Internet – Semantics

In the future, the Internet will cease to be a jumble of unstructured knowledge. Instead, it will provide information tailored to each user's requirements.

The above conversation is, of course, fictitious—but it's not science fiction. Researchers around the world are working all out on portals that understand and output speech in a flexible manner, automatically identify users and provide additional information without being asked.

One such researcher is Dr. Hans-Ulrich Block. Together with Dr. Dietrich Klakow from the Philips Research Laboratory, Block, a linguist from the Interactive Technologies department at Siemens Corporate Technology (CT) in Munich, Germany, has developed a concept that is expected to lead to a super-informed dialog system within four years. It already has a name: SInDia.

SlnDia should solve the troublesome out-of-domain problem of today's dialog systems. Such systems may know all the flights to Athens, for example; but they stop working when the caller leaves the "flight information" domain and tries to book a hotel or ask about striking taxi drivers. The goal of the SInDia concept is to include a small talk manager that actively slips interesting information into the dialog in a conversational tone—the weather forecast or tips on special events in Athens, for example. At the beginning, the applications will probably be limited to predefined semantic relationships like cities/weather or automobile/car rental fees. "It doesn't always have to be deadly serious," says Block. "It can be fun too."

In terms of technical requirements, a system of this kind is highly demanding. It must be able to scan texts, tables and interactive services like train schedules and price databases on the Web at incredibe speeds and evaluate them on the basis of their meaning.

"So far, it has been very costly to develop something as simple as a train-schedule service manually," says Klakow. "If we automate that, we can help dialog systems get off the ground and offer far more service." This requires that the computer "knows" what a certain piece of information means and in what context it occurs, so that it can fit it into the dialog appropriately later on.

Today, the business of tagging Web page program codes is still in its infancy and is generally done by hand—"It's work that many Webmasters are reluctant to do," says Block. "But this method would make sense where databases already exist, as is the case with train or flight information," he adds. For new pages, there are Web editors that ask for the tag information as soon as the author inputs text. Furthermore, Block believes tagging will eventually occur automatically by using parsers—programs that dissect Web copy into its grammatical constituents and use these to infer the meaning of sentence elements.

And Wahlster's team is going even further. In the REAL (Resource Adaptive Localization) project, it has developed an application for PDAs that uses data pertaining to changes in position and observations concerning user behavior to identify specific information that could be important for the user at any given moment.

During her flight to Athens, Cynthia Brown has to change planes in Frankfurt. She takes a walk through the shopping area at the airport. The PDA knows where she is and therefore displays information or special offers from the shops. Suddenly, Cynthia hears the last call for her flight to Athens. She looks down at her PDA, which now only displays arrows that point the way to her gate.

In the future, emotions and stress as well as information about shopping behavior, musical tastes or airplane seating preferences will be analyzed while a person surfs the Internet, and stored at the ISP or in the user's PDA. If the customer comes to an Internet portal or a voice response system, the content and/or the dialog will immediately be personalized. "After 20 years of research, we have now developed commercially viable techniques for automatically defining personal user profiles," says Wahlster.

Just as important for the mobile Internet of the future is the development of a new network structure. Despite faster transmission technologies, such as UMTS and wireless LAN, spectrum space is in short supply. One method that could be used to organize the Internet more efficiently is multicast. Thus far, data requested by a thousand users simultaneously has been sent out from the server a thousand times and distributed in a star-shaped pattern. Multicast, on the other hand, is more like a tree structure in which the data is sent only once and then duplicated en route at the appropriate nodes before being passed on to users.

Peer-to-peer techniques are another possibility. These have become well known mainly through the Napster music exchange platform. In this case, the data no longer comes from a server, but is exchanged directly among users .

"Peer-to-peer is becoming more and more important, because people don't just want to consume; they also want to share text, music, images and their knowledge with others," says Michael Finkenzeller of Siemens Corporate Technology (CT) in Munich. In the German research network, peer-to-peer applications already account for 60 % of total data traffic.

Jochen Grimminger, who also works at CT, thinks children might set up ad-hoc networks in schoolyards and exchange melodies or logos directly from cell phone to cell phone without an expensive detour through the mobile phone network. This would take some of the load off networks. Furthermore, if reception were poor, data might be passed to other cell phones that offer better connections to the network. Grimminger has considered billing models for ensuring that no one is taken advantage of in multihop scenarios of this kind. Use of multihop could involve a fee. For instance, network operators could cover their costs by collecting fees for exchanging melodies in ad-hoc networks. On the other hand, those who open up their cell phones for others would collect bonus minutes of telephone time in return.

In the mean time, Cynthia Brown has arrived in Athens. She uses the digicam in her PDA to take a picture of the temple that towers over the city. Shortly thereafter, "Acropolis" appears on the display. The photo is superimposed on a picture that portrays the building as it may have looked 2,000 years ago. Arrows direct her to the top of the hill.

This is very much in the future for the Acropolis, but it is already partly a reality at C-Lab in Paderborn, Germany, where Siemens and university scientists are conducting joint research. The basic idea is a world encyclopedia in which images serve as the input for a search engine. In Paderborn, however, this futuristic PDA scenario is now being investigated by a home appliance manufacturer to present its high-quality products in sales showrooms.

And indeed, the Paderborn researchers have come up with an appetizing solution. If the PDA with its digicam is pointed at a closed oven, the camera records a video of the oven and sends it to a computer in real time via a wireless radio network. The computer analyzes the video and uses the principle of augmented reality (see article Hello, I'm Pump 235) to generate additional image data that is superimposed with the right perspective. For example, on the display, the oven door then opens up, although it is still closed in the sales room. After a few moments, a virtual roasted chicken slowly emerges from the oven. The illusion is so perfect the customer's mouth waters.

Bernd Müller