A Play on Words


ALTHOUGH IT MAY BE THE MOTHER TONGUE for relatively few people on the planet, English has come to dominate the Internet. But that was then. This is now. Overseas World Wide Web users -- accent on the first W, s'il vous plaît -- are proliferating rapidly enough that Computer Economics of Carlsbad, Calif., predicts that by 2005, 57 percent of Internet users will be non-English speakers.

Real-time, on-demand translation of Web documents into other languages has been a fantasy of anyone buying or selling on the Web. Successful translation services will trigger a flood of additional overseas revenues for one-language sites. And a privately held company that sells everything from magazine subscriptions to fishing lures to hunting rifles to deer scent is leading the way.

EBSCO Publishing of Ipswich, Mass., a division of Birmingham, Ala.-based EBSCO Industries, last month unveiled a translate-on-demand feature for articles from 6,300 full-text journals. Think about all the non-English speakers who want an original English document translated into their native tongue.

Inside a Translation Machine

The Challenge: To provide terabytes of technical journal articles to a multilingual customer base without changing its back-end infrastructure or altering other aspects of its customer-facing technologies.

The Solution: Transparent Language's server software package linked to EBSCO's existing infrastructure, based on Microsoft's Windows NT.

The Pioneer: EBSCO Industries of Birmingham, Ala., is a privately held conglomerate that employs 4,000 people in 21 countries. Its EBSCO Publishing unit is located in Ipswich, Mass., and has 350 employees.

Locations: Canada; Taipei, Taiwan; Great Britain; Germany; Australia; the Netherlands and Japan.

The new automatic translation feature will enable Web users to click a "translate" button and have a specified article translated into Spanish, French, German, Italian or Portuguese from English; EBSCO eventually plans to add Chinese, Russian and Japanese to its repertoire. The on-the-fly translation capability is based on software from Transparent Language of Merrimack, N.H., running on a Windows NT server.

"EBSCO really is alone with this implementation," says Steven McClure, a research vice president of speech and natural language research at Framingham, Mass.-based IDC, a sister company of CIO's publisher, CXO Media. "[The technology] should significantly expand [EBSCO's] revenue in the existing translation markets and the new ones that will be created by the globalization of the Web population. Other than a few small European ones, there are no other major products supporting cross-language information retrieval."

Colorful History

Launching real-time document translation services was not an impulsive activity for EBSCO. While the privately held company has a colorful and eccentric history -- beginning in the 1930s, when "" Elton B. Stephens began selling magazine subscriptions door-to-door in the South to pay law school tuition -- its 4,000 employees working for 80 subsidiaries in 21 countries are anything but haphazard in their technology investment choices, according to EBSCO Publishing CIO Mike Gorrell.

"We don't do much on a wing and a prayer around here," Gorrell says. "We developed a solid business case that says the non-Engllish language initiatives, which translation on demand is a part of, will pay for themselves in two years and will make us tens of millions [of dollars] in five years." EBSCO officials predict that the company's overseas revenues from providing access to documents will rise to 40 percent of revenues from 10 percent, due to the increased number of articles to be accessible by non-English-speaking site visitors. The company's licensing cost for the translation software will run around $100,000 a year, according to a Transparent Language official.

In addition to the cost-benefit analysis, EBSCO's evaluation and implementation of translation software began with a competitive bake-off: It rated the ability of six software packages to translate English into Latin American Spanish, Brazilian Portuguese, French, German, Japanese and Chinese. To successfully test each language translation module, the company ran a series of full-text documents from a range of subject areas and 21 diverse sentences through the translation systems, and then 21 experienced translators evaluated the results. Technological compatibility and scalability were among the other criteria assessed in the EBSCO evaluation.

Interestingly, the EBSCO evaluators were not looking for the perfect translation score. As with horseshoes, close enough is good enough when it comes to these translations. EBSCO's goal was a "good" speed and an accuracy performance rating of roughly 60 percent to 65 percent, although a small number of packages achieved 80 percent in a few of the tests.

"It's very difficult to predict the accuracy of automatic machine-translated [AMT] documents," says Christoph Mosing, director of professional services for eTranslate, a San Francisco-based Web globalization solutions provider. "It all depends on the strength and the sophistication of the machine translation engine. "Depending on the needs of the target audience, there's a decision to be made between AMT and human translation. AMT can be done in real-time, but you lose accuracy. Humans can't translate in real-time, but they're much more accurate," Mosing says.

"If you look at the debates about AMT versus human, everyone prefers humans," says Judith Biewener, program manager of EBSCO's foreign language initiative. "But the scale of our data and translating it into five languages obviously prevents it. We know machine translation is less than perfect. But we have to accept it."

Not Perfect

According to Charles McGonagle, vice president of marketing for Transparent Language, foreign users need only the gist of an article, not a perfect 10. "It absolutely is not about getting every word translated correctly," McGonagle says. "It's about making the translation-on-demand product usable and accessible."

Transparent Language's ease of use and Spanish capabilities led to its selection, according to Pete Marsh, EBSCO's former director of Web development and design. "Spanish is our most used non-English language, and our market research indicated that we could expect the most access to our new package from people who wanted to translate English into Spanish."

EBSCO will monitor its website traffic and Web revenue stream to determine whether the translation-on-demand software is as big a revenue spinner as expected.

"We already know which users use which of our foreign language interfaces," notes Gorrell, explaining that EBSCO has tracked the use of its Spanish articles by Spanish site visitors, French articles by French speakers and so on. "So it will be very easy to determine which ones are used more frequently after we add the new product. We also can track which [areas'] sales volume increases."

EBSCO won't charge extra for the translation service. Gorrell notes the company's model is to sell content, so it will make more money by selling more content, thus improving service. "We don't get points for having the latest and the greatest," he says. "We get points for supporting customers."

This story, "A Play on Words" was originally published by CIO.

Free Course: JavaScript: The Good Parts
View Comments
You Might Like
Join the discussion
Be the first to comment on this article. Our Commenting Policies