In endowing a machine with huge amounts of knowledge we often hit a complexity ceiling Xerox has set itself the task of getting machines to understand us and ultimately to think for themselves

Teaching machines to be intelligent, to think with their own consciousness, has been a dream of scientists at least since Englishman Mr Charles Babbage envisioned his "Analytical Engine" in 1834.

Science fiction films and stories of the last century created a more general belief - and, perhaps, longing - that such a machine might not be too far off in the future.

But the optimism of the 1950s, when primitive Cold War computers translating Russian documents suggested that talkative robots were just around the corner, eventually gave way to the realisation that language itself was a massive barrier.

It is extremely hard to teach machines to do what humans do easily - understand and speak language. That's because language is subtle, re-arrangeable, and often takes its meaning from context. We're immersed in language, and we think, speak and understand on the fly. By contrast, machines need rules and definitions and dislike ambiguity.

But machines need to understand language because language is the way in which ideas, emotions and information are conveyed. That's made language one of the most important frontiers for work in the area of artificial intelligence.

At Xerox's Research Centre Europe, in Grenoble, France, a dedicated team is tackling the "natural language problem" - getting machines to understand us so that we can interact more easily with them, have them do intelligent searches for information, and use them for accurate multilingual translation. And ultimately, to think for themselves.

According to researcher Dr Pierre Isabelle, an amiable French-Canadian, "the natural language problem proved to be much harder than people thought in the 1950s". Computers have only recently begun to have the vast, cheap computing power needed to perform even the basics of natural language processing, he says. "We need to endow a machine with huge amounts of knowledge and in doing so, we often hit a complexity ceiling," he says.

Still, place some limits on the task - for example, ask a machine to work with a limited, defined vocabulary as in medicine, or handle telephone reservations for an airline - and systems perform very well, he says.

So Xerox researchers demonstrate a copier that, using software over a computer network, can "look" at even a ripped fragment of printed paper and then go out and find the complete document on the internet. It can perform searches in multiple languages. And, then, it will provide a translation of key vocabulary words and phrases, adjustable to your knowledge of the original language.

Xerox Research Centre Europe also has a variety of hierarchical search and information extraction engines that can be used for different tasks. Unlike existing search engines, a hierarchical engine understands to some extent how information is related to other bits of information. It has some knowledge of context and relevance, and so can return useful information with a higher level of success.

Elements of these two are combined in Xerox's virtual reality environment for learning a language. Students acquire a computer-generated three-dimensional avatar (animated character) and move through rooms in a building, having conversations with both synthetic characters, or other people's avatars. It's an addictive way to learn - before long, you're exploring rooms and cracking jokes with the other participants.

But why is Xerox - the self described "document company" - working in the area of artificial intelligence and natural language understanding? "Our view of the document today is very different from our view of the document in the past," says Ms Monica Beltrametti, Xerox vice-president and director of the research centre. The company wants to "redefine the document", which increasingly is in digital form.

Doing so at the Grenoble facility seems appropriate, as it combines the past and the future to spectacular effect. The centre nestles below the Alps in a chateau once owned by a pasta baron, but also includes a neighbouring, futuristic glass cube of a building. Grenoble itself is fast remaking itself into one of Europe's Silicon Valley-esque locations, with a concentration of technology research and industry.

"Research is the lifeblood of Xerox. If we have no development, we have no future," says Mr Jean-Noel Machon, president of Xerox Europe. He notes the "huge heritage" of the company, which almost single-handedly created all the most familiar elements of modern computing - from the graphical user interface to the laser printer to ethernet networking to the mouse, at its famed Palo Alto Research Centre - then failed to commercialise any of it.

The company has more than 45,000 patents - making it one of the top five in the world. But, as he readily admits, the recent past has not been easy for the company, with thousands of lay-offs as it axed 20 per cent of employees, company consolidation including the closure of its Cambridge University research centre (since folded into Grenoble), and the restating of some $2 billion in income.

Mr Machon is confident the worst is past: "We were one of the first into the crisis, and we will be one of the first to exit." Xerox has sold assets, restored its cash position, and securitised $5 billion of its business with GE Capital, he says.

"We have never cut research and development," he adds. It was always key, he says, remaining at 6 per cent of revenue - $1 billion in 2001, and the same in 2002.

Xerox Research Centre Europe is confident its natural language research (it also specialises in areas such as nanotechnology) will help boost the company's bottom line in the future. For now, the centre makes money through technology and patent licences, sponsored research and development, and consultancy work.

And as for solving the natural language problem? Dr Isabelle expects much progress, but in terms of an intelligent machine capable of its own consciousness? "I'm not sure if I will see it in my lifetime," he says wistfully.