Plans for automated translation of patents move ahead

Google is extending the reach of its automated translation technology in a formal deal with the European Patent Office (EPO) to set up a system for translating granted patents. The deal will enable EPO to provide free machine translation of patents into the languages of the 38 countries its serves, making it easier for patent examiners to look for prior art.

The two partners signed a long-term agreement to work together on machine translation of patents into multiple European, Slavonic and Asian languages today (24 March), following the announcement of the collaboration in November 2010. The deal is not exclusive and no money will change hands, but Google will get access to 1.5 million patents that EPO has translated manually.

EPO President Benoît Battistelli said the project will facilitate the development of the Single European patent. However, he told Science|Business, “It won’t solve the problem of the linguistics regime.” Where it will help is in easing and improving the access of anyone whose language is not one of the three official languages – French, German and English – in which EPO and the proposed Single Patent operates. Automated translation “breaks the language barrier”, Battistelli said, “so in that sense it will help the development of the unitary patent.”

Battistelli acknowledged that the translation system will not help in patent applications to EPO, which must still be in one of the three official languages. “But it will help someone who wants to apply, to find out, ‘is my invention really new’, he said.

EPO’s other motive is to improve access the rapidly increasing volumes of Japanese, Chinese and Korean patents, which it says is one of the biggest challenges facing the global patent system. “China is filing more and patents, and few people in Europe understand Chinese,” Battistelli said, “It’s a real problem because [people] don’t have access to prior art.”

Google’s automated system learns by looking at translated bodies of text and statistically determining what a given word or phrase means. Since patents are written in a very specific style and format, access to EPO’s manual translations will help improve Google’s machine translations. Battistelli said the number of patents in a particular language will not affect the quality of translation. Translations will be offered, “Only if we are convinced we have reached the necessary quality,” he said, adding, “If in a specific language we don’t have enough patents we will add technical documents that are not patents to reach the necessary quality.”

Using Google Translate technology, EPO plans to offer translation of patents on its website into 28 European languages, and Chinese, Japanese, Korean and Russian. As result of the deal EPO says that from this year it will be possible to search for patents on the EPO website in English, French and German and translate between them on the fly. The other languages will be made available in phases. The project is expected to be completed by the end of 2014.

Antoine Aubert, head of public policy at Google Brussels said the company will have a team of engineers dedicated to the project, and promised the quality of translations will improve as the number of documents it can reference increases. In addition, he said, “All the technology keeps improving.”

EPO has been working on machine translation since 2004, and Battistelli said that while it has made some headway, “It is clear it is not in EPO’s core business to develop a specific automated system.”