Talking translation technology, chez Xerox

By Richard Morochove

First published July 9, 1998

GRENOBLE, France -- It's not a chateau. It's merely a maison bourgeoisie, which is far less grand, or so I'm told. By whatever name, the historic landmark built in the 1820s is a wonderful setting for the headquarters of Xerox Research Centre Europe (XRCE).

Nestled in the foothills of the French Alps on the outskirts of Grenoble, I found it a marvel that any work gets done in this bucolic milieu. Surveying the green fields out back that stretch to the hills on the horizon, I felt like lord of the manor.

With four universities and 50,000 students, Grenoble attracts more than its fair share of idiosynchratic academics. The city is the closest thing to a Continental version of Silicon Valley with the European headquarters of Xerox rival Hewlett-Packard located nearby, along with ten national research facilities.

Almost 100 work at XRCE, mainly in Grenoble, with a couple of dozen at a satellite lab in Cambridge, England. The employees represent a wide range of nationalities and language backgrounds. XRCE Director Monica Beltrametti came to Xerox three years ago from the University of Alberta in Edmonton, where she was Director of Computing and Network Services. Beltrametti is responsible for the lab's research projects and for bringing the research results to market.

Opened in 1993, Xerox's European counterpart to the famous labs at Xerox PARC (Palo Alto Research Center) is responsible for some intriguing work in linguistic technology and human-computer interface design.

Much of the linguistic technology focuses on computer-aided translation of foreign languages.

The translation isn't perfect, but aims to accurately translate up to 80 per cent of a document written in any one of 15 different languages. It even handles the translation of idioms (e.g., take the bull by the horns) and proverbs (e.g., birds of a feather flock together.) Then an experienced human translator can take over, using the software to convert the remaining text.

The translation software can automatically guess the language of a document by examining just a sentence or two. I had to work pretty hard to fool it, by typing in a string of foreign words incorporated into English.

Xerox sees two main markets for the translation technology: Web pages and documents of multinational businesses. On the Web side, Xerox has licensed the technology to companies such as America Online, Infoseek, Oracle and Verity, which will incorporate it in search engines. On the corporate side, it will be called Xerox Translation & Authoring Systems (XTRAS) when it's launched in September.

It's easy for us to overlook the fact that most Web pages are written in English. The lack of native language content is one reason why Internet use in Europe isn't as high as in North America. Surveys of Internet use in Canada usually show lower market penetration in the province of Quebec. So automatic translation of Web pages could be a boon to Internet use and be particularly attractive to businesses that want to sell online to a worldwide market.

While I see a bright future for Web translation, Xerox faces established competition from the Systran software, also developed in France, used by the Babelfish translation service of Compaq's Altavista search engine. Babelfish translates Web pages one at a time. It's free, but relatively slow and handles fewer languages than Xerox's system.

Xerox is a longtime leader in human-computer interface design. Xerox PARC invented the windowing computer interface that was later adopted by Apple for its Macintosh and by Microsoft in Windows. LightWorks, the result of a research project at XRCE's Cambridge lab, aims to break down the barrier between the cyber and real worlds with a new desktop interface.

When you work on the LightWorks digital desk, you really work on a desk. A video camera scans paper pages on the scanning side of the desk and sends them to a computer. Image processing software enhances the quality of the video and projects the image back onto the other side of the digital desk. The software can also feed into Xerox's Textbridge OCR (optical character recognition) software to translate the image into computer text.

By placing paper documents in the scanning zone, you can convert them to the computer documents. LightWorks allows you to work effectively with paper and computer documents concurrently, so you don't need to switch between the real and cyber worlds.

LightWorks hasn't been incorporated into any consumer-level products yet. A next generation version that eliminates the computer mouse and lets you manipulate the data images by gesturing with your hands and touching the surface of the desk is under development. This interesting concept could be the next big step forward in computer interfaces. CW