What we work on
- PCT and national phase filings
- Claim sets and specifications
- Office action responses
- Prior art and literature for search and analysis
- Opposition and litigation material
- Freedom-to-operate documentation
Why patent content is the hardest case for automation
Claim language is engineered. Every word is chosen for scope. "Comprising" and "consisting of" are not synonyms; they are different rights. A translation engine optimises for natural expression, which is precisely the wrong objective — the source is not natural language, it is a legal instrument written in the shape of a sentence.
Term consistency is load-bearing. The same element must carry the same term throughout the specification and the claims. An engine that renders a component two ways across a document has created an ambiguity that an examiner or an opponent can use.
The errors are silent. In Chinese, verbs do not mark tense — a documented failure has machines rendering "has actively promoted" as "actively promotes". In a claim, a shift like that changes what is being asserted, and it is completely invisible to anyone reviewing in English or reading a fluent back-translation.
Consequences do not scale with document length. A three-word error in a claim can invalidate a filing. There is no other content type we work in where the ratio of consequence to volume is this extreme — which is why patent work carries a rate premium of roughly 60 to 200% over general translation, and why that premium is rational rather than opportunistic.
Where the volume is
China is the largest source of international patent filings in the world, which makes Chinese the highest-volume patent language and — given everything above — one where the gap between benchmark performance and filing-grade accuracy is widest. Japanese and Korean follow, both major filing jurisdictions with extensive technical documentation behind each application.
Those are the three languages where independent testing shows machine output is least reliable, in the content type where reliability matters most. That intersection is where we concentrate.
What we deliver
- Reviewers with technical field expertise as well as patent experience — a chemistry claim and a mechanical claim are not reviewed by the same person.
- Claim-critical language checked as a separate pass, distinct from the general accuracy review, with its own severity scale.
- Terminology held across the specification, the claims and the family, maintained as a termbase you keep.
- A structured error report, so your attorneys see exactly what was found and where, rather than receiving a silently amended file.
- Documented reviewer qualifications and a version trail, for the file and for any subsequent dispute.
What we will tell you honestly
Machine translation is genuinely useful in patent work — for prior art search, for freedom-to-operate reading, for deciding whether a document is worth a human's time. For that purpose it is excellent and you should use it.
For filing-grade text, it is a first draft. We are happy to tell you which of your patent content is which, and we would rather you spent the review budget on the claims than spread it evenly across everything.