Community patent: machines to the rescue!

First with the news, Managing Intellectual Property magazine reported yesterday that the EU Competitiveness Council is meeting on 29 and 30 May to discuss progress on technical issues relating to the fabled Community patent. Previous attempts to clinch agreement on the Community patent, which would provide a single unitary right covering the territory of all 27 EU member states, have foundered due to concerns over language, cost and court proceedings. With 23 official languages, a patent filed in any of those languages currently needs to be translated at great expense if country-by-country pan-European protection is sought, which is why most applicants cherry-pick the key jurisdictions and omit many of the minor ones. The latest plans however seek to employ machine translations, which are already available for German, Spanish and French at the European Patent Office.

These plans were previously discussed at meetings of an EU Council working party under the Slovenian Presidency, which expires at the end of June. The translation tools would include databases of several million standardised technical terms, which would be translated into all EU languages. The machines are capable of delivering translations in about 45 seconds. Under the draft plans, machine translations would be provided for all 23 languages, ensuring equal treatment. However, setting up the translations would take several years and cost at least €2 million. Existing technology would also need an upgrade. Machine translations would be for information purposes and would have no legal status. For the estimated 1% of patents that are litigated in infringement proceedings or the like, human translations would be needed. A working document with full details of the proposal is due to be published in a couple of weeks, in time for the Competitiveness Council meeting at the end of the month.

The IPKat looks forward to the new game of Community Patent Whispers. You take a claim, feed it into the translation machine for the first official language (taking them in alphabetical order); when you get the output, you feed it in for translation into the second, and vice versa. When the claim has been through each official language once, you put it before an EPO Board of Appeal and ask them to interpret it. Merpel says, my favourite term in European patent law is the phrase "as such". If you translate it from English to Dutch, Dutch to French, French to German and then German to English you get "ash look for".

Community patent: machines to the rescue! Community patent: machines to the rescue! Reviewed by Jeremy on Thursday, April 24, 2008 Rating: 5


  1. Lets not forget that human translators are not infallible. Significant errors can and do arise during a single translation, let alone if you translate the same phrase via 23 languages.

    As you point out - machine translations would be for information purposes only, the original would remain decisive in the case of differences. Current freely available translations are already good enough normally to get the gist of what a patent is about before making a decision as to whether it is important enough to get professionally translated. More use of this has to be a good thing.

  2. Jeremy,
    as i suggested recently in a guest editorial in World Patent Information, there is an easier and more accurate solution. Require electrically filed EP patents to have the disclosure (but not the claims) written in a word processor which constrains the writer to write text which is 'digestible' by the particular software the EPO uses. Then the discosure could be translated more accurately to any of the languages the software supports.

  3. With more and more member states, the EU sometimes has difficulty in finding every linguistic combination for interpreting and translation. If the number of languages is n, and each translator knows two languages, then in order to have one translator for every possible combination of languages the number of translators required is n x (n-1) (which is also expressed as n2-n). Assuming that the EU has 15 languages, this means they need 15x14 = 210 different translators just to cover every different possible combination.

    For example, they might not be able to find a translator who knowns both Finnish and Greek. To get around this problem, they translate the Greek text into a third language, say English, and then from English into Finnish. This has not thus far caused any major upsets within the EU (as such!).

    However, as the posting correctly points out, the greater inaccuracies of the machine translation, mean that the translation via a third language carries real risks since it disproprionately multiplies the errors made in the original translation, which being a machine translation was hardly perfect to begin with.

  4. Not providing is the patent (including the description) in the language of the country is just an offense to people who does not master foreign languages.

    The London Agreement is a pure disaster for people who wants to want what is patented, and this only widden the gap between the patent system and the citizens.

    Why citizens should be discriminated in function of the language they speak?

    And for the patent attorneys, no the claims are not enough to understand a patent.

  5. We need one official EU language - for sole use in all official documents, including patents. It should be Latin (as it used to be). This gives (roughly) equal pain to all member countries. It removes the need for a horde of official translators - 2(n-1)factorial, where n is the number of languages (if I've worked it out right). If a country wants an official document translated for the benefit of its citizens, it can pay for it (subsidiarity). We have one master text (so no need to argue whether 'animal varieties' is the same as 'Tierarten'). We sell it to the French on the basis that it will infuriate the Americans. And it clears the way for English to become the unofficial EU language.

  6. Jeremy, the important figure is not the 1% of patents which may be litigated, it's the 100% of patents which may be used as prior art during examination before any of the world's patent offices. Anyone who has seen a USPTO office action quoting a machine translation of a Japanese patent as a supposed prior art on your invention when the sentence quoted means nothing like what the examiner thinks it means knows this. The basic problem is it then ends up being your (Japanese-speaking collegue's) word against the examiners, unless your willing to pay for an official translation that is.

  7. Hi folks,

    there are at least two different machine translation options: rule based (RBMT) and statistical (SMT). RBMT for so many language pairs is beyond any hope, see EuroTra disaster. SMT for so many languages can be obtained easily and cheaply if (human) translations of similar texts exist (which happens to be the case for many languages, because of Art. 65 EPC).
    Have a look at Euromatrix and Google. Syntax is sometimes very bad, in particular if the word order in both languages differs. Moreover morphology such as agreement in gender and number is problematic. However, word sense disambiguation is typically fine.

    SMT works better between closely related languages (e.g. FR-PT-ES-IT-EN) and less good for languages with rich morphology (German, Finnish), e.g. concatenation of nouns.

    Proposal: the EU funds a project to scan/OCR available translations and makes them available to the SMT community. In addition a Euromatrix2 funding stipulates further research on statistical machine translation, in particular domain dependent SMT using IPC classification.

    See also Utiyama's paper on SMT of patents.

    Further remark: Things get easier if only one source language is used (English).

    Last remark: a bad MT is better than late and unsearchable human translations as it happens to be the case right now. So at the day of publication you may search prior art using CLIR, then obtain the original in one of the three official languages. Then you look for the original applicant. For EN, DE and FR you will find people which can judge if a patent is relevant.


All comments must be moderated by a member of the IPKat team before they appear on the blog. Comments will not be allowed if the contravene the IPKat policy that readers' comments should not be obscene or defamatory; they should not consist of ad hominem attacks on members of the blog team or other comment-posters and they should make a constructive contribution to the discussion of the post on which they purport to comment.

It is also the IPKat policy that comments should not be made completely anonymously, and users should use a consistent name or pseudonym (which should not itself be defamatory or obscene, or that of another real person), either in the "identity" field, or at the beginning of the comment. Current practice is to, however, allow a limited number of comments that contravene this policy, provided that the comment has a high degree of relevance and the comment chain does not become too difficult to follow.

Learn more here:

Powered by Blogger.