A post-editing audit should not begin by comparing a few translated segments against their source text. To audit post editing effectively, an organization must establish whether its full post-editing service is controlled, traceable, and consistently delivered in accordance with agreed requirements. For language service providers seeking ISO 18587 alignment, the audit must examine both the edited output and the system that produced it.

Post editing is frequently treated as a lighter version of translation review. That approach creates compliance risk. ISO 18587 addresses the full post-editing of machine translation output and places defined expectations on competence, client agreement, process control, and final quality. An audit should therefore test whether the organization can demonstrate objective evidence at each stage.

Define the Scope Before Auditing Post Editing

The first audit decision is scope. Confirm whether the organization provides full post-editing as a distinct service, whether it is offered only for selected customers or language pairs, and whether the scope includes subcontracted resources, technology platforms, and project management activities.

Full post-editing must be distinguished from light post-editing. A service intended only to make machine-translated content understandable is not equivalent to a service that requires output comparable to human translation. If the organization markets, contracts for, or delivers full post-editing, its procedures must reflect the quality level agreed with the client and the requirements applicable to ISO 18587.

The auditor should identify the documented process from quotation through delivery. This includes sales and client-facing personnel who define the service, project managers who assign work, post-editors who perform the activity, reviewers where applicable, and quality personnel who monitor performance. A process map is useful, but it is not sufficient on its own. The audit must follow real projects and verify that practice matches the documented process.

Audit Client Requirements and Project Preparation

A reliable audit starts with the customer agreement. The organization should be able to show how it determines whether machine translation and full post-editing are suitable for the requested content, intended use, target audience, and risk profile. Highly regulated, safety-critical, or legally sensitive content may require additional controls or may not be appropriate for a standard post-editing workflow without specific client approval.

Review the quotation, statement of work, purchase order, project instructions, and any quality agreement. These records should establish the required service level, language combination, subject-matter requirements, delivery format, confidentiality provisions, and quality expectations. Where a client provides a glossary, style guide, terminology database, or approved reference material, the project file should show that these resources were made available and used.

The audit should also test whether instructions are sufficiently clear for the assigned post-editor. Ambiguous instructions often produce inconsistent results, particularly when clients expect publishable output but have not defined terminology, tone, formatting, or permitted machine translation engines. A nonconformity may arise not because a post-editor made an isolated error, but because the organization failed to translate client requirements into workable project instructions.

Verify Competence and Resource Assignment

Competence is a central audit area. Organizations should maintain evidence that personnel assigned to full post-editing possess the qualifications, professional experience, linguistic competence, and subject-matter capability required for the project. A generic vendor profile or a declaration of language fluency is not adequate evidence of controlled competence.

Examine resource records alongside actual assignments. The assigned post-editor should be qualified for the source and target language combination, capable of evaluating machine translation output critically, and familiar with the relevant domain. Competence in post-editing includes more than the ability to revise language. It includes the ability to identify omissions, incorrect meaning, terminology errors, formatting defects, inappropriate register, and machine-generated content that appears fluent but is factually wrong.

Training records should demonstrate how the organization prepares personnel for its post-editing process. This may include training on client instructions, technology use, error classification, quality evaluation, data security, and escalation procedures. For external resources, equivalent controls may be established through qualification, documented onboarding, monitored performance, and periodic reassessment.

Examine the Post-Editing Process and Technology Controls

The audit must establish what happens between machine translation generation and final delivery. Review the workflow settings, handoff records, task instructions, and completed files. The organization should be able to identify which engine or approved technology was used, who performed the post-editing, which version was delivered, and how changes were controlled.

Technology controls matter because machine translation engines differ in capability, data handling, terminology support, and suitability for specific content. The audit should verify that the organization has a method for approving or restricting engines based on client requirements, confidentiality commitments, and project risk. If customer data is entered into a third-party platform, the organization should be able to demonstrate that this use is authorized and appropriately controlled.

Auditors should not assume that editing activity in a translation environment automatically proves conformity. Track changes, segment history, and system logs can support traceability, but they need to be interpreted alongside the project specification. A post-editor may make numerous changes while still failing to address a critical terminology or meaning issue. Conversely, low edit distance is not automatically evidence of inadequate work if the machine output was unusually strong and the final output meets all requirements.

Sample Output Using a Risk-Based Method

Output sampling remains necessary, but it should be planned rather than random in the casual sense. Select projects across language pairs, subject areas, client types, post-editors, delivery channels, and machine translation engines. Give greater attention to projects involving high-volume content, new customers, complaints, unusual terminology, tight turnaround times, newly approved resources, or prior corrective actions.

When reviewing samples, evaluate whether the final text meets the agreed quality level and whether the evidence supports the process followed. Look for recurring patterns rather than isolated stylistic preferences. Material findings commonly include meaning errors, omissions, untranslated content, terminology inconsistency, incorrect numbers or units, formatting defects, and failures to apply client instructions.

A complete audit file should normally allow an independent reviewer to trace at least the following evidence:

  • the confirmed client requirements and service level;
  • the qualification and assignment of the post-editor;
  • the technology or machine translation environment used;
  • the post-editing and any review records;
  • the final delivery, quality checks, and client feedback; and
  • any deviations, complaints, or corrective actions connected to the project.

The sampling method should be documented. The objective is not to create unnecessary paperwork, but to show that the audit reached a defensible conclusion based on representative and risk-relevant evidence.

Record Findings and Test Corrective Action

Audit findings should be written against defined criteria, not personal preference. Separate major process failures from isolated errors that were corrected before delivery. For example, assigning an unqualified post-editor to a specialized project, failing to define the agreed post-editing level, or using an unapproved machine translation platform may indicate a systemic control failure. A single typographical defect in an otherwise controlled file may require correction but does not necessarily demonstrate a system-wide nonconformity.

Each nonconformity should identify the requirement, the objective evidence, the extent of the issue, and the risk to service conformity. The organization should then investigate root cause rather than simply retraining the person involved. If repeated terminology errors occur, the cause may be incomplete client instructions, poor terminology management, inadequate resource selection, or an ineffective final quality check.

Corrective action is complete only when effectiveness has been verified. A later audit sample, performance review, or targeted project check should confirm that the revised control operates as intended. This is particularly important where the organization is preparing for certification or responding to a client complaint.

Use Internal Audits to Build Certification Readiness

An internal audit of post editing should be independent enough to challenge established practice. The auditor does not need to be external, but should be competent to assess ISO requirements and should not audit their own work without safeguards. For smaller language service providers, this may require a trained internal auditor from another function or an external assessment resource.

Management should review the resulting evidence as part of broader quality oversight. Trends in complaints, rework, post-editor performance, supplier performance, delivery timeliness, and quality findings can reveal whether the post-editing system is improving or merely reacting to individual failures.

The strongest post-editing audit does not attempt to prove that every machine-translated segment was changed. It demonstrates that the organization has defined the service correctly, assigned competent resources, controlled the technology, verified delivery against client requirements, and acted when evidence identifies risk. That is the foundation for credible ISO 18587 conformity and a post-editing service clients can rely on.