In 2008, a Chinese restaurant displayed a bilingual sign above its storefront to accommodate English-speaking patrons. To save a few pennies in creating the sign, someone working for the restaurant skipped use of a professional human translation. Instead, the restaurateur typed the restaurant name into a machine translation program, which quickly returned some English text that was printed and placed above the business storefront without review. Unfortunately, the translation utility was not working that day. Because of the misfortune, Internet users worldwide have seen photos of the restaurant sign displaying the supposed English translation, Translate server error.

Public interest in machine translation is growing as the nature of business becomes increasingly international, translation technology continues to improve and Google makes news each time it adds machine translation features to another of its online applications. With increased interest comes increased optimism. However, it is important to question whether this optimism is based on reality or hype.

Consumers' optimism about machine translation often cools when they notice glaring errors in machine-translated communications. Some of these anecdotes are rather humorous, like the incident involving the Chinese restaurant sign.

Earlier this year, the online periodical Kazakhstan Today was unable to convey its intended meaning through a machine-translated English article without producing a few chuckles. The article discussed how Kazakhstan's former president felt the country was not passing gas the way it should. While intending to report national news on energy, this short, two-paragraph article made four separate references to the important issue of "passing gas." Months later, the periodical altered the translation and removed the offending phrase.

In another instance, Israeli journalists nearly set off an international incident when they sent a list of machine-translated questions to a Dutch diplomat. The mistranslated message started with the greeting, "Helloh (sic) bud," and went downhill from there. At one point it appeared to insult the diplomat's mother.

Is machine translation worthless? No. Automatic translation is neither completely worthless nor perfect. However, in very specific business situations, machine translation can be quite useful and beneficial.

For example, when users merely need to understand the "gist" of a document, machine translation can provide low-quality translations much faster and more cost-effectively than a human ever could. This speed can make the technology appropriate for getting the gist of non-sensitive instant messaging. Machine translation "gisting" also succeeds when organizations need to quickly scour thousands or millions of documents for certain keywords and subjects.

The U.S. Government has used computers to quickly translate millions of pages of potential terrorist chatter, which are then searched in English to identify key documents that then require high-quality human translation for added certainty. Legal service firms do the same to identify potential evidence from thousands of documents in a foreign language, thus saving months of time and millions of dollars.

Automatic translation also produces high-quality results in situations where the source language is highly controlled and all ambiguity is removed. For example, weather reports and garment care labels have a limited and controlled terminology, and an automatic translation program can be trained to translate those phrases perfectly every time.

Such programs can be trained with human translations to handle other terminology, too. However, quality training is important, otherwise, the result is "garbage in, garbage out." For instance, a Chinese furniture manufacturer trusted a machine translation program to translate its furniture labels from Chinese to English. Unfortunately, the computer program contained a typo that added a racist slur to the label, thus seriously offending unsuspecting purchasers of a leather sofa.

Human translators also use various forms of technology to produce high-quality translations. Some machine translation technologies produce translations that can be edited cost-effectively by humans. Other technologies regularly help translators to quickly recall specific terminology or previously completed translations, resulting in what is called computer-assisted translation. Full automation will always be the most cost effective, but professional human translators are necessary to ensure the highest quality in subject areas that lack highly controlled language.

4 comments on this story

Ultimately, every quality translation originates from a human, whether that original translation is then used to train a machine or simply repurposed by the machine. Likewise, poor quality machine translations also originate from humans at times, because a human has used machine translation in an inappropriate business situation. When machine translation succeeds, it is because humans have used it wisely.

Adam Wooten is director of translation services at Lingotek. He also teaches a course on translation technology at BYU. E-mail: . Follow him on Twitter at AdamWooten..