Assess quality and cost before deciding how to use AI.
More content and shorter update cycles do not mean every piece should be handled by AI. LIG first evaluates representative samples for language quality, human editing effort and data-handling requirements. We then propose comparable options so you can decide whether AI is worthwhile and where it belongs in the process.
AI translation, machine-translation post-editing and AI-assisted post-editing serve different purposes. They can be assessed separately or combined in a workflow for ongoing updates.
AI Translation Services
Generate a first draft with machine translation or a large language model approved for the project, using terminology, context and reference materials to improve consistency. Suitable for assessed batches of content, knowledge bases, help centres and update materials.
The service can include sample comparisons, an assessment of suitable models or engines, preparation of terminology and prompt rules, translation output and agreed review. This helps identify which content can be processed in batches, which needs human retranslation and which checks are needed before public release.
Machine-Translation Post-Editing (MTPE)
Professional editors revise machine-translated text, checking meaning, omissions, terminology, numbers and expression. The depth of review and acceptance criteria depend on whether the content is for internal understanding, client communication or formal publication.
A project may include sample assessment, full post-editing, independent review and formatting checks. When source ambiguity, specialist risk or poor machine output warrants it, we switch to a more suitable human-led approach.
AI-Assisted Post-Editing
AI-assisted post-editing uses AI to check an existing translation and suggest changes to terminology, style, grammar and consistency, helping editors locate issues faster. People designated in the project agreement verify important meanings, product facts and key figures.
In human MTPE, a professional directly edits a machine translation. In AI-assisted post-editing, a tool first suggests changes, followed by human confirmation as agreed. The project plan specifies the scope of automated processing, the depth of human review and the final deliverable status separately.
Possible deliverables include suggested changes, an edited version, issue records and agreed results of human review.
Enterprise AI Translation Workflows
When content changes continuously, organisations need to manage source materials, terminology, versions and approvals. LIG can assess tool connections, file handling and task allocation for each project so automation and human work support one another.
Areas for discussion include content classification, terminology and reference setup, model choice, translation and editing steps, escalation of exceptions, sample quality checks, version records and client approval. For CMS, translation-management system or API connections, we first confirm system conditions and implementation scope.
Clients may ask us to assess a specified environment or restrict external AI use. Data storage, model inputs, logs, backups and staff access are each reviewed separately; an ‘enterprise’ tool label does not replace a specific data assessment.
Confirm AI use before processing materials.
Before choosing a tool, we verify whether client materials are used to train models, what inputs and logs the tool retains, and who can access them, then include the agreed conditions in the project plan. If a tool cannot meet the project’s requirements, we use a suitable alternative tool or a human workflow.
Start with a representative sample.
Provide anonymised samples suitable for assessment, along with target languages and intended uses. Obtain a proposed approach before expanding the project. The scope, price and timeline of the assessment are agreed in advance; claims about cost or efficiency are based on actual samples and project records.
One assessment helps answer four questions.
Is the quality suitable? Identify the main error types, terminology consistency and human work needed before public release.
How are costs built up? Separate translation, post-editing, review and formatting, and state the sample and scope used for the estimate.
How will delivery work? Set out the steps, estimated timing, client review points and approach to ongoing updates.
How will materials be handled? Explain the proposed tools, data conditions and restrictions; resolve outstanding questions before processing live materials.