SI
SI
discoversearch

We've detected that you're using an ad content blocking browser plug-in or feature. Ads provide a critical source of revenue to the continued operation of Silicon Investor.  We ask that you disable ad blocking while on Silicon Investor in the best interests of our community.  If you are not using an ad blocker but are still receiving this message, make sure your browser's tracking protection is set to the 'standard' level.
Technology Stocks : Intel Corporation (INTC) -- Ignore unavailable to you. Want to Upgrade?


To: Ali Chen who wrote (66972)10/18/1998 4:41:00 AM
From: stak  Respond to of 186894
 
Ali,>>Try first to design a talking computer so it will not
be boring.<< Try PowerMac. It's not too shabby.
stak



To: Ali Chen who wrote (66972)10/18/1998 5:28:00 AM
From: nihil  Read Replies (2) | Respond to of 186894
 
Hey, Ali --

Let me ask you, have you ever typed Navier-Stokes equations in spherical coordinates into a computer using an ascii keyboard and standard font? It is possible, fortunately, to create a very large computer or mental dictionary of more or less arbitary signs and to map them into phonetic words. Weirdly enough, there are parts of the brain that hold a collection of signs that the brain associates both with word and with specific real world objects (such as pig or screwdriver). Not yet clear whether all of these objects are the same across gender within culture or across cultures, or, certainly, over time. Thus it is possible to speak in English of the "integral of this from here to there" and the "partial of this with respect to that." -- it gets very boring, but even this is remarkably easier than using a Chinese typewriter. Few would try to memorize the sounds 20,000 or more hanzi without using a phonetic strategy (and the phonetic elements of each character). The problem of mapping speech into an arbitrary set of signs in a computer -- letters, hanzi or mathematical equations is much simpler than teaching a high school student calculus or how to write a decent English essay. And the computers and software get better and the students seem to get worse. We all appreciate the pioneering work of the Great Danish laureate Victor Borge in inventing "visual punctuation" which made Spoken Danglish so easy to understand. "Ladies and chentlemen, sput, I am here this efening ..." The use of "sput" for the comma, and "sput-sput" for double quotes was inspired, but "s-s-s-pew-sput" for the question mark was sheer genius. Today, of course we do not have to go such explicit detail to be understood y our computer. The computer, bless its heart, can learn our unique style of rising inflection in the ultima instead of insisting on a s-s-s-pew-sput which, in time, can even short out a keyboard.

I think the computer does just about as well today at transcribing speech and mapping it into a standard written language as a human court reporter (some of whom not only stenotype, but record and computer interpret to produce a near error free transcript. I believe that soon this will be overkill, and both parties will stipulate computer generated transcript (possibly backed up by audio or video tape). Computer generated text attempts to record the words not the sounds of the spoken input. It misses the nonverbal cues one generates while speaking ("let the record show that the deponent sneered or snickered"). Since commands to the computer are totally words or symbols (or names of symbols) it is duck soup to train the computer to interpret a wide variety of sequences of phonemes into (largely) unambiguous sequences of (computer) words. Going the other way -- from the computer text to spoken English or Chinese is no harder, but it will be a gas to put in the appropriate emotional content (John. Marcia. John? Marcia? John! Marcia-a-a). Until the computer learns to code affect into the computer interpretation, we will be unable to record or transmit it. Thus the ambiguities of email exchanges, the misunderstandings (complicated by inadequate or failing intelligence) of debates on the net, and the unexplained swerve of the robot automobile into oncoming traffic (I LEFT my heart in San Francisco -- hummed the drowsing driver shortly before his death). Let's bet. I say that in 5 years most users will speak more into the computer than we type. This includes remote entry over our portable intelligent phones and written voicemail. Winner takes the loser to a live poetry recitation in the language of his choice.



To: Ali Chen who wrote (66972)10/19/1998 1:46:00 AM
From: Paul Engel  Read Replies (1) | Respond to of 186894
 
Ail - Looks like the Chinese in China view Speech Recognition as very IMPORTANT.

You should read this, Ail - put down your screwdriver, turn off your mouth and try to learn something for a change.

Paul

{=======================================}
techweb.com
October 19, 1998, Issue: 1031
Section: International

Products that read Web pages, grasp Mandarin due out
soon -- IBM China lab sees gains in speech recognition

Richard Wallace

Beijing - IBM Corp.'s China Research Laboratory (CRL) has reported a series of major developments in Chinese speech- and handwriting-recognition software that are expected to accelerate the growth of China's burgeoning information-technology market.

Slated for introduction to the local computer market over the next several months, the advances include a machine-translation product that translates Web pages from English into Chinese, and software that recognizes
continuous speech in the Mandarin language as input for a PC.

IBM's new translation offering, called WebSphere, will include
English-to-Chinese Web translation among many other capabilities.

"The product is near completion," said George W. Wang, director of the IBM
China Research Lab, in an exclusive interview in his office here last week.
Wang hinted at a number of significant speech-recognition and
language-software advances that will be aimed at the Internet, PC and
embedded computing-device markets.

Among them was a new product based on IBM's ViaVoice
speech-recognition technology for Mandarin, which IBM introduced a year
ago. Dubbed HanWang 99 and sold by HanWang, a local vendor, the
product will combine IBM's ViaVoice speech-recognition engine with Chinese
handwriting-recognition software. Still in the product-development stage, the
technology is said to significantly improve Chinese handwriting recognition
beyond the basic Pinyon, or character-pronunciation systems now widely
used by Chinese computer users. HanWang 99 will be introduced early next
year.

"We want to help the Chinese IT industry to grow," Wang said, adding that
the productization of IBM Research's speaker-in-dependent, continuous
speech-recognition technology will help overcome some of the barriers to
market growth in the Chinese computer market.

Despite slumps elsewhere, China's PC market is one of the world's
fastest-growing computer markets. Over 1.5 million PCs were sold in
mainland China in the first half of 1998, according to Advanced
Forecasting-HuiCong (Cupertino, Calif.), a high-tech market researcher.

"Many Chinese, especially older people, are afraid of the keyboard. This will
help the Chinese computer industry take off," Wang predicted.

Located on the remote, northern periphery of Beijing and adjacent to
Tsinghua University-the MIT of China- IBM's 50-person China Research
Laboratory is the company's newest technology research facility and one of
only four research labs that IBM operates outside of the United States.

Though IBM's R&D capability has been thought to be on the wane, the
company's labs have actually been building strength in recent years, according
to Wang. And while down from their historic high of over 3,400 workers-and
eventually trimmed to 2,600 with budget cuts begun in 1993 by IBM chief
executive officer Louis Gerstner -the head count of IBM Research's labs is
now on the rebound and totals about 3,000.

In addition to the Yorktown Research Lab, IBM has technology research
facilities in San Jose, Calif. and Austin, Texas, and operates three other
international research laboratories in Haifa, Israel, Zurich, Switzerland, and
Tokyo.

"Rather than work independently, the labs work as an integrated team," Wang
said, noting that the CRL was founded in September 1995 and is the first
research facility established in China by a U.S. multinational corporation.
Since then it has become the springboard for speech-recognition technology-
a holy grail IBM has been pursuing since the early 1970s.

"Mr. Gerstner visited China last month and has never wavered in his
commitment to technology research and development," Wang said.

While copper-interconnect and silicon-germanium technology have generated
huge headlines for IBM researchers in the United States, IBM's China
Research Lab is just beginning to see the fruits of more than four years of
intensive research into what Wang calls "speaker independent, continuous
speech recognition" of Chinese language and dialects.

Products based on IBM's ViaVoice technology have already been introduced
in other markets, including those for English, Japanese and about six other
languages. Advanced versions of ViaVoice for Chinese Mandarin and other
dialects like Cantonese will be bundled into Chinese PCs by local vendors and
OEMs.

Accents and dialects

Localization of ViaVoice for specific regions of the Chinese market is a major
focus of the IBM China Research Lab, Wang said. "We're working on several
different accents and dialects" for ViaVoice, including the Beijing dialect,
which "we'll announce soon."

In addition to the English-to-Chinese version of IBM's WebSphere
machine-language translation technology, which Wang said is near completion,
the China Research Lab and other IBM Labs are developing versions based
on "multiple language pairs." The software performs "dynamic translation" of
Web content on a "line-by-line" basis and is said to be the first Web-based
English-to-Chinese Internet translation software. It's designed to run in a Web
browser, such as Netscape Navigator. When introduced later this year,
WebSphere will ship in both server and client versions.

Don't expect perfection

IBM began research into speech recognition and machine translation nearly 30
years ago at the Thomas J. Watson Research Laboratory (Yorktown Heights,
N.Y.). IBM invented the statistical methods for research into speech
recognition and holds more than 60 patents for Chinese speech recognition.

"We still have a lot of work to do to improve the accuracy. But the
[machine-language-translation] accuracy is good enough to understand Web
content a sentence at a time," Wang said.

Language translation is one of computing's most computationally intensive
functions, and is based upon complex modeling techniques. To master China's
bewildering array of accents and dialects, IBM CRL collects statistical
samples across China's many regions.

"We get a sample from one person who speaks or reads for 20 minutes and
we get 500 samples from every region. We're improving the way the system
deals with the accent," Wang said. The process includes refining the ability to
understand language within context by using acoustic modeling techniques and
by building acoustic language databases.

"We're looking into the Cantonese dialect, which is even more challenging than
Mandarin," Wang said. This southern Chinese language has nine "tones," as
opposed to four or five tones for Mandarin.

Chinese and other Asian languages have simple vocabularies, but achieve
nuance of meaning and diction through the complex variations of the
music-like pronunciation of spoken sounds, referred to as tones.

PC-based language translation is not CRL's only charter. "Our number-two
focus is pervasive computing," Wang said, including research on software
development for devices such as "smart cards, handheld computers,
embedded systems and set-top boxes." This work includes applications
software for devices using the Palm Pilot operating system. In this area, IBM
has licensed its Chinese machine-language recognition software to a large U.S.
communications company that's developing a Chinese-language pager
intended to work in conjunction with PDAs like the Palm Pilot.

Handwriting recognition

IBM has developed another handwriting-recognition capability for the local
market based on ThinkScribe, a pen-and-paper technology popularized in the
United States in the CrossPad products of Cross Pens.

"This technology is very relevant to the Chinese market," Wang said. IBM
plans to make an announcement about a Chinese-language pen-digitizer
product in the ThinkScribe line "within months," he said.

IBM's competitors are following Big Blue's lead into the market, and in typical
copycat fashion, there are a lot of would-be imitators.

Intel Corp. recently disclosed plans to set up a China R&D lab, and Lucent
opened a facility last year. And according to one industry report, Microsoft is
readying an announcement of a major investment in China for a new software
research facility. The focus of Microsoft's R&D efforts could not be learned,
but observers suggested that it will involve speech and handwriting-recognition
technology.

"When we heard that Lucent was coming here we asked what the target of
their research was and they told us 'speech,' " Wang said. "Then we heard
Andy Grove wants to start a research effort here. We asked what will be the
focus of Intel's effort? The answer, said Wang: speech. "Intel is building faster
and faster chips and speech uses a lot of Mips," Wang said. Heightened
interest in Chinese-language speech and handwriting R&D is creating a flurry
of interest in IBM's technology, he said.

But even in this remote section of Beijing, where IBM has fostered a close
working relationship with students and faculty of China's leading science and
technology university, Wang cannot avoid the bane of an industry where
seeking an edge in the market or technology is everything: high-tech recruiters.

Have any of his people been approached by the competition since Lucent set
up shop in China and Intel announced plans to do the same?

"One of my researchers got a call the other day," a slightly annoyed Wang
said, adding that he does everything he can to retain valuable employees. "I'd
like to complain but I don't know to whom I should complain."

Copyright ® 1998 CMP Media Inc.