Frequently Asked Questions: Omnilingua Technical Translation

For some purposes, yes. AI translation has genuine utility for low-stakes, non-contractual, internal content, such as getting the gist of a foreign-language document, early-stage research, or internal communications where precision is not critical.

For engineering manuals, technical specifications, safety documentation, and anything with legal, compliance, or contractual implications, AI translation alone is not a reliable option. The core problem is not that AI translation always produces poor results. It is that errors are unpredictable and often plausible-sounding. In technical content, a mistranslated specification, tolerance value, or contractual term can have real operational and financial consequences.

For an overview of the technical documentation types we handle, see our technical translation services page.

Yes, as a core part of our workflow, but in a way that is fundamentally different from using a generic online tool. We work with professional-grade Neural Machine Translation (NMT), currently among the most capable available for engineering and technical content, combined with Translation Memory Software (TMS) and specialist human oversight at every stage.

Our translators use NMT-generated suggestions as a starting point, applying their subject-matter expertise to review, refine, and approve every segment before delivery. For engineering, electronics, legal, and software content, that domain knowledge is not incidental: it is what allows errors to be caught that a purely automated process would pass through unchecked.

A full explanation of the tools and processes we apply is available on our technical translation services page.

It is possible and we can work with AI pre-translated content, but the cost saving is often considerably less than anticipated, and the approach can introduce complications that add time and cost back into the project.

The saving from submitting an AI pre-translated draft for professional post-editing, rather than commissioning a full translation, is roughly in the range of 30 to 50 per cent. That range is a guide only, not a guarantee, and reaching the upper end requires high-quality AI output on content that is not heavily terminology-specific. For complex technical, engineering, or legal documentation, savings are more typically towards the lower end of that range. The key variables are:

• The quality of the AI output. Technical documentation, engineering standards, and legal clauses are precisely where consumer AI tools produce their most unreliable results. A post-editor working with a poor AI draft may need to retranslate significant portions, leaving little or no saving.
• The subtlety of technical errors. Mistakes in technical translation are frequently not obvious: a tolerance in the wrong unit, a specification rendered ambiguously, or a contractual term applied with the wrong jurisdictional meaning. The reviewer must work methodically against the source text, which takes time.
• Loss of terminology database benefits. Translating outside our managed process means the work does not contribute to your TMS account. You lose the immediate savings on repeated content and the compound savings that build over time.

Our practical recommendation: for most technical documentation, the NMT and TMS combination we already apply in our standard workflow delivers better cost efficiency, greater terminology consistency, and fewer downstream complications than the pre-translate-then-proof approach.

These are the errors that occur most frequently in AI-translated technical content, and the ones that cause the most problems in practice:

• Terminology inconsistency across a document set. A single technical component may be referred to by three different translated terms across a manual, a specification, and a labelling document. Without a managed terminology database, AI tools make independent choices every time.
• Mistranslation of units, tolerances, and specifications. Numerical content, measurement units, and engineering tolerances are high-risk. An AI error here can affect product performance, safety conformance, or the terms of a contractual obligation.
• False cognates and near-miss technical terms. Engineering vocabulary is particularly vulnerable to words that look similar across languages but carry different technical meanings. AI tools operate at the surface level and do not catch these.
• Jurisdictional errors in legal and contractual content. Terms such as ‘warranty’, ‘liability’, ‘indemnity’, and ‘remedy’ have specific meanings in English law that may not have direct equivalents in other legal systems. A generic AI tool applies a surface translation without understanding the jurisdiction.
• Software and UI localisation errors. Text strings for software interfaces are typically fed to AI tools without surrounding context. The word ‘Cancel’ means something very specific in a payment workflow and something different in a device calibration sequence. Without context, AI tools guess, and the consequences for end users can be significant.
• Inconsistency across related documentation. Technical projects routinely span multiple documents: manuals, specifications, compliance certificates, tender documentation. AI translation without a managed terminology database makes cross-document consistency very difficult to achieve or verify.

Our software localisation blog post covers the specific challenges of translating UI and device software content in detail, including the context-loss problem that makes AI particularly unreliable in this area.

These practical questions will usually make the right approach clear:

• Does the document have legal, contractual, or compliance implications? If yes, specialist professional translation with human oversight is the appropriate choice.
• Will engineers, operators, or end users rely on this content? If yes, accuracy and consistency across the document set are essential.
• Does the content contain specialised technical terminology, references to engineering standards, or jurisdiction-specific legal language? If yes, a translator with domain expertise is required.
• Is this internal, low-stakes, or exploratory? AI tools may be appropriate in this case.

If you are uncertain, our technical translation team is happy to advise on the right approach for your documents and sector, with no obligation.