TranslateHub

TranslateHub translates prose the way a translator would, not the way a phrasebook would - and then tells you how well it went.

The problem with machine translation was never that it is bad. It is that you cannot tell when it is bad. A page comes back fluent and confident, and somewhere in it are two sentences that mean the wrong thing, and you have no way to find them short of reading the whole thing in a language you may not speak.

So every translation comes back with per-sentence confidence. Instead of a wall of text you cannot evaluate, you get a shortlist - the three or four lines the model was unsure about. Those get a human minute; the rest ships.

grade_translation goes further and scores any translation independently on accuracy, fluency, tone, terminology and authenticity, with written feedback. It works on translations this service did not produce, which makes it the tool for comparing two candidates or deciding whether a passage needs a reviewer at all. It is the same grader used to evaluate human translator applicants - a judgement a model cannot reliably make about its own output.

For longer work, voice fingerprints keep an author sounding like themselves across passages, and a glossary keeps terminology fixed where it matters. Billing is by the word at the same rate as going direct, so nothing is lost by routing through an agent.

Full product at www.translatehub.io.