The United States dollar is weak, driving an increasing orientation towards global exports. The Korean manufacturer Samsung and America’s Apple are in a patent war over mobile phones. The average annual income in China has quadrupled. Terrorists have taken root in remote countries, from those in the Sahel region of Africa to Chechnya.
Open up the news on any given day, and these are the global issues you’ll read about. Every one of these challenges impacts the US government. And each issue requires language translation, either into English or from English into another language.
But there are petabytes of data on the Internet to sift through and translate. The government doesn’t have the time or budget to hire an army of human translators for every job. Machine translation, while faster and more cost-effective than humans, only generates moderate to poor results for the majority of languages. The key lies in bridging the gap between translation memory — the databases of terms that feed machine translation — and human language intelligence.
Assimilation and Dissemination
Government agencies apply translation in two primary ways: assimilation and dissemination. Assimilation is used for intelligence purposes, economic research and information analysis activities. It involves gathering and translating content from countries around the world, then gauging the importance of each piece of content. Analysts can then judge the context of the information and escalate if necessary.
For example, SITE Intelligence Group, a private intelligence agency and government contractor, sifts through hours of video, blog posts, forums and other online sources every day to track and translate terrorist activity. In order to unveil important findings in a timely manner, SITE must translate quickly and effectively, not missing anything that might indicate suspicious actions. One of their more recent finds was a fatwa death sentence issued by an Egyptian clergyman, targeted at those involved with the “Innocence of Muslims” movie.
Dissemination, the other way that government uses translation, involves translating English documents into other languages for targeted dissemination. For example, the World Health Department has a substance abuse treatment program that is aiming to translate its processes for managing substance abuse for use in non-English speaking cultures. The initiative aims to “achieve different language versions of the English …[with a] focus on cross-cultural and conceptual, rather than on linguistic/literal equivalence.” In other words, WHO recognizes that translation is not just a matter of 1:1 language translation, but also of localization and coordination with differing cultural norms.
Language as Fuel
While different in nature, translation for assimilation and dissemination have one major thing in common. They both require the accumulation of linguistic resources, through translation memory, to propel future translations. This memory can be applied to machine translation to continue to improve upon the translation which occurs. Yet realistically, machine translation has years to decades more memory to build before it can be accurate across languages.
In the meantime, the ideal scenario would be for machines to generate decent baseline translations in all languages, then have humans edit and localize content if needed. With limited translation memory, how do we get there?
Humans and Machines Working Together
Human translators, whether professional or casual, must continue to feed high-quality translations into machine software. As more humans enter translations into their software, translation memory captures better information, enabling machine translation to produce higher-quality results. Machine translation generates garbage if it only receives bad translations, so in a sense, the onus is on humans to make machine translation smarter.
When machine translation has a large corpus of quality translation memory, it can generate instant and accurate translations. The government’s assimilation jobs will have readily available translations across languages, streamlining the process of intelligence, economic analysis and other tasks. For dissemination purposes, the government can use machines for basic translations, and humans to add cultural context.
We’re still a far way off from that reality today, but with enough human input, we can change the way translation is done in government.