Patent professionals spend months debating claim language, arguing over a single word’s scope, and negotiating antecedent basis clause by clause. Yet somewhere in that same document sits a string of numbers — concentrations, tolerances, temperature ranges, dimensions — that often receives far less scrutiny during translation, even though those figures can carry as much legal weight as any verb or adjective in the claim. 📐
Numbers feel objective. A digit looks like a digit in any language, so it’s easy to assume they translate themselves. They don’t. And when a number moves even slightly during localization — a decimal shifts, a unit converts incorrectly, a rounding convention changes — the scope of protection can shrink, expand, or collapse entirely without anyone noticing until litigation, an office action, or a competitor’s design-around forces the issue into daylight.
Why Courts Treat Numbers as Claim Boundaries, Not Decoration
Numerical ranges in patent claims aren’t illustrative — they function as hard boundaries of the invention, and courts have repeatedly had to decide exactly how “hard” those boundaries are. In a closely watched UK Court of Appeal dispute between ConvaTec and Smith & Nephew, the judges had to determine whether a claimed range of “between 1% and 25%” should be read as an exact mathematical boundary or whether it implied a wider range depending on how precisely the patentee had expressed the figures. The court ultimately held that a value of 1 could be read as encompassing anything from roughly 0.5 up to 1.5, depending on the degree of precision a skilled person would infer from the surrounding specification. Venner Shipley
That single interpretive question — how many decimal places does this number really mean? — decided whether a competing product infringed or fell safely outside the claim. Now layer a translation step on top of that kind of ambiguity. If a translator rounds, truncates, or reformats a number differently than the original drafter intended, the “degree of exactitude” a court later relies on may no longer match what the inventor actually disclosed. Courts have been clear that there is no justification for using rounding or approximation to change the disclosure of the prior art or modify the scope of an alleged infringement — which means a translation-introduced rounding error isn’t a harmless simplification. It’s a substantive alteration of the invention’s legal boundary. De Rebus
When a Mistranslated Number Becomes an Unfixable Trap
Numerical and technical mistranslation isn’t a hypothetical risk confined to conference presentations — it shows up in real prosecution histories. The Technical Boards of Appeal at the European Patent Office reviewed a case involving a European application originally filed in Chinese, where the subsequent English translation was found to be inaccurate in a way that extended the subject matter beyond what had originally been disclosed. Because European patent law prohibits both adding new subject matter after filing and later broadening a granted claim back to that added matter, the applicant landed in what practitioners call an inescapable trap between Articles 123(2) and 123(3) of the European Patent Convention — a scenario where the patent can become effectively unrepairable because of an error introduced at the translation stage, long before anyone thought to double-check it. 📉
This is precisely the danger with numerical content: unlike a stylistic phrase, a number that’s been altered during translation doesn’t just read awkwardly — it can constitute new technical information that wasn’t in the original filing, or it can quietly delete technical information that was. Either direction creates exposure under added-matter provisions that are notoriously difficult, and sometimes impossible, to cure once the patent has granted. The EPO’s own guidelines on added subject matter make clear how narrowly these corrections are permitted after grant.
The Quiet Culprit: Unit Conversion and Locale Formatting
Not every numerical error comes from a dramatic mistranslation of technical meaning. Many originate in something far more mundane: how numbers are formatted and converted across languages and measurement systems.
Consider a measurement expressed in feet in a source specification. A translator or machine translation engine converting that figure into metric units has to correctly compute the conversion and then correctly place the decimal — miss either step and a dimension that should read as roughly 30 meters can instead appear as 100 meters, an error large enough to describe a physically different structure. Research evaluating machine translation systems on exactly this kind of task found that unit conversion is one of the areas where automated engines are most likely to produce a fluent-sounding but numerically wrong result, precisely because the system has to “calculate” rather than simply carry a number across languages. 🌐
Locale formatting compounds the risk. Many languages use a comma as the decimal separator where English uses a period, and vice versa for thousands separators. A figure meant to read as one point five can be misread or mistranslated as fifteen hundred if the separator convention isn’t handled with precision — a transformation that changes a claimed tolerance by three orders of magnitude. These aren’t exotic edge cases; they’re everyday hazards in any specification containing dimensions, concentrations, pressures, or thermal ranges, which describes the overwhelming majority of mechanical, chemical, and pharmaceutical patent filings.
A Snapshot of Where Numerical Risk Concentrates
| Risk Area | What Can Go Wrong | Typical Consequence |
|---|---|---|
| Decimal placement | Digit shifted or dropped during formatting | Claim scope narrows or expands unintentionally |
| Unit conversion | Incorrect metric/imperial conversion | Physical dimension no longer matches invention |
| Rounding conventions | Significant figures altered without disclosure basis | Added or lost subject matter under Art. 123(2)/(3) |
| Range language | “Between,” “up to,” “approximately” translated inconsistently | Ambiguous boundary interpretation by examiners or courts |
| Locale separators | Comma/period decimal confusion | Order-of-magnitude misstatement of a claimed value |
Where This Surfaces Long After Filing
The unsettling part of numerical translation errors is timing. A misplaced decimal in a claim rarely triggers an objection at filing — examiners are reviewing technical merit and prior art, not auditing every digit against a source-language original filed months or years earlier in a different language. The error tends to surface only when it matters most: during an infringement dispute, a validity challenge, or a licensing negotiation, when opposing counsel or an examiner scrutinizes the translated numbers against the original filing and finds a mismatch that can’t be waved away as a typo.
By that point, correction options are limited and often unavailable. As industry commentary on patent translation practice has noted, once a patent has issued, translation-related errors in the claims may remain uncorrectable, leaving the patentee exposed to litigation risk, enforceability challenges, or outright invalidation depending on the jurisdiction and the nature of the discrepancy.
A Short Self-Check for Numerically Dense Filings
Before a translated specification moves to filing, a few targeted questions can catch what a general proofread often misses:
- Does every numerical range in the translation match the source document’s precision — same number of decimal places, same significant figures?
- Were any measurements converted between unit systems, and if so, was the conversion verified independently of the translation itself?
- Are decimal and thousands separators consistent with the target jurisdiction’s formatting conventions rather than carried over by default from the source language?
- Do qualifying terms around numbers — “approximately,” “at least,” “up to” — appear with the same frequency and placement as the original, since these words often define whether a range is open or closed?
- Has a subject-matter-competent reviewer, not just a linguistic proofreader, checked the translated numbers against the underlying technical disclosure?
None of these questions require exotic tooling. They require treating numerical content in a patent specification with the same rigor applied to claim language itself — because legally, that’s exactly what it is. ⚖️
Sources:
source: https://www.epo.org
source: https://www.vennershipley.com
source: https://www.kwm.com
source: https://www.ata-divisions.org