TransPerfect cuts translation costs up to 50% with Amazon Bedrock-powered automatic post-editing

Content localization is a multi-step process — asset handoff, preprocessing, machine translation, post-editing, quality review, and handback — that is largely manual, costly, and slow. Post-editing in particular, where a linguist refines machine-translated output to match meaning and style guides, could add days to the translation timeline. TransPerfect and its client AWS needed to process an ever-growing volume of multilingual content at lower cost and faster turnaround.

TransPerfect integrated Amazon Bedrock into its GlobalLink translation management system so that machine-translated segments are automatically post-edited by an LLM before being handed to a human linguist for lighter review or, in "no human touch" workflows, published directly. The LLM is prompted with style guides, approved translation examples, and examples of errors to avoid, and Amazon Bedrock Guardrails' contextual grounding checks help filter hallucinations to protect translation accuracy.

Over 95% of the edits suggested by Amazon Bedrock LLMs showed markedly improved translation quality. This automatic post-editing step delivered up to 50% overall cost savings on translations for TransPerfect and freed human linguists to focus on higher-value edits instead of routine corrections.

• Automating post-editing with LLMs let linguists focus on higher-value edits rather than routine corrections.
• Contextual grounding checks (guardrails) were critical to prevent hallucinations in a workflow requiring near-perfect translation accuracy.
• Layering translation memory, machine translation, and LLM-based post-editing before human review compounds savings across the pipeline.