The artificial intelligence-based machine translation tool for patent documents developed by the World Intellectual Property Organization, WIPO Translate, has extended its coverage to 10 languages, WIPO announced on Tuesday.

This marks "an important expansion of the highest-quality service yet available for accessing information on new technologies", WIPO said.

The online tool is now able to translate patent documents in one of the official languages of the Patent Cooperation Treaty - Arabic, Chinese, German, Spanish, Portuguese, Russian, French, Korean and Japanese - into English and vice-versa.

WIPO Translate uses cutting-edge neural machine translation technology to render highly technical patent documents into a second language in a style and syntax that closely mirrors common usage, according to WIPO.

?

"Extending WIPO Translate to more language pairs is a welcome development for innovators around the world," said WIPO Director-General Francis Gurry. "It ensures accessibility to the state-of-the-art knowledge created in the main languages of the production of technology.

"The speed and accuracy of translations through WIPO Translate is unique because this tool is trained by and focused solely on patent documents, instead of a more-disparate array of texts, thereby producing higher-quality translations," he added.

WIPO Translate powers Patentscope, a database with 65 million records of global patents. It is used by inventors to search for information before filing their own international patent applications through the PCT.

The launch of WIPO Translate means that "a vast and ever-increasing trove of patent documents will soon be more easily accessible to innovators who search these records for inspiration or technical know-how", Gurry said.

A WIPO news release said the tool is trained exclusively with huge amounts of patent texts and includes a "domain-aware-technique", which translates according to the specifics of the invention. The tool internally integrates 32 technical domains taken from the International Patent Classification to eliminate ambiguity in the translation process.

The specificity of neural machine translation compared to previous phrase-based statistical methods is that it produces more natural word orders, with particular improvements seen in so-called distant language pairs, such as Japanese-English and Chinese-English.

WIPO initially trained the tool to translate Chinese, Japanese and Korean patent documents into English. Patent applications in those languages accounted for about 55 percent of worldwide filings in 2014.

The first version of the translation system, launched in October last year, was available only for English and Chinese.

The high accuracy of the Chinese-English translation is the result of the training of the neural machine translation tool. It had compared 60 million sentences from Chinese patent documents, provided by China's State Intellectual Property Office to WIPO's database, with their translations as filed at the United States Patent and Trademark Office.

zhangzhao@chinadaily.com.cn

(China Daily 09/28/2017 page17)