Start with delivery requirements
A game combines system instructions, dialogue, quests and marketing copy. Each asks something different of terminology, context and creativity. Establish the target language, purpose, available references and quality requirements before comparing sample results. Repetition does not automatically make text easy: a short button may lack context, while a single line can define a relationship.
| Approach | When to consider it | Agree in advance |
|---|---|---|
| Human translation | Substantial work on creative intent, voice or context | Staffing, review requirements, and deliverables |
| AIPE | Samples show a useful benefit after editing | Post-editing and QA coverage |
| Hybrid | Modules have different requirements | Module ownership and shared language rules |
Understand the editing scope behind AIPE and MTPE
Here, AIPE means human post-editing of AI translation. PerfectLoc assigns translators who are native speakers of the target language to edit every AIPE translation and review its final quality. MTPE commonly refers to human editing of machine translation; the workflows can overlap, with differences depending on the tools and implementation. Ask about editing coverage, source comparison, glossary rules and acceptance criteria instead of judging quality from the acronym.
Let samples inform the workflow
Set content and quality goals
Confirm language pair, use, quality and launch timing.
Select representative material
Include systems, dialogue and context-dependent text with references.
Review quality and total effort
Apply the same quality criteria and record the work spent on editing, queries, and rechecks.
Assign responsibilities for each content type
Choose the approach and human involvement; reassess after changes.
Use representative samples to decide how to divide the work
Include routine text, context-dependent passages and creative language in the sample. Apply the same terminology and quality criteria, and account for translation, editing, queries and rechecking. Where repeated rewriting or missing context removes the expected benefit, improve the references or adjust human involvement. Different modules can use different workflows; reassess when content or language pairs change.