ad info


Asiaweek TIMEASIA.com CNN.com
 > technology
 home
 intelligence
 web features
 magazine archive
 technology
 newsmap
 customer service
 subscribe
 TIMEASIA.COM
 CNN.COM
  east asia
  southeast asia
  south asia
  central asia
  australasia
 BUSINESS
 SPORTS
 SHOWBIZ
 ASIA WEATHER
 ASIA TRAVEL

Other News
TIME.com
TIME Europe
FORTUNE.com
FORTUNE China
MONEY.com
Asiaweek Services
Contact Asiaweek
About Asiaweek
Media Kit
Get up to 3 months of Asiaweek free when you subscribe online!


SEPTEMBER 1 , 2000 VOL. 26 NO. 34 | SEARCH ASIAWEEK

Say it in Your own Words
Language software is opening new doors

With mobile communications and Internet use booming in Asia, voice recognition and translation software are becoming more important than ever. Today there are programs that can provide crude translations of webpages in various languages or take relatively accurate computer dictation from a designated speaker. There's still a long way to go before the Tower of Babel comes crashing down. But leading software makers like Belgium-based Lernout & Hauspie envision phones that can translate live conversations and systems that can read out e-mail and transcribe your response while you drive. Asiaweek's Yasmin Ghahremani caught up with Louis Woo, president of L&H Asia-Pacific, in Hong Kong recently.

What are the main developments in voice and language technology in Asia?

The two means of communication between humans usually are speech and text. Working from speech to text, we have what we call automatic speech recognition, which could be as sophisticated as maybe 10,000 words for a call center. We have an auto-attendant system that we provide to the largest bank in South Korea, Hanvit Bank. If I call up and say I'm looking for Louis Woo, it would be able to find that person. Or you could say, "I don't know the person's name but he's an account manager and his name could be W-O-O." It should be able to find the person. Speech-to-text is really just recognizing what you say. But more importantly, it will not only recognize what you say but understand what you say. That's what we call natural language understanding.

Another type of technology is dictation. We already cover 16 languages. In Asia we have Japanese, Korean, Mandarin and Cantonese. We want to expand to 36 within the next two or three years.

Text-to-text software includes machine translation and also intelligent content management [e.g., search engines that "understand" context]. Text-to-speech is technology that, for example, can read out text in a nearly natural-sounding synthesized voice.

Unified messaging services are an example [of how these technologies are used today]. There are some portals already in the U.S. today that, if I'm expecting a very important message from you and I don't have a PC with me, I can just call the portal service and say I'm looking for this e-mail. Get it and read it to me. They won't have a live person there. I don't want a live person reading my e-mail, which could be confidential. Even in Singapore we have three companies that are already incorporating our technology into unified messaging.

After 50 years of research, machine translations are only 80% accurate. Translations like, "I am like to meet you" are still common. What's the hold-up?
That 80% accuracy is for general topics - topics that the machine doesn't know exactly so it doesn't really have the context to start with. Machine translation today is more useful to domain-specific areas. For example if it knows the following [text] will focus on finance, or medicine, then the accuracy tends to increase tremendously. We are building special machine translation engines for particular corporations. We build special dictionaries because each company might have different definitions and names of their products and things like that.

Even humans don't have perfect understanding. If you talk about Shakespeare's works I would not be able to appreciate the breadth and depth of discussion of Shakespeare as literature. Similarly, it is very difficult to build all of this intelligence into just one machine translation engine.

We are also creating more domain-specific styles. Machine translations are enhanced by knowing what the style of writing is. It's easier to translate technical writings than poems, because poems are more unstructured, Structured is easier, more accurate to translate.

How are firms using your technology to help, say, a U.S. manager communicate with a supplier in China?
I think today most multinational corporations are building intelligent systems, at least for internal usage. Using machine translation to roughly translate the messages back and forth. So people in various countries could understand the gist of it.

Where will this technology be five years down the road?

My father lives in Hong Kong. I live in Singapore. The only language he speaks is Cantonese. And my oldest son is seven years old. My father always asks me how my son's doing in school and I sometimes say, "Why don't you talk to the teacher?" But then I think about it and he can't. The teacher is Malay. She doesn't speak Cantonese. Five years down the road I want my father to be able to pick up the phone, talk in Cantonese, and my son's teacher will pick up the phone and listen in whatever language she chooses, and answer too. And better yet, the voice my son's teacher hears would be my father's voice speaking in English. That could be far out. That's the kind of vision we have.

Write to Asiaweek at mail@web.asiaweek.com

Asiaweek Technology Home | Asiaweek.com Home

AsiaNow


Quick Scroll: More stories from Asiaweek, TIME and CNN

   LATEST HEADLINES:

WASHINGTON
U.S. secretary of state says China should be 'tolerant'

MANILA
Philippine government denies Estrada's claim to presidency

ALLAHABAD
Faith, madness, magic mix at sacred Hindu festival

COLOMBO
Land mine explosion kills 11 Sri Lankan soldiers

TOKYO
Japan claims StarLink found in U.S. corn sample

BANGKOK
Thai party announces first coalition partner



TIME:

COVER: President Joseph Estrada gives in to the chanting crowds on the streets of Manila and agrees to make room for his Vice President

THAILAND: Twin teenage warriors turn themselves in to Bangkok officials

CHINA: Despite official vilification, hip Chinese dig Lamaist culture

PHOTO ESSAY: Estrada Calls Snap Election

WEB-ONLY INTERVIEW: Jimmy Lai on feeling lucky -- and why he's committed to the island state



ASIAWEEK:

COVER: The DoCoMo generation - Japan's leading mobile phone company goes global

Bandwidth Boom: Racing to wire - how underseas cable systems may yet fall short

TAIWAN: Party intrigues add to Chen Shui-bian's woes

JAPAN: Japan's ruling party crushes a rebel ì at a cost

SINGAPORE: Singaporeans need to have more babies. But success breeds selfishness


Launch CNN's Desktop Ticker and get the latest news, delivered right on your desktop!

Today on CNN
 Search
  ASIAWEEK'S LATEST
Web-only Exclusives
November 30, 2000

From Our Correspondent: Hirohito and the War
A conversation with biographer Herbert Bix

From Our Correspondent: A Rough Road Ahead
Bad news for the Philippines - and some others

From Our Correspondent: Making Enemies
Indonesia needs friends. So why is it picking fights?


  TECHNOLOGY
THIS WEEK
Profits: A Chinese tech company must change its focus

Baang: South Korea's edgy computer cafes

Language: Computer scientists borrow from mother nature

ASIAWEEK.COM

ASIAWEEK.com
Vol. 2 No. 5


COVER: Playing the Modern Game
Improve your golf game with better technology

Face Off: Audio recorders

Healthcare: Take care on websites for the unwell

Net Gains: Be wary of stock tips in chat rooms

E-vesting: The high cost of online trading

Asiaweek/CNN Tech Index: The Asiaweek/CNN basket of 40 companies

B2B: Learning the job online

Wired Exec: A Manila publisher at work and play


Asiaweek Technology Home

Asiaweek/CNN Internet Index: Track our Asian high-tech stocks

Back to the top   © 2000 Asiaweek. All Rights Reserved.
Terms under which this service is provided to you.
Read our privacy guidelines.