Machine translation is improving, but the enterprise gap remains
Why impressive results in neural translation do not mean a company can remove translators from its workflows.
Machine translation has improved noticeably over the past few years. Neural approaches, which began appearing in research papers, produce more coherent and natural translations than the previous generation of systems. This is visible even on a quick comparison.
The improvements are real. But between "the translation reads better" and "the company can remove a professional translator from the process" lies a considerable distance. That distance is exactly what a manager should understand when evaluating where machine translation works and where it does not.
What changed technically
Neural approaches to translation work differently from the statistical systems of the previous generation. They handle word order, agreement in complex sentences, and idiomatic expressions better. The result looks more readable even where it is inaccurate - which is simultaneously an improvement and a problem.
"Looks readable" does not mean "accurate". Translation errors that previously were obvious from grammatical inconsistencies now hide behind smooth prose. A non-professional reader will not notice them where a professional would.
Where machine translation works well
For tasks where the overall meaning matters but precise detail does not, current systems are already useful. Understanding what a letter in an unfamiliar language is about. Getting oriented quickly in a document. Processing a large flow of communications to identify what needs priority attention.
For these scenarios, machine translation saves time and enables working with materials that were previously inaccessible due to language barriers.
Where the gap remains
Corporate contexts where accuracy is critical remain difficult to automate.
Legal and contractual documents. Terminology here has strict meaning. An error in translating one term changes the meaning of an obligation. There is no room for "roughly correct".
Technical documentation with high accuracy requirements. Safety instructions and manufacturing specifications - a translation error can have consequences beyond the text itself.
Communication where tone matters. Negotiations, correspondence with key partners, public materials. Machine translation conveys the meaning but loses the tonal nuances that are often more important than the literal content.
Texts with high density of domain-specific vocabulary. In narrow specialisations, machine systems often lack sufficient training material and produce inaccurate translations of professional terms.
A practical approach
A sensible position is to use machine translation as a first pass where it speeds up work, and bring in a professional translator where the consequences of an error are significant.
The question that helps draw the boundary: if the translation is inaccurate in a specific place, what happens? If the answer is "nothing serious" - machine translation is appropriate. If the answer is "legal consequences", "a technical incident", or "a damaged relationship" - a professional translator is still needed.
The technology is improving and this boundary will shift. But right now it is still where it was - just less visible behind a smoother output.