Machines No Match for Human Translators
The Washington Post recently reported that the growth of the internet has facilitated a revolution in translation technologies. The attempt to base machine translation on language rules has been replaced by statistics. With this method, vast volumes of text previously translated by humans are collected in a database. The probability that a set of words in one language is comparable to a set of words in another language can then be statistically calculated by a computer program. Google's Translate project is a good example of machine translation by statistics. The software can instantaneously translate text among 41 languages in any subject domain. According to Google research director Peter Norvig, the result is a translation good enough to know who did what to whom. He adds, "It will be very rare that you think a native speaker did the translation." In fact, Google still hires professional translators to generate its high-value web pages. The American Translators Association's Kevin Hendzel says that human translation is critical in areas where there is little room for error, such as nuclear power, the pharmaceutical industry, and disarmament negotiations. Norvig notes, however, that only 1 percent of the internet is currently accessible to speakers of minority languages. Google Translate opens up the web. And that's the point. "It's a lot more information, and it's also different worldviews."
Return to Headlines
From "Tongue in Check"
Washington Post (DC) (05/24/09) Garreau, Joel
Source: ATA Newsbriefs - May 2009