Advertisement

Hoping to be heard

NEW YORK - This is it. It's real. A decades-old doctoral dissertation has grown into an actual product supported by a fledgling company, which on this early autumn day is being unveiled in a windowless ballroom several stories above the cab-choked streets of Times Square.

The guys from Sonum Technologies Inc. are confident, but fidgety. They stand in their booth at the conference, jackets on, feet spread apart, eyes scanning the moving mass of vendors and middle-managers - a legion in golf shirts and dark suits.

More than 100 other companies are set up all around them, bearing giveaways like Frisbees and orange rubber balls. One company, uncertain of the persuasive power of free candy and pens, has set up a beer tap. Tension and noise run high as purveyors pitching products try to grab the attention of passers-by, especially anyone with a name tag that reads "Microsoft."

It's a make or break moment. Here - more than 200 miles from their Columbia home office - Sonum's people will find out whether they are as good as they think. The company, whose name is Latin for "sound," has spent the last year and a half honing its technology, and is looking for confirmation that the early results are the real deal.

They're not even certain anyone really wants what they could someday offer: Life in the computer age without a mouse or a keyboard.

A long romance

Human beings have long chased, or at least romanticized, the essence of Sonum's technology. Artificial intelligence - machines interacting and conversing with humans - has been glorified in science-fiction and popular culture for at least a half-century: the Jetsons' relationship with their robot maid in the 1960s cartoon series; K.I.T.T, the talkative Trans Am in the campy 1980s television series Knight Rider; mechanical children like Haley Joel Osment's character in the 2001 movie AI: Artificial Intelligence.

But reality hasn't yet come close to the fiction. Some machines can take dictation, but the programs are often unreliable and don't involve real communication or interaction. Others, like those associated with voice-activated telephone menus, accept spoken commands, but are typically driven by a few key words. They don't understand the stammers and stalls, "uhms," "likes," and grammatical errors of everyday - or natural - speech.

And no one has been able to figure out how to make them do so - until now, Sonum contends. The algorithms and programming processes beneath its software aim to teach machines the human language: how to understand it, interpret it and respond appropriately.

Some outside experts maintain that this is an impossible goal. Others have tried to attain it, failed and given up, or at the very least shifted direction. Today, Sonum's four-person full-time technical staff is competing with researchers at major education institutions and technology businesses like Microsoft, who are still working on the project. But most of their competitors no longer claim - as Sonum does - the goal of a type of artificial intelligence that would enable users to completely control computers through speech.

"Speech is the most natural form of interaction, but in some scenarios, it's just a lot simpler and easier to use a mouse and a keyboard," said James Mastan, director of marketing for Microsoft Speech Technologies, a division of the software giant. "It makes more sense."

Still, Sonum's all-male company - made up on the creative side of a bearded college professor, a 34-year-old new father and two 20-something graduate students - swears its "natural language processor" is the one that will make the difference.

During the next several months, The Sun will follow Sonum's infancy, chronicling the birth of a company. It is one of more than 1,000 technology companies created every year in Maryland. Nearly as many also die each year, because willpower and dreams aren't line items on balance sheets and can only sustain a business for so long.

Sonum's seed

In 1981, Sonum's creator, W. Randolph "Randy" Ford, was finishing work on a doctorate in artificial intelligence at the Johns Hopkins University. He had already concurrently earned a bachelor's degree in applied psychology and a master's degree in experimental psychology there.

At the time, few in the commercial realm had heard of the Internet. Only 1 percent of households owned a personal computer, and the only mouse most knew of was the kind invading pantries. But the education world was embracing burgeoning technologies, and Ford was determined to help others wrap their arms around computers.

The Ellicott City native, who graduated from Howard High School in 1968, had entered Hopkins intending to become a clinical psychologist. But he was drawn in another direction after he became entranced by statistical calculators, learning to manipulate the data and write number-crunching programs.

"I fell for computers out of the blue," said Ford, now 54 and living in Marriottsville.

At Hopkins, his newfound computer fascination and his interest in the human mind germinated the mental seed for a startup company some 25 years later.

Part of Ford's inspiration came from Bert F. Green Jr., a professor in the university's psychology department who is thought to be among the first to write a computer program designed to interpret everyday language. Ford's thesis adviser, Alphonse Chapanis, also had a major impact through his efforts within and without the university, which led him to be known as "the father of ergonomics" - the science of making work machines and equipment people-friendly. Chapanis, who died in 2002, championed the importance of the human side to computer interaction, which Ford took to heart.

"I got the roadmap for how to do something significant from them," Ford said. "It was no big light bulb, I took one and applied it to this and said, `That would be a good way to spend your life.'"

Bert Green's original program, written in the 1950s, used a very restricted database: baseball scores. It allowed users to ask typed questions such as "How often did the Red Sox beat the Yankees" and get an answer. But it was a cumbersome process that couldn't handle broad topics and didn't really solve the problem. Green gave up writing programs after that first try.

"Computers had not yet developed to the point where you could interact with them at all well," said Green, who retired from Hopkins in 1992 and lives in Baltimore. "It was too soon."

By the mid-1980s, Ford thought it was time for another try. He launched a company based on his doctoral dissertation, which, like Green's project, focused on finding ways to make communicating with computers easier by making the typed commands more customary. Personal computers were taking hold - they had grown to 14 percent of households by 1985 - and he thought he could further their adoption. But before he could get the company rolling, a new machine hit the market and, he said, stole his thunder: the automated teller machine.

"Bank machines taught everyone how to use a computer," Ford said. "It became very familiar, and that essentially had the effect of taking everything I'd ever done and putting it on a shelf."

His project was relegated to "intellectual hobby" status for two decades, while Ford worked as a commercial programmer. He would think about artificial intelligence from time to time, making mental notes of ways he could improve his ideas, perhaps by adding a speech component. But public and private-sector failures in the field made him leery about trying to resurrect the project, and speech recognition programs - which can recognize and transcribe words, if not necessarily understand them - were still far too elementary and unreliable.

Then, about 1995, in his basement office, Ford read that Japanese engineers were deploying an advanced process for speech recognition that might actually work. That energized him. It could be the breakthrough he was looking for, he thought.

"That put me in a watching mode for about two or three years," Ford said. By 1998, he began to devote more time to reviving his natural language processor. He began looking for financial support in 2000, about the time he also took a position as a computer science professor at Hood College in Frederick.

"I was in the game again," Ford said. "I just didn't know it was going to take almost 20 years for everything to come into place."

Making machines listen

The idea behind most artificial intelligence is to take normal, conversational speech and translate it into machine speak behind the scenes through something called a "natural language processor."

Some methods try to do this by teaching computers to diagram sentences and look for subjects and verbs to act upon. But they require immense processing power, gobble up computer memory and struggle to comprehend flawed grammar.

Other techniques train computers to look for keywords and ignore the rest, including sentence structure. But to use them, people have to follow a script of commands that can be annoying to learn and execute.

Some businesses are already using interactive computer technology. Amtrak and UAL Corp.'s United Airlines, for example, have automated their telephone information and reservations systems using software from a Boston company, ScanSoft Inc. The companies estimate they have saved $40 million or more combined in labor costs.

Those processors work because they're limited to a "particular domain," said Susan Feldman of International Data Corp., a technology research firm in Massachusetts. They only have to know a few words, such as "flight" and "departure."

Sonum's vision is much more encompassing. Its goal is to enable anything, even an entire operating system, to be run through natural speech.

The company's patent-pending processor, developed by Ford, is based on the way the human brain organizes information. It has been fed a huge vocabulary along with ways to interpret phrases through repeated simplification and classification. Words and their meanings are assigned numbered codes that the computer can understand and act upon.

For the New York conference six weeks ago, called SpeechTEK, Sonum applied the processor to a Microsoft Excel program for demonstration purposes.

"For Randy, I think that's just step number one so that he can move forward and get his company established," said Judith Markowitz, a technology consultant in Chicago who met Ford three years ago through a colleague at Northwestern University. The computer as "a true assistant, that's his goal," she said.

And he's "considerably ahead of other people in the field," said Walter Rolandi, a consultant for voice telephone systems and columnist for Speech Technology Magazine, which sponsored the New York conference. "The fact that Randy Ford is a psychologist and looking at it as a psychologist helps. He understands the functional nature of language."

Still, technology alone is not enough to sustain a business, and never was. If patents are an indication, tech ideas are created by the thousands every year, but most are never actualized because their inventors can't find backers or partners with business sense.

A half-dozen years ago, new technology companies were less worried about making it. With business plans scribbled on cocktail napkins, they sprouted like mushrooms, fed by investors enthralled by the unknown, seemingly infinite potential of the Internet and telecommunications. But after the technology investment bubble burst, erasing up to $5 trillion in stock value on paper, developing companies have a tougher time allaying skeptics and raising money.

"[Early] investors ran as far and as fast away as they could," said Christopher C. Foster, deputy secretary of the Maryland Department of Business and Economic Development. "They're starting to come back, but I think you're going to see [early] investors become a lot more sophisticated."

Investors are returning, slowly, more warily. Venture capital funding in the second quarter of this year was at its highest in two years, according to a survey by PricewaterhouseCoopers, Thomson Venture Economics and the National Venture Capital Association. More than $5.6 billion was invested in 761 companies, many of them early stage businesses. Headlines about Google Inc.'s rising stock and America Online Inc.'s $435 million purchase of Baltimore-based Advertising.com also rekindled tech dreams.

Building a team

Sonum knows it needs good guidance to survive. The company already has a seven-person board of directors made up of mostly investors, and a management team that has the benefit of hindsight from technology businesses gone awry.

"I don't like to fail at things I start," said Andy Meister, 41, Sonum's chief operating officer. He was formerly a co-founder and vice president at Aether Systems Inc., an Owings Mills company once described as the "Microsoft of wireless communications." For a time, it was the hottest new company in the region, but it eventually shriveled and exited technology entirely.

When Meister came on board, the Sonum team was still in a research and development phase, liable to take on a "deer in the headlights" glaze when business questions were posed, he said. But they believed fervently in their technology. "It reminded me a lot of the early days at Aether," Meister said.

Basement beginnings

The company was based in the basement of the house of its chief executive officer, Matthew B. Hitt. A 49-year-old Towson accountant, Hitt was the first member of Sonum's management team and the first investor - hence also a co-founder. He had been looking for a business to invest in and was smitten by Sonum's ideas. He persuaded a longtime friend, Harry E. Pollock III, to join as president.

"It's an awful lot of fun learning the business and new things. I just hope all of this stuff will come to fruition," said Pollock, 55, a former chief executive officer and president of the Barton-Gillet Company, a marketing communications firm in Baltimore. He now works in sales and financial printing at RR Donnelley's Baltimore offices.

New companies often pull their staffs from friends, families or co-workers. At Sonum, the technical team is led by Ford, who brought on two of his students and one of their friends.

"I left a job where I was getting benefits to go here," said David Gurzick, 26, the company's director of software engineering. He met Ford while earning his master's degree in computer science at Hood College. "He convinced me that this is the future."

After Meister arrived, things started happening fairly quickly. In January, the company set up in rented space in an office building in a Columbia business park, using funds from the $1 million Hitt had cobbled together in financing from "friends and family." The group started shopping their processor, showing it to consultants and pitching it to possible clients. But they were getting minimal feedback.

To reach a wider audience, they produced a demonstration video and headed to New York for the conference. They stocked a booth there with literature, informational DVDs and a $500 rented flat screen monitor for the video, in which Gurzick's wife - an emergency room nurse who was talked into being the impromptu star - repeatedly told an Excel program what to do.

Sipping from a coffee cup and looking relaxed - not at all as if she had only four hours sleep after a hospital shift - Martha Gurzick spoke into a microphone and led the computer through the paces: opening a new spreadsheet, entering figures, tabulating results. She didn't have to type or click on a thing.

Getting noticed

With so many others around them, it was tough to draw attention to their booth on the outskirts of the room, away from the prime center area real estate where Microsoft and IBM had set up. And Sonum only had Hershey Kisses to offer as enticement. Still, for those who saw it, the video was hard to ignore.

"I really don't know how much application it has, but it really has impressed me," said Ann Winn, a programmer for Dragon Systems, which creates voice recognition software. She was one of a few people who stopped by the Sonum booth on the first day of the conference, though most didn't have anything to offer the company other than well wishes and congratulations.

All around the room, brochures promised "conversational speech solutions" and "human-like user experience." But no other company was pitching a voice-computer technology as ambitious as Sonum's. At least that was the report back after a tour around the ballroom by the technical staff: Gurzick, who bubbles with optimism; Doug MacLean, his slightly more cynical counterpart; and Mark Newman, a 24-year-old math whiz. McLean, 34, was the second software engineer hired by Ford. Newman, another student of his, came on as the third in July.

"Within this universe, we know where we stand," said a satisfied Hitt. "We're a step above anybody."

What next?

Back home in Maryland, Sonum went to work with renewed vigor, laying plans for the future, which include another round of financing this fall to enable more hires and contract negotiations with a Seattle company - their potential first buyer. For now, the group is happy and optimistic about the future.

"Truth is, I'm having so much fun. I get to be scientifically creative every day," Ford said. "It's like being at Hopkins all over again."

What Ford doesn't know, is that he and his co-workers are about to find out something that has nothing to do with machines: That is, the value of their individual roles after one member is unexpectedly hospitalized for weeks, fighting for his life.
Advertisement
Advertisement