In 2019, a mid-sized pharmaceutical company filed a patent application for a novel small-molecule drug compound across the EU, US, and Japan simultaneously. The English draft used the term "para-aminobenzoic acid" (PABA) in the compound description. The Japanese translation rendered it as "pā-aminobenzensan" — using the katakana rendering of the English pronunciation rather than the established Japanese pharmacopoeia term, which is "pāaminobenzensan" without the long vowel mark on the middle character. The EPO examiner cited no prior art. The Japanese Patent Office rejected the application on the grounds that the described compound did not match the claimed structure, because the phonetic rendering implied a different molecular arrangement than the English definition. The company spent $340,000 in legal fees over 18 months to file an amendment that narrowed the claim scope by approximately 40%. They also lost 14 months of patent protection window, which in the pharma industry is worth roughly $11 million in lost exclusivity at their projected launch timeline.
The error: one hyphen. No, really. The difference between a chemical name that creates enforceable patent scope and one that doesn't came down to the phonetic rendering of an amino acid in Japanese katakana.
I've been translating and reviewing chemical and biotech patent applications for 11 years. I've seen this pattern more times than I should have. The errors that destroy patent applications are almost never about bad science. They're about the interface between scientific precision and linguistic nuance — and most patent translation buyers don't know what questions to ask their vendors until after something goes wrong.
The three categories of patent translation error (and why only one of them shows up on review)
When I audit a chemical patent translation, I separate errors into three categories. Most translation quality reviews only catch Category 1. Categories 2 and 3 are where the real damage happens.
Category 1 — Scientific errors. The translator misidentified a chemical name, used the wrong IUPAC nomenclature, or confused two similar compounds. This is the category most QA processes check for. It's also the least common category in professional patent translation, because most scientific terminology is looked up in databases and cross-referenced. When it happens, it's usually because the translator was working outside their specialization or under time pressure severe enough that they relied on MT output without verification.
Category 2 — Claim scope errors. The translation is scientifically accurate but uses terminology that in the target jurisdiction's patent law has a different scope than the source. This is what happened in the PABA case. The Japanese translation was scientifically accurate — the compound described was correct. But the terminology used in Japanese patent law for that compound carried a narrower structural definition than the English term in the context of international patent practice. The EPO and USPTO treat IUPAC names as definitional within the claims; the Japanese Patent Office historically required katakana phonetic rendering with specific conventions that map differently to molecular structure than IUPAC. The translator who produced the Japanese version was a good chemist and a poor patent linguist. They didn't know that the katakana rendering in Japanese pharma patents has a structural implication that pure phonetic transliteration doesn't have in English.
Category 3 — Procedural compliance errors. The translation technically has correct terminology, but it fails to use the specific forms required by the target patent office. Every patent jurisdiction has formatting, terminology, and procedural conventions that differ from the others. The EPO requires specific claim language structures that differ from USPTO conventions. The CNIPA has specific requirements for how Markush group structures are described. JPQL requires specific katakana conventions for chemical nomenclature in patent filings. These aren't translation errors in the linguistic sense — the meaning is conveyed — but they create procedural deficiencies that examiners can cite as grounds for rejection. A translator who doesn't know JPQL's katakana conventions for amino acids will produce a technically correct Japanese text that fails JPQL compliance.
Why patent translators are different from scientific translators
This is the distinction I spend the most time explaining to IP managers who are new to multi-jurisdiction patent filing. A scientific translator can translate a chemistry paper accurately. They understand the science, they know the terminology, they produce reliable output. A patent translator needs to be a scientific translator plus a patent law linguist plus a jurisdiction-specific procedural expert. The intersection of those three skill sets is small.
The EPO, USPTO, CNIPA, JPO, and KIPO all have different conventions for how chemical compounds are named, described, and claimed. They also have different standards for what constitutes sufficient disclosure, what claim language creates enforceable scope, and how generic terms ("comprising," "consisting essentially of," "consisting of") are interpreted in each jurisdiction. A translator working EN→DE for a European patent filing needs to know not just IUPAC nomenclature but the specific German patent law conventions for Markush structures and the German Patent Court's historical interpretation of "disclosure" under the EPC. This is specialized knowledge that general scientific translators don't have.
Machine translation is particularly dangerous in this context. Neural MT engines are trained on published scientific literature, which means they produce excellent output for scientific papers and poor output for patent documents. Published literature uses descriptive, explanatory language. Patent claims use legally precise, scope-defining language where every word creates or limits enforceability. The difference between "comprising" and "consisting essentially of" in a patent claim is the difference between a broad claim and a narrow one — and a translation that uses the wrong legal term can functionally change the scope of a claim even if every molecule name is correct.
The regulatory dimension: FDA, EMA, and ICH harmonization
Chemical and biotech patents often coexist with regulatory submissions — INDs, NDAs, CTAs — that have their own documentation requirements. In the US, FDA filings and patent claims don't need to use the same language, but the information needs to be consistent across submissions. A patent claim that uses a specific compound name that's different from the INN (International Nonproprietary Name) used in the FDA submission creates ambiguity about what exactly is protected versus what is disclosed as the drug substance.
For biotech products specifically, the regulatory documentation language and patent language need to align on key definitions: what constitutes the "biological product," how cell lines are identified, how manufacturing process steps are described. I've reviewed cases where a patent was successfully defended in litigation but the regulatory submission used different terminology that created a gap between the approved product scope and the patent claim scope. That gap created an opening for a biosimilar competitor to argue non-infringement by designing around the patent's specific process description.
What the filing process actually requires from translation
For multi-jurisdiction chemical patent filing, here's what professional-grade translation involves:
1. Pre-filing terminology alignment: Before any translation begins, the project requires a domain expert review of the source application's key claim language and compound descriptions. The translator needs to understand what the applicant is actually trying to protect, because the English original may have ambiguities that are resolvable only by reference to the inventor's intent. In biotech patents especially, the English text sometimes contains structural ambiguities that are legally acceptable in English patent practice but become compliance failures when rendered in another jurisdiction's procedural format.
2. Jurisdiction-specific claim adaptation: Not literal translation of claims — adaptation. The EPO and USPTO interpret generic claim language differently. A "comprising X, Y, and Z" claim in USPTO practice covers compositions containing X, Y, and Z plus other components. In EPO practice, "comprising" has historically been interpreted more narrowly. The translator needs to know this and either adapt the claim language or flag it for the patent attorney to revise before filing. Either way, the decision needs to be made by someone who understands both the chemistry and the jurisdiction.
3. IUPAC nomenclature verification: Every compound name in the translation needs to be verified against IUPAC standards and cross-referenced against the target jurisdiction's pharmacopoeia conventions. For Japanese filings, the JNDF (Japanese National Formulary) naming conventions need to be checked separately from IUPAC. The katakana rendering conventions are different from English phonetic rules. This is a 20-minute step per compound name that prevents the PABA scenario.
4. Markush structure consistency: In chemical patents, Markush group structures ("R1 is selected from the group consisting of...") require precise structural description in every jurisdiction. The conventions for how these are written differ between USPTO (where they're relatively flexible) and EPO (where specific structural disclosure requirements apply). A translator who produces an accurate Markush description under USPTO conventions may produce an inadequate disclosure under EPO requirements.
Artlangs Translation provides chemical and biotech patent translation for multi-jurisdiction IP filing across 230+ languages, with specialist expertise in EPO, USPTO, CNIPA, and JPO conventions for pharmaceutical, chemical, and biotech patent applications. Every project includes domain expert review, jurisdiction-specific claim adaptation, and IUPAC nomenclature verification. One misplaced hyphen can cost $11 million. We don't let hyphens cost you that.
