The IPKat has just seen this announcement on the European Patent Office (EPO) website:
In a major step to improve access to patent documents in multiple languages the EPO today launched a new machine translation service, called Patent translate, on the EPO's website. The service uses Google's Translate technology and enables translation from and to English for French, German, Spanish, Italian, Portuguese and Swedish, covering approximately 90% of all patents issued in Europe. By the end of 2014, the service will also be able to translate patents from and into all 28 languages of the EPO member states, as well as Chinese, Japanese, Korean and Russian.
This collaboration between the EPO and Google was announced a little over a year ago, and this is the first stage of the implementation. The EPO website also has (apparently in today’s world obligatory) FAQs. The translation can be accessed either via Espacenet or the EPO Publication Server. Instructions for use can be found here.
Basically, once you have up the text that you want, you just click a button and the translation appears. The IPKat has performed a quick test accessing via Espacenet, and has found that procedurally it works well. It will take him a little longer to check what is the quality of the translation. This of course is the important point. If the machine translation does not result in a text which can actually be understood, then it is of very little value indeed. If any readers are trying the system, please let the IPKat know your experience.
Patent Translate – EPO and Google launch new service
Reviewed by Darren Smyth
on
Wednesday, February 29, 2012
Rating:
Reviewed by Darren Smyth
on
Wednesday, February 29, 2012
Rating:
Follow the IPKat on LinkedIn 
![[Guest post] EUIPO Second Board of Appeal refers case to Grand Board of Appeal following earlier refusal to register a photograph of a man’s head/face as an EU trade mark](https://blogger.googleusercontent.com/img/b/R29vZ2xl/AVvXsEiUj9v-k3DYaSVMR4lsmKDLHVB5nIoINuvZNbH2TQCelrfaAE9qmGSAQs-plzQ7HnZ7Svk_gRdrIODp0Nat5vpWajduit3xkArNhK76AHhX770RIJ62thfCVI5fnvcXU-5zH4NkUSfPY9FjRMFiK7OKJ7m7YSo2FzhyPSoA8ZWCsxaOXb0bBCoPQA/s72-c/Picture%201.png)
![[Guest post] Benelux Office rules on opposition proceedings prepared by ChatGPT](https://blogger.googleusercontent.com/img/b/R29vZ2xl/AVvXsEi29VY0rNNP1Hqyz8RgB47r4q7fhqQfYHxsEbX_RKmT4KU5Y_vg7vB6dsAW91CidjMfJ38klTvD4mMtZb3QmwFPAHaGF88BPttWSbfYvo7UUg7MpFRTNtEjThmG0QaX2qZFFHSnN71i3YQZSBFC0SL7PgkAlme-XH8cSNWRFbOAOuLr_LMOGcImLQ/s72-c/Screenshot%202024-12-04%20at%2020.45.58.png)
Posts
Those who like to feel smug at the expense of a machine may enjoy the effect of repeated translations. For example, the EPO's announcement translated from English to German, then to French, and then back to English is:
ReplyDelete"To improve documentation of patent information in an important step to enter into several languages, the EPA announced today a new machine translation service, patent translation, called on the EPO website. The service uses Google Translate technology to provide translation to and from English into French, German, Spanish, Italian, Portuguese and Swedish, which gave about 90% of all patents in Europe. At the end of 2014, the service also the possibility of patents will lead to and from all 28 languages of the EPO member states, as well as Chinese, Japanese, Korean and Russian."
...which results in the usual machine translation style - something that you kind of know what they probably mean, albeit in a very difficult-to-read form.
I've always found online translation services to be very bittersweet - immensely valuable in bringing some kind of sense to something which was meaningless, but always with a nagging worry that something vital has been missed.
They've improved a lot, though. Perhaps once Google has sorted out translation, they could produce a browser that will analyse CAPTCHAs automatically...
I tried it out with EP1186189, translating to German. The result is ... Gerglish or Engman, at best:
ReplyDelete"Verfahren und Teilnehmerstation für die Zuweisung von Zugriffsrechten auf einen Telekommunikationskanal eines Telekommunikationsnetzes mindestens einen Teilnehmerstation (5, 10, 15, 20) des Telekommunikationsnetzes.
Gemäß dem Verfahren Informationssignale zu dem mindestens einen Teilnehmerstation (5, 10, 15, 20) und Zugriffsberechtigungsdaten (45, 50, 55) übertragen wird, um der mindestens einen Teilnehmerstation (5, 10, 15 übertragen , 20) über der Informationssignale.
Eine Auswerteeinheit (60) des mindestens einen Teilnehmerstation (5, 10, 15, 20) die Zugriffsberechtigungsdaten (45, 50, 55) und überprüft, ob die Zugriffsberechtigungsdaten (45, 50, 55) enthält ein Zugang Schwellenwert (S)." The words are basically ok but the grammar is outlandish.
I have tried this from German to English, from English to German and from English to French and it looks remarkably comprehensible. Those who make their living from translating patents will rue today.
ReplyDeleteI guess it is just coincidental that the translation service starts on the day that Google throws overboard any pretence of preserving the privacy of its users ...
ReplyDeleteHere's something I don't understand: if Google infringe privacy by using machine analysis why are Google not opening themselves up, (at some point in the unforeseen future) to triple damages if they are accused of infringing a patent that they have machine translated?
ReplyDeleteMT: one more useful tool for IP lawyers (including Gmail and Blogger) despite the seemingly collective hope that someone, preferably Apple, will give Google a deep dose of grief as just reward for seeking to reduce the overbearing effect of IP on our lives
Gentoo, I am not an expert in US patent law, but triple damages (a feature of US, not European, patent law) are reserved to cases of "willful infringement". I believe that even the most wily US patent litigator would have trouble persuading a jury that having the computers in one of its server farms machine translate a patent on an outside request made Google aware of that patent, let alone infringed it. No actual will involved, therefore no willful infringement.
ReplyDeleteIf it was found, however, that Google was data mining those translation requests to pick out particularly interesting technologies and then go and infringe patents in those technologies, then it would be a different matter.
As for your comment that "IP lawyers" (presumably including all sorts of other IP professionals, like patent agents and European patent attorneys) have the "seemingly collective hope" that someone will give Google "a deep dose of grief", well...firstly you appear to ignore that Google does itself employ very significant numbers of "IP lawyers" (who presumably don't wish their employer or client any sort of grief). More importantly, though, you appear to confuse the general feeling of IP-aware, impartial third parties that, in all matters IP, Google keeps pushing the envelope in various risky manners, sometimes using its hefty weight against weaker opponents without much apparent remorse. Google also raises quite a few eyebrows among not just "IP lawyers" but also the antitrust authorities, when, after complaining of others' aggressive enforcement of UI patents, it supports the even more aggressive enforcement of standards-essential patents by Samsung and Motorola.
Mind you, on the other hand, eyebrows are also raised when a surprisingly well-informed former anti-software-patent "non-lobbyist" keeps hammering on Google in his blog...
Anyway, those are all rich, highly profitable corporations, not NGOs. If Google is offering this service, it's certainly because it stands to gain some profit from it.
It has to be borne in mind that these machine translations are made "on the fly" and are not downloaded from a translations database. My experience of a decade's use of the Japanese Patent Office's machine translation service is that translations made of the same document at different dates can differ as the translation engine is updated. Therefore I always make a note, usually by means of footer text, of the date the translation was downloaded.
ReplyDeleteTo make sure that we are not talking at cross purposes, it is my practice to send the examiner a copy of the version I have downloaded, or ask him to send me a copy of the version he has downloaded, to ensure that we are both considering identical texts.
Last month I received a communication where the examiner gave the URL of a machine translation of a Japanese patent that he had referred to in the communication. The URL occupied two lines, and when I did type it out and double check for errors, the server was perpetually busy over a period of 2 weeks. I ended up telephoning the examiner, who kindly emailed me a copy of his download.
The URL occupied two lines, and when I did type it out and double check for errors, the server was perpetually busy over a period of 2 weeks.
ReplyDeleteNo wonder. EPO examiners normally use the machine translation service of the Japanese Industrial Property Digital Library, which uses dynamic URLs. Hence the length of the URL and why it was completely useless afterwards...
The only way to retrieve that machine translation is to go to the IPDL's website linked above, click on "PAJ", then click on the righ-hand button "Number Search", select "Publication Number", enter the publication number, and then click the top button "Detail". Not the most user-friendly interface around (which is not surprising: it is at least a dozen years old...)
Thanks for the dynamic URL explanation. The examiner did confirm that the source was the same JIP source that I am familiar with, although I didn't ask how he got the entire document in one go.
ReplyDeleteI normally find it is necessary to download as three sections that have to be pasted into a WORD document in the right order. As both the translation and the Japanese language original use paragraph numbering, it is easy enough to compare the two to ensure that all sections have been translated.