Post-editing machine translation, known as MTPE, has settled into a permanent place in the language industry. For some content it is the most efficient route to a usable translation; for others it is a false economy. The difference lies in knowing how to select, brief and execute. This practical guide sets out how professional teams approach MTPE in practice.

Start from the content, not the tool

The first question is never "which engine?" but "what is this text for?" MTPE is well suited to high-volume material where the value of professional polish is modest and the cost of delay is real: internal documentation, support content, certain catalogue pages and texts needed mainly for comprehension. It is poorly suited, without genuinely full post-editing, to marketing that must persuade, legal documents that must have precise effect and clinical content where errors carry safety risks.

Professionals also consider how repetitive the source is and whether approved terminology exists. Machine output improves dramatically when guided by glossaries and previous high-quality translations.

Light versus full post-editing

MTPE comes in distinct quality levels, and conflating them causes most of the friction in client relationships.

  • Light post-editing produces text that is accurate and readable, delivered quickly. The editor fixes meaning errors, confusing wording and obvious mistakes, but does not restyle every sentence. The output is openly a post-edited product.
  • Full post-editing aims for output equivalent to professional human translation: correct, consistent, stylistically polished and fit for external publication. It takes correspondingly longer.

The level should be agreed before work begins and reflected in pricing and schedules. Asking for light post-editing while expecting human-quality prose is a recipe for disappointment.

What post-editors actually do

Effective post-editing is not simply reading the target text and smoothing it. Editors compare against the source systematically, because fluent machine output can silently omit clauses, mistranslate numbers or choose the wrong sense of a term. Their checklist typically includes:

  • completeness: nothing added, dropped or distorted;
  • terminology: approved terms and client glossaries applied consistently;
  • facts and figures: numbers, dates, units and names verified;
  • grammar and readability: natural sentences in the target language;
  • formatting: markup, tags and layout preserved;
  • consistency across the document and with previous content.

Organising the workflow

Mature teams treat MTPE as a managed process rather than ad hoc use of a website translator. Engines are selected for language pair and domain, customised with client data where permitted, and connected to CAT environments so that translation memory, termbases and quality tools work together. Data security is explicit, especially for confidential material, with policies on whether content may pass to public services.

Post-editors receive real briefings: audience, purpose, quality level and reference material. Their productivity expectations are realistic, because full post-editing is not dramatically faster than translation from scratch for difficult text, while light post-editing of simple material can be substantially quicker.

Common MTPE pitfalls

Most problems with MTPE follow recognisable patterns. One is post-editing the target text without comparing against the source, which allows silent omissions and invented details to survive precisely because they read naturally. Another is inconsistent quality expectations: reviewers applying human-translation standards to work priced as light post-editing, or accepting material marked for full editing that still contains machine-like phrasing. Misuse of terminology is equally common, especially when engines are not guided by client glossaries.

Operational pitfalls matter too. Teams sometimes reuse engines and settings indiscriminately across language pairs and domains, although performance varies considerably between them. Productivity figures are occasionally borrowed from marketing claims rather than measured on real projects, making schedules unrealistic. A short feedback loop between post-editors, reviewers and project managers addresses these issues quickly: recurring engine weaknesses can be logged, glossaries corrected and expectations reset before a small problem becomes a client complaint.

Measuring honestly

The success of MTPE should be judged against its actual goals: turnaround time, fitness for purpose and total cost including any downstream correction. Professional teams track edit distances and quality feedback, both to price honestly and to identify content where machine translation simply does not pay.

Used with discipline, MTPE is one of the most useful developments in years, making large volumes of content accessible that would never previously have been translated at all. Used carelessly, it publishes fluent-looking errors under the banner of efficiency. The dividing line is expertise: in choosing the right content, defining the right level and entrusting the output to qualified linguists who hold the machines to account.