Machine translation has improved dramatically over the years. Tools like DeepL, Google Translate, and the translation features embedded in major document platforms now produce output that is often grammatically correct, frequently fluent, and occasionally impressive.
For informal use like understanding a foreign-language article, communicating roughly with a counterpart in another language, these tools are genuinely useful as they have changed the landscape of quick communication in ways that are hard to overstate.
For professional documents, however, the picture is more complicated. And the costs of getting it wrong are significantly higher than most organisations realise.
What Machine Translation Is Actually Doing
To understand where machine translation fails, it helps to understand what it is doing. Modern translation tools, particularly those based on large language models(LLMs), are pattern-matching systems. They predict what words should follow other words based on vast amounts of training data. They are exceptionally good at producing plausible-sounding text.
*Plausible-sounding is not the same as accurate. In professional translation, the gap between the two can be extremely costly.
Machine translation systems do not understand context the way a human translator does. They do not know the jurisdiction, the relationship between the parties, the regulatory environment, or the conventions of the document type. They produce text that looks right far more reliably than text that is right.
Where Legal Translation Goes Wrong
Legal documents from contracts, agreements, compliance filings, and regulatory submissions all depend on precision that machine translation cannot guarantee. A single term translated incorrectly can alter the meaning of an obligation. A subtle difference between "shall" and "may", between "indemnify" and "hold harmless", between "material breach" and "substantial breach", all of which these distinctions have legal consequences that machine translation systems are not equipped to navigate.
Legal language is also highly jurisdiction-specific. Terms that have a precise legal meaning in one country may not have an equivalent in another legal system. A competent legal translator does not just translate words; they translate concepts across legal frameworks. This is a task that requires both linguistic and legal expertise. No current machine translation system has it.
The NGO and Development Sector Problem
In the development sector, translation errors carry a different kind of risk. Programme documents, beneficiary-facing materials, monitoring tools, and donor reports all depend on accurate, culturally appropriate language.
Machine translation tends to struggle with two things that are particularly prevalent in development documents: specialised sector terminology and local language nuance. A term like "participatory needs assessment" may be translated in ways that are technically accurate but carry connotations in the target language that undermine the approach being described. A concept that is well understood in English development discourse may not translate cleanly into French, Spanish, or Swahili, and the gap requires a translator who understands both the language and the sector.
In beneficiary-facing materials, errors are not just embarrassing. They can cause confusion, erode trust, and in extreme cases give incorrect information to vulnerable populations.
The Business Cost of Machine Translation Errors
In commercial contexts, investor documents, partnership agreements, and client proposals, the reputational risk of poorly translated materials is underestimated. A document with translation errors signals to a foreign counterpart that the organisation did not consider the relationship important enough to invest in proper translation.
Beyond reputation, there are operational risks. A contract translated inaccurately by a machine creates ambiguity that can lead to genuine disputes about obligations and terms. Resolving those disputes, in legal fees, time, and relationship damage, typically costs far more than the translation would have.
The Specific Failures to Watch For
Machine translation errors in professional documents tend to cluster around several categories:
Terminology: sector-specific or legal terms translated using general-language equivalents that carry different meanings
Register: formal documents translated into an informal register or vice versa, creating a mismatch with the document's purpose
Ambiguity in source text: machine translation resolves ambiguous source language by guessing; a human translator flags the ambiguity and seeks clarification
False cognates: words that look similar across languages but mean different things, a persistent source of error in European language pairs
Cultural reference: idioms, examples, and cultural references translated literally rather than adapted
The Appropriate Use of Machine Translation
This is not an argument that machine translation has no place in professional workflows. It does when used appropriately. Machine translation can significantly accelerate the work of a professional translator doing post-editing. It can provide a working draft that a qualified human then reviews, corrects, and refines. In this workflow, quality is maintained while costs and turnaround times are reduced.
What it cannot do is replace the review step. A machine-translated document submitted as final is not a professionally translated document. It is a draft that has not been checked, and in professional contexts, that distinction matters.
What That Costs You
The question for any organisation is not whether machine translation is impressive. It is whether the risk of error, be it legal, reputational, operational, or relational, is acceptable in the context where the document will be used.
For internal notes and quick communications, often yes. For pitch decks, grant proposals, contracts, compliance documents, and beneficiary-facing materials: the cost of getting it wrong is almost always higher than the cost of doing it properly the first time.