Why Does ISO 18587 Treat the Post-Editor as a Distinct Professional?

posteditors > Q. Isn't a post-editor just a translator who fixes machine output? Why does ISO 18587 spend so much effort defining their qualifications?

A. Because post-editing is not translation with a head start, it is a distinct skill that combines translation competence with the ability to diagnose AI-specific failure patterns. A bilingual person can read MT output and feel that something is off, but ISO 18587 requires a professional who can identify exactly what is wrong, why the engine produced it, and how to fix it without introducing new errors.

ISO 18587 makes this explicit: the standard does not certify the MT engine, the translation memory, or the workflow alone. It certifies that a qualified human is positioned at the center of the production process, applying judgment that no AI model can currently replace.

Q. What is the practical risk of treating post-editing as a low-skill task?

A. If an LSP staffs MTPE projects with under-qualified editors, the most common failure mode is output that looks fluent but contains latent errors no one catches before delivery. As covered in Part 1, modern LLM-based MT fails in fluent, grammatically correct ways, which means surface-level review is no longer sufficient. ISO 18587 exists precisely because this risk became systemic once neural MT made output fluent enough to disguise its own mistakes.

The Seven Competence Areas ISO 18587 Requires

Q. What are the actual competence requirements ISO 18587 lists for a post-editor?

A. ISO 18587 requires a post-editor to demonstrate competence across seven areas. Each one addresses a different failure mode that can occur during MTPE.

Competence Area	Core Requirement	What Happens Without It
Translation competence	Full professional ability to produce a target-language text that conveys source meaning accurately and idiomatically	Edits look plausible but drift from source meaning (Mistranslation, 1.1)
Linguistic & textual competence	Command of grammar, syntax, and discourse-level coherence in both languages	Sentence-level fixes that break paragraph-level flow or register
Research competence	Ability to verify terminology, facts, and domain references rather than accept MT defaults	Approved glossary terms get overwritten by MT's plausible-sounding alternative
Cultural competence	Understanding of source and target culture well enough to localize, not just translate	Literal renderings of idioms, humor, or cultural references that confuse the target reader
Technical competence	Fluency with CAT tools, TM systems, QA checkers, and the specific MT platform in use	Tag corruption, broken placeholders, or lost formatting go unnoticed
Domain / subject-matter competence	Sufficient expertise in the relevant field, legal, medical, gaming, technical	Domain-specific severity thresholds are misjudged, e.g. treating a Critical legal error as Minor
MT literacy	Familiarity with how MT systems fail, including hallucination, omission, and overconfident fluency	Hallucinated content (Addition, 1.3) is accepted because it reads naturally

Key insight: the first five competence areas overlap heavily with what ISO 17100 already requires of a human translator. The last two, domain competence and MT literacy, are where ISO 18587 adds something genuinely new. A fully qualified translator who has never been trained on how MT specifically fails is, by ISO 18587's logic, not yet qualified to post-edit.

Translation, Linguistic, and Research Competence

Q. These three sound like standard translator qualifications. Why does ISO 18587 restate them instead of just pointing to ISO 17100?

A. Because MTPE changes how these competencies get applied, even if the competencies themselves are unchanged. A translator working from a blank page constructs meaning. A post-editor works from an existing draft and must resist the pull of MT phrasing, an effect sometimes called anchoring: once a sentence is in front of you, even a low-quality one, it is psychologically harder to deviate from than it would be to write from scratch.

ISO 18587 restates translation, linguistic, and research competence specifically to make clear that post-editing demands the same rigor as translation, applied under different cognitive conditions.

Cultural and Domain Competence

Q. Why are cultural and domain competence listed separately if they are also part of normal translator training?

A. Because MT engines fail differently depending on domain and culture, and a generalist post-editor without domain grounding cannot reliably catch domain-specific errors. As covered in Part 3, severity itself shifts by domain: a Minor stylistic issue in marketing copy can be a Critical error in legal or medical text. A post-editor without domain competence cannot apply the correct severity judgment, even if their general language skills are excellent.

Cultural competence plays a similar role for content like gaming, marketing, and entertainment, where MT routinely produces text that is grammatically correct but culturally flat, a literal joke translation, a reference that does not transfer, or a slogan that falls flat in the target market.

Technical Competence

Q. Why does ISO 18587 require technical, tool-level competence as a separate category?

A. Because the most damaging MTPE errors are often technical, not linguistic. A post-editor can produce a perfectly fluent, accurate sentence and still break the product by deleting a placeholder variable, corrupting a tag, or losing a formatting code. In the gaming example reviewed earlier, where "The hero has slain the dragon!" dropped the {hero_name} variable, this single technical slip is rated Critical regardless of how polished the surrounding prose is.

ISO 18587 requires post-editors to be proficient with the specific CAT tools, TM systems, and QA checkers used on the project, not just generally computer-literate.

MT Literacy

Q. Of all seven, which competence is the most unique to post-editing specifically?

A. MT literacy. This is the one competence area with no real equivalent in traditional translator training, because it did not need to exist before neural MT. MT literacy means understanding the characteristic failure patterns of machine translation: hallucination, dropped negation, mistranslation of low-frequency terms, and overconfident fluency that hides semantic drift.

A post-editor with strong MT literacy reads MT output with productive suspicion. They know which sentence types are statistically more likely to contain an error, even when the sentence reads smoothly. This is the direct human countermeasure to the paradox raised in Part 1: the better MT becomes at sounding fluent, the more this specific competence matters.

How Does an LSP Verify These Competencies Exist?

Q. These are good principles, but how does an LSP actually test for them before assigning a post-editor to a project?

A. ISO 18587 does not mandate one specific test format, but it does require documented evidence of competence. In practice, LSPs verify each area differently:

Competence Area	Common Verification Method
Translation & linguistic competence	Translation certification, degree, or documented professional experience, the same evidence ISO 17100 accepts
Research competence	Sample task requiring terminology verification against a provided glossary or reference set
Cultural & domain competence	Domain-specific test content; track record in the relevant vertical
Technical competence	Practical assessment inside the actual CAT tool or platform used on the account
MT literacy	Error-spotting test using real MT output samples with seeded errors across MQM categories

Practical note: the error-spotting test for MT literacy is often the most revealing. A candidate who only flags grammar issues but misses a dropped negation or a hallucinated addition has a translation skill set but not yet an MT literacy skill set, and ISO 18587 treats these as separable.

Q. Can a highly experienced human translator skip the MT literacy assessment, since they are already clearly qualified overall?

A. No, and this is one of the more counterintuitive implications of the standard. ISO 18587 treats MT literacy as its own competence axis precisely because translation skill and MT literacy are not correlated. An excellent translator with no MT exposure can still systematically miss hallucinated content, because their trained instinct is to evaluate fluency and accuracy, not to suspect that an entirely plausible-sounding clause was never in the source at all. Experience post-editing AI-MT specifically is what builds this skill, general translation seniority does not substitute for it.

What Happens When a Specific Competence Is Missing?

Q. Can you map each competence gap to the kind of error it tends to produce?

A. Yes, and this connects directly to the MQM error categories from earlier in this series:

Missing Competence	Typical Resulting MQM Error
Translation competence	Mistranslation
Linguistic & textual competence	Linguistic Conventions, Fluency
Research competence	Glossary Compliance, mistranslation of technical terms
Cultural competence	Cultural Adaptation, Tone / Register
Technical competence	Tags & Placeholders
Domain competence	Severity misjudgment, e.g. under-rating a Critical error as Minor
MT literacy	Addition (hallucination), Omission

This is the strongest practical argument for taking ISO 18587's competence requirements seriously. Every gap in the table above is not a hypothetical risk, it is a specific, recurring error type that this series has already documented as something automated metrics, and even careless human review, can miss.

Closing thought: ISO 18587 ultimately makes one quiet but important claim. The value an LSP adds in the AI-MT era is not the act of producing target-language text, AI can already do that fluently. The value is a qualified human who knows exactly where that fluent text is most likely to be wrong, and has the competence across all seven areas to catch it before the client does.