George Risko is a new member of VIPACE and this mailing list.  He is 80
years old and got his first computer system last May.  George was able to
use Omni to read this cover story from Business Week from a photocopy sent
to him by his son.   He and I were happy to find the story archived at
http://www.businessweek.com/1998/08/b3566001.htm

There are many related articles in that issue of Feb. 23, 1998.

Anyone who feels he or she is too old to learn about computers should
write [log in to unmask] - George is looking forward to the day his talking
computer will listen as well.

---- from www.businessweek.com -----

                                LET'S TALK!

   Speech technology is the next big thing in computing. Will it put a PC
                               in every home?

   It's payoff time at IBM's T.J. Watson Research Center in Yorktown
   Heights, N.Y., and the excitement is palpable. Since the 1960s,
   scientists here have been struggling--Henry Higgins-like--to teach
   computers to talk with humans. They've invented powerful programs that
   can recognize what people say with more than 95% accuracy.
   Impressively, last summer, IBM beat most of its competitors to market
   with a jazzy and affordable speech program called ViaVoice Gold. It
   transforms spoken sentences into text on a computer screen and lets
   users open Windows programs by voice command.

   But at Watson, no one seems content with this feat. Instead,
   scientists are scrambling to perfect the next generation of speech
   technology, which will have a profound impact on the way we work and
   live. In one corner of the main speech lab, an intent staff member
   tests an automated ticket-reservation system by asking the computer
   for flight information. Another researcher addresses a computer that
   accesses a database full of digitized CNN news clips. Using nothing
   but spoken words, without any arcane search commands, he plucks out
   video broadcasts on land mines. Down the hall, 34-year-old Mark
   Lucente rotates 3-D images of molecules, cylinders, and topographic
   maps on a wall-size display merely by gesturing and speaking to the
   images.

   With these prototypes, IBM is taking a giant step toward a
   long-cherished ideal of computer scientists and sci-fi fans the world
   over: machines that understand ''natural language''--meaning sentences
   as people actually speak them, unconstrained by special vocabulary or
   context. Computers have been listening to humans and transcribing what
   they say for years. Since the 1980s, a host of startups, including
   Kurzweil Applied Intelligence and Dragon Systems Inc., have sold
   specialized speech-recognition programs that were snapped up by
   doctors and lawyers who could pay a fat premium. But often, such
   programs had small vocabularies, required speaker training, and
   demanded unnatural pauses between words.

   Now, after decades of painstaking research, powerful
   speech-recognition technology is bursting into the marketplace. The
   plummeting cost of computing and a competitive frenzy among speech
   researchers is fueling the long overdue phenomenon. Carnegie Mellon
   University (CMU), Massachusetts Institute of Technology, SRI
   International, Lucent Technologies' Bell Labs, and a welter of small
   companies in Boston, San Francisco, and Seattle are racing to refine
   the mathematics of computer-based speech, license the programs to
   industry, and, in some cases, sell products as bold as Big Blue's
   prototypes. These technologies are no longer pie-in-the-sky, insists
   IBM's top speech researcher, David Nahamoo. ''Without question, 1998
   will be the year of natural-language products,'' he says. ''I feel
   very aggressive about this and very down-to-earth.''

   Speech could be the ultimate bridge between humans and machines.
   Mouse-clicking is fine for firing up a spreadsheet. But few enjoy
   clicking for hours through Internet Web sites, dialogue boxes, online
   application forms, and help menus to find some scrap of information.
   Worse, tasks that require hard-to-memorize commands, or creating and
   finding files you only use on occasion, can be onerous, even
   intimidating. And today's computers lock out those who lack digital
   skills or education, not to mention people with disabilities. Little
   wonder that nearly 60% of U.S. households still don't have a personal
   computer.

   LONG WAIT. Yet suppose, for one golden moment, that people could
   instead say the words: ''Take me to the Titanic Web page,'' and the
   computer would do just that. Suddenly, millions more could be drawn
   into computing and out into the vast reaches of cyberspace. Software
   startup Conversa Corp. in Redmond, Wash., has taken a step in that
   direction with a voice-controlled Web-browsing program--though it's
   still limited to specific phrases, and far from the ultimate dream.
   IBM's 200 speech engineers are working feverishly on natural language
   for products that will locate information--when you say the
   word--either on the Net or in other databases. And Microsoft Corp. is
   spending millions to give future versions of its Windows software the
   gift of gab (page 78). ''Speech is not just the future of Windows,''
   says Microsoft Chairman William H. Gates III, ''but the future of
   computing itself.''

   Machine comprehension of human conversation may never be perfect. And
   PCs driven purely by voice won't hit store shelves this year. In the
   coming months, however, speech pioneers and their pool of early
   adopters will demonstrate, more and more, how voice power can make our
   lives easier. For years, phone companies have used limited
   speech-recognition technology in directory-assistance services. Now,
   Charles Schwab, United Parcel Service, American Express, United Air
   Lines, and dozens of other brand-name companies are testing programs
   that liberate call-in customers from tedious, ''press-one, press-two''
   phone menus. The computer's voice on the line either talks them
   through choices or asks the equivalent of: ''How can I help you?''

   For road warriors, the news is even better: Speech recognition could
   actually save lives. Dozens of companies now offer versions of
   dial-by-voice phone. In some, the driver speaks a key word followed by
   a name, and his cellular phone dials a stored phone number. Other
   types of speech systems tailored to people who can't see or physically
   manipulate keyboards could bring millions off government assistance
   programs and into the workforce (page 74). ''Speech technology shapes
   the way you live,'' says Rob Enderle, senior analyst at Giga
   Information Group in Cambridge, Mass. ''It has a huge impact.''

   Voice power won't be the next megabillion-dollar software market--at
   least not overnight. Total sales of speech-recognition software for
   call centers and other telecom uses--the biggest single
   niche--amounted to just $245 million in 1997, according to Voice
   Information Associates in Lexington, Mass. Because they're so new,
   dictation programs from IBM, Dragon Systems, and others racked up even
   less. Giga reckons sales of all speech-technology companies combined
   won't exceed $1 billion in 2000.

   Beyond 2000, the market for products that use speech could be
   astronomic. But it's unclear what role today's vanguard startups will
   play. Even as use of the technology explodes, demand for
   ''shrink-wrapped'' speech software could dwindle, dragging some of the
   market pioneers along with it. Why? As speech recognition becomes
   cheaper and more pervasive, it will be designed into hundreds of
   different kinds of products, from computers and cars to consumer
   electronics and household appliances to telephones and toys. Like the
   magic of digital compression or high-end graphics capability, speech
   technology may become ubiquitous. In that scenario, companies that
   sell speech-enhanced products--rather than those that developed the
   speech software--hold most of the cards. That could force small speech
   startups to merge, fold, or be snapped up by one of the giants.

   All of this is years in the future. For now, enthusiasm for the new
   technology is drowning out most other concerns. IBM's ViaVoice and
   another dictation program called Naturally Speaking from Dragon
   Systems in Newton, Mass., have won raves from reviewers. William
   ''Ozzie'' Osborne, general manager of IBM's speech business, says unit
   sales in 1997's fourth quarter were greater than the previous two
   quarters combined. Lernout & Hauspie, a Belgian marketing powerhouse
   in the speech field, has seen its stock surge on news of strong sales.

   The best buzz on speech, however, is coming from the telecom crowd.
   Companies desperately need new tricks to spring their customers from
   Help Desk Hell and its voice-mail and call-center equivalents. Lucent
   Technologies Inc., whose Bell Laboratories created the first crude
   speech-recognizer in 1952, is customizing applications for call
   centers at banks and other financial-services firms. In December, it
   completed a trial with the United Services Automobile Assn., a
   privately held financing firm serving mostly military families.
   Customers calling in could discuss their needs with the computerized
   operator, which asks simple questions about the desired type of loan
   and then transfers callers to the appropriate desk.

   This saves each customer 20 to 25 seconds compared with menus and
   keypad tapping, figures USAA Assistant Vice-President Edgar A.
   Bradley. The system sailed through tests even when up against regional
   accents and stammers. The only glitch: Some customers were rattled
   when they realized it was a machine and either slowed down or talked
   too loudly. Bradley's team is working on that. Given time, he says,
   ''we could deploy this throughout the organization.''

   UPS has similar hopes about a speech-recognition system that it
   installed last Thanksgiving. Normally, UPS hires temps in its call
   centers at that time of year to deal with customers worried about
   their Christmas packages. Last year, UPS turned to speech software
   from Nuance Communications, a Silicon Valley spin-off of SRI
   International. Throwing in hardware from another company, the price
   tag ''was in the low six figures,'' says Douglas C. Fields,
   vice-president for telecommunications. Unaided by humans, the software
   responds by voice to customer's inquiries on the whereabouts of their
   parcels. By not adding staff, ''we've already gotten our money back,''
   he says. Operating costs are about one-third what the company would
   have had to pay workers to handle the same number of calls, he adds.

   At UPS, Internet-based tracking has proved even cheaper--and that
   poses a dilemma to speech companies. Potential users may simply prefer
   to beef up their Net capability. On the other hand, argues Victor Zue,
   associate director of MIT's Laboratory for Computer Science, about 90
   million households in the U.S. have phones, vs. some 30 million with
   Net access. ''The future of information-access must be done on the
   phone,'' he declares. Research at Lucent Bell Labs in New Jersey
   supports the point. When incoming calls must be transferred among a
   hundred different locations, ''you can't automate it with a keypad
   menu,'' says Joseph P. Olive, a top speech researcher at Lucent Bell.
   And live operators, he says, will make almost as many mistakes as a
   speech-recognition system.

   Traveling executives are thrilled with speech power. For over a year
   now, Pacific Bell Mobile Services has been testing a voice-activated
   mobile system from Wildfire Communications Inc. in Lexington, Mass. It
   lets drivers place calls or retrieve voice-mail messages without
   taking their hands off the wheel. Sivam Namasivayam, a network
   engineer at Gymboree Corp. in Burlingame, Calif., uses the system
   during his 45-minute commutes, dialing up associates by calling out
   their names and getting voice-mail by speaking key words. He's already
   looking forward to Wildfire's advanced package, in which ''you have
   one phone number, and Wildfire will find you wherever you are,'' says
   Namasivayam.

   PROMISES, PROMISES. Of course, the information that mobile workers
   crave is not always sitting in their voice mail. By harnessing a
   branch of voice technology known as text-to-speech, Wildfire, General
   Magic Inc. in Sunnyvale, Calif., and others have begun demonstrating
   hands-free fax and E-mail from the car.

   General Magic's product, called Serengeti, is a new type of network
   service that users can access by phone or PC, at the office of on the
   road. It communicates with the user via a slick voice interface and
   will carry out your bidding, much like a human assistant, retrieving
   calendar items or reading aloud faxes and E-mail messages that are
   stored in a universal in-box. Chatting with the software agent, ''you
   really feel you are talking to a person,'' says Dataquest Inc.
   principal analyst Nancy Jamison. ''While it's reading, you can order
   it to back up or stop what it's doing and look up a phone number.''

   Some analysts are wary of Serengeti, given General Magic's poor track
   record for popularizing its earlier agent-based products. But there
   are even better reasons for skepticism: More than 100 years of efforts
   in automated speech recognition have left a trail of dashed hopes and
   expectations. Eloquent sci-fi cult characters such as HAL in 2001: A
   Space Odyssey and C3PO in Star Wars make it look so easy. In fact,
   language presents devilishly tough challenges for computers. The
   sounds that make up words consist of dozens of overlapping frequencies
   that change depending on how fast or loud a speaker talks. And when
   words slur together, frequency patterns can change completely.

   Computers cut through a lot of this by referring to stored acoustic
   models of words--digitized and reduced to numerical averages--and
   using statistical tricks to ''guess'' what combinations are most
   likely to occur. Machines can also learn clear rules of syntax and
   grammar. Humans, however, often don't speak grammatically. And even
   when they do, what is a machine supposed to make of slang, jokes,
   ellipses, and snippets of silliness that simply don't make sense--even
   to humans?

   Considering these hurdles, it's impressive that dictation programs
   such as ViaVoice can achieve 95% accuracy. But they can only pull this
   off under ideal conditions. Try putting a bunch of people in a room
   and sparking a lively debate--what scientists call ''spontaneous
   speech.'' Then flick on a dictation program. ''All of a sudden, error
   rates shoot from a respectable level of 10% all the way up to 50%,''
   says D. Raj Reddy, dean of the school of computer science at CMU.
   ''That means every other word is wrong. We have to solve that
   problem.'' Ronald A. Cole at the Oregon Graduate Institute of Science
   & Technology articulates just how high the bar still needs to be
   raised: ''Speech technology must work, whether you have a Yiddish
   accent, Spanish, or Deep South, whether you are on a cell phone, land
   line, in the airport, or on speakerphone. It doesn't matter. It should
   work.''

   Huge technical hurdles are one reason some analysts question the
   viability of today's mushrooming speech startups. There are some
   simple economic reasons as well. For the past several decades,
   universities have spawned many of the key breakthroughs in speech and
   publicized them broadly. So large swaths of the technology are now in
   the public domain.

   For a fee, any company wishing to hone its own speech technology can
   turn to the University of Pennsylvania's ''tree bank''--a collection
   of 50,000 natural English sentences carefully annotated to teach
   machines about syntactic relationships. Ron Cole and his team at the
   Oregon Graduate Institute are posting tool kits that anyone can
   use--for free--to create speech-recognition systems. The only
   stipulation: If they use the tools for commercial gain, they must pay
   a moderate license fee. As computing power gets cheaper,
   speech-recognition technology will be widely available and cheap, if
   not free.

   Knowing that, IBM has tailored its strategy accordingly. Its ingenious
   software comes as a $99 shrink-wrapped product, Via-Voice. But the
   real future of speech recognition, says General Manager Osborne, is as
   an enabling technology. That's why, long-term, IBM's intense effort in
   natural language is geared more to creating products that make use of
   speech rather than selling packaged software.

   One top priority is managing the oceans of information that will
   reside in the multitrillion-byte databases of the 21st century. Within
   10 years, it will be humanly impossible to keypunch or mouse-click
   your way through such mind-boggling repositories, which will store
   everything from 50 years of global currency and interest-rate data to
   the entire sequenced DNA code of every living animal and plant species
   on the planet. IBM wants to sell the database-management software and
   hardware to handle such systems--and give its customers the option to
   address them by voice. ''Speech will drive growth for the entire
   computer industry,'' predicts Osborne.

   Phone companies see it the same way. Lucent, AT&T, Northern Telecom,
   and GTE all own their own speech technology, use it in their products,
   and refine it in their own labs. Some may also license technology from
   speech startups, but none intends to surrender control of the
   technology.

   It's easy to see why. AT&T reports that by managing collect and
   credit-card calls with speech-recognition software from Bell Labs, it
   has saved several hundred million dollars in the past six years.
   Nortel, meanwhile, provides Bell Canada with a system that can service
   4 million directory-assistance callers a month. For now, callers must
   answer prompts such as: ''What city are you calling?'' But a version
   of the software in Nortel's labs goes far beyond this. Armed with
   programs that can handle natural language, the system breezes through
   messy situations where a caller starts out with the equivalent of
   ''yeah...um, gee, I was trying to get, um, John Doe's number.''

   What will the startups do as voice power is increasingly folded into
   products made by the giants? If they aim to be independent, their only
   hope is to stay one step ahead with cutting-edge developments. So far,
   they have done this by collaborating with university laboratories and
   teaming up in the market with other scrappy startups. Consider the
   competitive arena of stockbroking. Startups such as Nuance
   Communications and Applied Language Technologies Inc. (ALTech), an MIT
   spin-off, have attacked this sector in partnerships with nimble
   developers of call-center software, known as interactive voice
   response (IVR) systems.

   Together, they've beaten out potential rivals such as IBM, Lucent, and
   Nortel in pioneering voice-automated stockbroking systems. First out
   the door was Nuance and its IVR partner, Periphonics Corp. of Bohemia,
   N.Y. At the end of 1996, they installed a system for the online arm of
   Charles Schwab. With 97% accuracy, it now handles half the company's
   80,000 to 100,000 daily calls from customers seeking price quotes. And
   the system has begun to handle mutual-fund trading. ''Nuance really
   jumped out ahead with the application at Schwab,'' says John A.
   Oberteuffer, president of Voice Information Associates.

   Rival E*Trade Group Inc. of Palo Alto, Calif., also offers voice-based
   trading, in league with IVR startup InterVoice Inc. The company has
   integrated its call-handling gear with speech-recognition software
   from ALTech. Only 5% of E*Trade's volume is now handled by phone, but
   the number is growing fast, executives there say.

   So Round 1 in the speech contest goes to the welterweights. All that
   could change, though, as IBM gets more aggressive in the
   natural-language arena and as Microsoft folds its speech technology
   into its wide range of products. So far, the software giant's market
   presence has been confined to toys and low-level systems for the car
   dashboard. But Microsoft's high-powered research team, deep pockets,
   and proven savvy about consumer products virtually guarantee the
   company a leadership role once the technology is ready for prime time
   (page 78).

   MONEY TALKS. What will consumer applications look like? MIT's Zue
   suggests four ingredients that prove an application is worth pursuing.
   ''First, it must have information that millions of people care about,
   like sports, stocks, weather,'' he says. The information must change,
   so people come back for more. The context must be clearly defined--air
   travel, for example. And not to be ignored: ''It must be something you
   can make money off of.''

   One system he has constructed meets the criteria, though it isn't yet
   commercial. Called Jupiter, it's an 800 number people can dial for
   weather information on 500 cities worldwide. Jupiter doesn't care much
   what words the speaker chooses--as long as the topic is weather. You
   can ask ''Is it hot in Beijing?'' or ''What's the forecast for
   Boston?'' or ''Tell me if it's going to rain tomorrow in New York,''
   and you get the appropriate reply. Ask about London, and it will ask
   if you mean London, England, or London, Ky.

   Zue humbly points out that Jupiter lacks the kind of whizzy artificial
   intelligence that might help a computer reason its way to a
   conclusion. Nonetheless, ''behind the scenes, something very tough and
   subtle is going on,'' says Allen Sears, program manager for speech
   technology at the

   Defense Advanced Research Projects Agency, which funded Jupiter.
   Several times a day, Jupiter's software connects to the Web and reads
   current weather info from two domestic and two international weather
   computer servers. ''Weather forecasters get pretty poetic, and Jupiter
   has to understand,'' Sears says. ''It's dog dumb, but it is amazing.''

   Sears would like to see a lot more applications like Jupiter. And
   given DARPA's clout, he probably will. For the past 10 years, the
   agency has pumped $10 million to $15 million a year into speech
   research, mainly at research institutes such as MIT, CMU, and GTE's
   BBN subsidiary. It sponsors yearly competitions, in which grantees get
   to pit their latest systems against one another--and use their test
   scores in public-relations wars. DARPA defines the types of
   challenges, or ''tasks,'' to be tested. In the past, these have
   included transcribing newspaper articles with a vocabulary of 64,000
   words, read at normal speed by a human speaker or transcribing
   broadcasts directly from the radio.

   Until recently, the tasks served mainly to refine well-known
   statistical tools that computers use to turn language into text. The
   goals have been incremental--to cut error rates. But DARPA is shifting
   gears. In reviewing future grant proposals, Sears says he will place a
   lot more weight on the dynamics of conversation--something he calls
   ''turn-taking.''

   It's an area where even the best experimental systems today don't
   shine. Most dialogues with machines consist of just one or two turns:
   You ask about a stock, or a movie, and the machine asks you for
   clarification. You provide one more bit of information, and the
   computer completes the transaction. ''From now on, I'm not interested
   unless it's 10 turns,'' says Sears. And for a machine to do something
   really useful--such as help a traveler arrange air tickets involving
   three different cities over a five-day period, ''I see a minimum of 50
   or 60.''

   PIECES OF THE PUZZLE. When will machines finally meet expectations
   like those of Sears or CMU's Reddy? For computers to truly grasp human
   language, they must deal with gaps that can only be filled in through
   an understanding of human context. ''They need a lot of knowledge
   about how the world works,'' says William B. Dolan, a researcher in
   Microsoft's labs.

   This is the type of problem that specialists in artificial
   intelligence have spent entire careers struggling with. One of them is
   Douglas B. Lenat, president of Cycorp Inc. in Austin, Tex. For the
   past decade, he has been amassing an encyclopedia of common-sense
   facts and relationships that would help computers understand the real
   world. Lenat's system, called Cyc, has now progressed to the point
   where it can independently discover new information. But Cyc is still
   years from being a complete fountain of the common sense that
   underlies human exchanges. ''These problems are not remotely solved,''
   muses Bell Lab's Olive. ''It's scary when you start thinking of all
   the issues.''

   That's why most scientists grappling with natural language concentrate
   on small pieces of the puzzle and use tricks to simulate partial
   understanding. Columbia University computer-science department chair
   Kathleen R. McKeown uses something called ''shallow analysis'' to
   elicit machine summaries of long texts. By looking at relationships
   among grammatical parts of speech, such as subject, object, and verb,
   ''we get information about the actor, the action, and the purpose,''
   she says.

   At Rutgers University, Vice-President for Research James L. Flanagan
   and his colleagues take a different tack. They build systems that
   study a person's gestures and eye movements to shed light on the
   meaning of spoken words--similar to Mark Lucente's efforts at IBM
   Watson. If a speaker points when he says ''this,'' a machine must see
   or feel the hand, to make sense of it. Scientist James Pustejovsky at
   Brandeis University, meanwhile, is working on ways to tag information
   on the Internet so that it is presented to individual users in ways
   that suit them. A medical clinician and a biochemist, for example,
   probably are not looking for the same things in a body of biological
   data. ''People require multiple perspectives on the same data,''
   Pustejovsky says.

   Speech is the ideal tool for mining information, in all its forms. And
   most computer scientists believe that the tools will improve on a
   steep trajectory. After all, huge resources are being thrown at the
   problems. In addition to deep pockets at multinationals, such as IBM
   and Microsoft, and at DARPA, there is massive support from governments
   in Europe and Japan and from the Computer Science Directorate of the
   National Science foundation in Arlington, Va. This arm of the NSF is
   funded each year to the tune of $300 million, ''and one of the main
   goals is to make computing affordable and accessible to everyone,''
   says Gary W. Strong, deputy division director for information and
   intelligence systems.

   The NSF has its eye on other emerging technologies. But speech is the
   most promising means for making information universally accessible.
   And it's the only one that is direct, spontaneous, and intuitive for
   all people. We can't guess what kinds of dialogues will evolve among
   humans and machines in the next century. But it's certain that we'll
   all soon be spending a lot more time chatting with computers.

   By Neil Gross in New York and Paul C. Judge in Boston, with Otis Port
   in Redmond, Wash., and Stephen H. Wildstrom in Indian Wells, Calif.
                    ___________________________________


Subscribe to vicug-l at http://trfn.clpgh.org/vipace/vicug/subscribe.html
Archives on-line at http://maelstrom.stjohns.edu/archives/vicug-l.html