Voice Recognition To Ease Travel Bookings

Feeling left out of the online era because your travelers can't type? This may just be your year. Among the major construction projects under way on the Travel superhighway are new access ramps designed for travelers who carry telephones instead of laptops.

Technology companies like IBM and BBN Hark are rolling forward with speech recognition products that understand not just individual words, but full sentences. And at travel industry suppliers like Aqua Software, Carlson Wagonlit Travel, Omega World Travel and Via World Network, speech recognition systems are on the agenda for 1997.

"This is going to be the year we go from research and development to actual implementation of voice technology," predicted Claude Guay, director of travel distribution for IBM's Travel and Transportation Industries Group. "Our mission is to enable customers to have access to as many channels as possible in the online environment, and voice is the most natural interface for human beings. If we enable the legacy systems of the travel industry to take advantage of this fact, we can provide real consumer convenience."

Behind the scenes, what has made the new systems possible is a leap forward from early systems that translated one word at a time to new ones based on "natural language understanding," where the computer understands speech in its natural context, said IBM natural language systems manager Salim Roukos.

Where early systems prompted users with questions that required one-word answers, newer technology allows the user to take the initiative, and understands phrases like, "I want the first flight to Atlanta," Roukos explained. And now the technology is focusing on a third phase: dialogue systems that not only understand words, but act on them.

"We have demo systems that are accurate 80 or 90 percent of the time, and their understanding improves with the level of the user," Roukos said. "Talking to a machine is a new tool for all of us, and people have to learn how to do it. After they try it a few times, the level of understanding goes up even higher."

The earliest applications of speech recognition software were dictation machines, which relied on the user to speak slowly, one word at a time, while the software transcribed the words. Four months ago, IBM introduced a "continual dictation product" designed to allow doctors to speak into the microphone and have their report print out when they are finished.

Roukos expects travel to be a prime application for voice systems because "the travel domain is constrained largely to talking about flights and prices, so the technology works well. It's rich enough to support a voice system, but not too complex."

Guay said IBM is "in discussion with multiple customers, both vendors and agencies, to partner with us in taking the technology to a business application and demonstrating its applicability to the travel industry." Once contracts are signed, IBM will begin testing immediately.

With airline counter space in major airports at a premium, IBM already is experimenting with laptops that communicate remotely with mainframes, so airline agents with cellular phones can check a traveler's photo ID, handle an electronic ticket and distribute a boarding pass from anywhere in the airport. The goal, according to Guay, is to have you say "This is John Smith, flight 123, seat 2A" to an agent or a telephone, "with no counter space or technical infrastructure."

The mission now is to apply these systems to interfaces in the real world--and especially to those involving the CRSs. IBM's Transaction Processing Facility already handles 90 percent of the world's travel transactions, Guay said.

Also in the works among developers are systems that use voice technology to access online booking systems over the Internet and corporate intranets.

"Those will take a little more time to develop, but they're on the radar screen," Guay said. "That's a different piece of the same puzzle, but the technologies are converging. One of the big issues in corporate intranet products is how to make them accessible to everyone. As providers roll out systems, the challenge is to deploy them to everyone in the corporate environment."

For situations where voice alone does not suffice, IBM also soon will be offering its new "Collaborative Live Information Voice Environment"--or CLIVE--which will allow customers to call into a Help Desk and share data on their computers with a customer-service agent they can see right on the screen. Clients who get lost in the online booking process, for example, could call for help and have an agent appear on their computer screen to complete the online transaction for them. To make such systems viable, "there needs to be an easy user interface on the customer side, and all the work should be done by the agent," said IBM's Tetsu Fujisaki, senior manager of human centric solutions.

Voice technology is already beginning to play a role in the travel booking process at agencies like Omega World Travel in Fairfax, Va. Omega, the eighth-largest corporate travel agency in America, has been working on a voice-based system using software from BBN Hark, which built the first major voice application for the travel industry, the Thomas Cook Travel system that debuted at the National Business Travel Association trade show in 1994.

"We're taking the approach that 20 percent of travelers will use an online booking system, and that speech recognition products can access another 20 percent that wants to use the phone," said Omega MIS director Tim Todd.

Omega has been working with BBN Hark for about 18 months (BTN, Nov. 6, 1995), and has been testing a voice recognition system with a limited number of users at the federal government's Department of Veteran Affairs. That product allows travelers to use a telephone to inquire about flights, but is limited in that it only faxes responses back to them, rather than makes bookings in real time, Todd said.

The agency is now adding a booking capability so that the system will understand such full sentences as "I want to travel from Washington National to Chicago on March 15 at about 9 a.m." and follow up with a confirmed reservation. The agency hopes to begin rolling out that product to the government market by the end of the year, and then to offer it to corporate customers--all at no charge. Three or four months later, Omega plans to market the complete system to other agencies.

BBN Hark also is working with Via World Network, the new limited partnership from Andersen Consulting, which is unveiling its travel system this week (BTN, Feb. 24). Using the BBN Recognizer voice technology engine, the Via Voice product is taking the application to a new level, focusing on teaching the computer to recognize and understand the vocabulary of the travel industry.

"We saw a nugget of good technology in the Thomas Cook system, but we knew it needed to be more scalable, more open, more high-performance, so we completely rebuilt the application," said Via president Elmer Baldwin. "We also didn't feel that the way a person would talk on the telephone to make an airline reservation was at all similar to the dialogue a travel agent needed to enter into an airline reservation system, and decided those were the functions we needed to rebuild. We came up with a list of 30 or 40 functions that needed to interact with the reservation system so the traveler could interact with it in a quick time frame."

Once travelers entered their requests into the computer, however, Via found that while the telephone was suitable for changing a reservation or for ordering a stored trip, it did not work well for making a reservation by selecting from a list of flight options. In the Via system, travelers can call in a request on the phone but have the list of flight choices appear on their computer screen.

In developing its system, Via is tapping Andersen Consulting's own technology expertise as well as linguistics specialists from BBN Hark, who will continue to "train" the computer to understand words that are specific to the travel industry--names of airports, cities and travel suppliers, for example, as well as such linguistic shortcuts as "a week from Tuesday" or "the first flight out." Speech recognition systems need to be "90-plus percent recognition perfect, or people won't use them," Baldwin said.

"People won't use a system that doesn't relate to the way they naturally approach a task, and that's the importance of natural language," agreed former BBN Hark travel manager Terry Sullo, who has moved over to the marketing side of the business.

In the next few months, Baldwin said, Via will look for partners in using Via Voice to distribute airline seats and hotel rooms at the last minute. Travelers interested in booking one of the suddenly ubiquitous "Wednesday specials for the following Saturday" could pick up the phone, identify themselves and say, "I want the E-Saver flight to Chicago on Saturday morning at 10," and have the system book it with no human intervention.

Across the street from BBN Hark in Cambridge, Mass., voice technology developer PureSpeech also has discovered the travel industry.

"Voice technology is very hot this year, and particularly for travel applications," said Jamie Goldstein, business development vice president of PureSpeech. In addition to inking a contract with online travel provider E-Travel in Concord, Mass., PureSpeech has received calls from six travel companies--including car rental companies, airlines and hotels--within the past month, asking about the company's system, Goldstein said. The system can recognize 20,000 words with an accuracy rate in the mid-90 percent range. "Some people are interested in simple things like flight arrival and departure status, but others have more aggressive reservations systems in mind," Goldstein said.

On the more aggressive side, no doubt, will be the mega agencies, many of which expect to begin implementing speech-based systems in 1997. New York-based American Express, for example, has "looked at voice technology from a couple of different angles," said president of corporate services Ed Gilligan. "I think it is going to play a role for the agency this year."

American Express has been following the developments in voice systems, and in particular the developments at BBN Hark, since acquiring Thomas Cook, Gilligan said. "The way I would characterize what happened is that a year ago the voice recognition software probably wasn't producing a high enough quality of service to replace a phone call," he said. "It was maybe at 85 to 90 percent accuracy, and it's got to get to well over 95 percent. But what is available today is much better, and I think in 1997 you will see us and probably many other firms jump in."

The nation's largest corporate agency has looked at 10 different providers of voice recognition, including BBN, which is an Amex customer as well as a supplier, Gilligan noted.

"We're going to have to make a decision in the next month or so on which product we're going to test," he said. "In order for us to succeed, a key goal is to help liberate employees from time-consuming, pain-in-the-neck stuff, such as making a travel reservation. Expense vouchers are another area that can be automated. What I'm trying to decide right now is how much to invest in 1997, in what areas, and with whom to partner--and voice is one of those areas."

Also enthusiastic about voice is Carlson Wagonlit Travel of Minneapolis, which is preparing to roll out a voice-based system using software from technology partner SHL, a division of MCI. Code-named the Traveler Management System, the product allows the agency to "attach to the CRS and talk in a graphical way, bringing in corporate and traveler profiles, and a list of trips taken in the past year," said Carlson chief information officer Dick Smith.

The system will allow connections to travelers via the phone as well. For repeat trips, a traveler can simply say, "I want trip number 3," and the system will say "That's Chicago to L.A.; on what dates do you want to travel?"

The Carlson "telephony" system is scheduled for deployment in the third quarter of 1997, and "will be unique in that the system will operate over a single call center, so no matter where you call in from, the system will route your call, your profile and your travel history to the agent," Smith said. "Our goal is to let our customers talk to us via laptop, telephone, cellular phone or hand-held device, via the Internet or an intranet. And not far in the distant future, we'll support talking to you via cable through your television system, so you can look up whether your plane is leaving on time before you leave your house."

Smith expects that 5 to 10 percent of calls to Carlson will be automated using voice telephony in 1997, and that at some point the agency will market parts of the system to its competitors.

Carlson Wagonlit plans to introduce voice technology to its European customers in their native languages by the second or third quarter of 1998. By then, Smith predicted, "the competition for voice-based systems will be significant."

Another company exploring voice recognition is Aqua Software, which is beta testing two systems.