Well-resourced, and still not solved
Russian has enormous training data behind it and engines produce fluent output. On real commercial content it still scores near the bottom of the major languages — because the difficulty here is grammatical and stylistic, not a shortage of examples.
Where the difficulty comes from
Russian is not a low-resource language. There is a great deal of digitised parallel text and engines have been trained on it for as long as they have existed. The problems are structural, and more data has not resolved them.
Case morphology carries the sentence. Six cases mark what English marks with word order and prepositions. Every noun, adjective, pronoun and participle in a phrase has to agree. Engines get short phrases right and break agreement in long, qualified, nested sentences — which is precisely the sentence type that legal, technical and regulatory documents are built from.
Verbal aspect has no English equivalent. Every Russian verb makes a choice between perfective and imperfective, encoding whether an action is completed, repeated, ongoing or attempted. English does not mark this, so the system must infer it from context — and in an instruction, a specification or a contract, choosing wrongly changes what is being required.
Word order carries emphasis, not structure. Because case marks grammatical role, word order is free to carry information structure — what is new, what is known, what is being contrasted. Engines default to a mechanical order that is grammatical and flattens the emphasis, producing text that reads as translated to any native reader.
No articles. Definiteness is conveyed by word order, context and particles. Translating from English means deciding what to do with every "the" and "a"; translating into English means recovering distinctions the Russian never marked. Both directions lose information silently.
Register is sharply divided. Russian technical, official and legal writing follows conventions considerably more prescriptive than their English counterparts. Fluent conversational output in an official document is a visible failure.
What we do
- We check aspect and modality as a named category, at critical severity in instructional and contractual content, because these are the errors that change an obligation while reading perfectly.
- We check agreement across long sentences as a discrete pass rather than trusting a fluency read.
- We set register by document type and score deviations separately from accuracy.
- We hold terminology across document families, particularly in technical and energy content where the same component recurs across thousands of pages.
- We use reviewers matched to the domain, not generalists.
Where Russian work is worth most
Energy, mining and heavy industry documentation, where the technical vocabulary is deep and the safety content is consequential. Legal and contractual material. Technical and engineering documentation. Pharmaceutical and clinical content for Russian-language markets.
Where sanctions, export control or counterparty screening affect a piece of work, tell us at the outset — it is a compliance question about your organisation and ours, not a linguistic one, and it is better raised in the first email than the last.