AppTek Provides Ford With Machine Translation

Translation software is at last good enough to help companies do business in other languages

Thanks to the Internet, companies can leap over most geographical barriers to conduct business globally. But language barriers remain a tough hurdle.

Increasingly, though, translation software is making it easier to do business in other languages. While computer translation isn’t perfect — human input is still needed to ensure complete accuracy — the latest programs are faster and more accurate than earlier generations of translation tools.

That’s helping businesses decipher foreign-language Web pages, news reports and documents, translate company communications for foreign workers and reach customers around the world in their own languages.

Saving a Lot

Ford Motor Co. uses translation software from Systran SA of France, along with some human input, to convert vehicle-assembly instructions written in English into four languages: Spanish, German, Portuguese and Dutch. It also uses software from Applications Technology Inc., of McLean, Va., for translations from English to Turkish. Nestor Rychtyckyj, a Ford technical specialist in artificial intelligence, says that while machine translation still isn’t 100% accurate, it has improved over the years and is good enough to convey the substance of instructions to foreign workers.

“Machine translation just makes the process more efficient” than it would be using human translators alone, Mr. Rychtyckyj says. “We’re saving a lot of time and effort.”

Ford, based in Dearborn, Mich., writes new assembly instructions not only when it comes out with a new model, but also when it changes the manufacturing process for existing models. Sometimes, for example, the company decides to use different parts or tools to build its cars. Instructions for each vehicle may change three to four times during each model year.

The new instructions are usually written in English and then fed into Ford’s translation systems. The software includes a customized glossary of about 5,000 terms compiled by Ford, with each term preprogrammed to be translated into one of the five languages. If the new instructions contain words or phrases that don’t match the Ford glossary, the software runs them through a more general, built-in dictionary. In addition, Ford uses human translators periodically to monitor the translations for accuracy.

Workers in foreign plants can see the results on Ford’s computer system the next morning and print them out. Mr. Rychtyckyj says the results are mostly accurate, up to 94% in German, for example. Inaccurately translated instructions haven’t led to any significant mistakes in how a car is built, he says, because people usually spot the translation errors before they become a problem.

At plants where workers speak languages not covered by the current system, there are engineers and other employees who speak English and can provide oral or written translations of new instructions, Mr. Rychtyckyj says. But Ford plans to expand its use of machine translation beyond the current five languages, with Russian next on the menu, he says.

Keeping the Human Touch

Translation software also offers greater efficiency when it comes to translating material that customers will see, like Web sites or product packaging. But when dealing with the public, where mistakes can embarrass a company and discourage customers, the human element becomes more critical, industry experts say.

“No one thinks you can depend solely upon a machine,” says Chris Boorman, chief marketing officer for SDL PLC, a British maker of translation software. “I do not believe you can ever take the human out of language when it comes to publishable” content, he says.

Dimitris Sabatakakis, chief executive of Systran, concurs. “The first purpose is to provide users with immediate access to translation,” he says of translation software. “It’s not a perfect translation. You can’t publish it as is, but it’s useful.” Systran sells desktop software packages for individual and corporate users, ranging from $49 to $999 per user. It also offers a subscription online service and licenses its technology to Web sites including Google Inc.’s popular search engine.

IT’S NOT PERFECT

Computer translation has come a long way, but even translation-software vendors say humans are still needed to polish anything meant for public consumption. Here’s an excerpt from one translation performed by Google’s automatic translation service:

From the Google translation of a press release on the Web site of French company Dassault Aviation: In a context of exacerbated competition, Dassault Aviation differs from its competitors by taking the head of a revolution which will extend, in the long term, on the whole of world industry. The numerical design, which governs the production of Falcon 7X, is based on the modeling of all the parts of a plane in 3D.

From Dassault’s English Web site: In an extremely competitive environment, Dassault Aviation has set itself apart from its rivals by pioneering a revolution that will eventually spread throughout the worldwide industry. Digital design, which controls production of the Falcon 7X, is based on 3D modelling of all aircraft parts.

Google has used Systran’s technology to provide its users with translations of Web sites for several years. If search results include a foreign-language Web site, the site’s listing will have a “Translate this” tab next to it. Clicking on the tab will, for instance, display a French Web site in English.

It’s not perfect, but at least it’s comprehensible. Google’s translation of the main Web site of French aircraft maker Dassault Aviation SA, for example, displays the title of a press release as “Dassault Aviation entered of full foot a new industrial era.” Clicking on the “English” tab built into Dassault’s own Web site produces a more fluent title: “Dassault Aviation steps into a new industrial area.”

Google also has been developing its own translation software, which it uses to translate Web sites written in Chinese and Arabic. Google’s technology is different from other translation software. Google feeds massive volumes of existing translations of text into a program, which uses that material to perform new translations by determining the statistical probability that a word or phrase in one language is equivalent to that of the other, says Peter Norvig, Google’s director of research. The source text can include matching articles from news sites written in both Chinese and English, or European Union documents that are translated into the languages of the group’s member countries, he says. Other translation technologies rely on preprogrammed dictionaries and grammatical rules to perform translations.

Google’s translation tools aren’t designed to help businesses publish information, Mr. Norvig says. “It might be good for a first draft, but you’d want someone to look it over.”

Working From Memory

SDL and some other makers of translation software supplement their offerings with a technology called translation memory, which stores translated words, phrases and sentences so that they don’t have to be translated again every time they recur. That speeds up the translation process.

For instance, a manufacturer of television sets can store the translations of the manuals for every set it makes in the SDL program, along with the original documents. Each time the company needs to translate a manual for a new television set, the software will automatically reproduce the translations of any wording that matches past manuals, as well as translating any new material. A human translator can then fix any mistakes or make any desired modifications in the automated translations, rather than being employed to do the translation from scratch.

Lionbridge Technologies Inc., of Waltham, Mass., offers a service that incorporates all three elements of translation — machine translation, translation memory and human translation. Customers submit content through the Internet to Lionbridge and pay by the word — anywhere from a few pennies to 40 cents a word, depending on the language and format of the content, says Rory Cowan, Lionbridge’s chief executive.

A big Lionbridge customer is Microsoft Corp. The software giant used Lionbridge’s technology and human translators to help translate about 15 million words of documentation connected to Microsoft’s 2005 edition of Visual Studio, a suite of tools used by software developers. The documentation includes an electronic user manual and other information posted on a Microsoft Web site or included in CDs.

About 60% of the material was handled by translation-memory software because it had appeared in earlier translated versions of Visual Studio documentation, says Fabrice Fonck, a general manager for developer content at Microsoft. His group is responsible for translating content from English into eight languages, including French, Spanish and German. The rest of the material was run through machine translation by Lionbridge and then verified by Lionbridge’s human translators, Mr. Fonck says.

This was the first time Mr. Fonck’s group added machine translation to the process — previous translations had been performed with a combination of translation memory and human translators. The savings on the multimillion-dollar project, from relying less on human translators, were in the 6% to 8% range, he says. “We wanted to try this new process to drive productivity, drive down costs, and in the end come out with a quality product,” Mr. Fonck says.