A question from a patent outsider: is there any point in the patent process in which a patent might be translated (by a human, not automatically)? I am looking for patents in multiple languages to make a multilingual corpus (for NLP research). Do you know if a similar corpus is available or where one could get useful data in this direction? In none of that is available, is there a way to retrieve patents for the same invention in different languages, even if their text is not a direct translation?

  • 1
    In line with an answer and comments, some will be very professional direct translations and some may be significantly re-written and not suitable for inclusion. As a U.S. patent agent, I got a few poor English translations of applications originally filed in China and needed to restructure them to fit U.S. practices. The resulting U.S. applications' claims were supported by the original priority document but there was no textual correspondence between the applications in the U.S. and in China.
    – George White
    Commented May 22, 2022 at 23:01
  • Thanks a lot for your comment. For what I need, direct correspondence would be much better, but I hope some of the bad translations might be excluded automatically with some tools. Do you know if there is any public dataset that makes it easy to understand if two patents are a translation or are from the same family? Or, do patents have linking metadata if they are translated or from the same family?
    – ImAUser
    Commented May 23, 2022 at 15:46
  • 1
    Take a look at the INPADOC database epo.org/searching-for-patents/data/bulk-data-sets/inpadoc.html It collects patents into families of patents.
    – George White
    Commented May 23, 2022 at 17:38

1 Answer 1


There are some sources you could use for that corpus:

  • Different members of a same patent family: Sometimes an invention is to be protected in multiple jurisdictions, each of which has particular language requirements. If, for instance, the patent family has a US patent application, a French patent application, a Chinese patent application, etc. you can take all their publications for the corpus. Sometimes there are differences between the descriptions/claims of the different family members, so care must be taken in this regard. For the most part, these translations are prepared by skilled translators.
  • Translations required for the validation of European patents: Some states of the European Patent Convention require the translation of the claims and/or description of a European patent into one of its official languages, e.g. Czech, Danish, etc. These are supposed to be verbatim translations, and are typically prepared by skilled translators. The European patent will be in English, French or German. And European patents publish the claims in all three official languages: English, French and German, which are typically prepared by skilled translators as well.
  • Priority translations: Sometimes, but not very often, patent offices request the translation of a priority application into one of its official languages. Said translation is available through the public register of the patent office, and as far as I know, no OCR version is available. Although a verbatim translation, it could be a machine translation.
  • You might point out that (in my experience) these translations are done manually by skilled translators. In my old company we used to hire firms especially for this task. Also, versions of patent family may not be verbatim translations, especially between US and European applications and even more especially about claims.
    – Eric S
    Commented May 22, 2022 at 16:14
  • @EricS You are absolutely right. Thank you. I am adding those insights to my answer. Commented May 22, 2022 at 16:26
  • @theEuropeist - would you recommend excluding claims from the project the OP is doing since that may be where the most changes might occur beyond translation?
    – George White
    Commented May 22, 2022 at 22:55
  • Thanks a lot! Do you know if the data are publicly available in a form that is easy to query? I will look at the EPO and USPTO public datasets to see if there is anything available (I know the data are there, but I find them super difficult to query: eg., is there a way t know that two documents are from the same family -- eg through a code -- or that they are a direct translation?)
    – ImAUser
    Commented May 23, 2022 at 15:18

You must log in to answer this question.

Not the answer you're looking for? Browse other questions tagged .