What MTPE means
The machine proposes
a linguist decides
Machine translation post-editing, or MTPE, is the human correction of a machine-generated translation against the source text and an agreed quality specification.
Fluent output can conceal a missing negative, an invented connection or a changed instruction. Post-editing therefore requires comparison with the source, not a quick polish of the target text. The post-editor must decide whether the meaning is complete, the terminology is appropriate and the sentence serves its intended purpose.
The initial output determines how much work follows. A usable draft may retain substantial wording; a poor segment may need complete rewriting. Keeping machine-generated wording is never a useful objective in itself. The relevant question is whether post-editing is a sensible route to the required finished text.
ISO 18587:2017 addresses full human post-editing of machine translation output and post-editor competences. Babylon Translations holds this certification. It is distinct from the agency’s ISO 17100 translation and revision process; procurement specifications should identify which route applies.
Quality specification
Light and full
mean different targets
| Approach | Intended result | Important boundary |
|---|---|---|
| Light post-editing | A comprehensible, accurate text for a defined limited purpose; stylistic imperfections may remain. | The brief must define acceptable limitations. Light post-editing is outside the full post-editing scope of ISO 18587. |
| Full post-editing | A complete, accurate and linguistically appropriate text comparable in quality to human translation. | Meaning, grammar, terminology, style and relevant conventions require attention; effort depends on the draft. |
| Human translation | A translation created by a linguist from the source rather than by correcting a machine draft. | Often preferable where machine output offers little useful material or the content needs extensive editorial judgement. |
The comparison defines approaches rather than interchangeable service grades. The intended use and required quality must be agreed before selecting a method.
Suitability assessment
Assess the content
before choosing MTPE
Clear, repetitive business or technical content can be a reasonable candidate for MTPE when the language pair and machine output support the required result. Stable terminology, complete sentences and an unambiguous source make errors easier to identify. A representative sample is more informative than the total word count alone.
Creative advertising, slogans, humour and highly compressed interface text often give the machine too little reliable context. Ambiguous source material can produce confident but incorrect interpretations. In such cases, human translation may provide a better starting point than repairing a draft whose fluency hides its weaknesses.
High-consequence content needs particular care. For medical translation or legal translation, a readable draft is insufficient evidence of suitability. Intended use, terminology, specialist review needs and acceptance requirements determine the appropriate process.
A useful MTPE brief describes the final use
The brief should state the audience, publication status, subject, language combination and quality target. Internal reference material and externally published instructions do not create the same reading conditions. A representative source sample, any existing machine output and approved terminology help establish whether post-editing offers a practical advantage.
Quality expectations also need explicit language. ‘Understandable’ leaves room for disagreement about terminology, omissions and style. An agreed specification should make clear whether the text must be ready for publication, which reference materials govern terminology and how questions about the source will be resolved.
Confidentiality is part of method selection
Source texts and supporting information are subject to confidentiality under the agency’s general terms, including the same obligation for subcontractors. For an MTPE project, restrictions on machine processing, permitted handling and sensitive content should be identified before a method is agreed.
The presence of a human post-editor does not by itself answer questions about where an earlier machine translation was processed. Any supplied output should therefore be accompanied by the relevant handling requirements. A confidentiality requirement may rule out a proposed workflow regardless of how well the draft reads.
Human assessment remains the acceptance boundary
Dedicated QA software can support checks for consistency and detectable mismatches. It cannot establish that every relationship, implication or instruction is correct. A post-editor must read beyond individual terms, compare the source and target, and correct output that is plausible in language but wrong in substance.
MTPE should be selected because it can achieve the agreed result efficiently for the actual content. It is not a universal saving, a fixed speed multiplier or permission to reduce human attention. If the draft requires extensive reconstruction, translation from the source may be the more rational route.
MTPE questions
Choosing the
right translation route
Is Babylon Translations ISO 18587 certified?
Babylon Translations is certified to ISO 18587:2017 for post-editing machine translation output. The standard concerns full human post-editing and post-editor competences. The project specification should still identify the content, intended use and required result; certification does not make every machine draft suitable or remove the need for human comparison with the source.
What is the difference between light and full post-editing?
Light post-editing aims at an accurate, comprehensible text for an agreed limited purpose, while full post-editing targets quality comparable to human translation. Light work may leave stylistic imperfections that full post-editing should address. ISO 18587 covers full human post-editing, so a light-editing specification must not be presented as equivalent to that certified process.
When is machine translation post-editing suitable?
MTPE can suit clear, repetitive content when the machine output provides a useful starting point for the required quality. Suitability depends on the language combination, subject, source quality and intended use. A representative assessment should include difficult passages as well as routine ones; an easy sample can give a misleading impression of the work.
When is human translation preferable to MTPE?
Human translation is often preferable when machine output needs extensive rewriting or the content relies on creative intent, ambiguity or missing context. Slogans and fragmentary interface strings are common examples. Sensitive or high-consequence material also requires careful method selection, with specialist requirements and permitted data handling established before any machine processing is considered.
Does MTPE cost less than human translation?
MTPE can reduce effort when the draft is useful, but a lower cost cannot be assumed before assessment. Poor output, difficult terminology and demanding publication requirements can remove the advantage. A meaningful estimate considers the source, existing output, language pair and required quality rather than treating all machine-generated words as equally easy to edit.
Is machine-translated text reviewed by a person?
Human assessment is essential to post-editing: the linguist compares machine output with the source and corrects it against the agreed requirements. Reading only the target text is insufficient because fluent sentences may omit or change meaning. Automated checks support this work but do not replace the post-editor’s decisions about accuracy, completeness and suitability.