE): You’re aiming for publication quality. Every term is checked against a glossary. Grammar and style are polished. Cultural appropriateness is verified. Facts and numbers are validated. You correct not just errors but also clumsy phrasing. Full PE can take 50-70% of translation time—sometimes as much as a fresh translation, which is why some translators argue that heavily flawed MT output should just be re-translated from scratch.

Publishable: This one’s tricky because it depends on context. A translated Instagram caption is “publishable” when it’s clear and on-brand. A certified medical document is “publishable” only when it’s been reviewed by a subject-matter expert and meets regulatory standards. Publishable = meets the end-use requirement and client spec, nothing more.

The industry standard for defining these levels is ISO 18587:2017, which sets out the competencies, processes, and quality expectations for machine translation post-editing. It distinguishes between light and full PE formally, but it doesn’t mandate which one you use—that’s a client decision based on the document’s purpose.

ISO 18587: The Post-Editing Standard¶

If you’re working with a serious client or agency, they likely reference ISO 18587 in their specs. This standard defines what professional post-editing looks like.

What ISO 18587 requires:

• A qualified post-editor (someone with translation training, familiarity with the subject domain, and experience with post-editing specifically).
• Clear definition upfront of whether you’re doing light or full PE.
• A documented process: translation engine selected, MT output generated, PE performed, QA checks done, delivery.
• For full PE, a second review or QA pass is recommended (though not mandatory, unlike ISO 17100).
• Documented feedback to the MT engine or vendor for future improvement.

The standard doesn’t set a single “quality level.” Instead, it says: define your quality target, document your process, and consistently meet that target. If the client says “light PE, internal use only,” and you deliver that, you’ve met ISO 18587. If they later complain “but it wasn’t publication-ready,” the contract backs you—because light PE was agreed upfront.

This is why contracts matter. A professional post-editing engagement spells out: light or full PE? Glossary provided? Who handles terminology questions? What counts as an error? By how much can you exceed the PE time estimate?

ISO 17100: The Broader Translation Quality Standard¶

ISO 17100:2015 covers all professional translation work, not just post-editing. But it’s crucial for understanding what “quality” means in a formal setting.

The core ISO 17100 requirement: Every translation must be revised by someone other than the translator. This isn’t negotiable. A self-check by the translator doesn’t count. You need a second qualified linguist to catch what the first one missed.

What qualifies as qualified? ISO 17100 defines competence as:

• Five years of professional translation experience (or two years if you hold a postgraduate diploma in translation).
• Native or near-native fluency in both source and target languages.
• Subject-matter knowledge for the domain.
• Technical competence (using CAT tools, managing projects, understanding the client’s specs).

The reviser’s job isn’t just to proofread. They check:

• Terminology accuracy and consistency.
• Grammatical correctness.
• Compliance with the client’s style guide and glossary.
• Completeness (nothing lost or added without reason).
• Cultural and legal appropriateness.

Under ISO 17100, if a translation ships with an error that independent revision should have caught, the translator and the agency are liable. That’s why many agencies require revision as a separate step with sign-off, and why individual post-editors working for agencies often aren’t the ones who deliver—they hand off to a reviser.

For solo translators using AI tools or post-editing MT: ISO 17100 still applies if you’re working under a professional contract. You either do independent revision yourself (time-consuming) or partner with another translator to swap revisions. Many freelancers skip this for cost reasons, which means they’re not ISO 17100 compliant—and client contracts often require it.

MQM: How Errors Are Counted and Weighted¶

If ISO 18587 and ISO 17100 define who checks quality and when, MQM (Multidimensional Quality Metrics) defines what counts as an error and how much it matters.

MQM breaks errors into five main categories:

Accuracy — Does the translation convey the source meaning? Errors include: - Mistranslation (wrong meaning conveyed). - Omission (content left out). - Addition (content added without source justification). - Untranslated (foreign word or phrase left in the source language).

Fluency — Does the target text read naturally? Errors include: - Grammar errors. - Unnatural phrasing. - Inconsistent spacing or punctuation.

Terminology — Are terms consistent and correct? Errors include: - Wrong term used for a glossary entry. - Inconsistent terms within the document (using three different words for one concept). - Domain-inappropriate terminology.

Style — Is the tone, register, and voice appropriate? Errors include: - Overly formal when casual is needed (or vice versa). - Brand voice not matched. - Inconsistent formality across the document.

Locale — Are formatting, dates, currencies, and cultural references correct? Errors include: - Date format wrong (DD/MM/YYYY vs MM/DD/YYYY). - Currency symbol or placement incorrect. - Cultural references that don’t land in the target language.

MQM also assigns severity weights:

• Minor (weight 1) — Doesn’t affect readability or meaning. Example: a comma in the wrong place.
• Major (weight 5) — Affects readability or clarity. Example: a mistranslation of a product feature.
• Critical (weight 25) — Makes the text unusable or misleading. Example: a legal clause mistranslated or omitted.

How to use MQM in your workflow: When you’re reviewing a translation, categorize each error. If most errors are minor fluency issues (grammar), that’s different from a handful of critical accuracy errors. A translation with 50 minor errors might still be publishable; one with a single critical omission is not. This helps you prioritize what to fix and how much time to spend.

Many QA tools now score translations using MQM principles. You’ll see output like: “3 critical errors, 12 major, 40 minor” rather than just a single “quality score.” That’s much more actionable.

TAUS QE Scores: Predicting Post-Editing Effort¶

TAUS (Translation Automation User Society) developed a quality estimation tool that predicts how much work a segment needs—before a human even touches it.

The TAUS QE score ranges from 0 to 1:

• 0.84 and above: “Light review needed.” The MT output is solid; a quick pass catches any rough edges.
• 0.60 to 0.84: “Standard post-editing.” Expect to spend 30-60% of translation time on edits.
• Below 0.60: “Re-translate or heavy PE.” Often faster to translate from scratch than to salvage this output.

The score is based on semantic similarity—how close is the target meaning to the source meaning—calculated using AI embeddings. It’s not a human judgment, so it’s consistent and fast. You can run it on 1,000 segments in seconds.

Why this matters for your workflow: If you’re deciding whether to post-edit a batch of MT output, run a TAUS QE check first. If 80% of segments score above 0.84, you’ll finish fast with light PE. If 60% score below 0.60, you’re re-translating most of it anyway—your client should know that PE won’t be cost-effective.

Many agencies now build TAUS QE into their pipelines. They auto-route high-scoring segments to junior post-editors (faster turnaround, lower cost) and low-scoring segments to senior translators (who make the call: salvage or re-translate). This isn’t yet standard practice for solo translators, but some MT platforms (like DeepL or Systran) are starting to offer QE scores alongside translations.

Practical: How to Build a Post-Editing Checklist¶

Frameworks are great, but here’s what a real checklist looks like:

Before you start: - Do you have the client’s glossary? If not, ask. Missing glossaries cause 20-30% of PE edits. - Do you know light or full PE? Ask, in writing. - Do you know the end-use? (internal doc, customer-facing, legal, marketing) That changes priorities.

During PE (in priority order):

1. Critical errors (MQM Critical weight 25): - Any mistranslation of a term in the glossary. - Omitted sentences or paragraphs. - Numbers, dates, or proper names that are wrong. - Stop here if doing light PE. Fix these and you’re done.

2. Major accuracy errors (weight 5): - Sentences that changed meaning (even if technically readable). - Terminology inconsistencies within the document. - Grammatical errors that affect clarity.

3. Fluency and style (weight 1-5): - Awkward phrasing that reads unnatural. - Tone mismatches (too formal for marketing copy, too casual for technical docs). - Formatting issues (spacing, punctuation). - For light PE, skip these unless they block understanding. - For full PE, fix them all.

4. Locale and cultural fit: - Date/time formats wrong for the target country. - Currency symbols or formats wrong. - Idioms or cultural references that don’t translate.

After you finish: - Run a glossary check: search for your glossary terms and make sure each is used consistently. Tools like memoQ or XBench can automate this. - Fact-check numbers, URLs, dates. Especially important for contracts, invoices, and tech docs. - Read aloud one full paragraph to catch rhythm and flow issues.

Terminology Consistency: Your Biggest Lever¶

Here’s a concrete lever you control: terminology management. It’s the single biggest factor in reducing post-editing time.

When a client provides a glossary, enforce it ruthlessly. If the glossary says “cloud-based solution,” not “cloud computing” or “cloud services,” use only that phrase every time. This removes post-editing decisions. A glossary-compliant translation takes 15-25% less time to edit.

How to build a glossary if the client hasn’t provided one:

1. Scan the source document for repeated terms (especially nouns and phrases that carry meaning).
2. Identify the 50-100 most critical terms (product names, legal terms, domain-specific jargon).
3. Propose translations, ideally checked by a subject-matter expert (SME) in the target country.
4. Document each term’s context and usage notes.
5. Get the client’s sign-off.

A shared glossary cuts PE effort and also prevents back-and-forth debates. “Is it X or Y?” is answered by the glossary, not by email ping-pong.

Many translators skip glossary work to save time upfront. Then they spend twice as long post-editing and revising because terminology is all over the place. It’s a false economy.

Real-World Example: When PE Isn’t Worth It¶

Here’s a case where post-editing fails:

You get an MT output of a 50-page technical manual. It’s in a language pair where the MT engine is weak (e.g., Japanese to Polish). You start PE:

• Missing entire paragraphs (additions/omissions).
• Technical terms translated inconsistently (cloud/network/server all use wrong terminology).
• Sentences so mangled that you have to re-read the source to understand what was meant.
• No glossary provided.

Your options: 1. Post-edit it (estimated 60-80 hours at $35/hour = $2,100-2,800). 2. Re-translate from scratch (estimated 40-60 hours at $50/hour = $2,000-3,000).

Option 2 is faster and better. Option 1 is a trap—you spend twice as long because you’re trying to salvage unusable output.

Professional translators make this call based on TAUS QE scores or a quick sample PE (edit the first 2-3 pages). If the effort is unclear, you get paid for the assessment—time spent figuring out if PE is feasible is billable. Don’t do it for free.

FAQ¶

What’s the difference between light post-editing and full post-editing?¶

Light PE targets readability and basic accuracy—corrects grammar and major errors for internal use—taking ~20% of translation time. Full PE ensures publication quality—checking terminology, style, cultural fit, and completeness—taking ~50-70% of translation time. Use light PE for low-stakes content, full PE for client-facing materials.

How do translators know when a translation is publishable?¶

Publishable means it meets the client’s specs and end-use requirements. ISO 17100 requires independent revision by a second linguist. MQM frameworks check for zero critical errors and minimal major errors. Practical checkers use glossary alignment, fact accuracy, and cultural appropriateness. The definition varies—a social media post and a contract have different “publishable” standards.

Why do some translators reject machine translation output as un-post-editable?¶

Some MT engines produce output so flawed (massive omissions, contradictory glossary terms, nonsensical phrasing) that rewriting is faster than editing. TAUS scores below 0.60 typically fall into this category. Professional post-editors make a judgment call: if edits would take longer than a fresh translation, it’s more cost-effective to re-translate.

How do terminology glossaries reduce post-editing effort?¶

Consistent terminology cuts editing time by reinforcing expected word choices. Without a glossary, a post-editor spends time deciding which variant (e.g., “cloud-based” vs “cloud computing” vs “cloud solution”) is correct. With a glossary approved by SMEs, that decision is pre-made. This alone can reduce PE effort by 15-25% on technical content.

What does a TAUS quality score tell you?¶

TAUS (Translation Automation User Society) scores predict how much editing a segment needs. Scores 0.84+ need only light review. Scores 0.60-0.84 need standard post-editing (terminology, fluency fixes). Scores below 0.60 likely need re-translation. The score is based on semantic similarity between source and target, not on human judgment.

Who is responsible for errors found after a translation is published?¶

Under ISO 17100, the translator/agency who signed off is responsible if the error would have been caught by proper revision. If both translator and reviser signed off, the agency (not individual translators) bears responsibility. For client-supplied glossaries, the client shares responsibility for terminology errors. Contracts typically outline post-delivery correction timeframes (24-48 hours for minor fixes).