Tomedes Cuts German Documentation Costs Without Cutting Quality

Machine translation post-editing is not a compromise between speed and accuracy. For high-volume technical content, it is the workflow that makes accuracy sustainable at scale.

About the Client

A mid-size German manufacturing company with operations across Central Europe had a documentation problem that grew every time it launched a product. Each new machine, each updated component specification, each revised safety procedure generated technical documentation in German that needed to be available in English for its international distribution network — user manuals, maintenance guides, installation instructions, technical data sheets, and safety notices.

The volume was not the problem at launch. It became the problem over time. As the product line expanded and existing documentation was revised, the English translation backlog grew faster than the company's translation budget. The per-word rate that had been manageable for the initial documentation program had become difficult to sustain across an expanding catalog of technical content that required updates on a rolling basis.

The company's options, as its procurement team saw them, were to reduce the volume of content it translated, to find a lower per-word rate, or to find a different approach entirely. The first option meant gaps in English documentation that its distribution partners had started flagging. The second meant accepting lower quality on safety-critical content where lower quality was not an option. They came to Tomedes looking for the third option.

Industry and Services

• Manufacturing / Engineering
• Machine Translation Post-Editing (MTPE)
• German to English

What Is Machine Translation Post-Editing?

Machine translation post-editing is a workflow in which AI-generated translation output is reviewed, corrected, and approved by a professional human translator before delivery. Unlike raw machine translation (where output is used directly without human review), MTPE combines the throughput of AI translation with the judgment and accountability of a qualified linguist. According to ISO 18587:2017, the international standard governing MTPE, post-editing encompasses full post-editing (bringing machine output to the same standard as human translation) and light post-editing (correcting errors while accepting stylistic differences appropriate for the content type).

For technical documentation where consistency and accuracy matter but where creative register is not a factor, full post-editing consistently delivers human-translation-grade output at a cost structure that makes high-volume programs sustainable.

Tomedes holds ISO 18587:2017 certification (the specific standard for MTPE), making it one of a small number of translation providers whose post-editing workflow is independently audited against the international quality benchmark for this service type.

The Challenge

The company's documentation program had three characteristics that made it a strong candidate for MTPE, and one that made it a candidate only if the MTPE workflow was correctly configured for the content type.

1. A catalog where most content was stable but a significant portion changed on every product cycle

Technical documentation for manufactured products is not uniformly dynamic. A safety notice for a component that has not changed in three years does not need to be retranslated from scratch when the manual it appears in is updated. A revised installation procedure for a new component version does need fresh translation, and it may share substantial text with the previous version. Identifying which content was genuinely new, which was modified, and which was unchanged was the first efficiency problem the company's existing translation program had never properly solved. It was sending entire documents for retranslation when only sections had changed, paying full per-word rates for content that had already been translated and was being translated again because no translation memory was being maintained.

2. Terminology consistency across a growing product line

A manufacturing company's technical documentation uses a defined set of product terms (component names, process designations, measurement specifications, safety classifications) that must be rendered consistently across every document in the catalog. A torque specification rendered one way in a maintenance guide and a different way in an installation manual creates a discrepancy that a technician in the field notices. The company's existing translation program had no shared terminology reference. Each document went to translation as an independent job, with no mechanism to enforce consistency across the catalog.

3. Safety content within a predominantly non-safety catalog

Not all of the company's technical documentation carried the same stakes. Data sheets and product descriptions could tolerate light post-editing, efficient output that communicated the information accurately without requiring the same level of scrutiny as a safety procedure. Safety notices, hazard warnings, and emergency shutdown procedures could not. A single post-editing error in a safety notice has a different consequence than a single post-editing error in a product description. The workflow had to differentiate between these content types and apply the appropriate post-editing depth to each, not treat the entire catalog as a uniform translation task.

Why Tomedes?

The company selected Tomedes for two reasons that had nothing to do with price. The first was ISO 18587:2017 certification — evidence that Tomedes' MTPE workflow was not a cost-cutting improvisation but a structured, independently audited process with defined quality standards. The second was Tomedes' ability to configure the MTPE workflow differently for different content types within the same documentation program — applying full post-editing to safety content and light post-editing to non-critical descriptive content, within a single engagement.

The engagement was scoped around the documentation catalog as a whole, not around individual documents. That distinction matters, it is the difference between a translation vendor and a documentation partner.

The Solution

1. A translation memory audit before MTPE began

The first action was not translation. It was a translation memory audit, a review of the company's existing English documentation to extract and structure the translations that had already been completed. From the existing English catalog, Tomedes built a translation memory covering the established English renderings of the company's core technical content: component descriptions, process steps, specification language, and safety notice text.

Before any new document entered the MTPE workflow, it was run against this translation memory. Content that matched existing translations (unchanged safety notices, stable component descriptions, repeated procedural steps) was carried over directly from the translation memory without re-translation. Content that was genuinely new or modified entered the MTPE workflow. The company stopped paying for translations it had already paid for.

Based on Tomedes' project data across German manufacturing documentation programs, translation memory leverage on established technical catalogs typically recovers between 25% and 40% of the word volume that would otherwise enter the translation workflow as new content — a structural cost reduction that has nothing to do with per-word rates and everything to do with not duplicating work already done.

2. Content classification driving post-editing depth

Before MTPE began, every document type in the company's catalog was classified into one of two post-editing tiers. Safety notices, hazard warnings, emergency procedures, and regulatory compliance content were designated for full post-editing — the post-editor reviewed every segment against the source, corrected accuracy errors, resolved ambiguities, and confirmed that the English output met the same standard as a human translation. User-facing descriptive content, data sheets, and product descriptions were designated for light post-editing — the post-editor corrected errors that would affect comprehension or accuracy while accepting stylistic differences that did not compromise the technical meaning.

This classification meant the company was not applying the same scrutiny (and therefore the same cost) to a product description that it applied to a lockout/tagout safety procedure. The workflow matched the investment to the risk.

3. A German-English technical terminology reference governing the machine translation engine

The machine translation engine used in the MTPE workflow was configured with the company's approved German-English terminology pairs before any document was processed. Component names, process terms, measurement designations, and safety classification language were all pre-loaded as approved equivalents — so the MT engine was not generating its own renderings of established terms and requiring the post-editor to correct them. The post-editor's time was focused on genuine translation judgment, not on fixing terminology decisions the engine should never have been making independently.

The terminology reference also served as the governing standard for the post-editors — when the MT output used an approved term, the post-editor accepted it. When the MT output deviated from an approved term, the post-editor corrected it. Terminology consistency across the catalog was enforced at the engine configuration level and confirmed at the post-editing level.

4. A structured review process for safety content that matched regulatory expectations

Safety documentation for manufactured products sold in international markets is subject to regulatory scrutiny — in the EU, safety notice language for machinery must meet the requirements of the Machinery Directive, and the English translations used in markets with English-language regulatory environments must accurately reflect those requirements. The full post-editing workflow for safety content included a review step specifically focused on regulatory language accuracy: confirming that hazard classifications, signal words (DANGER, WARNING, CAUTION), and required disclosure language were correctly rendered in English and consistent with the regulatory conventions of the target market.

This review step was not an addition to the MTPE workflow. It was built into the post-editing brief for safety content — the post-editor knew before they opened the document that regulatory language review was part of their responsibility, not an optional check.

5. A quarterly translation memory refresh tied to product updates

As the company launched new products and revised existing documentation, new approved translations entered the catalog. Each quarter, the translation memory was updated to incorporate the new translations from that period's documentation program — so the leverage available for the next quarter's work reflected the full accumulated translation history, not just the initial catalog.

This quarterly refresh is what prevents translation memory from becoming a static asset that gradually loses relevance as the product line evolves. A translation memory maintained against the current documentation catalog compounds its value over time. One maintained against a catalog from two years ago is a liability — it generates false matches that the post-editor then has to reject, adding time without adding value.

The Result

Twelve months into the Tomedes MTPE engagement, the company's German-to-English documentation program was covering a larger catalog than it had been at the start (more products, more document types, more languages added to the program) at a total cost lower than the previous year's translation spend on a smaller catalog.

The translation memory leverage recovered a substantial share of the word volume that had previously been re-translated from scratch on each documentation cycle. The content classification workflow directed post-editing investment to the content types that required it. The terminology reference eliminated the category of post-editing corrections that had no value, fixing MT output that had deviated from established terms the engine should have rendered correctly from the start.

The distribution partners who had flagged English documentation gaps at the start of the engagement had stopped raising the issue. The English catalog was current, consistent, and complete.

The company's procurement team had come looking for a lower per-word rate. What they found was a workflow that made the per-word rate largely irrelevant, because the work entering the workflow at a per-word rate was a fraction of what it had been before the translation memory and terminology configuration were in place.

Are you a manufacturing, engineering, or technical company managing high-volume documentation translation across multiple languages? Explore Tomedes' machine translation post-editing services or contact Tomedes for a free consultation.

Frequently Asked Questions

Q: What is machine translation post-editing (MTPE)?
A: MTPE is a workflow where AI-generated translation is reviewed and corrected by a professional human translator before delivery. It combines machine translation speed with human accuracy and judgment. ISO 18587:2017 defines two levels: full post-editing, which brings output to human translation standard, and light post-editing, which corrects errors while accepting minor stylistic differences appropriate for the content type.

Q: Is MTPE suitable for technical documentation?
A: Yes. Technical documentation is one of the strongest use cases for MTPE because it is repetitive, terminology-driven, and register-neutral — the qualities that machine translation handles best. Safety content within technical documentation should always receive full post-editing. Descriptive content, data sheets, and product specifications typically qualify for light post-editing without compromising accuracy.

Q: How much does MTPE cost compared to human translation?
A: MTPE costs less than human translation per word because the machine translation engine handles the initial output and the post-editor corrects rather than creates. The cost difference varies by language pair, content type, and post-editing depth. The larger cost reduction in an established technical documentation program typically comes from translation memory leverage, not re-translating content that has already been translated and has not changed.

Q: What is translation memory and how does it reduce cost?
A: Translation memory is a database of approved source-to-target segment pairs built from previously completed translations. When a new document contains text that matches an existing translation memory segment, the approved translation is applied without re-translation. Based on Tomedes' project data, translation memory leverage on established technical catalogs typically recovers 25% to 40% of word volume that would otherwise enter the translation workflow as new content.

Q: What is the difference between light and full post-editing?
A: Full post-editing brings machine translation output to the same standard as a human translation — correcting all accuracy errors, resolving ambiguities, and ensuring the output reads naturally in the target language. Light post-editing corrects errors that affect comprehension or accuracy while accepting stylistic differences that do not compromise meaning. Content type and risk level determine which is appropriate.

Q: How does MTPE handle safety-critical content?
A: Safety-critical content (hazard warnings, emergency procedures, regulatory compliance notices) should always receive full post-editing in an MTPE workflow. The post-editor reviews every segment against the source, confirms that signal words and hazard classifications are correctly rendered, and checks that the output meets the regulatory language conventions of the target market. Safety content should never be processed through light post-editing.