How to Run AI Document Translation on S1000D and ASD-STE100 Technical Manuals

To run AI document translation on S1000D and ASD-STE100 technical manuals, lock the data modules, protect the XML tags, build the term base, run the AI translation, have an engineer review it, then validate the output. Skip the tag step and the file won't load back into your Common Source Data Base.
AD VERBUM is an EU-hosted AI translation company that runs client-tuned open-weight models on structured technical publications for defence and engineering clients, with certified subject-matter engineers reviewing every procedure under ISO 17100 and ISO 18587. We hold AQAP 2110, ISO 27001 and ISO 42001, nothing goes to public cloud tooling, and we never train on your data.
Military technical manuals are where this work gets hard, because one wrong maintenance step is a safety event. This guide builds on our work on defence engineering documentation and military technical manuals for defence electronics, and it follows the structure that the S1000D specification defines.
What S1000D and ASD-STE100 are, and why they change the job
S1000D is an international specification for technical publications, produced jointly by ASD in Europe, the AIA in the United States, and the ATA e-Business Program. It breaks a manual into data modules: self-contained units of information, each stored in a Common Source Data Base (CSDB) under a unique data module code. A procedure, a description or a parts list is one module, reused wherever a publication needs it. The official S1000D overview lists the information types, from descriptions and procedures to fault information and illustrated parts data.
ASD-STE100, Simplified Technical English, is a controlled language with 53 writing rules and a dictionary of about 900 approved words. Issue 9 came out in January 2025. STE is a recommendation in S1000D, and EDSTAR recommends it as best practice for defence contracting by European Defence Agency participating member states, as the ASD-STE100 site explains.
Here is what that means for translation. STE controls the English source, so your target languages don't follow STE rules, and a translator can't repair a badly written source. What the structure gives you is leverage: a module that hasn't changed is translated once and reused, and a short sentence with one meaning per word is the best input an AI model gets. The risk sits in the markup. Touch a tag, a code or an applicability value and the module breaks.

The six steps, in order
Follow these in order. Each one protects the next:
Lock the data modules at a fixed issue, so the source can't shift.
Protect the tags, so only the readable text gets translated.
Build the term base from your approved terms and your STE technical nouns and verbs.
Run the AI translation on a client-tuned model constrained by that term base.
Have a certified engineer review every procedure, warning and value.
Validate the output against your schema and archive it with the review record.
The sections below take each step in turn, then list where projects fail.
1. Lock the data modules
Export a defined set of modules at a fixed issue and record each data module code, issue number and date. Translate from that frozen set only. If the source team revises a module mid-project, treat it as a new issue with its own translation and review cycle. Modules unchanged since the last release carry over from the archive, so you pay for the delta and not for the whole manual. The same freeze-first discipline runs through our functional safety documentation method.
2. Protect the tags
A data module is XML, and only the text between the tags is for translators. Element names, attributes, data module codes, cross-references, applicability values and identifiers stay exactly as exported. Agree the export format with your CSDB team and settle it on one pilot module before you scale, because the format decides how cleanly the tags survive. Schema validation belongs to your CSDB toolchain, so plan to run it there.
3. Build the term base
Pull your approved vocabulary into a term base first: equipment nomenclature, part names, procedure verbs and warning phrases. Add the STE technical nouns and technical verbs your programme uses, since STE permits them outside the dictionary. A client-tuned model constrained by that term base keeps one term for one concept in every language. Our note on self-hosted open-weight models for regulated translation explains why that beats a stock engine on controlled content.
4. Run the AI translation on a client-tuned model
Generate the draft on a model that already holds your terminology, not a public chatbot. We run client-tuned open-weight models on EU-hosted infrastructure, constrained by your term base and Translation Memory, with no public cloud tooling and no training on your data. Military manuals can carry controlled technical data, and under Regulation (EU) 2021/821, Article 2(9), technical assistance includes transmitting working knowledge by electronic means. Where that applies, route the classification call to your export compliance officer. Our notes on dual-use technical data and on secure translation with EU data residency cover the controls.

5. Have an engineer review every procedure
This step makes the output usable, and it needs a certified engineering linguist who knows the equipment, not a general proofreader. The reviewer checks:
Every warning, caution and note, so a hazard never softens in translation.
Every step of a maintenance procedure, in the original order, with no step merged or dropped.
Every value, unit and tolerance, against the source module.
Every equipment name and part reference, against the term base.
This is full human post-editing under ISO 18587, where the reviewer takes responsibility for the final text. A warning never ships on machine output alone.
6. Validate the output and archive it
Run the translated modules through your CSDB's schema validation, then check the data module codes, cross-references and applicability against the source set. Archive each language version with its frozen source issue, reviewer and change record, so you can show which text went into which publication. That traceability matches the record-keeping logic in our machinery manual migration guide, and in defence work it is a contract term.
Where it goes wrong
Most failures on structured manuals trace back to a skipped step. The ones we see most:
Sending whole XML files to a general engine, so tags, codes or attribute values change and the module fails to load.
Translating a module that was still under revision, so the delivered text cites an issue that no longer exists.
Skipping the term base, so one part name appears as three different terms across a manual.
Expecting the translation to carry STE rules, when STE governs the English source only.
Sending controlled technical data through public cloud tooling, so a transfer question lands on your desk after the fact.
Each one is avoidable with the source frozen first and the review owned by a certified engineer. The procedure someone follows with tools in hand is the text that never ships on machine output alone.
Our defence technical publication translation services
Our translation services for regulated sectors run on ISO 27001 and ISO 42001 certified, EU-hosted infrastructure, with no reliance on public cloud tooling for core processing. Every project runs through our AI+HUMAN hybrid workflow: we ingest client Translation Memories and Term Bases first, our proprietary LLM-based LangOps System generates output constrained by client terminology on client-tuned open-weight models, and our certified subject-matter experts review for technical accuracy and regulatory compliance. Our QA is aligned to ISO 17100 and ISO 18587, with sector-specific requirements such as AQAP 2110 quality assurance and Regulation 2021/821 controlled-data handling applied where relevant. We serve Life Sciences, Legal, Finance, Defense, and Manufacturing clients across 150+ languages with 3,500+ subject-matter linguists. For teams managing audit-sensitive content, contact us to discuss your security and compliance requirements directly.
FAQ
Can AI translation be used on S1000D technical manuals?
Yes, when the tags stay protected and a certified engineer reviews the output. AI document translation produces the draft, ISO 18587 full human post-editing puts a qualified reviewer in charge of the final text, and your CSDB validates the result against its schema.
What is S1000D?
S1000D is an international specification for technical publications, produced jointly by ASD, the AIA and the ATA e-Business Program. It splits content into data modules stored in a Common Source Data Base under a unique data module code, which is why one module can be reused across publications.
What is ASD-STE100, and does it apply to translated text?
ASD-STE100 is Simplified Technical English, a controlled language with 53 writing rules and about 900 approved words, current at Issue 9 from January 2025. It governs the English source only. Target-language text follows your own style rules and term base, not STE.
Which parts of an S1000D data module get translated?
Only the readable text between the tags: titles, procedural steps, warnings, descriptions and captions. Element names, attribute values, data module codes, cross-references and applicability values stay exactly as exported, and your CSDB validates them after translation.
Is it safe to send military manuals to an AI translation workflow?
Only inside a controlled, EU-hosted workflow with no public cloud and no training on client data. Manuals can carry dual-use technical data under Regulation (EU) 2021/821, so the classification call goes to your export compliance officer. AD VERBUM runs under ISO 27001 and AQAP 2110.
Which certifications matter for AI translation of defence manuals?
ISO 17100 and ISO 18587 for certified review, ISO 27001 for information security, ISO 42001 for AI governance, and AQAP 2110 for NATO quality assurance requirements. AD VERBUM is certified for all of them, with ISO 42001 and AQAP 2110 certified by Bureau Veritas.
Recommended
Best AI Translation Companies for Defence Engineering Documentation
Best AI Translation Services for Defence Electronics and Military Technical Manuals
How to Use AI Translation for Functional Safety Documentation Under IEC 61508 and ISO 26262
How to Move Machinery Manuals to AI Document Translation Before 20 January 2027
