Helping the web to understand

By Mark Ward Technology correspondent, BBC News

Airbus has been using semantic technology to help design aircraft

Sometimes using a search engine can be maddening. Especially when the keywords you expect to lead straight to the pages detailing what you seek turn out to be only tangentially related to that subject.

Completing the task means finding the words a search engine associates with a subject, rather than those used by ordinary folk to describe the same thing.

The slipperiness of language is at the heart of such troubles.

Many expect that improvements in the way the web works and, in particular, greater use of the so-called semantic web will make those frustrations a distant memory.

Early work on the semantic web stressed the importance of tagging - labelling all the elements on a webpage so computers can work out what they are for and how they inter-relate.

But, said Dr Christian Hempelmann from semantic search engine Hakia, that approach has its limitations.

"It works well on pre-structured data," he said. "It works well in limited domains where the corpus of tags that exists is very well defined.

"But," he added, "it does not do well with the free rein of language."

What scuppers the approach is the ambiguity of language and the fact that people rarely tag consistently.

"It has to try to represent the dirtiness of natural language," he said.

Making meaning

Instead some are using semantic technology in a way that does not try to impose meaning on data. Instead it teases out the sense by seeing how it is used.

Semantic technology can handle the ambiguity of language

"It's all about trying to find a way to give data a consistent meaning," said Keith Walker, spokesman for semantic web firm Metatomix.

Some of the first users of this novel application of semantic technology have been large corporations and organisations that generate huge amounts of data as they go about their business.

Workers can be stumped when searching through that pile of data for the information they need, said Mr Walker.

"It's not easy to find what you are looking for when you are bombarded with raw data and too little understanding," he said.

To help with that, Metatomix builds a database known as a semantic ontology, which attempts to capture how all the different parts of an organisation understand a particular thing.

Some courts in the US have become the first users of semantic ontologies to help all those involved in the judicial process manage the information collected about the people that pass through the courts.

Mr Walker said a "criminal" means very different things to the police, defence lawyers, prosecution team and victims - even though it is the same person under scrutiny.

Understanding that ambiguity can help smooth the flow of data across formerly incompatible computer systems and ensure that nothing is lost as a case comes to a conclusion.

A more tangible example is aerospace giant Airbus, which has created a semantic ontology to help it understand what a wing means to the different groups of engineers engaged in making new aircraft.

For Airbus, data about a wing is generated by many different groups involved in modelling and design.

"Airbus has no formal way of consistently sharing information across these different disciplines today," said Mr Walker.

"There's a need to share so it optimises designs and short circuits the design life cycle which is hugely long and complicated," he added.

Early work using semantic technology to understand the knock-on effects of design choices has helped Airbus work out which will be the most costly, said Mr Walker.

And, he said, sharing that data is not just about helping to cut costs. There are other benefits to developing a greater understanding of what a "wing" means.

"Engineers tend to take design choices they have already done as opposed to investigating alternatives," said Mr Walker. "However, innovation comes from iterating around design choices."

Smarter web

John Davies from telecoms giant BT said this semantic technology could make search engines far more useful.

Efforts to make AI useful could be aided by semantic technology

Rather than typing in a few keywords, the semantic technology will be able to glean the meaning behind the query and reach accordingly, he said.

"It's an information-centred approach in a form you want rather than leaving you to do the analysis when you get the list of links back," he said.

Typing a restaurant name could return a link to a website, add maps, reviews, user ratings and recommended dishes from the latest menu. The search engine will be hard to fool and should be able to discern, much more readily, the intent behind search terms.

For Dr Hempelmann from Hakia this use of semantic technology goes further than just helping companies get to grips with the mountains of data they produce.

"We are giving the machine knowledge and that knowledge enables the machine to understand and act on that," he said.

Applying this technology could mean machines achieve significant insights into the way people understand the world. It could feed a breakthrough in an area of science that has long frustrated mankind.

"What we are talking about is AI," he said. "After all, to fool a true semantic web engine means fooling a human-like intelligence."