you'd think that if a big system could do this, by laptop could
reliably do speech-to-text.
i'm not holding my breath here....
---
http://www.forbes.com/technology/2005/10/19/language-translation-cell-
cz_as_1019language.html
Breaking Down Language Barriers
Andy Stone, 10.19.05, 3:00 PM ET
Much sooner than you might expect, it will be possible for an
American in New York to pick up a cell phone and have a conversation
with a Chinese co-worker in Shanghai--with both people speaking their
native languages. Cell phones, using sophisticated translation
technology, will translate the conversation in real time.
One company leading the way to such a smaller world is TransClick, a
systems integrator that puts together all of the technological
components necessary to make real-time translation a reality.
TransClick Chief Executive Robert Levin does not have modest plans;
he describes his technology as breaking down barriers for
international commerce. "The last barrier for global e-commerce is
the language barrier," Levin says.
But the service does not currently handle voice-to-voice translation.
During the first quarter of 2006, TransClick will introduce real-time
short-message service (SMS) for Verizon Wireless (nyse: VZ - news -
people ) corporate customers, who will be able to send SMS messages
from their cell phones to French-, Spanish-, German-, Italian- and
Portuguese-speaking colleagues in Europe.
TransClick plans to expand the service to Sprint Nextel (nyse: S -
news - people ) and Cingular in 2006, with translated connections to
Asia and Latin America.
By November, TransClick will introduce a Web-based instant messaging
program for $5 per month. Levin expects interoperability with popular
SMS programs, like Yahoo! (nasdaq: YHOO - news - people ) Messenger,
by next year, as well as the introduction of translation services for
mobile e-mail.
No single mad scientist buried deep in a university basement is going
to make quality machine translation a reality. Instead, Levin relies
on best-of-breed products from several disciplines. Bilingual
dictionaries form the basis of all translations, giving computers the
raw material needed to correlate words and phrases in one language
with those in another.
The second level involves what's called a translation engine. These
are the grammar- and algorithm-based programs that decipher the
meaning of a particular sentence, breaking it down into its
individual parts and reconstructing it in the target language.
Today's engines concentrate on specific language pairs, such as
Arabic-English, so Levin picks the best engine for any given language
pair. He relies on the Bleu scoring system, originally developed by
IBM (nyse: IBM - news - people ) in the 1990s, to gauge translation
accuracy.
In fact, IBM is also working on its own real-time translation system,
called Multilingual Automatic Speech to Speech Translator, or MASTOR,
which handles Mandarin and English. The technology is in its early
stages.
Companies, like Microsoft (nasdaq: MSFT - news - people ) and General
Motors (nyse: GM - news - people ), use specific engines to translate
their technical manuals into many languages. These programs focus on
limited vocabularies, helping to avoid contextual confusion and
keeping down errors. The best engines, developed over many years and
with large reserves of source data, can translate with 80% accuracy:
Human translators do the final editing.
"Ninety-nine percent of revenue from engine vendors is document
translation. Less than 1% is from gee-whiz applications," says Levin.
"So we're on the cutting edge."
Verizon's SMS service will cost users $3 per month plus 20 cents per
message sent. The technology is simple to implement: Users download a
small application from Verizon's Web site, which connects users to
TransClick's servers.
The international market for machine translation technology could be
huge: 65% of the world's 800 million Internet users don't speak
English comfortably--or at all--a promising statistic for companies
that can get their Web sites translated automatically into as many of
the world's 4,900 languages as possible.
Speech-to-text and speech-to-speech translations present additional
technical hurdles, because voice recognition must be added. Current
voice recognition software has trouble getting even clearly spoken
words correct. Just consider voice-activated customer calling menus.
Nonstandard accents and background noise bring their accuracy down
even further.
But Levin says the first systems, developed for "dirty travel
survival," like using the cell phone to give cab drivers directions
in Chinese, will be ready within two years. Levin is currently
working on voice-to-text translations that will allow tourists to
speak a question into their cell phones. Translated Chinese text will
then appear on the cell screen for another person to read. True
speech-to-speech translation, albeit with limited vocabulary, will be
ready by the end of the decade.
But none of this means that high school students will be able to blow
off their foreign language homework anytime soon. Robust translation
for everyday conversation, taking into account different dialects and
the incredible breadth of spoken language and all of the ambiguities
that currently confound computers, is far off. And as translation
technology becomes more commonly used, the need for qualified human
editors to tidy up translations for publishing will grow. For
example, an English-Chinese translator specializing in medical texts
gets paid $600 or more for the final editing of a short journal article.
---
You are currently subscribed to telecom-cities as: [EMAIL PROTECTED] To
unsubscribe send a blank email to [EMAIL PROTECTED]
Manage your mail settings at
http://forums.nyu.edu/cgi-bin/nyu.pl?enter=telecom-cities
RSS feed of list traffic:
http://www.mail-archive.com/telecom-cities@forums.nyu.edu/maillist.xml