The Hidden Costs of Free Online Translation
It is quite difficult to resist the draw of “free” or “instant” services. Any company searching for translation services would be attracted to these offers from a variety of online providers, but these online tools often sacrifice quality. Here are two examples illustrating issues these tools can create.
In July 2013, NW News Network, a collaboration of public radio stations in Washington, Oregon and Idaho, reported on the translations of the Washington State and Oregon websites created to help families navigate the Affordable Care Act. The Oregon website, Cover Oregon, intended to make it easier for limited English speakers to access the information displayed by adding localization features. As a temporary solution, the organization relied on Google Translate to do the work. Google Translate is a statistical machine translation tool using large libraries of existing translations to detect patterns and derive translation algorithms. When Austin Jenkins, correspondent at NW News Network, contacted CETRA’s Director of Translation Services, Angèle Surault, she reviewed the French translations on the site and concluded that the translation was unprofessional and of low-quality. For example, the word “cover” in “Cover Oregon” was always translated, either as a noun (bed cover) or as a verb (to cover), rendering the text incomprehensible.
Translation mistakes created by free online machine translation tools can also have far-reaching consequences. Last year, the Government Accountability Institute (GAI) reported that foreign donations had flowed to President Obama’s re-election campaign. A further review found that the campaign’s sole contribution from a foreign-national donor was, in fact, based on a translation error. The GAI reported that a Norwegian blogger initially posted a link to the Obama campaign’s donation page which sparked a reader to comment that it would be illegal for a non-U.S. citizen to contribute to Obama’s campaign. The blogger responded in Norwegian, “Hadde jeg i praksis kunne gitt penger til Obama hadde jeg gjort det.” which the GAI, with the help of Google Translate, relayed as “I have in practice given money to Obama, I had done it.” However, ThinkProgress confirmed based on the feedback of three Norwegian speakers, including a professor of Norwegian language from the University of North Dakota, that the quote actually meant quite the opposite. It should have been translated as “If I actually could have given money to Obama, I would have done it.”
In conclusion, these tools are extremely useful on a consultative basis, but unless the machine is trained for specific content and the translation is post-edited by a human translator, their output should not be relied upon to publish information or reach conclusions.
