Human post-editing of machine translation

What full post-editing involves

The linguist compares the output with the source, corrects omissions and shifts of meaning, checks terminology and improves the target text for its intended use. A sentence may read smoothly while reversing a condition or attaching a figure to the wrong product. Editing what sounds plausible without checking the original would leave those errors in place.

ISO 18587 concerns requirements for full human post-editing of machine translation output. Referring to the standard does not mean that any quick correction fulfils it, or that every document is suitable for machine translation in the first place. If your procurement process requires a particular certification scope, tell us so it can be checked. Our quality approach also depends on source preparation, terminology and file checks.

Assessing suitability rather than assuming it

Language direction, source clarity, subject matter and raw output quality all affect the decision. A large collection of consistently structured product descriptions may behave differently from a negotiated agreement or a clinical consent form. Post-editing is not automatically faster or cheaper than translation. A representative sample can reveal the density and type of errors before we recommend a workflow.

Technical material requires attention to part names and warnings; see technical translation. A legal document may require so much contextual judgement that legal translation from the source is more appropriate. Medical and pharmaceutical content warrants a separate assessment of its readers and risks. One procedure should not be applied to all these documents merely because raw machine output is available.

Source files, output and glossary

Supply the source and corresponding machine output in clearly identified versions. Segment alignment, when available, can make comparison easier. Include approved terminology, style instructions and previous publications. An unclear source, disordered table or label hidden inside an image can create errors that cannot be fixed by polishing visible sentences alone.

Changes to the source after machine output was generated should be identified explicitly. We need to know whether a segment reflects the current version before editing it. Website material may also require localisation checks for buttons and forms on screen. Designed brochures can need multilingual typesetting after the linguistic work is complete.

Raw output should be supplied without silently accepting suggestions in some passages and leaving others untouched. That distinction helps us assess the error pattern and set an appropriate scope. If the system omitted text in tables or images, include the source assets so those parts can be accounted for separately rather than disappearing from the quoted volume.

Post-editing and bilingual review

Translation review evaluates an existing target against its source within an agreed scope. Post-editing is specifically concerned with raw machine output and an agreed standard for its human improvement. Some checks overlap, but the effort depends on the actual files, not the name used in a request. If a professional has already worked on the target, reviewing that version may be more direct.

The finished text should be readable for its audience and accurately convey the source. If you need substantial brand adaptation or rewriting for another culture, describe that objective before selecting the route. The work may go beyond post-editing and should be planned and priced accordingly.

Quotation and schedule

Send a representative sample, total volume, source and target languages and intended use. We will examine the error pattern before confirming whether post-editing is appropriate. Specify the return format and any internal approval stage as well. The wider translation services provide alternatives, and the quote form is the place to send your material for assessment.