Make sure important applications
are translated by a human expert.
Modern AI translations are generally produced by statistically trained neural translation models, including specialized NMT systems and, increasingly, large language models. These models learn conditional probabilities for target tokens or target sequences from corpora and generate target text by decoding. The result may be linguistically highly plausible, but it still does not guarantee the technical or legal equivalence of a patent claim.
Whether a claim translation reflects the protected subject-matter in the same legal sense cannot currently be determined reliably by the model alone; this requires expert review and, where appropriate, post-editing.
For this reason, for translations of applications that are not merely intended to inform the public, we recommend a translation prepared and/or reviewed by a human translator.
Sichern Sie wichtige Anmeldungstexte mit einer Übersetzung durch einen Fachübersetzer ab !
Modern AI translations are generally produced by statistically trained neural translation models, including specialized NMT systems and, increasingly, large language models. These models learn conditional probabilities for target tokens or target sequences from corpora and generate target text by decoding. The result may be linguistically highly plausible, but it still does not guarantee the technical or legal equivalence of a patent claim.
Whether a claim translation reflects the protected subject-matter in the same legal sense cannot currently be determined reliably by the model alone; this requires expert review and, where appropriate, post-editing.
Aus diesem Grund empfehlen wir Ihnen für Übersetzungen von Anmeldungen, die nicht nur zur Information der Öffentlichkeit dienen, eine menschlich erstellte und/oder geprüfte Übersetzung
Human patent translation
A human patent translation does not start with a machine-generated proposal; it starts with understanding. An experienced specialist translator analyses claim structure, technical teaching, reference signs, embodiments and legal effect before the first sentence is formulated. The translation is therefore built semantically from the patent itself — not retrofitted as a repair of a statistically or AI-generated text surface.
Post-editing entails the following risk: the machine pre-translation already determines sentence structure, terminology and semantic direction. Even a qualified translator will often correct within that framework instead of fully re-examining the technical and legal statement from first principles. In patent claims in particular, such subtle cues can be problematic because small deviations can alter the scope of protection.
By contrast, the traditional workflow — initial translation by a specialist translator followed by review by a second qualified patent translator — provides two independent expert perspectives. The first translator constructs the target version in a controlled manner based on the technology, terminology and claim logic. The second checks consistency, semantics and completeness.
The result is not an optimized machine version, but a responsible expert translation — precise, traceable and focused on protecting the invention.