Machine Translation for Patents?

The Euromachine-translation-225x300pean Patent Office’s (EPO) project with Google Translate – Patent Translate – is now complete one year early. The online tool covers translations between English and 31 other languages, namely Albanian, Bulgarian, Chinese, Croatian, Czech, Danish, Dutch, Estonian, Finnish, French, German, Greek, Hungarian, Icelandic, Italian, Japanese, Korean, Latvian, Lithuanian, Macedonian, Norwegian, Polish, Portuguese, Romanian, Russian, Serbian, Slovak, Slovenian, Spanish, Swedish and Turkish. Translation from and into French and German is also available for 27 of these languages.

Has machine translation (MT) come of age? Can multinational companies finally trim translation from their over-stretched patent budgets? Well, not quite. By EPO’s own admission, “the engine cannot provide legally binding translations.” However, it certainly can make the discovery process for patent applications quicker and more simple, which is what it is designed to do. Patent Translate can give you the gist of any patent or patent-related document and help determine if it is relevant. Then you can determine better where you need to invest in human translation.

Machine translation is a common theme. We’re tempted to believe that machine translation will be so good in a few years time that there will be no need for human translators. Indeed, rapid progress is being made as machines learn from texts they are given. Every year, millions of documents are added to EPO’s databases. These new documents will also be fed into the system, continually improving their Patent Translate engine.

At MultiLing, we love translation technology. Computer-aided translation makes our job faster and more accurate with tools such as translation memory and terminology management. It also saves money for our customers. We’ve found that it is not an uncommon scenario at large global corporations for different departments to request translation for the exact same document. Document management on our end and using translation memory as a first pass easily identifies these duplicate requests and can literally save the client $100,000+ per year.

We see a distinct difference between computer-aided translation and machine translation, and here are two main reasons why we don’t worry about machine translation taking over our work:

Machine translation uses the past as its guide.

Every year we hear the news of what the new words are. For example, Oxford’s 2013 word of the year is “selfie” – a term that increased 17,000 percent in use over the past two years. So even if you compile a massive database of all the language used everywhere today, and update it in real-time, it can only tell you how people have used language, not how they are using it today, and definitely not how it is being used tomorrow. In the patent world we are always working with new inventions, many of which may need to use words in new ways, and perhaps even new words – language you won’t find in yesterdays’ documents.

Language is bound to culture and context.

Since language is bound so tightly to culture, “literal” translations are often incredibly misleading. In one language the culture may dictate using a verbose greeting. If it is translated literally it will sound ridiculous in another language or culture. It is possible for an advanced statistical machine translation to look for an equivalent, rather than literal translation. But then there is the problem that an “equivalent” word will likely be different for different contexts, so the translation decision is linked to context and culture more so than to language. The complexity of understanding culture and context limits machine translation.

So where might the future lead? Well, just as human translators are becoming more specialized in their language, culture and field of expertise, so will MT engines. Using human translators to edit after machine translation is a distinct possibility in certain fields. For example, patents do use consistent document types, style and language. However, as long as attorneys are paid to prosecute patents, we are pretty sure that human translators will be paid too. The art of language is what law and translation use every day.