Preparing Your Content for Machine Translation

Page 1 of 3

Article ImageThe history of machine translation reaches far back into the past-long before preteens in the mid-1990s were using Babelfish in the middle school library to translate bad words into other languages. Some trace the history of machine translation all the way back to the French philosopher René Descartes, who, in the 17th century, proposed the creation of a universal language in which common symbols would be used to represent equivalent ideas in different languages.

It was in the 1950s, though, that computer scientists began to teach grammatical rules to computers in an effort to create artificial translation machines. In 1951, a team from IBM and Georgetown University demonstrated its machine-translation achievements in a demonstration wherein an assistant typed pithy Russian phrases into IBM cards and the so-called brain returned accurate responses such as, "We transmit thoughts by means of speech." As it turns out, those early machine-learning pioneers who were attempting to teach grammar rules to computers had their process a little bit backward. But we'll get back to that in a moment.

Any organization with an interest in disseminating its content to a global audience needs to familiarize itself with the basics of machine translation. At a time when the globalization of commerce is yesterday's news, the number of companies with the capacity for international ecommerce has grown considerably. Machine translation is not necessarily reserved for the titans of industry, as many translation service providers are beginning to offer more scalable services to assist even small companies with translation projects.

Once you've decided that your business model needs to incorporate an element of translation, you'll begin to start thinking about what'll you need to do to prepare your content for the process of translation. What follows is an exploration of some tips and techniques for getting your content ready.


Kevin Nelson is the SVP of strategy and technology at MultiLing, a company that has been providing machine-translation services out of Provo, Utah, since 1988. The first tip Nelson would give to anyone getting ready for a machine-translation project is to collect as many examples of text that you've previously translated as you possibly can. (And here's where that part about the history of machine translation being a little bit backward comes in.)

"If you look at the ways humans learn, we learn from memorization first," Nelson says. The machine-translation industry has been moving away from exclusively teaching rules and structures of grammar to the translation engine and moving toward starting by just feeding as much previously completed translation as possible to the server and allowing it to learn from that.

"In the last number of decades, we have started storing previously done translations and feeding them into the servers as a way of teaching them," Nelson says. "Before, translations weren't being stored in a database for future access. Doing everything in a digital way rather than on paper had to come first. But now that we're virtually unlimited in terms of storage space and archive capabilities, this is the direction that we're headed."

The specificity of the translation server you use-and the specificity of the training you implement with that server-is also immensely important to the quality of the translation output, Nelson says. Training your translation server to recognize the terms and constructions of writing that are unique to your industry is vastly preferable to using a machine-translation server that is more broadly trained for general translation.

"If you keep the realm narrow, and you keep content you're creating within that same realm, you'll have a lot better chance of getting good content," Nelson says. "This is part of the big difference between what you get when you use Google Translate, as opposed to using a specific server that is trained for your material. Google is really working on being the translator for the world, but it's still a ways down the road for it to get there. To hand your content to a translator as vast as that, the odds are that it hasn't seen enough content that's related to your industry yet, and the output won't be quite up to business standard." By feeding as much of your archive of previously translated material into your machine-translation server as possible, the server will learn to tailor its translations to the particularities of your translation needs-not just of the language into which you're translating, but the terminology of your industry, the style of your writers, and the voice your company.

Page 1 of 3

Related Articles

For much of history, the primary obstacle preventing people from connecting with one another was distance. But in the mobile, social, always-on world in which we live today, distance is no longer the major challenge. Instead, our biggest challenge is the lack of a common language. It's the inability to understand one another that prevents us from making meaningful connections. Isn't there an app for that? Yes, there is. Machine translation (MT) hails from the discipline of computational linguistics and aims to help humans who speak different languages communicate with one another. And while MT is not actually an "app," many app makers are attempting to harness the power of automated translation by building MT into their products and services. But there are challenges.