Machine translation is now part of daily work for many language specialists. Clients send files with a machine draft already in place, or ask for faster delivery at lower cost. Post-editing, the skill of improving that draft, has become a profession in its own right, and doing it well requires method rather than guesswork.
What post-editing means
Post-editing is the process of correcting and improving machine-translated text. The goal is not to rewrite everything but to bring the text to an agreed level of quality as efficiently as possible. A good overview of machine translation post-editing explains the main approaches.
Light and full post-editing
Light post-editing fixes errors that block understanding and leaves style alone. Full post-editing aims for text that reads like a human translation. Always agree with the client which level is required, because the difference in time and cost is large.
Typical machine errors
Common problems include wrong terminology, mistranslated idioms, inconsistent names, missing negations, wrong gender and agreement and invented text. Learning to scan for these patterns quickly is a major part of professional skill and keeps editing time under control.
Choosing the right system
Quality varies by language pair and subject. A configurable machine translation tool that applies your glossary, protects tags and stores approved edits gives the editor a better starting point than a generic public website.
Using memory and glossaries
Editors should reuse approved sentences from the translation memory before accepting machine suggestions. Glossaries ensure that specialised terms are correct, which reduces the most expensive kind of errors in technical, legal and medical text.
Quality standards
The ISO 18587 standard describes requirements for full post-editing of machine translation output, including competences of post-editors. It is a helpful reference when discussing quality expectations with clients and agencies.
Measuring effort and fair pricing
Track how long editing takes compared with translating from scratch, and how many corrections each segment needs. These measurements support fair pricing and show where the machine performs well or poorly for a particular client and subject.
Keeping your own judgement
Reading machine output can make errors look normal. Take breaks, compare against the source carefully and read difficult passages aloud. The editor is responsible for the final text, regardless of who or what produced the first draft.
Briefing the client
Ask clients what the text is for, who will read it and how long it will stay in use. A document for internal information needs less polish than a public brochure. Clear expectations prevent disputes and let you choose the right level of editing from the start.
Handling terminology
Check the glossary before starting and add new terms as you meet them. Terminology errors cluster in specialised texts, and fixing them consistently across a document is easier when the tool highlights every occurrence for review.
Style and tone
Machines often produce correct but flat sentences. In marketing and customer-facing text, tone matters. Decide whether the project needs creative adaptation, and if so, consider a different approach such as full translation instead of editing a poor draft.
When not to post-edit
If the draft is so poor that fixing it takes longer than starting again, say so. Some texts, such as poetry, humour and highly creative copy, are unsuitable for machine drafts. Honest advice protects both your reputation and the client budget.
Confidentiality
Ask where the machine translation was produced and whether the content was shared with third parties. Confidential documents require secure systems and clear agreements. Never paste sensitive material into public tools without the client agreement.
Working with large projects
Split big jobs into batches, review the first batch carefully and give feedback on recurring machine errors. Sharing these findings with the project manager helps improve glossaries and settings for the remaining batches and for future work.
Training and career development
Post-editing skills can be learned. Practise on varied texts, learn the typical error patterns of the engines you use and take courses on quality evaluation. Specialists who document their method and results find it easier to win long-term clients.
Automatic quality checks
Run checks for numbers, tags, punctuation and consistency after editing. They catch small mistakes that remain after careful reading. Combine them with a final human proofreading of the most visible text, such as headings and calls to action.
Feedback loops with engine owners
Where possible, send corrected segments back to the system owner so the engine can improve for your domain. Even simple lists of recurring errors help. Over time, better engines mean less editing effort and a more pleasant working experience.
Planning your workflow
Define a standard sequence: check the brief, run memory matches, edit machine output, run quality checks, proofread and deliver. Writing it down makes the process repeatable, helps colleagues join your projects and supports consistent quality.
Final thoughts
Post-editing rewards specialists who combine linguistic skill with a systematic approach. By agreeing levels of quality, using memories and glossaries and measuring effort, language professionals can deliver speed without sacrificing accuracy.
One skill underpins everything in this guide: judging the draft in front of you before you touch a single segment. When machine translation runs inside your editor, the scores, confidence marks and error highlights it shows are only a starting point, not a verdict. Specialists learn to sample a few segments from each section, compare them against the source and decide whether the engine deserves light editing, full editing or rejection. That habit turns raw engine output into a reliable production decision.