Man v. Machine
One of the major decisions a company faces when managing multilingual content is if they wish to use a software translation, human translation, or both. There is no right answer, according to industry experts; it all depends on your company's needs and how accurate the translation has to be. Even companies that make a living selling translation software admit there is no perfect software solution, but for certain types of content, machine-based translation can work quite well.
"Machine translation does have its limitations, the first being it does not provide accurate out-of-the-box results," says Reba Rosenbluth, director of corporate sales at Systran, a translation software company that works with the U.S. government, Google, CMS vendors like EMC/Documentum, and others. In her approach, you deal with certain types of content by building a glossary, or dictionary, of terminology, and the software checks the content against this glossary. A software solution, she says, is well-suited to dealing with this type of scenario.
There are two types of software translations, explains Rosenbluth—gisting and publishing. With gisting, you get the basic idea of what the content means. She says, "This could be used for scanning the web, for patent research, or even getting a rough first-draft translation." With publishing, it has to be close-to-perfect translation. She offers an example of each type. A web self-service portal does not necessarily have to be perfect, only good enough for the individual to solve the problem alone without a call to technical support, she explains.
A case that needs to be near-perfect involves instructions on how to build a car. The Ford Motor company has Systran software integrated into its ERP system to translate vehicle build instructions. Although Ford has manufacturing facilities worldwide, the engineers are in Dearborn, Michigan, and write in English. When build instructions reach the factory floor, they need to be translated into the appropriate language for the given country.
Lawlor points out that in many instances, a human translator needs to see the translation after it goes through a machine translation, but even for a customer-facing website, it is possible that there will be cases in which the content is fairly standard and translation can be automated. "At the end of the day," he says, "no matter how good the automated translation is, you will never get perfect translation, but you can customize the automatic translation for the needs of a particular business." For example, he explains, the Best Western hotel chain needed to automate a system that understood how to translate terms like king-size bed into multiple languages. Since hotel terminology remains fairly standard, they built an automated workflow so that any time content on the site changes, it goes through machine translation and is published live automatically, without human intervention until after it has been published. This won't work for everyone, Lawlor explains, but for companies with a standard approach to content, it can be very effective.
In many instances, the translation is not so clear-cut, however, and in these instances, you need human translators who not only speak the language, but also understand the language subtleties—idioms, word plays, double entendres—cultural sensibilities, and in some cases understand a particular industry, DePalma says. "Depending on the type of activity, there is a level of translation and information adaptation that is going to be helpful in a given market." As such, you need to understand more than linguistic issues; you also need someone who understands local customs, norms, and cultural attitudes.
This involves working with people on the ground in the country where you are translating. "We call it localization, not translation," Bolen says. "Translators need to be embroiled in local culture where they are. We are not going to have a person living in the U.S. for 20 years translating. We are going to have someone in that country, so they know the latest cultural trends," which allows the translator to strip language that might have been written with an American audience in mind.